RF Linear Accelerators - PDF Free Download (2024)

Thomas P. Wangler RF Linear Accelerators

Related Titles Padamsee, H.

Reiser, M.

RF Superconductivity

Theory and Design of Charged Particle Beams

Science, Technology & Applications 2008

ISBN: 978-3-527-40572-5

Brown, I. G. (ed.)

The Physics and Technology of Ion Sources

1994

ISBN: 978-0-471-30616-0

Edwards, D. A., Syphers, M. J.

An Introduction to the Physics of High Energy Accelerators

2004

1993

ISBN: 978-3-527-40410-0

ISBN: 978-0-471-55163-8

Thomas P. Wangler

RF Linear Accelerators 2nd completely revised and enlarged edition

The Author Thomas P. Wangler Michigan State University USA

Cover Peter Hesse Design

All books published by Wiley-VCH are carefully produced. Nevertheless, authors, editors, and publisher do not warrant the information contained in these books, including this book, to be free of errors. Readers are advised to keep in mind that statements, data, illustrations, procedural details or other items may inadvertently be inaccurate. Library of Congress Card No.: applied for British Library Cataloguing-in-Publication Data A catalogue record for this book is available from the British Library. Bibliographic information published by the Deutsche Nationalbibliothek Die Deutsche Nationalbibliothek lists this publication in the Deutsche Nationalbibliograﬁe; detailed bibliographic data are available on the Internet at .  2008 WILEY-VCH Verlag GmbH & Co. KGaA, Weinheim All rights reserved (including those of translation into other languages). No part of this book may be reproduced in any form – by photoprinting, microﬁlm, or any other means – nor transmitted or translated into a machine language without written permission from the publishers. Registered names, trademarks, etc. used in this book, even when not speciﬁcally marked as such, are not to be considered unprotected by law. Printed in the Federal Republic of Germany Printed on acid-free paper Composition Laserwords Private Ltd, Chennai, India Printing betz-druck GmbH, Darmstadt Bookbinding Litges & Dopf GmbH, Heppenheim ISBN: 978-3-527-40680-7

Contents Preface to the Second Edition XI Preface to the First Edition

1 1.1 1.2 1.3 1.4 1.5 1.6

1.7 1.8 1.9 1.10 1.11 1.12 1.13 1.14

2 2.1 2.2 2.3 2.4 2.5 2.6 2.7 2.8

XIII

Introduction 1 Linear Accelerators: Historical Perspective 2 Linac Structures 6 Linac Beam Dynamics 10 Multiparticle Effects 12 Applications of Modern RF Linacs 13 Accelerator-Physics Units, Unit Conversions, and Physical Constants 15 Useful Relativistic Mechanics Relationships 16 Maxwell’s Equations 17 Conducting Walls 19 Group Velocity and Energy Velocity 20 Coaxial Resonator 22 Transverse-Magnetic Mode of a Circular Cylindrical Cavity Cylindrical Resonator Transverse-Magnetic Modes 26 Cylindrical Resonator Transverse Electric Modes 27 References 30 RF Acceleration in Linacs 32 Particle Acceleration in an RF Field 32 Energy Gain on Axis in an RF Gap 33 Longitudinal Electric Field as a Fourier Integral 36 Transit-Time-Factor Models 39 Power and Acceleration Efﬁciency Figures of Merit 42 Cavity Design Issues 44 Frequency Scaling of Cavity Parameters 46 Linac Economics 47 References 52

RF Linear Accelerators. 2nd, completely revised and enlarged edition. Thomas P. Wangler Copyright  2008 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-40680-7

Contents

3 3.1 3.2 3.3 3.4 3.5 3.6 3.7 3.8 3.9 3.10 3.11 3.12 3.13 3.14 3.15

Periodic Accelerating Structures 53 Synchronous Acceleration and Periodic Structures 53 Floquet Theorem and Space Harmonics 54 General Description of Periodic Structures 57 Equivalent Circuit Model for Periodic Structures 59 Periodic Array of Low-Pass Filters 61 Periodic Array of Electrically Coupled Circuits 62 Periodic Array of Magnetically Coupled Circuits 63 Periodic Array of Cavities with Resonant Coupling Element 64 Measurement of Dispersion Curves in Periodic Structures 65 Traveling-Wave Linac Structures 68 Analysis of the Periodic Iris-Loaded Waveguide 69 Constant-Impedance Traveling-Wave Structure 72 Constant-Gradient Structure 74 Characteristics of Normal Modes for Particle Acceleration 76 Physics Regimes of Traveling-Wave and Standing-Wave Structures 79 References 81

4 4.1 4.2 4.3 4.4 4.5 4.6 4.7 4.8

Standard Linac Structures 83 Independent-Cavity Linacs 83 Wider¨oe Linac 87 H-Mode Structures 89 Alvarez Drift-Tube Linac 91 Design of Drift-Tube Linacs 96 Coupled-Cavity Linacs 98 Three Coupled Oscillators 99 Perturbation Theory and Effects of Resonant-Frequency Errors 101 Effects from Ohmic Power Dissipation 103 General Problem of N + 1 Coupled Oscillators 105 Biperiodic Structures for Linacs 108 Design of Coupled-Cavity Linacs 111 Intercell Coupling Constant 114 Decoupling of Cavities Connected by a Beam Pipe 116 Resonant Coupling 117 Accelerating Structures for Superconducting Linacs 121 λ/4 Superconducting Structures 121 λ/2 Superconducting Structures 121 TM Superconducting Structures 122 RF Properties and Scaling Laws for TM and λ/2 Superconducting Structures 125 Shunt Impedance for TM and λ/2 Superconducting Structures 127 Stored Energy for TM and λ/2 Superconducting Structures 129

4.9 4.10 4.11 4.12 4.13 4.14 4.15 4.16

Contents

Scaling Formulas for λ/4 Superconducting Structures References 133

131

5 5.1 5.2 5.3 5.4 5.5 5.6 5.7 5.8 5.9 5.10 5.11 5.12 5.13 5.14 5.15

Microwave Topics for Linacs 135 Shunt Resonant Circuit Model 135 Theory of Resonant Cavities 137 Coupling to Cavities 138 Equivalent Circuit for a Resonant-Cavity System 139 Equivalent Circuit for a Cavity Coupled to two Waveguides 144 Transient Behavior of a Resonant-Cavity System 146 Wave Description of a Waveguide-to-Cavity Coupling 148 Microwave Power Systems for Linacs 156 Multipacting 159 Electron Field Emission 162 RF Electric Breakdown: Kilpatrick Criterion 163 Adiabatic Invariant of an Oscillator 164 Slater Perturbation Theorem 165 Quasistatic Approximation 167 Panofsky–Wenzel Theorem 168 References 173

6 6.1 6.2

Longitudinal Particle Dynamics 175 Longitudinal Focusing 175 Difference Equations of Longitudinal Motion for Standing-Wave Linacs 177 Differential Equations of Longitudinal Motion 178 Longitudinal Motion when Acceleration Rate is Small 178 Hamiltonian and Liouville’s Theorem 182 Small Amplitude Oscillations 186 Adiabatic Phase Damping 187 Longitudinal Dynamics of Ion Beams in Coupled-Cavity Linacs 189 Longitudinal Dynamics in Independent-Cavity Ion Linacs 190 Longitudinal Dynamics of Low-Energy Beams Injected into a v = c Linac 192 Rf Bunching 194 Longitudinal Beam Dynamics in H-Mode Linac Structures 196 References 199

6.3 6.4 6.5 6.6 6.7 6.8 6.9 6.10 6.11 6.12

7 7.1 7.2 7.3 7.4 7.5

Transverse Particle Dynamics 201 Transverse RF Focusing and Defocusing 201 Radial Impulse from a Synchronous Traveling Wave 203 Radial Impulse near the Axis in an Accelerating Gap 204 Including Electrostatic Focusing in the Gap 207 Coordinate Transformation through an Accelerating Gap 208

VII

VIII

Contents

7.6 7.7 7.8 7.9 7.10 7.11 7.12 7.13 7.14 7.15 7.16 7.17 7.18

Quadrupole Focusing in a Linac 209 Transfer-Matrix Solution of Hill’s Equation 211 Phase-Amplitude Form of Solution to Hill’s Equation 213 Transfer Matrix through One Period 214 Thin-Lens FODO Periodic Lattice 215 Transverse Stability Plot in a Linac 217 Effects of Random Quadrupole Misalignment Errors 218 Ellipse Transformations 221 Beam Matching 222 Current-Independent Beam Matching 224 Solenoid Focusing 225 Smooth Approximation to Linac Periodic Focusing 226 Radial Motion for Unfocused Relativistic Beams 227 References 230

8 8.1 8.2 8.3 8.4 8.5 8.6 8.7 8.8 8.9 8.10 8.11 8.12 8.13 8.14 8.15 8.16 8.17 8.18

Radiofrequency Quadrupole Linac 232 Principles of Operation 232 General Potential Function 236 Two-Term Potential Function Description 238 Electric Fields 240 Synchronous Acceleration 241 Longitudinal Dynamics 242 Transverse Dynamics 243 Adiabatic Bunching in the RFQ 245 Four-Vane Cavity 248 Lumped-Circuit Model of Four-Vane Cavity 249 Four-Vane Cavity Eigenmodes 251 Transmission-Line Model of Quadrupole Spectrum 254 Radial-Matching Section 260 RFQ Transition Cell 265 Beam Ellipses in an RFQ 271 Tuning for the Desired Field Distribution in an RFQ 273 Four-Rod Cavity 274 Four Vane with Windows RFQ 276 References 280

9 9.1 9.2 9.3 9.4 9.5 9.6 9.7 9.8

Multiparticle Dynamics with Space Charge 282 Beam Quality, Phase Space, and Emittance 283 RMS Emittance 285 Transverse and Longitudinal Emittance 287 Emittance Conventions 288 Space-Charge Dynamics 289 Practical Methods for Numerical Space-Charge Calculations RMS Envelope Equation with Space Charge 296 Continuous Elliptical Beams 297

292

Contents

9.9 9.10 9.11 9.12 9.13 9.14 9.15 9.16 9.17 9.18 9.19 9.20 9.21

10 10.1 10.2 10.3 10.4 10.5

10.6 10.7

11 11.1 11.2

11.3 11.4 11.5 11.6 11.7 11.8 11.9

Three-Dimensional Ellipsoidal Bunched Beams 299 Beam Dynamics Including Linear Space-Charge Field 300 Beam-Current Limits from Space Charge 302 Overview of Emittance Growth from Space Charge 303 Emittance Growth for rms Matched Beams 306 Model of Space-Charge-Induced Emittance Growth in a Linac 314 Emittance Growth for rms Mismatched Beams 316 Space-Charge Instabilities in RF Linacs from Periodic Focusing: Structure Resonances 318 Longitudinal-Transverse Coupling and Space-Charge Instabilities for Anisotropic Linac Beams 319 Beam Loss and Beam Halo 325 Los Alamos Beam Halo Experiment 329 Scaling of Emittance Growth and Halo 331 Longitudinal Beam Dynamics Constraint on the Accelerating Gradient 332 References 338 Beam Loading 341 Fundamental Beam-Loading Theorem 342 The Single-Bunch Loss Parameter 344 Energy Loss to Higher-Order Cavity Modes 344 Beam Loading in the Accelerating Mode 345 Equations Describing a Beam-Loaded Cavity 347 General Results 348 Optimum Detuning 350 Extreme Beam-Loaded Limit 351 Numerical Example of a Beam-Loaded Cavity 351 Example of a Heavily Beam - Loaded Superconducting Cavity with Bunches Injected on the Crest of the Accelerating Wave 352 Generator Power when the Beam Current is Less than Design Value 352 Transient Turn-On of a Beam-Loaded Cavity 354 References 360 Wakeﬁelds 361 Image Force for Line Charge in Round Pipe 362 Fields from a Relativistic Point Charge and Introduction to Wakeﬁelds 364 Wake Potential from a Relativistic Point Charge 367 Wake Potentials in Cylindrically Symmetric Structures 368 Scaling of Wake Potentials with Frequency 370 Bunch Wake Potentials for an Arbitrary Charge Distribution 371 Loss Parameters for a Particular Charge Distribution 376 Bunch Loss Parameters for a Gaussian Distribution 377 Beam-Coupling Impedance 378

Contents

11.10 11.11 11.12 11.13 11.14 11.15 11.16 11.17 11.18

12 12.1 12.2 12.3 12.4 12.5 12.6

12.7

12.8

Longitudinal- and Transverse-Impedance Deﬁnitions 380 Impedance and Wake Potential for a Single Cavity Mode 381 Short-Range Wakeﬁelds-Parasitic Losses 383 Short-Range Wakeﬁelds: Energy Spread 383 Short-Range Wakeﬁelds: Compensation of Longitudinal Wake Effect 384 Short-Range Wakeﬁelds: Single-Bunch Beam Breakup 384 Short-Range Wakeﬁelds: BNS Damping of Beam Breakup 386 Long-Range Wakeﬁelds and Multibunch Beam Breakup 389 Multipass BBU in Recirculating Electron Linacs 397 References 402 Special Structures and Techniques 405 Alternating-Phase Focusing 405 Accelerating Structures Using Electric Focusing 406 Coupled-Cavity Drift-Tube Linac 410 Beam Funneling 411 RF Pulse Compression 413 Superconducting RF Linacs 414 Brief History 415 Introduction to the Physics and Technology of RF Superconductivity Examples of Operating Superconducting Linacs 419 Atlas 419 CEBAF 419 Spallation Neutron Source 421 Future Superconducting Linac Facilities 423 International Linear Collider 423 Next-Generation Rare Isotope Facility 426 Free-Electron Lasers 427 References 430 Index

433

416

Preface to the Second Edition In the nine years since this book was published there has been continuing progress in many areas of the RF-linac ﬁeld. This is not surprising because linacs are a critical enabling technology for high-energy physics, nuclear physics, neutron spallation sources, and free electron lasers (FELs). My motivation for the ﬁrst edition was that the RF-linac ﬁeld needed a textbook. The second edition is motivated by two considerations: 1) the need to add additional material that is important for those who are designing today’s new linacs, and 2) to include new developments and new linacs that have been built during the past decade, as well as proposed new linacs for which major design efforts have been underway. This is still a textbook, but addition of the new material makes the second edition more useful as a reference book. Several chapters have been expanded to address many developments during the past decade. New developments in the linac ﬁeld during this time have included increased development of H-mode structures (Chapter 4), accelerating cavities for superconducting linacs (Chapter 4), multipass BBU in recirculating linacs and energy-recovery linacs (Chapter 11), and two new promising electric-focused structures under development (Chapter 12), the RF-DTL or RFD and the RF-Interdigital or RFI structure. The last paragraph of Section 4.15 in this new addition describes a simple basic principle for resonant coupling of two cavities. In addition, Chapter 6 includes a simple model for RF bunching and longitudinal beam dynamics for long H-mode structures. Chapter 7 includes a new section on current-independent matching. Chapter 8 on the Radiofrequency Quadrupole Linac is expanded to include treatment of the RFQ radial matching section, beam ellipses in the RFQ, RFQ tuning, RFQ transition cells, and the Four-Vane with windows RFQ. Chapter 9, which covers multiparticle beam dynamics with space charge, is considerably expanded to include emittance growth for both rms-matched and rms-mismatched beams, space-charge instabilities from periodic focusing in RF-linacs, longitudinaltransverse coupling, and space-charge instabilities for anisotropic beams and how to avoid them. The Los Alamos beam-halo experiment which tested the two main analytic models of beam-halo formation is described in Chapter 9; RF Linear Accelerators. 2nd, completely revised and enlarged edition. Thomas P. Wangler Copyright  2008 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-40680-7

XII

Preface to the Second Edition

also included in Chapter 9 is the longitudinal beam-dynamics constraint on accelerating gradient. Discussed in Chapter 12 are the new linacs that have been built and operated within the past decade. Included is the spallation neutron source (SNS), an accelerator based neutron source that was designed by six National Laboratories and constructed at Oak Ridge National Laboratory for the U.S. Department of Energy. The majority of the SNS linac uses superconducting cavities for beam acceleration. At its full beam power of 1.4 MW, SNS will provide the worlds most intense pulsed neutron beams for scientiﬁc research and industrial development. In February, 2007 the SNS accelerated beam to its design energy of 1 GeV, a new world record for a proton linear accelerator. Another important machine built and operated during the past decade is the Jefferson Laboratory FEL, which operates as an RF-superconducting energyrecovery linac (ERL). JLAB has demonstrated ERL operation in continuous wave (CW) mode at high power. In July, 2004 10 kW of radiated power was produced at 6 micron wavelength, and in October, 2006, 14.2 kW CW was achieved at 1.6 microns wavelength. Included among proposed future linacs are the International Linear Collider (ILC), a proposed facility based on RF-superconducting cavities that would produce high-energy collisions between electron and positron beams at centerof-mass energies from 200 to 500 GeV. Included for future nuclear physics is a next-generation radioactive ion beam facility. As was remarked in the preface to the ﬁrst edition, superconducting RF linacs will become increasingly important in the RF-linac ﬁeld for many applications. This is clearly illustrated by the examples in Chapter 12 of both new operating linacs and proposed future linacs. In all cases superconducting linacs are an important part of these machines. The discussion of superconducting RF technology in Chapter 12 is intended to provide a brief and systematic presentation of the physics and technology for the new student. A more comprehensive treatment is available in the excellent textbook RF Superconductivity for Accelerators, by Hasan Padamsee, Jens Knobloch, and Thomas Hays. I wish to acknowledge a very productive collaboration with Jim Billen in teaching the USPAS linac courses. Jim’s contribution included developing and documenting the problem solutions, and working with the students. In putting together this new edition, I have beneﬁted from many colleagues. I especially want to acknowledge Jim Billen, Ken Crandall, Jean Delayen, Bob Gluckstern, Ingo Hofmann, Andy Jason, Martin Reiser, Uli Ratzinger, Alwin Schempp, Don Swenson, and Peter Walstrom. I want also to acknowledge the support and helpful suggestions of my wife Julie. Los Alamos, 2007

Thomas P. Wangler

XIII

Preface to the First Edition This book is based on my RF linear accelerator course presented through the US Particle Accelerator School (USPAS), sponsored by the Department of Energy; and organized by Mel Month. The course has been presented through the USPAS at the University of Texas, Harvard University, Duke University, and at University of New Mexico, Los Alamos. The material has been written assuming that the student has a basic knowledge of classical and relativistic mechanics, as well as electromagnetism. Thus, the material should be suitable for engineering students as well as advanced undergraduate or graduate students of physics. The ﬁeld of charged-particle accelerators is an impressive product of twentieth-century science and technology. It is an interdisciplinary ﬁeld using state-of-the-art technologies, and draws on the skills of experts from different science and engineering ﬁelds. The modern era of radio-frequency linear accelerators began just after the end of World War II. Since that time, the RF linac has emerged as a device of major scientiﬁc and technical importance, addressing a broad range of important applications in basic research, energy, medicine, and defense. The ﬁeld of RF linacs is undergoing rapid development, stimulated particularly by the interest in linear colliders for high-energy physics, high-power proton linacs for advanced neutron sources, and smaller linacs for medical and commerial applications. In many respects the RF linac is well suited to these new applications, especially those which emphasize high-intensity or high-brightness output beams. Continuing developments, especially in RF superconducting technology, are contributing signiﬁcant advances in the linac ﬁeld. Because of the diversity of the future applications, and the steady advances of accelerator technology, we can expect the ﬁeld of RF linacs to continue to grow as we move into the twenty-ﬁrst century. My main motivation for writing this book has been that the RF-linac ﬁeld needs a textbook, and until now there has been none. A textbook is invaluable to new students, who need a systematic and comprehensive introduction to basic principles. Additionally, the textbook can also be of value to accelerator experts. The accumulating body of knowledge in the ﬁeld continues to be communicated mostly in laboratory technical reports and in acclerator conference proceedings. There is an international linac conference, RF Linear Accelerators. 2nd, completely revised and enlarged edition. Thomas P. Wangler Copyright  2008 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-40680-7

XIV

Preface to the First Edition

as well as international particle-acclerator conferences in the U.S. and Europe, each of which takes place every two years. Nevertheless, this published material is generally very condensed, and is written primarily for the specialist in a particular sub-discipline. The nature of the RF linac ﬁeld is such that at the major laboratories, experts can be found in the main sub-disciplines, including beam dynamics, accelerating structures, mechanical engineering, microwave engineering, radio-frequency power systems, vacuum technology, and cryogencis. To contribute more effectively to an RF-linac project, an expert often needs information from sources outside his or her area of expertise. Without a textbook, both the new student and the working scientist or engineer, often struggle to ﬁnd the material needed to broaden their knowledge. The objective of this textbook is to provide a systematic presentation of both the science principles and the technology aspects that are the basis of the RF-linac ﬁeld. I have organized the book in the way that I have found most useful for teaching an introductory course on the principles of RF linacs. The core material consists of the basic introductory linac course. Those sections marked with an asterisk (*) are presented as optional material, which can be omitted without disrupting the overall continuity of the book. Selected optional sections can be added to the basic course to meet the speciﬁc needs of the students. Detailed presentations of speciﬁc applications have been omitted in the book, not because I think these are unimportant, but because it is relatively easy to ﬁnd good and up-to-date review articles on the applications. For all of the major topics, the basic analytic theory is presented, because I believe that understanding the analytic theory is the best way to obtain the insight required for producing a modern linac design. Computer codes and simulation methods are the next step, and I have included introductory presentations of the physics algorithms for some of the most commonly-used beam-dynamics codes that are used to predict the performance of modern linacs. I believe that analytic theory and the computer codes are both important tools; however, in these exciting times of rapid development in the computer ﬁeld, there can be a temptation for overreliance on the computer. Even with such powerful computer codes, there is still no substitute for understanding the basic physics principles. If the trends of the past decade continue, superconducting RF linacs will become increasingly important in the RF linac ﬁeld for many applications. In this text, the superconducting RF technology topics are distributed throughout the chapters; technical information is provided as is needed for each speciﬁc discussion. The discussion of superconducting RF technology in Chapter 12 is indended to provide a brief and systematic presentation for the new student, including the physics and the technology together with some examples of superconducting RF linacs. I recommend to the student that a reading of this introductory section should be followed by the more detailed treatment in the excellent textbook in this same Wiley series, RF Superconductivity for Accelerators, by Hasan Padamsee, Jens Knobloch, and Thomas Hays.

Preface to the First Edition

The problems at back of each chatere have been chosen to develop additional insight to the material presented, to illustrate further a particular point, or to provide the student with a better feeling for typical numerical magnitudes of important quantities. I believe it is important that the latter capability be developed by anyone who needs to apply the material in this book to the design or real accelerators. The references provided are not claimed to be complete, but represent those that I have found most useful. Throughout the book, I have tried to use conventional notation in the accelerator ﬁeld, even thoug this results in an unfortunate duplication of symbols. Such is the nature of the accelerator ﬁeld, much of whose language has been adopted from the notation of older ﬁelds of physics and engineering that form its basis. Although no previous text book on RF linacs exists, books and articles have been published over the years that can be valuable resources for the serious student. Most notable is the 1970 red bible title Linear Accelerators, a compilation of articles edited by P. Lapostolle and A. Septier, which has served over the years as an important reference source for many in the linac ﬁeld. Also abailable are the 1959 Handbuch der Physik article on ‘‘Linear Accelerators’’ by L. Smith, the 1965 book Theory of Linear Accelerators by A. D. Vlasov, the 1982 lectures thorugh the USPAS by G. Loew and R. Talman in A.I.P Conference Proceedings No. 105, and the 1985 book Theory of Resonance Linear Accelerators by I. M. Kapchinskiy. The books Theory and Design of Charged Particle Beams by M. Reiser, and The Physics of Charged Particle Beams by J. D. Lawson also contain much useful material related to linac beam dynamics. I wish to acknowledge a very productive collaboration with Jim Billen in teaching the USPAS linac courses and his substantial efforts in developing and documenting solutions to the problems. I am grateful to Jim Billen, Dick Cooper, and Pierre Lapostolle, who read much of the manuscript and made valuable suggestions, and to Ken Crandall for helpful suggestions on the beam dynamics material. I am grateful to a long list of colleagues in the accelerator ﬁeld from whom I have learned, and who have made important indirect contributions to this book. Thanks to my daughter, Anne, who helped me with many of the ﬁgures. Thanks to Jim Billen, Frank Krawczyk, and Peter Kneisel, who provided some key ﬁgures. Thanks to Mel Month, who asked me to develop and teach a linac course for the USPAS and encouraged me to write this book. Los Alamos

Tom Wangler

1 Introduction My little machine was a primitive precursor of this type of accelerator which today is called a ‘linac’ for short. However, I must now emphasize one important detail. The drift tube was the ﬁrst accelerating system which had earthed potential on both sides, i.e. at both the particles’ entry and exit, and was still able to accelerate the particles exactly as if a strong electric ﬁeld was present. This fact is not trivial. In all naivet´e one may well expect that, when the voltage on the drift tube was reversed, the particles ﬂying within would be decelerated, which is clearly not the case. After I had proven that such structures, earthed at both ends, were effectively possible, many other such systems were invented. – Rolf Wider¨oe [From The Infancy of Particle Accelerators, edited by Pedro Waloschek, see ref. [4]]

During the second half of the twentieth century, the linear accelerator has undergone a remarkable development. Its technological base is a consequence of the science of both the nineteenth and twentieth centuries, including the discoveries of electromagnetism by Faraday, Maxwell, and Hertz in the nineteenth century and the discovery of superconductivity in the twentieth century. The design of a linear accelerator requires an understanding of the major areas of classical physics, especially classical mechanics, and electromagnetism, as well as relativity theory. The linear accelerator has developed as a great tool for learning about the world of subatomic particles. The linear accelerator provides beams of high quality and high energy, sufﬁcient to resolve the internal structure of the nucleus and of its constituent subnuclear particles. Like a microscope, it has probed the internal structure of the atomic nucleus and of the nuclear constituents, the proton and neutron. Measurements made using the beams from an electron linear accelerator have given us our present picture of the proton, that it is made of pointlike particles called quarks. Furthermore, the linear accelerator has been applied in hospitals throughout the world as a source of X rays for radiation therapy to treat cancer. This application may represent the most signiﬁcant spin-off of high-energy and nuclear physics research for the beneﬁt of mankind. The linear accelerator is truly one of the most signiﬁcant examples of high-technological development in the postindustrial era. The sizes of linacs range from a few meters to a few kilometers, and the costs range from a few million to a billion dollars, depending on the ﬁnal energy. RF Linear Accelerators. 2nd, completely revised and enlarged edition. Thomas P. Wangler Copyright  2008 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-40680-7

1 Introduction

As a research tool alone, we can expect that the linear accelerator will have a great future in the twenty-ﬁrst century. The straight-line trajectory avoids power losses caused by synchrotron radiation that accompanies circular radio frequency (RF) accelerators. The capability for providing strong focusing allows high-quality and high-intensity beams that enable precision measurements to be made, and provides high-power beams for many applications. We can anticipate continuing progress in areas such as radio-frequency quadrupole (RFQ) linacs, colliding beams, high-power beams, high-frequency RF power and microwave technology, and RF superconductivity. Further developments in these areas will lead the linac to new performance levels with higher currents, better beam quality, and lower power requirements. We can conﬁdently expect an expansion to new applications in the medical and industrial areas. The purpose of this book is to present the scientiﬁc and technical foundations of the linear accelerator, how it works, and why it will continue to serve as a powerful tool for the study of nature, and for many other practical applications.

1.1 Linear Accelerators: Historical Perspective

It might be expected that the term linear accelerator should refer to any device in which particles are accelerated along a straight line. However, through common usage in the accelerator ﬁeld the term linear accelerator has been reserved for an accelerator in which charged particles move on a linear path, and are accelerated by time-dependent electromagnetic ﬁelds. The abbreviation linac is commonly used for the term linear accelerator. In a RF linac, the beam is accelerated by RF electromagnetic ﬁelds with a harmonic time dependence [1]. The ﬁrst formal proposal and experimental test for a linac was by Rolf Wider¨oe in 1928, but linear accelerators that were useful for research in nuclear and elementary particle research did not appear until after the developments of microwave technology in World War II, stimulated by radar programs. Since then, the progress has been rapid, and today, the linac is not only a useful research tool, but is also being developed for many other important applications. A particle accelerator delivers energy to a charged-particle beam by application of an electric ﬁeld. The ﬁrst particle accelerators were electrostatic accelerators in which the beam gains energy from a constant electric ﬁeld. Each particle acquires an energy equal to the product of its electric charge times the potential drop, and the use of electrostatic ﬁelds led to a unit of energy called the electron volt (eV), equal to the product of the charge times the voltage drop. The main limitation of electrostatic accelerators is that the maximum energy obtainable cannot exceed the product of the charge times the potential difference that can be maintained, and in practice this potential difference is limited by electric breakdown to no more than a few tens of megavolts. RF accelerators bypass this limitation by applying a harmonic

1.1 Linear Accelerators: Historical Perspective

time-varying electric ﬁeld to the beam, which is localized into bunches, such that the bunches always arrive when the ﬁeld has the correct polarity for acceleration. The time variation of the ﬁeld removes the restriction that the energy gain be limited by a ﬁxed potential drop. The beam is accelerated within electromagnetic-cavity structures, in which a particular electromagnetic mode is excited from a high-frequency external power source. For acceleration, the beam particles must be properly phased with respect to the ﬁelds, and for sustained energy gain they must maintain synchronism with those ﬁelds. The latter requirement has led to the name resonance accelerators, which includes the linac, cyclotron, and synchrotron. The ideal particle orbit in an RF accelerator may be either a straight line for a linac, a spiral for a cyclotron, or a circle for a synchrotron. In 1924, Gustav Ising of Stockholm proposed the ﬁrst accelerator that used time-dependent ﬁelds, consisting of a straight vacuum tube, and a sequence of metallic drift tubes with holes for the beam [2]. The particles were to be accelerated from the pulsed voltages that were generated by a spark discharge and applied across adjacent drift tubes. Synchronism of the applied voltage pulses with the beam particles was to be obtained by introducing transmission lines, chosen to delay the pulse from the voltage source to each of the drift tubes. The concept proposed by Ising was not tested at that time, but the publication was very important because it inﬂuenced the young Norwegian student, Rolf Wider¨oe. The ﬁrst RF linear accelerator was conceived and demonstrated experimentally by Wider¨oe in 1927 at Aachen, Germany. It was reported in a paper [3] that is one of the most signiﬁcant in the history of particle accelerators,[4] and which inspired E. O. Lawrence to the invention of the cyclotron [5]. The linac built by Wider¨oe was the forerunner of all modern RF accelerators. The Wider¨oe linac concept, shown in Fig. 1.1, was to apply a time-alternating voltage to a sequence of drift tubes, whose lengths increased with increasing particle velocity, so that the particles would arrive in every gap at the right time to be accelerated. In the ﬁgure, D are drift tubes connected to an alternating voltage source V that applies equal and opposite voltages to sequential drift tubes, G are the gaps between the drift tubes in which the electric force acts to accelerate the particles, and S is the source of a continuous ion beam. For efﬁcient acceleration the particles must be grouped into bunches (shown by

Figure 1.1

¨ drift-tube linac. Concept of the Wideroe

1 Introduction

the black dots), which are injected into the linac at the time when the polarity of the drift tubes is correct for acceleration. The bunching can be accomplished by using an RF gap B between the dc source and the linac. This gap impresses a velocity modulation on the incoming beam that produces spatial bunching at the end of a suitable drift space L. The net effect of the sequence of voltage kicks is to deliver a total voltage gain to the beam, which is greater than the impressed voltage V in any single gap. In Wider¨oe’s experiment, an RF voltage of 25 kV from a 1-MHz oscillator was applied to a single drift tube between two grounded electrodes, and a beam of singly charged potassium ions gained the maximum energy in each gap. A ﬁnal beam energy of 50 keV was measured, which is twice that obtainable from a single application of the applied voltage. This was also the ﬁrst accelerator that had ground potential at both the entrance and the exit ends, and was still able to deliver a net energy gain to the beam, using the electric ﬁelds within. The experiment established the principle that, unlike an electrostatic accelerator, the voltage gain of an RF accelerator could exceed the maximum applied voltage. There was no reason to doubt that the method could be repeated as often as desired to obtain unlimited higher energies. In 1931 Sloan and Lawrence [6] built a Wider¨oe-type linac with 30 drift tubes, and by applying 42 kV at a frequency of 10 MHz, they accelerated mercury ions to an energy of 1.26 MeV at a beam current of 1 µA. By 1934 the output energy had been raised to 2.85 MeV [7] using 36 drift tubes. The original Wider¨oe linac concept was not suitable for acceleration to high energies of beams of lighter protons and electrons, which was of greater interest for fundamental physics research. These beam velocities are much larger, approaching the speed of light, and the drift-tube lengths and distances between accelerating gaps would be impractically large, resulting in very small acceleration rates, unless the frequency could be increased to near a gigahertz. In this frequency range, the wavelengths are comparable to the ac circuit dimensions, and electromagnetic-wave propagation and electromagnetic radiation effects must be included for a practical accelerator system. For example, for an electron linac the lengths of the drift tubes and supporting stems would equal nearly a half a wavelength, and instead of isopotential electrodes they would function more like resonant antennas with high power losses. Thus, linac development required higher-power microwave generators, and accelerating structures better adapted for high frequencies and for acceleration requirements of high-velocity beams. High-frequency power generators, developed for radar applications, became available after World War II. At this time, a new and more efﬁcient high-frequency protonaccelerating structure, based on a linear array of drift tubes enclosed in a high-Q cylindrical cavity, was proposed by Luis Alvarez [8] and coworkers at the University of California. The drift-tube linac (DTL) concept was to excite a mode with a uniform electric ﬁeld in the gaps and zero ﬁeld inside the drift tubes to avoid deceleration when the ﬁeld was reversed. A 1-m diameter, 12-m DTL with a resonant frequency of 200 MHz was built,[9] which accelerated

1.1 Linear Accelerators: Historical Perspective

protons from 4 to 32 MeV. At about the same time at Stanford a new, efﬁcient accelerating structure for relativistic electrons was proposed, consisting of an array of pillbox-cavity resonators with a central hole in each end wall for propagation of both the beam and the electromagnetic energy. The structure was called the disk-loaded or iris-loaded waveguide,[10] and this development led eventually to the 3-km Stanford Linear Accelerating Center (SLAC) linac. From these two projects the ﬁrst modern proton and electron linacs were born. [11] The RF linear accelerator is classiﬁed as a resonance accelerator. Because both ends of the structure are grounded, a linac can easily be constructed as a modular array of accelerating structures. The modern linac typically consists of sections of specially designed waveguides or high-Q resonant cavities that are excited by RF electromagnetic ﬁelds, usually in the VHF and UHF microwave frequency ranges. The accelerating structures are tuned to resonance and are driven by external, high-power RF-power tubes, such as klystrons, or various types of gridded vacuum tubes. The ac (wall plug) to RF efﬁciencies of these tubes typically range from about 40 to 60%. The output electromagnetic energy from the tubes is transported in conventional transmission lines or waveguides to the linac structure. The accelerating structures must efﬁciently transfer the electromagnetic energy to the beam, and this is accomplished in two important ways. First, the resonant buildup of the ﬁelds in the high-Q structure transforms the low ﬁeld levels of the input waveguide into high ﬁelds within the structure and produces a large ratio of stored electromagnetic energy relative to the ohmic energy dissipated per cycle. Second, through an optimized conﬁguration of the internal geometry, the structure can concentrate the electric ﬁeld along the trajectory of the beam promoting maximal energy transfer. The most useful ﬁgure of merit for high ﬁeld concentration on the beam axis and low ohmic power loss is the shunt impedance. One of the main advantages of the linear accelerator is its capability for producing high-energy, and high-intensity charged-particle beams of high beam quality, where high beam quality can be related to a capacity for producing a small beam diameter and small energy spread. Other attractive characteristics of the linac include the following: (1) strong focusing can be easily provided to conﬁne a high-intensity beam; (2) the beam traverses the structure in a single pass, and therefore repetitive error conditions causing destructive beam resonances are avoided; (3) because the beam travels in a straight line, there is no power loss from synchrotron radiation, which is a limitation for high-energy electron beams in circular accelerators; (4) injection and extraction are simpler than in circular accelerators, since the natural orbit of the linac is open at each end; special techniques for efﬁcient beam injection and extraction are unnecessary; (5) the linac can operate at any duty factor, all the way to 100% duty or a continuous wave (CW), which results in acceleration of beams with high average current.

1 Introduction

1.2 Linac Structures

A simpliﬁed block diagram of a linac in Fig. 1.2 shows a linac structure with accelerating cavities and focusing magnets, and supplied with electromagnetic energy by an RF-power system. Beam is injected from a dc injector system. A vacuum system is required for good beam transmission. Electric power is used primarily by the RF-power system. A cooling system (water for normalconducting linacs and liquid helium for superconducting linacs) removes the heat generated by the resistive wall losses. Because the linac uses a sinusoidally varying electric ﬁeld for acceleration, particles can either gain or lose energy depending on the beam phase relative to the crest of the wave. To provide efﬁcient acceleration for all the particles, the beam must be bunched as shown in Fig. 1.3. The bunches may be separated longitudinally by one or more RF periods. Figure 1.4 shows the electric- and magnetic-ﬁeld patterns in a simple cylindrical cavity operated in a transverse-magnetic resonant mode. Such a

Klystron RF power system RF control system

Input beam DC particle injector

Linac structure (Accelerating cavities and focusing magnets)

Electric power system

Figure 1.2

Output beam

Vacuum system

Water cooling system

Simpliﬁed block diagram of a linac.

Accelerating field

Time

Beam bunch

Figure 1.3

Beam bunches in an RF linac.

1.2 Linac Structures

B Z

Figure 1.4 Electric (E) and magnetic (B) ﬁelds for the transverse-magnetic resonant mode in a cylindrical cavity.

mode is characterized by a longitudinal electric ﬁeld on axis, which is ideal for acceleration of a charged-particle beam. An important practical consideration is how to construct an efﬁcient linac using these cavities. There have been several solutions. First, an array of independent cavities can be used, each driven by its own RF generator, and each phased independently to provide acceleration along the entire length. This solution is used for superconducting linacs, where its main advantage is operational ﬂexibility. Another solution is to launch an electromagnetic traveling wave in a long structure consisting of many electromagnetically coupled cells. The travelingwave structure was used for the 50-GeV electron linac at the SLAC. Although the simplest accelerating structure might appear to be a uniform cylindrical waveguide, it cannot provide continuous acceleration of electrons, because the phase velocity of an electromagnetic wave in a uniform waveguide always exceeds the velocity of light, so that synchronism with the beam is not possible. A structure with modiﬁed geometry is required to lower the phase velocity to that of the beam. At SLAC, the linac structure consists of a cylindrical waveguide that contains a periodic array of conducting disks with axial holes, as shown in Fig. 1.5. Each individual cell within a pair of disks is essentially identical to the basic cavity of Fig. 1.4, and the whole structure is equivalent to an array of coupled cylindrical cavities. It can be shown that for this structure the phase velocity can be reduced below the speed of light, as required for particle acceleration. The electrical characteristics of the disk-loaded waveguide structure will be described in more detail in Chapter 3. The other common method of producing acceleration in a linac is to excite a standing wave in a multicell or coupled-cavity array. Several types of multicell structures have been invented for optimum application over speciﬁc ranges of beam velocity. One type of structure is the Alvarez DTL, discussed earlier and shown in Fig. 1.6, which is used to accelerate protons and other ions in the

1 Introduction

Figure 1.5 The disk-loaded traveling-wave structure, also showing the input waveguide through which the electromagnetic wave is injected into the structure at the end cell. The beam propagates along the central axis and is accelerated by the electric ﬁeld of the traveling wave.

S G

Figure 1.6 Alvarez drift-tube linac structure used for acceleration of medium-velocity ions. The beam particles are bunched before injection into the drift-tube linac. The beam bunches being accelerated in the gaps G are shown. They are shielded from the ﬁeld by

the drift tubes, when the ﬁeld has the wrong polarity for acceleration. The drift tubes D are supported by the stems S. The cavity is excited by the RF current ﬂowing on a coaxial line into the loop coupler C.

velocity range from about 0.04 to about 0.4 times the speed of light. Unlike the Wider¨oe structure, in the DTL the ﬁelds in adjacent gaps are in phase, and the spacing of the accelerating gaps is nominally equal to the distance the beam travels in one RF period. The DTL structure is not used for electrons, because electrons are so light that their velocity is already above the applicable velocity region at injection from the dc electron gun. Other coupled-cavity linac structures are used for both electrons and protons in the velocity range above about 0.4 times the speed of light. This velocity corresponds to kinetic energies near 50 keV for electrons, the typical injection energy from an electron gun, and near 100 MeV for protons. For example, a coupled-cavity structure called the side-coupled linac (SCL) is used at the Los Alamos Neutron Science Center

1.2 Linac Structures

(LANSCE) linac at Los Alamos to accelerate the proton beam from 100 to 800 MeV. The transverse-focusing requirements are provided by magneticquadrupole lenses mounted within the drift tubes of the DTL, and between structures in a coupled-cavity linac. The newest accelerating structure for the very-low-velocity range from about 0.01 to 0.06 times the velocity of light is the (RFQ), shown in Fig. 1.7. An electric-quadrupole mode is excited in a cavity resonator loaded with four conducting rods or vanes, placed symmetrically about the beam axis. The RFQ electric ﬁeld provides strong transverse electric focusing, which is an important requirement for low-velocity protons and heavy ions. Acceleration in the RFQ is obtained by machining a longitudinal-modulation pattern on these four elements to create an array of effective accelerating cells and a longitudinal accelerating ﬁeld. The RFQ bunches and captures a dc beam injected from the ion source, and then accelerates the beam to high-enough energies for injection into the DTL . The overall result is a signiﬁcant increase in the focusing strength at low velocities, which enables acceleration of higher-current beams in linacs. In pulsed linacs, one must distinguish between micropulses and macropulses. We will see later that, within each RF cycle, the longitudinal electric ﬁeld produces a stable region (the bucket) for the beam. Consequently, the linac ﬁelds form a sequence of stable RF buckets separated by one RF period. Each bucket may contain a stable bunch of particles called a micropulse.

Figure 1.7 The radio-frequency quadrupole (RFQ), used for acceleration of low-velocity ions, consists of four vanes mounted within a cylindrical cavity. The cavity is excited in an electric-quadrupole mode in which the RF electric ﬁeld is concentrated near the vane

tips to produce a transverse RF electric-restoring force for particles that are off-axis. The modulation of the vane tips produces a longitudinal electric-ﬁeld component that accelerates the beam along the axis.

1 Introduction

When the RF generator itself is pulsed, with a period that is generally very long compared with the RF period, the generator pulses are called macropulses. The linac may be operated continuously, which is called continuous-wave operation. The choice to operate pulsed or continuously depends on several issues. One important issue is the total RF efﬁciency. If the accelerated beam current is small, most of the power in CW operation is not delivered to the beam, but is dissipated in the structure walls. Instead, if the accelerator is operated pulsed, and the current per RF bucket is increased while maintaining the same average beam current, then a larger fractional power is delivered to the beam, and the efﬁciency is improved. Another important advantage for pulsed operation is that the peak surface electric ﬁeld attainable is generally larger for shorter pulses. Thus, if high accelerating ﬁelds are required, pulsed operation may be preferred. The main advantage for either longer pulse or CW operation is to reduce the space-charge forces or other beam-current-dependent effects associated with acceleration of beam with high average currents. These effects can be reduced by spreading the total beam charge over more RF buckets, as is done in longer pulse or CW operation. Because the linac is a single-pass device, the linac length and the ohmic power consumption in the cavity walls may be large compared with circular accelerators, which use the same accelerating cavities over and over. To shorten the accelerator for a given energy gain, it is necessary to raise the longitudinal electric ﬁeld, but this increases the power dissipation and increases the risk of RF electric breakdown. For high-duty-factor operation, the average power density from RF losses on the cavity walls can produce challenging cooling requirements for the conventional copper-cavity technology. Another approach to these problems that has become increasingly successful in recent years is the use of superconducting niobium cavities.

1.3 Linac Beam Dynamics

Multicell ion linacs are designed to produce a given velocity gain per cell. Particles with the correct initial velocity will gain the right amount of energy to maintain synchronism with the ﬁeld. For a ﬁeld amplitude above a certain threshold, there will be two phases for which the velocity gain is equal to the design value, one earlier and the other later than the crest, as shown in Fig. 1.8. The earlier phase is called the synchronous phase and is the stable operating point. It is a stable point because nearby particles that arrive earlier than the synchronous phase experience a smaller accelerating ﬁeld, and particles that arrive later will experience a larger ﬁeld. This provides a mechanism that keeps the nearby particles oscillating about the stable phase, and therefore provides phase focusing or phase stability. The particle with the correct velocity at exactly the stable phase is called the synchronous particle, and it maintains exact synchronism with the accelerating ﬁelds. As the particles approach relativistic

1.3 Linac Beam Dynamics L S

Figure 1.8

Stable (S) and unstable (U) phases, early (E), and late (L) phases.

velocities, the phase oscillations slow down, and the particles maintain a nearly constant phase relative to the traveling wave. After beam injection into electron linacs, the velocities approach the speed of light so rapidly that hardly any phase oscillations take place. With the electromagnetic wave traveling at the speed of light, electrons initially slip relative to the wave and rapidly approach a ﬁnal phase, which is maintained all the way to high energy. The ﬁnal energy of each electron with a ﬁxed phase depends on the accelerating ﬁeld and the value of the phase. In contrast, the ﬁnal energy of an ion that undergoes phase oscillations about a synchronous particle is approximately determined not by the ﬁeld, but by the geometry of the structure, which is tailored to produce a speciﬁc ﬁnal synchronous energy. For an ion linac built from an array of short independent cavities, each capable of operating over a wide velocity range, the ﬁnal energy depends on the ﬁeld and the phasing of the cavities, and can be changed by changing the ﬁeld, as in an electron linac. Longitudinal focusing, obtained by injecting the beam on the leading edge of the wave, is essential for nonrelativistic beams of high intensity. However, RF transverse electric ﬁelds also act on the beam as shown by the radial ﬁeld lines near the edges of the gap in Fig. 1.9, and except for some special cases, the particles that are focused longitudinally experience transverse defocusing forces. Furthermore, additional defocusing effects arise because the injected

Figure 1.9 Electric-ﬁeld lines in an accelerating gap.

1 Introduction

beam particles always have ﬁnite transverse velocities, and the beam particles also exert mutually repulsive Coulomb forces. Thus, provision for transverse focusing must be provided. The most successful solutions for transverse focusing have been either to include separate magnetic-quadrupole focusing lenses or to invent accelerating structures that can provide focusing from the RF transverse electric ﬁelds, such as the RFQ.

1.4 Multiparticle Effects

Some applications require beams of high quality that occupy a small volume of phase space, called the emittance. Small beam phase volume is necessary if a small output focal spot or small output energy spread is required. As the beam intensity increases, several effects begin to increase the phase volume occupied by the beam, and these may eventually lead to loss of the beam. The most serious intensity limitation in ion linacs is caused by the repulsive space-charge forces, which are usually the most important at lower velocities, where the beam density is highest. The repulsive space-charge forces cause additional defocusing, and because these forces are nonlinear, they distort the particle distribution. Space-charge forces can also produce an extended halo of large-amplitude particles surrounding the main core of the beam. The halo particles can strike the walls and contribute to beam loss that causes radioactivity along the accelerating structure. The radioactivity increases the difﬁculty of providing routine maintenance of the linac, and thereby reduces the overall operational availability of the linac. For applications requiring high average beam current, control of the halo and beam losses through strong focusing, adequate aperture radius, and proper matching of the beam distribution to the focusing system becomes an important design requirement. For relativistic electron linacs, the electric (space charge) and magnetic self-ﬁelds from the beam tend to cancel, nearly eliminating the total effective space-charge effect. But, short bunches of relativistic particles produce a highly Lorentz-compressed ﬁeld distribution, and these ﬁelds from the beam interact with conducting-boundary discontinuities, producing scattered radiation, called wakeﬁelds, that act on trailing charges in both the same and later bunches. The wakeﬁelds can also increase the beam emittance. Wakeﬁeld effects can be reduced by damping the higher-order modes that are the major contributors and reducing discontinuities whenever possible. Certain cavity modes, called deﬂecting modes, can be excited by an off-axis beam, and are the most dangerous. These modes can cause further deﬂection of trailing particles and under certain conditions lead to an instability known as the beam-breakup instability, which results in loss of the beam. Finally, beam loading occurs, as the beam itself excites the accelerating mode in the cavities. The beam-induced ﬁeld adds vectorially to the contribution from the generator to produce a modiﬁed amplitude and phase. From another

1.5 Applications of Modern RF Linacs

viewpoint, energy is transferred from the beam to each cavity, which unless otherwise corrected, reduces the cavity ﬁeld and may shift the phase. Beamloading compensation methods are used successfully to maintain the correct amplitudes and phases in the presence of the beam. Solutions for controlling all these high-intensity effects can signiﬁcantly inﬂuence the design choices for the main accelerator parameters, including frequency, aperture radius, focusing characteristics, cavity tuning, and RF system operation.

1.5 Applications of Modern RF Linacs

For electron linacs, applications of recent interest include (1) electron-positron colliders for elementary-particle-physics research, (2) high-quality electron beams for free-electron lasers, (3) pulsed neutron sources for nuclear physics and material sciences, and (4) X-ray sources for radiotherapy. Electron-positron linear colliders are preferred over circular colliders because synchrotron radiation losses, experienced by relativistic electrons in circular accelerators, are avoided. Furthermore, because of the strong focusing in a linac, high beam quality is achieved, which is required for high luminosity and a high collision rate. Design studies, and research and development for linear colliders in the tetraelectron volt range, are being carried out within the framework of an international collaboration [12]. The most successful commercial application of RF accelerators is the small 10–20-MeV electron linacs for cancer therapy. A few thousand electron linacs are used worldwide for medical irradiations, and this number is growing. Small electron linacs are also used for industrial radiography and radiation processing, including radiation sterilization. For proton linacs, modern applications include (1) injectors to high-energy synchrotrons for elementary-particle-physics research; (2) high-energy linacs for CW spallation neutron sources used for condensed matter and materials research, production of nuclear fuel, transmutation of nuclear wastes, and accelerator-driven ﬁssion-reactor concepts; (3) CW neutron sources for materials irradiation studies related to fusion reactors; and (4) low-energy neutron sources for medical applications such as boron–neutron capture therapy. Design studies for large proton linacs have been carried out for the Accelerator Production of Tritium (APT) project [13] and the neutron spallation source projects in Europe, the European Spallation Source (ESS) [14] and in the United States, where the Spallation Neutron Source (SNS) was recently constructed [15]. There are also linac applications for heavy ions, including (1) linacs for nuclear physics research, (2) ion implantation for semiconductor fabrication, and (3) multigigaelectron volt linacs for heavy-ion-driven inertialconﬁnement fusion. The most recently commissioned heavy-ion linac is the lead-ion linac at CERN [16]. A recent worldwide compendium of existing and planned scientiﬁc linacs [17] listed 174 linacs distributed over the Americas, Europe, and Asia.

1 Introduction

Historically, two signiﬁcant large linac projects are the SLC electron–positron linear collider at SLAC, shown in Fig. 1.10, and the LANSCE linac at Los Alamos, shown in Fig. 1.11. The main parameters of these two linacs are summarized in Table 1.1. The SLC at SLAC is the ﬁrst linear collider, built to produce the Z0 vector boson near a center-of-mass energy of 91 GeV. It used the 2-mile SLAC electron linac, which was built in the 1960s [18]. Positrons were produced by bombarding a target with a 30-GeV e− beam. More details of this unique facility are summarized elsewhere [19]. The LANSCE linac,[21] formerly known as LAMPF, began operation in the early 1970s as a pion factory for research in nuclear and high-energy physics. It delivers the highest average proton beam power of any existing accelerator. It can deliver 1-mA, 800-MeV proton beams to a ﬁxed target, or an H beam for multiturn injection into the proton storage ring, where the accumulated beam is extracted in short pulses for neutron scattering research.

Positron beam dump

Electron beam dump Final focus

Electron beam transport

Positron beam transport

50 GeV accelerator 0.2 Gev accelerator Positron production target 33 Gev electron beam transpor 0.2 GeV positron beam transport

Electron damping ring

Positron damping ring 1.0 GeV accelerator 0.2 GeV accelerator Electron source

Figure 1.10 The SLAC linear collider.

1.6 Accelerator-Physics Units, Unit Conversions, and Physical Constants

(a) H+ Source 750 KeV Drift-tube linac 100 MeV Bunching & chopping

(b)

800 MeV Side-coupied linac Experimental areas & proton storage ring

Transition region

H– Source

Figure 1.11 The 800-MeV LANSCE proton linac.

Table 1.1 Parameters of SLC and LANSCE linacs [20]. Facility

SLC at SLAC

e− /e+ collider for high-energy physics research First beam 1967 Species Electrons and positrons Beam intensity 2–3.5 × 1010 particles per pulse Application

Beam pulse 120 Hz, 0.06–3 µs Output energy 46.6 GeV Accelerating 960 3-m structures for structure data traveling-wave acceleration at 2856 MHz; 60–130 MW peak power, 25 kW average power Length

3000 m

LANSCE at Los Alamos (H− data) Linac for high-intensity beams of H− and H+ 1972 H+ and H− 11 mA peak H− (average over 825 µs macropulse) 120 Hz, 825 µs 800 MeV 201.25-MHz DTL from 0.75 to 100 MeV

805-MHz SCL from 100 to 800 MeV 62-m DTL, 731-m SCL

1.6 Accelerator-Physics Units, Unit Conversions, and Physical Constants

In this book we will use the SI or MKS units, with two notable exceptions. The magnetic ﬂux density will sometimes be expressed in gauss. The conversion

1 Introduction

factor between the SI unit tesla and gauss is 1 Gauss = 10−4 tesla. Beamparticle energy will be given in electron volts or eV units, rather than joules. The electron volt is deﬁned as the energy acquired by a particle with charge equal to the electron charge that has been accelerated through a potential difference of 1 V. The conversion factor between electron volts and joules is approximately 1.602 × 10−19 = 1 eV. Finally, instead of the particle mass, we will usually give the rest energy, mc2 in megaelectron volt units. Some frequently used physical constants are given in Table 1.2 [22].

1.7 Useful Relativistic Mechanics Relationships

We assume that the reader has a basic knowledge of classical and relativistic mechanics. In this section we present a brief review of basic formulas from relativistic mechanics that will be useful. Consider a particle of mass m and speed v. If c is the speed of light, it is customary to deﬁne a normalized velocity β as β = v/c, and a relativistic mass factor γ , deﬁned as γ = 1/ 1 − β 2 . Some other important deﬁnitions include the relativistic momentum of a particle, p = γ mv, the kinetic energy, W = (γ − 1)mc2 , the rest energy, mc2 , and the total energy, U = W + mc2 = γ mc2 . The nonrelativistic limit applies when β 1. It is often convenient to convert between velocity, energy, and momentum, and the following relationships are helpful. The conversion from velocity β to kinetic energy W is γ = 1/ 1 − β 2 , W = (γ − 1)mc2 (1.1) The inverse conversion is γ = (W + mc2 )/mc2 , β =

1 − 1/γ 2

(1.2)

The following relationships between small differences are sometimes useful: δγ = γ 3 βδβ, δγ = βδ(βγ ), δW = mc2 δγ , δp = mcδ(βγ ). Particle dynamics

Table 1.2 Physical constants. Speed of light Elementary charge Electron mass Proton mass Atomic mass unita Permeability of free space Permittivity of free space DC resistivity of copper (293 K) a

Mass of (12 C)/12.

c e me mp mu µ0 ε0 1/σ

2.99792458 × 108 m/s 1.60217733 × 10−19 C 0.510 999 06 MeV/c2 938.272 31 MeV/c2 931.494 32 MeV/c2 4π × 10−7 T-m/A 1/µ0 c2 = 8.854187817 . . . × 10−12 F/m 1.7 × 10−8 -m (nominal)

1.8 Maxwell’s Equations

is obtained from Newton’s law relating the force and the rate of change of momentum: F=

d(γ v) dp =m dt dt

(1.3)

For a particle of charge q in an electromagnetic ﬁeld, the Lorentz force on particle with charge q and velocity v in an electric ﬁeld E and a magnetic ﬁeld B, is given by F = q(E + v × B)

(1.4)

1.8 Maxwell’s Equations

The laws describing all classical electromagnetic phenomena are known as Maxwell’s equations. These equations relate the electric and magnetic ﬁelds, and the charge and current sources. Maxwell’s four equations in vacuum (where D = ε0 E and B = µ0 H), expressed in differential form, are as follows: ∇ · E = ρ/ε0 , Gauss’s law

(1.5)

∇·B=0

(1.6)

∇ × E = −∂B/∂t, Faraday’s law

(1.7)

∇ × B = µ0 J + µ0 ε0 ∂E/∂t, Amp´ere’s law

(1.8)

where ρ is the charge density and J is the current density. A charge-continuity equation ∇ · J = −∂ρ/∂t is derived from these equations. Maxwell’s equations can also be expressed in what is often a more convenient integral form: 1 ρdV, Gauss’s law (1.9) E · dS = ε0 B · dS = 0 (1.10) ∂B E · dl = − · dS, Faradays law ∂t ∂E B · dl = µ0 · dS, Amp´ere’s law J + ε0 ∂t

(1.11) (1.12)

It can be shown that the electric and magnetic ﬁelds can propagate as electromagnetic waves. When the charges and currents are zero, the wave equations in Cartesian coordinates are as follows: 1 ∂ 2E 1 ∂ 2B = 0; ∇ 2 B − 2 2 = 0 2 2 c ∂t c ∂t √ where c = 1/ µ0 ε0 is the speed of light in vacuum. ∇2 E −

(1.13)

1 Introduction

Maxwell’s equations are composed of four coupled ﬁrst-order partial differential equations. Two of the equations, Eqs. (1.6) and (1.7), have no charge or current source terms, and may be called the hom*ogeneous equations. The other two equations, Eqs. (1.5) and (1.8), do contain source terms and may be called the inhom*ogeneous equations. Although in principle the four equations may be solved for any given problem, it is often convenient to solve a problem using potentials from which the ﬁelds may be derived. It is common to deﬁne the scalar potential φ, and the vector potential A, which are functions of space and time, such that B = ∇ × A,

E = −∇φ −

∂A ∂t

(1.14)

With these deﬁnitions it can be shown [23] that the two hom*ogeneous equations are automatically satisﬁed. The potentials are not uniquely speciﬁed from Eq. (1.14), and uncoupled source equations may be obtained by substituting Eq. (1.14) into Eqs. (1.5) to (1.8), and by imposing what is called the Lorentz condition,[24] ∇·A+

1 ∂φ =0 c2 ∂t

(1.15)

The resulting equations for the potentials have the symmetric, decoupled form of inhom*ogeneous wave equations [25] ∇2 φ −

1 ∂ 2φ = −ρ/ε0 c2 ∂t2

(1.16)

∇2 A −

1 ∂ 2A = −µ0 J c2 ∂t2

(1.17)

and

In the course of some of our discussions on RF cavities, we will consider the solution of Eqs. (1.15) to (1.17) within a closed region of space containing no free charges, surrounded by an equipotential surface. In this case Eq. (1.6) can be satisﬁed by choosing φ = 0 [26]. Then, from Eq. (1.15) we have ∇·A=0

(1.18)

and Eq. (1.14) reduces to [27] B = ∇ × A,

E=−

∂A ∂t

(1.19)

For this case, the electric and magnetic ﬁelds are obtained from the vector potential alone, which must satisfy Eqs. (1.17) and (1.18).

1.9 Conducting Walls

The boundary conditions at the interface between vacuum and an ideal perfect conductor can be derived by applying the integral forms of Maxwell’s equations to small pillbox-shaped volumes at the interface. One ﬁnds that only the normal electric-ﬁeld component and the tangential magnetic-ﬁeld component can be nonzero just outside the conductor surface. If nˆ is a vector normal to the interface, is the surface charge density on the conductor, and K is the surface-current density, the boundary conditions that must be satisﬁed by the ﬁelds just outside the conductor are nˆ · E = /ε0 nˆ × H = K nˆ · B = 0 nˆ × E = 0

(1.20)

No perfect conductors exist, but certain metals are very good conductors. Copper, with a room-temperature resistivity of ρ = 1/σ = 1.7 × 10−8 -m, is the most commonly used metal for accelerator applications. For a good but not perfect conductor, ﬁelds and currents are not exactly zero inside the conductor, but are conﬁned to within a small ﬁnite layer at the surface, called the skin depth. In a real conductor, the electric and magnetic ﬁelds, and the current decay exponentially with distance from the surface of the conductor, a phenomenon known as the skin effect. The skin depth is given by 2 (1.21) δ= σ µ0 ω Because of the skin effect, the ac and dc resistances are not equal. It is convenient to deﬁne the ac or RF surface resistance Rs = 1/σ δ, and using √ Eq. (1.21), we ﬁnd Rs = µ0 ω/2σ , which shows that the ac surface resistance is proportional to the square root of the frequency. If dS is the area element on the cavity walls, the average power dissipation per cycle is Rs (1.22) H2 dS P= 2 Physically, the skin effect is explained by the fact that RF electric and magnetic ﬁelds applied at the surface of a conductor induce a current, which shields the interior of the conductor from those ﬁelds. For frequencies in the 100 MHz range and for a good conductor like copper, the skin depth δ is of the order 10−6 m, and Rs is in the milliohm range. The use of superconducting materials dramatically reduces the surface resistance. For the RF surface

1 Introduction

resistance of superconducting niobium, we will use an approximate formula 2 Tc −5 f (GHz) exp −α Rs () = 9 × 10 + Rres (1.23) T(◦ K) T where α = 1.83, and Tc = 9.2 K is the critical temperature. Rres is known as the residual resistance; it is determined by imperfections in the surface, and typically is approximately 10−9 to 10−8 . The superconducting surface resistance is roughly 10−5 lower than that of copper.

1.10 Group Velocity and Energy Velocity

Linac technology requires the propagation of electromagnetic waves in transmission lines, waveguides, and cavities. There are no truly monochromatic waves in nature. A real wave exists in the form of a wave group, which consists of a superposition of waves of different frequencies and wave numbers. If the spread in the phase velocities of the individual waves is small, the envelope of the wave pattern will tend to maintain its shape as it moves with a velocity that is called the group velocity. The simplest example of a wave group, shown in Fig. 1.12, consists of two equal-amplitude waves, propagating in the +z direction, with frequencies ω1 and ω2 , and wave numbers k1 and k2 , which we can express in complex exponential form as V(z, t) = e j(ω1 t−k1 z) + e j(ω2 t−k2 z) (ω1 − ω2 )t − (k1 − k2 )z j[(ω1 +ω2 )t−(k1 +k2 )z]/2 e = 2 cos 2

(1.24)

The exponential factor describes a traveling wave with the mean frequency and mean wave number, and the ﬁrst factor represents a slowly varying

Figure 1.12 Wave composed of two components with different frequencies and wave numbers.

1.10 Group Velocity and Energy Velocity

modulation of the wave amplitude. The phase velocities of component waves are ω1 /k1 and ω2 /k2 , and the mean phase velocity is vp =

ω1 + ω2 ω¯ = k1 + k2 k¯

(1.25)

The group velocity is deﬁned as the velocity of the amplitude-modulation envelope, which is vg =

ω1 − ω2 dω → k1 − k2 dk

(1.26)

Generally, the mean phase velocity and the group velocity are not necessarily equal; they are equal when there is a linear relation between frequency and wave number, as for the ideal transmission line. The waveguide dispersion curve is a plot of ω versus k. Figure 1.13 shows an example of a dispersion curve for a uniform waveguide. The phase velocity at any point on the curve is the slope of the line from the origin to that point, and the group velocity is the slope of the dispersion curve, or tangent at that point. For the uniform guide, one ﬁnds vp vg = c2 , where vg < c and vp > c. A more general example of wave group is the wave packet, which describes a spatially localized wave group, as shown in Fig. 1.14. Again, the group velocity, rather than the phase velocity, must be used to characterize the motion of a wave packet. For example, the transient ﬁlling of a waveguide with w

w = kzc

w = –kzc wc

p Vg Vp

Figure 1.13 Example of dispersion curve for uniform waveguide, ω2 = ωc2 + (kz c)2 , showing graphically the meaning of phase and group velocity at the point p on the curve. The group velocity at point p is the tangent to the curve at that point. The phase velocity is the slope of the line from the origin to the point p.

Group velocity

Figure 1.14 Wave packet.

1 Introduction

electromagnetic energy must be described in terms of the motion of a wave packet, which will have a leading edge that moves at approximately the group velocity for practical cases where dispersion is not too large. The phase velocity does not appear in the description, because it describes the speed of individual waves that make up the wave packet, rather than the wave packet itself, which really consists of an interference pattern of these waves. The energy velocity is deﬁned as the velocity of electromagnetic energy ﬂow, which, for a traveling wave moving in the +z direction, is vE =

P+ U

(1.27)

where P+ is the wave power, the electromagnetic energy per unit time crossing a transverse plane at ﬁxed z, and U is the stored electromagnetic energy per unit length. For cases of practical interest, the energy velocity is equal to the group velocity [28]. This result is useful for evaluating the energy velocity because the group velocity at the operating point is easy to determine from the slope of the dispersion curve. 1.11 Coaxial Resonator

Some accelerating cavities, especially for relatively low frequencies below about 100 MHz, are variants of the simple coaxial resonator. Without worrying now about the speciﬁc modiﬁcations needed to produce a practical accelerating cavity, we consider the properties of a coaxial resonator. A coaxial resonator, shown in Fig. 1.15, is formed by placing conducting end walls on a section of coaxial line formed by an inner conductor of radius a and an outer conductor with radius b. When the enclosed length is an integer multiple of half wavelengths, transverse electromagnetic (TEM) resonant standing-wave modes exist, where both the electric and magnetic ﬁelds have only transverse components. Resonance occurs when the boundary condition on the end walls, Er = 0, is satisﬁed. This condition occurs when the conducting walls are separated by a distance = pλ/2, p = 1, 2, 3, . . . . To obtain the solution, ﬁrst imagine a current wave on the inner conductor traveling in the +z direction, I0 e j(ωt−kz) . From the integral form of Ampere’s law, the current produces an azimuthal magnetic ﬁeld given by Bθ = I0 e j(ωt−kz) µ0 /2π r. Given the magnetic ﬁeld, the radial electric ﬁeld can be obtained from the differential form of Faraday’s law as Er = I0 e j(ωt−kz) µ0 c/2πr. Likewise, we ﬁnd that a wave traveling in the −z direction has components I0 e j(ωt+kz) , Bθ = I0 e j(ωt+kz) µ0 /2π r, and Er = −I0 e j(ωt+kz) µ0 c/2πr. Adding these two waves produces a standing wave satisfying the boundary condition that the tangential electric ﬁeld Er vanishes on the end walls at z = 0 and . The nonzero ﬁeld components are Bθ =

µ0 I0 cos(pπz/) exp[jωt] πr

(1.28)

1.11 Coaxial Resonator

Er = −2j

µ0 I0 sin(pπz/) exp[jωt] ε0 2πr

(1.29)

pπc , p = 1, 2, 3, . . . . We note that the complex j factor in ◦ Eq. (1.29) denotes a 90 phase shift in time between the left and right sides of the equation, which can be obtained explicitly by substituting the identity j = e jπ/2 . The electromagnetic stored energy is where ω = kz c =

µ0 I02 ln(b/a) 2π

(1.30)

and the quality factor or Q, including the losses on the end walls, is pπ µ0 ln(b/a) Q0 = 1 1 b Rs ε0 + + 4 ln a b a

(1.31)

The lowest mode corresponds to p = 1, the half-wave resonator. Figure 1.15 shows the peak current and voltage distributions for p = 1, where the voltage b is V = a Er dr. The cavity in Fig. 1.15 could be modiﬁed to make it suitable for acceleration by introducing beam holes in the inner and outer conductors at z = /2 where the voltage is maximum. The beam, moving along a radial path, will see no ﬁeld when it is within the inner conductor, and can see an accelerating ﬁeld in the region on both sides between the inner and outer conductors. The injection phase could be chosen so that the beam travels across the inner conductor while the ﬁeld reverses sign, so that the beam can be accelerated on both the entrance and exit sides of the inner conductor. The conductor radii could be chosen so that the beam receives the maximum energy gain on each side. Another widely used type of resonator for accelerator applications is the coaxial line terminated at one end by a short and at the other end by a capacitance, as shown in Fig. 1.16. The capacitive termination can be

Voltage

Current

Figure 1.15 Coaxial resonator with voltage and current standing waves for p = 1.

1 Introduction z=l

Gap

Z=0 Z0

Figure 1.16 Coaxial resonator.

accomplished in practice with a coaxial line that has a gap at one end between the center conductor and the conducting end wall. An electric ﬁeld suitable for acceleration may exist between the inner conductor and either the end wall or the cylindrical wall. Thus, beam holes can be introduced near the gap, allowing either a radial or an axial trajectory. Resonant modes correspond approximately to the length λ equal to an odd multiple of a quarter wavelength, the lowest mode being a quarter-wave resonator. Design formulas for the quarter-wave resonator, including the contribution to the capacitance from fringe ﬁelds, are given by Moreno [29].

1.12 Transverse-Magnetic Mode of a Circular Cylindrical Cavity

Most cavity resonators used in electron and proton linacs are derived from the simple cylindrical or pillbox cavity. Fortunately, an analytic solution exists for the ﬁelds in a pillbox cavity. Beginning with a cylinder of radius Rc , we place conducting end plates at the axial coordinates z = 0 and . In the pillbox cavity, the holes on the end plates that must be provided for the beam are ignored. We assume a simple azimuthally symmetric trial solution of the form Ez (r, z, t) = R(r)e jωt . This solution must satisfy the wave equation with the condition that Ez vanishes at the cylindrical boundary, r = Rc , where it is tangential. The wave equation in cylindrical coordinates is 1 ∂Ez 1 ∂ 2 Ez ∂ 2 Ez ∂ 2 Ez + − 2 2 =0 + 2 2 ∂z r ∂r ∂r c ∂t

(1.32)

Substituting the trial solution into Eq. (1.32), we obtain a differential equation for the radial function R(r), which is the well-known Bessel’s equation of order zero. The magnetic ﬁeld is obtained from Amp´ere’s law from Section 1.8. The nonzero ﬁeld components of the complete solution are given by Ez = E0 J0 (kr r) cos(ωt) Bθ = −

E0 J1 (kr r) sin(ωt) c

(1.33)

The radial ﬁeld distributions are shown graphically in Figures 1.4 and 1.17. To satisfy the boundary condition, the resonant frequency of this mode must be ωc = kr c = 2.405c/Rc , which is independent of the cavity length. The mode

1.12 Transverse-Magnetic Mode of a Circular Cylindrical Cavity

is called a transverse-magnetic mode because the z component of the magnetic ﬁeld is zero, and in the conventional nomenclature the mode is called a TM010 mode for reasons that will be explained shortly. The total electromagnetic stored energy can be calculated from the peak electric stored energy, and the result is U=

πε0 R2c 2 2 E0 J1 (2.405) 2

(1.34)

The average power dissipated on the cylindrical walls and the end walls is ε0 (1.35) P = πRc Rs E02 J 2 (2.405)( + Rc ) µ0 1 The quality factor is Q=

√ 2.405 µ0 /ε0 ωc U 1 = P 2Rs 1 + Rc /

(1.36)

The electric ﬁeld is maximum at r = 0, where J0 is maximum. Two useful values of J1 are the maximum value, which is J1 (1.841) = 0.5819, and the value of J1 at the cylindrical wall, which is J1 (2.405) = 0.5191. The magnetic ﬁeld is maximum at kr R = 0.5819, where J1 (kr R) is maximum. Therefore, Bmax /Emax = 0.5819/c = 19.4 G/MV/m. Ez 1 0.8 2.405

0.6 0.4 0.2 (a) 0

0.5

1.5

wr /c

cBq

0.8

2.405

0.6 0.4 0.2 (b) 0

0.5

1.5

wr /c

Figure 1.17 Fields for a TM010 mode of a cylindrical (pillbox) cavity resonator.

1 Introduction

1.13 Cylindrical Resonator Transverse-Magnetic Modes

There are other transverse-magnetic modes with the same radial Besselfunction solution, corresponding to ﬁtting a half-integer number of guide wavelengths within the length . We label the different longitudinal modes with the index p, and adopt the conventional nomenclature TM01p , p = 1, 2, 3, . . . . The dispersion relation is the same as for a uniform waveguide, except that the longitudinal wave number is restricted to those discrete values required to satisfy the boundary conditions at the two ends. The modes lie on the curve given by ω2 /c2 = k2r + k2z , where kr = 2.405/Rc and kz = 2π/λg = πp/. Then, the dispersion relation becomes a discrete spectrum of points that are sprinkled on a hyperbolic curve, as shown in Fig. 1.18. The TM010 mode, discussed in Section 1.12, is the lowest mode with p = 0. The dispersion relation gives the resonant frequency of this mode as the cutoff frequency ωc = kr c = 2.405c/Rc . There exist additional transverse-magnetic modes of a cylindrical cavity, corresponding to different radial and azimuthal solutions. The general expressions for the ﬁeld components are as follows: Ez = E0 Jm (kmn r) cos mθ cos(pπz/) exp[jωt] pπ a E0 Jm (kmn r) cos mθ sin(pπz/) exp[jωt] Er = − xmn Eθ = −

pπ ma2 E J (k r) sin mθ sin(pπz/) exp[jωt] 2 r 0 m mn xmn

Bz = 0 ma2 E J (k r) sin mθ cos(pπz/) exp[jωt] 2 rc 2 0 m mn xmn a Bθ = −jω E0 Jm (kmn r) cos mθ cos(pπz/) exp[jωt] xmn c2 Br = −jω

(1.37)

2.5 2 1.5 1 0.5 −4

−2

0 p

Figure 1.18 Dispersion curve for the TM01p family of modes of a circular cylindrical cavity.

1.14 Cylindrical Resonator Transverse Electric Modes

The general dispersion relation is ω2 /c2 = k2mn + k2z , where kmn = xmn /Rc and kz = 2π/λguide = pπ/, p = 0, 1, 2, . . .. Some values of the zeros of the Bessel functions, xmn , are given in Table 1.3. The nomenclature of the TMmnp modes is deﬁned as follows. The subscript m (m = 0, 1, 2, . . .) is the number of full period variations in θ of the ﬁeld components. The subscript n (n = 1, 2, 3, . . .) is the number of zeros of the axial ﬁeld component in the radial direction in the range 0 < r ≤ Rc , excluding r = 0. The subscript p (p = 0, 1, 2, . . .) is the number of half period variations in z of the ﬁelds.

1.14 Cylindrical Resonator Transverse Electric Modes

Similarly, there exist additional transverse electric modes of a cylindrical cavity, corresponding to solutions with the zero axial component of the electric ﬁeld. The general ﬁeld-component expressions for the transverse electric modes are as follows: Bz = B0 Jm (kmn r) cos mθ sin(pπz/) exp[jωt] pπ a B0 Jm (kmn r) cos mθ cos(pπz/) exp[jωt] Br = xmn Bθ = −

pπ ma2 B J (k r) sin mθ cos(pπz/) exp[jωt] 2 r 0 m mn xmn

Ez = 0 ma2 B J (k r) sin mθ sin(pπz/) exp[jωt] 2 r 0 m mn xmn a Eθ = jω B0 Jm (kmn r) cos mθ sin(pπz/) exp[jωt] xmn Er = jω

(1.38)

The general dispersion relation is ω2 /c2 = k2mn + k2z , where kmn = xmn /Rc and kz = 2π/λguide = pπ/, p = 0, 1, 2, . . . . The xmn are the zeros of the derivatives of the Bessel functions and are given in Table 1.4. The nomenclature of the TEmnp modes is deﬁned as follows. The subscript m (m = 0, 1, 2, . . .) is the number of full period variations in θ of the ﬁeld components. The subscript n (n = 1, 2, 3, . . .) is the number of zeros of the axial ﬁeld component in

Table 1.3 Zeros of Jm (x) or xmn . m

xm1

xm2

xm3

0 1 2

2.405 3.832 5.136

5.520 7.016 8.417

8.654 10.173 11.620

1 Introduction Table 1.4 Zeros of Jm or xmn .

xm1

xm2

xm3

0 1 2

3.832 1.841 3.054

7.016 5.331 6.706

10.174 8.536 9.970

the radial direction in the range 0 < r ≤ Rc , excluding r = 0. The subscript p (p = 0, 1, 2, . . .) is the number of half period variations in z of the ﬁelds.

Problems

1.1. 1.2.

1.3.

1.4.

1.5.

1.6.

1.7.

What is the kinetic energy in units of both joules and electron volts for an electron accelerated through a dc potential of 1 MV? Find an expression for the fractional error when the nonrelativistic approximation for kinetic energy as a function of β is used. (a) At what values of β and γ does the error in kinetic energy equal 1%? (b) To what kinetic energy does this correspond, for electrons and for protons? If the only nonzero components of the electromagnetic ﬁeld in cylindrical coordinates are Er , Ez, and Bθ , write the nonzero components of the Lorentz force for a particle of mass m and charge q moving along the z direction with velocity v. The rate of work done by a force F acting on a particle with velocity v is F · v. Using the deﬁnition of the Lorentz force and the appropriate vector relationship, derive the expression for the rate of kinetic energy gain for a particle of charge q, and show that the magnetic force does not contribute. A cylindrical resonator has a diameter of 1.5 in. (3.81 cm) and length of 1 in. (2.54 cm). (a) Calculate the resonant frequency of the TM010 , TM110 , TE011 , TE111 , and TE211 modes, and list in order of increasing frequency. (b) For the two lowest-frequency modes, plot the dispersion relation, f (= ω/2π) versus kz (=pπ/), both on the same graph. For simplicity, label the abscissa with the longitudinal mode index p (i.e., units of π/ for p = 0 to p = 5). (Recall that the TE modes have no resonance at p = 0.) Repeat the exercise of Problem 1.5 for the same diameter resonator but with different lengths. (a) = 7.725 cm. Note the frequency of the TE112 mode compared to the TM010 mode. How did it change relative to the result of Problem 1.5? (b) = 25.4 cm. Note that all of the ﬁrst ﬁve TE11p modes now lie below the TM010 frequency. Calculate the RF surface resistance and skin depth of room-temperature copper at 400 MHz. Use a dc resistivity σ −1 = 1.7 × 10−8 -m.

Problems

1.8.

Calculate the RF surface resistance of superconducting niobium at 400 MHz. Assume a residual resistance Rresidual = 100 × 10−9 . What is the ratio of the RF surface resistance of superconducting niobium to that of room-temperature copper? (a) Assume T = 4.2 K; (b) assume T = 2.0 K. 1.9. Design a room-temperature cylindrical cavity that operates in the TM010 mode at 400 MHz with an axial electric ﬁeld E0 = 1 MV/m, and a length = λ/2, where λ is the RF wavelength in free space. (a) Calculate the length and diameter of the cavity. (b) Calculate the maximum B and H ﬁelds on the cavity wall. Where does this occur? (c) Calculate the B and H ﬁelds on the cylindrical wall. (d) Calculate the electromagnetic stored energy in the cavity. (e) Use the value of Rs from Problem 1.7 for a room-temperature copper surface to calculate the power loss P, the quality factor Q0 , and the decay time τ . (f) Repeat part (e) using Rs from Problem 1.8 for a 4.2-K niobium surface. 1.10. Design a half-wave coaxial cavity to be used as a 100-kW resonant load at 400 MHz. To absorb the RF power, use a 20-cm-diameter stainless steel pipe as the center conductor inside a 60-cm-diameter copper cylinder with copper end walls. This type of cavity is easily cooled by ﬂowing water through the center conductor. (a) Ignoring any effects of the coupling loop and probe, calculate the length of the cavity. (b) Assume that the room-temperature surface resistance of stainless steel is 6.5 times that of copper. (From Problem 1.7 the copper surface resistance at 400 MHz is 0.0052 .) Calculate the power dissipated on the center conductor, end walls, and outer wall. (c) Calculate the cavity stored energy and the unloaded Q. (d) What is the peak power density in watts per square centimeter on the inner and outer conductors? 1.11. A 25-MHz quarter-wave coaxial-cavity resonator with characteristic impedance 50 is designed as a buncher for heavy-ion beams. (a) If the impedance at the open end of the cavity (where the electric ﬁeld is maximum) could really be made inﬁnite, what would be the length of the inner conductor? (b) If we want to reduce the size by restricting the length of the inner conductor to = 1 m, what lumped capacitance would be required at the open end? 1.12. Accelerator cavities require ports through the cavity walls, not only for RF drive and RF pickup probes, but also for the beam and for vacuum pumping. Cylindrical pipes are commonly used, and such pipes will support waveguide modes. Consider a cylindrical cavity operating in a TM010 mode with beam pipes connected at the center of each end wall. Assume each beam pipe has the same inner radius b, which is much less than the cavity radius. (a) Why do you expect the ﬁelds from the cavity to attenuate in the pipes? (b) Show that if the cavity excites a TM01 waveguide mode in the beam pipes at a frequency well below cutoff, the wave power and the E and B ﬁelds attenuate with distance x along the pipe according to the approximate formula dB = −20.9x/b. Note

1 Introduction

that it is convenient to describe the attenuation of a wave with power P and ﬁeld E in decibels, or dB, where dB = 10 log10 P/P0 = 20 log10 E/E0 , where P0 and E0 are the input reference values. (c) The TE11 mode has the lowest cutoff frequency of the modes in a cylindrical pipe, and below the cutoff frequency the attenuation will be the slowest. If this mode is excited in the pipe, show that at a frequency well below cutoff, the attenuation is described by the approximate formula dB = −16.0x/b. (d) Assuming a pipe with radius b = 0.5 inches that is excited in a TE10 mode by the cavity, calculate the total attenuation in decibels if the pipe length is 2 in. Also express the answer as the fractional attenuation of the ﬁeld. (e) The attenuation in a waveguide below cutoff frequency was derived ignoring ohmic losses. What do you think has happened to the energy in the wave?

References 1 The betatron and the related induction

2 3 4

5 6 7 8 9

linac accelerate beams using nonharmonic time-dependent ﬁelds. These machines produce pulsed electric ﬁelds by induction from a magnetic pulse in accordance with Faraday’s law. The orbit of the betatron is circular, and that of the induction linac is straight. Ising G., (1924) Ark. Mat. Fys. 18, (No. 30), 1–4. Wider¨oe R., (1928) Arch. Electrotech. 21, 387. (1994) For a discussion of this and other work by Wider¨oe, see P. Waloschek, The Infancy of Particle Accelerators-Life and Work of Rolf Wider¨oe, DESY 94–039, Deutsches Elektronen-Synchrotron, Notkestrasse 85–22603, Hamburg, March. Lawrence E.O. and Edlefsen N.E., (1930) Science 72, 376. Sloan D.H. and Lawrence E.O., (1931) Phys. Rev. 38, 2021. Sloan D.H. and Coate W.M., (1934) Phys. Rev. 46, 539. Alvarez L.W., (1946) Phys. Rev. 70, 799. Alvarez L.W., Bradner H., Franck J.V., Gordon H., Gow J.D., Marshall L.C., Oppenheimer F., Panofsky W.K.H., Richmond C., and Woodward J.R., (1955) Rev. Sci. Instr. 26, 111, and (1955); Rev. Sci. Instr. 26, 210. Hansen W.W., Kyhl R.L., Neal R.B., Panofsky W.K.H., Chodorow M., and

Ginzton E.L., (1955) Rev. Sci. Instrum. 26, 134. For a more complete early history of linacs, see Blewett J.P., in Linear Accelerators, Lapostolle P.M., and Septier A.L., North Holland Publishing, 1970 pp. 1–16. For an overview of the issues and of some very interesting advanced electron-linac concepts for linear-collider applications, see International Linear Collider Technical Review Committee Report, prepared for the Interlaboratory Collaboration for R&D Towards TeV-scale Electron-Positron Linear Colliders, Committee Chairman, G. A. Loew (SLAC), December, 1995; see also Design Issues of TeV Linear Colliders, CERN/PS 96–30 (LP), 1996 European Part. Accel. Conf., ed. J. P. Delahaye(CERN). Lawrence G.P., et al., (1996) Conventional and Superconducting RF Linac Designs for the APT Project, Proc. 1996 Int. Linac Conf., Geneva, August 26–30. Outline Design of the European Spallation Neutron Source, ed. Gardner I.S.K., Lengeler H., and Rees, G.H., ESS 95–30-M, September, 1995. Holtkamp, N., Status of the SNS Linac: an Overview, Proc. of EPAC 2006, Edinburgh, Scotland, June 26–30, pp. 29–33.

References 16 Haseroth H.R., The Lead Ion Injector at

21 Compendium of Superconducting Linacs,

CERN, Proc. 1996 Int. Linac Conf., Geneva, August 26–30, 1996. Compendium of Scientiﬁc Linacs, Provisional Edition, 18th Int. Co., CERN-PS Division, Geneva, August, 26–30, 1996. The Stanford Two-Mile Accelerator, Neal R.B., Benjamin W.A., New York, (1968); Panofsky W.K.H., The Creation of SLAC Leading to 30 Years of Operation, Proc. 18th Int. Linear Accel. Conf., August, 26–30, 1996, Geneva, ed. Hill C. and Vretenar M., CERN 96–07, Geneva, November 15, 1996 p. 3. SLC Design Handbook, Erickson R., Stanford 1984; also Seeman J., (1991) Ann. Rev. Nucl. Part. Sci. 41, 3891; also Seeman J.T., Advances of Accelerator Physics and Technologies, Advanced Series on Directions in High Energy Physics, Vol. 12, Schopper H., 1993 p. 219. Erickson R.E.O., Hughes V.W., and Nagel D.E., The Meson Factories, University of California Press, Los Angeles, 1991; also Livingston M.S., LAMPF, A Nuclear Research Facility, LA-6878-MS, September 1977; also Livingston M.S., Origins and History of the Los Alamos Meson Physics Facility, LA-5000.

Provisional Edition, 18th Int. Linac Conf. Geneva, August, 26–30, 1996. (1987) Rev. Mod. Phys. 57, 1121. Jackson J.D., 1975 Classical Electrodynamics, 2nd ed., John Wiley & Sons, New York, pp. 219–223. Potentials that satisfy Eq. (11) are said to belong to the Lorentz gauge. The beauty of this result has been previously noted. See Feynman R.P., Leighton R.B., and Sands M., The Feynman Lectures on Physics, Addison Wesley, 1963 Vol. 2, p. 18–11. See Condon E.U., (1940) J. App. Phys. 11, 502; Condon E.U., (1941) J. App. Phys. 12, 129. Potentials that satisfy Eq. (14) are said to belong to the Coulomb gauge. Note that for this case, because φ = 0, Eq. (11) is also satisﬁed and the potentials belong simultaneously to the Lorentz and Coulomb gauges. Watkins D.A., 1958 Topics in Electromagnetic Theory, John Wiley & Sons, New York, pp. 12–14; Bevensee R.M., 1964 Electromagnetic Slow Wave Systems, John Wiley & Sons, New York, pp. 17–19. Moreno T., 1948 Microwave Transmission Design Data, Dover, New York, pp. 227–230.

22 23

24 25

2 RF Acceleration in Linacs

In all RF accelerators energy is delivered to the beam from an RF electric ﬁeld, which must be synchronous with the beam for a sustained energy transfer. In this chapter we introduce some important parameters that characterize the energy-gain process.

2.1 Particle Acceleration in an RF Field

First, consider an electromagnetic wave propagating along the +z direction in a waveguide, whose electric ﬁeld along the axis is given by z dz k(z) + φ (2.1) Ez (z, t) = E(z) cos ωt − 0

where the wave number k is expressed in terms of the phase velocity vp by k(z) = ω/vp (z). For efﬁcient particle acceleration, the phase velocity of the wave must be closely matched to the beam velocity. If we consider a particle of charge q moving along the +z direction, whose velocity v(z) at each instant of time equals the phase velocity of the traveling wave, the particle arrives at z position z at time t(z) = 0 dz/v(z), and the electric force on the particle is given by Fz = qE(z) cos φ. This particle is called a synchronous particle, and the phase φ is called the synchronous phase. Energy can also be transferred to a particle from an electromagnetic standing wave in an RF cavity. The electric ﬁeld along the axis is given by Ez (z, t) = E(z) cos(ωt + φ)

(2.2)

In general, the cavity may be constructed as a periodic or quasiperiodic array of cells, each with a single accelerating gap, where the parameters in the quasiperiodic structure vary monotonically along the structure to maintain synchronism with the beam particles, whose velocities are increasing. The ﬁeld acting on a particle of charge q and velocity v at position z is obtained by RF Linear Accelerators. 2nd, completely revised and enlarged edition. Thomas P. Wangler Copyright  2008 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-40680-7

2.2 Energy Gain on Axis in an RF Gap

substituting t(z) = 0 dz/v(z) into Eq. (2.2). It is simplest to consider cavities with cells that are similar to the pillbox cavity that we have studied in Chapter 1. Each cell can be excited in a TM010 -like standing-wave mode. For this mode, a particle is acted upon by the nonzero Ez and Bθ ﬁelds, and when holes are included in the end walls for the beam, a nonzero Er component is also present. The general equations of motion are dpz = q Ez (r, z) cos(ωt + φ) + vr Bθ (r, z) sin(ωt + φ) dt

(2.3)

dpr = q Er (r, z) cos(ωt + φ) − vz Bθ (r, z) sin(ωt + φ) + Fex dt

(2.4)

and

where Fex includes any external radial ﬁeld that might be applied for focusing. Particle trajectories can be calculated by numerical integration of Eqs. (2.3) and (2.4), using measured ﬁelds or ﬁelds calculated from an electromagnetic ﬁeld–solver code.

2.2 Energy Gain on Axis in an RF Gap

Consider the accelerating gap shown in Fig. 2.1, and express the electric ﬁeld on the axis experienced by a particle with a velocity v, as (2.5) Ez (r = 0, z, t) = E(0, z) cos[ωt(z) + φ] z where t(z) = 0 dz v(z) is the time the particle is at position z. We choose time t = 0 when the particle is at the origin, which can be arbitrarily chosen within the gap. At t = 0, the phase of the ﬁeld relative to the crest is φ. Suppose that the ﬁeld is conﬁned within an axial distance L containing the gap.

E(r,z) r=a rc

r=0

z g −L /2

Figure 2.1

L /2

Gap geometry and ﬁeld distribution.

−L /2

L /2

2 RF Acceleration in Linacs

The energy gain of an arbitrary particle with charge q traveling through the gap is W = q

L/2

−L/2

E(0, z) cos(ωt(z) + φ) dz

(2.6)

The use of a common trigonometric identity allows us to write the energy gain as W = q

L/2

−L/2

E(0, z) [cos ωt cos φ − sin ωt sin φ] dz

(2.7)

Next, we write Eq. (2.7) in the form W = qV0 T cos φ

(2.8)

where we have introduced an axial RF voltage, deﬁned by V0 ≡

L/2

(2.9)

E(0, z) dz

−L/2

which is assumed to be nonzero, and the transit-time factor, deﬁned by [1] T≡

L/2 −L/2

E(0, z) cos ωt(z) dz

− tan φ

L/2

−L/2

E(0, z) dz

L/2

−L/2

E(0, z) sin ωt(z) dz

(2.10)

L/2 −L/2

E(0, z) dz

Equation (2.8) is an important result that we will use frequently. The phase φ = 0 if the particle arrives at the origin when the ﬁeld is at a crest. It is negative if the particle arrives at the origin earlier than the crest, and positive if it arrives later. Maximum energy gain occurs when φ = 0, which is often the choice for relativistic electrons. But we will see later that for nonrelativistic particles, where longitudinal focusing is important, it is necessary to operate with negative phases for the particles in the bunch. It is useful to deﬁne an average axial electric-ﬁeld amplitude E0 = V0 /L, where V0 is the axial RF voltage given by Eq. (2.9). V0 is the voltage gain that would be experienced by a particle passing through a constant dc ﬁeld equal to the ﬁeld in the gap at time t = 0. The value of E0 , which is an average ﬁeld over the length L, does depend on the choice of L. Therefore, when a value of E0 is quoted for a cavity, it is important to specify the corresponding length L. For a multicell cavity, the natural choice for L is the geometric cell length. We will call the quantity E0 T the accelerating gradient. In terms of E0 , the energy gain may be expressed as W = qE0 T cos φL

(2.11)

2.2 Energy Gain on Axis in an RF Gap

This result is sometimes called the Panofsky equation [2]. It is deceptively simple in appearance, but much of the physics is still contained within the transit-time factor. Regardless of the phase φ, the energy gain of a particle in a harmonically time-varying ﬁeld is always less than the energy gain in a constant dc ﬁeld equal to that seen by the particle at the center of the gap. This is known as the transit-time effect, and the transit-time factor T is the ratio of the energy gained in the time-varying RF ﬁeld to that in a dc ﬁeld of voltage V0 cos(φ). Thus, T is a measure of the reduction in the energy gain caused by the sinusoidal time variation of the ﬁeld in the gap. The phase and the transit-time factor depend on the choice of the origin. It is convenient to simplify the transit-time factor, and remove its dependence on the phase, by choosing the origin at the electrical center. E(z) is usually at least approximately an even function about a geometric center of the gap. We will choose the origin at the electrical center of the gap, deﬁned to give L/2 E(0, z) sin ωt(z) dz (2.12) 0= −L/2

When E(z) is an even function about the geometric center of the gap, the electrical center and the geometric center coincide. When Eq. (2.12) applies, the transit-time factor simpliﬁes to L/2 E(0, z) cos ωt(z) dz −L/2 T= (2.13) L/2 E(0, z) dz −L/2

Then the transit-time-factor expression in Eq. (2.13) is the average of the cosine factor weighted by the ﬁeld. The transit-time factor increases when the ﬁeld is more concentrated longitudinally near the origin, where the cosine factor is largest. Thus, the larger the gap between drift tubes and the more the ﬁeld can penetrate into the drift tubes, the smaller the transit-time factor. The practice of constructing drift tubes with nose cones that extend further into the gap forces a concentration of the ﬁeld near the center of the gap, and raises the transit-time factor. In most practical cases the change of particle velocity in the gap is small compared with the initial velocity. If we ignore the velocity change, we have ωt ≈ ωz/v = 2πz/βλ, where β = v/c and βλ is the distance the particle travels in an RF period. The transit-time factor simpliﬁes to the form most often seen in the literature, L/2 E(0, z) cos(2πz/βλ) dz −L/2 T= (2.14) L/2 E(0, z) dz −L/2

2 RF Acceleration in Linacs

E(z) 1

0.8

0.6 0.4 −g /2

0.2 0.5 −0.2

Figure 2.2

1 g /bl

1.5

g /2

Transit-time factor for square-wave electric-ﬁeld distribution.

We note that the transit-time factor is ﬁnite only if the denominator of Eq. (2.14) is nonzero. For most practical cases this is true, but exceptional cases arise. A TM011 cavity mode, which has a node in Ez at the origin, is an example for which the integral of the axial ﬁeld is zero, and a transit-time factor for the whole cell would be inﬁnite. When this happens, the energy gain can still be calculated directly from Eq. (2.6). We will usually consider TM010 -like modes where this is not a problem. A ﬁrst approximation for T assumes that the electric ﬁeld has a square proﬁle, as shown in Fig. 2.2. Thus, E(0, z) = Eg is constant over a gap of length g, and falls immediately to zero outside the gap. This would be the exact result for a simple TM010 pillbox cavity of length g, and would be an approximate result for a simple gap with beam holes at the ends in the limit that the beam-aperture radius is very small. Then we ﬁnd E0 = Eg if L = g, and Eq. (2.14) becomes T=

sin πg/βλ πg/βλ

(2.15)

The result for T versus g/βλ is plotted in Fig. 2.2. To achieve maximum energy gain for a given V0 , it is clear that we want T = 1, which corresponds to g = 0. However, other considerations must be taken into account to determine the optimum gap geometry, such as the risk of RF electric breakdown, and RF power efﬁciency. 2.3 Longitudinal Electric Field as a Fourier Integral

A description of the ﬁelds in an accelerating gap in terms of a Fourier integral allows us to deduce some useful relationships for the transit-time factor. For a

2.3 Longitudinal Electric Field as a Fourier Integral

single gap, where the ﬁelds vanish at some ﬁnite distance outside the gap, the value of L in Eq. (2.14) is arbitrary as long as it is large enough to include the nonzero ﬁeld region. Then, we are free to let L → ∞, without affecting any of the physics. We may write ∞ V0 T(k) = E(0, z) cos(kz) dz (2.16) −∞

where k = 2π/βλ, depends on the particle velocity. The integral has the form of the Fourier cosine integral, and its Fourier transform is V0 ∞ T(k) cos(kz) dk (2.17) E(0, z) = 2π −∞ We can extend these results to include the transit-time factor and energy gain for off-axis particles. For the ﬁeld at a displacement r from the axis, assuming azimuthal symmetry, we write Ez (r, z, t) = E(r, z) cos(ωt + φ)

(2.18)

and V0 T(r, k) =

L/2 −L/2

E(r, z) cos(kz) dz

(2.19)

where in a single gap we can let L → ∞. The Fourier transform of Eq. (2.19) is V0 ∞ E(r, z) = T(r, k) cos(kz) dk (2.20) 2π −∞ Now, we substitute this solution for Ez (r, z, t) into the wave equation in cylindrical coordinates: ∂ 2 Ez ∂ 2 Ez 1 ∂Ez 1 ∂ 2 Ez + + − 2 2 =0 2 2 ∂z r ∂r ∂r c ∂t

(2.21)

Integrating over k and using the orthogonality of the cosine functions, we obtain 1 ∂T ∂ 2T + +T =0 2 ∂x x ∂x where x2 ≡ K 2 r 2 and 2π 2 2 2 −k K = λ

(2.22)

(2.23)

Equation (2.22) is the Bessel equation. Thus, the sign of K 2 can be either positive or negative, depending on whether k is larger or smaller than 2π/λ.

2 RF Acceleration in Linacs

Substituting the physical result k = 2π/βλ, we have 2

2 2

2π 1 2π β −1 2π 2 2 K = 1− 2 = =− λ β λ β2 γβλ

(2.24)

and we ﬁnd K 2 < 1. But, for the nonphysical velocities and k values, which also are included mathematically in Eqs. (2.17) and (2.20), we can have K 2 ≥ 1. To simplify the notation we will we will always write K as a positive number given by K = |K 2 |. The solution is T(k)J0 (Kr), k ≤ 2π/λ (2.25) T(r, k) = T(k)I0 (Kr), k ≥ 2π/λ where J0 is the Bessel function and I0 is the modiﬁed Bessel function of order zero. We use the notation T(k) = T(0, k). For most beams in linacs, the paraxial approximation applies, which means that the particle radius or displacement from axis is nearly constant through the gap. We can calculate the energy gain of a particle at constant radius r. It is convenient to express the Fourier integrals in exponential form. Thus V0 ∞ E(r, z) = dk T(r, k) e jkz (2.26) 2π −∞ ∞ V0 T(r, k) = dz E(r, z) e −jkz (2.27) −∞

The cosine forms are obtained by taking the real part. The energy gain for a particle with radius r and velocity β0 is obtained by calculating ∞ dz Ez (r, z, t = z/β0 c) (2.28) W = q −∞

We ﬁnd W =

qV0 2π

∞

−∞

∞

dk T(r, k)

dz −∞

e j[(k+k0 )z+φ] + e j[(k−k0 )z−φ] 2

where k0 = 2π/β0 λ. We apply a familiar result for the δ function, ∞ 1 δ(k) = dx e jkx 2π −∞

(2.29)

(2.30)

and W =

qV0 T(r, −k0 ) e jφ + T(r, k0 ) e−jφ 2

(2.31)

We assume T(r, k) is an even function of k, so T(r, −k) = T(r, k). Also, for k corresponding to particle velocities β < 1, T(r, k) = T(k)I0 (Kr). Substituting these results into Eq. (2.31), and taking the real part, we obtain W = qV0 T(k0 )I0 (K0 r) cos φ

(2.32)

2.4 Transit-Time-Factor Models

where K0 = 2π/γ0 β0 λ. This equation is like the axial energy gain result of Eq. (2.8), except now we have included the radial dependence of the energy gain through the modiﬁed Bessel function. For additional insight to the energy-gain result, using a familiar trigonometric identity, we can write V0 ∞ Ez (r, z, t) = dk T(r, k) cos(kz) cos(ωt + φ) 2π −∞ V0 ∞ = dk T(r, k){cos(ωt − kz + φ) + cos(ωt + kz + φ)} 4π −∞ (2.33) This equation describes the standing-wave ﬁeld pattern as resulting from an inﬁnite sum of forward and backward traveling waves with phase velocities given by β = ±ω/kc. If we imagine a particle traveling from z = −∞ to +∞, only the single wave with the same phase velocity as the particle will be synchronous with the particle, and will deliver net energy to it. The effects of all other waves average to zero, because they do not maintain a constant phase relationship with the particle. From this we conclude that the radial dependence of the energy gain is the same as that of the synchronous wave. The z component of the electric ﬁeld can also be written as follows:

V0 ∞ I (Kr) Ez (r, z, t) = dk T(k) 0 cos(kz) cos(ωt + φ) (2.34) J0 (Kr) 2π −∞ where we use I0 (Kr) when |k| ≥ 2π/λ, and J0 (Kr) when |k| ≤ 2π/λ. Then, the radial electric-ﬁeld component can be derived from Gauss’s law as

V0 ∞ k I (Kr) Er (r, z, t) = dk T(k) 1 sin(kz) cos(ωt + φ) (2.35) J1 (Kr) 2π −∞ K and using Amp´ere’s law, we obtain,

−V0 ∞ T(k) I1 (Kr) dk cos(kz) sin(ωt + φ) Bθ (r, z, t) = J1 (Kr) cλ −∞ K

(2.36)

Equations (2.34) to (2.36) comprise a complete solution to Maxwell’s equations together with the remaining components Bz , Br , and Eθ , which are zero, as can be shown by direct substitution.

2.4 Transit-Time-Factor Models

We can improve on the approximation for T in Eq. (2.15) by accounting for penetration of the ﬁeld into the axial bore holes of drift tubes. Field penetration into the drift tubes reduces T, because it reduces the concentration of the ﬁeld near the center of the gap. We assume that the electric ﬁeld at the drift-tube

2 RF Acceleration in Linacs

bore radius (r = a) is constant within the gap and zero outside the gap within the metallic walls, as shown in Fig. 2.1. This is a more realistic approximation, because it constrains the ﬁeld to vanish only at the bore radius inside the conductor, rather than on the axis at the end of the gap. Thus, we require that Eg , 0 ≤ |z| ≤ g/2 E(r = a, z) = (2.37) 0, g/2 < |z| From Eq. (2.19) this leads to ∞ 2Eg E(a, z) cos(kz) dz = sin(kg/2) V0 T(a, k) = k −∞

(2.38)

From Eq. (2.25), T(a, k) = T(k)I0 (Ka), where K = 2π/γβλ, and we obtain Eg g sin(kg/2) (2.39) I0 (Ka) kg/2 ∞ Also,V0 T(0, 0) = −∞ dz E(0, z) = V0 , or T(0) = T(0, 0) = 1, and at k = 0 and r = a, Eq. (2.25) gives T(a, 0) = J0 (K0 a), where K0 = 2π/λ. Then ∞ dz E(a, z) = Eg g (2.40) V0 T(a, 0) = V0 T(k) =

−∞

or V0 =

Eg g J0 (2πa/λ)

(2.41)

For β ≤ 1 we obtain from Eqs. (2.25) and (2.39), T(r, k) = J0 (2πa/λ)

I0 (Kr) sin(kg/2) I0 (Ka) kg/2

(2.42)

and the axial transit-time factor is T(k) =

T(r, k) J0 (2πa/λ) sin(πg/βλ) = I0 (Kr) I0 (Ka) πg/βλ

(2.43)

It is usually a good approximation that J0 (2πa/λ) ≈ 1. In Eq. (2.43) the axial transit-time factor has been expressed as a product of an aperture-dependent factor and a gap-dependent factor, where the aperture factor takes into account the penetration of the ﬁeld into the borehole. The result of Eq. (2.43) applies for a single gap. For a multigap periodic structure, the transit-time factor depends on both the synchronous velocity βs and the velocity β of the particle. One ﬁnds that in the argument of the Bessel functions, the quantity K becomes Ks , where Ks = 2π/γs βs λ. Instead of Eq. (2.43), the result becomes [3] T(k) =

J0 (2πa/λ) sin(πg/βλ) I0 (Ks a) πg/βλ

(2.44)

2.4 Transit-Time-Factor Models

An empirical correction to the gap factor can be made [4] to account for the ﬁnite chamber radius ρc that is put on a drift tube to remove the sharp edge, and avoid electric breakdown. The effective gap is obtained by replacing g in the formula for the transit-time factor by gc = g + 0.85ρc

(2.45)

For π-mode structures an additional effect occurs. The axial electric ﬁeld changes sign from cell to cell and must cross zero at the cell boundaries, as shown in Fig. 2.3. For a large gap and a large borehole, the axial ﬁeld distribution will be more strongly inﬂuenced by the boundary condition that the ﬁeld vanishes at the beginning and end of the cell. An example is the typical multicell superconducting elliptical cavity structure with a large aperture and without any nose cone that extends into the gap. To construct a simple model for this situation, we assume that in a large-bore, π-mode structure, the axial ﬁeld in the gap of a cell of length L has a cosine dependence, given by E(r = 0, z) = Eg cos(ks z)

(2.46)

where ks = π/L. The ﬁeld in Eq. (2.46) vanishes when z = ±L/2, satisfying the π-mode constraint. Generally, the cell length will have been designed to give synchronous acceleration for a particle with velocity βs , by choosing L = βs λ/2. Substituting Eq. (2.46) into Eq. (2.19) evaluated on axis, we ﬁnd V0 T(0, k) =

L/2 −L/2

= Eg

dz E(0, k) cos(kz)

L/2

dz cos(kz) cos(kz) −L/2

(2.47)

Figure 2.3 Axial electric-ﬁeld distribution showing the effect of the π -mode boundary conditions, which causes the ﬁeld to cross the axis at the boundaries of each cell.

2 RF Acceleration in Linacs

The axial transit-time factor for a particle with β = βs , or k = ks is Eg L/2 πEg T(ks ) = T(0, ks ) = dz cos2 (ks z) = V0 −L/2 2ks V0 Also we have L/2 Eg L/2 2Eg V0 = dz E(0, z) = dz cos(ks z) = k ks 0 −L/2 −L/2

(2.48)

(2.49)

Substituting Eq. (2.49) into Eq. (2.48), we obtain the axial transit-time factor for the synchronous particle T(ks ) =

π ≈ 0.785 4

(2.50)

Unlike the result of Eq. (2.44), the axial transit-time factor is independent of both the gap, and the radial aperture of the borehole. For completeness the radial dependence of the transit-time factor from the π-mode boundary condition is given from Eq. (2.25), by T(r, ks ) = T(ks )I0 (Ks r) =

π I0 (Ks r) 4

(2.51)

where Ks = 2π/γs βs λ. For the case of a particle with velocity β which is not equal to βs , a general result analogous to Eq. (2.44) is

π sin[(k/ks − 1)π/2] sin[(k/ks + 1)π/2] + (2.52) T(r, k) = I0 (Ks r) 4 (k/ks − 1)π/2 (k/ks + 1)π

2.5 Power and Acceleration Efﬁciency Figures of Merit

There are several ﬁgures of merit that are commonly used to characterize accelerating cavities, and we will deﬁne them in this section. Some of these depend on the power, which is dissipated because of electrical resistance in the walls of the cavities. The well-known quality factor of a resonator is deﬁned in terms of the average power loss P as Q = ωU/P

(2.53)

It is also convenient to deﬁne the shunt impedance, a ﬁgure of merit that is independent of the excitation level of the cavity and measures the effectiveness of producing an axial voltage V0 for a given power dissipated. The shunt impedance rs of a cavity is usually expressed in megohms, and is deﬁned by [5] rs =

V02 P

(2.54)

2.5 Power and Acceleration Efﬁciency Figures of Merit

In an accelerating cavity we are really more interested in maximizing the particle energy gain per unit power dissipation. The peak energy gain of a particle occurs when φ = 0, and is Wφ=0 = qV0 T. Consequently, we deﬁne an effective shunt impedance of a cavity as

Wφ=0 2 1 [V0 T]2 (2.55) r= = = rs T 2 q P P This parameter in megohms measures the effectiveness per unit power loss for delivering energy to a particle. For a given ﬁeld both V0 = E0 L and P increase linearly with cavity length, as do both r and rs . For long cavities we often prefer to use a ﬁgure of merit that is independent of both the ﬁeld level and the cavity length. Thus, it is also convenient to introduce shunt impedances per unit length. The shunt impedance per unit length, Z, is simply Z≡

E2 rs = 0 L P/L

(2.56)

Similarly, the effective shunt impedance per unit length is ZT 2 =

r (E0 T)2 = L P/L

(2.57)

The units of shunt impedance per unit length and effective shunt impedance per unit length are usually megohms per meter. Especially for normalconducting cavities, one of the main objectives in cavity design is to choose the geometry to maximize effective shunt impedance per unit length. This is equivalent to maximizing the energy gain in a given length for a given power loss. Another useful parameter is the ratio of effective shunt impedance to Q, often called r over Q, (V0 T)2 r = Q ωU

(2.58)

We see that r/Q measures the efﬁciency of acceleration per unit stored energy at a given frequency. One may wish instead to quote the ZT 2 /Q. Either of these ratios is useful, because they are a function only of the cavity geometry, and are independent of the surface properties that determine the power losses. Some fraction of the RF power that goes into the cavity is also transferred to the beam as a result of the interaction of the beam with the cavity ﬁelds. The power PB delivered to the beam is easily calculated from PB = IW/q

(2.59)

where I is the beam current and W is the energy gain. The sum of the dissipated power plus that delivered to the beam is PT = P + PB , and the efﬁciency of the structure is measured by the beam-loading ratio εs = PB /PT . The overall system efﬁciency must also include the power losses in the input

2 RF Acceleration in Linacs

waveguide or transmission line, reﬂected power from the input coupler, which ends up in a load, and power dissipation in the RF generator. The latter is usually the largest contributor. The RF generator efﬁciency, which is the ratio of the output RF power to the total ac input power, typically lies in the range from about 40 to 60%. The overall efﬁciency increases with increasing beam current.

2.6 Cavity Design Issues

Cavity design generally requires electromagnetic ﬁeld–solver codes that numerically solve Maxwell’s equations for the speciﬁed boundary conditions. The optimization procedures depend on the constraints of the problem. Suppose the objective is to maximize the effective shunt impedance of a single cavity. If we begin with a simple pillbox cavity with beam holes on the end plates, adding nose cones to create a region of more concentrated axial electric ﬁeld, as shown in Fig. 2.4, reduces the gap and raises the transit-time factor. The nose cone is part of the drift tube, and its shape may be constrained by the requirement that it must hold a focusing quadrupole magnet. If the gap between drift tubes is too small, then the voltage gain for a given peak surface electric ﬁeld becomes small. Also a capacitance between the opposite nose cones increases with decreasing gap size, increasing the ratio of wall current to axial voltage. Some designers prefer to optimize in steps. First, the gap size and the shape of the nose cone can be chosen to maximize the r/Q parameter to increase the energy gain per unit stored energy. Then, the geometry of the outer walls can be adjusted to maximize Q, thus minimizing the power dissipation per unit stored energy. This latter procedure usually leads to a spherical shape for the outer wall, which corresponds to the smallest surface area for a given volume.

Figure 2.4 Cross section of cavity with nose cones and spherical outer wall.

2.6 Cavity Design Issues

Figure 2.5 Cross section of elliptical cavity used for a superconducting proton linac. The cross section for each cell consists of an outer circular arc, an ellipse at the iris, and a connecting straight line.

In superconducting cavities, the RF power efﬁciency may not be as important as other requirements such as minimizing the peak surface ﬁelds, which limit the performance. The spherical or elliptical cavity shape without nose cones, shown in Fig. 2.5, is the universally accepted choice for a superconducting cavity shape for high-velocity particles, because of its low ratio of peak surface ﬁeld to E0 , its lack of strong multipacting levels, and practical advantages for drainage of ﬂuids during chemical processing. The peak surface electric ﬁeld and magnetic ﬁeld are important constraints in cavity design. In normal-conducting cavities, too large a peak surface electric ﬁeld can result in electric breakdown. For a cavity with nose cones or drift tubes, a typical ratio of peak surface electric ﬁeld Es to average axial ﬁeld E0 , is Es /E0 = 6, whereas the ratio is unity for a pillbox cavity. The Kilpatrick criterion, discussed later, is often used as the basis for the peak surface electric-ﬁeld limit in normal-conducting cavities. In superconducting cavities high electric ﬁelds cause ﬁeld emission, which produces electrons in the cavity volume that absorb RF energy and create additional power loss. The surface magnetic ﬁelds correspond to surface currents that produce resistive heating. For normal-conducting cavities, cooling requirements restrict the peak magnetic ﬁeld, and average power densities of about 20 W/cm2 may begin to produce cooling difﬁculties in drift-tube linacs. Superconducting cavities can quench and make a transition to the normal-conducting state, when a critical magnetic ﬁeld is exceeded. This ﬁeld depends on temperature, and is near B = 0.2 T in the range of operating temperatures from about 2 to 4.2 K. The presence of normal-conducting impurities lowers this limit. Few designers of superconducting cavities would exceed about B = 0.05 to 0.1 T. The ratio of peak surface magnetic ﬁeld to average axial ﬁeld E0 is 19.4 G/MV/m for a pillbox cavity. This value can increase by a factor of 2 to 3, especially in drift tubes.

2 RF Acceleration in Linacs

The aperture radius is usually chosen to satisfy the beam-dynamics requirements for high transmission. Large aperture results in a reduced transit-time factor. As the aperture radius increases, the axial electric ﬁeld decreases for a given peak surface ﬁeld near the aperture. Introducing nose cones or drift tubes increases the capacitance, and introducing stems that carry a net RF current to the drift tubes increases the local magnetic ﬂux. Loading a cavity with lumped elements produces more localized ﬁelds. This approach reduces the ﬁelds at the outer cavity walls, and the diameter of those walls can be reduced without producing much change in the frequency. The size reduction can make the structures easier to build and handle. This approach is especially useful for very low-frequency structures and is widely used for heavy-ion accelerating structures.

2.7 Frequency Scaling of Cavity Parameters

One of the most important parameters to choose when designing a linac is the operating frequency. To make this choice, it is important to know how the cavity parameters vary with frequency. Consider a cavity with ﬁxed accelerating ﬁeld E0 and ﬁxed total energy gain W, so that the total length is ﬁxed, independent of frequency. Suppose we scale all other cavity dimensions with wavelength or as f −1 . How do the other parameters behave? The transit-time factor and the ﬁelds are independent of frequency. At a ﬁxed total length, the surface area is inversely proportional to the frequency, and the total cavity volume and stored energy are inversely proportional to the frequency squared. Surface-resistance and power loss scaling depend on whether the linac is normal conducting or superconducting. In what follows we will ignore the residual-resistance term in the superconducting RF resistance, which is a good approximation at high frequency. One ﬁnds that the RF surface resistance scales as 1/2

f normal conducting Rs ∝ (2.60) f2 super conducting The RF power losses are not zero in the superconducting case. They are caused by normal-conducting electrons that are present at any ﬁnite temperature. Although these losses are small in terms of RF power, they affect the requirements for cryogenic refrigeration and must be taken into account. The RF power dissipation scales as

−1/2 Rs B 2 normal conducting f (2.61) dA ∝ P= f super conducting 2 µ0 Thus, higher frequency gives reduced power loss for normal-conducting structures, but increased power losses for the superconducting case. The value

2.8 Linac Economics

of Q scales as Q=

ωU ∝ P

f −1/2 f −2

normal conducting super conducting

The effective shunt impedance per unit length scales as

+1/2 (E0 T)2 L normal conducting f ZT 2 = ∝ f −1 super conducting P

(2.62)

(2.63)

The shunt impedance increases at higher frequencies for normal-conducting structures, and at lower frequencies for superconducting structures. The ratio ZT 2 /Q scales as

1 ZT 2 normal conducting f (2.64) ∝ f 1 super conducting Q This is the same for normal and superconducting, as it should be since this is a ﬁgure of merit that is independent of the surface properties.

2.8 Linac Economics

Accelerator designers must understand the impact of their design choices on the total costs. Included may be the capital costs of construction, and the costs of operation over some period of time, perhaps several decades. To illustrate the issues, we present a simpliﬁed linac cost model, which includes two main cost factors. First, we introduce a capital cost per meter, CL , which includes the costs of fabrication of the accelerating structure, and any other costs that scale in proportion to the length of the linac. We introduce a capital cost per watt of RF power, CP . We deﬁne the total cost C, the total length L, and the total RF power P, which is expressed as the sum of the resistive losses in the accelerating structure PS , and the power PB delivered to the beam. We can express the total cost for the design as C = CL L + CP (PS + PB )

(2.65)

Suppose we want to choose the quantities L and PS to minimize C. Before we can do so, we must include some constraints. First, we must ensure that the accelerator design produces the correct ﬁnal energy. Assuming that the energy gain of the linac must be W, we write W = qEL cos φ

(2.66)

where we have introduced E = E0 T, the charge q, and the RF phase φ. We introduce two other relationships. From the deﬁnition of effective shunt impedance, we write PS = E 2 L/ZT 2 , where ZT 2 is the effective shunt

2 RF Acceleration in Linacs

impedance, and the beam power PB = IW/q, where I is the beam current. Expressing L and PS as functions of E, we substitute L = W/qE cos φ, and PS = EW/qZT 2 cos φ into Eq. (2.65) to obtain

W CP E CL C(E) = (2.67) + + CP I q E cos φ ZT 2 cos φ An example is shown in Fig. 2.6 for the three separate cost terms and the total, as a function of the accelerating ﬁeld E. The ﬁrst term is the structure length cost, the second is the RF structure power cost, and the third term is the RF beam power cost. It can be seen that there is an optimum value of E that minimizes the cost. We can calculate the minimum cost by differentiating C with respect to E, which leads to CP E CL = E ZT 2

(2.68)

The left side of Eq. (2.68) is proportional to the structure length cost, and the right side to the RF structure power cost. The result is that the minimum total cost corresponds to equal costs for these two contributions. The RF beam power gives a ﬁxed cost that does not affect the minimum. Solving Eq. (2.68) for the optimum ﬁeld, we ﬁnd CL ZT 2 E= (2.69) CP From Eqs. (2.66) and (2.69), we ﬁnd that the corresponding optimum length L is CP W (2.70) L= q cos φ CL ZT 2

Total cost Cost

Structure power cost Beam power cost Structure length cost

E (MV/m)

Figure 2.6 Costs versus accelerating ﬁeld. The structure power cost, the beam power cost, structure length cost, and the total cost are shown.

Problems

The structure power for the optimum solution is PS =

CL L CP

(2.71)

and the minimum cost can be written as CMIN = 2CL L + CP PB = CP (2PS + PB )

(2.72)

These results show that if E is too large, L is small but this pushes the RF power costs too high. Likewise, if E is too small, L is large and the length cost will be too high. Cost parameters vary depending on many factors, but for rough cost numbers, one might choose Cp = $2 per watt for the installed cost of an RF system, and CL ≈ $250,000 per meter to include the installed cost of accelerating structure, support stand, vacuum system, diagnostics, and tunnel. Other constraints, especially technological ones, may enter that prevent achieving the optimum choices. For example the optimum ﬁeld E may be too large for reliable operation. An accurate cost estimate requires a detailed accounting of the cost of all the components with careful attention to many details. Furthermore, complete estimates for large projects include additional costs such as engineering and design, project management, and allowance for contingency. These costs are far from negligible, and may increase the total cost by a factor of more than two. In recent years, operational costs are also being included, summed over a few decades of operation.

Problems

2.1.

2.2.

Consider a cylindrical pillbox cavity of length L operated in a TM014 mode. Complete the following steps to determine whether one can use this mode to achieve synchronous acceleration at each antinode. (a) Sketch the wave pattern for Ez versus z. Express the spacing between adjacent antinodes as a function of the cavity length. How many antinodes are there? (b) Express a synchronism requirement by equating the spacing between adjacent antinodes to the distance the particle travels in one half an RF period. This should relate the cavity length L to βλ. (c) But the cavity length and λ are already related by the dispersion relation. Write the dispersion relation and calculate the frequency f for the TM014 mode. Substitute the synchronism condition from part (b). Solve for β, and determine the range of allowed β values. Can one use this mode to achieve synchronous acceleration at each antinode? Consider a 2.5-cm-long TM010 pillbox cavity that is resonant at 400 MHz and is excited with an axial electric ﬁeld of E0 = 2 MV/m. Suppose a proton with an initial kinetic energy of 5 MeV is accelerated along the axis of the cavity and arrives at the center when the phase of the ﬁeld is 30◦ before the crest. (a) Ignoring the effects of the aperture, and

2 RF Acceleration in Linacs

2.3.

2.4.

2.5.

2.6.

2.7.

2.8.

2.9.

assuming the velocity β remains constant at its initial value, calculate the transit-time factor. (b) Calculate the ﬁnal energy of the particle as it leaves the cavity. Also calculate the fractional change in the velocity. Consider a room-temperature copper TM010 pillbox cavity that is resonant at 400 MHz with an axial electric ﬁeld E0 = 1 MV/m and a length λ/2, where λ is the RF wavelength. Assume the cavity accelerates relativistic electrons with velocity β = 1. (a) Use the power loss formula for the pillbox cavity from Section 1.12 to calculate the RF power dissipated in the walls. (b) Ignoring the effects of the apertures calculate the transit-time factor. (c) Calculate the shunt impedance, the effective shunt impedance, and the effective shunt impedance per unit length. (d) Calculate the energy gain of an electron with synchronous phase φ = 0. Consider a room-temperature 10-m-long electron linac with synchronous phase φ = 0, that accelerates the beam from nearly zero kinetic energy to 100 MeV. Suppose that the effective shunt impedance per unit length is ZT 2 = 50 M /m. (a) Assuming the power delivered to the beam is negligible, how much RF power is required? (b) Repeat part (a) assuming the accelerator length is 100 m. Use the deﬁnition of effective shunt impedance per unit length and the expression for energy gain in a cavity to show that the resistive power loss for ﬁxed energy gain and ﬁxed ZT 2 is proportional to E0 T. For a 100-MeV room-temperature electron linac with synchronous phase φ = 0, and ZT 2 = 50 M /m, plot the curve of power (MW) required versus E0 T (MV/m). Suppose that you have to design a TM010 pillbox cavity with a constant axial electric ﬁeld, but you are free to choose the length. If the cavity is too short the voltage gain across it is small; if it is too long the transit-time factor is small. (a) Ignoring the effect of the beam aperture, ﬁnd the ratio of length to βλ that maximizes the energy gain for the cavity. (b) At this length, what is the transit-time factor? (c) Show that a zero-length cavity is required to maximize the energy gain per unit length. (In practice you would be limited by peak surface electric ﬁelds.) Use the analytic expressions for the pillbox cavity given in Section 1.12 to derive expressions for r, r/Q, and ZT 2 for a cylindrical pillbox cavity in a TM010 mode. Express the results as the product of a constant and algebraic combinations such as length, radius, and transit-time factor. Calculate numerical values for r, r/Q, and ZT 2 for a cylindrical pillbox cavity in a TM010 mode, assuming a room-temperature cavity for acceleration of relativistic electrons. Assume a cavity resonant frequency of 400 MHz and a cavity length of λ/2. Use the formula for transit-time factor and ignore the aperture factor. Suppose that we want to design a continuous-wave (CW) roomtemperature drift-tube linac to accelerate a 100-mA proton beam from 2.5 to 20 MeV. Assume we can purchase 350-MHz klystron tubes of 1-MW capacity each for the RF power. Suppose that we run the SUPERFISH

Problems

cavity code and obtain the following results for all β values: T = 0.8, ZT 2 = 50 M /m, and Es /E0 = 6. Let us choose to restrict the peak surface electric ﬁeld at a bravery factor b = Es /EK = 1. For adequate longitudinal acceptance we choose the synchronous phase φ = −30◦ . (a) Calculate the average axial ﬁeld E0 . (b) Calculate the length of the linac assuming it consists of a single tank. (c) Calculate the structure power (power dissipated in the cavity walls), the beam power, and the total RF power required (assuming the klystron generator is matched to the cavity structure with the beam present). (d) What is the structure efﬁciency (ratio of beam power to total RF power)? Assuming a generator efﬁciency of 60%, what is the overall RF efﬁciency (ac power to beam power)? (e) How many klystrons do we need? 2.10. For largest effective shunt impedance (highest power efﬁciency), we prefer high frequencies for room-temperature linacs. But if the frequency is too large, the radial variation of beam parameters may become intolerable. Choose the linac frequency for the cases below, by using the criterion that at the injection energy the accelerating-ﬁeld variation with radius over the beam aperture must not exceed 10%. Assume for all cases the radial aperture a = 1 cm for acceptable transmission through the entire accelerator. Assume that Ez (r) ∝ I0 (2πr/γβλ) and use the approximation I0 (x) = 1 + x2 /4. Choose the linac frequency for the following injection conditions for the structure phase velocity: (a) electron linac near 200 keV (use βγ = 1, or β = 0.707 instead of β = 1 which is really used), (b) proton linac at 750 keV (β = 0.04), (c) proton linac at 3 MeV (β = 0.08), and (d) proton linac near 100 MeV (β = 0.4). 2.11. To illustrate the choice of the gap length to optimize the cavity shunt impedance, suppose we represent an accelerating cavity by a shunt √ LRC circuit, where the gap capacitance has the form C = α/ g, α is a constant, and g is the gap length. Assuming ω is the angular resonant frequency, and V is the required gap voltage, the magnitude of the peak current charging the capacitance is I = |ωCV|. Suppose that the gap length is varied, and the inductance L is varied to keep the resonant frequency constant. (a) If β is the particle velocity and λ = 2πc/ω is the RF wavelength, use the simple expression for transit-time factor T to determine what gap length g/βλ maximizes T. (b) If R is the resistance of the circuit, show that the average power dissipated over an RF cycle is P = R(αωV)2 /2g, which is inﬁnite at g = 0. (c) If g is very small the power dissipation is too large, and if g is very large the transit-time factor is too small. There exists some value of g that maximizes the effective shunt impedance ZT 2 . Using the expression for the effective shunt impedance, given in the form ZT 2 = (VT)2 /(P), where l is the cell length, and using the simple expression for T versus g, ﬁnd the dependence of ZT 2 versus g/βλ, and plot the result. What approximate value of g/βλ gives the maximum ZT 2 .

2 RF Acceleration in Linacs

2.12. Prove that the quantity ejkz E(z) dz

2 sin(kz)E(z) dz cos(kz)E(z) dz + = E(z) dz is independent of the choice of origin, and is equal to the transit-time factor when the origin is at the electrical center of the gap. This provides a convenient way of calculating the transit-time factor corresponding to the origin is at the electrical center, without having to ﬁnd the exact location of the electrical center. (Hint: First, consider the integrand in the numerator on the left side as a phasor with a phase kz and magnitude E(z). The integral represents a sum of all these vectors, and the numerator gives the magnitude of this sum. Then consider an arbitrary coordinate transformation to a new origin displaced by . Consider what happens to the magnitudes and phases of each of the transformed vectors. Does the magnitude of the new sum change? Finally, does the denominator on the left side depend on the choice of origin?)

References 1 The case φ = ±π/2 can cause problems

Acceleration in a Gap, in Linear Accelerators, P.M. Lapostolle and with this deﬁnition. When in doubt the A.L. Septier, John Wiley & Sons, New reader can return to Eq. (2) to avoid York (1970), p. 755. singularities. 4 Lapostolle, P.M. Proton Linear 2 Panofsky, W.K.H. Linear Accelerator Accelerators: A Theoretical and Historical Beam Dynamics, University of California Introduction, Los Alamos Report Radiation Laboratory Report LA-11601-MS, (1989) p. 74. UCRL-1216, University of California, 5 Sometimes one ﬁnds a factor of two in Berkeley, Calif., February, 1951, the denominator for this deﬁnition, so available from the Technical be careful. Information Service, Oak Ridge, Tenn. 3 Carne, A., Lapostolle, P., Schnitzer, B. and Prom´e, M. Numerical methods,

3 Periodic Accelerating Structures

For an electromagnetic wave to deliver a continuous energy gain to a moving charged particle, two conditions must be satisﬁed: (1) the wave must have an electric ﬁeld component along the direction of particle motion and (2) the particle and wave must have the same velocity to maintain synchronism. The ﬁrst condition is not satisﬁed by electromagnetic waves in free space, but can be satisﬁed by a transverse magnetic wave propagating in a uniform waveguide. However, the second condition is not satisﬁed for a uniform waveguide, because the phase velocity vp > c. The most widely used solution for obtaining phase velocity vp < c in linacs has been the use of accelerating structures with periodic geometries. A periodic structure has the property that its modes are composed of a Fourier sum of waves, some of which are suitable for synchronous-particle acceleration. 3.1 Synchronous Acceleration and Periodic Structures

Two pictures can be used to describe periodic accelerating structures: (1) considering the structure as a periodically loaded waveguide in which reﬂections from the periodic loading elements reduce the phase velocity compared with the uniform guide and (2) considering the structure as a periodic array of coupled resonant cavities. The periodic waveguide picture leads to a qualitative description of how the dispersion curve of a uniform waveguide is modiﬁed by the presence of the periodic loading elements [1]. However, the coupled-cavity picture of a periodic accelerating structure provides a more quantitative description of the dispersion curve. The irisloaded structure can be viewed as an array of pillbox cavities that are coupled through the irises. An analytic treatment based on Bethe’s theory of coupling of cavities through apertures, the Slater perturbation theorem, and the Floquet theorem leads to an analytic description for the dispersion curve and a simple formula for the cavity coupling constant. In the remainder of this chapter we will look more closely at periodic structures and their properties. RF Linear Accelerators. 2nd, completely revised and enlarged edition. Thomas P. Wangler Copyright  2008 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-40680-7

3 Periodic Accelerating Structures

3.2 Floquet Theorem and Space Harmonics

In a lossless uniform waveguide with azimuthal symmetry, the axial electric ﬁeld for the lowest transverse-magnetic mode, the TM01 mode, is Ez (r, z, t) = EJ0 (Kr)e j(ωt−k0 z)

(3.1)

This describes a wave propagating in the +z direction, with wavenumber k0 = 2π/λg , where λg is the guide wavelength. The uniform waveguide has a dispersion relation, shown in Fig. 3.1, given by ω2 = (Kc)2 + (k0 c)2

(3.2)

where K is the cutoff wavenumber for the TM01 mode, related to the cutoff angular frequency by ωc = Kc. Then the phase velocity is expressed as vp =

ω c = >c k0 1 − (Kc)2 /ω2

(3.3)

Because the phase velocity is always larger than c, the uniform guide is unsuitable for synchronous-particle acceleration, and we must modify the structure to obtain a lower phase velocity. The solution that has been uniformly adopted is the periodic structure, Fig. 3.2. One might expect that converting the uniform guide to a periodic structure might perturb the ﬁeld distribution by introducing a z-periodic modulation of the amplitude of the wave, giving a TM01 propagating-wave solution of the form E(r, z, t) = Ed (r, z)e j(ωt−k0 z)

(3.4)

where Ed (r, z) is a periodic function with the same period d as the structure. Equation (3.4) is indeed found to be a correct solution, and it satisﬁes the

Figure 3.1

Dispersion curve (Brillouin diagram) for uniform waveguide.

3.2 Floquet Theorem and Space Harmonics

Figure 3.2

The periodic iris-loaded structure.

Floquet theorem stated in the following form. In a given mode of an inﬁnite periodic structure, the ﬁelds at two different cross sections that are separated by one period differ only by a constant factor, which in general is a complex number. We will ﬁnd that there are intervals of ω called stopbands where the constant is real with magnitude less than 1. For these regions the modes are evanescent – hence the name stopband. There are also passbands in which waves will propagate, within which in the loss-free case the complex constant is exp(jk0 d), which physically represents a cell-to-cell phase shift k0 d of the ﬁeld. To include losses, the constant factor can be written as e−γ z , where γ = −α + jk0 corresponds to propagation with attenuation. At the ends of the passband solutions are found where the constant is ±1, representing the k0 d = 0 and π modes. The Floquet theorem [2] in a passband is expressed as E(r, z + d) = E(r, z)e±jk0 d

(3.5)

where the sign depends on the sign of the wave propagation. The plausibility of the Floquet theorem for a propagating mode can be outlined for the loss-free periodic structure. Physically, one expects that the form of the time-independent amplitude function Ed (r, z) depends only on the details of the cell geometry. If this function is a solution for any one cell, it must also be a solution for any other cell in the structure, since all cells are identical. Therefore, a solution should exist using a periodic function for Ed (r, z). The exponential factor accounts for differences in the initial conditions that result in different phases of the oscillation in different cells. Because Ed (r, z) is periodic, it can be expanded in a Fourier series as Ed (r, z) =

∞ n=−∞

an (r)e−j2πnz/d

(3.6)

3 Periodic Accelerating Structures

We obtain the radial solution by requiring Eq. (3.4) to satisfy the wave equation. We ﬁnd an (r) +

an (r) − Kn2 an (r) = 0 r

(3.7)

for all n, where the prime notation refers to differentiation with respect to r, and where Kn2 = (ω/c)2 − (k0 + 2πn/d)2 . For Kn2 > 0, the solution for a propagating wave is ∞

Ez (r, z, t) =

En J0 (Kn r)ej(ωt−kn z)

(3.8)

n=−∞

where kn = k0 +

2πn d

(3.9)

We can interpret the solution of Eq. (3.8) as representing an inﬁnite number of traveling waves, which are called space harmonics, each of which is denoted by the index n. We refer to the principle wave as the one with n = 0. The space harmonics have the same frequency but different wavenumbers, and each has a constant amplitude En independent of z. The waves for n > 0 travel in the +z direction, and those for n < 0 travel in the −z direction. The wavenumber for the nth space harmonic is shifted from the wavenumber k0 of the principle wave by 2πn/d. The phase velocity for the nth space harmonic is βn =

ω β0 = kn c 1 + (nβ0 λ/d)

(3.10)

and by choosing n sufﬁciently large, one can obtain an arbitrarily low phase velocity. Not only has the introduction of periodicity led to the generation of space harmonics, but the phase velocity of the principle wave β0 = ω/ck0 is also generally modiﬁed. Physically, this can be attributed to the addition of reﬂections from the periodic elements to the principle wave. Computationally and experimentally, the resulting phase velocity for the principle wave at a given frequency is obtained from the slope of the dispersion curve. Generally, it is found that the value of β0 generally decreases as the perturbation of the periodic elements increases, as is discussed by Slater. When Kn2 > 0, the other nonzero components of the TM01 solution are Er (r, z, t) = j

∞

kn J1 (Kn r)e j(ωt−kn z) Kn

(3.11)

ω J1 (Kn r)e j(ωt−kn z) Kn c 2

(3.12)

n=−∞

Bθ (r, z, t) = j

∞ n=−∞

When Kn2 < 0, the Bessel functions J0 and J1 are replaced with modiﬁed Bessel functions I0 and I1 . The integrated effect on a beam particle that is

3.3 General Description of Periodic Structures

synchronous with one of the space-harmonic waves is obtained by assuming that only the synchronous space harmonic waves act on the beam. The effects of the nonsynchronous waves are assumed to average to zero and are ignored. To see this for the case where the synchronous-particle velocity matches that of the n = 0 principle wave, we use Eq. (3.8) and calculate the energy gain of the synchronous particle as it experiences the force from the wave over one structure period d. We assume that the synchronous particle is at an arbitrarily chosen origin z = 0 at time t = t0 , when the phase of the ﬁeld relative to the wave crest is φ0 = ωt0 . Substituting t = z/β0 c + t0 , the energy gain of the synchronous particle is d dzRe[Ez (r, z, t = z/β0 c + t0 )] = qE0 I0 (K0 r) cos(φ0 )d (3.13) W = q 0

where after substituting Eq. (3.8) only the n = 0 term survives. In Eq. (3.13), we have assumed that Kn2 < 0, resulting in the modiﬁed Bessel function. The principle wave usually has the largest Fourier amplitude and consequently most periodic accelerating structures are designed so that the principle wave n = 0 is synchronous with the beam (i.e., β0 = βs ). The effective ﬁeld components experienced by the synchronous particle can be identiﬁed from the n = 0 terms in Eqs. (3.8), (3.11), and (3.12). From the deﬁnition of the phase velocity, for n = 0 we have k0 = ω/β0 c, and we ﬁnd K02 = (ω/c)2 − k20 = −ω2 /γs2 βs2 c2 . The effective ﬁeld components seen by the beam are the components of the synchronous wave, given by Ez = E0 I0 (K0 r) cos(φ)

(3.14)

Er = −γs E0 I1 (K0 r) sin(φ)

(3.15)

γs βs E0 I1 (K0 r) sin(φ) c where we have written K0 = |K02 |, and Bθ = −

φ = ωt −

2π λ

dz βs (z)

(3.16)

(3.17)

For a structure with phase velocity equal to c, we have K0 = 0 and there is no radial dependence of the ﬁelds. These effective ﬁeld components will be used in later chapters when we discuss linac beam dynamics.

3.3 General Description of Periodic Structures

In Chapter 1 we have seen that a single cavity such as the pillbox cavity has an inﬁnite number of resonant modes, which we will call cavity modes. For the pillbox cavity, these modes were labeled as transverse-electric (TEmnp )

3 Periodic Accelerating Structures

and transverse-magnetic (TMmnp ) cavity modes. Typical accelerator cavities are constructed as periodic or almost periodic arrays of coupled cavities, and we must understand the characteristics of coupled-cavity periodic arrays. It is well known that for any system of coupled oscillators, there exists a family of so-called normal modes, each mode behaving like an independent harmonic oscillator with its own characteristic resonant frequency. In general, when any normal mode is excited by a suitable driving force at the right frequency, each of the individual oscillators participates in the motion, and each oscillates at the same frequency with a characteristic phase difference from one oscillator to the next. For an array of coupled cavities, each of the individual cavity modes generates its own family of normal modes. Each such family lies within a deﬁnite frequency band called a passband, which is centered near the resonant frequency of the uncoupled cavity mode. Each passband includes all the normal modes associated with a single cavity mode, such as the familiar TM010 mode of the pillbox cavity. One can describe each normal-mode solution in terms of a characteristic wave that can propagate through the cavity array with a characteristic frequency, and a characteristic wavelength or wavenumber. For a periodic array, the wave solutions are expressed graphically in the form of a so-called Brillouin diagram or dispersion relation, which is a plot of resonant frequency versus wavenumber. Each normal mode is represented by a single point on this plot. For an inﬁnite periodic array, there are an inﬁnite number of modes, and the passband is a continuous curve. For a ﬁnite length array there are a ﬁnite number of modes, equal to the number of coupled oscillators. The dispersion relation for an inﬁnite periodic structure typically shows passbands, frequency bands within which waves associated with a given family of normal modes can propagate with little attenuation, and stop bands, frequency bands where waves do not propagate. Figure 3.3 shows an example of the lowest passband corresponding to the TM010 resonant mode of the individual cavities. At any frequency within a passband, there are an inﬁnite number of waves, each corresponding to a different space harmonic, as described in Section 3.2. The n = 0 waves propagating in the +z direction correspond to the range from 0 < kz < π/d, where d is the spatial period, the n = 1 waves correspond to the range from 2π/d < kz < 3π/d, and so on. For a periodic structure, the dispersion curves express the frequency as a periodic function of wavenumber kz . Each cycle or zone in kz space represents the behavior of ω versus kz for the nth space harmonic. Because of the symmetry of the curve, all the information is presented if the plot is simply restricted to the range 0 ≤ kz ≤ π/d. At any given frequency, each space-harmonic component of a normal mode has a unique phase velocity, corresponding to the slope of the line from the origin to that point on the dispersion curve. All space-harmonic components have the same group velocity, corresponding to the same tangent on the dispersion curve.

3.4 Equivalent Circuit Model for Periodic Structures w

−2p/d

Figure 3.3

−p/d

p /d

2p /d

Dispersion curve of the lowest passband of an inﬁnite periodic structure.

3.4 Equivalent Circuit Model for Periodic Structures

More insight into the characteristics of periodic structures is obtained by considering some examples of periodic electrical circuits. Consider an inﬁnite chain of identical cells, each with a series impedance Z and a shunt admittance Y as shown in Fig. 3.4. Given a voltage Vn and current In for the nth cell, as shown in the ﬁgure, we can use Ohm’s law to write In−1 − In = YVn In − In+1 = YVn+1 Vn − Vn+1 = ZIn

(3.18)

Substituting the ﬁrst two equations into the third to eliminate the voltage, we ﬁnd In−1 2 In+1 − In Z + + =0 (3.19) Y Y Y

In −1

Zn −1 Yn −1

Figure 3.4

Zn Yn

Periodic chain of identical electrical cells.

Yn +1

3 Periodic Accelerating Structures

Suppose the period of the structure is , and consider a propagating wave with wavenumber kz = 2π/λg . Then, the phase advance of the wave per period

is φ = kz . Applying the Floquet theorem, we have for a wave traveling in the +z direction In = In−1 e−jφ , and In+1 = In e−jφ . Substituting this result into Eq. (3.19), we have jφ e 2 e−jφ − Z+ + =0 (3.20) In Y Y Y The solution is obtained when the square bracket is zero, or cos φ = 1 +

YZ 2

(3.21)

Particular choices of Z and Y introduce a frequency dependence, and then Eq. (3.21) relates frequency ω to phase advance per period φ. Equation (3.21) is an important result, because it yields the dispersion relation, and deﬁnes the passband from the requirement that −1 ≤ cos φ ≤ 1. The limits of the passband are at φ = 0 and π. Equation (3.21) can also be expressed in a compact form using 2 × 2 matrix notation. The effect of the series impedance Z on an input voltage and current (Vi , Ii ) can be written as Vf 1 −Z Vi = (3.22) 0 1 If Ii where (Vf , If ) are the voltage and current at the output end of the impedance. Likewise, the effect of the shunt admittance Y can be written in terms of a 2 × 2 transfer matrix, as 1 0 Vi Vf = (3.23) −Y 1 If Ii The transformation of the voltage and current through the whole cell can be written in terms of the product transfer matrix M, as Vn Vn+1 =M (3.24) In+1 In where M is the transfer matrix through one period. 1 0 1 −Z 1 −Z M= = −Y 1 0 1 −Y 1 + YZ

(3.25)

The trace of the matrix through one period is Tr(M) = 2 + YZ. Thus, Eq. (3.21) can be written as cos φ = 1 + ZY/2 = Tr(M)/2

(3.26)

3.5 Periodic Array of Low-Pass Filters

from which the dispersion relation is obtained for any particular choice of impedance and admittance. Next we apply the equivalent circuit model to four examples.

3.5 Periodic Array of Low-Pass Filters

As the ﬁrst example, we study the simple low-pass ﬁlter periodic array, consisting of an array of series inductances coupled by shunt capacitances, as shown in Fig. 3.5. We identify Z = jωL1 , and Y = jωC2 . From Eq. (3.26) cos φ = 1 + ZY/2 = 1 − ω2 /2 ω02

(3.27)

where ω02 = 1/L1 C2 . The dispersion relation can be written as ω = ω0 2(1 − cos φ)

(3.28)

and is plotted in Fig. 3.6. The limits of the passband are ω = 0 at φ = 0 (0 mode), and ω = 2ω0 at φ = π(π mode). The π mode lies higher in frequency than the 0 mode, and the bandwidth of the passband is ωπ − ω0 ∼ = 2 ω0 . As shown in Fig. 3.5, the low-pass ﬁlter looks like an ideal transmission line with ﬁnite rather than inﬁnitesimal cell lengths. This suggests that the dispersion relation should reduce to that of an ideal transmission line, when the phase advance per period becomes small. When φ 1 we

L1 C2

Figure 3.5

Basic cell of a periodic low-pass ﬁlter.

3.5 3 2.5 2 1.5 1 0.5 −7.5

−5

−2.5

f 0

2.5

7.5

Figure 3.6 Dispersion curve for periodic low-pass ﬁlter.

3 Periodic Accelerating Structures

√ ∼ kz / L1 C2 . Identifying L1 / and C2 / as the ﬁnd ω ∼ = ω0 φ = ω0 kz , or ω = inductance and capacitance per unit length, we ﬁnd that the dispersion relation does reduce to that of the ideal transmission line. 3.6 Periodic Array of Electrically Coupled Circuits

We consider in Fig. 3.7 an example that may serve as an equivalent circuit for a periodic array of electrically coupled cavities, such as the iris-loaded waveguide structure. The ﬁgure shows a single period of a periodic circuit in which an inductance and capacitance in series form a resonator that is coupled to the next identical period by a shunt capacitance. We have Z = jωL1 + 1/jωC1 = jωL1 (1 − ω02 /ω2 ), where ω02 = 1/L1 C1 , and Y = jωC2 . Then, cos φ = 1 + ZY/2 = 1 −

C2 2 2 (ω /ω0 − 1) 2C1

Solving for ω2 , we ﬁnd the dispersion curve ω = ω0 2(C1 /C2 )(1 − cos φ) + 1

(3.29)

(3.30)

The dispersion curve of Eq. (3.30) is plotted in Fig. 3.8. The limits of the passband correspond to φ√= 0 and π. At φ = 0 (0 mode) ω = ω0 . At φ = π (π mode), ω = ωπ ∼ = ω0 4(C1 /C2 ) + 1, which for C1 /C2 1 becomes ωπ ∼ = ω0 (1 + 2C1 /C2 ). The π mode lies higher in frequency than the 0 mode.

C1 C2

Figure 3.7 Basic cell of a periodic array of electrically coupled TM circuits.

3.5 3 2.5 2 1.5 1

0.5 −7.5

−5

−2.5

2.5

7.5

Figure 3.8 Dispersion relation for an electrically coupled periodic TM cell array.

3.7 Periodic Array of Magnetically Coupled Circuits

It is sometimes convenient to normalize frequency differences to the frequency of the φ = π/2 mode, which is in the center of the passband. 2 = ω02 [1 + 2C1 /C2 ], and can write We have ωπ/2 2 ω2 = ωπ/2 [1 − k cos φ]

(3.31)

where k = 2C1 /(C2 + 2C1 ) is deﬁned as the intercell coupling constant, a measure of the strength of the coupling. The parameter k is not to be confused with the wavenumber. For k 1, the fractional bandwidth of the passband is (ωπ − ω0 )/ωπ/2 = k. Therefore the coupling constant is also the fractional width of the passband. If the capacitance C2 → ∞, the intercell coupling vanishes. In this limit the cells become independent, and the bandwidth of the passband is zero.

3.7 Periodic Array of Magnetically Coupled Circuits

We now modify the circuit of Section 3.6 by changing the shunt admittance, which served as the coupling element between adjacent cells, from a capacitance to an inductance as shown in Fig. 3.9. This circuit represents a periodic array of coupled cavities with magnetic rather than electrical coupling, representing a model for a periodic array of magnetically coupled cavities that are coupled through slots cut in the iris walls. We have Z = jωL1 + 1/jωC1 = jωL1 (1 − ω02 /ω2 ), where ω02 = 1/L1 C1 and Y = 1/jωL2 . Then, cos φ = 1 + ZY/2 = 1 +

L1 (1 − ω02 /ω2 ) 2L2

(3.32)

Solving for ω2 , we ﬁnd the dispersion curve ω= √

ω0 2(L2 /L1 )(1 − cos φ) + 1

(3.33)

The dispersion curve is plotted in Fig.√3.10. At φ = 0, the 0 mode, ω = ω0 . At φ = π, the π mode, ω = ωπ ∼ = ω0 / 4(L2 /L1 ) + 1, which for L2 /L1 1 becomes ωπ ∼ = ω0 (1 − 2L2 /L1 ). Now, the π mode lies lower in frequency than the 0 mode. By comparison in Sections 3.5 and 3.6, electric coupling gave a

C1 L2

Figure 3.9 Basic cell of a periodic array of magnetically coupled TM circuits.

3 Periodic Accelerating Structures 4

3.5 3

2.5 2 1.5 1 0.5 −7.5

−5

−2.5

f 0

2.5

7.5

Figure 3.10 Dispersion relation for circuit model of magnetically coupled periodic TM cavity array.

π mode higher in frequency than the 0 mode. Comparing phase velocity, represented by the slope of the line from the origin to any point on the dispersion curve, with the group velocity, represented by the tangent to the curve at that point, we see that these two velocities have opposite sign, which, by deﬁnition, 2 = ω02 /[1 + 2(L2 /L1 )], and Eq. (2) is called a backward-wave mode. We have ωπ/2 can be written ω2 =

2 ωπ/2

(3.34)

1 − k cos φ

where k = 2L2 /(L1 + 2L2 ) is deﬁned as the cavity coupling constant. For k 1, the fractional bandwidth of the passband is (ωπ − ω0 )/ωπ/2 = −k. The minus sign indicates only that the π mode lies lower in frequency than the 0 mode. If the inductance L2 → 0, the intercell coupling vanishes. In this limit the cells become independent and the bandwidth of the passband is zero.

3.8 Periodic Array of Cavities with Resonant Coupling Element

As a ﬁnal example, we now consider the case of a shunt resonant admittance as the coupling element, as shown in Fig. 3.11.

C1 L2

Figure 3.11 Basic cell of a periodic array of circuits with resonant coupling element.

3.9 Measurement of Dispersion Curves in Periodic Structures

We now have two resonating elements within the same basic cell, and we will see that this leads to an interesting and important effect, when the series and shunt resonant frequencies are equal. We begin by writing Z = jωL1 + 1/jωC1 = jωL1 (1 − ω12 /ω2 ), where ω12 = 1/L1 C1 and Y = jωC2 + 1/jωL2 = jωC2 (1 − ω22 /ω2 ), where ω22 = 1/L2 C2 . Then, cos φ = 1 + ZY/2 = 1 −

ω 2 L 1 C2 (1 − ω12 /ω2 )(1 − ω22 /ω2 ) 2

Solving this as a quadratic equation for ω2 , we ﬁnd b±

ω=

where

b2 − ω12 ω22

ω12 + ω22 2

(3.35)

(3.36)

(1 + k2 (1 − cos(φ)))

(3.37)

and we have deﬁned k2 = 2/[L1 C2 (ω12 + ω22 )] The dispersion relation has two branches, depending on the choice of the sign in Eq. (3.36). Figure 3.12 shows the dispersion relation for several choices of the parameters with ω2 ≥ ω1 . Because Eq. (3.35) is symmetric with respect to interchange of ω1 and ω2 , there is again a stopband when ω1 > ω2 . Figure 3.12 shows an important general effect that occurs as the frequencies ω1 and ω2 approach each other; one can observe a steepening of the slopes near the 0 mode. When ω1 = ω2 , the two curves intersect, an effect known as conﬂuence. Each curve continues into the other branch with the same ﬁnite slope. As a consequence, the group velocity near the 0 mode increases from zero to a ﬁnite value at conﬂuence. This phenomenon has been applied to standing-wave accelerating structures to produce increased mode spacing at conﬂuence, giving signiﬁcantly greater ﬁeld stability in a long array of coupled cavities. Resonantly coupled accelerating structures are described in Chapter 4.

3.9 Measurement of Dispersion Curves in Periodic Structures

The dispersion relation of a structure contains the detailed information needed to determine its suitability for particle acceleration, such as phase and group velocities as a function of frequency. Analytic calculations or models, such as the ones given earlier, can be helpful for obtaining physical insight, but are generally not adequate for design purposes. To determine a dispersion relation for a periodic structure, designers rely on either laboratory measurements or numerical calculations using electromagnetic ﬁeld–solving codes. In either case the usual method for determining the dispersion

3 Periodic Accelerating Structures 5

w w2 = 3

w w2 = 2.5

1 f

(a)

−7.5 −5 −2.5 0

2.5

7.5

−7.5 −5 −2.5 0

(b) 5

2.5

7.5

w2 = 2

4 3 2 1

f (c)

−7.5 −5 −2.5 0

2.5

7.5

Figure 3.12 Dispersion curve for circuit model of a periodic array of cavities √ with a resonant coupling element. Deﬁning ω0 = 1/ L1 C2 , the curves are plotted for the following parameters ω0 = 1, ω1 = 2, and the three values ω2 = 3, 2.5, and 2.

curve is to measure the resonant frequencies for different sections of the structure, deﬁned by the placement of two shorting planes, as shown in Fig. 3.13. Recall that each standing-wave mode is the superposition of two traveling waves of equal amplitude moving in opposite directions as described by the equation Re{V0 ej(ωt−kz z) + V0 ej(ωt+kz z) } = 2V0 cos(kz z) cos(ωt)

(3.38)

where kz is the wavenumber. In the laboratory, by inserting weakly coupled radio-frequency (RF) drive probes, the resonant standing-wave modes can be excited, and their frequencies can be measured for any position of the shorting planes. To excite the correct cavity mode, the shorting planes are generally placed at a symmetry plane, where the electric ﬁeld of the mode is normal to the plane. For example, in the electrically coupled iris-loaded structure with periodic iris spacing , the longitudinal electric ﬁeld is maximum at the midpoints between the irises. If the shorting planes are located at these midpoints and include just one iris, or one full cell, two normal modes, based on the TM010 cavity mode can be excited, the lower frequency mode

3.9 Measurement of Dispersion Curves in Periodic Structures p mode

p/2 mode

(a)

(b) p/3 mode

(c)

2p/3 mode

(d)

Figure 3.13 Determining the dispersion curve by measurement of the resonant frequencies for different sections of the structure, deﬁned by the placement of two shorting planes. (a) π mode, (b) π/2 mode, (c) π/3 mode, and (d)2π/3 mode.

is the 0 mode, where the ﬁelds are in phase on each side of the iris and the higher frequency mode is the π mode, where the ﬁelds reverse sign at the location of the iris [3]. From this measurement alone, the bandwidth is determined, which, as was shown in the previous sections, gives a measure of the intercell coupling strength k. Next, if the shorting planes are positioned to include two irises, or two cells, the two previous modes will be excited, and also there will be a new normal mode, which is the kz = π/2 mode. Likewise, when three cells are included, one can excite the kz = π/3 and kz = 2π/3 normal modes. As more cells are included, more normal modes can be excited, and more points on the dispersion curve can be determined. The same procedure can also be followed using the computer, when designing a periodic structure. This procedure allows the effect of design changes in the cell geometry to be evaluated in terms of their effects on the dispersion curve.

3 Periodic Accelerating Structures

3.10 Traveling-Wave Linac Structures

Next we describe the disk-loaded or iris-loaded waveguide structure (Fig. 3.14) used to produce traveling-wave acceleration of relativistic electrons [4]. The basic principles of operation of a traveling-wave accelerator are simple conceptually. The linac consists of a sequence of identical tanks, each consisting of an array of accelerating cavities or cells separated by the irises. The electromagnetic wave is launched at the input cell of each tank; the wave propagates along the beam direction, and beam bunches are injected along the axis for acceleration by the wave. The electromagnetic energy is absorbed by the conductor walls, and by the beam, and the ﬁeld amplitude attenuates along each tank. At the end of each tank the remaining energy is delivered to an external resistive load. The structural parameters should be generally chosen to achieve the acceleration with high power efﬁciency. It is desirable to maximize the energy gain of the beam over a given distance and to minimize the power lost to the walls and to the external load. Perhaps the most important design parameter and usually the ﬁrst parameter chosen is the frequency. Some basic considerations for the choice include: (1) higher power efﬁciency at higher frequencies, because of the ω1/2 dependence of shunt impedance and (2) tighter beam-positioning tolerances at higher frequencies, because of smaller apertures. Other considerations are often equally important. For applications requiring acceleration of very short intense bunches of electrons, it is desirable to provide large stored energy per unit length, which scales as ω−2 , favoring lower frequencies. For linacs used in linear collider applications, where the beam emittance of high-intensity bunches must be controlled, the undesirable effects of wakeﬁelds must be considered. These effects scale as ω2 for the longitudinal case, and as ω3 for the transverse case. Thus, the choice of frequency is inevitably a compromise, because the frequency seems to have a pervasive impact effecting nearly every linac-design issue. The most common frequency chosen has been 2856 MHz, which is the frequency of the SLAC linac.

Figure 3.14 Iris(disk)-loaded traveling-wave structure.

3.11 Analysis of the Periodic Iris-Loaded Waveguide

Another important parameter to choose is the group velocity vg that can be controlled by the choice of the aperture radius. The group velocity is important because (1) it affects the electromagnetic ﬁll time tF = L/vg , where L is the length of the accelerator section and (2) it affects the ﬁeld amplitude and the power dissipation for any given input RF power. Thus, for any travelingwave power Pw , one can express the basic relation Pw = Uvg , where U is the electromagnetic stored energy per unit length in the wave. For the same wave power one can consider two extreme cases. By choosing a low group velocity, one has a large stored energy and high ﬁelds. This results in a high initial acceleration rate and can also lead to rapid decay of the wave power, because of the large resistive power dissipation in the structure walls. Alternatively, by choosing a high group velocity, the stored energy and the ﬁelds may be too low for efﬁcient transfer of power to the beam. We will see that the group velocity must be chosen to balance these considerations, and we will consider two ways of doing this in the following two sections. Other parameters must also be optimized. For example, the disk separation must be optimized for maximum shunt impedance per unit length. If the disks are too far apart, the amplitude of the synchronous harmonic waves is reduced relative to the other harmonics. If they are too close, the power loss on the disks becomes signiﬁcant. The disk thickness is not a critical parameter, and is chosen large enough for good mechanical strength and good tolerance for electrical breakdown. The radius b of the waveguide and the aperture radius a determine the phase velocity of the synchronous wave. As was mentioned earlier, the radius a also strongly inﬂuences the group velocity. 3.11 Analysis of the Periodic Iris-Loaded Waveguide

In this section, we obtain an approximate formula for the dispersion curve for the iris-loaded structure, shown in Fig. 3.15. The results presented in this section [5] are useful for providing scaling formulas and physical insight, rather than to replace the accurate numerical results, which could be obtained from the methods introduced in Section 3.9. We describe the iris-loaded periodic structure as an inﬁnite array of pillbox cavities, each excited in a TM010 mode, coupled through small apertures on the axis. The Slater perturbation theorem (discussed later), the Bethe theory [6] of the coupling of cavities through small holes, and the Floquet theorem can be used to obtain an expression for the dispersion curve and other useful results. The Slater perturbation theorem relates the resonant-frequency change, resulting from a cavity perturbation, to the change in the electric and magnetic stored energies of the cavity. The theorem states that the perturbed resonant frequency ω is given by ω − ωr 1 = ( Um − Ue ) ωr U

(3.39)

3 Periodic Accelerating Structures

Figure 3.15 The iris-loaded traveling-wave accelerating structure including the input waveguide through which the electromagnetic wave is injected. The beam moves along the central horizontal axis.

where ωr is the unperturbed frequency, U is the unperturbed electromagnetic stored energy, and Um and Ue are the time-averaged magnetic and electric stored energies removed as a result of the perturbation. Bethe’s theory describes the perturbing effect on the cavity ﬁelds of a hole in the cavity wall. It can be shown that, when the hole size is small compared with the wavelength, the effect of a hole is equivalent to adding electric and/or magnetic dipoles located at the center of the hole, whose dipole moments are proportional to the unperturbed electric and/or magnetic ﬁelds at the center of the hole. For a small circular aperture, the equivalent electric-dipole moment representing the ﬁeld inside the cavity caused by the perturbation of the hole is P = −2a3 ε0 E0 /3, where a is the aperture radius, and E0 is the unperturbed electric ﬁeld at the location of the coupling aperture. The electric-dipole direction is normal to the plane of the hole; the sign of the dipole moment is negative because the induced dipole moment is opposite the direction of the unperturbed ﬁeld [7]. For a wall of inﬁnitesimal thickness, the ﬁeld outside the cavity resulting from the hole is also represented by an electric-dipole moment located at the hole, which has the same magnitude but opposite sign, compared with the dipole moment describing the ﬁeld within the cavity. [8] The ﬁnite thickness of the wall can be represented by a factor that represents the decay of the ﬁeld below cutoff of a waveguide, whose length equals the wall thickness. The basic assumption of the model is that the interaction energy of the dipoles will change the stored energy, and as a consequence of Eq. (3.39), there will be a resonant frequency shift of each normal mode. For an array of identical TM010 -mode cavities separated by small circular apertures of radius a, one obtains Um = 0, and Ue = −a3 ε0 E02 (1 − e−αh cos ψ)/6 where α=

(2.405/a)2 − ω2 /c2 ≈ 2.405/a

(3.40)

(3.41)

is the attenuation per unit length of the ﬁeld for the TM01 waveguide mode [9] through a hole in a wall of thickness h, and where ψ is the phase advance per

3.11 Analysis of the Periodic Iris-Loaded Waveguide

cavity of the traveling wave. For a ﬁnite array of coupled cavities only discrete values of ψ will be possible. The cavities are indistinguishable and the total frequency shift for each is the same; from Eq. (3.39) it is given by −N Ue ω − ωr = ωr U

(3.42)

where N is the number of apertures per cavity. For two cavities separated by a single wall, N = 1. For an array of coupled cavities (for proper termination the cells on the ends should be half cells), each cavity sees two apertures, so this is N = 2. Substituting Eq. (3.40) and N = 2 into Eq. (3.42), the expression for the frequency change is ω − ωr a3 ε0 E02 (1 − e−αh cos ψ) = ωr 3U

(3.43)

Recall that the stored energy for the TM010 mode of an unperturbed pillbox cavity is U=

πb2

ε0 E02 J12 (2.405) 2

(3.44)

where b is the cavity radius, is the axial length of the cavity, and J1 (2.405) = 0.5191. Also for the TM010 mode, b is related to the unperturbed resonant frequency by ωr = 2.405c/b. Substituting Eq. (3.44) into Eq. (3.43), we obtain the basic dispersion relation 2.405c 1 + κ(1 − cos(ψ)e−αh ) (3.45) ω= b where κ≡

4a3 3πJ12 (2.405)b2

(3.46)

The phase advance per cavity can be written as ψ = kz , where kz = ω/vp and vp is the phase velocity. The phase velocity can be expressed as ω 2.405c

vp ≡ = 1 + κ(1 − cos(ψ)e−αh ) (3.47) kz bψ Generally κ 1 so that vp is inﬂuenced mostly by ψ/ and the radius b. The group velocity is obtained from Eq. (3.45) by substituting ψ = kz and differentiating is vg =

dω 2(2.405)c a 2 = sin(ψ)e−αh dkz 3πJ12 (2.405) b

(3.48)

The group velocity is a sensitive function of the radii a and b. If b is used to deﬁne the phase velocity according to Eq. (3.47), then a can be used to deﬁne

3 Periodic Accelerating Structures

the group velocity. In the approximation that κ 1, we ﬁnd that the fractional bandwidth of the TM010 passband is ω(ψ = π) − ω(ψ = 0) ∼ κe−αh ∼ −αh =k = = κe ω(ψ = π/2) 1 + κ/2

(3.49)

where we deﬁne k = κe−αh . The quantity k should not be confused with the wavenumber kz . On the basis of the result discussed earlier in this chapter that the fractional width of the passband is equal to the intercell coupling constant, we identify k as the intercell coupling constant and express it as k≡

4a3 e−αh 3πJ12 (2.405)b2

(3.50)

If we specify the operating mode parameters as ω0, ψ0, kz0 , and v0 is the velocity of the synchronous particle, the required cavity length is

ψ 0 v0 ψ 0 λ 0 v0 = ω0 2π c

(3.51)

Typical parameter values are ω0 /2π = 3 GHz, v0 = c, ψ0 = 2π/3, b/λ0 = 0.4, a/λ0 = 0.1, h/λ0 = 0.05, which leads to /λ0 = 1/3, a coupling constant k = 0.009 and a group velocity vg /c = 0.008.

3.12 Constant-Impedance Traveling-Wave Structure

It is convenient to introduce some important relationships between the longitudinal accelerating ﬁeld amplitude Ea for the traveling wave, the stored energy per unit length U, and the traveling wave power Pw . The traveling-wave power is obtained by integrating the Poynting vector over the aperture, and a is Pw = 0 Er Hθ 2πrdr, where a is the radial aperture. The resistive power dissipation per unit length in the walls of the structure is −dPw /dz. The quality factor is Q = ωU/(−dPw /dz), and the shunt impedance per unit length is rL = Ea2 /(−dPw /dz). The group velocity, vg , is also the energy-ﬂow velocity, and relates the traveling-wave power to the stored energy per unit length according to Pw = vg U. Eliminating U from this expression and from the expression for Q yields a differential equation for traveling-wave power ωPw dPw =− dz Qvg

(3.52)

We deﬁne the ﬁeld attenuation per unit length as α0 = ω/2Qvg , so Eq. (3.52) becomes dPw = −2α0 Pw dz

(3.53)

3.12 Constant-Impedance Traveling-Wave Structure

Now we consider the simplest case of an iris-loaded traveling-wave structure with uniform cell geometry independent of z, and identical parameters for each cell including Q, vg , rL , and α0 . This is called a constant-impedance structure in the literature. The attenuation per unit length is constant, so the solution to Eq. (3.53) is Pw (z) = P0 e−2α0 z

(3.54)

which shows that the wave power is exponentially damped. We can obtain a similar expression for the accelerating ﬁeld amplitude. Thus, from the basic deﬁnitions given above, it is straightforward to show that the accelerating ﬁeld and the traveling-wave power are related by Ea2 = ωrL Pw /Qvg and from Eq. (3.53) dEa ωEa = −α0 Ea =− dz 2Qvg

(3.55)

The solution is Ea (z) = E0 e−α0 z . At the end of a tank of length L, we have Pw (L) = P0 e−2τ0 , and Ea (L) = E0 e−τ0 , where τ0 = α0 L =

ωL 2Qvg

(3.56)

is the total power attenuation parameter for that tank. The energy gain of a synchronous particle riding at a phase φ relative to the crest of the wave is L (1 − e−τ0 ) Ea (z) dz = qE0 L cos φ (3.57) W = q cos φ τ0 0 Using the relation between the input power P0 and input ﬁeld E0 evaluated at z = 0, we have E02 = 2rL α0 P0 , which leads to (1 − e−τ0 ) cos φ W = q 2rL P0 L √ τ0

(3.58)

If the input power and shunt impedance are ﬁxed, the energy gain over a tank of length L depends on the total attenuation parameter τ0 . The total attenuation parameter can be controlled by choosing the group velocity, which is a strong function of the aperture radius a. If the value of τ0 is to be chosen to maximize the energy gain W in the length L, we ﬁnd that the maximum occurs when τ0 = (eτ0 − 1)/2, which has for its solution τ0 ∼ = 1.26, and Eq. (3.58) yields the maximum energy gain per tank Wmax = 0.903q rL P0 L cos φ (3.59) Also, from the deﬁnition of τ0 , if L is ﬁxed and Q is known, the value of τ0 determines the optimum group velocity. If vg is too small, the attenuation of the wave is too great, and the ﬁeld in the latter part of the structure is very small. In this case, most of the power that is traveling relatively slowly through the

3 Periodic Accelerating Structures

structure is dissipated in the walls of the structure, resulting in poor transfer of power to the beam. If vg is too large, the initial accelerating ﬁeld for a given input power is too small, and therefore the accelerating ﬁeld throughout the structure is also too small for efﬁcient energy transfer to the beam. In this case, most of the beam power passes rapidly through the structure and is delivered to the external resistive load. The value of τ0 also affects the ﬁlling time of the waveguide, which is calculated as the time for the energy to propagate at the group velocity from the input to the output end of the guide. The ﬁll time is tF =

2Q L = τ0 vg ω

(3.60)

which is tF = 2.52Q/ω for the optimum value of τ0 . One may need to choose τ0 to be less than the optimum value, if the ﬁlling time needs to be reduced.

3.13 Constant-Gradient Structure

In the constant-impedance structure with uniform cell geometry and uniform parameters, we found that the RF power and the electric ﬁeld decay exponentially as the wave propagates away from the input. This raises the question of whether we could do better by varying the transverse geometry to keep the accelerating ﬁeld constant along the structure. This design approach is called the constant-gradient structure in the literature, where the term gradient refers to the voltage gradient, that is, accelerating ﬁeld. The group velocity and the attenuation per unit length α0 , which depends on the group velocity, are very sensitive to the aperture radius, and their variation will be included explicitly. In the approximate treatment that follows, the quantities Q and rL , which are not as sensitive to the transverse geometry, will be assumed constant. Beginning with dPw = −2α0 (z)Pw dz

(3.61)

we can write L PL dPw = −2 α(z)dz Pw 0 P0

(3.62)

where PL is the traveling-wave power at the end of the section, which is delivered to a resistive load. Integrating Eq. (3.62), we obtain PL = P0 e−2τ0

(3.63) L

where P0 is the input power, and the total power attenuation is τ0 ≡ 0 α0 (z)dz. From the deﬁnition of rL , which is assumed to be essentially constant, we

3.13 Constant-Gradient Structure

conclude that for Ea to be constant, so must be dPw /dz. Then Pw (z) must be linear in z, or Pw (z) = P0 +

PL − P0 z L

Substituting Eq. (3.63), we ﬁnd

z Pw (z) = P0 1 − (1 − e−2τ0 ) L

(3.64)

(3.65)

Therefore, for a constant ﬁeld, the traveling-wave power decreases linearly with z, rather than exponentially. Next we want an expression for α0 (z). We ﬁrst differentiate Eq. (3.65) dPw P0 = − (1 − e−2τ0 ) dz L

(3.66)

Using Eqs. (3.61) and (3.66), and eliminating dPw /dz, we ﬁnd α0 (z) =

1 (1 − e−2τ0 ) 2L [1 − (z/L)(1 − e−2τ0 )]

(3.67)

The group velocity is vg (z) =

ω ωL [1 − (z/L)(1 − e−2τ0 )] = 2Qα0 (z) Q (1 − e−2τ0 )

(3.68)

The group velocity required for a constant accelerating ﬁeld also decreases linearly with z. From Eqs. (3.65) and (3.68), the stored energy per unit length, like the accelerating ﬁeld, is constant, given by U = P0 Q(1 − e−2τ0 )/ωL. The energy gain in a section of length L for a particle riding at a phase φ relative to the crest of the wave is L Ea (z) dz = qE0 L cos φ (3.69) W = q cos φ 0

The relation between the constant accelerating ﬁeld amplitude and the input power is E02 = −rL

dPL rL P0 = (1 − e−2τ0 ) dz L

Using Eq. (3.70) to eliminate E0 from Eq. (3.69) gives W = q rL P0 L(1 − e−2τ0 ) cos φ

(3.70)

(3.71)

Comparing Eq. (3.71) with Eq. (3.58), one ﬁnds that for a given τ0 , the energy gain of the constant-gradient structure is larger than for the constantimpedance structure, although for τ0 < 1 the differences are not large. In the constant-gradient case the optimum τ0 is inﬁnite, corresponding to

3 Periodic Accelerating Structures

transferring all the energy in the structure. In practice, we also need to consider the effect of τ0 on the ﬁlling time of the waveguide. The ﬁlling time is L L Q dz dz tF = = (1 − e−2τ0 ) (3.72) v (z) ωL 1 − (z/L)(1 − e−2τ0 ) g 0 0 Carrying out the integration gives tF = τ0

2Q ω

(3.73)

which is equal to the result for the constant-impedance structure, given in Eq. (3.60). The constant-gradient design was chosen over the constant-impedance design at SLAC for reasons which include the uniformity of the power dissipation, and the lower value for the peak surface electric ﬁeld for the same energy gain. To achieve the constant-gradient design, the structure radius was tapered from about b = 4.2 to 4.1 cm, the iris radius was tapered from about 1.3 to 1.0 cm, and τ0 = 0.57 was chosen as a compromise between maximizing the energy gain and minimizing the ﬁll time [10]. 3.14 Characteristics of Normal Modes for Particle Acceleration

In this section, we explore the question of which normal modes are most suitable for efﬁcient particle acceleration [11]. For simplicity, suppose that an accelerating structure consists of an array of N + 1 identical coupled cavities, each with a short accelerating gap, so that we can approximate the transit-time factor T ∼ = 1. We assume that the ﬁeld is zero outside the gaps, and that the gaps are spaced at a distance apart. As will be discussed in Chapter 4, the standing-wave ﬁeld in each cavity of a periodic structure with N + 1 cells is described by

nπq En = E0 cos cos ωt (3.74) N where n = 0, 1, 2, . . . , N, is the cavity number and q = 0, 1, . . . , N is the normal-mode number. The usual nomenclature for each mode is to identify it by the quantity πq/N. Thus, q = 0 corresponds to the 0 mode, and q = N corresponds to the π mode. A standing wave can be expressed as a sum of two traveling waves moving in opposite directions; the forward wave is proportional to ej(ωt−nπq/N) , and for the backward wave we have ej(ωt+nπq/N) . The wavenumbers of these two waves are identiﬁed as kz = ±πq/N , and the corresponding phase velocities are given by vp = ω/kz = ±ωN /πq. If vs is the synchronous-particle velocity, synchronism with the forward wave requires vs =

ωN

πq

(3.75)

3.14 Characteristics of Normal Modes for Particle Acceleration

Assuming that vs is constant and that the synchronous particle starts at the center of gap n = 0 at time t = 0, the time for the particle to travel to gap n is given by t = nπq/ωN. Then, from Eq. (3.74) the ﬁeld in the nth gap seen by a synchronous particle is En = E0 cos2 (nψ)

(3.76)

where ψ = πq/N is the phase advance of the wave per cavity. It is instructive to look at this result from the equivalent point of view of two traveling waves. Expressed as the sum of traveling waves, we have jnψ e + e−jnψ jωt e En = E0 cos(nψ)ejωt = E0 (3.77) 2 Taking the real part and introducing the synchronism condition, again we ﬁnd the ﬁeld in the nth gap seen by the synchronous particle, where ωt = nψ, En =

E0 E0 [cos(ωt − nψ) + cos(ωt + nψ)] = [1 + cos(2nψ)] 2 2

(3.78)

This result is identical to the standing-wave result of Eq. (3.76) as can be seen by applying a simple trigonometric identity for the half angle. The ﬁrst term in the last bracket of Eq. (3.78) corresponds to the contribution from the forward wave, and the cosine term is the contribution from the backward wave. Next, the total voltage gain of the synchronous particle, assumed for simplicity to arrive at each gap at the crest of the wave, is (N + 2) N N E0

E0 , 0 < q < N En = [1 + cos(2nψ)] = V = 2 2 (N + 1)E0 , q = 0, N n=0 n=0 (3.79) Note that if the backward wave contributed zero energy gain, the cosine term in Eq. (3.79) would not be present and we would have V = (N + 1)E0 /2

(3.80)

Comparison of Eqs. (3.79) and (3.80) shows that the backward wave does contribute to the voltage gain. For the case 0 < q < N, the fractional contribution of the backward wave is small, and vanishes in the limit N → ∞, since in that limit the result of Eq. (3.79) equals that of Eq. (3.80). For the case q = 0 (the 0 mode), or q = N (the π mode), the energy gain is twice that of the forward wave alone, which means that the backward wave contributes the same amount to particle acceleration as the forward wave. This result is less surprising when one realizes that for the 0 mode and π modes, both the backward and forward waves move (in opposite directions) from one gap to the next at the same time as the particle moves that distance. Then, both the backward and the forward waves can be said to be synchronous with the

3 Periodic Accelerating Structures

beam, arriving at the gaps at the right time for acceleration. That the backward wave can deliver a net energy gain to the forward-moving beam is possible, because the beam only sees the ﬁeld from the backward wave when it is in the gap. From the point of space harmonics, one ﬁnds that both the forward and backward waves contribute space harmonics of equal strength that are synchronous with the beam [12]. Standing-wave operation in the 0 and π modes and traveling-wave operation in any mode are equally efﬁcient, as can be conﬁrmed by calculation of the shunt impedances. To show this, we rewrite the time dependent voltage for the nth cavity as Vn = Vn0 cos(ωt), where the nth amplitude is Vn0 = V0 cos(nψ), 2 and V0 = E0 . The effective shunt impedance per cavity is r = Vn0 /Pn , where Pn is the power dissipated in the nth cavity. We assume that all the cavities are identical and have the same shunt impedance. The total power dissipation for the entire array of N + 1 cavities is N+2 N N N 2 Vn0 V02 cos2 (nψ) V02 , 0 − ln(0.01N/k0 QL )/α

(4.55)

4.15 Resonant Coupling

The general approach of using resonant oscillators as coupling elements is called resonant coupling. We have already discussed the case of resonant coupling in a π/2-like mode of a biperiodic cavity chain. In general the dispersion relation is described by two passbands, separated by a stopband, as shown in Fig. 4.20. For an inﬁnite structure, the stopband is eliminated or closed, when the frequency of the highest mode of the lower passband coincides with the frequency of the lowest mode of the upper passband. The modes that are joined may be either the zero or the π modes, shown in Fig. 4.21. The joining of the two passbands in this way is called conﬂuence. The method is sometimes called structure compensation or structure stabilization, because when conﬂuence is obtained, the slopes of the two dispersion curves where the passbands join are increased from zero to a maximum value.

w2 = 2.5

−7.5 −5 −2.5 0 (a )

f 2.5

7.5

w2 = 2

−7.5 −5 −2.5 0

f 2.5

7.5

(b )

Figure 4.20 Dispersion curves of a structure with resonant coupling. (a) The curves when not at conﬂuence, and (b) the curves at conﬂuence.

117

118

4 Standard Linac Structures w

0 (a )

0 (b )

Figure 4.21 (a) Conﬂuence at zero mode and (b) conﬂuence at π mode.

The steepening effect of the slope at the point of conﬂuence reﬂects the increase in the group velocity associated with the improved power ﬂow along the tuned structure. At ﬁrst it may seem counterintuitive that we can improve the acceleratingmode ﬁeld stability by tuning the structure for conﬂuence, which brings modes closer in frequency. What happens as a result of conﬂuence is that the tuning process improves the power ﬂow, which is directly related to the local mode spacing. The creation of a ﬁnite slope where the passbands join, produces the desired increased mode spacing at the operating point, and usually this procedure produces a symmetrical or nearly symmetrical spacing between the nearest modes. The symmetry produces a ﬁrst-order cancellation of the error-induced contributions from the nearest modes according to perturbation theory. Another important example of resonant coupling is the stabilization of the ﬁelds in the zero-mode DTL cell array, where the accelerating mode of the cells is a TM010 -like mode, and where instead of external coupling cavities, the resonant-coupling elements are internal posts called post couplers, of electrical length λ/4. The post couplers, shown in Fig. 4.22, mount on the cylindrical wall across from the center of each drift tube, with typically one post coupler per cell. The posts extend radially inward, leaving a small gap near the body of the drift tube. Each post may be thought of as being capacitively coupled through the small gap to each end of the opposing drift tube. An unbalanced excitation of the two adjacent cells results in a net excitation of the post. The posts do provide additional coupling between adjacent cells, but unlike the example of the side-coupled linac, the accelerating cells are already strongly coupled without the posts. Nevertheless, the posts function effectively as resonantcoupling elements. A distinctive feature of the post-coupled structure is that the post ﬁelds share the same structure volume as the ﬁelds of the accelerating cell. This sharing of the same cavity volume results in a more complicated interaction between all the different elements than one ﬁnds in

4.15 Resonant Coupling

Post

Drift tube Stem

Figure 4.22 Post-coupled drift-tube linac structure.

a coupled-cavity structure. It is found experimentally that if the azimuthal positions of sequential posts are alternated by 180◦ from one side of the drift tubes to the other, the post-mode passband is widened, resulting in a larger mode spacing near the accelerating mode at conﬂuence, and a better-stabilized structure. The drift tubes are each supported by a stem, which is oriented at 90◦ to the post to reduce the coupling between the posts and stems. Because of their location at the symmetry points of the structure, the posts have little effect on the 0 TM010 accelerating mode, [31] but they capacitively load the 0 TE110 -like mode of each accelerating cell. At conﬂuence the upper frequency of the post-mode passband is tuned to the accelerating-mode frequency, which is the lowest frequency of the TM010 passband. To maintain a high degree of ﬁeld stabilization it is found that one should avoid letting the 0 TE110 -mode frequency from becoming larger than the accelerating-mode frequency [32]. We note that the resonant-coupling elements for the DTL and side-coupled linac function in different ways. The coupling of the DTL accelerating cells is very large, as is evidenced by a large bandwidth when no posts are present. But this large coupling does not promote electromagnetic energy ﬂow for the 0 TM010 accelerating mode. For this mode, the Poynting vector is directed transversely, which means that the electromagnetic-ﬁeld energy cycles between the inside and the outsides of the cavity volume. The ﬂow of energy longitudinally is zero, which means that the axial group velocity is also zero. To produce a large mode spacing and a nonzero axial group velocity in the vicinity of the accelerating mode, the accelerating mode must be perturbed to create a longitudinal component of Poynting vector. This is accomplished very effectively with the post couplers after bringing the TE post mode into conﬂuence with the accelerating mode. The nonzero Poyntingvector longitudinal component, resulting from the electric ﬁeld of the post mode and the magnetic ﬁeld of the accelerating mode, produces a ﬁnite axial group velocity and a dispersion relation with a nonzero slope. The posts provide the mechanism that generates the electromagnetic energy ﬂow, needed to produce a longitudinal component of the Poynting vector. For the CCL, the direct coupling of adjacent accelerating cavities through the beam holes is

119

120

4 Standard Linac Structures

usually very small. The main coupling between adjacent accelerating cavities is produced through the coupling slots and through the coupling cavities. For the side-coupled linac, the longitudinal energy ﬂow is a consequence of providing a bypass for the electromagnetic energy ﬂow through the resonant side cavities. If the direct coupling between the individual coupling elements is very strong, it is more convenient to describe the coupling resonators using the normal modes of an array of coupling elements. The ﬁelds of the normal modes are then distributed throughout the structure. An example of this case is the disk and washer structure. This is a structure with disks that connect to the outer wall, and washers, supported by longitudinal stems from the disks, that are concentric with the beam axis and are mounted halfway between the disks. The accelerating mode is the π TM020 -like mode, where the electric ﬁeld is concentrated in the accelerating gaps that are centered in the same plane as the disks, and with a node in the plane of the washers. The coupling mode is a π TM010 mode that has most of the electric ﬁeld concentrated between the disks at large radius. There is a considerable fringe ﬁeld near the washers at large radii, with a node in the plane of the disks. The electric-ﬁeld patterns of these modes are shown in Fig. 4.23.

(a )

(b)

Figure 4.23 Electric-ﬁeld pattern of disk and washer modes that are degenerate at conﬂuence. (a) The π TM020 -like accelerating mode and (b) the π TM010 -like coupling mode. For a structure with ﬁnite length, the boundary conditions prevent the excitation of the π TM010 mode in the error-free case. It is not easy to look at the detailed ﬁeld distribution in this structure and to see a π /2-like mode pattern, since the entire physical space of the structure contains the superposition of the ﬁelds from both

modes, if both were excited. The coupling modes must have low ﬁelds near the beam axis to avoid undesirable perturbations on the beam. This is generally not considered to be a problem for the post-coupled DTL, or the disk and washer structure. The concept of resonant coupling is an important development in the ﬁeld of RF linacs that has enabled the construction of long standing-wave structures, which is an economical choice for many accelerator applications.

4.16 Accelerating Structures for Superconducting Linacs

Another basic application of resonant coupling is to electromagnetically couple together two cavities that nominally have the same frequency, by introduction of a resonant-coupling element that is coupled to both cavities. As seen by the RF system this effectively creates a single very stable electromagnetic structure, which can be excited by a single RF drive. The basic principle to make this work is to provide a resonantcoupling element, such as a post in a DTL or a coupling cell in a CCL, that is unexcited when driven by the two cavities to be coupled together. Generally, it is necessary to tune the three-resonator assembly to operate in a π/2 mode. With perturbations present an unbalanced net drive to the resonant-coupling element will produce excitation of that element and promotes power ﬂow between the two cavities that compensates for the imbalance. 4.16 Accelerating Structures for Superconducting Linacs

Accelerating structures based on a TEM mode are commonly used for acceleration of low- and medium-velocity beams. We now discuss these TEM mode structures, making use of the excellent review of mediumβ superconducting accelerating structures that has been presented by Delayen [33, 34]. There are two general types of TEM resonant structures, those based on electrical length λ/4, and those based on electrical length λ/2. λ/4 Superconducting Structures

Delayen [33] provides examples of λ/4 resonators. Figure 4.24 shows an example of a superconducting λ/4 resonator with a single center conductor, more generally called a loading element. Delayen classiﬁes the split-ring resonator shown in Fig. 4.25, as an example of a quarter-wave resonator containing two loading elements. The λ//4 resonators have been typically designed for use over a velocity range of β = 0.05 to about β = 0.15. λ/2 Superconducting Structures

Half-wave resonators include both the coaxial geometry, as shown in Fig. 4.26, and the spoke geometry as shown in Figs. 4.27 and 4.28. Spoke resonators with multiple spokes that are sequentially rotated by 90◦ as shown in Fig. 4.28 are also called crossbar H-type resonators. Half-wave resonators have been designed for use from near β = 0.1 to near to β = 0.5.

121

122

4 Standard Linac Structures

Figure 4.24 University of Washington, 150-MHz β = 0.2 coaxial quarter-wave resonator (courtesy of J. R. Delayen, Ref. 33).

Figure 4.25 Example of a split-ring resonator concept for a 115-MHz β = 0.13 resonator with two loading elements for a rare-isotope accelerator (courtesy of J. R. Delayen, Ref. 33).

TM Superconducting Structures

The Alvarez DTL and the CCL may be considered as examples of normal-conducting coupled multicell TM010 cavities. Examples of coupled superconducting multicell TM010 cavities are shown in Fig. 4.29. These are also called elliptical cavities. Cavities of this type are designed for use over a velocity range near β = 0.5 to β = 1.0. Some important differences between the transverse magnetic (TM) and TEM resonators include (1) a signiﬁcant size difference, (2) a difference in Lorentz detuning and microphonics, and (3) a difference in multipacting. The inner diameter of TM resonators is about 0.9λ, whereas λ/2 structures have

4.16 Accelerating Structures for Superconducting Linacs

Figure 4.26 350-MHz β = 0.12 coaxial half-wave resonator with a single loading element (courtesy of J. R. Delayen, Ref. 33).

Figure 4.27 850-MHz, β = 0.28 spoke resonator (courtesy of J. R. Delayen, Ref. 33).

an inner diameter of nearly a factor of 2 less. The λ/2 resonator is smaller and lighter. For the same transverse size it would have half the frequency, and for the same length would have fewer cells, and would have a larger velocity acceptance. The λ/2 resonators are less sensitive to microphonics and Lorentz force detuning than TM resonators of the same frequency and same β. Multipacting has limited the performance of TM resonators, but is nearly eliminated in the TM elliptical resonators. Furthermore simulation programs

123

124

4 Standard Linac Structures

(a )

(b )

(c )

Figure 4.28 Spoke cavities with multiple loading elements. (a) An 850-MHz, β = 0.28 double spoke concept. (b) A 345-MHz, β = 0.4 double spoke concept. (c) A 700-MHz, β = 0.2 eight-spoke concept (courtesy of J. R. Delayen. Ref. 33).

4.16 Accelerating Structures for Superconducting Linacs

(a )

(b )

Figure 4.29 TM010 cavities for β < 1. (a) 805-MHz, β = 0.82, six-cell cavity, (b) 805-MHz, β = 0.62, six-cell cavity (courtesy of J. R. Delayen, Ref. 33).

have been developed that allow the designer to avoid multipacting. But λ/2 resonators almost always exhibit multipacting, and no such simulation codes are available that are reliable enough to guarantee no multipacting. Nevertheless, multipacting almost always occurs in TEM resonators, but no cases are known for which multipacting prevented operation of the cavity.

RF Properties and Scaling Laws for TM and λ/2 Superconducting Structures

Delayen [33] has compiled data and calculations for TM and λ/2 superconducting resonators, shown in Figs. 4.30 to 4.35. Figure 4.30 shows the peak surface electric ﬁeld to accelerating-gradient ratio versus the design velocity β. The peak surface electric ﬁeld is important because it can be a source of ﬁeld emission that absorbs RF power. This electron loading is an important practical limitation for achieving high gradients. In both TM and TEM structures the peak surface electric ﬁeld occurs near the iris. For TM structures the ratio is about 2 at β = 1 and increases as β decreases. For TEM structures the geometry of the loading element near the beam aperture can be optimized to obtain a nearly constant peak surface ﬁeld value around its circumference that results in a ratio of about 3.3 independent of β. This optimization can be made with minimal effect on the peak surface magnetic ﬁeld. The peak surface magnetic ﬁeld to accelerating-gradient ratio versus design velocity β is shown in Fig. 4.31. The peak surface magnetic ﬁeld is important

125

4 Standard Linac Structures 6 5

Ep /Ea

×+

+ ×

2 1 0 0

0.1

0.2

0.3

0.4

0.5

0.6

0.7

0.8

0.9

Figure 4.30 Peak surface electric ﬁeld to accelerating-gradient ratio versus design velocity β compiled by Delayen. Points joined by lines are for TM resonators, while the squares not connected by lines are for λ/2 resonators (courtesy of J. R. Delayen, Ref. 33).

14 12

Bp /Ea (mt /(MV/ m))

126

10 +

2 0 0

0.1

0.2

0.3

0.4

0.5 0.6

0.7

0.8 0.9

Figure 4.31 Peak surface magnetic ﬁeld to accelerating-gradient ratio versus design velocity β, compiled by Delayen. Points joined by lines are for TM resonators, while the squares not connected by lines are for λ/2 resonators (courtesy of J. R. Delayen, Ref. 33).

because too large a value will ultimately result in a quench that destroys the superconductivity. For TM structures the peak surface magnetic ﬁeld occurs at the equator. The ratio typically increases with decreasing β from about 4 to 4.5 mT at 1 MV/m at β = 1 to 6 to 8 mT at 1 MV/m for β = 0.5. For TEM structures the maximum occurs where the loading element intersects the outer wall, and can be optimized to achieve nearly constant peak surface magneticﬁeld values around the circumference of the loading element. Delayen quotes achievable values of 7 to 8 mT, independent of β. The quality factor Q = ωU/P is proportional to the ratio of stored energy to power dissipation per cycle. The geometry factor, deﬁned as G = QRs , where

4.16 Accelerating Structures for Superconducting Linacs

Rs is the RF surface resistance, is a very useful parameter for comparing different cavity shapes. This is because QRs depends on the cavity’s shape, but not its size or wall material. The quality factor Q does depend on the cavity size because it depends on Rs which depends on the resonant frequency, and the cavity size also depends on frequency. For TM structures the stored energy is approximately proportional to the cavity volume, which means that since the length depends on β, the stored energy will be approximately proportional to β. The power dissipation is approximately proportional to surface area, and for a typical TM structure like a low-β elliptical structure, the surface area is dominated by the end-wall surface area and is more weakly dependent on β. Neglecting this weak β dependence, QRs should be approximately linearly dependent on β for a TM structure, and Delayen gives QRs ≈ 275β( )

(4.56)

independent of the number of cells. In Ref. 34, Delayen uses a transmission line model to obtain a scaling formula for λ/2 structures, QRs ≈ 270β( )

(4.57)

Figure 4.32 shows the results for QRs as a function of β.

Shunt Impedance for TM and λ/2 Superconducting Structures

Following Delayen [33] we use the deﬁnition of shunt impedance Rsh = (V0 T)2 /P where V0 T is the maximum voltage gained by a particle as

300 250 ×

QRs (Ω)

200 ×

150 100 50 0 0

0.1

0.2

0.3

0.4

0.5

0.6 0.7

0.8

0.9

Figure 4.32 Geometry factor QRs versus design velocity β, compiled by Delayen. Points joined by lines are for TM resonators, while the squares not connected by lines are for λ/2 resonators. The dashed line shows

the scaling formula for TM resonators, and the dashed-dotted line shows the scaling formula for λ/2 resonators (courtesy of J. R. Delayen, Ref. 33).

127

128

4 Standard Linac Structures

a function of input velocity at the crest of the RF phase. The shunt impedance is a measure of the voltage gained by a particle of optimum velocity to the power dissipation averaged over an RF cycle. It depends on the wall material. There are two shunt-impedance-related quantities that are independent of wall material, Rsh /Q and Rsh Rs . These parameters also depend only on the shape and not on size or wall material. As we shall see, for TM structures these quantities are proportional to the number of cells and are quoted per cell. For TEM structures they are proportional to the number of loading elements and are quoted per loading element. Suppose we begin with an inﬁnitely long periodic structure operating in a π mode, and we wish to obtain from this a ﬁnite length cavity by inserting two conducting end walls that are perpendicular to the beam axis. These end walls would properly terminate the ﬁnite length structure if the ﬁelds are unchanged as a result of inserting the plates. If the end walls contained no beam aperture, the way to terminate the cavities without distorting the ﬁelds would be to end with half-cells that intersect the outer wall at the equator. However, perturbations such as those caused by the beam aperture can complicate matters, depending on the magnitude of the perturbation. Delayen argues that the importance of the perturbation to the ﬁelds from the beam aperture on the end walls depends on the strength of the cell-to-cell coupling. For the case of the TEM structure where the outer cell-to-cell openings and the corresponding magnetic intercell coupling are very large, typically about 20% or more, the ﬁeld proﬁle is very insensitive to perturbations such as the end-wall beam aperture. Since the ﬁelds are approximately unchanged by such a perturbation, the proper endwall placement for the TEM structures is still nearly that which corresponds to half-cell terminations. However, for the TM structures that use electric cell-to-cell coupling through the beam holes, the coupling is small, typically about 2%. In this case the end-cell perturbation caused by the beam aperture has a large effect on the ﬁelds. One ﬁnds that the resulting termination for the TM structures is no longer a half-cell, but in this case the end cells are approximately the same as the interior cells, which are approximately full cells. Figure 4.33 shows Rsh /Q per cell or loading element versus design velocity β. Delayen obtains a scaling formula for Rsh /Q using the following arguments. For a single cell of a TM structure, Rsh (V0 T)2 (E0 TL)2 = = Q ωU ωU

(4.58)

The stored energy is approximately proportional to volume and therefore proportional to β. The cell length L is also proportional to β. The simplest argument is that Rsh /Q is proportional to β. But, Delayen ﬁnds that this argument is not correct. One needs to take into account a decrease in the on-axis ﬁeld E0 with decreasing cell length, relative to the electric ﬁeld at an

4.16 Accelerating Structures for Superconducting Linacs 350

Rsh /Q per cell (Ω)

300 250 200 150 100 50 0 0

0.1

0.2

0.3

0.4

0.5 b

Figure 4.33 Rsh /Q per cell or loading element versus design velocity β, compiled by Delayen. Points joined by lines are for TM resonators, while the squares not connected by lines are for λ/2 resonators. The dashed

0.6

0.7

0.8

0.9

line shows the scaling formula for TM resonators, and the dashed-dotted line shows the scaling formula for λ/2 resonators (courtesy of J. R. Delayen, Ref. 33).

approximately ﬁxed iris radius. Rsh /Q decreases more rapidly than linearly, and the TM data are best ﬁt by a quadratic dependence, Rsh ≈ 120β 2 ( ) Q

(4.59)

per cell. For λ/2 structures Delayen’s transmission line model gives Rsh ≈ 205( ) Q

(4.60)

per loading element. Figure 4.34 shows the product of shunt impedance Rsh times RF surface resistance Rs per cell or loading element versus design velocity β. By multiplying the scaling formulas already obtained for Rsh /Q and QRs , we have per cell for TM or per loading element for λ/2 Rsh Rs ≈ 33 × 103 β 3 ( )3 , Rsh Rs ≈ 55 × 103 β( )3 ,

TM λ/2

(4.61)

Stored Energy for TM and λ/2 Superconducting Structures

Figure 4.35 shows the electromagnetic stored energy per cell or per loading element at Eacc = 1 MV/m and at 500 MHz. Delayen also obtains scaling

129

4 Standard Linac Structures 45,000 40,000

R /Q QRs (Ω2)

35,000 30,000 25,000 20,000 15,000 10,000 5000 0 0

0.1 0.2

0.3 0.4 0.5

0.6 0.7 0.8 0.9

Figure 4.34 Product of shunt impedance Rsh times RF surface resistance Rs per cell or loading element versus design velocity β, compiled by Delayen. Points joined by lines are for TM resonators, while the squares not

U /E 2 at 1 MVm, 500 MHZ (mJ)

130

connected by lines are for λ/2 resonators. The dashed lines show the scaling formulas for the two types of resonators (courtesy of J. R. Delayen, Ref. 33).

300 250 200 150 100 50 0 0

0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9

Figure 4.35 Stored energy per cell or loading element versus design velocity β at 1 MV/m for 500-MHz structures, compiled by Delayen [33]. Points joined by lines are for TM resonators, while the squares not

connected by lines are for λ/2 resonators. The dashed line is the scaling law developed by Delayen for λ/2 resonators (courtesy of J. R. Delayen, Ref. 33).

formulas for these which can be written as

3 500 2 U(J) ≈ 0.2Eacc (MV/m) , TM f (MHz)

3 500 2 (MV/m) , λ/2 U(J) ≈ 0.2β 2 Eacc f (MHz)

(4.62)

Problems

Scaling Formulas for λ/4 Superconducting Structures

Delayen [35] obtains scaling formulas for λ/4 resonators from his scaling formulas for λ/2 resonators given above by using a simple model where a λ/4 resonator is assumed to be a λ/2 resonator cut in two equal parts. His resulting formulas are QRs ( ) ≈ 270β

(4.63)

independent of the number of loading elements. Rsh /Q ≈ 410( )

(4.64)

per loading element. Rsh Rs ≈ 111β( 2 )

(4.65)

per loading element.

Problems

4.1. Plot the transit-time factor for a π-mode two-gap cavity versus β/βs in the range 0.5 < β/βs < 3. Assume a square-wave ﬁeld distribution in the gaps, and assume g/βs λ = 1/4. Is the transit-time factor maximum when β = βs ? 4.2. Suppose you want to accelerate deuterons synchronously in a DTL that has been designed and operated for protons in a 1βλ mode. Assume the deuteron and proton masses are related by mD = 2mP , and assume the synchronous phase φ is unchanged. Calculate the required value of E0 T for the following two cases. (a) Use a 1βλ mode for the deuteron operation. If E0 was already as high as you could reliably operate, is this solution practical? (b) Use a 2βλ mode for the deuteron operation. (c) Suppose the ratio of gap to cell length is 1/4. Calculate the transit-time factor for the 1βλ and 2βλ cases using the simple formula T = (sin x)/x, where x = πg/βλ. What is the ratio of E0 for the 2βλ deuteron case to that for proton operation? 4.3. Plot the dispersion relation (mode frequency q /2π versus phase advance per cavity πq/N) for N + 1 = 5 coupled cavities with uncoupled cavity frequency ω0 /2π = 400 MHz and coupling constant k = 0.05. 4.4. Consider N + 1 coupled cavities with an uncoupled cavity frequency ω0 and coupling constant k 1. (a) Write the expressions for the frequencies of the π mode and the mode closest to the π mode. (b) Calculate the fractional difference between the two mode frequencies. Obtain an approximation for this result when N 1. How does this depend on k and N? (c) If k = 0.02 and N + 1 = 30, calculate the fractional difference

131

132

4 Standard Linac Structures

4.5. 4.6.

4.7.

4.8.

between the frequencies of the two modes. Compare this with the width of the π mode assuming the Q value for the π mode is 20,000. Are the two modes adequately separated? Repeat parts (a), (b), and (c) of Problem 4.4. for the π/2 mode of N + 1 coupled oscillators. A CCL module that operates in a π/2 mode has 135 excited (accelerating) cells and 134 nominally unexcited coupling cells, and is driven in the central excited cell by a single klystron. The parameters are Qa = 20,000 for the accelerating cells, Qc = 2000 for the coupling cells, the frequency is 1280 MHz, and the coupling constant is k = 0.05. Assume the beam power is negligible. (a) Calculate the power-ﬂow droop in the ﬁeld excitation of the end cavities and sketch the ﬁeld excitation along the module relative to the drive cell (for the accelerating cells only). (b) Calculate the power-ﬂow phase shift in the end cavities and sketch the power-ﬂow phase shift along the module (for the accelerating cells only), assuming the stopband is open by δω/2π = 100 kHz. (c) Calculate the ﬁeld excitation of the two coupling cells adjacent to the drive cell relative to the drive cell amplitude. Sketch the excitation of the coupling cells along the module relative to the drive cell excitation. Resonant-frequency errors in a coupled-cavity system that are small compared with the spacing of the normal modes might not seem to be serious, because the modes still remain well separated. But even small cavity-frequency errors can cause nonuniformity of the ﬁeld distribution that may be unacceptable for accelerators. Consider the weakly coupled three-cavity system operating in π mode, with an end-cell frequency error δω0 /ω0 = 10−3 (2δω0 is the frequency difference of the two end cells), a middle-cell frequency error δω1 = 0 (the middle cell has same frequency as the average of the end cells: ω0 ), and coupling constant k = 0.01. (a) Ignoring the frequency error, calculate the frequency spacing of the three normal modes relative to ω0 /2π. Is the cavity-frequency error small compared with the frequency spacing of the modes? (b) Calculate the perturbed π-mode ﬁeld vector for this case. Describing the perturbed distribution as a linear ﬁeld tilt, what is the fractional end-to-end ﬁeld tilt? (c) Calculate the perturbed π/2-mode ﬁeld vector. If only the end cells are used for acceleration, would the π/2 mode give a ﬁeld distribution that is more uniform in the presence of the same cavity-frequency errors? We consider the effects of power losses in the coupled three-cavity system. Assume there are no frequency errors (δω0 = δω1 = 0), a coupling constant k = 0.01, and Q = 104 for each mode. Assume the RF drive is located in one end cell. (a) Calculate the ﬁeld vector for the zero, π/2, and π modes. (b) Calculate the power-ﬂow phase shift in the end cell without the drive for the zero, π/2, and π modes. Calculate the π/2-mode power-ﬂow droop in the end cell without the drive. (c) Assume the loaded Q is reduced to Q = 103 , and repeat part (b).

References

References 1 Bohne, D., Proc. 1976 Linear Accel.

3 4 5

9 10

Conf., September 14–17, (1976), Chalk River, Ontario, Canada, AECL-5677, p. 2. For a review of superconducting heavy-ion linacs, see Bollinger, L.M., Ann. Rev. Nucl. Part. Sci. 36, 375 (1986); or Bollinger, L.M. Proc. 1992 Linear Accel. Conf., August 24–28, 1992, Ottawa, Ontario, Canada, AECL-10728, p. 13. Wider¨oe, R., Arch. Electrotech. 21, 387 (1928). Sloan, D.H. and Lawrence, E.O., Phys. Rev. 38, 2021–2032 (1931). Kaspar, K., Proc. 1976 Linear Accel. Conf., September 14–17, 1976, Chalk River, Ontario, Canada, AECL-5677, p. 73. Ratzinger, U., Proc. 1988 Linear Accel. Conf., 1988, CEBAF∼Rep. 89–001, p. 185. Alvarez, L.W., Phys. Rev. 70, 799 (1946); Alvarez, L.W. et al., Rev. Sci. Instrum. 26, 111–133 (1955). Calculated by Billen, J. from the electromagnetic-ﬁeld-solver program SUPERFISH. Halbach, K. and Holsinger, R.F., Part. Accel. 7, (No. (4)), 213–222 (1976). Prior to 1995, the linac at Los Alamos was called LAMPF for Los Alamos Meson Physics Facility. The new name is LANSCE, which means Los Alamos Neutron Science Center. Warner, D.J., Proc. 1988 Linear Accel. Conf., 1988, CEBAF Rep. 89–001, p. 109. Swenson, D.A. and Stovall, J., Los Alamos Internal Report, MP-3-19, January, 1958; Boicourt, G.P. AIP Conf. Proc. 177: Linear Accelerator and Beam Optics Codes, ed. Eminhizer, C.R., American Institute of Physics, New York, 1988, pp. 1–21. Nagel, D.E. Knapp, E.A. and Knapp, B.C., Rev. Sci. Instrum. 38, 1583–1587 (1967). Strictly speaking, in an accelerator the end oscillator would correspond to half-length cells with conducting

18 19

planes inserted at the center plane of symmetry and with no beam holes. Mathews, Jon. and Walker, R.L., Mathematical Methods of Physics, 2nd ed., W. A. Benjamin, Inc., 1970, p. 286; Wangler, T.P. Calculations for Three Coupled Oscillators-Including Eigenvalue Problem, Perturbation Theory and Losses, Los Alamos Group AT-1 Report AT-1 : 84–299, August 29, 1984. Note that the second-order terms appear in the eigenfrequency and in the ﬁrst and third elements of the eigenvector. Nagel, D.A., unpublished Los Alamos technical memo P-11/DEN-2, August 21, 1963. Knapp, E.A., Proc. 1964 Linac Conf., MURA, Stoughton, Wis. 1964, p. 35. Schriber, S.O., Heighway, E.A. and Funk, L.W., Proc. 1972 Linear Accel. Conf., Los Alamos Laboratory Report LA-5115, p. 140; Schriber, S.O., Proc. 1976 Linear Accel. Conf., September 14–17, 1976, Chalk River, Ontario, Canada, AECL-5677, p. 405. Knapp, E.A., Knapp, B.C. and Potter, J.M., Rev. Sci. Instrum. 39, 979–991 (1968). The annular coupled structure is discussed in Radio-Frequency Structure Development for the Los Alamos/NBS Racetrack Microtron, compiled and edited by Jameson, R.A., Stokes, R.H., Stovall, J.E. and Taylor, L.S., Los Alamos National Laboratory Report LA-UR-83-95, January 21, 1983. Andreev, V.G. et al., Proc. 1972 Linear Accel. Conf., Los Alamos, N.M., Los Alamos Scientiﬁc Laboratory Report LA-5115, p. 114. Schriber, S.O., Fitting of an Ordered Set of Mode Frequencies, Chalk River Nuclear Laboratories Report AECL-3669, Chalk River, Ontario, October, 1970. Knapp, E.A., High energy structures, in Linear Accelerators, Lapostolle, P.M. and Septier, A.L., North-Holland Publishing Company, Amsterdam, and John Wiley & Sons, New York, 1970, p. 615.

133

134

4 Standard Linac Structures 25 Dome, G. Review and survey of

27 28

accelerating structures, in Linear Accelerators, eds. Lapostolle, P.M. and Septier, A.L., North-Holland Publishing Company, Amsterdam, and John Wiley & Sons, New York, 1970, p. 615. Lapostolle, P.M., Proton Linear Accelerators: A Theoretical and Historical Introduction, Los Alamos Report, LA-11601-MS, July, 1989, p. 64. Bethe, H.A., Phys. Rev. 66, 163 (1944). See Eqn. (44) from Gao, J., Nucl. Instrum. Methods A311, 437–443 (1992). In the case of TM010 cavities that are electrically coupled through a hole on the axis, the symmetry of the unperturbed ﬁeld on the axis, prohibits the excitation of the TE11 waveguide mode in the hole. For the case treated in Section 3.11, we assumed that the waveguide mode with the smallest attenuation factor was the TM01 mode.

30 Spalek, G., Los Alamos Group AT-1

memorandum, AT-1 : 91–124, March 26, 1991. We adopt a nomenclature for the operating mode, denoted by 0 TM010 , where the superscript gives the speciﬁc mode of the passband corresponding to normal mode spectrum of the TM010 cavity mode. Lapostolle, P.M., Proton Linear Accelerators: A Theoretical and Historical Introduction, Los Alamos Report, LA-11601-MS, July 1989, pp. 64–65. Delayen, J.R., Medium-β Superconducting Accelerating Structures, Proc. of the 10th Superconducting RF Workshop, Tsukuba, Japan, 2001. Delayen, J.R., Design of low velocity superconducting accelerating structures using quarter-wavelength resonant lines, Nucl. Instrum. Methods A259, 341–357 (1987). Delayen, J.R. private communication.

135

5 Microwave Topics for Linacs

In this chapter, we present some general topics from the ﬁeld of microwaves and microwave systems that are important for the linac application. We will begin with descriptions of the resonant cavity that constitutes the most fundamental element in an RF linac. This is followed by discussions of coupling of waveguides to cavities, including a waveguide model of an iriscoupled cavity that illustrates the wave-interference effects in the drive line, and the build up of the cavity ﬁelds as a result of reﬂections of waves inside the cavity. This is followed by elementary discussions of the klystron, multipacting, ﬁeld emission, and electric breakdown. Finally, we present some special accelerator-physics topics including the Boltzmann–Ehrenfest theorem of adiabatic invariants of an oscillator, the Slater perturbation theorem used as the basis of cavity-ﬁeld measurements, the quasistatic approximation for solving Maxwell’s equations, and the Panofsky–Wenzel theorem describing transverse deﬂections of particles from RF ﬁelds. 5.1 Shunt Resonant Circuit Model

A parallel resonant circuit driven by a current generator is the simplest model for describing a single mode of an accelerating cavity. This is because a cavity behaves like a parallel resonant circuit, when viewed from the plane of a detuned short circuit in the input waveguide, as is discussed in Section 5.4. We consider a circuit, shown in Fig. 5.1, with an inductance L, a capacitance C, and circuit shunt resistance R, driven by a simple current generator that supplies a current I(t) at frequency ω. In Section 5.4, we will improve the model to include the input waveguide and the waveguide-to-cavity coupling mechanism. The driving current from the generator produces a voltage V(t) across the circuit that can be identiﬁed with the axial voltage in the cavity. The total current is V dt V I(t) = CV˙ + + (5.1) L R RF Linear Accelerators. 2nd, completely revised and enlarged edition. Thomas P. Wangler Copyright  2008 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-40680-7

136

5 Microwave Topics for Linacs V (t )

I (t )

Figure 5.1

Circuit model of cavity resonator.

and by differentiation I˙ ω0 V˙ (5.2) = V¨ + + ω02 V C Q √ where ω0 = 1/ LC is the resonant frequency. The stored energy U = CV02 /2, the dissipated power P = V02 /2R, where V0 is the peak voltage, and the quality factor Q = ω0 U/P = ω0 RC. Equation (5.2) is the equation for a damped, driven oscillator. To derive the steady-state solution, we express I(t) = I0 e jωt and V(t) = V0 e j(ωt+φ) , where I0 and V0 are real amplitudes, and φ is the phase of the resonator voltage relative to the driving current. Substituting these expressions into Eq. (5.2), we obtain the steady-state solution, RI0 e j(ωt+φ) V(t) = 1 + y2 where φ = − tan−1 y, and y is called the detuning factor, deﬁned by ω ω0 y≡Q − ω0 ω

(5.3)

(5.4)

We deﬁne δω = ω − ω0 and a useful approximation when Q 1 is y∼ = 2Qδω/ω0 . From Eq. (5.3) the circuit shunt impedance [1] is identiﬁed as Z=

V Re jφ = I 1 + y2

(5.5)

On resonance, y = φ = 0, and from Eqs. (5.3) and (5.5) we can see that the cavity voltage and impedance increase with the circuit shunt resistance R. Off resonance, if ω < ω0 , φ is positive, the resonator voltage leads the driving current. By convention, the bandwidth of an oscillator is deﬁned as the frequency difference ω between the two points on each side of the peak√of the resonance curve, where the voltage is lower than the peak value by 2, or where y ∼ = 2Qδω/ω0 = 1. These two points are called the half-power points, and the resonator bandwidth is deﬁned as ω0 (5.6) ω ≡ 2δω = Q

5.2 Theory of Resonant Cavities

In a real cavity, this result provides a useful way to measure the Q; measurement of the resonance curve allows the determination of both ω0 and ω, and Q = ω0 /ω. The general solution of Eq. (5.2) can be written as RI0 e j(ωt+φ) V(t) = e−ω0 t/2Q [c1 e jω1 t + c2 e−jω1 t ] + 1 + y2

(5.7)

where c1and c2 are constants that depend on the initial conditions, and ω1 ≡ ω0 1 − 1/(2Q)2 . The resonator time constant is deﬁned as τ ≡ 2Q/ω0 . If we choose c1 = c2 = −RI0 /2, assume Q 1, and take the real part of Eq. (5.7), we ﬁnd V(t) = RI0 (1 − e−t/τ ) cos(ω0 t)

(5.8)

which shows the time dependence of the resonator voltage, when the driving current is turned on at t = 0 with an amplitude I0 . After an interval of several time constants, the voltage is nearly the steady-state value of V(t) = RI0 cos(ω0 t). Returning to Eq. (5.7), suppose that at t = 0, the driving current is suddenly turned off with V(0) = RI0 cos(δ), where δ is an arbitrary phase angle. In Eq. (5.7) the last term is zero for t > 0, and the initial condition is satisﬁed by choosing c1 = RI0 e jδ /2 and c2 = RI0 e−jδ /2. The solution is V(t) = RI0 e−t/τ cos(ω0 t + δ)

(5.9)

that describes the decay of the voltage, when the generator is turned off at t = 0.

5.2 Theory of Resonant Cavities

Another description of an accelerating cavity is based directly on Maxwell’s equations, following the approach of Condon [2]. We consider an ideal cavity with zero power losses, and no charge sources in the volume. As described in Section 1.8, the ﬁelds can be derived from a vector potential A(r, t) that satisﬁes the wave equation ∇ 2 A(r, t) −

1 ∂2 A(r, t) = −µ0 J(r, t) c2 ∂t2

(5.10)

The ideal loss-free cavity will have an inﬁnite number of eigenmodes that satisfy Eq. (5.10) with J(r, t) = 0, so that each mode n has a harmonic time dependence at a frequency ωn, and a characteristic vector potential An (r, t) = An (r)e jωn t that satisﬁes the Helmholtz equation ∇ 2 An (r) + (ωn /c)2 = 0. An arbitrary ﬁeld within the cavity that satisﬁes the boundary conditions can be expanded in

137

138

5 Microwave Topics for Linacs

terms of the complete set of eigenmodes, A(r, t) = qn (t)An (r)

(5.11)

√ The An (r) satisfy Am · An dv 0 ωm ωn , where δmn is the =2 2δmn Um Un /ε Kronecker δ, and Un = (ε0 /2) En (r) dv = (ε0 ωn2 /2) A2n (r) dv is the stored energy for mode n. Suppose that an external generator excites the cavity through some coupling mechanism, [3] that establishes a current-density distribution J(r, t) inside the cavity. To determine the effect of the driving current on amplitude qn , we substitute Eq. (5.11) into Eq. (5.10), multiply by A(r), and integrate to obtain J(r, t) · An (r) dv 1 (5.12) q¨ n + ωn2 qn = ε0 A2n dv The effect of ohmic losses in the cavity walls introduces a damping term that depends on the quality factor for the mode, Qn = ωn Un /Pn , where Pn is the power loss averaged over a cycle [4]. Then, J(r, t) · An (r) dv ωn 1 q˙ n + ωn2 qn = q¨ n + (5.13) Qn ε0 A2n dv which is the equation for a damped, driven oscillator. The integral on the right side of Eq. (5.13) determines the effectiveness of the current distribution for exciting the nth mode. Physically, Eq. (5.13) shows that the driving term on the right is proportional to the overlap of the current density produced by the coupling device with the vector potential for the mode. The differential equation, Eq. (5.13), describing the dynamical performance of the cavity, has the same form as the differential equation Eq. (5.2) that describes the shunt resonant circuit model of the cavity. Thus, a physical interpretation of Eq. (5.13) can be made, where the right side of Eq. (5.13) is interpreted as a quantity proportional to the effective current that drives the mode, and the coefﬁcient qn is proportional to the effective voltage for the mode.

5.3 Coupling to Cavities

The most common methods for coupling electromagnetic energy into or out of a cavity are shown in Fig. 5.2 as (1) a magnetic-coupling loop at the end of a coaxial transmission line, (2) a hole or iris in a cavity wall to which a waveguide is connected, (3) an electric-coupling probe or antenna, using an open-ended

5.4 Equivalent Circuit for a Resonant-Cavity System

center conductor of a coaxial transmission line. The loop-coupling method can be analyzed in terms of Eq. (5.13). Suppose we assume that the driving current I(t) is carried by a loop coupler with a single-turn wire loop. Then, J(r, t) dv = I(t)d, where d is an element of length along the wire. The right side of Eq. (5.13) contains an integral over the wire, which is J(r, t) · An (r) dv = I(t) An (r) · d (5.14) where An (r) is the vector potential of the unperturbed mode in the wire. Using Stoke’s theorem (5.15) An · d = ∇ × An · dS = n where dS is the area element in the plane of the loop and n is the magnetic ﬂux for the mode that threads the loop. Substituting the results of Eqs. (5.14) and (5.15) into Eq. 5.13, we ﬁnd a simple expression for the steady-state amplitude, |qn | =

Qn I n 2Un

(5.16)

The quantities Qn , n , and Un are all constants that are properties of the mode and the coupler geometry. Another common method of coupling is the electric probe or antenna. The probe may consist of an open center conductor of a transmission line. The general treatment, using the methods of Section 5.2, is complicated by the need to include a scalar potential to describe the ﬁelds produced by the charges at the end of the probe, as discussed by Condon. Physically, this coupling method may be pictured as a conduction current I on the center conductor that ﬂows as displacement current from the end of the conductor. Based on this physical picture, one might expect that the steady-state result for an antenna coupler, corresponding to the result of Eq. (5.16) for a coupling loop, should have a form such as qn ∝ Qn I E,n /cUn , where E,n = En · dS is the electric ﬂux from mode n that is intercepted on the surface of the probe.

5.4 Equivalent Circuit for a Resonant-Cavity System

While the direct application of Maxwell’s equations is useful for obtaining detailed information about the resonant frequencies and ﬁeld distributions of real cavities, the method is cumbersome for describing the complete system consisting of an RF generator, a transmission line or waveguide to transport the electromagnetic wave from the generator to the cavity, and a coupling mechanism that couples electromagnetic energy between the guide and the cavity. Instead, we use an equivalent ac-circuit model similar to that of Section 5.1 to describe the steady-state behavior of the system. As before,

139

140

5 Microwave Topics for Linacs

we represent a cavity by a shunt resonant circuit with components L, C, and R. The accelerator shunt impedance, deﬁned in Section 2.5, is related to R by rs = 2R. For the external circuit we will introduce a matched current generator, a coupling mechanism (see Fig. 5.2), a waveguide with characteristic impedance Z0 , and a circulator, as shown in Fig. 5.3. The coupling mechanism and the waveguide are represented by a transformer with a turns ratio of 1 : n. The external circuit with a circulator or equivalent isolator transmits forward waves going into the cavity but absorbs all backward waves in the matched load Z0 . In general, the cavity does not present an exactly matched load to the input waveguide; the input waveguide between the circulator and the cavity contains a standing wave with voltage minima (and maxima) separated by λ/2, and whose locations vary as the drive frequency is varied. The end of the input waveguide in this model is deﬁned to correspond to the plane of a detuned short circuit in the waveguide, from which the cavity appears as a shunt resonant circuit. Figure 5.3a shows generator, circulator, matched load, power coupler, and cavity. We denote quantities in the external circuit with a primed symbol, and in the resonator circuit with no primes. The impedance Zc of the shunt resonant circuit is given by Eq. (5.5). The resonator voltage V is given by V = Zc i, where i is the driving current, transformed into the resonator circuit. From the transformer coupling, we write V = nV , i = i /n

(5.17)

and Zc =

V V = n2 = n2 Zc i i

(5.18)

Cylindrical cavity Coaxial line

Cylindrical cavity Waveguide

Cylindrical cavity

Antenna probe

Figure 5.2

Methods of coupling to cavities.

5.4 Equivalent Circuit for a Resonant-Cavity System Matched load Cavity

Power coupler

Circulator RF generator ( a)

I′

Waveguide

V′

i′

1:n

Matched generator

Power coupler

Cavity

(b)

Figure 5.3

(a) Block diagram of RF system components and (b) the equivalent circuit.

where Zc is the resonator impedance transformed into the generator circuit. Then, Zc =

Zc 1 = n2 jωCn2 + n2 /jωL + n2 /R

(5.19)

From Eq. (5.19) we identify the transformed components as C = n2 C, L = L/n2 , and R = R/n2 , and the circuit as transformed into the primary circuit can be represented as shown in Fig. 5.4. The resonator stored energy is U = n2 CV 2 /2, the average power dissipated in the resonator is Pc = n2 V 2 /2R, and the power dissipated in the external load is Pex = V 2 /2Z0 [5]. The quality factor, unloaded by the external resistance, is called the unloaded Q, and is written Q0 =

I′

ω0 U = ω0 RC Pc

(5.20)

V ′, i ′

Zc /n 2 = bZ0Zc /R

Figure 5.4 Equivalent circuit transformed into the generator circuit.

141

142

5 Microwave Topics for Linacs

The quality factor associated with the external load is called the external Q, and is written Qex =

ω0 U = ω0 n2 Z0 C Pex

(5.21)

The average power dissipated is P = Pc + Pex and the loaded Q is deﬁned as QL =

ω0 U P

(5.22)

From Eqs. (5.20) to (5.22), it can be shown that the Q values are related by 1 1 1 = + QL Q0 Qex

(5.23)

In the circuit problem, the coupling between the generator and resonator circuits is determined by the transformer–turns ratio n that enters into the expression for Qex in Eq. (5.21). For a cavity coupled to a waveguide, it is conventional to deﬁne a general parameter that is a measure of the waveguide-to-cavity coupling strength, known as the parameter β that is deﬁned by [6] β≡

Q0 Pex = Pc Qex

(5.24)

When β < 1 the waveguide and cavity are said to be undercoupled. When β > 1, the waveguide and cavity are said to be overcoupled. When β = 1, the waveguide and cavity are said to be critically coupled, and we have Qex = Q0 and QL = Q0 /2. Substituting Eq. (5.24) into Eq. (5.23), we ﬁnd QL =

Q0 1+β

(5.25)

Substituting Eqs. (5.20) and (5.21) into Eq. (5.24), we ﬁnd that for the equivalent circuit, the relationship between n and β is β = R/n2 Z0 . Thus, larger n implies smaller β or smaller coupling strength, because for larger n, the external load resistance transforms into the resonator as a larger shunt resistance, compared with the resonator shunt resistance, and produces a smaller loading of the QL . It is convenient to express the circuit model results in terms of the waveguide-to-cavity coupling parameter β, instead of the transformer–turns ratio n. We ﬁnd R = βZ0 , C = RC/βZ0 , and L = βZ0 L/R. It is also convenient to associate the forward power from the generator with the available power to a matched load in the equivalent circuit. The available power to a matched load corresponds to setting β = 1 and driving on resonance, so that Zc = R. Then, referring to Fig. 5.4, the matched load impedance is Z0 , the peak current ﬂowing through the matched load impedance is I /2, and the peak voltage across the load is I Z0 /2. The average available power to the load

5.4 Equivalent Circuit for a Resonant-Cavity System

nV ′

I = l ′/n

Z0n 2

C Figure 5.5 Equivalent circuit transformed into the resonator circuit.

= R /b

is P+ = Z0 I2 /8. It is also useful to be able to transform the equivalent circuit of Fig. 5.3 into the resonator circuit. The result is shown in Fig. 5.5. Using the result that I = I /n = I βZ0 /R, where in general I depends on the coupling strength β, the available power from the generator can be expressed in terms of quantities transformed to the resonator circuit as P+ = I2 R/8β. When the generator drives the resonator at the resonant frequency, the steady-state resonator voltage is V=

IR 2β 1/2 = P+ rs 1+β 1+β

(5.26)

The equivalent circuit has made it possible to discuss the characteristics of the wave in the input guide, because the model has given us an expression for the cavity load impedance as seen in the generator circuit. For simplicity we restrict our treatment to consider a cavity driven on resonance, where the load impedance for the guide is ZL = R = Z0 β. If we consider a wave that is emitted from the generator into the guide, the reﬂection coefﬁcient produced by the cavity load impedance, becomes =

ZL − Z0 β−1 = ZL + Z0 β+1

(5.27)

The standing wave ratio is S = (1 + | |)/(1 − | |). For the undercoupled case with β < 1, | | = (1 − β)/(1 + β) and S = 1/β. For the overcoupled case with β > 1, | | = (β − 1)/(1 + β) and S = β. For the critically coupled case with β = 1, | | = 0, and S = 1, = 0 is deﬁned as the condition for the matched state. Thus, critical coupling results in a matched cavity load as seen from the waveguide. As β → 0, → −1, which corresponds to a short-circuit load, and as β → ∞, → +1, which is an open-circuit load. If P+ is the incident power from the generator, the power propagating back from the input coupler toward the generator is [7] P− = P+ 2 = P+

β −1 β +1

2 (5.28)

143

144

5 Microwave Topics for Linacs

From energy conservation, the power delivered into the cavity through the input power coupler is Pc = P+ (1 − 2 ) = P+

4β (1 + β)2

(5.29)

It is easy to show that for a given P+ , Pc is maximum, when β = 1. The voltage response of the complete circuit to the driving current I that is supplied by the generator is V = Zc i , where Zc can be expressed as a function of β by returning to Eq. (5.19), and substituting β = R/n2 Z0 . We can write a result that is analogous to Eq. (5.5), R e jφ Zc = L 1 + yL2

(5.30)

where RL = R/(1 + β), yL = QL (ω/ω0 − ω0 /ω) ∼ = 2QL (ω − ω0 )/ω0 , and φ = − tan−1 yL . We note that QL , rather than Q0 , determines the voltage response of the total circuit. It can also be shown that the resonator time constant becomes τ = 2QL /ω0 .

5.5 Equivalent Circuit for a Cavity Coupled to two Waveguides

A more common conﬁguration consists of a cavity coupled to external two transmission lines or waveguides. One external circuit usually contains the RF generator, and the other may serve as an RF pickup to monitor the ﬁelds in the cavity. The method of Section 5.4 can be used to treat this problem, using the equivalent ac circuit of Fig. 5.6. Both external guides are assumed to be terminated with a matched load resistance Z0 as shown in the ﬁgure. The input guide and generator are represented in the circuit on the left, and the output guide is contained in the circuit on the right. We can simplify the equivalent circuit by reﬂecting the external circuit on the right of Fig. 5.6 into the resonator circuit, and then reﬂecting all of the components into the circuit on the left that contains the generator. The ﬁnal equivalent circuit is shown in Fig. 5.7.

1:n1

n2:1

Figure 5.6 Equivalent circuit of a cavity coupled to two external circuits, which are transformers coupled to the cavity resonator in the center.

5.5 Equivalent Circuit for a Cavity Coupled to two Waveguides

n22Z0/n12

L /n12

R/n12

C /n12

Figure 5.7 Equivalent circuit with all components reﬂected into the external circuit containing the generator.

In this section, we will drop the prime symbols that were used in the previous section to denote quantities seen in the primary circuit. Referring to Fig. 5.7, the resonator power dissipated is Pc = n21 V 2 /2R. The power dissipation in the external circuits is Pex1 = V 2 /2Z0 , and Pex2 = n21 V 2 /2n22 Z0 , and the stored energy is U = n21 CV 2 /2. The Q values are Q0 = ω0 U/Pc = ω0 RC, Qex1 = ω0 U/Pex1 = ω0 n21 Z0 C, and Qex2 = ω0 U/Pex2 = ω0 n22 Z0 C. The total power dissipated is P = Pc + Pex1 + Pex2 , and the loaded Q is QL = ω0 U/P. The Q values are related by 1 1 1 1 = + + QL Q0 Qex1 Qex2

(5.31)

Now, we introduce the resonator-to-guide coupling factors, deﬁned as β1 =

Pex1 Q0 Pex2 Q0 = , β2 = = Pc Qex1 Pc Qex2

(5.32)

From the equivalent circuit model, we ﬁnd that β1 = R/n21 Z0 , β2 = R/n22 Z0 , Qex1 = Q0 /β1 , Qex2 = Q0 /β2 , and Eq. (5.31) can be expressed as QL =

Q0 1 + β1 + β2

(5.33)

Following the derivations of the previous section, we ﬁnd that on resonance the resonator presents a load impedance to the input guide, ZL = Z0 β1 /(1 + β2 ). Now we take this result for the load impedance from the circuit model, and use it to calculate the reﬂection coefﬁcient for the input guide that is 1 =

β1 − β2 − 1 1 + β1 + β2

(5.34)

If the input guide is matched to the resonator, we must have 1 = 0, and the matched solution is β1 = 1 + β2

(5.35)

145

146

5 Microwave Topics for Linacs

Because β2 ≥ 0, the matched condition of the input guide implies that β1 ≥ 0. Only in the weakly coupled limit for guide 2, where β2 ∼ = 0, does β1 ∼ = 1 for the matched case.

5.6 Transient Behavior of a Resonant-Cavity System

The steady-state solution of a resonant-cavity system was treated with an equivalent ac-circuit model in Section 5.4, and in this section we discuss the transient behavior. The average power in the incident wave is P+ = V+2 /2Z0 , t ≥ 0. Equation (5.28) expresses the result that for an incident wave from the generator of amplitude V+ , the effective reﬂected wave [8] from the effective load impedance at the coupling mechanism has a steady-state amplitude V− =

β −1 V+ β +1

(5.36)

and the total steady-state voltage across the effective resonator load impedance at the coupler is V = n(V+ + V− ) =

2βn V+ β +1

(5.37)

This represents the effective voltage on resonance across the resonator impedance. Now, suppose that at time t = 0, the generator of Fig. 5.3 is turned on to launch an incident wave with amplitude V+ . We assume that the wave will be partially reﬂected at the coupler with amplitude V− and partially transmitted to the resonator. The time dependence of the resonator ﬁelds during the turn-on transient was shown in Section 5.1 to contain the factor 1 − e−t/τ , where τ is the cavity time constant, τ = 2QL /ω0 . Combining the known time dependence with the steady-state result of Eq. (5.37), the time-dependent resonator voltage is V0 (t) = V+ (1 − e−t/τ )

2βn 1+β

(5.38)

We have assumed that V+ is constant after t = 0, so the reﬂected wave must have a time-dependent amplitude given by V0 (t) 2β −t/τ V− (t) = ) − V+ = V+ (1 − e −1 (5.39) n 1+β which corresponds to a time-dependent reﬂection coefﬁcient (t) =

2β V− (t) = (1 − e−t/τ ) −1 V+ 1+β

(5.40)

5.6 Transient Behavior of a Resonant-Cavity System

These results can also be expressed in terms of traveling-wave power. The average reﬂected power is 2 2β P− (t) = P+ 2 (t) = P+ (1 − e−t/τ ) (5.41) −1 1+β This power will be dissipated in the load resistor Z0 , shown in Fig. 5.3. The time-dependent power delivered to the resonator is

2 2β 2 −t/τ −1 ) Pc (t) = P+ (1 − (t)) = P+ 1 − (1 − e (5.42) 1+β We consider the limiting cases predicted by the equations of this section. At t = 0, when the incident wavefront reaches the coupler, V0 (0) = 0, V− (0) = −V+ , (0) = −1, P− (0) = P+ , and Pc (0) = 0. These results tell us that at t = 0 the resonator presents a short-circuit load, and all of the incident power is reﬂected. Similarly, as t → ∞ the steady-state results are recovered. The transient results depend on β and are illustrated in Fig. 5.8. For β ≤ 1, the backward power P− (t) decreases monotonically as a function of time from its initial value P+ to the steady-state value. For the overcoupled case with β > 1, P− (t) decreases as a function of time until it reaches zero; thereafter it increases and approaches the steady-state value. From Eq. (5.42) we ﬁnd that when t τ , the power delivered to the resonator, as well as the resonator stored energy is proportional to 4βtP+ /(β + 1)τ . This means that it is more difﬁcult to excite a resonator with a pulse that is short compared to the resonator time constant. The results of Eqs. (5.39) to (5.42), which show a time-dependent reﬂected wave and reﬂection coefﬁcient need further explanation. This effect is caused

P−(t )/P+

0.8

b = 0.1

0.6 0.4

b=1

b = 10

0.2 0

t /t

Figure 5.8

Backward power versus time for different values of the coupling factor β.

147

148

5 Microwave Topics for Linacs

by the time dependence of the resonator. We are interested in understanding this effect at the high frequencies of interest for accelerators, where the dimensions of the circuit are comparable to or greater than the wavelength, and the ﬁelds vary spatially as well as temporally. For this case, an explicit traveling-wave description that is presented in Section 5.7, is needed to provide an adequate explanation of the physics. There we will learn that the cavity builds up stored energy as a result of multiple internal reﬂections of the injected traveling wave, reinforced at each pass by a continuous ﬂow of energy from the generator through the coupler. The cavity not only receives an input wave from the generator, but radiates a wave back through the coupler. As the cavity stored energy increases, the amplitude of the radiated wave also increases. The effective reﬂected voltage and power in the input waveguide is really the sum of two traveling waves; one is the direct-reﬂected wave from the coupler, and the other is the radiated wave from the cavity. These two waves are out of phase and whether the coupling regime is undercoupled, critically coupled, or overcoupled depends on the relative strengths of the two waves. The time dependence of the effective reﬂected wave is associated with the radiated wave. For a simple model, we describe the total wave for times t > 0 as the sum of three monochromatic waves that propagate in the input waveguide. The incident wave from the generator is Vin = V+ e j(ωt−kz)

(5.43)

The wave reﬂected from the coupler is approximately [9] Vref ∼ = ±V+ e j(ωt+kz)

(5.44)

The wave radiated from the cavity is Vrad = ∓V+

2β (1 − e−t/τ )e j(ωt+kz+φ) 1+β

(5.45)

where τ 2π/ω, and where in general φ = − tan−1 (2QL δω/ω) is the phase shift of the cavity ﬁeld relative to its value when driven on resonance. The model, described by Eqs. (5.43) to (5.45), ignores the details of the wavefront propagation near t = 0. The signs in Eqs. (5.44) and (5.45) are always opposite, and the choice relative to Eq. (5.43) depends on the type of coupler. The backward wave given in Eq. (5.39) is really V− (t) = Vref + Vrad (t). 5.7 Wave Description of a Waveguide-to-Cavity Coupling

In this section, we reexamine the cavity system from a more physical point of view without reference to the ac-circuit model. In the circuit model the true wave behavior of the cavity system is concealed. In this section we

5.7 Wave Description of a Waveguide-to-Cavity Coupling

will be describing the waves, their propagation, transmission, reﬂection, and interference, both in the input waveguide and in the cavity. We begin the discussion by introducing an important tool for the analysis, called the scattering matrix [10]. Consider a waveguide that contains an obstacle or discontinuity. To be speciﬁc, we will describe the voltage across the waveguide resulting from a transverse electric ﬁeld, although the general description is not restricted to this case. An incident voltage wave coming from the left can be expressed as v+ e j(ωt−kz) , and a reﬂected wave from the obstacle is v+ e j(ωt−kz) , where is the reﬂection coefﬁcient. Some microwave devices have more than two incident channels, and some are more complex elements than simple obstacles or discontinuities. For each channel there is generally an incident wave and an outgoing wave. The total outgoing wave in any channel is the sum of the reﬂected wave in that channel plus the contributions of the transmitted waves from all the other channels. The general description has been formulated in terms of a scattering matrix [11]. To deﬁne the scattering matrix for an n-port device, we begin by deﬁning normalized incident waves, such that the incoming traveling waves are represented as ai , i = 1, 2, . . . , n, and the outgoing waves are represented by bi , i = 1, 2,√. . . , n. The ith normalized i j(ωt−kz) incident voltage wave / Z0 , and the outgoing wave √ is deﬁned as ai = v+ e i j(ωt+kz) / Z0 . The powers in the ith incident and outgoing waves is bi = v− e i i 2 i i 2 are P+ = (v+ ) /2Z0 = |ai |2 /2, and P− = (v− ) /2Z0 = |bi |2 /2. The scattering matrix S is deﬁned as the transformation of the column vector a of the incident waves into the column vector b of the outgoing waves. Thus,  b  s s ... s  a  1

 b2   s21 s22 . . . s2n   a2   =   ... ... ... ... ... ... bn sn1 sn2 . . . snn an

(5.46)

Each diagonal element sii is the reﬂection coefﬁcient i in the ith channel. Each off-diagonal element sij is the transmission coefﬁcient from channel i into channel j. The number of channels is also called the number of ports. Thus, an n-port element has n input channels, and is described by an n × n scattering matrix. Reciprocal n-port elements are symmetric and satisfy sij = sji . It can be shown that S is a unitary matrix when the element is free of power losses. Next, we describe a simple cavity. We know that two conducting plates can be inserted into a waveguide to create a microwave cavity in the enclosed volume. To excite such a cavity there must be some coupling mechanism to provide a way of delivering the electromagnetic energy from a microwave generator into the cavity. One simple way of doing this is to place a hole in one of the plates, through which at least some fraction of the incident wave in the transmission line can be transmitted into the enclosed cavity. The scattering matrix can be used to describe the effects of a conducting plate with a small hole or iris that is inserted into a waveguide perpendicular to the axis, as shown in Fig. 5.9. We consider the iris to be a two-port device that has two input channels, one from the left and one from the right. We describe the iris by a 2 × 2 scattering

149

150

5 Microwave Topics for Linacs b1 a1

b2 a2 Cavity

Waveguide

Iris

l = nlg/2

Figure 5.9 Waveguide and cavity section separated by a conducting plate with an iris.

matrix,

b1 b2

s = 11 s21

s12 s22

a1 a2

(5.47)

We can determine the properties of the scattering matrix for the iris, using a simple treatment based on two assumptions. The iris is assumed to be left–right symmetric, so that if the iris orientation is reversed, its behavior is unchanged. This property is known as reciprocity, and implies that s11 = s22 , and s21 = s12 . We have already noted that the diagonal elements are equal to the reﬂection coefﬁcient . As the hole size increases until the obstacle recedes into the waveguide wall, the magnitude of will be expected to decrease to zero. If the hole size vanishes, we know from transmission-line theory that the reﬂection coefﬁcient for the voltage wave from a perfectly conducting short circuit is = −1. In general we write s11 = s22 = , where is a negative number. The iris itself is assumed to be free of energy dissipation. Thus, energy conservation implies that the total wave power coming into the iris equals the total wave power going out, or |a1 |2 + |a2 |2 = |b1 |2 + |b2 |2

(5.48)

To see what energy conservation implies about the scattering matrix, we form b1 = s11 a1 + s12 a2 = a1 + s12 a2 . The complex conjugate is b∗1 = a∗1 + s∗12 a∗2 . Then we obtain b1 b∗1 = 2 |a1 |2 + s12 s∗12 |a2 |2 + s∗12 a1 a∗2 + s12 a2 a∗1

(5.49)

and interchanging subscripts 1 and 2 we obtain the corresponding result for b2 b∗2 . Then, b1 b∗1 + b2 b∗2 = ( 2 + s12 s∗12 )|a1 |2 + ( 2 + s12 s∗12 )|a2 |2 + (s12 + s∗12 )a1 a∗2 + (s12 + s∗12 )a2 a∗1

(5.50)

To satisfy energy conservation as in Eq. (5.48), we require 2 + |s12 |2 = 1 and s12 + s∗12 = 0

(5.51)

5.7 Wave Description of a Waveguide-to-Cavity Coupling 2 2 Equation √ (5.51) are satisﬁed if Re(s12 ) = 0, and |s12 | = 1 − . A solution is s12 = j 1 − 2 , and the scattering matrix for the iris becomes j 1 − | |2 S= (5.52) j 1 − | |2

The properties of symmetry or reciprocity, and conservation of energy in the waves, imply that S is determined in terms of a single parameter, the reﬂection coefﬁcient. Suppose that on the left side of the iris is an input waveguide through which an RF generator delivers a traveling electromagnetic wave to drive the cavity. When the wave from the generator is incident upon the iris there will be some fraction that is reﬂected back into the waveguide, and the remaining fraction that is transmitted into the cavity. The transmitted wave will propagate in the cavity to the plane of the short circuit, where it will be reﬂected back to the iris. At the iris the wave is again partially reﬂected back into the cavity, and partially transmitted back into the incident waveguide. The input waveguide will have both incident and outgoing traveling waves. If the generator frequency is varied, there will exist discrete resonant frequencies at which we will observe the build up of a high-ﬁeld standing wave within the cavity region as a result of the constructive interference of the multiply reﬂected waves inside the cavity. Eventually, a steady state will be reached, where the energy lost from ohmic power dissipation in the cavity walls, and wave energy propagating out of the iris, becomes equal to the energy entering the cavity from the generator through the iris. From Eqs. (5.47) and (5.52) we write b1 = a1 + j(1 − 2 )1/2 a2 b2 = j(1 − 2 )1/2 a1 + a2

(5.53)

The expressions for b1 and b2 have two terms, a reﬂected wave and a transmitted wave through the iris. To examine the build up of the ﬁeld, suppose that at time t = 0, the wave from the generator appears at the iris with amplitude a1 , and we will also assume that a2 = 0 initially. Shortly afterwards, at time t = 0+ , the outgoing amplitudes, obtained from the scattering matrix are b1 = a1 and b2 = j(1 − 2 )1/2 a1 . One cycle later at time T = 2/vG , the wave b2 will have propagated to the short, where it is reﬂected with a phase reversal, and propagated back to the iris, where it becomes an incoming wave from the right with amplitude a2 . A small fraction of the a2 wave is transmitted through the iris to contribute to the outgoing wave b1 , and the largest fraction is reﬂected back with a phase reversal and produces a step in the wave b2 . If we assume an ohmic attenuation for the round trip is given by e−2α , where α is the attenuation constant per unit length, we have a2 = −b2 e−(2α+j2φ) , where 2φ is the round-trip phase shift of the wave. The phase shift φ = 2π/λ is determined by /λ, the electrical length of the cavity, where λ is the free-space wavelength. Subsequent round-trip passes

151

152

5 Microwave Topics for Linacs

of the internal wave will reinforce each other and produce resonance when φ = 2πn, where n is a positive integer. At resonance we have a2 = −b2 e−2α

(5.54)

−2α At time √ t = 0+ + T, the updated amplitudes become a2 = √−e −2α 2 −2α 2 b2 = −je 1 − a1 , b1 = ( + (1 − )e )a1 , and b2 = j 1 − 2 −2α (1 − e )a1 . Following the buildup of the waves, after N cycles we ﬁnd

N−1 1 − xN xn = − j 1 − 2 a1 e−2α a2 (N) = −j 1 − 2 a1 e−2α 1−x n=0

b1 (N) = a1 + (1 − 2 )e−2α

N−1

(5.55)

n=0

1 − xN = a1 + (1 − 2 )e−2α 1−x

(5.56)

and

b2 (N) = j 1 −

2 a

1 − xN+1 2 x =j 1 − a1 1 1−x n=0 N

(5.57)

where x ≡ − e−2α is a positive number. The steady-state solution corresponds to N → ∞, and since |x| < 1, xN → 0 and the steady-state solution is √ −j 1 − 2 a1 e−2α (5.58) a2 (∞) = 1−x (1 − 2 )e−2α (5.59) b1 (∞) = a1 + 1−x and b2 (∞) =

√ j 1 − 2 a1 1−x

(5.60)

Thus, we have an expression for the growth of the traveling-wave amplitude in the cavity as a stairstep proﬁle given by b2 (N)/b2 (∞) = 1 − xN+1 , and as shown in Fig. 5.10. Assuming ≈ −1, we can approximate (− )N ∼ = e−N(1+ ) or N −2α N ∼ −(1+ +2α)N ) =e . We approximate this result as a continuous x = (− e function of time using a smooth approximation t = NT, where T = 2/vG is the round-trip transit time of the wave or twice the ﬁll time, and vG is the group velocity. The cavity time constant τc is deﬁned as the time required for the exponent to increase from zero to unity; thus xN ∼ = e−t/τc , where τc =

T 1 + + 2α

(5.61)

5.7 Wave Description of a Waveguide-to-Cavity Coupling 3 Exponential approximation

2.5

Ia2/a1I

2 1.5 Stairstep function

1 0.5 0

t /T

Figure 5.10 Transient build up of the cavity ﬁeld seen at the input iris, and the exponential approximation for the buildup. The parameters = −0.95 and 2αl = 0.05 were chosen to facilitate an illustration of the stairstep buildup. The cavity time constant is τ = 10T.

We can also write x = e−T/τc , so that b2 (N) = 1 − xt/T+1 = 1 − e−(t+T)/τc ∼ = 1 − e−t/τc b2 (∞)

(5.62)

where the last step is valid when t T. Therefore, we ﬁnd that the stairstep growth of the traveling-wave amplitude of the cavity is approximated by an exponential growth. It will be shown that the time constant is τc = 2QL /ω0 , the same result that we obtained from the equivalent circuit model in Section 5.1. Critical coupling corresponds to the value of that makes the outgoing-wave amplitude b1 = 0. For this condition all of the incident power is absorbed in the cavity. At critical coupling, b1 = 0 is the result of the reﬂected wave in the input waveguide being exactly canceled by destructive interference with the wave emitted from the cavity. Any other condition results in an outgoing wave in the input waveguide, and less power absorbed in the cavity. The steady-state value of b1 from Eq. (5.59) can be re-expressed as + e−2α (5.63) a1 b1 = 1 + e−2α Thus, b1 = 0 for critical coupling occurs when = −e−2α

(5.64)

For critical coupling the reﬂection coefﬁcient equals the round-trip ﬁeld attenuation factor for the traveling wave in the cavity. Suppose 2α ≈ 10−4 , and we approximate ∼ = −1 + 2α. Equations (5.58) and (5.60) give |a2 | ∼ = √ 2 ∼ ∼ |b2 | = |a1 |/ 4α = 10 a1 . Then the iris is nearly a short circuit that almost

153

154

5 Microwave Topics for Linacs

completely reﬂects the incident wave. The outgoing wave b1 is the sum of a reﬂected wave plus an emitted wave from the cavity. The emitted wave is a small fraction of the large internal wave a2 that has built up within the cavity. Because is negative, the reﬂected and emitted waves are of opposite phase. The phase difference can be understood by adding the phase shifts that occur in a round-trip through the cavity, including the 180◦ shift at the reﬂections, and the 90◦ phase shift that occurs both at injection into the cavity and at emission from the cavity. When = −e−2α , the reﬂected and emitted waves cancel. When | | > e−2α , the reﬂected wave dominates, which is the undercoupled case. When | | < e−2α , the emitted wave from the cavity dominates, which is the overcoupled case. We now relate these results to conventional cavity parameters for comparison with the results from the equivalent circuit model. The standing wave in the cavity is composed of the two waves, traveling in opposite directions. For the wave propagating in the +z direction, the traveling-wave power is P+ = U+ vG /, where U+ is the stored energy in the +z wave. Similarly, for the wave propagating in the −z direction, P− = U− vG / is the power in the wave propagating in the −z direction. We have P+ ∼ = P− , so the total stored energy in the cavity is U = U+ + U− =

2P+ vG

(5.65)

The fractional decrease in the ﬁeld amplitude of the traveling wave in one round-trip transit of the cavity is (V+ − V+ )/V+ = e−2α , and the fractional decrease in the traveling-wave power is (P+ − P+ )/P+ = e−4α . The wave energy crossing any plane decreases by U = P+ T ∼ = 4αP+ T in a roundtrip transit time T = 2/vG . The power dissipation is P = U/T = 4αP+ . Then, the unloaded Q, the quality factor associated only with cavity power dissipation is Q0 =

ω0 ω0 U = P 2αvG

(5.66)

We note that Q0 is independent of and inversely proportional to vG , because smaller vG implies more stored energy per unit wave power. Another important quality factor is related to the strength of the waveguideto-cavity coupling through the iris, and this is deﬁned by the external quality factor or Qex , deﬁned at resonance as the ratio of ω0 U to the energy radiated from the cavity through the iris, when the generator is turned off from P+ to 0. From Eq. (5.53), when a1 = 0, the emitted wave is b1 = j(1 − 2 )1/2 a2

(5.67)

Then the emitted or externally dissipated power is Pex =

|b1 |2 |a2 |2 = (1 − 2 ) = (1 − 2 )P+ 2 2

(5.68)

5.7 Wave Description of a Waveguide-to-Cavity Coupling

The external Q is Qex =

ω0 U 2 ω0 = Pex vG (1 − 2 )

(5.69)

Note the important requirement that the generator be turned off. If this condition is forgotten, one might mistakenly conclude that for critical coupling Qex is inﬁnite, since for critical coupling, when the generator is on, there is no power radiated into the external circuit. The absence of radiated power is because of the complete destructive interference of the radiated wave with the reﬂected wave from the iris, when the generator is on. The external Q would seem to be an artiﬁcial concept for describing the performance of the cavity, when the generator drives the cavity exactly on resonance. Nevertheless, we will see that Qex contributes to the loaded Q that determines the real cavity time constant, when the cavity is loaded by the external waveguide [12]. Whereas Q0 depends on the attenuation constant α, Qex depends on the power transmission through the iris, 1 − 2 . The waveguide-to-cavity coupling parameter is β≡

Q0 1 − 2 ∼ 1 − 2 = = Qex 1 − e−4α 4α

(5.70)

Physically, β corresponds to the ratio of power that would be radiated out through the iris, if the incident wave from the generator suddenly collapsed to the power dissipated in the walls of the cavity. Recall that the condition for critical coupling is − = e−2α , so given the assumption that α 1, we have 1 − 2 ∼ = 4α, and from Eq. (5.70) this corresponds to β = 1, and Qex = Q0 . Similarly, undercoupling corresponds to β < 1, and overcoupling corresponds to β > 1. Finally the loaded Q or QL , is the ratio at resonance of cavity stored energy to energy dissipated in the cavity per radian plus the energy radiated from the cavity per radian through the iris, when the generator is turned off. The loaded Q is given by QL =

ω0 U ∼ ω0 1 = P + Pex vG (1 − 2 ) 2α + 2

(5.71)

Using 1 − 2 ∼ = (1 − )(1 + ) ∼ = 2(1 + ), we ﬁnd that ω0 1 QL ∼ = vG [1 + + 2α]

(5.72)

It is easy to show that QL = Q0 /(1 + β), and for critical coupling where β = 1, we have QL = Q0 /2. Comparing Eqs. (5.61) and (5.72), and using the deﬁnition T = 2/vG , shows that the cavity time constant is τc =

2QL ω0

(5.73)

155

156

5 Microwave Topics for Linacs

which is the same result already deduced from the equivalent circuit model. Thus, the equivalent circuit model of Section 5.1 is seen to be a valid representation of the average or smoothed time dependence of the cavity ﬁelds. Analysis of the cavity decay process also shows a stairstep solution for the decay that is approximated by a smooth exponential decay. The same argument that was used above for the growth of the ﬁelds, shows that the decay time constant is again τc = 2QL /ω0 , the same as for the growth.

5.8 Microwave Power Systems for Linacs

The material of this section will provide a short introduction to the subject of microwave power systems. We begin with a discussion of microwave power sources. The interested reader is referred to the other books that deal with this subject, such as Smith and Phillips (1995), and Gandhi (1987). RF power must be supplied to the linac cavities to establish and maintain the electromagnetic ﬁelds [13]. The choice of an RF power source for an accelerator cavity depends on many factors, including frequency, peak and average power, efﬁciency, reliability, and cost, and generally, the RF power system is a major contributor to the cost of a linac. High-power vacuum-tube ampliﬁers, designed for operation in the frequency range of importance for linacs (typically from a few tens of megahertz to a few tens of gigahertz) include triodes, tetrodes, and klystrons. The gridded tubes, including triodes and tetrodes, are used as ampliﬁers at frequencies below about 300 MHz, and the application of the gridded tube has been almost exclusively for pulsed operation. Magnetrons have been used for accelerators with a single section, but their inherent lack of phase stability has prevented their application to multisection linacs. The most commonly used linac RF power source is the klystron ampliﬁer, which is normally used at frequencies above about 300 MHz, where its size is practicable. Klystrons have been operated at pulse lengths ranging from about 1 µs to a continuous wave or CW. In the basic two-cavity klystron, the beam is injected from an electron gun, accelerated by a dc potential V, and focused along the length of the klystron by an axial magnetic ﬁeld. The input cavity is excited at the resonant frequency by a low-voltage generator. As the dc electron beam enters the input cavity, the electrons are either accelerated or decelerated by the RF axial electric ﬁeld of the input cavity, depending on the phase of the accelerating ﬁeld when the electrons enter the gap. As the velocity-modulated electron beam travels along the drift tube following the input cavity, the accelerated electrons from one RF cycle tend to catch up with decelerated electrons from the previous cycle. This results in the formation of bunches of electrons at the frequency of the input cavity ﬁelds. If the output cavity is tuned to the same frequency, it will be excited by the bunched beam, and if the output cavity is coupled to an output waveguide, RF output power will be radiated into the guide. The output power

5.8 Microwave Power Systems for Linacs

level can be much larger than the input power, resulting in ampliﬁed RF power. Maximum gains of up to 50 to 60 dB are typical. Klystrons have been designed to operate successfully from a few hundred megahertz up to 100 GHz. For room-temperature accelerators operating at accelerating ﬁeld levels of a few megavolts per meter, and over lengths of a few meters, the RF power required is about 1 MW. At frequencies of a few hundred megahertz, CW klystrons near 1 MW are commercially available. For pulsed accelerators in the 3-GHz frequency range, klystrons with peak power of 60 MW are used; the SLAC linac uses a total of 240 such klystrons. Klystron efﬁciencies, deﬁned as the ratio of RF output power to dc beam power, near about 40% are typical for pulsed operation, and efﬁciencies near 60% have been achieved for cw operation. The output circuit must be designed to limit the reﬂected power from the cavity load to prevent deterioration of the performance of the device, and to avoid arcing damage from the enhanced ﬁelds. It is common in high-power systems to install an arc detector, and a reﬂected power monitor in the output line, and to switch off the dc power within microseconds of an abnormal condition. Such a condition can develop when the accelerator cavity that represents the load impedance for the klystron, experiences an arc-down condition. Another method of protection is to install a waveguide device called a circulator to isolate the klystron from power reﬂected from the load impedance. The circulator, installed between the ampliﬁer and the cavity, is a special microwave device that transmits RF power traveling from the RF ampliﬁer to the cavity. Any power coming back from the cavity to the circulator is delivered to a matched load rather than back to the RF ampliﬁer, where it could cause damage or problems with the performance. Because of the circulator, the drive system appears like a matched RF generator. Klystrons are reliable devices with a mean time before failure that is greater than about 30,000 h. A low-level RF control system is used to maintain the phase and amplitude of linac-cavity ﬁelds, given the thermal and mechanical perturbations that cause resonant frequency variations, and beam-current variations that directly perturb the cavity ﬁelds [14]. Because an accelerator cavity has a high Q and can only accept power over a narrow frequency band, the low-level control system must also provide a resonance tracking and adjustment system that retunes the cavity, allowing it to accept RF power at the design frequency. Figure 5.11 shows a schematic drawing of an RF system for an accelerator cavity. Conceptually, the cavity RF pickup signal is fed back to the ﬁeld-control electronics, where it is subtracted from a stable RF reference signal to produce an error signal. The ﬁeld-control electronics may augment the error signal with a feedforward signal that allows disturbances in the beam current that are detected upstream of the cavity to be anticipated and compensated. This approach is based on the use of past error information to predict a suitable feedforward correction function. Feedforward regulation works for nonrelativistic beams, where the signal can propagate from the point of detection to the point of application, faster than the beam. The resulting RF error signal has an amplitude that

157

158

5 Microwave Topics for Linacs High power RF amplifier

RF reference signal

Field control electronics

Feedforward signal

Cavity pickup signa l

Load Power coupler

Cavity

Circulator

Frequency Tuner

Resonance detection electronics

Figure 5.11 Schematic drawing of an RF system for an accelerator cavity.

is adjusted in proportion to the cavity-amplitude error, and a frequency that allows the cavity phase to catch up to the reference phase. This error signal is applied to the input of the high-gain, high-power RF ampliﬁer that delivers RF power to the cavity. The details by which the error signal is produced may be realized with some standard microwave components such as a doubly balanced mixer, used to compare two RF signals and generate a voltage level proportional to the phase difference, and a voltage-controlled oscillator that produces an RF output wave with a frequency that depends on the input phase-error level. The RF pickup signal is also used by the resonance-detection electronics, which compares the relative phases of the pickup signal and the drive signal, and produces an error signal proportional to the phase difference. The difference can be used to implement a slow retuning correction to the cavity. For a normal conducting cavity, the resonant frequency is usually maintained by regulating the temperature of the cooling water. For a superconducting cavity, the resonant frequency is usually maintained by applying a mechanical force, for example, to push or pull on the ends of the cavity. Most perturbations that affect the cavity ﬁelds occur at frequencies below about 1 MHz. The typical 1-MHz upper limit occurs because noise frequencies corresponding to periods that are short compared with the cavity time constant will be ineffective at producing signiﬁcant ﬁeld variations. In feedback control, phase shifts in the loop, especially at high frequencies, which cause a roll off in the loop gain, establish a characteristic time constant and an associated bandwidth limit above which the feedback is positive. Consequently, feedback is effective only for regulating low-frequency ﬁeld perturbations; in practice feedback works well for frequencies below about 10 kHz. Feedforward is especially helpful for control of ﬁeld variations induced by beam noise at frequencies above about 10 kHz, where feedback control is only partially effective. Using feedback and feedforward, the amplitude and phase of the cavity ﬁeld may typically be controlled to less than 1% and 1◦ .

5.9 Multipacting

Electrons can be driven from the cavity walls by bombardment of charged particles or X rays, or produced by ionizing processes in the residual gas of the imperfect vacuum. These electrons can absorb energy from their interaction with the electromagnetic ﬁelds. Such effects are often referred to as electron loading, a process that can limit the ﬁelds, and result in X-ray emission. Three effects can be identiﬁed as contributing to electron loading: (1) multipacting, (2) ﬁeld emission, and (3) RF electric breakdown. The word multipacting [15] is a contraction of the phrase multiple impact. To understand this phenomenon, consider an ac electric ﬁeld perpendicular to two planeparallel metallic surfaces with a ﬁeld strength such that an electron originating from one surface at x = 0 takes exactly one-half period to travel across the gap between the surfaces, as shown in Fig. 5.12. Suppose the conditions are such that this electron knocks out more than one secondary electron from the second surface at x = x0 . The ﬁeld will reverse, and these electrons will return to the ﬁrst surface, where they may each knock out more electrons. Once this process has begun, it will continue until the avalanche is limited by the space charge of the electrons. Multipacting requires two conditions: (1) a kinematic condition for resonant buildup must be satisﬁed (i.e., the electrons must travel from surface to surface in an integer number of half periods of the ac voltage and (2) a physical condition must be satisﬁed, i.e., the secondary electron coefﬁcient must exceed unity). The simple theory of two-conductor or two-point multipacting is based on nonrelativistic electron dynamics in a pure ac electric ﬁeld between two planeparallel conducting plates. Because electrons can be born at different phases of the RF voltage, the kinematic conditions for multipacting can occur at a range of voltages V = E0 x0 between Vmax =

mω2 x02 2e

(5.74)

Secondary electrons

Primary e −

E0sin(wt) x=0

x = x0

Figure 5.12 Initial stage of two-point multipacting between two plane surfaces. Electrons traverse gap in half an rf period.

159

5 Microwave Topics for Linacs

and Vmin =

mω2 x02 1 e 4 + (2n + 1)2 π 2

(5.75)

where e and m are the electron charge and mass. The index n is the order of the multipacting level, and is related to the transit time between the two surfaces t = (n + 1/2)T, where T is the RF period, and n = 0, 1, 2, . . . . There is a maximum electric ﬁeld, above which multipacting cannot occur, given by Emax =

Vmax mω2 x0 = x0 2e

(5.76)

For larger ﬁelds the electron transit times are too short to satisfy the resonant condition. There is no minimum ﬁeld below which the kinematic conditions for multipacting cannot be satisﬁed, because the order n can be increased indeﬁnitely. It may seem surprising that the maximum voltage is independent of n. This is because multipacting at the maximum voltage corresponds to the electron leaving a surface, when the electric ﬁeld is maximum, and arriving at the opposite surface, when the ﬁeld is maximum in the reversed direction. This electron is accelerated and decelerated by the same amount, and the integrated force is zero, independent of n. Electrons born at other phases require less peak voltage, because they are given a net velocity increase by the ﬁeld. Regions where kinematic conditions for multipacting between two parallel plates are satisﬁed are shown in Fig. 5.13, where Eq. (5.74) is plotted. For values of V/f 2 larger than that given by Eq. (5.74), multipacting is not kinematically possible. For values less than the Eq. (5.74) result, multipacting can always exist for some order n. It is useful to summarize some characteristics of multipacting. 1. Multipacting often occurs at much lower electric ﬁeld levels than those typically used for acceleration of the beam. It is often encountered when raising the cavity ﬁeld to the operating level. 1

V (V )/f 2 (MHz)

160

0.1

No multipacting possible

Multipacting possible

0.01

0.001 0.01

0.1

1 x0 (cm)

100

Figure 5.13 Regions where kinematic conditions between two parallel plates are or are not satisﬁed. The quantity f = 1/T is the RF frequency in megahertz, V is in volts, and x0 is in centimeters. The curve that separates the two regions is Eq. (5.74).

5.9 Multipacting

2. The simple theory does not account for the effects of the RF magnetic ﬁelds on the electrons. 3. There is evidence that dc magnetic ﬁelds perpendicular to the surfaces can enhance multipacting, whereas magnetic ﬁelds parallel to the surface can suppress the effect. 4. Overcoupling to reduce the time constant for the buildup of the cavity ﬁelds can suppress the effect by bringing the ﬁeld past the multipacting region, before the resonance has time to build up. 5. Keeping the surfaces thoroughly clean and oil-free by using good vacuum practice such as use of oil-free vacuum pumps, can suppress multipacting by keeping the secondary electron coefﬁcient small. 6. RF conditioning, which means operating at the multipacting resonance with relatively high-power levels, may help by letting the multipactor discharge burn until the secondary electron coefﬁcient reduces to a sufﬁciently small value. 7. The most serious multipacting condition arises when multipacting occurs at a level near the operating level of the cavity. A problem has occurred in the low-ﬁeld coupling cavities of coupled-cavity linacs, sometimes exacerbated by the presence of solenoidal dc magnetic ﬁelds used to focus the beam. The solution for this case has been to detune the coupling cavities, which raises the ﬁelds above the resonant values. 8. Surface coatings of titanium and titanium nitride can reduce the secondary electron coefﬁcient. 9. A plot of the secondary electron coefﬁcient versus electron kinetic energy shows that the coefﬁcient typically exceeds unity over a range of impact kinetic energies extending from a few tens of electron volts to several kilovolts. 10. Symmetry in both the cavity geometry and the cavity ﬁelds can result in strong multipacting, because electrons from a large surface area can be involved. Another form of multipacting, known as single-point multipacting, was discovered in superconducting cavities, and occurs at the outer wall of cavities operating in the TM010 -like mode, where the RF magnetic ﬁelds are large. Characteristics of single-point multipacting include the following. (1) The electrons are emitted from, and return to the same surface, where they knock out secondary electrons and build up a discharge. (2) The time between surface hits is an integer multiple of an RF period, rather than an odd number of half

161

162

5 Microwave Topics for Linacs

periods, as for the two-point multipacting [16]. Multipacting can occur only if the electrons impact the surface with sufﬁcient energy that the secondary electron coefﬁcient exceeds unity. The single-point multipacting theory yields an expression for the impact energy 2 3 E⊥ (MV/m) (5.77) Wimp (eV) = 7.6 × 10 f (GHz) Single-point multipacting has limited the performance of pillbox-like superconducting cavities operating in the TM010 mode. The method that has been most successful in suppressing single-point multipacting for the superconducting case is to use a spherical cavity geometry at the outer wall. This cavity geometry effectively suppresses single-point multipacting by creating a conﬁguration such that electrons, emitted from the walls, drift toward the equator, where E⊥ ≈ 0.

5.10 Electron Field Emission

Electron ﬁeld emission is an effect that limits the high-ﬁeld performance of superconducting cavities. Electrons are emitted from material surfaces in the presence of a strong surface electric ﬁeld. These electrons are further accelerated by the cavity ﬁelds, and when they strike the cavity walls their kinetic energy is converted to heat and X rays. Electron loading absorbs RF energy and lowers the Q of superconducting cavities at high ﬁelds. Classical ﬁeld emission follows the Fowler–Nordheim law for the emitted current density E2 a 3/2 j∝ exp − (5.78) E where E is the electric ﬁeld and is the work function. For niobium = 4.3 eV. An approximate formula for the ﬁeld-emitted current density for niobium is 6 j(1012 A/m2 ) ∼ (5.79) = (6E)2 exp − E where E is in units of 1010 V/m. In practice E is not the ideal surface ﬁeld Es , but is a surface-ﬁeld-enhanced value E = βES, , where β is a ﬁeld enhancement factor that can be as large as β ≈ 250. It is believed that the onset of ﬁeld emission at lower than expected ﬁelds is often associated with low-workfunction dust particles as well as needles or other protrusions. The most important rule for suppressing ﬁeld emission in superconducting cavities is cleanliness. This may include rinsing of cavity surfaces in ultrapure water, and assembly in laminar dust-free airﬂow systems as used in the semiconductor industry. Statistically, the probability of ﬁeld emission is expected to increase

5.11 RF Electric Breakdown: Kilpatrick Criterion

with the surface area that is exposed to high ﬁelds. Thus, smaller cavities tend to achieve higher ﬁelds. An approximate scaling formula for peak surface electric ﬁeld in superconducting elliptical cavities is Epeak ∝ (surface area)−1/4 .

5.11 RF Electric Breakdown: Kilpatrick Criterion

At sufﬁciently high ﬁelds, room-temperature copper cavities will suffer electric breakdown or sparking. The detailed mechanism of this breakdown is not well understood, but it may be initiated by electron ﬁeld emission and it has been suggested that protons, originating on the surfaces or perhaps from hydrogen in the residual gas, are involved in the discharge. In the 1950s, W. D. Kilpatrick [17] analyzed the data on RF breakdown, and deﬁned the conditions that would result in breakdown-free operation. The Kilpatrick results were expressed in a convenient formula by T. J. Boyd [18] given as f (MHz) = 1.64EK2 e−8.5/EK

(5.80)

where f is the frequency, and EK in megavolts per meter is known as the Kilpatrick limit. This is plotted in Fig. 5.14. Note that for a given frequency, the equation must be solved iteratively for EK . Also, notice the similarity of the Kilpatrick-limit formula and the Fowler–Nordheim ﬁeld-emission formula. In the Kilpatrick formula, EK increases with increasing frequency. The Kilpatrick criterion is based on experimental results that were obtained in an era before clean vacuum systems were common. Therefore, the Kilpatrick criterion is considered conservative by today’s standards. Nevertheless, the same expression is commonly used for choosing the design ﬁeld level for

1500

f (MHz)

1250 1000 750 500 250 15

20 25 EK (MV/m)

Figure 5.14 The Kilpatrick formula from Eq. (5.80).

163

164

5 Microwave Topics for Linacs

accelerating cavities, except that the actual peak surface ﬁeld Es is expressed as Es = bEK

(5.81)

where b is known as the bravery factor. Typical values chosen for b, by accelerator designers, range from 1.0 to 2.0. Larger values of b are often chosen for pulsed accelerators with pulse lengths less than about 1 ms. For very short pulses, valid near 1 µs or less, an empirical scaling has been observed, given by [19] Es ∝ f 1/2 /t1/4

(5.82)

5.12 Adiabatic Invariant of an Oscillator

If a perturbation is introduced in the cavity geometry, either from the displacement of the cavity walls, from the introduction of a hole or aperture, or from the introduction of some object, the resonant frequency will generally change, and we will be interested in calculating this change. First we look at a general theorem concerning the effect of adiabatic changes on the frequency, and then we will present the Slater perturbation theorem, which is the basis of microwave ﬁeld measurements. Consider an oscillator in which a parameter varies as a result of some external action at a rate that is slow compared to the period of the oscillation. Such changes are said to be adiabatic [20]. It can be shown [21] that for a frictionless or Hamiltonian system, pdq dq dp I≡ = (5.83) 2π 2π is invariant when any parameter is varied adiabatically. To illustrate this consider a one-dimensional simple harmonic oscillator with displacement q, momentum p, mass m, and resonant frequency ω0 . The Hamiltonian is H(q, p) =

mω02 q2 p2 + 2m 2

(5.84)

and from conservation of energy H(q, p) = U, where U is the constant total energy. The path of a trajectory in phase space is an ellipse, given by p2 mω02 q2 + =1 2mU 2U

(5.85)

U πp0 q0 = 2π ω0

(5.86)

√ corresponding to semiaxes p0 = 2mU, and q0 = 2U/mω02 . If there are adiabatic changes of the parameters ω0 or m, the energy is no longer constant, and it can be shown that an adiabatic invariant is I=

5.13 Slater Perturbation Theorem

or U ω0 = ω0 U

(5.87)

Thus, when the parameters of the oscillator vary slowly, the frequency is proportional to the energy of the oscillator. This result, also known as the Boltzmann–Ehrenfest theorem, [22] applies to resonant cavities as well as mechanical oscillators, and describes the effects of frequency tuners used in microwave cavities. It is easy to see from the result, that when the resonant frequency changes, so also do the amplitudes q0 and p0 , where we have p0 ∝ U and q0 ∝ U −1 . An adiabatic process can be thought of as proceeding approximately through a sequence of equilibrium steps. The maximum kinetic and potential energies remain equal to each other and equal to the total energy U.

5.13 Slater Perturbation Theorem

From the discussion of the previous section, we have learned that perturbations of a simple oscillator resulting in a change in the stored energy will generally result in a resonant frequency shift. For a cavity on resonance, the electric and magnetic stored energies are equal. If a small perturbation is made on the cavity wall, this will generally produce an unbalance of the electric and magnetic energies, and the resonant frequency will shift to restore the balance. The Slater perturbation theorem [23] describes the shift of the resonant frequency, when a small volume V is removed from the cavity of volume V. The general result is (µ0 H2 − ε0 E 2 ) dV ω0 Um − Ue V = (5.88) = ω0 U (µ0 H2 + ε0 E 2 ) dV V

where U is the total stored energy, given in terms of the unperturbed ﬁeld amplitudes E and H by 1 U= (µ0 H2 + ε0 E 2 ) dV (5.89) 4 V is the time average of the stored magnetic energy Um = V µ0 H2 dV/4 removed, and Ue = V ε0 E 2 dV/4 is the time average of the stored electric energy removed as a result of the reduced volume [24]. The frequency increases if the magnetic ﬁeld is large where the walls are pushed in, and it decreases if the electric ﬁeld is large there. This result is easier to remember if one identiﬁes a decrease in the effective inductance where the magnetic ﬁeld is large and an increase in the effective capacitance where the electric ﬁeld is

165

166

5 Microwave Topics for Linacs

large. There can be cases where the electric and magnetic effects cancel. For example, this would occur if the end walls of a pillbox cavity, excited in a TM010 mode, are uniformly pushed in. The contribution from the electric ﬁelds near the axis is exactly canceled by the dominant magnetic ﬁeld effect at larger radius. This is why the frequency is independent of the cavity length for this case. The Slater theorem provides the basis for ﬁeld measurements in cavities [25]. If a small bead is inserted into the cavity, the perturbation shifts the resonant frequency. For a spherical bead of volume V, the shift is given as a function of the unperturbed ﬁeld amplitudes E and H, which are assumed to be constant over the bead, by 3V εr − 1 µr − 1 ω0 2 2 =− (5.90) ε0 E + µ0 H ω0 4U εr + 2 µr + 2 where εr and µr are the dielectric constants, and magnetic permeability of the material relative to vacuum. For a spherical dielectric bead, we choose µr = 1, and 3V εr − 1 ω0 =− ε0 E 2 ω0 4U εr + 2

(5.91)

Thus, if the shift in the resonant frequency is measured, and the stored energy is known from measurement of the Q and the power dissipation, the magnitude of the electric ﬁeld can be calculated. Sapphire is a commonly used dielectric that has a relative dielectric constant εr ≈ 9. The results for a spherical perfectly conducting bead are obtained by letting εr → −j∞ (Inside a conductor the relative dielectric constant is εr = 1 − jσ/ε0 ω.), and µr → 0 for a diamagnetic metal, and we ﬁnd 3V µ0 H2 ω0 =− ε0 E 2 − (5.92) ω0 4U 2 Because this result depends on both E and H, unless it is already known which ﬁeld is dominant, it would be necessary to do measurements using both dielectric and metallic beads to determine both ﬁeld values. Similar equations have been derived for beads of other geometries. For example, a metallic needle allows one to obtain some directional information about the ﬁelds. In the standard bead-perturbation technique for measuring the RF electric and/or magnetic ﬁelds versus position, a small bead is moved through the cavity using a typical setup as shown in Fig. 5.15. Typically, the beam is attached to a thin nylon line that is driven through the cavity by a motor. The perturbation at any beam position causes a frequency shift that is proportional to the square of the local ﬁeld. The frequency shift can easily be measured, by driving the cavity using a phase-locked loop that locks the driving frequency to the resonant frequency of the cavity. The measurement determines the ratio of squared ﬁeld to the stored energy. The stored energy can be determined from

5.14 Quasistatic Approximation Nylon line

Bead RF drive Motor driven pully

RF pickup Cavity

Generator

Frequency detector

Figure 5.15 A schematic diagram of a ﬁeld-measuring apparatus based on the Slater perturbation method.

the relation U = QP/ω0 , and from separate measurements of Q and the cavity power dissipation P, using standard microwave measurement techniques. 5.14 Quasistatic Approximation

For some applications involving time-dependent electromagnetic ﬁelds, the spatial distribution of the ﬁelds is nearly the same as for the static problem, even though the frequency may be in the microwave region. In these cases the electric-ﬁeld components can be derived from a time-dependent scalar potential that satisﬁes Laplace’s equation. This method is called the quasistatic approximation, and is usually valid when the geometrical variations are small compared to the free-space wavelength. This result follows from consideration of the wave equation in charge-free space, which for the case of harmonic time dependence becomes the Helmholtz equation ∇ 2 E + k2 E = 0

(5.93)

where the wave number is k = ω/c = 2π/λ. Expanding the Helmholtz equation into its three components gives 2 ∂ 2E ∂ 2E ∂ 2E 2π + 2 + 2 + E=0 (5.94) ∂x2 ∂y ∂z λ When ﬁeld variations develop on a scale determined by structural elements, whose spacings are small compared to the wavelength, the values of derivative terms near those elements will dominate over the last term, and the equation then reduces approximately to Laplace’s equation ∇2E ∼ =0

(5.95)

167

168

5 Microwave Topics for Linacs

The quasistatic approximation consists of neglecting the last term of the Helmholtz equation, and assuming that the electric-ﬁeld components satisfy Laplace’s equation. The electric ﬁeld must also satisfy the basic vector relationship ∇ × (∇ × E) = ∇(∇ · E) − ∇ 2 E. Since the last term is approximately zero, and in charge-free space ∇ · E = 0, Eq. (5.95) implies that ∇ × (∇ × E) = 0. The latter result is satisﬁed if E = −∇U, where U is a scalar potential, because of the vector identity ∇ × ∇U = 0. Therefore, in the quasistatic approximation, the time-dependent electric-ﬁeld components are derivable from a scalar potential, and since ∇ · E = −∇ 2 U = 0

(5.96)

the potential must also satisfy Laplace’s equation. Notice that the quasistatic approximation decouples the magnetic and electric ﬁelds in Maxwell’s equations, and this implies that the contribution to the electric ﬁelds from Faraday’s law is negligible. For the radiofrequency quadrapole or RFQ the quasistatic approximation applies very accurately to the region near the four poles, including the region of the beam, where the spacings of the poles in that region are typically only a few thousandths of the wavelength.

5.15 Panofsky–Wenzel Theorem

The Panofsky–Wenzel theorem [26] is a general theorem pertaining to the transverse deﬂection of particles by an RF ﬁeld. A transverse deﬂection in a linac is undesirable if it leads to the beam-breakup instability, or it may be desirable if the cavity is to be used as a beam chopper to pulse the beam. Consider a particle with velocity v and charge q, traveling through an RF cavity near and parallel to the axis, and assume that the particle is moving fast enough that the transverse forces, which produce a transverse momentum impulse, have a negligible effect on the trajectory in the cavity. The transverse momentum change over a length d, is approximately d q [E⊥ + (v × B)⊥ ] dz z=0 (5.97) p⊥ = v where the ﬁelds are those seen by the particle. In a cavity where the electric and magnetic ﬁelds are time-dependent, generally not in phase, and where cancellation of the electric and magnetic deﬂections is possible, it is not obvious what kind of cavity mode gives the largest deﬂection. Following the original derivation, we introduce the vector potential. In the cavity volume with no free charges, we may write the electric and magnetic ﬁelds as E=−

∂A , ∂t

B=∇ ×A

(5.98)

5.15 Panofsky–Wenzel Theorem

where A is the vector potential. We can write (v × B)⊥ = [∇(v · A) − (v · ∇)A]⊥ = ∇⊥ (v · A) − (v · ∇)A⊥

(5.99)

We substitute Eqs. (5.98) and (5.99) into Eq. (5.97), and using the fact that v is a constant, and directed along the z-axis, we have d 1 ∂A⊥ ∂A⊥ p⊥ = q − (5.100) + + ∇⊥ Az dz v ∂t ∂z 0 The total derivative of A⊥ along the path of the particle is 1 ∂A⊥ ∂A⊥ dA⊥ = + dz v ∂t ∂z Substituting Eq. (5.101) into Eq. (5.100), we ﬁnd A⊥ (z=d) d dA⊥ + q ∇⊥ Az dz p⊥ = −q A⊥ (z=0)

(5.101)

(5.102)

If we assume an e jωt harmonic time dependence for E, Eq. (5.98) requires that A = jE/ω, and Eq. (5.102) becomes [27] jq A⊥ (z=d) jq d p⊥ = −q dE⊥ + ∇⊥ Ez dz (5.103) ω A⊥ (z=0) ω 0 The ﬁrst term of Eq. (5.103) vanishes for any case with a conducting wall at z = 0 and d, such as for a pillbox cavity without beam holes. It also vanishes for any cavity with a beam pipe at each end, if z = 0 and d are chosen far enough beyond the cavity end walls that the ﬁelds have attenuated, so that approximately E⊥ = 0. For these cases jq d p⊥ = ∇⊥ Ez dz (5.104) ω 0 Equation (5.104) is the Panofsky–Wenzel theorem that says that the total momentum impulse is proportional to the transverse gradient of the longitudinal electric ﬁeld. Because Ez is the ﬁeld seen by the particle, one should substitute t = z/v. The quantities A, B, and p⊥ are in phase, and are 90◦ ahead of E. The right side of Eq. (5.104) is zero for a pure TE mode because Ez and ∇⊥ Ez = 0 are then zero everywhere in the cavity. Physically, the pure TE mode of a pillbox cavity gives zero deﬂection, because the electric and magnetic forces cancel. A pure TM mode can give a nonzero ∇⊥ Ez , and can give a nonzero transverse deﬂection. For example, the TM010 mode produces a nonzero ∇⊥ Ez for an off-axis particle, and the TM110 mode produces a large nonzero ∇⊥ Ez on axis. For a cavity with beam holes, there will generally be neither pure TE nor pure TM modes [28]. Note that the derivation did not assume that the particle had to be traveling at the extreme relativistic limit;

169

170

5 Microwave Topics for Linacs

only that it traveled fast enough that the particle trajectory was approximately unperturbed by the deﬂecting ﬁeld.

Problems

5.1.

5.2.

5.3.

A circular cylindrical cavity with radius R and length l operating in a TM010 mode is excited by a coupling loop inserted through a hole in the cylindrical wall. Assume that the loop is located at the wall and that the effective loop area A is small enough so that the magnetic ﬁeld is uniform over the loop. (a) Use the expression for the power dissipated in the TM010 mode and show that the transformed load resistance presented by the cavity at resonance to the line is RL = [ωµ0 A]2 /[2πR( + R)Rs ]. (b) Consider a room-temperature copper cavity operated in the TM010 mode with a radius of 0.287 m, and a length of 0.345 m. Calculate the effective loop area required so that at resonance the cavity presents a matched load to a coaxial line with a 50 characteristic impedance. (c) Suppose we consider the cavity and loop as an equivalent transformer that transforms the axial cavity voltage V0 = E0 to the loop voltage V = ωBA. Show that the effective transformer–turns ratio n = V0 /V = c/(J1 (2.405)ωA). Using the result J1 (2.405) = 0.5191 and the other numerical values of this problem, calculate the numerical value of n. A spherical conducting bead of diameter 3/16 in. is placed near the outer cylindrical wall of a cylindrical cavity of length l = 0.5 m, operating in a TM010 mode at a resonant frequency of 400 MHz. (a) Assuming we can ignore any effect from the electric ﬁeld near the outer wall, and evaluating the magnetic ﬁeld at the outer wall, show that the fractional frequency shift is ω0 /ω0 = (3/4)V/(πR2 ) where V is the volume of the bead, and R is the cavity radius. (b) Calculate the numerical value of the shift in the resonant frequency f0 = ω0 /2π. (c) Suppose the bead is moved to a new radius where the unperturbed magnetic ﬁeld is larger by 1% for the same excitation level of the cavity. Calculate the change in the resonant frequency shift. (d) If the cavity length was increased by a factor of 4 and the cavity-ﬁeld level remains the same, how would the frequency shifts be affected? Would you expect the relative ﬁeld changes obtained from the measurements of frequency shifts to be as easy to measure for a longer cavity? An accelerator cavity operates in the TM010 mode at a resonant frequency of 400 MHz, and has a measured (unloaded) Q0 of 40,000. The cavity is matched to a transmission line with a characteristic Z0 = 50 , through which the generator supplies an average power P = 12 kW to excite the cavity to a peak axial voltage of V0 = 0.5 MV. (a) Use the axial voltage and the power to calculate the cavity shunt resistance R, and the value of R/Q0 . (b) The input power coupler behaves like a transformer with turns ratio n, transforming the low voltage in the input transmission

Problems

5.4.

5.5.

5.6.

5.7.

line to the high voltage in the cavity. Use the shunt resistance R and the cavity load resistance at the transmission line to calculate the effective transformer–turns ratio n. (c) Use the value of n to calculate the peak voltage of the incident wave in the input transmission line. What is the corresponding peak forward current needed to supply the cavity power? A cavity, operating at a resonant frequency of 350 MHz with an unloaded quality factor Q0 = 20,000, is excited on resonance by an RF source through a transmission line and an input power coupler. The input coupler is not adjustable, and to measure the degree of coupling the RF source is turned off and the time constant of the cavity decay is determined from measurement to be τ = 12.0 µs. What are the loaded and external quality factors, QL and Qex , and the coupling factor β. Assuming all other ports connected to the cavity are weakly coupled, is the cavity undercoupled, matched, or overcoupled to the input transmission line? A cavity is characterized by Q0 = 20,000, an R/Q parameter (ratio of shunt resistance of R to Q0 ) = 250 , and the input transmission line is matched at its characteristic impedance value of Z0 = 50 , when looking back from the power coupler. The cavity coupling to the transmission line is characterized by the coupling factor β = 2. (a) Draw the equivalent circuit as seen by the generator circuit (primary) and show the numerical values of the resistances. (b) Draw the equivalent circuit as seen by the cavity (secondary) and show the numerical values of the resistances. Consider a cavity with R/Q = 240 and Q0 = 20,000 and assume that the driving generator emits a forward wave P+ that supplies the cavity power Pc with reﬂected power P− at the power coupler. (a) Using the coupling factor β = 2, calculate the reﬂection coefﬁcient of the input transmission line. (b) Assuming the cavity voltage is V = 1 MV, calculate the power dissipated in the cavity. (c) Calculate the forward and reﬂected power in the transmission line. (d) Calculate the voltages in the transmission line, including the forward and reverse waves and the voltage at the load (equal to the transformed cavity voltage). A cavity operating on resonance is coupled to an input transmission line with a coupling factor β1 , and to an output transmission line with a coupling factor β2 . The cavity resistance transformed into the generator circuit can be shown to equal Z0 β1 , and the transformed output resistance is Z0 β1 /β2 . These two resistances combined in parallel constitute the load resistance for the input line. (a) Express this input load resistance as a function of Z0 , β1 , and β2 , and show that the reﬂection coefﬁcient is given by = (β1 − (1 + β2 ))/(β1 + β2 + 1). (b) If β2 = 0.01, calculate the value of β1 that makes the reﬂection coefﬁcient equal to zero. (c) Suppose the input and output coupling factors are adjusted so that β1 = β2 = 1. Calculate the reﬂection coefﬁcient in the input transmission line. Is the cavity matched to the input line? If β1 could be adjusted for a match, what value should it be?

171

172

5 Microwave Topics for Linacs

5.8.

A room-temperature pillbox cavity is designed to operate in a TM010 mode at 400 MHz at the highest surface electric ﬁeld Es allowed by the Kilpatrick criterion, modiﬁed to allow an additional bravery factor b = 2.0, such that Es = b EK . (a) Calculate the Kilpatrick ﬁeld EK , and the peak surface electric ﬁeld Es . Where is the electric ﬁeld maximum for this mode? (b) Calculate the corresponding maximum value for the maximum surface magnetic ﬁeld. Where on the cavity wall does this occur? 5.9. Suppose the gap length of the ﬁrst gap of the LANSCE 201.25-MHz drift-tube linac is g = 1.05 cm. Assume a plane-parallel geometry is a good approximation for the opposing drift tubes. (a) Calculate the minimum and maximum gap voltage and electric ﬁeld for which the lowest n = 0 multipacting mode can occur. (b) What is the maximum ﬁeld value, above which no multipacting of any order can occur in the gap? 5.10. (a) Derive the nonzero ﬁeld components in complex exponential form for the TM010 mode of a cylindrical cavity resonator using the vector potential Ax , = 0, Ay = 0, and Az = E0 J0 (kr r)je jωt /ω, where ω = kr c, and show that they are Ez = E0 J0 (kr r)e jωt , and Bθ = E0 J1 (kr r)je jωt /c. (b) Consider a cavity with length d βλ, and a particle crossing the cavity parallel to the axis at ﬁxed radius r. Calculate the transverse momentum delivered to the particle by the cavity RF ﬁelds, directly from the Lorentz force, Eq. (5.97), and from the Panofsky–Wenzel theorem, Eq. (5.104), and show that these are equal. Although the Panofsky–Wenzel theorem expresses the transverse deﬂection as a function of an electric-ﬁeld gradient, is the transverse deﬂection for this problem caused by the electric or the magnetic force? 5.11. Suppose a generator launches an RF wave in a coaxial transmission line to drive an initially unexcited cavity. Assume the line is coupled to the cavity through a coaxial antenna-type coupler, which, in the extreme undercoupled limit, is assumed to be an open circuit at its electrical end (z = 0). (a) As was discussed in Section 5.6, ignoring the details of the wavefront propagation, write the expression for the total wave in the line for times t > 0 as a sum of three traveling waves, an incident wave, a direct-reﬂected wave, and a time-dependent radiated wave from the cavity. Assume that the peak of the total voltage wave at t = 0 occurs at z = 0. (b) Express the total wave in the undercoupled limit, when β = 0, and show that it is a pure standing wave. (c) Express the total wave at critical coupling, when β = 1. What is the nature of this wave at both t = 0 and ∞? (d) Express the total wave in the overcoupled limit, β → ∞. What is the nature of this wave at both t = 0 and ∞? Show that momentarily at time t = τ ln 2, there is zero effective reﬂected wave (only an incident wave). 5.12. An RF window is used to separate the beam-line high vacuum from atmospheric-pressure air in the RF waveguide that electromagnetically

References

connects the generator to the cavity. The windows are made from an insulating material such as ceramic, and their location in the waveguide should be chosen to minimize high electric ﬁelds, which can produce damage from thermal stress caused by dielectric heating. Consider the RF system and cavity in problem 5.11, and describe the locations of the voltage minima in the coaxial line for the different coupling regimes: (a) undercoupled limit, (b) critical coupling, and (c) overcoupled limit. For any of these cases do the voltage minima change with time t? References 1 To conform to common terminology,

4 5

the accelerator shunt impedance is deﬁned as twice the circuit shunt impedance. Condon, E.U., J. Appl. Phys. 11, 502 (1940); Condon, E.U., J. Appl. Phys. 12, 129 (1941). A common example of a suitable coupling mechanism is a loop at the end of a coaxial transmission line that is inserted through a small hole in the cavity wall. Energy conservation can be used to justify the damping term. We will learn later that for a waveguide-cavity system, this is a virtual power that is dissipated when the generator is suddenly turned off. For a waveguide-cavity system, this deﬁnition needs to be qualiﬁed. The power Pex that enters in the deﬁnition of β is deﬁned as the external power radiated from the cavity, when the generator is turned off. P− is generally not equal to Pex , unless the generator is off and the cavity ﬁelds are decaying. Strictly speaking, we will ﬁnd that this is not simply a reﬂected wave, but is really the superposition of a reﬂection from the input coupler, plus the emitted wave from the cavity. This amplitude is slightly less than V+ because a small fraction of incident energy is transmitted to the cavity. The approach of this section closely follows that given in Altman, J.L., Microwave Circuits, D. Van Nostrand, Princeton, N.J., 1964.

11 Ramo, S., Whinnery, J.R. and Van

Duzer, T., Fields and Waves in Communication Electronics, 2nd ed., John Wiley & Sons, New York, 1984. The cancellation of reﬂected and radiated waves that produces the critical-coupling condition occurs only for the resonant frequency. For the transient behavior, while the cavity ﬁelds are building or decaying, there is power dissipated in the external load, because other frequencies besides the resonant frequency will contribute. A summary of accelerator issues for RF power systems is given by Taylor, C.S., in Linear Accelerators, Lapostolle, P.M. and Septier, A.L., North Holland Publishing, 1970, pp. 905–933. Ziomek, C.D., Regan, A.H., Lynch, M.T. and Bowling, P.S., Low-level RF control system issues for an ADTT accelerator, Int. Conf. Accelerator Driven Transmutation Technol. Appl., Las Vegas, NV, July, 1994, ed., Arthur, E.D., Rodruiguez, A. and Schriber, S.O., AIP Conf. Proc. 346, p. 453. Hatch, A.J. and Williams, H.B., Phys. Rev. 112, 68 (1958). It is common usage to refer to this phenomenon as either multipacting or multipactoring. Klein, U. and Proch, D., Proc. Conf. on Future Possibilities for Electron Accel., eds. McCarthy, J.S. and Whitney, R.R., Charlottesville, VA, 1979. Kilpatrick, W.D., Rev. Sci. Instr. 28, 824 (1957).

173

174

5 Microwave Topics for Linacs 18 Boyd, T.J. Jr., Kilpatrick’s Criterion, Los

Alamos Group AT-1 report AT-1:82-28, February 12, 1982. Palmer, R.B., Prospects for high energy e+ e− linear colliders, Ann. Rev. Nucl. Part. Sci. 40, 568 (1990). Adiabatic in this context should not be confused with its common use in thermodynamics, where it is associated with rapid processes. However, the two seemingly contradictory uses of this term may not really be incompatible. In physics textbooks, adiabatic processes are deﬁned as being sufﬁciently rapid that the body is thermally isolated. However, the condition of slowness compared to the internal time constants is also tacitly assumed for any reversible process that takes the system through quasi-equilibrium states. The reader is referred to the discussion by Landau, L.D. and Lifsh*tz, E.M., Statistical Physics, 2nd ed., Pergamon Press, 1970, p. 38. Landau, L.D. and Lifsh*tz, E.M., Mechanics, 3rd ed., Pergamon Press, 1989, p. 154. See Boltzmann, L., Vorlesungen uber Mechanik II, J. A. Barth, Leipzig, 1903, p. 48; Ehrenfest, P., Ann. Phys. 36, 91 (1911); Ehrenfest, P., Proc. Amsterdam Acad. 16, 591 (1914); Maclean, W.R., Q. Appl. Math. 2, 329 (1945). Slater, J.C., Microwave Electronics, D. Van Nostrand Company, New York, 1950, pp. 80–81.

24 Slater’s derivation does not assume

the process is adiabatic. However, a presentation of the theorem in the context of an adiabatic theorem is given by Klein, H., CERN Accelerator School on RF Engineering for Particle Accelerators, CERN 92-03, Vol. 1, 1992, p. 112–118. Maier, L.C., Jr. and Slater, J.C., J. Appl. Phys. 23(No. 1), 68–83 (1952); Klein, H., CERN Accelerator School on RF Engineering for Particle Accelerators, CERN 92-03, Vol. 1, 1992, pp. 112–118. Panofsky, W.K.H. and Wenzel, W.A., Rev. Sci. Instrum. 27, 967 (1956). The treatment presented here follows that of M. Jean Browman, Los Alamos, 1991, private communication. The phase of A is obtained simply by using the identity j = exp[jπ/2] so that A is proportional to exp[j(ωt + π/2)]. Thus A is π/2 ahead of E. The original paper discusses the theorem only in the context of pure TE and TM modes, which at that time may have represented the only available solutions, and those solutions only applied to simple pillbox cavities without beam holes. In general, the introduction of axially symmetric beam holes results in solutions with mixed TE and TM terms, except for axially symmetric modes such as the TM010 mode.

175

6 Longitudinal Particle Dynamics An ion linac is designed for acceleration of a single particle, which remains in synchronism with the accelerating ﬁelds and is called the synchronous particle. For acceptable output beam intensity, restoring forces must be present so that those particles near the synchronous particle will have stable trajectories. Longitudinal restoring forces are produced when the beam is accelerated by an electric ﬁeld that is increasing with time, and these forces produce phase and energy oscillations about the synchronous particle. The ﬁnal energy of an ion that undergoes phase oscillations is approximately determined, not by the ﬁeld, but by the geometry of the structure that is tailored to produce a speciﬁc ﬁnal synchronous energy. An exception is an ion linac built from an array of short independent cavities; each one is capable of operating over a wide velocity range. In this case, the ﬁnal energy depends on the ﬁeld and the phasing of the cavities. The longitudinal dynamics is different for relativistic electron linacs. After beam injection into electron linacs, the velocities approach the speed of light so rapidly that hardly any phase oscillations take place. The electrons initially slip relative to the wave and rapidly approach a ﬁnal phase that is maintained all the way to high energy. The ﬁnal energy of each electron with a ﬁxed phase depends on the accelerating ﬁeld and on the value of the phase. In this chapter we will develop the general equations of longitudinal dynamics in a linac, including the Liouvillian gap transformations used in computer simulation codes. This will be followed by special treatments of the dynamics in coupled-cavity ion linacs, independent-cavity ion linacs, and low-energy electron beams injected into conventional electron-linac structures. 6.1 Longitudinal Focusing

Longitudinal focusing is provided by an appropriate choice of the phase of the synchronous particle relative to the crest of the accelerating wave. A longitudinal restoring force exists when the synchronous phase is chosen corresponding to a ﬁeld that is rising in time, as shown in Fig. 6.1. For this case, the early particles experience a smaller ﬁeld and the late particles a larger ﬁeld RF Linear Accelerators. 2nd, completely revised and enlarged edition. Thomas P. Wangler Copyright  2008 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-40680-7

176

6 Longitudinal Particle Dynamics

than the synchronous particle. The accelerated particles are formed in stable bunches that are near the synchronous particle. Those particles outside the stable region slip behind in phase and do not experience any net acceleration. To develop these ideas, we consider an array of accelerating cells (Fig. 6.2), containing drift tubes and accelerating gaps, designed at the nth cell for a particle with synchronous phase φsn , synchronous energy Wsn , and synchronous velocity βsn . We express the phase, energy, and velocity of an arbitrary particle in the nth cell as φn , Wn , and βn . The particle phase in the nth cell is deﬁned as the phase of the ﬁeld when the particle is at the center of the nth gap, and the particle energy for the nth cell is the value at the end of the nth cell at the center of the drift tube. We now investigate the motion of particles with phases and energies that deviate from the synchronous values. We assume that the synchronous particle always arrives at each succeeding gap at the correct phase, and we consider particles with velocities that are close enough to the synchronous velocity that all particles have about the same transit-time factor.

Late particle

Stable particle

Early particle

Figure 6.1

Stable phase.

ln −1

bn −1

fn −1

Figure 6.2

Accelerating cells for describing the longitudinal motion.

6.2 Difference Equations of Longitudinal Motion for Standing-Wave Linacs

We consider the particle motion through each cell to consist of a sequence of drift spaces where the particles see no forces, follow thin gaps where the forces are applied as impulses. The thin gaps are assumed to be located at the geometrical center of each cell. From gap n − 1 to gap n, the particle has a constant velocity βn−1 , as indicated in Fig. 6.2. The RF phase changes as the particle advances from one gap to the next according to 2n−1 π for π mode (6.1) + φn = φn−1 + ω 0 for 0 mode βn−1 c where the half-cell length is 1 n−1 = Nβs,n−1 λ/2,

where N =

for π mode

2 1 for 0 mode

(6.2)

By deﬁnition of the synchronous velocity, the cell length, measured from the center of one drift tube to the center of the next, must be Ln = N(βs,n−1 + βs,n )λ/2. The phase change during the time an arbitrary particle travels from gap n – 1 to gap n relative to that of the synchronous particle is 1 1 (6.3) − (φ − φs )n = φn − φs,n = 2πNβs,n−1 βn−1 βs,n−1 Using the Taylor expansion to write 1 1 1 ∼ δβ 1 − − = = − 2, β βs βs + δβ βs βs

for δβ 0 and defocusing when K < 0. Denoting the transfer matrix by R, the following results can be derived. (a) Drift space (K = 0): 1 R= (7.50) 0 1 where is the drift length.

|qG| (b) Focusing quadrupole K = >0 : mcβγ √   √ sin K √   cos K K  R= √ √ √ − K sin K cos K

(7.51)

where is the length, and G is the quadrupole gradient. qG (c) Defocusing quadrupole K = 2 correspond to overfocusing and are unstable. It is customary, following the deﬁnitions introduced by Smith and Gluckstern, but changing the notation to correspond to modern usage, to introduce the ˜ quantities β˜max , the value of β˜ at the center of the focusing lens, and βmin , the value of β˜ at the center of the defocusing lens. We also deﬁne ψ = β˜max /β˜min , which is called the ﬂutter factor, = /( + 2d) the quadrupole ﬁlling factor, and N, the polarity reversal index, where N corresponds to the number of quadrupoles between polarity alternations. For example, N = 1 for FODO, N = 2 for FOFODODO, and so on. The Smith and Gluckstern stability charts for a DTL with = 0.5 for the FODO, or N = 1 case is shown in Fig. 7.7a, and for N = 2 is shown in Fig. 7.7b. In the notation of Smith and Gluckstern, the symbol γN is used, where γN = β˜max c/vλ and µN is used instead of σ . We are only concerned with the lowest passband, because it alone has sufﬁcient width for a practical focusing system. The parameters can be chosen by using the stability chart to pick an operating line as a function of particle phase, so that all particles in the bunch lie within the stable region. The use of larger N tends to reduce the required quadrupole gradients, but the height and width of the stable region also decrease with increasing N. If one uses the criterion that the strongest focusing is obtained when the phase advance per unit length is maximum, it is found that larger N is preferable. However, including the additional constraint that for stability, the phase advance is limited to σ < π, it is found that N will be limited. The most typical choices are N = 1 or 2. Finally, there is a resonance between the radial and phase motion, when the longitudinal oscillation frequency is twice the transverse oscillation frequency. In general it is not difﬁcult to avoid this resonance [15].

7.12 Effects of Random Quadrupole Misalignment Errors

During the commissioning of a newly constructed accelerator, most large errors are discovered and corrected. Nevertheless, there are always practical or economic limits to an attainable precision, and within such limits, the parameters may exhibit random deviations from their design values. To achieve the desired performance, the effects on the beam, caused by such random errors, must be understood. Some very useful results, showing the effects of random errors, have been obtained by Crandall for periodic or quasiperiodic accelerator systems [16]. In this section we discuss random quadrupole misalignments.

7.12 Effects of Random Quadrupole Misalignment Errors

cos m1 = −1 g1 = ∞ y = ∞ 1

g1 = 5

q 20

y1 = 2

g1 = 3.5

g1 = 5

g1 = 3

y1 = 1.5

cos m1 = 1 g1 = ∞

N = 1, Λ = 0.5 Stability limits g1 = Constant y1 = Constant −0.20

−0.15

y1 = 1

−0.10

−0.05

0.05

0.10

0.15

∆

(a)

2.0

cos m2 = −1 g2 = 10

1.6

q 20 1.2

g2 = ∞

y2 = ∞

y2 = 3 g2 = 10 cos m2 = 1

g2 = 7

g2 = ∞

0.8

0.4

g2 = 1.5

N = 2, Λ = 0.5 Stability limits g2 = Constant y2 = Constant

g2 = 4

y2 = 1 0 (b )

−0.20

−0.15

−0.10

−0.05

0.05

0.10

0.15

∆

Figure 7.7 Stability chart for transverse motion from Smith and Gluckstern for (a) N = 1 and (b) N = 2. In the notation of Smith and Gluckstern, γN = β˜max c/vλ and µN is used instead of σ . The other notation

First, we consider the effect of random transverse misalignments of thinlens quadrupoles in a beam transport line, when no acceleration is present. If there are no misalignments, but the injected beam centroid is displaced from the axis, the centroid trajectory will behave as a single particle. If the beam

219

220

7 Transverse Particle Dynamics

line comprises a sequence of randomly misaligned quadrupoles, the beam centroid will receive deﬂections in every quadrupole, and the evolution of the centroid may be viewed as a random walk. It is found that on average the area of the trajectory ellipse on which the centroid moves will grow as the beam passes through more quadrupoles. The process is a statistical one, and a mean ellipse area may be deﬁned at any point along the beam. If the amplitude A of the centroid motion in the x plane is deﬁned as the maximum displacement in a period, we can write A2 = εβ˜max , where β˜max is the maximum value of the Courant–Snyder ellipse parameter, and ε is the ellipse area divided by π. If the mean value of ε is ε, the mean square value of the amplitude at any location along the beam line is

A2 = β˜max ε

(7.78)

For a system of identical thin-lens quadrupoles, after N periods one obtains

εN = ε0 + N (δx)2

n β˜i f2 i=1 i

(7.79)

where fi is the focal length of the ith lens in a period, given by 1/fi = qG /mcβγ , β˜i is the Courant–Snyder ellipse parameter at the ith lens, ε0 is the initial ellipse area, which is zero if the beam is aligned on the axis initially, δx is the quadrupole displacement error, and n is the number of lenses in a single period. Equation (7.79) shows that misalignment of the focusing lenses is more serious than that of the defocusing lenses, because β˜i is larger in the focusing lenses. Furthermore, misalignments are more serious as the strength of the lenses, as measured by 1/fi , increases. Substituting Eq. (7.79) in Eq. (7.78), the root-mean-square (rms) value of the amplitude after N periods is n β˜i (7.80) AN,rms = A2N = A20 + N β˜max (δx)2 f2 i=1 i For identical thin-lens quadrupoles in a FODO lattice, Eq. (7.79) simpliﬁes to

εN = ε0 +

N (β˜max + β˜min ) (δx)2 f2

(7.81)

where β˜max and β˜min are evaluated at the focusing and defocusing lenses, respectively. The rms value of the amplitude after N periods is AN,rms =

A2N =

A20 +

N β˜max

(δx)2 (β˜max + β˜min ) f2

(7.82)

These equations can be modiﬁed to include the effects of quasiperiodicity, and of acceleration. For quasiperiodic systems, where the parameters are

7.13 Ellipse Transformations

slowly changing with energy, the second term of Eq. (7.80) is modiﬁed. Also, when acceleration is included, the ellipse area is no longer constant in the absence of errors, but decreases, since it is the ellipse area in the phase space of position and momentum that is constant. The general result, including all these effects, is n β˜N (β0 γ0 ) β˜max (δx)2 qG 2 β˜i 2 2 AN,rms = AN =

A0 + (βN γN ) mc (βi γi ) β˜0 (βN γN ) i=1 (7.83) Crandall has studied numerically the distribution of rms amplitudes of the centroid, associated with a randomly displaced FODO sequence of thin lenses. The results are only weakly dependent on the number of lattice periods. It was found that between 60 and 65% of the cases have amplitudes less than AN,rms , about 90% have amplitudes less than 1.5AN,rms , and about 97.5% have amplitudes less than 2AN,rms . Although the amplitude of the centroid oscillation produced by misaligned quadrupoles is generally larger for stronger quadrupoles, weaker quadrupoles result in an increase in the width of the beam, which suggests that when errors are taken into account, there is an optimum focusing strength. 7.13 Ellipse Transformations

We have discussed the transfer-matrix formalism to transport particle coordinates between two points in an accelerator. The formalism is valid when the forces acting on the beam are linear. The matrix formalism can also be extended to transport the beam ellipses between two locations, as shown in Fig. 7.8. The general equation of an ellipse in x − x phase space is ˜ 2 = ε γ˜ x2 + 2αxx ˜ + βx

(7.84)

where γ˜ β˜ − α˜ 2 = 1. It is convenient to express this equation in matrix form as X T σ −1 X = ε x where X = , X T = [xx ] is the transpose matrix, and we deﬁne x γ˜ α˜ σ −1 = α˜ β˜ The inverse matrix of σ −1 is β˜ −α˜ σ = −α˜ γ˜

(7.85)

(7.86)

(7.87)

221

222

7 Transverse Particle Dynamics ~

~ a 1,b1,g1

x′

a2,b2,g~ 2

x′ e

Figure 7.8 Phase-space ellipse transformation between two locations.

We know that for any particle the coordinate transformation from position 1 to 2 can be expressed in the form of a 2 × 2 transfer matrix R, such that X2 = RX1

(7.88)

The equation for the ellipse at location 1 is X1T σ1−1 X1 = ε, and that for the ellipse at location 2 is X2T σ2−1 X2 = ε, where, because of the linear transformation, the ellipse emittance is assumed to be unchanged. It can be shown that σ2 is related to σ1 , as σ2 = Rσ1 RT

(7.89)

An equivalent expression, obtained by carrying out the above matrix multiplication, is

β˜ R2 −2R11 R12 R212 β˜1 2 11 (7.90) α˜ 2 = −R11 R21 1 + 2R12 R21 −R12 R22 α˜ 1 γ˜2 R221 −2R21 R22 R222 γ˜1 Equation (7.89) or (7.90) is useful for designing the matching optics to inject a matched beam ellipse into an accelerator with periodic focusing. They are valid even in the presence of space-charge forces, as long as those forces are linear. 7.14 Beam Matching

If the injected beam ellipse is not matched to the focusing system, there will be additional oscillations of the rms beam projections, which produce a larger beam at some locations and a smaller beam at other locations. Generally, beam matching means that the beam-density contours coincide with the ellipses corresponding to particle trajectories. For this discussion, we assume that the beam is described by a well-deﬁned phase-space ellipse. In Chapter 9, this is reﬁned in terms of an rms deﬁnition for the beam ellipse. At every point in the focusing lattice, there exists a matched beam condition, deﬁned by the Courant–Snyder ellipse parameters α˜ m , β˜m , and γ˜m . In general, the beam will have ellipse parameters that may differ from the matched values.

7.14 Beam Matching

It is convenient to deﬁne a mismatch factor, [17] which is a measure of the increase in the maximum beam size, resulting from a mismatch. Suppose that the matched beam ellipse is deﬁned by γ˜m x2 + 2α˜ m xx + β˜m x2 = ε

(7.91)

and a mismatched beam ellipse with the same area is deﬁned by ˜ 2 = ε ˜ + βx γ˜ x2 + 2αxx The mismatch factor is deﬁned by √ 1/2 + ( + 4) M = 1+ −1 2

(7.92)

(7.93)

where ˜ γ˜ = (α) ˜ 2 − β

(7.94)

and α˜ = α˜ − α˜ m , β˜ = β˜ − β˜m = γ˜ = γ˜ − γ˜m . It is useful to express as = − det σ where

σ =

β˜ −α˜

(7.95)

−α˜ γ˜

(7.96)

is the difference between the actual beam matrix σ , and the matched beam matrix σm , given by β˜m −α˜ m σm = (7.97) −α˜ m γ˜m Although all physics effects associated with beam mismatch cannot necessarily be described in terms of the mismatch factor alone, the mismatch factor is nevertheless a very useful parameter. It can be shown that the mismatch factor has the property that the maximum beam size for the mismatched beam is larger than that for a matched beam by a factor 1 + M. Thus, a mismatched beam with M = 0.1 would have a 10% larger maximum projection. Finally, we note that another deﬁnition of mismatch factor is found in the literature, [18] based on an increase in the effective phase-space area resulting from mismatch. With this deﬁnition, Eq. (7.93) is replaced by an equation of the form √ + ( + 4) (7.98) M = 2 √ and when 1, M ≈ .

223

224

7 Transverse Particle Dynamics

7.15 Current-Independent Beam Matching

Crandall has discussed the problems associated with beam matching in ion linacs, especially when there are few knobs or adjustable elements [19]. He considered both transverse and longitudinal matching and proposed guidelines for intertank matching that lead to matching solutions that are insensitive to beam current and emittance. These can be achieved by avoiding sharp discontinuities in the focusing systems along the linac. If the linac is designed in this way, it is found that a beam that is matched at the radiofrequency-quadrupole (RFQ) entrance will remain matched throughout the linac approximately independent of beam current and emittance. This often implies that the initial accelerating gradient of a DTL that follows an RFQ must be reduced to make the longitudinal focusing strength comparable with that of the ﬁnal RFQ focusing strength. Another consequence is that longitudinal matching between DTL tanks may require changes in the geometry of the cells at the ends. Fortunately, this method of designing linacs can be applied at zero beam current; space-charge forces do not need to be considered. Because the method results in approximate current-independent matching, the results of the zero-current analysis still apply at full beam current. Figure 7.9 is a schematic depiction of the typical matched beam conditions at the interface between the RFQ and the ﬁrst DTL tank, where the drift tubes contain quadrupole lenses (FODO), and the distance between the gap centers is βλ. The matched transverse ellipses at location A are similar to those at location C, which is at center of the ﬁrst gap of the DTL. Figure 7.10 shows longitudinal matching between two DTL tanks.

RFQ A

DTL D

Plane x – x′ y – y′ f–w

Figure 7.9 Matched phase-space conﬁgurations at the end of an RFQ and the beginning of a DTL (courtesy of K. R.Crandall).

7.16 Solenoid Focusing Tank N + 1

Tank N A

∆Φ

Figure 7.10 Longitudinal matching between two DTL tanks. Solid lines show normal drift tubes and phase proﬁles, and dashed lines show drift tubes and proﬁles as modiﬁed for matching (courtesy of K. R.Crandall).

7.16 Solenoid Focusing

The longitudinal magnetic ﬁeld produced by a solenoid has often been used for ion and electron focusing. Comparison of a solenoid lens with a single quadrupole is not valid, since at least two quadrupoles are needed to produce focusing in both planes. Compared with quadrupole doublets or triplets of the same performance, it is usually found that the power dissipation in a solenoid is much higher. Solenoid focusing is a consequence of the interaction between the azimuthal velocity component induced in the entrance fringe-ﬁeld region, and the longitudinal magnetic ﬁeld component in the central region [20]. At the ends of the solenoid, the ﬁeld ﬂares out and there is an interaction between the radial ﬁeld component and the axial velocity component, producing an azimuthal acceleration. In this way the particle acquires the azimuthal velocity component in the entrance fringe-ﬁeld region. In the central region, particles traveling parallel to the ﬁeld are unaffected, but those with an azimuthal velocity component will experience a force causing them to describe an orbit that is helical in space, and circular when viewed from the end of the solenoid. The radial motion can be approximately described in the smooth approximation as d2 r + ds2

qBrms 2γ mv

k2 0 − r=0 2

(7.99)

225

226

7 Transverse Particle Dynamics

where Brms is the rms value along the axis, and k 0 is the wave number for longitudinal motion, deﬁned earlier, which is the equation for a simple harmonic oscillator. 7.17 Smooth Approximation to Linac Periodic Focusing

To simplify the description of the periodic transverse focusing, it is useful to introduce a smooth approximation, which averages over the rapid ﬂutter associated with the individual focusing lenses. There are different versions of the smooth approximation. In this section, an approximate formula for the transverse phase advance per focusing period is obtained by using the matrix method to form the transport matrix over one full period, and then treating both the RF gaps and the quadrupoles as thin lenses. Retaining only the lowest-order terms, we obtain simple approximate formulas. Recall that the thin-lens focal length for the RF gap is given by πqE0 T sin(−φ)L 1 x = = fg x mc2 λ(γβ)3

(7.100)

where L is the length over which the RF impulse is integrated, which is the distance between the gaps. The quadrupole thin-lens focal length is given by qG x 1 =± = fQ x mcγβ

(7.101)

For a FODO lattice, where the length L is equal the spacing between lenses, and the period is 2L; we obtain to lowest order an expression for the phase advance per period qG L 2 πqE0 T sin(−φ)(2L)2 σ02 ∼ − (7.102) = mcγβ mc2 λ(γβ)3 The ﬁrst term is the quadrupole term, and the second is the RF-defocus impulse, which subtracts from the quadrupole term when φ is negative. For stability, the quadrupole term must be larger than the RF-defocusing term. We see that the RF-defocusing term falls off with βγ faster than the quadrupole term, so that the RF defocusing is relatively more important at low velocities. The phase advance per unit length, k0 = σ0 /2L, is an effective wave number for transverse or betatron oscillations. In a smooth approximation, the trajectory is sinusoidal, corresponding to simple harmonic motion, and the local ﬂutter of the trajectories caused by the individual lenses is averaged out. The smoothed wave number is σ 2 qG 2 πqE T sin(−φ) 0 0 ∼ − (7.103) k20 = = 2L 2mcγβ mc2 λ(γβ)3

7.18 Radial Motion for Unfocused Relativistic Beams Table 7.1 Comparison of lens and lattice properties Lens

Lattice

Magnetic quad singlet

FODO

Magnetic quad singlet

FOFODODO

Magnetic solenoid

Magnetic quad doublet

FDO

1/f

Period

qG γ mv

qG γ mv qB 2 2γ mv qG γ mv

L L+D

σ0

qG L γ mv √ 2 2qG L γ mv √ qB L 2γ mv √ qG LD γ mv

σ0 qG = 2L 2γ mv √ 2 qG σ0 = 4L 2 γ mv ! σ0 qB = L 2γ mv L √ σ0 qG LD = L+D γ mv L + D

This result shows that in lowest order the phase advance per unit length is independent of L and of the focusing period. But, as L increases, σ0 increases and eventually σ0 reaches the upper limit for stability at σ0 < π. In Table 7.1, we present thin-lens approximations for various lens and lattice properties. The phase-advance formulas for the quadrupole term are to be squared and added to the RF-defocus term to obtain σ02 ∼ = σQ2 − k2 0 P2 /2. The formulas are generally accurate to within about 10% when σ0 < π/2. The RF-defocus term in Eq. (7.103) does not change for a different lattice, but in Eq. (7.102), the period 2L must be replaced by the appropriate one, denoted by the symbol P in the table. For doublet focusing, the two interlens spacings, L and D, alternate along the lattice, one is larger and the other is smaller. Electric-quadrupole results can be obtained from any of the magnetic-quadrupole results by substituting E/va for the magneticquadrupole gradient G, where E is the pole-tip electric ﬁeld and a is the aperture radius. Table 7.1 is useful for answering questions such as whether, based on overall focusing strength, a singlet or a doublet focusing lattice is preferred. The wave number k0 = σ0 /P, which corresponds to the average of the inverse of the Courant–Snyder β function, is a measure of the focusing strength; the average beam size decreases with increasing k0 . Note that if σ0 is ﬁxed, the largest k0 corresponds to the lattice with the shortest period.

7.18 Radial Motion for Unfocused Relativistic Beams

The radial RF forces approach zero for a relativistic beam, because in that limit the RF magnetic force cancels the RF electric force. Consequently, transverse focusing is not as important for low-intensity relativistic-electron linacs. For high-intensity electron beams, focusing is required to compensate

227

228

7 Transverse Particle Dynamics

for space-charge effects at low energies, and for controlling the beam-breakup instability at higher energies. In this section, we consider the radial trajectories for electrons in a linac that experience no RF transverse ﬁelds and no transverse focusing. With no transverse ﬁelds, the divergence angle of each particle will decrease as the longitudinal momentum increases. Suppose we let β = 1, and if there is no transverse force, Newton’s law gives dx d γ =0 (7.104) dz dz or γ

dx = γ x = constant dz

(7.105)

which is a statement of transverse momentum conservation. We assume that the injected beam at z = 0 is relativistic, and we write the initial values as γ = γ0 , and x = x0 . Therefore transverse momentum conservation implies γ x = γ0 x0 . Assuming the particle energy increases linearly with z, we write γ = γ z + γ0

(7.106)

Now we integrate Eq. (7.105) and obtain the total transverse displacement at the end of the linac, resulting from the particle divergence, z z γ dz γ0 x0 x dz = γ0 x0 ln x = = γ γ γ 0 0 0 W0 W = x0 z ln (7.107) W − W0 W0 Because transverse momentum is conserved, the divergence angle decreases as the beam is accelerated. The trajectory is a logarithmic function of energy, as shown in Fig. 7.11. One ﬁnds that for z = 3 km (the length of the SLAC accelerator) and for an initial maximum divergence of 0.25 mrad at the energy W0 = 40 MeV, the change in displacement at a ﬁnal energy of W = 40 GeV is only x = 5 mm. This is a remarkably small deﬂection, and additional insight is obtained by seeing what the accelerator looks like in the moving beam frame. Because of Lorentz contraction, if dz is a longitudinal element of length in the laboratory frame, where the accelerator is at rest, the element in the moving frame dzm is dzm = dz/γ (z)

(7.108)

The length of the accelerator m in the moving frame is given in terms of the length in the laboratory frame by m mc2 W dz dzm = = ln (7.109) m = W − W0 W0 0 γ z + γ0 0

Problems ∆x

gmc 2

Figure 7.11 Transverse deﬂection versus energy in relative units for an accelerated relativistic beam.

For the SLAC linac using = 3 km, W0 = 40 MeV, and W = 40 GeV, we ﬁnd m = 0.27 m. Because the accelerator appears so short in the moving frame, it is not as surprising that the transverse displacement of the particles is so small.

Problems

7.1. A quadrupole magnet of effective length = 3.25 cm is operated at a gradient of 100 T/m, and is used for focusing a 5-MeV proton beam. Calculate the focusing strength K and the thin-lens focal length. 7.2. A 5-MeV proton is focused by the quadrupole magnet of Problem 7.1. The initial coordinates of the proton at the entrance plane of the quadrupole are x = 3 mm, x = 0, y = 0, and y = 0. Calculate the ﬁnal divergence angle x at the exit plane of the quadrupole, using the following assumptions: (a) The quadrupole has constant ﬁeld over the effective length and is focusing in the x plane. (b) The quadrupole has constant ﬁeld over the effective length and is defocusing in the x plane. (c) The quadrupole obeys the thin-lens approximation and is focusing in the x plane. 7.3. Consider a FODO lattice composed of the same beam and quadrupoles as in Problems 7.1 and 7.2. Assume the thin-lens approximation is valid, and assume the center-to-center spacing of the quadrupoles is 10 cm. (a) Calculate the phase advance per focusing period σ of the beam particles. ˜ and γ˜ of the trajectory ellipses in (b) Calculate the Twiss parameters α, ˜ β, the middle of a focusing lens. (c) Repeat the Twiss parameter calculation in the middle of a defocusing lens. (d) Repeat the Twiss parameter calculation in the middle of the two drift spaces. 7.4. Assume the emittance in the x plane of the ellipse on which the largestamplitude particle lies is ε1 = 30 mm-mrad (3 × 10−5 m-rad). (a) Use the values of theTwiss parameters calculated in Problem 7.3 to tabulate the " " ˜ ˜ projections ε1 β and ε1 γ˜ and the intercepts ε1 /β and ε1 /γ˜ at each of the four locations in Problem 7.3, beginning at the middle of the focusing lens. How large must the aperture be in the quadrupoles so that

229

230

7 Transverse Particle Dynamics

the largest-amplitude particle is transported? (b) Use the projections and intercepts calculated in part (a) to make a rough sketch of the ellipse at each of the four locations (middle of the lenses and of the drift spaces), beginning with the middle of the thin focusing quadrupole, with the same scales for all plots. 7.5. Suppose the particle with the maximum amplitude of Problem 7.4 has coordinates x = ε1 β˜ = 3.22 mm and x = 0 at the middle of a thin focusing lens. (a) Use the transfer matrix through a period to calculate the particle coordinates x and x at the middle of the next thin focusing lens. (b) Plot these two points on the ellipse at the thin focusing lens sketched for Problem 7.4. The phase advance from one focusing lens to the following one was calculated for Problem 7.3. Do the plotted coordinates look qualitatively consistent with this? 7.6. A FODO array of permanent-magnet quadrupole lenses with gradient G = 100 T/m and effective length = 3 cm is used to focus a low-current proton beam in a 400-MHz DTL. Assume the array is periodic with lens spacing equal to βλ (quadrupoles in every drift tube) at the injection energy of 2.5 MeV. (a) Use the smooth approximation formula in Section 7.17 for transverse phase advance per focusing period σ to calculate σ for the quadrupoles alone (as if E0 T = 0). (b) Calculate σ when E0 T = 1 MV/m and φ = −30◦ . Are the particles stable transversely? (c) For the same −30◦ phase and the same quadrupole array, what is the maximum accelerating ﬁeld E0 T for transverse stability?

References 1 Adlam, J.H., AERE-GP/M-146,

2 3

4 5

Harwell, Berkshire, England, 1953; Good, M.L., Phys. Rev. 92, 538 (1953). McMillan, E.M., Phys. Rev. 80, 493 (1950). Smythe, W.R., Static and Dynamic Electricity, 2nd ed., McGraw-Hill, 1950, pp. 13–14. Stratton, J.A., Electromagnetic Theory, McGraw-Hill, 1941, p. 116. This section outlines a treatment carried out by Gluckstern, R.L. private communication. Lapostolle, P.M., Meot, F., Valero, S. and Tanke, E., Proc. 1990 Linear Accel. Conf., Albuquerque, NM, September 10–14, 1990, LA-112004-C, p. 315. Carne, A., Lapostolle, P., Prome, M. and Schnizer, B. in Linear Accelerators, ed. Lapostolle, P.M. and Septier, A.,

11 12 13

North Holland, Amsterdam, 1970, pp. 747–783. Prome, M., thesis presented at Universite-Paris-Sud, Centre d’Orsay (1971), English translation in Los Alamos Report LA-R-79-33, 1979. Lapostolle, P.M. Proton Linear Accelerators: A Theoretical and Historical Introduction, Los Alamos Report LA-11601-MS, July, 1989. Courant, E.D., Livingston, M.S. and Snyder, H.S., Phys. Rev. 88, 1197 (1952); Courant, E.D. and Snyder, H.S. Ann. Phys. 3, 1–48 (1958). Blewett, J.P., Phys. Rev. 88, 1197 (1952). Teng, L.C., Rev. Sci. Instrum. 25, 264 (1954). Smith, L. and Gluckstern, R.L., Rev. Sci. Instrum. 26, 220 (1955).

References 14 Brown, K.L. and Servranckx, R.V. ,

AIP Conf. Proc. 127, 102, 103 (1985). 15 Prome, M., in Linear Accelerators, Lapostolle, P.M. and Septier, A.L., North-Holland Publishing and John Wiley & Sons, 1970, pp. 790–794; Hereward, H.G. and Johnsen, K., CERN Symp. High Energy Accel. Pion Phys., 1956, p. 167. 16 Crandall, K.R., Linear Error Analysis of Beam Transport Systems, AOT Division Technical Note No. LA-CP-96-16, January 22, 1966. 17 Crandall, K.R., TRACE 3-D Documentation, 2nd ed., Los Alamos

Report LA-UR-90-4146, December, 1990. 18 Guyard, J. and Weiss, M., Proc. 1976 Proton Linear Accel. Conf., Chalk River, Ontario, Canada. 19 Mills, R.S., Crandall, K.R. and Farrell, J.A., Design Self-Matching Linacs, Proc. of 1984 Linear Accel. Conf., Seeheim, Germany, 7-11, 1984, GSI-84-11, pp. 112–114. 20 Reiser, M., Theory and Design of Charged Particle Beams, John Wiley & Sons, 1994, p. 98.

231

232

8 Radiofrequency Quadrupole Linac

The radiofrequency quadrupole or RFQ is relatively a new type of linear accelerator, and its recent development was a major innovation in the linac ﬁeld. The RFQ is especially well suited for the acceleration of beams with low velocities in the typical range of about 0.01 to 0.06 times of the speed of light. As such, it is an important accelerator for ions, but not for electrons, which are already emitted from a typical electron-gun source with velocities approaching half the speed of light. All high-energy beams begin at low velocities, and the acceleration technique used at low velocities is an essential contributor to the overall accelerator performance. The RFQ can be used to accelerate high-current proton beams to several megaelectron volts, and can serve as the initial linac structure for a linac system that produces even higher energies. In this chapter, we present the principles of RFQ operation including the vane geometry and the potential function, from which we obtain analytic formulas for the ﬁelds that describe the beam dynamics. This is followed by a discussion of the design procedures including adiabatic bunching. We discuss the properties of the four-vane and four-rod structures and introduce an approach on the basis of perturbation theory to understand the effects of ﬁeld errors.

8.1 Principles of Operation

The principles of operation of the RFQ were ﬁrst presented by the inventors, Kapchinskiy and Tepliakov (K–T) in their 1969 publication [1]. The RFQ fourvane structure is shown in Fig. 8.1. K–T proposed to modify the shapes of the four electrodes of an RF quadrupole to achieve both acceleration and focusing from RF electric ﬁelds. By using a potential function description, K–T showed how to shape the electrodes to produce the ﬁelds required by the beam. The achievement of practicable means of applying velocityindependent electric focusing in a low-velocity accelerator gave the RFQ a signiﬁcant strong-focusing advantage compared with conventional linacs that RF Linear Accelerators. 2nd, completely revised and enlarged edition. Thomas P. Wangler Copyright  2008 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-40680-7

8.1 Principles of Operation

Figure 8.1 Four-vane RFQ accelerator section. The four electrodes are excited with quadrupole-mode RF voltages to focus the beam. The electrodes are modulated to produce longitudinal electric ﬁelds to accelerate low-velocity ions.

used velocity-dependent magnetic lenses. This allowed the RFQ to extend the practical range of operation of ion linacs to low velocities, thus eliminating the need for large, high-voltage dc accelerators for injection of the beam into the linac. In a later publication [2], K–T showed how to introduce speciﬁc slow variations of the RFQ parameters to bunch the beam adiabatically. This allowed the beam to be injected into the RFQ and to be bunched over many spatial periods, while the beam is contained transversely by the electric-quadrupole forces. Adiabatic bunching allows a large fraction of the beam to be captured, and converted into stable bunches that can be accelerated efﬁciently to the ﬁnal energy. Adiabatic bunching result in very compact bunches with minimal tails in longitudinal phase space, and increases the beam-current capacity, because it avoids unnecessary longitudinal compression of the beam at low velocities, which would increase the transverse space-charge effects. In the RFQ, although no drift tubes or magnetic quadrupole lenses are used, the beam is accelerated by longitudinal RF electric ﬁelds, and transversely focused by RF electric-quadrupole ﬁelds that are determined by the electrode geometry of the RFQ structure. These electrodes are described as either rods or vanes depending on the type of geometry. The operating principles of the RFQ are most easily explained by ﬁrst describing how the transverse focusing is accomplished. Suppose we consider four equally spaced conducting electrodes, symmetrically placed about the beam axis, as illustrated in Fig. 8.2. Suppose an ac voltage is applied to each electrode, whose polarity changes sign as we move from one electrode to the next. Thus, a voltage ±V0 cos(ωt)/2 is applied to the four electrodes in a quadrupolar-symmetric pattern. Off-axis particles experience a transverse

233

234

8 Radiofrequency Quadrupole Linac

−v

Figure 8.2 Electric-quadrupole cross section with electric-ﬁeld lines in the transverse plane and magnetic-ﬁeld lines in the longitudinal direction, perpendicular to the page.

electric ﬁeld that alternates as a function of time. The particles experience a time-varying electric ﬁeld with a quadrupole pattern that provides the focusing. An advantage of the RFQ focusing, compared with the more conventional magnetic focusing is that for low-velocity particles, the electric force on a particle is stronger than the magnetic force. Other advantages of the RFQ focusing will be appreciated after the more detailed treatment that follows. If the electrode geometry is uniform along the axis, there is no axial electricﬁeld component for acceleration of the particles. To see how the geometry can be modiﬁed to obtain a nonzero longitudinal electric-ﬁeld component, consider the geometry shown in Fig. 8.3, where the transverse displacements relative to the axis of the horizontal and vertical electrodes are unequal. The potential on the axis will be nonzero, because it is inﬂuenced more by the electrodes that are nearer to the axis than those of the opposite polarity that are further away. If the perturbed geometry of Fig. 8.3 is maintained constant along the axial direction, there will still be no axial electric ﬁeld, because there is no change of the axial potential. However, if the transverse-electrode displacements are varied along the axis, the axial potential will change as a

−

V0 2

ma +

V0 2

−

V0 2

Figure 8.3 Electric-quadrupole geometry with unequal electrode spacing.

8.1 Principles of Operation

function of the longitudinal position, and a corresponding axial electric ﬁeld will be produced. Suppose we consider the voltages at the time when they have their maximum value, and suppose the transverse displacement of the electrodes is varied in a sinusoidal-like pattern. If the horizontal and the vertical electrode displacements are out of phase spatially, as shown in Fig. 8.4, one would expect a sinusoidal voltage distribution along the axis, depending on the relative displacements of the horizontal and vertical electrodes. Indeed, the axial potential is maximum at the location where the electrodes have their extreme displacements, and the axial electric ﬁeld, shown in Fig. 8.5, is maximum at the locations halfway between these extreme-displacement points, where the gradient of the potential is maximum and where the four electrodes also have equal displacements.

C−C

ro a bl 2

B− B

Figure 8.4 Two views of an RFQ accelerator showing the arrangement of the four electrodes, and the unit cell of length βs λ/2 along with the parameters a, ma, and r0 that are deﬁned in the text.

235

8 Radiofrequency Quadrupole Linac 2

x /a

236

0.1

0.2

z /bsl

0.3

0.4

0.5

Figure 8.5 Electric-ﬁeld vectors in the x − z plane over half an electrode modulation cycle. This shows how the pole-tip modulation creates a longitudinal electric-ﬁeld component.

If the potential on the electrodes did not depend on time, beam particles would be accelerated and decelerated by equal amounts, and the structure would not be a practical accelerator. With time-dependent ﬁelds, what is needed to obtain a sustained acceleration of the beam particles is to produce an axial-ﬁeld pattern that results in synchronism between the accelerating ﬁelds and the particle motion. If the pattern of the electrode displacements changes through half a period during the time that the electrode voltages change sign, the axial voltage will maintain the correct sign for sustained acceleration of a synchronous particle. The spatial period of the electrode displacements must match the axial distance traversed by a synchronous particle during one RF period.

8.2 General Potential Function

Next, we analyze the RF quadrupole structure, using the quasistatic approximation. The quasistatic approximation, described in Section 5.14, is valid when the main source of the RF electric ﬁeld is the electric charge on the four conducting electrodes and the contribution to the RF electric ﬁeld from the time-changing magnetic ﬁeld, as given by Faraday’s law, can be ignored. It can be shown that the quasistatic approximation is valid, when the electrode displacements from the axis are small compared with the RF wavelength. Because of the particle-velocity increase, the vane geometry is not exactly periodic along the axis. However, it is convenient to make the approximation that the structure is periodic that corresponds to assuming that the rate of acceleration is small, compared with the distance the particle moves in one RF period. Figure 8.6 shows a cut through the x − z plane, and shows the mirror symmetry of the opposite poles about the beam axis. The minimum radius of the electrode tips is a, the maximum radius is ma, where

8.2 General Potential Function

ma x

V 2 z

βl (unit cell) 2

Figure 8.6

RFQ pole-tip geometry.

m is called the modulation parameter. The unit cell has a length βλ/2, and the longitudinal electric-ﬁeld lines ﬂow between the tip of the x electrode near z = 0 to the y tip that has a minimum radius at z = βλ/2. There are two unit cells per modulation period, and adjacent unit cells will have oppositely directed longitudinal electric ﬁelds. At a given time, every other cell can contain a bunch for acceleration. Using cylindrical-polar coordinates, the time-dependent scalar potential is written as U(r, θ, z, t) = V(r, θ, z) sin(ωt + φ)

(8.1)

where ω/2π is the RF frequency, φ is the initial phase of the potential, and V(r, θ, z) is a solution of Laplace’s equation, given in cylindrical coordinates by 1 ∂V ∂ 2V ∂ 2V 1 ∂ 2V + + 2 2 + 2 =0 2 ∂r r ∂r r ∂θ ∂z

(8.2)

where in Cartesian coordinates x = r cos θ and y = r sin θ. When the electrode displacements vary along the z-axis, we are interested in 3-D solutions that depend on z. Otherwise we are interested in 2-D solutions that are independent of z. Using the method of separation of variables for each case and adding the two solutions, one obtains V(r, θ, z) =

∞

As r 2(2s+1) cos(2(2s + 1)θ)

s=0

∞ ∞

Ans I2s (knr) cos(2sθ) sin(knz)

(8.3)

n=1 s=0

Eq. (8.3) is the general K–T potential function, from which the electric ﬁeld in the vicinity of the beam may be calculated.

237

238

8 Radiofrequency Quadrupole Linac

8.3 Two-Term Potential Function Description

As a ﬁrst impression, the RFQ may look like a complicated threedimensional structure, for which a simple analytical form for the ﬁelds would be difﬁcult to obtain. An analytical solution would have great value, because it would allow us to see immediately how the ﬁelds depend on the geometry, and this would greatly simplify our task of choosing optimized electrode geometries. Yet, we have written a general solution for the potential, which has an inﬁnite number of terms, and one wonders whether this is of any practical use. An arbitrary electrode geometry will generally require many terms from the general solution of the potential for an accurate description of the potential and the ﬁelds. However, if we could choose the electrode geometry, so that only a few terms are large near the beam, the potential description would be greatly simpliﬁed. One way to obtain a self-consistent solution is to select only the lowest-order terms from the general solution for the potential, and then construct the electrode shapes that conform to the resulting equipotential surfaces. This is the approach that was used by K–T in their original paper. They chose the pure quadrupole term (s = 0) from the ﬁrst summation of Eq. (8.3) and the monopole term (s = 0, n = 1) from the second summation. We call this the two-term potential function, which is the starting point for our discussion of the RFQ dynamics, and we write it as V(r, θ, z) = A0 r 2 cos(2θ) + A10 I0 (kr) cos(kz)

(8.4)

where A0 and A10 are constants that are determined by the electrode geometry, and k = 2π/L, where L is the period of the electrode modulation. We choose L = βs λ, where βs is the velocity of the synchronous particle to satisfy the requirement of synchronism, and we approximate the modiﬁed Bessel function as I0 (v) ∼ = 1 + v2 /4. We determine the two constants A0 and A10 as follows. Suppose that at some time the horizontal and vertical vane potentials are +V0 /2 and −V0 /2, respectively. At z = 0, where the modulation term has its maximum value, we express the horizontal (θ = 0) vane displacement as r = a, and the vertical vane displacement as r = ma, where we assume that m ≥ 1. At θ = 0 the two-term potential function is evaluated at the horizontal vane tip, which gives V0 = A0 a2 + A10 I0 (ka) 2

(8.5)

and at θ = π/2 the potential is evaluated at the vertical vane tip, which gives −

V0 = −A0 (ma)2 + A10 I0 (kma) 2

(8.6)

8.3 Two-Term Potential Function Description

Solving Eqs. (8.5) and (8.6) for the constants A0 and A10 yields A0 =

V0 I0 (ka) + I0 (kma) 2a2 m2 I0 (ka) + I0 (kma)

(8.7)

A10 =

V0 m2 − 1 2 2 m I0 (ka) + I0 (kma)

(8.8)

and

It is convenient to deﬁne the dimensionless constants X and A as X=

I0 (ka) + I0 (kma) m2 I0 (ka) + I0 (kma)

(8.9)

m2 − 1 m2 I0 (ka) + I0 (kma)

(8.10)

Then A0 = XV0 /2a2 and A10 = AV0 /2. The complete time-dependent potential is r 2 V0 cos(2θ) + AI0 (kr) cos(kz) sin(ωt + φ) X U(r, θ, z, t) = 2 a (8.11) where the time-dependent voltages on the horizontal and vertical electrodes are +V0 sin(ωt + φ)/2 and −V0 sin(ωt + φ)/2, respectively. It is convenient to express Eq. (8.11) in Cartesian coordinates x = r cos θ and y = r sin θ, which results in V0 X 2 2 U(x, y, z, t) = [x − y ] + AI (kr) cos(kz) sin(ωt + φ) 0 2 a2 (8.12) In principle, the geometry of the electrodes is now speciﬁed from the ±V0 /2 equipotential surfaces. The transverse cross sections are approximately hyperbolas. At z = βs λ/4, half way through the unit cell, the RFQ has exact quadrupole symmetry, and the tips of the x and y electrodes have a radius equal to r0 = aX −1/2 . In practice, the electrode contours must deviate from the ideal shape to control the peak surface electric ﬁeld, and to facilitate the machining. However, the electrode tips at θ = 0 and π for the horizontal electrodes, and at π/2 and 3π/2 for the vertical electrodes, can still be chosen to correspond to the correct equipotential curves of the two-term potential function. From Eq. (8.12), the x tip at y = 0 is given by 1=

X 2 x + AI0 (kx) cos(kz) a2

(8.13)

and the y tip at x = 0 is given by −1=−

X 2 y + AI0 (ky) cos(kz) a2

(8.14)

239

240

8 Radiofrequency Quadrupole Linac

Away from the tips, different transverse proﬁles can be used, including circular arcs with the same local radius of curvature as at the tip, or with a constant transverse radius of curvature throughout the entire cell. The peak surface electric ﬁeld is important, because the probability of electric breakdown increases with increasing surface ﬁeld. For a circular transverse-electrode geometry with no longitudinal modulations, the peak surface electric ﬁeld does not occur at the electrode tip, but at the point where the electrodes have minimum separation. The ﬁeld at the tip is V0 /r0 and the peak ﬁeld is Es = αV0 /r0 , where α = 1.36. For modulated poles, the value of α is slightly modiﬁed by the dependence on the transverse and longitudinal radii of curvature. The spatial dependence of the potential function within a cell is completely determined by the three parameters X, A, and k, which vary from cell to cell along the RFQ, and must be chosen to provide the desired RFQ performance. For purposes of numerical beam dynamics calculations in the four-vane RFQ structure, eight multipoles are often used to deﬁne the ﬁelds in an RFQ. The eight-term potential function is: 2 6 r r V A01 U(r, θ, z) = cos 2θ + A03 cos 6θ 2 r0 r0 + A10 I0 (kr) cos kz + A30 I0 (3kr) cos 3kz + [A12 I4 (kr) cos kz + A32 I4 (3kr) cos 3kz] cos 4θ

+ [A21 I2 (2kr) cos 2θ + A23 I6 (2kr) cos 6θ] cos 2kz (8.15) In the above equation, k = π/L, where L is the length of the cell. The Amn coefﬁcients, which depend on m and L/r0 , have been calculated and tabulated for ‘normal’ cells having several vane-tip geometries. These coefﬁcients vary along the RFQ, and are deﬁned at the end of each cell.

8.4 Electric Fields

The particle dynamics is determined from the electric-ﬁeld components. The RF magnetic ﬁeld is small in the vicinity of the beam, and produces only a small force on nonrelativistic particles. The electric ﬁeld may be obtained from the gradient of the potential function, or E = −∇U. By differentiating Eq. (8.12), the components in Cartesian coordinates are EX = −

XV0 kAV0 x x− I1 (kr) cos(kz) 2 a 2 r

(8.16)

8.5 Synchronous Acceleration

XV0 kAV0 y I1 (kr) cos(kz) y− a2 2 r kV0 A Ez = I0 (kr) sin(kz) 2 Ey =

(8.17) (8.18)

where each component above is multiplied by sin(ωt + φ). We will be interested in the lowest-order approximation for the modiﬁed Bessel function, I1 (v) ∼ = v/2. The Ez component in Eq. (8.18) provides the accelerating force on the beam. The ﬁrst term of both Eqs. (8.16) and (8.17) is associated with quadrupole focusing, and the second term leads to the transverse RF defocusing force that acts on the beam, when the phase φ is chosen for longitudinal focusing. We call A the acceleration efﬁciency and X the focusing efﬁciency. When m = 1, A = 0, X = 1 and the RFQ becomes a pure quadrapole transport channel with no acceleration. As m increases an accelerating ﬁeld is produced on axis. The quantity XV0 /a2 is the quadrupole gradient that is a measure of the quadrupole focusing strength. The most accurate way to calculate the beam dynamics in the RFQ is by numerical integration of the equations of motion through these electric ﬁelds. However, considerable insight is obtained by making some simple approximations to examine the effect of these electric ﬁelds on the beam. 8.5 Synchronous Acceleration

In the approximation that the radial position and velocity of a particle within a unit cell are constant, the energy gain W for a particle with arbitrary normalized velocity β can be calculated by replacing ωt = 2πz/β λ, and integrating the electric ﬁeld seen by the particle, Ez sin(ωt + φ), over the unit cell. We obtain

qkAV0 I0 (kr)

W = sin(kz) sin(k z + φ) dz (8.19) 2 0 where k = 2π/β λ, k = 2π/βs λ, and = βs λ/2 is the length of the unit cell. Maximum acceleration occurs when the particle travels from the center of one cell to the center of the next in the time that the ﬁeld reverses its polarity that is exactly a half of an RF period. This is the synchronous condition, and the particle that satisﬁes this condition is a synchronous particle. Clearly, synchronism requires that β = βs , and the energy gain of the synchronous particle is W =

qπAV0 I0 (kr) cos φ 4

(8.20)

The phase φ when the particle is at the center of the unit cell is called the particle phase; at φ = 0, the electric ﬁeld is at the peak value. The synchronousenergy gain can be written in a form that corresponds to more conventional

241

242

8 Radiofrequency Quadrupole Linac

linac terminology. First, we calculate the spatial average of the peak axial accelerating ﬁeld over the RFQ unit cell:

2AV0 (8.21) Ez dz = E0 =

0 βλ The resulting average axial ﬁeld can be interpreted as the ratio of an effective axial voltage AV0 applied over the length of the unit cell. The transit-time factor for the synchronous particle is

Ez sin(kz) dz π = T= 0

(8.22) 4 Ez dz 0

Using these results the energy gain of a synchronous particle per cell of length has the familiar form W = qE0 TI0 (kr) cos φs

(8.23)

8.6 Longitudinal Dynamics

The RFQ electrode geometry and ﬁelds are chosen to produce a speciﬁc velocity gain for a synchronous particle that will continuously gain energy along the RFQ. When the synchronous phase is chosen to correspond to the time when the ﬁeld is increasing, particles that arrive earlier than the synchronous particle experience a smaller ﬁeld, and those that arrive later experience a larger ﬁeld. As was described in Chapter 6, this produces the longitudinal restoring force that provides phase-stable acceleration. The oscillating particles constitute a bunch, and every period in the RFQ contains a single bunch. It is straightforward to obtain the equations that describe the longitudinal focusing. First, we calculate the energy gain of a particle of phase φ, and subtract the energy gain of the synchronous particle, assumed to be on-axis with phase φs . On expressing this result in differential form as an average rate of change of the relative energy W − Ws , we obtain d(W − Ws ) = qE0 T(I0 (kr) cos φ − cos φs ) dz

(8.24)

where the independent variable z is the axial coordinate. Second, we write the equation that describes the average rate of change of the phase difference between an arbitrary particle and the synchronous particle, as d(φ − φs ) 2π(W − Ws ) =− dz mc2 βs3 λ

(8.25)

8.7 Transverse Dynamics

When −π < φs < 0, one obtains simple harmonic motion at a longitudinal wavenumber k2l =

π 2 qAV0 I0 (kr) sin(−φs ) mc2 βs4 λ2

(8.26)

One ﬁnds that Eq. (8.26) can be put in exactly the same form as Eq. (6.43), the expression for longitudinal wavenumber in conventional linacs. By comparison with standard linear accelerator theory, one ﬁnds that the longitudinal dynamics of the RFQ are really the same as in conventional linacs.

8.7 Transverse Dynamics

Perhaps the most fundamental impact of the RFQ, as it affects the beam dynamics of low-velocity ions, results from the increased transverse-focusing strength, compared with other methods. The capability of the RFQ to provide this focusing is the reason why the RFQ can be operated at low velocities, and this enables some very attractive features of the RFQ, such as adiabatic bunching, and a high beam-current limit. The focusing or conﬁnement of the off-axis particles in the RFQ is an example of the alternating gradient-focusing principle, already discussed in Chapter 7, and which is the basic beam-focusing principle used in most modern accelerators. The RFQ focusing is different than that described in Chapter 7, only because the polarity of the focusing ﬁeld varies in time instead of in space. Now we examine the transverse-focusing properties of the RFQ more quantitatively, using the electric-ﬁeld components derived in Section 8.4. Suppose we consider the motion in the x plane for a particle with charge q and mass m, and limit ourselves to small displacements from the axis. Nonrelativistically, the equation of motion is qk2 AV0 qXV0 + cos(kz) x sin(ωt + φ) = 0 (8.27) x¨ + ma2 4m where x¨ = d2 x/dt2 . The ﬁrst term in brackets is the quadrupole term, and the second term is the transverse ﬁeld associated with the electrode modulation that produces the longitudinal ﬁelds, that is, the familiar RF defocus term. Because the time dependence of the axial position of a particle is given by kz = ωt, the second term in brackets is proportional to cos(ωt) sin(ωt + φ) = [sin φ + sin(2 ωt + φ)]/2

(8.28)

The term with twice the RF frequency goes through one complete cycle as the synchronous particle moves through a unit cell, and if x is assumed to be constant throughout the cell, this term averages to zero. As a ﬁrst

243

244

8 Radiofrequency Quadrupole Linac

approximation we ignore the contribution of this term, and we write the equation of motion as qk2 AV0 qXV0 sin(ωt + φ) + x¨ + sin φ x = 0 (8.29) ma2 8m This result has the form of the well-known Mathieu equation. A smoothapproximation solution is obtained by considering a trial solution of the form x = [C1 sin t + C2 cos t][1 + ε sin(ωt + φ)]

(8.30)

where C1 and C2 are constants, and and ε are two new parameters, such that ω, and ε 1. The factor in the ﬁrst bracket is assumed to vary slowly over the unit cell, and represents the average or smoothed particle trajectory. The angular frequency is the oscillation frequency for the smoothed motion, known as the betatron frequency. The factor in the second bracket is a periodic function of time, which varies at the RF frequency, and is called the ﬂutter factor, where ε is the ﬂutter amplitude. Values for the two parameters and ε are obtained from the theory. For simplicity we choose C1 = 1 and C2 = 0, differentiate the trial solution twice, and neglecting the smaller terms of order ε/ω and 2 /ω2 , we ﬁnd x¨ ∼ = −εω2 sin(t) sin(ωt + φ)

(8.31)

An approximate solution of Eq. (8.29) is obtained by choosing the ﬂutter amplitude 1 qXV0 λ2 qXV0 = ε∼ = mω2 a2 4π 2 mc2 a2

(8.32)

Next, we look at the smoothed properties of the motion, by substituting Eq. (8.30) into the equation of motion, Eq. (8.29), and averaging over an RF period. We ﬁnd that x¨ ∼ = −2 sin(t)

(8.33)

Thus, the smooth-approximation solution implies that the average particle displacement satisﬁes the equation of a simple harmonic oscillator x¨ + 2 x

(8.34)

1 qXV0 2 qk2 V0 A sin φ + 2 ∼ = 2 mωa2 8m

(8.35)

where

The ﬁrst term is always positive and represents the contribution of the quadrupole focusing, and the second term represents the RF defocus term.

8.8 Adiabatic Bunching in the RFQ

If the quadrupole term is large compared with the RF defocus term that depends on the acceleration efﬁciency A, the transverse motion becomes decoupled from the longitudinal motion, and is approximately the same for particles of all phases. While the magnitude of 2 is a measure of the effective focusing force, it is customary to refer to the betatron phase-advance per focusing period, deﬁned as σ0 = λ/c. Substituting the expression for σ0 into Eq. (8.35), we obtain 1 σ02 ∼ = 8π 2

qXV0 λ2 mc2 a2

2 +

π 2 qAV0 sin(φ) 2mc2 β 2

(8.36)

It is found that the stability criteria depend on σ0 . In the smooth approximation, the beam is stable when σ02 > 0. When −π/2 ≤ φ ≤ 0, which is the condition for simultaneous acceleration and longitudinal focusing, the second term, representing the RF defocusing effect, is negative, which reduces the net focusing. The magnitude of the second term vanishes at the peak of the accelerating waveform, when φ = 0, and is maximum for φ = −π/2. If the second term exceeds the ﬁrst term, σ02 is negative, and the beam is unstable. We remarked earlier that Eq. (8.29) has the form of the Mathieu equation. A more accurate treatment, on the basis of the solutions to the Mathieu equation, shows that the beam is stable when 0 < σ0 < π. In standard notation, the Mathieu equation has the form d2 x + [P + 2Q sin(2τ )], dτ 2

x=0

(8.37)

The lower stability limit at σ0 = 0 is satisﬁed when −P < Q 2 /2, and the upper limit at σ0 = π is satisﬁed when Q < 1 − P. The equation of motion is brought into the standard notation by identifying τ = ωt/2, P = qAV0 sin φ/2mc2 β 2 , and Q = qXV0 λ2 /2π 2 mc2 a2 . Finally, we note that the transverse focusing in the RFQ is strong for the following reasons: (1) the use of electric rather than magnetic ﬁelds for focusing is superior for low-velocity particles, (2) the use of RF rather than dc ﬁelds allows higher peak surface electric ﬁelds, and larger electrode voltages, (3) the spatially uniform quadrupole focusing increases the fraction of space used by the focusing ﬁelds, especially compared with the use of discrete quadrupole lenses in drift-tube linacs at low velocities, and (4) the short focusing period, βλ, helps to reduce σ0 , allowing a larger phase-advance per unit length and stronger focusing without loosing beam stability.

8.8 Adiabatic Bunching in the RFQ

In any RF accelerator the beam must be longitudinally bunched so that all particles will be accelerated. In conventional accelerators such as the drift tube

245

246

8 Radiofrequency Quadrupole Linac

linac (DTL), bunching is accomplished prior to injection into the linac, by the use of one or more RF cavities placed in front of the linac. In buncher cavities the RF electric ﬁelds are applied to the unbunched input beam to produce a velocity modulation in which early particles are decelerated and late particles are accelerated. After a suitable drift space, the beam becomes bunched, ready for injection into the linac. The bunching is usually not very efﬁcient and the resulting beam quality is also poor. For high-intensity beams, the bunching process causes an increase in the beam density, which increases the space-charge forces, and often results in a blow up of the transverse beam emittance [3]. These problems were fundamental limitations for the performance of conventional linacs, especially at high beam currents. However, the RFQ can be used to adiabatically bunch the input dc beam, and if the parameters are chosen properly, RFQ bunching nearly eliminates these problems. The approach to RFQ adiabatic bunching is to inject the beam into the RFQ at low energy, typically 50 to 100 keV, with an initial synchronous phase of about φs ∼ = −π/2, where the separatrix has the largest phase width, and the longitudinal acceptance is maximum. After having begun to capture and bunch the beam, the synchronous φs is gradually increased. While continuing to accelerate the beam, the synchronous phase is moved toward the crest of the accelerating waveform, where the acceleration is more efﬁcient. As the beam is accelerated to higher velocity, the center-to-center bunch spacings get further apart. The phase width of the bunch shrinks, so bunching in phase is accomplished. However, the parameters can be chosen so that, as the velocity increases the geometric length of the bunch remains nearly constant. This keeps the bunch from being spatially compressed, which would result in large space-charge forces within the beam. The electrode modulation and the accelerating ﬁeld are small initially, so that the focusing is dominated by the quadrupole term, and the longitudinal and transverse dynamics are almost decoupled. As the beam is bunched more in phase, the amplitude of the electrode modulations is increased. Some unique features of the RFQ are useful for adiabatic bunching. (1) The large axial accelerating ﬁeld at low velocities, which is a result of the inverse proportionality of the ﬁeld to the synchronous velocity, allows the bunching to be both adiabatic, and to be accomplished in a relatively short length. (2) The velocity-independent electricquadrupole focusing that is very effective for low-velocity particles, provides the transverse beam conﬁnement even at high currents. (3) The control of the axial ﬁeld and the synchronous phase is accomplished by the machining of the four vane tips. Next, we describe the RFQ adiabatic bunching in more detail [4], and deﬁne the conditions that lead to a constant geometrical bunch length during the acceleration. It is convenient to change variables from phase and energy differences to position and momentum differences. For small velocity differences relative to the synchronous particle, we can write W − Ws ∼ = βcδp, where δp ∼ = p − ps is the axial momentum difference. The phase difference

8.8 Adiabatic Bunching in the RFQ

can be converted to an axial position difference by using φ − φs ∼ = −kδz, where dz = z − zs . The solution for small amplitude motion is of the form z = Z sin ω t, and the momentum is p = P cos ω t, where P = mω Z. From Section 5.12 the adiabatic invariant for small oscillations is

p dz = πPZ = πmω Z2 (8.38) Suppose that to control the space-charge effects during acceleration, we want to maintain a constant density for the beam. Constant density can be achieved, if we maintain a constant-amplitude Z for each particle, and we can do that for the particles that undergo small oscillations, because from the constancy of the adiabatic invariant in Eq. (8.38), we see that Z is constant, if ω is kept constant. For the RFQ the longitudinal small oscillation frequency is given by ω 2 =

π 2 qAV0 sin(−φs ) mβs2 λ2

(8.39)

which means, we must choose the RFQ parameters to maintain AV0 sin(−φs ) = cons tan t βs2

(8.40)

Equation (8.40) is the ﬁrst condition for RFQ adiabatic bunching, and assuming V0 is constant along the RFQ, Eq. (8.40) determines A as a function of φs and βs . We can construct the electrodes to produce the desired A by machining the electrodes to have the correct value of the modulation parameter m. Next we determine a relation for φs versus βs , as follows. We would like to require that the constant bunch density extends beyond the region of small oscillations, and includes the particles that undergo large-amplitude longitudinal oscillations. A simple approximate way to accomplish this is to require that the geometric length of the separatrix is constant. The geometric length Zψ of the separatrix is converted from the phase length ψ by using the relation Zψ = ψβs λ/2π. If the synchronous phase is varied along the RFQ, so that βs ψ = constant

(8.41)

the length of the separatrix will also be constant. The angular width ψ has already been given in Section 6.4, and depends only on the synchronous phase. The synchronous phase |φs | is controlled by controlling the center-to-center spacing of the unit cells. Combining Eqs. (8.40) and (8.41) gives a prescription for specifying both A(βs ) and φs (βs ) to maintain a constant bunch length. This adiabatic bunching approach is the basis of the bunching section of the RFQ, known as the gentle buncher. Although the space-charge forces have been neglected in this discussion, numerical simulation studies that include space-charge forces have shown that this procedure leads to an approximately

247

8 Radiofrequency Quadrupole Linac Cell number Vane profile (cm)

248

4 6 RM 3

73 Shaper

119 Gentle buncher

135 Accelerator

0 −3 −6

100

200 Vane length (cm)

Figure 8.7 A schematic drawing of the pole tips of an RFQ with adiabatic bunching. The transverse dimensions are magniﬁed compared with the longitudinal ones. The beam goes from left to right. Four sections are shown, a radial-matching (RM) section,

300

388

a shaper section, a gentle-buncher section, and an accelerator section. The bunching is started in the shaper section, and the adiabatic bunching, described in this section, is done in the gentle-buncher section.

constant bunch density, and provides excellent control of space-charge-induced emittance growth. In practice, all of the bunching of an initial dc beam cannot be done adiabatically without making the RFQ too long. The prebunching is usually started in a section called the shaper, using a prescription that ramps the phase and the acceleration efﬁciency linearly with axial distance. A schematic drawing of the pole tips of an RFQ designed for adiabatic bunching is shown in Fig. 8.7.

8.9 Four-Vane Cavity

The RFQ electrodes must be charged periodically at the RF frequency. For typical frequencies ranging from tens to hundreds of megahertz, this is accomplished by placing the electrodes in an RF cavity that is resonant at the desired frequency to charge the vanes in the desired + − +− quadrupole pattern (electric-quadrupole mode). In this section we will discuss the fourvane cavity that is mostly used in the high-frequency range, above about 200 MHz, and is the most common structure for light ions, especially protons. The most popular lower-frequency RFQ structure is the four-rod cavity that will be discussed in Section 8.14. The four-vane cavity consists of four vanes symmetrically placed within a cavity. The cavity is operated in a TE210 -like mode, which is obtained from the natural TE211 mode, [5] by tuning specially conﬁgured end cells to produce a longitudinally uniform ﬁeld throughout the interior of the cavity. The transverse electric ﬁeld is localized near the vane tips, and the magnetic ﬁeld, which is longitudinal in the interior of the cavity, is localized mostly in four outer quadrants. The efﬁciency of the four-vane cavity is relatively high,

8.10 Lumped-Circuit Model of Four-Vane Cavity

because the vane charging currents are distributed very uniformly along the length of the vanes. There have been many improvements in the mechanical design of the four-vane cavity, since the ﬁrst cavities were built. The vane-tip patterns are usually produced using a computer-controlled milling machine. The individual vanes can be easily cooled by incorporating suitable cooling channels. Monolithic four-vane RFQ structures with integral vacuum vessels have been fabricated by joining quadrants using both electroforming, and hydrogen-furnace brazing [6].

8.10 Lumped-Circuit Model of Four-Vane Cavity

The separation of the electric- and magnetic-ﬁeld regions of the four-vane cavity suggests a description based on a simple lumped-circuit model. It is convenient to assume that the cavity has a cloverleaf geometry, shown in Fig. 8.8, consisting of four quadrants, each of which could be analyzed as a resonant cavity. We represent the quadrants with a capacitance C and an inductance L , shown in Fig. 8.9. (To include dipole modes it would be necessary to include the capacitance between the diametrically opposite vanes.) Because the four gaps provide separate parallel electrical paths between the vanes of opposite potential, the total capacitance per unit length is C = 4C / V . We can derive a number of formulas in terms of the single free parameter C . To determine the inductance, we assume that B is approximately uniform over the outer part of the cavity, and write the magnetic ﬂux in each quadrant as = BA = µ0 AI/ V , where I is the total transverse current over the vane length

V , and A is the effective cross-sectional area per quadrant. The inductance of each quadrant is the ratio of the ﬂux to the current, or L = µ0 A/ V . To develop the model further, we assume that the effective quadrant area consists of three quarters of a circle of radius r, plus a square with sides of length r (see

+v 2 −v 2

−v 2 +v 2

Figure 8.8 Cross section of four-vane RFQ in cloverleaf geometry, used in the lumped-circuit model. The electric-ﬁeld lines are shown.

249

250

8 Radiofrequency Quadrupole Linac +V /2

L′ C′

−V /2

C′

L′

Figure 8.9 Equivalent circuit of the quadrupole mode of a four-vane cavity.

L′

+V /2

Figure 8.10 Shape of idealized quadrant for inductance calculation.

Fig. 8.10) to represent the area nearest the beam axis, yielding a total quadrant area of A = (4 + 3π)r 2 /4, and a quadrant inductance L = µ0 (4 + 3π )r 2 /4 V . The cavity resonant frequency is ω02 = (L C )−1 =

16 µ0 (4 + 3π )r 2 C

(8.42)

Equation (8.42) implies that at a ﬁxed frequency, an increase in the capacitive loading reduces the quadrant radius r and the transverse dimensions. Assuming an ejωt time dependence of the currents and voltages, the peak current on the outer wall is given by I = jω0 C V V/4, and the magnetic ﬁeld in the quadrants is B = jµ0 ω0 C V/4. We calculate the power loss by assuming the conducting surface area is the perimeter of the quadrant cross section comprised of three quarters of a circle plus a square. Taken over the length

V , the total surface area for all four quadrants is S = 2(4 + 3π)r V . The √ power dissipation is P = RS (B/µ0 )2 S/2, where RS = µ0 ω0 /2σ is the surface resistance, expressed in terms of the conductivity σ . Substituting for B and eliminating r, we obtain the power per unit length for the whole cavity 4 + 3π P = (8.43) (ω0 C )3/2 V 2 32σ

8.11 Four-Vane Cavity Eigenmodes

This shows that for a given vane voltage, the power is proportional to the −1/2 3/2 power of the frequency. This is different than the usual ω0 scaling of the power dissipation for typical accelerating cavities, because for this case the ratio of surface magnetic ﬁeld to vane voltage is proportional to the frequency. The stored energy per unit length is W = C V 2 /2, and the quality factor is ω0 W

8σ = (8.44) Q= P

(4 + 3π)ω0 C

The value of the parameter C can be estimated in several ways. As a ﬁrst approximation, an electrostatic calculation for four rods of circular cross section, whose radius of curvature equals the aperture radius is C ∼ = 90 × 10−12 F/m, independent of aperture radius. Using the electromagnetic code SUPERFISH, the value of C obtained from the calculations of the resonant frequency of a four-vane cloverleaf cavity, and using the inductance formulas from this section, is found to be a weak function of the vane radius, and for typical values of vane radius to wavelength is approximately C = 120 × 10−12 F/m. Using the latter result for the capacitance per unit length, this model can be used to estimate the properties of a four-vane cavity, and shows the approximate dependence of the cavity properties on the parameters. For an accurate calculation of the cavity properties for any speciﬁc geometry, an electromagnetic-ﬁeld-solver code must be used.

8.11 Four-Vane Cavity Eigenmodes

The ideal quadrupole-mode dispersion curve of the four-vane cavity has the classic hyperbolic shape that is characteristic of a uniform structure. The linac structure is normally operated at the cutoff frequency of the quadrupolemode passband, where for a given length, the mode separation between the operating mode and the nearest higher longitudinal mode is relatively small. The dispersion curve, showing the longitudinal-mode spectrum for the quadrupole family of modes, is given approximately by the expression

See Also

NeurIPS 2022 Call for Papers

ω 2 ω 2 nπ 2 n 0 = + , n = 0, 1, 2, . . . (8.45) c c

V where ωn is the frequency of the nth mode, ω0 is the frequency of the operating quadrupole mode, which has a guide wavelength λn = 2 V /n for an effective vane length V . At the vane ends, the magnetic ﬂux from each quadrant splits in half, each half ﬂows around the end of a vane, and returns in the adjacent quadrants. The magnetic ﬁeld is shown in Fig. 8.11. This produces a continuous path for the ﬂux, and at any cross section the magnetic ﬂux summed over the four quadrants is zero.

251

252

8 Radiofrequency Quadrupole Linac

(a ) Cross section at end

(b )

(c )

Side

Top

Figure 8.11 Magnetic ﬁeld lines in quadrupole mode at the end: (a) cross section, (b) side view, and (c) top view.

There are three independent azimuthal modes for which the magnetic ﬂux sums to zero, the quadrupole mode and two dipole modes, shown in Fig. 8.12. Suppose that each quadrant has the same cross-sectional area, and that the magnetic ﬁeld in each quadrant n is determined from a perturbation measurement to have the value Bn , as shown in Fig. 8.13. If the cavity is not properly tuned, the magnitudes of the Bn will not necessarily be equal, and the cavity mode will be an admixture of the quadrupole mode and the two dipole modes. From ﬂux conservation B1 + B2 + B3 + B4 = 0. The quadrupole-mode amplitude is proportional to AQ = |B1 − B2 + B3 − B4 |/4. The amplitudes of the two dipole modes are AD1 = |B1 − B3 |/2, and AD2 = |B2 − B4 |/2. One of the objectives of the cavity tuning is to minimize the magnitude of the dipolemode amplitudes, and produce the best approximation to a pure quadrupole operating mode. An idealized mode spectrum of a four-vane RFQ is shown in Fig. 8.14. The longitudinal-mode spectra for the pair of degenerate dipole modes overlap the quadrupole-mode spectrum. The cutoff frequency of the dipole modes typically

(a )

Quadrupole TE21n

(b )

Dipole 1 TE11n

Figure 8.12 Azimuthal modes of the RFQ four-vane cavity: (a) quadrupole, (b) dipole 1, and (c) dipole 2. It is convenient to label the RFQ modes by analogy with modes of the pillbox cavity that would exist with

(c )

Dipole 2 TE11n

open-circuit boundary conditions. The quadrupole modes are labeled as TE21n modes, where n is the longitudinal-mode number, and the dipole modes are TE11n modes.

8.11 Four-Vane Cavity Eigenmodes

Figure 8.13 General azimuthal magnetic ﬁeld pattern in four-vane RFQ.

500

Frequency (MHz)

480

460

TE213 Quadrupole modes

440 TE210 420

400

TE212

TE211

TE113 TE112

TE110

Dipole modes

TE111 1

2 3 Mode number

Figure 8.14 Idealized mode spectrum of a four-vane RFQ plotted against a longitudinal-mode number. The RFQ operating mode is labeled the TE210 mode because of the appearance of the central ﬁeld distribution.

lies a few percent lower in frequency than the quadrupole operating mode. This introduces the possibility of accidental degeneracy between the operating quadrupole mode, and a higher longitudinal mode of the dipole family. These effects can make the ﬁeld distribution of a long, four-vane structure especially sensitive to dimensional construction errors. Field-tilt effects can become important for structures with the length of a few wavelengths. Various methods have been devised to deal with these problems, including vane coupling rings that connect opposing vanes, movable ﬁeld tuners distributed along the structure, tuning rods that selectively shift the frequency of the dipole-mode spectrum, and resonant coupling of individual RFQ cavities [7].

253

254

8 Radiofrequency Quadrupole Linac

8.12 Transmission-Line Model of Quadrupole Spectrum

We now construct a model of the four-vane-cavity quadrupole modes that will help us to understand the effects of errors on the ﬁeld distribution. As shown in Fig. 8.15, we let L0 (x) and C0 (x) be the inductance and capacitance of a shunt resonant circuit that carries the charging current to deliver charge to each vane tip. The variable x is the axial distance along the RFQ. Let L(x) be the inductance associated with longitudinal current that ﬂows along a vane tip, and deﬁne V(x) and I(x) as the intervane voltage and the longitudinal current along the vane tip. For a properly tuned RFQ, I(x) = 0. From transmission-line theory ∂V = −jωLI(x) ∂x

(8.46)

1 ∂I V(x) = − jωC0 + ∂x jωL0

(8.47)

and

Substituting Eq. (8.47) into Eq. (8.46), we obtain the wave equation ∂ 2V + k2 V(x) = 0 ∂x2

(8.48)

where the dispersion relation is ω2 = ω02 + k2 /LC0 , and ω02 = 1/L0 C0 is the cutoff frequency. To represent the known dispersion curve, we must identify c2 = 1/LC0 , where c is the speed of light. Next we choose open-circuit boundary conditions at each end of the line, x = 0 and x = V . Thus I(0) = I( V ) = 0 and from the transmission-line equation it follows that dV(0)/dx = dV( V ) dx = 0.

I (x + ∆x )

l (x ) V (x )

V (x + ∆x )

L(x )

C 0(x )

L0(x )

Figure 8.15 Transmission-line model of the RFQ quadrupole mode for the four-vane cavity.

8.12 Transmission-Line Model of Quadrupole Spectrum

The solutions to the wave equation that satisfy the boundary conditions are 2 0 Vn (x) = cos kn x, n = 0, 1, 2, . . . (8.49)

where kn = 2π/λg = nπ/ V , and the normalization is such that 0 V 0 Vn2 (x) dx = 1. The modes are characterized by the number of half wavelengths of the voltage cosine waves that ﬁt between the boundaries; the frequency of the nth mode is ωn2 = ω02 + (nπc/ V )2 . The voltage distribution for the three lowest modes is shown in Fig. 8.16. We now examine the effects of perturbations on the ﬁeld distribution, by applying standard perturbation theory for the eigenvalue problem [8]. First we review the theory, recalling that the eigenvalue problem consists in ﬁnding the solutions for the eigenvectors Vn and the corresponding eigenvalues γn of the operator M that satisﬁes the equation MVn = γn Vn . The unperturbed solutions satisfy M0 0 Vn = 0 γn 0 Vn , or ∂2 (8.50) c2 2 − ω02 0 Vn = 0 γn 0 Vn ∂x which is the wave equation, if we identify 0 γn = −ωn2 . The solutions are given by Eq. (8.49). The perturbed problem is described by M = M0 + P, where the perturbation matrix is P M0 . The theory tells us that perturbed solutions can be expressed in the lowest order by amn 0 Vn (8.51) Vn = 0 Vn + m =n

Vn(x )

n=0

n=1

n=2

Figure 8.16 Voltage distribution of the three lowest unperturbed modes of the four-vane RFQ.

255

256

8 Radiofrequency Quadrupole Linac

and

amn =

0 0γ n

Vm P0 Vn (8.52)

− 0 γm

We identify the unperturbed operator as M0 = c 2

∂2 − ω02 ∂x2

(8.53)

and the perturbation operator is P = − ω02 (x), where ω02 (x) is the squared resonant-frequency error, as a function of x. Now suppose we have a small resonant-frequency error distribution ω0 (x) that depends on x. We assume ω02 ω02 , where ω02 = 2ω0 ω0 . In ﬁrst order the perturbed cavity frequency is

V 2 (1) 2 ωn = ωn2 + ω02 (x) cos2 (kn x) dx (8.54)

V 0 where kn = nπ/ V . The perturbed ﬁeld in the lowest order is    amn cos(km x)  2 (1) cos(kn x) + Vn (x) = 2 

V  ωn2 − ωm

(8.55)

m =n

where

amn

2 =

cos(km x) cos(kn x) ω02 (x) dx 2 ωn2 − ωm

(8.56)

and

ωn2

−

2 ωm

πc = (n − m )

V 2

For the RFQ operating mode, n = 0 and the perturbed ﬁeld is 2 ∞

V 2 am0 cos(km x) (1) 1− V0 (x) =

V πc m=1 m2 where am0 = −

2 V (mπc)2

V 0

cos(km x) ω02 (x) dx

(8.57)

(8.58)

(8.59)

and Eq. (8.54) gives the perturbed frequency with kn = 0 for the n = 0 mode. The perturbed ﬁeld error is expressed in Eq. (8.58) as a sum over the unperturbed modes. Because of the denominator, the unperturbed modes

8.12 Transmission-Line Model of Quadrupole Spectrum

that are nearest the operating mode will tend to contribute more to the ﬁeld error. The contribution of any given mode also increases in proportion to the overlap between the squared-frequency-error distribution ω02 (x), and the unperturbed eigenfunction cos km x. Consider the example of a δ-function frequency error at some point x0 , expressed as ω02 (x) = δ(x − x0 ), where determines the magnitude of the error, and

V δ(x − x0 ) dx (8.60) 1= 0

The frequency perturbation induced by the error is given by

2 V 2 (1) 2 2 ω0 = ω0 + δ(x − x0 )dx = ω02 +

V 0

(8.61)

and the frequency shift is δω0 =

V ω0

(8.62)

This means we can also write ω02 (x) = V ω0 δω0 δ(x − x0 ). The perturbed ﬁeld distribution is

V       δ(x − x ) cos(k x) dx ∞ 0 m   2 2 0 (1) V0 (x) = cos(k x) 1+ m 2 

V 

V m=1 ω02 − ωm     (8.63) or

(1)

V0 (x) =

δω0 1−8 ω0

V λ

2 ∞

cos(mπx0 / V ) cos(mπx/ V ) m2 m=1

(8.64) Thus, δV0 (x) δω0 = −8 V0 ω0

V λ

2 ∞

cos(mπx0 / V ) cos(mπx/ V ) m2 m=1

(8.65)

is the fractional-ﬁeld error in terms of the cavity resonant-frequency shift δω0 caused by the δ-function error at x0 . From Eq. (8.65) we see that the local frequency error adds small amounts of all the other unperturbed-mode ﬁelds to the unperturbed mode that corresponds to the error-free state, an effect known as mode mixing. We should not think of the error as exciting these other modes, but rather as an error-induced mixing of modes of the ideal system to create a new ﬁeld distribution for the operating mode. Note that each of the unperturbed modes contributes a term proportional to the ﬁeld value of each

257

258

8 Radiofrequency Quadrupole Linac

mode at the point of the perturbing error, divided by the mode index squared. Modes with m closest to m = 0 tend to contribute more to the ﬁeld error. An analytic solution exists for the summation in Eq. (8.65), which gives the ﬁnal result δω0 V 2 δV0 (x) = −4π 2 V0 ω0 λ  1 x 2 1 x0 2 x 1   − + + , x > x0  3 V 2 V 2 V × (8.66) 2  1 x0 2   1 − x0 + 1 x + , x0 > x 3 V 2 V 2 V The ﬁeld error varies as a second-order polynomial in x; it is proportional to δω0 and to ( V /λ)2 . An interesting property of the solution is that if an error at some position x0 causes the local resonant frequency to increase, the ﬁeld at x0 decreases. Likewise, if the local frequency decreases, the local ﬁeld increases. This turns out to be a general rule of thumb, when the modes that are mixed lie higher in frequency than that of the pure mode representing the unperturbed state. Figures 8.17 and 8.18 show some perturbed ﬁeld distributions for two examples, (a) a δ-function error at the vane end, where x0 / V = 0 and (b) a δ-function error at middle of the vane, where x0 / V = 0.5. For both examples, we have chosen V /λ = 2 and δω0 /ω0 = 0.01. Finally, we consider δ-function frequency errors at each end that are equal and opposite, expressed as ω02 (x) = δ(x) − δ(x − V ), where again

V0(x )

1.75 1.5 1.25

m = 0, 1

m=0

1 0.75 0.5

m = 1 – 20

0.25

X/lv 0

0.2

0.4

0.6

0.8

Figure 8.17 Perturbed ﬁeld distributions for problem with a δ-function error at the vane end, where x0 / V = 0, V /λ = 2, and δω0 /ω0 = 0.01. (a) unperturbed m = 0 mode, (b) adding the m = 1 mode, (c) sum of the ﬁrst 20 modes.

8.12 Transmission-Line Model of Quadrupole Spectrum 2

V0(x )

1.75 1.5

m = 0, 1

1.25 1 0.75

m=0–2

m = 1 – 20

0.5 0.25

X/lv 0

0.2

0.4

0.6

Figure 8.18 Perturbed ﬁeld distributions for problem with a δ-function error at the middle of the vane, where x0 / V = 0.5,

V /λ = 2, and δω0 /ω0 = 0.01. (a)

0.8

unperturbed m = 0 mode plus m = 1 mode, which contributes zero, (b) adding the m = 2 mode, (c) sum of the ﬁrst 20 modes.

determines the magnitude of the error. The induced frequency shift with errors at both ends is V

V 2 (1) 2 ω0 = ω02 + δ(x) dx − δ(x − lV ) dx

V 0 0 = ω02 +

2 [1 − 1] = ω02

(8.67)

Thus, the resonant frequency is unchanged. The perturbed ﬁeld distribution is

(1)

V0 (x) =

δω0 1−8 ω0

V λ

1 − cos(mπ) cos(mπx/ V ) m2 m=1

2 ∞

(8.68) There is an analytic solution for the summation that is given by ∞ π2 1 − cos(mπ) x cos(mπx/

) = 1 − 2 V m2 4

V m=1

(8.69)

Using this result we may express the fractional change in the ﬁeld distribution as x δω0 V 2 δV0 (x) = −2π 2 1 − 2 (8.70) V0

V ω0 λ The perturbed ﬁeld distribution of Eq. (8.70) has a linear ﬁeld tilt. The error term is proportional to δω0 and to ( V /λ)2 . We show in the Fig. 8.19 the ﬁeld

259

260

8 Radiofrequency Quadrupole Linac

distribution for the problem with V /λ = 2 and δω0 /ω0 = 0.01, summing all the modes.

8.13 Radial-Matching Section

The RFQ can be designed to capture, bunch, and accelerate a continuous unbunched input beam. Matching of the input beam into the RFQ presents a special problem. The matched ellipse parameters in the RFQ vary with the RF phase and are relatively independent of position along the linac. Therefore, the orientation of the acceptance ellipse depends on time. For proper matching into the RFQ, one must provide a transition from a beam having time-independent characteristics to one that has the proper variations with time. At the input, a time-independent set of ellipse parameters is required, which will depend on the beam current. The solution is to taper unmodulated vanes at the input to the RFQ so that the radius decreases and the focusing strength increases from near zero to its full value over a distance of several cells. Quadrupole symmetry is maintained throughout the RM section. The procedure allows a time-independent beam to adapt to the time structure of the focusing system. Phase-space plots and beam envelopes for a RM section design are shown in Fig. 8.20. In this RM section the focusing strength B is increased linearly as a function of longitudinal position from its value at the input to its full value at the end of the RM section. First, the matched ellipse parameters are found in the interior cells for various phases. Three matched ellipses corresponding to phases 90◦ apart are shown in the two phase-space plots in the upper right side of the ﬁgure for the x − x and y − y planes, respectively. This

V0(x )

1.75 1.5 1.25 1

0.75 0.5 0.25

X/lv 0

0.2

0.4

0.6

0.8

Figure 8.19 Perturbed ﬁeld distribution for problem with equal and opposite δ-function errors at the ends of the vane, x0 / V = 0.0 and x0 / V = 1.0, V /λ = 2, and δω0 /ω0 = 0.01.

8.13 Radial-Matching Section

V 2. mm × 100. MRAD

8.0 mm

2 3 4 5

Figure 8.20 Beam envelopes at phases 90◦ apart in a radial-matching section.

shows graphically how different the matched ellipses are as a function of the phase, and the relatively small area of overlap is common to all the ellipses. The phase-space plots at the upper left are the result of tracking these same three ellipses backward through the tapered RM section, ﬁve-periods long (10 cells). These ellipses at different phases are similar and have a high degree of overlap. To obtain a matched beam within the RFQ, the average of these ellipses will be taken as the matched ellipse. The x − x ellipses and the y − y ellipses are essentially identical. Thus, the input beam injected into the RM section is axisymmetric. If the beam from the ion source is axisymmetric, the external lenses required for matching the beam would normally be solenoids or quadrupole triplets. The bottom graph shows the horizontal and vertical proﬁles for the three ellipses that are being tracked through the RM section. The RFQ can also be designed to accelerate a bunched beam. In this case the RFQ could begin with an m > 1 accelerator section. The input beam would generally not be axisymmetric. One or more external cavities would be used to bunch the input beam before injection into the RFQ. With a bunched beam, which has a limited range of input phases, conventional external quadrupole lenses could be used to provide transverse matching that would be suitable for all beam phases. Therefore, for the bunched beam case, a RM section would

261

262

8 Radiofrequency Quadrupole Linac

not be needed. However, if m > 1 initially, there would be a time-varying on-axis potential at the input, which is not present with pure quadrupole symmetry when m = 1. Consequently, beginning with m > 1 would introduce an energy kick at the input, which may be undesirable. However, this could be eliminated by beginning the RFQ with an entrance transition cell, to be discussed in the following section. The transition cell begins with m = 1, and makes a smooth transition to the full m > 1 value. This transition cell would be immediately followed by the accelerator section. RM section designs have been improved considerably since the early days when a linear ramp of the focusing strength B was used. A remarkably effective and commonly used RM section is the design proposed by Crandall [9]. This RM section is normally used at the entrance of the RFQ, but can also be used at the exit. It uses a four-term potential function given by U(r, θ, z) =

3 V An Tn (r, z) cos(2nθ ), 2 n=0

Tn (r, z) = I2n (kr) cos(kz) + I2n (3kr) cos(3kz)/32n+1

(8.71)

where z = 0 is at the interface between the RM section and the adjacent normal RFQ cell, and k = π/2L, where L is the length of the RM section, the distance from the end wall at z = −L to the end of the RM section at z = 0. The length L can be chosen by the designer. Crandall considers a range from L = 2βλ to 12βλ, and based on the quality of the match, concludes that there is no reason for choosing L any longer than 3βλ. Short matching sections are an advantage since it is found that they require a less convergent injected beam that is easier to achieve. We treat the four amplitudes An for n = 0, 1, 2, and 3 as unknown amplitudes for the RM section geometry. They must be chosen so that the RM section geometry matches the known geometry of the adjacent normal RFQ cell at the z = 0 interface. Four equations are needed to determine the four amplitudes. The four equations correspond to requiring at the z = 0 interface: (1) the same x-vane displacements, ±x0 , (2) the same y-vane displacements, ±y0 , (3) the same transverse radius of curvature of the x vanes, and (4) the same transverse radius of curvature of the y vanes. The properties of Crandall’s four-term potential function are: (1) Each potential term satisﬁes Laplace’s equation. (2) Each potential term is zero at the end wall at z = −L. (3) Each potential term has a physically reasonable s-shaped z dependence, starting with zero at the end wall, and ending at a maximum at the z = 0 interface. (4) The terms for n > 0 vanish at r = 0 contributing zero potential on-axis, as would be appropriate for an adjacent

8.13 Radial-Matching Section

normal RFQ cell with no modulation. A nonzero n = 0 term provides for an adjacent normal RFQ cell with modulation m > 1. (5) The longitudinal electric ﬁeld −∂U/∂z = 0 at the z = 0 interface for all r. (6) The longitudinal electric ﬁeld −∂U/∂z ∼ 0 for small kr at z = −L end wall. (7) A smooth transition is obtained between the ﬁeld-free region in the end wall and the ﬁeld within the RFQ. Four equations must be solved to determine the four unknown amplitudes, An . The ﬁrst requirement which expresses equality of the two horizontal vane displacements at the z = 0 interface is U(±x0 , 0, 0) = +V/2

(8.72)

Likewise, for the vertical vane displacements at z = 0, U(±y0 , π/2, 0) = −V/2

(8.73)

We only need to consider the coordinates with the plus sign. Substituting Eq. (8.71) into Eqs. (8.72) and (8.73) gives 3

An Tn (x0 , 0) = 1

n=0 3

An Tn (y0 , 0) cos(nπ) = −1

(8.74)

n=0

Next we obtain the equations involving the two transverse curvature radii of the vane tips at z = 0. By inspection, the transverse slopes at the two horizontal vane tips at x = ±x0 and y = 0 are dx/dy = 0, and at the two vertical vane tips at y = ±y0 and x = 0 are dy/dx = 0. The transverse radius of curvature for each horizontal vane at y = 0 is 1/x0 , where x0 = d2 xp /dy2 . The transverse radius of curvature for the y vane at x = 0 is 1/y0 , where y0 = d2 y/dx2 . These transverse radii of curvature are given by their values at the z = 0 location of the adjacent normal RFQ cell. The ﬁrst equation we obtain that involves a transverse radius of curvature is obtained by differentiating twice both sides of Eq. (8.71) with respect to x and evaluating at z = 0. We will be interested in the y vane at x = 0 and y = y0 . Differentiating the ﬁrst time gives 3 n=0

∂θ ∂Tn (r, 0) ∂r cos(2nθ) − 2nTn (r, 0) sin(2nθ) =0 ∂r ∂x ∂x (8.75)

263

264

8 Radiofrequency Quadrupole Linac

To avoid a lot of algebra, Crandall notes that the cross section of the four vanes shows that on the y-pole tip at x = 0 and y = y0 , we have ∂r/∂x = 0. Also, on the y-pole tip at x = 0, we have θ = π/2 and sin(2nθ) = sin(nπ) = 0. Then, when differentiating Eq. (8.75), any terms multiplying ∂r/∂x or sin(2nθ) will be multiplied by zero. Then, to differentiate Eq. (8.75) the only nonzero terms correspond to differentiating ∂r/∂x and sin(2nθ). Changing from polar to Cartesian coordinates yields ∂r yy + x = ∂x r

(8.76)

Differentiating again and evaluating at the vertical pole tip where y = y0 and x = 0 results in ∂ 2r 1 = + y0 2 ∂x y0

where y0 = d2 y/dx2 at x = 0 and r = y0

(8.77)

Also since tan(θ) = y/x 1 ∂θ xy − y =− = ∂x r2 y0

(8.78)

Now, differentiating Eq. (8.75) once more, 3

n=0

2 ∂θ ∂Tn (r, 0) ∂ 2r 2 =0 cos(2nθ) 2 − 4n cos(2nθ)Tn (r, 0) ∂r ∂x ∂x

(8.79) Substituting Eqs. (8.77) and (8.78) into Eq. (8.79), and using the results that on the vertical pole tip where x = y = 0, y = y0 , and θ = π/2, we obtain 3

n=0

∂Tn (y0 , 0) ∂r

1 1 + y0 − 4n2 Tn (y0 , 0) 2 cos(nπ) = 0 y0 y0

(8.80) Repeating this procedure for the x vane tip, where θ = 0, y = 0, x = dx/dy = 0, r = x0 , we obtain 3 n=0

∂Tn (x0 , 0) ∂r

1 1 + x0 − 4n2 Tn (x0 , 0) =0 x0 x0

(8.81)

We now have four equations, Eqs. (8.72), (8.73), (8.80), and (8.81), that must be solved simultaneously to determine the four unknown amplitudes, An , for n = 0, 1, 2, and 3. Once these are determined, the potential and ﬁelds are determined everywhere in the RM section.

8.14 RFQ Transition Cell

The vane tips in the ﬁrst RFQs that were built ended abruptly at the end of a full accelerating cell, where one pair of vanes is closer to the axis than the other pair. However, it is not obvious that the two-term potential function, which is valid for a quasiperiodic structure, and therefore a good approximation for the interior cells of the RFQ, is still a good model in the vicinity of the last accelerating cell where the vanes end abruptly. Also, for an RFQ where the vanes end abruptly at the end of the last cell, one would expect physically that the unequal spacing of the vanes at the end of the cell would result in time-varying on-axis potential at the end. In this case, the potential variation at the end would cause an undesirable variation in the energy of the output beam. Furthermore, ending the vanes at the end of a full cell results in an asymmetric output beam, which is strongly converging in one plane and strongly diverging in the other plane, and these characteristics can make output-beam matching difﬁcult. For these reasons it was important to develop a more realistic model for the ﬁelds at the ends of the vanes. Motivated by these arguments, in 1994 Crandall [10] introduced a new type of cell called a transition cell to be used after the last full accelerating cell. The transition cell makes a smooth transition from full modulation at the last accelerating cell to no modulation and pure quadrupole symmetry at the end of the transition cell. The transition cell is described using a three-term potential solution of Laplace’s equation, and its length is slightly less than that of a normal accelerating cell. The transition-cell parameters are chosen to be continuous with respect to both the potential and the ﬁelds from the two-term description used for the ﬁnal accelerating cell. Crandall pointed out that the RFQ vane tips at the end of the transition cell could also be extended for a short distance with no modulation. The ability to choose the length of the vane extension provides additional ﬂexibility to change the output transverse phase-space ellipses as required for output-beam matching. In a normal acceleration cell, the ﬁelds are derived from a time-varying two-term potential function (Eq. 8.82) that satisﬁes Laplace’s equation: r 2 V cos 2θ + AI0 (kr) cos kz sin ωt (8.82) U(r, θ, z; t) = 2 r0 where V is the maximum potential difference between vanes, k = π/L, L is the cell length, and A and r0 are constants that depend on the boundary conditions. Typical vane-tip proﬁles are shown in Fig. 8.21. The period of these proﬁles is 2L, and the slopes of the proﬁles (x and y ) are zero at the end of each cell. If the RFQ linac vanes begin with modulation at the beginning of a full cell (z = 0), one pair of vanes, either the x or the y vanes, will begin closer to the axis than the other pair. This means that on-axis beam particles will see a nonzero on-axis ﬁeld in the entrance fringe-ﬁeld region, and this means that

265

266

8 Radiofrequency Quadrupole Linac x

ma a

y z n

Figure 8.21 Vane-tip proﬁles from the two-term potential function of Eq. (8.82) (courtesy of K. R. Crandall).

the particles will generally gain or lose energy depending on time or phase. The same argument applies in the exit fringe-ﬁeld region. The magnitude of the effect can be estimated in the approximation that the two-term potential is valid at the beginning and ends of the vanes. The two-term potential on-axis in the RFQ is U(0, 0, z, t) =

AV cos kz sin φ 2

(8.83)

where φ = ωt is the phase (using the sine convention where zero phase corresponds to the zero crossing of the sine wave at 90◦ before the crest), V is the intervene voltage, and A is the acceleration efﬁciency, for which we will assume A = 0.5 as a typical value. The synchronous phase, which is the RF phase when the synchronous particle is at the middle of the cell, is typically 30◦ before the crest or φ = 60◦ in the sine convention. When the synchronous particle enters the cell the RF phase is typically 120◦ before the crest (φ = −30◦ in the sine convention) and when it exits the cell the phase is about 60◦ after the crest (φ = 150◦ in the sine convention). At the entrance of the RFQ, using z = 0, A = 0.5, and φ = −30◦ on the sine wave, the on-axis potential is ◦

U(0, 0, z = 0, φ = −30 ) = −V/8

(8.84)

Outside the fringe ﬁeld the on-axis potential is U = 0. In a hard-edged approximation the on-axis potential falls from zero to −V/8 at z = 0, and a synchronous particle of charge q gains kinetic energy qV/8 in the entrance fringe ﬁeld. At the exit of the RFQ, z = L, A = 0.5, and φ = 150◦ on the sine wave, the on-axis potential is ◦

U(0, 0, z = L, φ = 150 ) = −V/8

(8.85)

In the hard-edged approximation, the on-axis potential increases from −V/8 to 0 at Z = L, and a synchronous particle loses kinetic energy qV/8 in the exit fringe ﬁeld. These kinetic energy changes are only estimates because of the uncertain validity of the two-term potential function at the ends of the RFQ. These energy changes and their uncertainty can be eliminated by using a transition cell.

8.14 RFQ Transition Cell

267

Crandall’s approach for a better analytic model was to ﬁnd a solution of Laplace’s equation for a transition cell, valid after the last accelerating cell, which at the exit makes a smooth transition from full vane-tip modulation at the last accelerating cell to zero modulation with an optional zero-modulation vane segment with pure quadrupole symmetry (equal spacing of the four vane tips from the beam axis). The optional segment with no modulation will be discussed later. To accomplish this, Crandall introduced a three-term potential function to describe the transition cell of length L . The three-term potential function is given by V r 2 U(r, θ, z; t) = cos 2θ ± A10 I0 (Kr) cos Kz 2 r0 ± A30 I0 (3Kr) cos 3Kz sin ωt

(8.86)

π 2L Periodic vane-tip proﬁles for the three-term potential function with the period length 4L are shown in Fig. 8.22. The minus sign was used for the second and third terms so that the x vane would start at ma and the y vane would start at a when z = 0. Because the second and third terms are zero at z = L , both vane-tip displacements are equal to r0 at this point. The three unknown parameters are A10 , A30 , and K (or L ), and these are determined by three conditions. Two conditions are determined at the z = 0 interface, where the values of the x- and y-vane-tip potentials in the transition cell must be matched to those at the end of the last accelerating cell. The third condition results from requiring zero slope for both the x- or y-vane-tip proﬁles at the point of pure quadrupole symmetry at the end in the transition cell, z = L . The horizontal and vertical vane-tip proﬁles, where U(r, θ = 0, z) = V/2 and U(r, θ = π/2, z) = −V/2, respectively, are given by: 2 x − A10 I0 (Kx) cos Kz − A30 I0 (3Kx) cos 3Kz = 1 (8.87) r0 where K =

y z 0

L′

2L′

3L′

Figure 8.22 Periodic vane-tip proﬁles satisfying the three-term potential function (Eq. 8.86) (courtesy of K. R. Crandall).

4L′

5 L′

268

8 Radiofrequency Quadrupole Linac

and

−

y r0

2 − A10 I0 (Ky) cos Kz − A30 I0 (3Ky) cos 3Kz = −1

(8.88)

Following Crandall’s derivation, we ﬁrst impose the zero-slope condition at z = L . Differentiating Eqs. (8.87) and (8.88) with respect to z gives expressions for the slopes: 2x − K[A I (Kx) cos Kz − 3A I (3Kx) cos 3Kz] x = 10 1 30 1 r02

− K[A10 I0 (Kx) sin Kz + 3A30 I0 (3Kx) sin 3Kz]

(8.89)

−2y − K[A10 I1 (Ky) cos Kz − 3A30 I1 (3Ky) cos 3Kz] y = r02 − K[A10 I0 (Ky) sin Kz + 3A30 I0 (3Ky) sin 3Kz]

(8.90)

Both x and y are zero at z = 0. For these also to be zero at z = L , the right-hand side of Eqs. (8.89) and (8.90) must be zero there. Since both x and y equal r0 at z = L , the second terms in both equations are the same and are equal to zero if A30 (A10 , K) =

α(K) A10 3

(8.91)

where α(K) =

I0 (Kr0 ) I0 (3Kr0 )

(8.92)

The vane-tip proﬁles, showing the ﬁnal accelerating cell and the transition cell that follows are shown in Fig. 8.23. Note that because Kz = π/2 at z = L , U(r0 , 0, L ) = V/2 and U(r0 , π/ 2, L ) = −V/2 are satisﬁed automatically by Eq. (8.86). Next we return to the conditions at the z = 0 interface, where the values of the x- and y-vane-tip potentials of the transition cell must be matched to those at the end of the last accelerating cell. At the z = 0 interface between the two cells we must have U(ma, 0, 0) = V/2 ma

(8.93)

x Acceleration cell z −L

Transition cell 0

L′

Figure 8.23 Vane-tip proﬁles in ﬁnal accelerating cell followed by a transition cell (courtesy of K. R. Crandall).

8.14 RFQ Transition Cell

and U(a, π/2, 0) = −V/2

(8.94)

Substituting Eqs. (8.93) and (8.94) into the three-term potential function (Eq. 8.86), we ﬁnd

ma r0

2 − A10 I0 (mKa) − A30 I0 (3mKa) = 1

(8.95)

and −

a r0

2 − A10 I0 (Ka) − A30 I0 (3 Ka) = −1

(8.96)

Multiplying Eq. (8.96) by m2 and adding the result to Eq. (8.95) and leads to A10 [m2 I0 (Ka) + I0 (mKa)] + A30 [m2 I0 (3 Ka) + I0 (3mKa)] = m2 − 1 (8.97) or A10 T10 (K) + A30 T30 (K) = m2 − 1

(8.98)

where we identify T10 and T30 as the terms in brackets multiplying A10 and A30 , respectively. Substituting Eq. (8.91) to eliminate A30 , and solving for A10 gives A10 (K) =

m2 − 1 T10 (K) + α(K)T30 (K)/3

(8.99)

Substituting Eqs. (8.91), (8.92), and (8.99) into Eq. (8.96) yields an expression for K: 2 m2 − 1 a + (I0 (Ka) + α(K)I0 (3Ka)/3) = 1 r0 T10 (K) + α(K)T30 (K)/3 (8.100) After K is determined, A10 (K) is determined by Eq. (8.99), and A30 (A10 , K) can be determined from Eqs. (8.91) and (8.92). Also, the length of the transition cell is given by L = π/2K. One can obtain an initial value of K for an iterative solution of Eq. (8.100) in the following way. First, we make the approximation that the modiﬁed Bessel functions are given by the leading term in the expansion, so that I0 ∼ = 1. This allows us to obtain approximate expressions for α, A10 , A30 , and the acceleration efﬁciency A of the two-term potential function. The results are α = 1, A30 = A10 /3, and A10 = 3(m2 − 1)/4(m2 + 1). For comparison with the acceleration efﬁciency parameter A of the two-term potential function, the

269

270

8 Radiofrequency Quadrupole Linac

same approximation I0 ∼ = 1 gives A=

m2 − 1 m2 − 1 ∼ = m2 I0 (ka) + I0 (mka) m2 + 1

(8.101)

Next, the starting value for K can be found by equating the potential functions at the z = 0 interface. At the vertical vane tip that is displaced by a, we have 2 2 α a a − AI0 (ka) = − − A10 I0 (Ka) + I0 (3Ka) − r0 r0 3 (8.102) The ﬁrst terms on each side cancel. This time, to obtain an expression for K, we must use the ﬁrst two terms in the expansion for I0 because otherwise the expressions for K cancel. Thus, A+A

K 2 a2 k 2 a2 α = A10 1 + + A10 (1 + 3α) 4 3 4

(8.103)

Using the approximation for A given by Eq. (8.101), using α = 1, and A10 = 3A/4, we obtain 1 K≈ √ k 3

(8.104)

Eq. (8.104) may be used as an initial value of K for an iterative solution of Eq. (8.100). Also, Eq. (8.104) can be used to estimate the length of the transition cell, since the length, L , of the transition cell and the length L = π/k of the accelerating cell are related approximately by 1 π π =√ 2L 3L or

L =

3 L 4

(8.105)

(8.106)

By differentiating the slopes x and y , one can derive the second derivatives, x and y . These can be evaluated for both potential functions at the interface between the accelerating cell and the transition cell, and also for the three-term potential at the end of the transition cell. Crandall shows that the second derivatives agree to ﬁrst order at the interface, which implies that a smooth transition is made at the interface. Also, the second derivatives of the threeterm potential function are zero at the end of the transition cell. Thus, at the end of the transition cell both the ﬁrst and the second derivatives of the vane proﬁles are zero, which means that a smooth transition can be made to unmodulated vanes at the end of the transition cell. The potential and ﬁelds at the beginning of this transition cell blend smoothly with those at the end of the previous accelerating cell. The potential and ﬁelds at the end of the transition

8.15 Beam Ellipses in an RFQ

cell are the same as for unmodulated vanes. The vane tips at the end of the transition cell have quadrupole symmetry (four vanes equally spaced from the beam axis), so the on-axis potential is zero, and this removes the uncertainty in the output energy of the beam. Crandall adds that by continuing the vanes with zero modulation for a short distance, the designer can control the point at which the beam exits from the periodic focusing system of the RFQ. This provides more control over the output transverse phase-space ellipses. The beam can be made to exit the RFQ having the characteristics it would normally have in the center of a quadrupole. This would allow an external matching quadrupole to be spaced farther from the end of the RFQ, and the strength requirement for this quadrupole would also be reduced. Another transition-cell application is to allow a long RFQ to be divided into two shorter ones. The ﬁrst RFQ would terminate with a transition cell followed by a short zero-modulation section; the second RFQ would begin with a short zero-modulation section followed by a transition cell followed by a normal acceleration cell. A low-current beam would require nothing but a short drift distance between these two RFQs. An axisymmetric RFQ output beam with identical phase-space ellipses in x and y beam can also be obtained by adding an output RM section at the end of the transition cell. One particular application of an axisymmetric output beam is to allow both positive and negative beams accelerated by the RFQ to be matched to a following magnetic transport system.

8.15 Beam Ellipses in an RFQ

Ellipse conﬁgurations in the RFQ depend on time or phase. The intervane voltage is minimum at −180◦ , assuming a cosine wave, and maximum at 0◦ . The x and y beam ellipses will be upright at −180◦ (with, e.g., maximum x projection and minimum y projection), and upright with x and y projections interchanged at 0◦ . At −90◦ and +90◦ , the ellipses are tilted and the beam is round. Now consider an accelerating cell and a synchronous particle with phase φ = −30◦ . This means that the synchronous particle is at the middle of the cell when f = −30◦ . It is at the beginning of the cell when φ = −120, and is at the end of the cell when φ = +60◦ . The phases of the synchronous particle range from −120◦ to +60◦ . At 30◦ after the beginning of the cell the phase of the synchronous particle is −90◦ where the phase-space ellipses are tilted and the beam is round. At 30◦ after the particle reaches the center, the phase of the synchronous particle is 0◦ , where the ellipses are upright. Thus, the ellipses are almost upright when the synchronous particle is at the middle of the cell. The synchronous particle is at the end of the cell at 30◦ before +90◦ , where the ellipses are tilted and the beam is round. Thus, for a synchronous phase of

271

272

8 Radiofrequency Quadrupole Linac

−30◦ , the beam is almost round when the synchronous particle is at the end of the cell. Next, consider a transition cell that follows an accelerating cell. Again we consider a synchronous phase φ = −30◦ , and we ignore any change of β in the transition cell and treat β as a constant. We use the approximate value √ for the transition-cell length of LT = 3βλ/4 = 0.433βλ, where βλ/2 is the length of the last accelerating cell. The total phase change in the transition √ cell is φ = 360 LT /βλ = 90◦ × 3 = 155.9◦ . Thus, the synchronous-particle phase changes from −120◦ at the beginning of the transition cell to +35.9◦ at the end. At the end of the transition cell, this phase is 35.9◦ away from 0◦ where the ellipses are upright. Thus, the ellipses are almost upright when the synchronous particle is at the end of the transition cell. Suppose we want the beam to be round and have tilted ellipses when the synchronous particle is at the output. We could achieve this if we add an m = 1 vane extension, as discussed in the previous section, whose length changes the phase of the synchronous particle from +35.9◦ to +90◦ . The required phase shift for the m = 1 extension is 90 − 35.9 = 54.1◦ . The corresponding length for the m = 1 extension is L1 = 54.1βλ/360 = 0.15βλ. Another round output-beam solution with x and y ellipses interchanged is to add an m = 1 extension that gets the synchronous particle from +35.9◦ to +270◦ . The required phase shift for the m = 1 extension is 270 − 35.9 = 234.1◦ . The corresponding length for the m = 1 extension is L1 = 234.1βλ/360 = 0.65βλ. The round output beams are usually good solutions for cases where the beam must be matched into a FODO (+−) quadrupole-focusing system, as shown in Fig. 8.24 Another possibility is the case where a beam must be matched into a FODODOFO (+ − −+) focusing system. In this case good solutions are usually obtained when the output-beam phase is 0◦ or 180◦ , when the beam ellipses are upright with a larger size in one plane and smaller size in the other plane. We could achieve the 180◦ solution, if we add an m = 1 extension whose length changes the synchronous-particle phase from +35.9◦ to +180◦ . The required phase shift for the m = 1 extension is 180 − 35.9 = 144.1◦ .

RFQ A

Plane x − x′ y − y′ f−w

DTL D

Figure 8.24 Matched phase-space drawings at end of RFQ and beginning of DTL (courtesy of K.R. Crandall).

8.16 Tuning for the Desired Field Distribution in an RFQ

The corresponding extension length is L1 = 144.1βλ/360 = 0.40βλ. The upright-ellipse solution with the x − x and y − y ellipses interchanged corresponds to an output phase of 0◦ or 360◦ . For this case we can add an m = 1 extension whose length changes the synchronous-particle phase from +35.9◦ to +360◦ . The required phase shift for the m = 1 extension is 360 − 35.9 = 324.1◦ . The corresponding length for the extension is L1 = 324.1 βλ/360 = 0.90 βλ

(8.107)

8.16 Tuning for the Desired Field Distribution in an RFQ

Consider an array of adjustable tuning stubs, also called slug tuners, installed in the outer cylindrical walls in each quadrant of an RFQ. Adjusting the radial penetration of these tuning stubs allows us to adjust the magneticﬁeld distributions. By proper adjustment of these stubs both longitudinal and azimuthal distributions can be adjusted in an RFQ. There would normally be four tuners at each longitudinal position, one in each quadrant. The longitudinal squared magnetic-ﬁeld distribution near the outer wall of an RFQ can be measured by the standard bead pull method using a metallic bead as described in Section 5.13. The problem to be solved by one who tunes the cavity is to ﬁnd the mechanical settings of the tuners that produce the desired magnetic-ﬁeld distributions in the operating quadrupole mode. The magnetic-ﬁeld distribution at each longitudinal position is described in terms of the quadrupole and dipole amplitudes of the ﬁeld, which can be obtained as discussed in Section 8.11. If B1 , B2 , B3 , and B4 are the measured ﬁelds in the quadrants at a given longitudinal position, then the quadrupole amplitude is AQ = (B1 − B2 + B3 − B4 )/4 and the dipole amplitudes are AD1 = (B1 − B3 )/2 and AD2 = (B2 − B4 )/4. A good check on the measurement is that B1 + B2 + B3 + B4 = 0. The desired values of the dipole amplitudes are AD1 = AD2 = 0. These equations imply that the four ﬁelds in each quadrant at each longitudinal position should all be equal for a pure quadrupole mode. The desired longitudinal proﬁle for AQ (z) is a known function speciﬁed by the designer. It is sometimes ﬂat or uniform, but may have a tilt or a more complicated proﬁle depending on the beam dynamics requirements. Suppose we assume a linear relationship between each tuner setting and the quadrupole or dipole amplitudes at any given longitudinal position, and assume that there are no measurement errors. If the problem is linear and error-free, we could use a familiar method described below to solve a set of simultaneous linear equations to ﬁnd a solution for the tuner settings that give the desired amplitudes. However, in the real world one may expect some errors in setting the tuner positions and in measuring the ﬁelds. Even for these cases, we can still ﬁnd a solution for the tuner settings by iterating on the simultaneous linear equation solution.

273

274

8 Radiofrequency Quadrupole Linac

From the linearity assumption, we write Vi − V0i =

N ∂Vi j=1

∂Tj

(Tj − T0j ),

i = 1, P

(8.108)

where Vi are the desired AQ , AD1 , or AD2 quadrupole or dipole ﬁeld amplitudes at longitudinal locations i, which are known. The desired amplitudes are AQ (z), and AD1 (z) = AD2 (z) = 0. These V0i are the actual ﬁeld amplitudes at locations i, which are known from measurement. The Tj is the unknown desired setting for the tuners at locations j. The T0j is the actual setting for the tuners at locations j. Each ∂Vi /∂Tj is the measured derivative of ith amplitude with respect to jth tuner. In matrix form this equation is V = MT, where V is the amplitude vector, which is known, T is the tuner vector which is unknown since we do not know the correct tuner settings, and Mij = ∂Vi /∂Tj is the derivative matrix, which is obtained from measurements. The tuner vector is obtained from the matrix equation above by inverting the M matrix. Then, one obtains the solution for the tuner settings by multiplying by the inverse of the derivative matrix to obtain T = M−1 V. From the resulting T vector, one can then adjust the tuners to new settings, and then remeasure the ﬁelds and to obtain a new amplitude distribution. If the new amplitude vector distribution is not close enough to zero, one can use the new V0i and T0j to carry out a new iteration.

8.17 Four-Rod Cavity

The four-rod cavity [11] is used mostly in the lower-frequency range, below about 200 MHz, and is the most commonly used RFQ structure for very low-velocity heavy ions. In principle the four-rod structure, shown in Figs. 8.25 and 8.26, is similar to the Wider¨oe structure. The four rods are charged from a linear array of conducting support plates or inductive stems. In the ideal

Figure 8.25 Four-rod RFQ (courtesy of A. Schempp).

8.17 Four-Rod Cavity

Figure 8.26 Cross section of the four-rod RFQ (courtesy of A. Schempp).

quadrupole geometry, adjacent rods are at opposite potentials, and opposite rods are at the same potential. The opposite pairs of rods are connected to the same plates and the plate connections alternate from one rod pair to the next along the length of the cavity. Because of the low-impedance path between the opposite rods (opposite rods are shorted together), dipole-mode frequencies are much higher than that of the quadrupole mode. Therefore, accidental degeneracy caused by the dipole modes is not a problem for the four-rod cavity, which helps to improve the stability of the ﬁelds. The electric ﬁeld is concentrated near the rods, and the magnetic ﬁelds are concentrated near the inductive stems. The charging currents ﬂow longitudinally along the base from one stem to the next. The current density on the stems is higher than that on the vanes of the four-vane structure, which tends to reduce the efﬁciency compared with the four-vane cavity. This can be offset by the reduced capacity between the rods, which reduces the required charging current for a given voltage. The outer walls of the cavity carry little current, and the placement of the outer walls does not have a large effect on the resonant frequency. Consequently, the transverse size of the four-rod cavity can be very compact, an advantage for a low-frequency structure. The modulation pattern in the rods can easily be machined on a lathe, but can be machined on a milling machine in the form of short vanes. At higher frequencies, the rods become smaller and can be more difﬁcult to cool than vanes. This can be a disadvantage for high-frequency applications, and is one of the main reasons why the four-vane structure is usually preferred for high frequencies at high duty factor.

275

8 Radiofrequency Quadrupole Linac

276

8.18 Four Vane with Windows RFQ

At present the most commonly used RFQ structures are the four-vane and the four-rod structures. Usually the four-vane structure is preferred at frequencies above about 200 MHz because of its high RF power efﬁciency, and the four-rod structure is preferred below 200 MHz because of its compact size at low frequencies. Another type of RFQ structure has attracted interest more recently [12–16], which may be called a four vane with windows RFQ. The four vane with windows RFQ may be described as an intermediate resonator conﬁguration between the four-rod and the four-vane RFQ. It may be conﬁgured as a four-vane RFQ with holes or windows in the vanes (Fig. 8.27), or as a four-rod RFQ with individual stems supporting each of the four rods, where the stems are oriented 90◦ apart (Fig. 8.28). With its close relationship to both the four-vane and the four–rod structures, there would seem to be no reason why this structure could not be used over a broader frequency range that encompasses the range of both the fourvane and the four-rod structures. Most importantly, because of the strong

Type Monitor Maximum Maximum arrow Frequency Phase

= = = = = =

x H−Field (peak) Mode 1 16204.2 A/m 5815.06 A/m 57.6829 90°

Figure 8.27 Four vane with windows RFQ conﬁgured with windows in the vanes and with a fourfold antisymmetric window arrangement where the windows in adjacent vanes are placed longitudinally 180◦ apart (courtesy of P. Ostroumov).

5.02e + 003 A/m

8.18 Four Vane with Windows RFQ

Figure 8.28 Four vane with windows RFQ conﬁgured as a structure with radial stems that support the four vanes or rods. This structure has a fourfold symmetric window arrangement (courtesy of D. Swenson).

magnetic coupling through the windows between neighboring quadrants, the introduction of windows can lower the frequency of the quadrupole operating mode well below that of the dipole modes, reducing the undesirable dipole-mode degeneracy, which complicates the RFQ tuning in a four-vane resonator. At the same frequency, the four vane with windows RFQ is more radially compact than the four-vane RFQ, and is less compact than that the four-rod RFQ. At a given frequency, the four vane with windows RFQ is less efﬁcient than the four-vane RFQ and is more efﬁcient than the four-rod RFQ. The antisymmetric window conﬁguration, shown in Fig. 8.27, where adjacent windows are shifted by 180◦ is more effective in separating the dipole modes from the quadrupole mode. However, the four vane with windows RFQ can also be built to maintain fourfold symmetry of the windows. This allows the design of an RFQ with zero on-axis potential at the entrance and exit gaps, making matching easier at the entrance and avoiding uncertainty in the output energy at the exit. Delayen [12] gives quadrupole and dipole-mode frequency results obtained from the electromagnetic-ﬁeld-solver code MAFIA for different windowgeometry parameters. Delayen has also developed a transmission-line model [12] for the case where the windows extend to the outside diameter of the structure, corresponding to b = 0 in Fig. 8.29. The resonant wavelength λ for the quadrupole mode satisﬁes πt ε0 2t 2π L − t 2πh tan + = +α (8.109) 2 tan λ λ 2 λ C g

277

278

8 Radiofrequency Quadrupole Linac

a /2

t /2 r

d L /2

Figure 8.29 Drawing of one quadrant of the periodic four vane with windows RFQ structure with period L (courtesy of J. R. Delayen).

The capacitance per unit length of the four rods, C , is given in picofarad per meter, by 31.045 39.365 f + + 25.28 ln 1 + , C = √ d d+r d+r 1− 2+ cosh−1 √ r r 2 (8.110) √ where a is a constant of order 3, and g ∼ = [(2R − h)/2 2 − r]. The right side of Eq. (8.109) is valid when t < L − αg/2. When L − αg/2 < t < L, the right side of Eq. (8.109) is replaced by 2ε0 L/C g.

Problems

8.1. Assume an RFQ cell has the following parameters: βs = 0.025, λ = 0.75 m, a = 1.875 mm, and V = 50 kV. (a) Assume the vane-modulation parameter is m = 2. Calculate the focusing efﬁciency X, the characteristic radius r0 , the acceleration efﬁciency A, and the average axial ﬁeld E0 . If the synchronous phase is φ = −30◦ , what is the energy gain of the synchronous particle in the cell? (b) Repeat part (a) assuming m = 1. 8.2. The alternating-gradient focusing principle allows the time-varying electric quadrupole to produce a net focusing effect in both planes. Suppose that the time-dependent electric ﬁelds for the electric quadrupole are derivable in the quasistatic approximation from a scalar potential with

Problems

harmonic time dependence, given by V0 [x2 − y2 ] U(x, y) = sin ωt 2 a2 where ω is the RF frequency. We will see that by adding the timedependent factor, a time-dependent polarity variation is produced, which results in a net focusing effect in both planes. (a) Write the equation of motion for the y plane. (b) Substitute a trial solution in the form of an amplitude-modulated sinusoid, y = C[1 + ε sin ωt] sin t, where ε is called the ﬂutter amplitude of the rapidly oscillating RF factor, C is a constant, and is the frequency of the slowly varying average trajectory. Assume that /ω 1 and ε 1. By substituting this solution into the equation of motion, identify in lowest order the value of ε that makes the trial solution approximately valid. (c) Show that the averaged trajectory is conﬁned by the time-dependent focusing system, by considering the equation of motion averaged over an RF period, and showing that the averaged trajectory satisﬁes the equation of simple harmonic motion. Calculate the frequency of this average motion. 8.3. To understand how an RFQ quadrupole-focusing geometry can be modiﬁed to function as a linear accelerator, we need to see how to perturb the electrodes to obtain a longitudinal electric ﬁeld, or equivalently to obtain a spatially varying axial potential. (a) Suppose that the x-pole tips at voltage +V0 /2 have radius a, and the y-pole tips at voltage −V0 /2 have radius ma with m ≥ 1. Show that the potential function for this geometry is given by X 2 V0 2 [x − y ] + A U(x, y) = 2 a2 where X and A are constants (called focusing and acceleration efﬁciencies), by ﬁnding the expressions for X and A as a function of the parameter m that satisfy the boundary conditions. How do X and A change as a function of m, if the radii of the x and y pole tips are interchanged to ma and a respectively, but their voltages remain the same? 8.4. Consider the three-dimensional time-dependent potential function with a spatially oscillating longitudinal term (x2 − y2 ) V0 + AI (kr) cos(kz) sin(ωt + φ) X U(x, y, z, t) = 0 2 a2 which, by direct substitution, can be shown to be a solution of Laplace’s equation. The parameters X and A are constant focusing and acceleration efﬁciencies, I0 is the modiﬁed Bessel function of order zero, r = x2 + y2 and k = 2π/βλ, where βc is the velocity of the synchronous beam particle (c is the speed of light), λ = 2πc/ω is the RF wavelength and βλ is equal to the distance traversed by a synchronous particle in one RF period.

279

280

8 Radiofrequency Quadrupole Linac

The voltage of the horizontal poles is V0 sin(ωt + φ)/2, and that of the vertical poles is −V0 sin(ωt + φ)/2. The geometry in the y − z plane is shifted axially by 180◦ . At z = 0 the horizontal pole-tip coordinates are x = a, y = 0, and the vertical pole-tip coordinates are x = 0, y = ma, where m ≥ 1. (a) Derive the expression for X and A as a function of the geometry parameters a, m, and k, by assuming that the boundary conditions on the pole tips are satisﬁed at z = 0. (b) Derive the expression for the pole tips in the x − z and y − z planes. (c) Obtain the expressions for the three electric-ﬁeld components. (d) In the approximation that the radial position and the velocity of a particle within a period are constant, calculate the energy gain W for a synchronous particle with velocity βc, by replacing ωt = 2πz/βλ and integrating the axial electric ﬁeld seen by the particle over one period. 8.5. An RFQ is designed to operate at a frequency of 400 MHz and has a vane length of 2.25 m. An RF-drive port that is introduced at the midpoint of the RFQ cavity lowers the resonant frequency by 4 MHz. Assuming that this perturbation can be approximated by a δ-function error at the midpoint, and that ﬁrst-order perturbation theory is valid, calculate the fractional-ﬁeld error at both ends and at the midpoint of the cavity, and plot the three values versus distance along the cavity. You may use the series summations given in the text material. 8.6. An error in the fabrication of a 400-MHz RFQ, whose vane length is 1.5 m, results in unequal dimensions for the two end cells. Assume that this perturbation can be approximated by equal and opposite δ-function errors at each end of the RFQ, such that the cavity frequency error at one end alone would produce a resonant-frequency error of 2 MHz and that at the opposite end would produce a frequency error of −2 MHz. Using ﬁrst-order perturbation theory: (a) What is the resonant frequency of the cavity? (b) Calculate the fractional-ﬁeld error at both ends and at the midpoint of the cavity, and plot the three values versus distance along the cavity. Use these three points versus distance along the RFQ to calculate a slope that characterizes this ﬁeld tilt. You may use the appropriate series summations as given in the text material. References 1 Kapchinskiy, I.M. and Tepliakov, V.A.,

4 Crandall, K.R., Stokes, R.H. and

Prib. Tekh. Eksp. 2, 19–22 (1970). 2 Kapchinskiy, I.M. and Tepliakov, V.A., Prib. Tekh. Eksp. 4, 17–19 (1970). 3 The emittance of a beam, which will be deﬁned in Chapter 9, is a measure of its phase space extent, and emittance growth reduces the minimum achievable beam size for a given focusing system.

Wangler, T.P., Proc. 1979 Linac Conf., Brookaven National Laboratory Report BNL-51134, 1979, pp. 205–216. 5 There is no TE210 mode of a cylindrical cavity with conducting end walls, because the conducting end-wall boundary condition requiring a zero transverse electric-ﬁeld component is not satisﬁed.

References 6 Schrage, D., et al., Proc. of 1988 Linac

Conf., Williamsburg, VA, CEBAF Report 89-001, p. 54; Schrage, D., et al., Proc. of 12th Int. Conf. on Applications of Accelerators in Research and Industry, NIM B79, 1993, p. 372; Schrage, D., et al., Proc. of 1994 Int. Linac Conf., Tsukubu, Japan. Some references to RFQ tuning methods may be found in Stokes, R.H. and Wangler, T.P., Radiofrequency quadrupole accelerators and their applications, Annu. Rev. Nucl. Part. Sci. 38, 97–118 (1988); A resonant-coupling method is described in Young, L.M., Proc. 1993 Particle Accel. Conf., May 17–20, 1993, Washington, D.C., p. 3136; Proc. 1994 Linear Accel. Conf., August 21–26, 1994, Tsukubu, Japan, p. 178. Mathews, J. and Walker, R.L., Mathematical Methods in Physics, 2nd ed., W. A. Benjamin, 1970, Chapter 10. Crandall, K.R., ‘‘RFQ Radial Matching Sections and Fringe Fields’’, Proc. 1984 Linear Accelerator Conf., Seeheim, Germany, May 7–11. Crandall, K.R., ‘‘Ending RFQ Vanetips With Quadrupole Symmetry’’, Proc. of 1994 Linac Conf., Tsukuba, Japan, pp. 227–229.

11 For a review of the four-rod RFQ

structure, see Schempp, A., ‘‘Radio-Frequency Quadrupoles’’, CERN Accelerator School, CERN 92-03, Vol II, p. 522. Delayen, et al., ‘‘Design and Modeling of Superconducting RFQ Structures’’, Proc. of the 1992 Linear Accel. Conf., Aug. 24–28, 1992, Ottawa, Ontario, Canada, pp. 692–694. Shepard, et al., ‘‘Design for a Superconducting Niobium RFQ Structure’’, Proc. of the 1992 Linear Accel. Conf., Aug. 24–28, 1992, Ottawa, Ontario, Canada, pp. 441–443. Donald A., Swenson, Wayne, Cornelius and Young, P.E., ‘‘Segmented Vane Radio-Frequency Quadrupole Linear Accelerator’’, United States Patent 5,430, 359, ﬁled Nov. 2, 1992, Date of Patent July 4, 1995. Andreev, V.A. and Parisi, G., ‘‘90◦ Apart Stem RFQ Structure for Wide Range of Frequencies’’, Proc. of 1993 Part. Accel. Conf., Washington, DC, May 13–18, pp. 3124–3126. Ostroumov, P.N. and Kolomiets, A.A., Design of 57.5 MHz CW RFQ for medium energy heavy ion superconducting linac, Phys. Rev. Spec. Top. Accel. Beams 5, 060101 (2002).

281

282

9 Multiparticle Dynamics with Space Charge

Particle motion in a linac depends not only on the external or applied ﬁelds, but also on the ﬁelds from the Coulomb interactions of the particles, and the ﬁelds induced by the beam in the walls of the surrounding structure. In Chapters 6 and 7, we treated the longitudinal and transverse dynamics, but we included only the applied forces from the radio frequency (RF) and from static focusing ﬁelds. The Coulomb forces play an increasingly important role as the beam-current increases, and that topic is covered in this chapter. The effects caused by the interactions with the structure are the subjects of beam loading and wake ﬁelds, which are treated in Chapters 10 and 11. The Coulomb effects in linacs are usually most important in nonrelativistic beams at low velocities, because at low velocities the beam density is larger, and for relativistic beams the self-magnetic forces increase and produce a partial cancellation of the electric Coulomb forces. The net effect of the Coulomb interactions in a multiparticle system can be separated into two contributions. First is the space-charge ﬁeld, the result of combining the ﬁelds from all the particles to produce a smoothed ﬁeld distribution, which varies appreciably only over distances that are large compared with the average separation of the particles. Second are the contributions arising from the particulate nature of the beam, which includes the short-range ﬁelds describing binary, small impact-parameter Coulomb collisions. Typically, the number of particles in a linac bunch exceeds 108 , and the effects of the collisions are very small compared with the effects of the averaged space-charge ﬁeld [1]. To describe the space-charge ﬁeld, we need to understand the properties of an evolving particle distribution, which requires a self-consistent solution for the particles and the associated ﬁelds. This is a problem, which has been formulated in terms of the coupled Vlasov–Maxwell equations, for which there are no generally successful analytic solutions in a linac, and computer simulation is the most reliable tool. In this chapter we include an introduction to the most common methods used for computer simulation of the space-charge forces. We also ﬁnd that among the most important properties of the distribution besides the beam current are the emittances, which are deﬁned from the second moments of the particle distribution and are measures of the phase-space areas occupied by RF Linear Accelerators. 2nd, completely revised and enlarged edition. Thomas P. Wangler Copyright  2008 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-40680-7

9.1 Beam Quality, Phase Space, and Emittance

the beam in each of the three projections of position-momentum phase space. The emittances are important measures of the beam quality; they determine the inherent capability of producing, by means of a suitable focusing system, small sizes for the waists, angular divergences, micropulse width, and energy spread. The signiﬁcance of the space-charge ﬁelds is not only that they reduce the effective focusing strength, but also the nonlinear terms, a consequence of the deviations from charge-density uniformity, cause growth of the rms emittances, which degrades the intrinsic beam quality. One consequence of space-charge-induced emittance growth is the formation of a low-density beam halo surrounding the core of the beam, which can be the cause of beam loss, resulting in radioactivation of the accelerating structure.

9.1 Beam Quality, Phase Space, and Emittance

A beam bunch in a linac consists of a collection of particles, moving in approximately the same direction, with approximately the same positions, phases, and energies. The term beam quality, although not precise, refers to a coherence property of the beam: the degree to which the beam particles have nearly the same coordinates as the reference particle, which travels on the accelerator axis with a speciﬁed energy and phase. The ideal beam with highest beam quality is called the laminar beam because it exhibits laminar-like ﬂow. A laminar beam represents the ideal of a highly ordered and coherent beam, which is never exactly realized. As a quantiﬁer of beam quality, it is more convenient to work with a directly measurable quantity known as beam emittance, which is introduced in this section. Every particle in the beam can be described by three position and three momentum coordinates for a total of six coordinates per particle. Then, each particle is represented by a single point in the six-dimensional phase space of coordinates and momenta. In practice, it is more convenient to work with the two-dimensional phase-space projections of the beam, which correspond more directly to what can be measured. For a linac beam, one can refer to a collection of points in the three normalized phase-space projections x − px /mc, y − py /mc, and z − pz /mc, where x, y, and z are the coordinates, and px , py , and pz are the momentum components. Phase-space areas can be associated with the collection of points inside speciﬁed density contours in each of the two-dimensional phase-space projections, as shown in Fig. 9.1. Instead of the transverse momenta, it is convenient to measure the divergence angles, dx/ds and dy/ds. Plots of x − dx/ds and y − dy/ds are known as the trace-space or unnormalized phase-space projections. In longitudinal phase space, position and momentum relative to the synchronous particle can be used, but more often, these are replaced by the phase and energy variables φ and W. Before deﬁning beam emittance, we need to recognize the special signiﬁcance of elliptical phase-space distributions. The beam-phase-space

283

284

9 Multiparticle Dynamics with Space Charge x′ x

(a)

∆w

px /mc

∆f

(b)

(c)

Figure 9.1 Isodensity contours of the beam in (a) unnormalized transverse phase space, (b) normalized transverse phase space, and (c) longitudinal phase space. The emittance associated with any elliptical contour is the area of that contour divided by π .

contours in a linac often have the approximate shape of an ellipse. One reason for this shape is related to the predominance in most accelerators of linear focusing forces. With linear focusing, the trajectory of each particle in phase space lies on an ellipse, which may be called the trajectory ellipse. In general, a beam is said to be matched when the phase-space isodensity contours of the beam are concentric and geometrically similar to the trajectory ellipses. Therefore matched beams in linear focusing systems have elliptical density contours. Furthermore, with linear focusing, elliptical distributions in phase space, even for unmatched beams, remain elliptical. Because of the tendency of linac beams to exhibit approximately elliptical phase-space distributions, it has become conventional to deﬁne for each two-dimensional projection, a quantity called the emittance, which is proportional to the area of a chosen beam ellipse. The general equation for an ellipse is written in Chapter 7 as ˜ 2 = ε ˜ + βx γ˜ x2 + 2αxx

(9.1)

where γ˜ β˜ − α˜ 2 = 1, and where the area of the ellipse is πε. The deﬁnition of the emittance ε associated with any two-dimensional elliptical phase-space area is ε=

Area π

(9.2)

Having deﬁned the emittance in terms of an elliptical area in 2D phase space, we return to the question of which phase-space area to choose to characterize a real beam. If the beam distribution in phase space had a welldeﬁned boundary, the emittance could be deﬁned simply by the area within that boundary. Generally, real beams do not have well-deﬁned boundaries. One method for assigning an emittance is to choose a speciﬁc density contour in phase space, such as one at 50 or 90 or 95% of the maximum density. It can be shown [2] that under certain conditions such emittances are conserved: (1) when Liouville’s theorem is satisﬁed in the six-dimensional phase-space, and (2) when the forces in the three orthogonal directions are uncoupled. As discussed later in this chapter, Liouville’s theorem is satisﬁed when there are no

9.2 RMS Emittance

dissipative forces, no particles lost or created, and no small-impact-parameter binary Coulomb collisions between particles. When these conditions hold, the volume of six-dimensional phase space deﬁned by any ﬁxed-density contour of the beam is invariant. While this is an important characteristic, the most useful deﬁnition of emittance in a linac is the rms emittance, which is described in the following section. 9.2 RMS Emittance

The presence of nonlinear forces can produce a considerable departure from elliptical trajectories, and can distort the phase-space contours. If the phase-space projections are described by effective phase-space ellipses, such distortions cause an effective-emittance growth, resulting from a ﬁlamentation process in six-dimensional phase space, in which outer ﬁlaments develop, enclosing regions with rareﬁed density and resulting in a dilution of the phase-space density. Furthermore, forces that couple the motion between the three directions can cause real growth in the two-dimensional projected areas, even though the six-dimensional volume is constant. Consequently, it becomes important to introduce a deﬁnition of an effective emittance. The deﬁnition is most conveniently based on mean-square values, or second moments of the coordinates and momenta. An rms emittance deﬁnition is more useful because it can easily be deﬁned for an arbitrary particle distribution, being determined only by the rms characteristics of the beam distribution. As shown in Fig. 9.2, the general form of the rms ellipse is deﬁned by Courant–Snyder parameters α˜ r , β˜r , and γ˜r , and an rms emittance εr . To be speciﬁc we describe the deﬁnitions for an unnormalized transverse emittance in x –x space. The point (x, x ) lies on the rms ellipse if γ˜r x2 + 2α˜ r xx + β˜r x2 = εr

(9.3)

where γ˜r β˜r − α˜ r2 = 1. We deﬁne the ellipse parameters in the following manner. First, we require that the ellipse projections on the x and x axes are equal to the rms values of the distribution. Thus, we require that x2 = β˜r εr

and x2 = γ˜r εr

(9.4)

Next we deﬁne α˜ r . For the case of an ideally matched beam with linear focusing, where the isodensity contours of the beam are concentric and similar to the trajectory ellipses, we would like to deﬁne α˜ r so that the rms ellipse also coincides with a trajectory ellipse. Then, because the trajectory ellipses relate α˜ and β˜ by α = −β˜ /2, we require the same relationship for αr and βr . Thus, we deﬁne α˜ r = −

xx β˜r 1 dx2 =− =− 2 2εr ds εr

(9.5)

285

286

9 Multiparticle Dynamics with Space Charge x′

Slope = −

x ′rms = eg∼r

∼ a r ∼ br

x xrms =

∼ ebr

Figure 9.2 The rms ellipse.

where we have assumed that the emittance of the matched beam is constant. The Courant–Snyder parameters for an rms ellipse have been deﬁned in terms of the second moments of the distribution from Eqs. (9.4) and (9.5). We recall that the three ellipse parameters are not independent, but are related by γ˜r β˜r − α˜ r2 = 1. The rms emittance as a function of the second moments is obtained by substituting Eqs. (9.4) and (9.5) into this relationship, which leads to (9.6) εr = x2 x2 − xx2 A simple exercise may be helpful in understanding an important characteristic of the rms emittance. Consider an idealized particle distribution in phase space representing a beam that lies on some line that passes through the origin. This is illustrated by two examples, shown in Fig. 9.3. Assume that, for any x, the divergence x of the particles is given by x = Cxn

(9.7)

where n is positive, and C is a constant. The second moments of this distribution are easily calculated using Eq. (9.7), and the squared rms emittance is given by 2 2 εr2 = x2 x2 − xx = C2 x2 x2n − xn+1 (9.8)

x′

(a)

(b)

Figure 9.3 The rms emittance εr for two distributions with zero area: (a) straight line where εr = 0, (b) curved line where εr = 0.

9.3 Transverse and Longitudinal Emittance

When n = 1, the line is straight and the rms emittance is εr = 0. When n = 1, the relationship is nonlinear, the line in phase space is curved, and the rms emittance is in general not zero. Both distributions have zero area. Therefore, we conclude that, even when the phase space area is zero, if the distribution lies on a curved line, its rms emittance is not zero. The rms emittance depends not only on the true area occupied by the beam in phase space, but also on the distortions produced by nonlinear forces.

9.3 Transverse and Longitudinal Emittance

Transverse emittance can be determined from measurements of the particle distribution as a function of displacement and angular divergence. This distribution determines what is called the unnormalized emittance. For example, an upright ellipse, with semiaxes x and x , has an unnormalized emittance ε = xx

(9.9)

For an accelerator, it is more convenient to deﬁne what is called the normalized emittance, for which the dimensionless transverse momentum variable px /mc is used instead of the divergence. The normalized and unnormalized emittances are related by εn = εβγ

(9.10)

The reason for introducing a normalized emittance is that the transverse momenta of the particles are unaffected by acceleration, but the divergences of the particles are reduced during acceleration because x = px /p. Thus, acceleration reduces the unnormalized emittance, but does not affect the normalized emittance. The typical unit of either unnormalized or normalized emittance is meter-radians or millimeter-milliradians and typical normalized rms emittance values range from a few tenths to a few millimeter-milliradians. The variables used for longitudinal emittance can be chosen in several ways. First, we write a normalized longitudinal emittance for an upright ellipse as εn =

zp tW φW = = mc mc ωmc

(9.11)

where we used the relationships vp = W, z = vt, and φ = ωt. This emittance is expressed in the same units of length as transverse emittances. However, phase difference φ and energy difference W are more convenient longitudinal variables in a linac, and a more common form for longitudinal emittance in a linac is ε = φW

(9.12)

287

288

9 Multiparticle Dynamics with Space Charge

The units for Eq. (9.12) are usually expressed in deg-MeV. This emittance is proportional to the normalized emittance and does not change because of acceleration; unfortunately, it does depend on the RF frequency. For example, if the frequency of the linac doubles at a certain energy, the emittance ε also doubles. For a linac with a frequency transition, an emittance deﬁned by tW would not change when the frequency changes.

9.4 Emittance Conventions

Lack of agreement on emittance conventions has led to much confusion. One ambiguity is whether emittance is deﬁned as area/π as we have done, or as area. One convention in the literature is to quote the emittance value as the area, showing an explicit factor of π; for example, ε = 1π mm-mrad. For those who interpret the emittance as area/π, as done in this book, the emittance is really the factor multiplying the π; thus in the example, the emittance is really 1 mm-mrad. The rms emittance has other areas of confusion. Some people multiply the rms emittance that we have deﬁned, which is the Sacherer [3] convention, by 4 and call this the rms emittance. This is the Lapostolle [4] convention, which is useful because, for continuous beams, it gives the total emittance of an equivalent uniform beam. Also, for almost all other real beams, the value of the rms emittance is close to the emittance containing 90% of the beam. Use of the Lapostolle convention becomes confusing, only when it is also called the rms emittance. One way to avoid confusion is to call this the 4-rms emittance. Some have also found it convenient to quote an emittance of 6 times the rms emittance. This contains 95% of the beam for Gaussian beams. If this is called the 6-rms emittance instead of the rms emittance, it should lead to no confusion. The only safe solution for these problems is to carefully deﬁne the emittance. Because of the lack of a convention for deﬁning and expressing emittance, the reader of accelerator literature must beware, especially if the authors have not carefully deﬁned the emittance they quote. It is sometimes useful to refer to the total emittance of a distribution that has been generated for computer simulations. A useful result relates the total emittance to the rms emittance for uniform hyper-ellipsoids in n-dimensional space. This was calculated by Weiss [5] and is εr =

εtotal n+2

(9.13)

Some examples include n = 6, uniform density in 6D space, known as the 6D Waterbag distribution where εr = εtotal /8; n = 4, uniform density in 4D space, known as the 4D Waterbag distribution where εr = εtotal /6; n = 3, uniform in all three-dimensional projections where εr = εtotal /5; and n = 2, uniform density

9.5 Space-Charge Dynamics

in all two-dimensional projections, known as the Kapchinsky–Vladimirsky distribution where εr = εtotal /4.

9.5 Space-Charge Dynamics

Before beginning our discussions of space-charge and other high-intensity effects, we note that in the usual nomenclature for bunched beams in linacs, there are several different deﬁnitions of current that are used, which often creates confusion, even among linac experts. For proton linacs there are two deﬁnitions of current that are most frequently speciﬁed for a pulsed machine. One deﬁnition is the value averaged over the RF macropulse, given by I = qNf , where q is the charge per particle, N is the average number of particles per bunch, and f is the bunch frequency. This is usually called the peak current, because it is the current that is present during the macropulse. A second common deﬁnition of current in proton linacs is called the average current. The average current is the current averaged over a full RF macropulse cycle, including both the macropulse time and the time between the macropulses. For a CW proton linac (100% duty factor), the peak and average currents are equal, while for any pulsed linac, the average current is always less than the peak current. A typical high value for the peak current in a proton linac is about 100 mA, and average currents above about 1 mA are usually considered high values. Generally, these two deﬁnitions are sufﬁcient to describe the beam intensity in a proton linac. A more complicated situation occurs when the pulsed-linac beam is chopped, which is done for applications where the linac beam is injected into and stacked in an accumulator ring. An example is an H− linac injector for a neutron-spallation source, where the RF period is about 1 ns, the chopper period is several hundred nanoseconds, and the RF macropulse period is about 1 ms. For this case, one must distinguish between the peak current deﬁned so as to either exclude or include the chopping cycle. The former current is relevant to the beam dynamics, and the latter is of relevance for the performance of the RF system, which has a longer time constant than the chopping period. In addition, for some of the space-charge models one may refer to a dc current over the time duration of the passage of bunch, which equals the average current while the bunch is passing a given point, given by I = qN v, where N is the number of particles per unit length and v is the beam velocity. For electron linacs, the current is often deﬁned differently than for proton linacs. This difference is probably because for the most common electron-linac applications, such as free-electron lasers and linear colliders, what matters most is the charge per bunch, or the charge divided by the time spread of the bunch. Often the electron linacs bunch the beam using a subharmonic-frequency buncher, such that ratio of ﬁlled to total buckets is small. The averaging, which is done for the proton-linac current deﬁnitions, is of little interest for some

289

290

9 Multiparticle Dynamics with Space Charge

electron linacs. A common deﬁnition of electron-linac peak current is the total charge per bunch divided by the full-width half maximum of the micropulse duration. Typical peak currents for high-current electron linacs range from a few tens to a few hundreds of amperes. For some electron-linacs, a current is not quoted, but merely the total charge per bunch. The overall conclusion is that because of the different deﬁnitions of current, one must be careful to determine the precise deﬁnition that is being used. Now, we are ready to begin a general discussion of space-charge effects. With increasing beam intensity the interaction between the charged beam particles becomes more important. The charges produce mutually repulsive electric ﬁelds that act in opposition to the focusing forces, and also magnetic ﬁelds that produce attractive forces. The magnetic forces are smaller than the electric forces, and are unimportant except for relativistic particles. The total Coulomb ﬁeld experienced by any particle is the sum of the ﬁelds due to all the other particles in the beam. For real linac beams with the order of 108 or more particles per bunch, the method of directly summing the ﬁelds from this many particles does not lend itself to a practical computational approach. One may classify the self-interaction of the particles into two categories, the collisional regime, dominated by binary collisions caused by close encounters, and the collective or space-charge regime, the result of many particles that produce an average or collective ﬁeld that can be represented by a smooth ﬁeld as a function of space and time. The space-charge regime occurs when there are enough neighboring particles to shield the effects of density ﬂuctuations, a phenomenon known in plasma physics as Debye shielding. Ignoring the effect of the ﬁnite-sized beam, the criterion for the collective Debye shielding for a beam with number density n is that the number of particles ND in a Debye sphere, [6] a sphere with radius equal to the beam-Debye length λD , is ND = n

4πλ3D 1 3

(9.14)

where λD = ε0 kB T/nq2 and kB is Boltzmann’s constant, and for a spherical bunch with rms projections a, and kB T = mc2 εn /a2 . For accelerator beams this criterion is usually satisﬁed and the smoothed space-charge ﬁeld describes the main effect of the Coulomb interactions. With collisions neglected, Liouville’s theorem, written in the form that expresses continuity of particles in phase space, is ∂f ∂f ∂f df = + x˙ · + p˙ · =0 dt ∂t ∂x ∂p

(9.15)

where f (x, p, t) is the particle density in phase space. Expressing p˙ in terms of the sum of the external ﬁelds plus the smoothed self ﬁelds yields the Vlasov equation, [7] also known as the kinetic equation, or the collisionless Boltzmann

9.5 Space-Charge Dynamics

equation, p ∂f p×B ∂f ∂f + · +q E+ =0 · ∂t γ m ∂x γm ∂p

(9.16)

Rather than describing the motion of single particles, the object is to follow the evolution of the distribution function that is consistent with the electromagnetic ﬁelds. The Vlasov equation and Maxwell’s equations form a set of closed equations, which determine the self-consistent dynamics of a distribution of charges satisfying Liouville’s theorem. Unfortunately, there is no general method for obtaining analytic solutions of this set of equations for linac beams. In practice, equivalent solutions are usually obtained numerically on the computer from the more direct approach of simulating the particle interactions, typically using 104 to 105 simulation particles, as discussed in Section 9.6. As the previous discussion suggests, space-charge-dominated beams behave in some respects like plasmas. Similarities include particle shielding from the effects of density variations and external focusing ﬁelds as described above, and density oscillations at the beam-plasma frequency for space-charge-dominated beams. At ﬁrst, it may seem surprising that a beam with a single charge species can exhibit plasma oscillations, which are characteristic of plasmas composed of two-charge species. But the effect of the external focusing force can play the role of the other charge species to provide the necessary restoring force that is the cause of such oscillations. It is important to bear in mind that beams in linacs are also different from plasmas in some important respects. One is that beam transit time through a linac is too short for the beam to reach thermal equilibrium. Also, because Liouville’s theorem and phase-space conservation are a good approximation, the phase-space volume of the beam is generally a more important parameter than beam temperature. Also, unlike a plasma, the Debye length of the beam may be larger than or comparable to the beam radius, so shielding effects may be incomplete. The space-charge ﬁelds can be separated into linear and nonlinear terms as a function of the displacement from the centroid. The linear space-charge term defocuses the beam and leads to an increase in the beam size. The nonlinear space-charge terms also increase the effective emittance by distorting the phase-space distribution, further increasing the spread of the displacements, divergences, and energy. One must account not only for the electric force between the particles but also for a magnetic force associated with the motion of the charges in the beam. While the electric force causes defocusing, the magnetic force causes attraction. This can be seen by considering the simple example of a long bunch, approximately represented by a cylinder of charge with azimuthal symmetry and charge density qn(r) moving at velocity v. The electric ﬁeld is directed

291

292

9 Multiparticle Dynamics with Space Charge

radially outward and from Gauss’s law is r q n(r)r dr Er = ε0 r

(9.17)

From Ampere’s law, the magnetic ﬁeld is azimuthally directed, and is qvµ0 r n(r)r dr (9.18) Bθ = r The radial Lorentz force is Fr = q(Er − vBθ ) = qEr (1 − β 2 ) = qEr /γ 2 . Thus the attractive magnetic force, which becomes signiﬁcant at high velocities, tends to compensate for the repulsive electric force. Therefore space-charge defocusing is primarily a nonrelativistic effect. It is of great concern for intense low-velocity ion beams, and for electron beams at injection into a linac, but not for relativistic electron beams where β ≈ 1. In the paraxial approximation, Newton’s second law for transverse motion is Fr = mc

d(γβr ) = γ mcβ˙r , dt

d2 r qEr = 3 dt2 γ m

(9.19)

While the magnetic-self force does not affect the longitudinal motion, the longitudinal motion is affected by a relativistic longitudinal mass γ 3 m resulting in an equation of motion d2 z qEz = 3 dt2 γ m

(9.20)

In both cases but for different reasons, the space-charge term in the equation of motion has γ 3 in the denominator.

9.6 Practical Methods for Numerical Space-Charge Calculations

Beam-dynamics codes used for simulation of intense linac beams normally include a subroutine for calculating the space-charge forces. In principle the effects of space-charge can be calculated by adding the Coulomb forces between all the particles. However, to represent bunches with a typical number of particles near 108 or more, this approach is impractical. Several methods for calculating space-charge forces have been developed [8]. The ﬁrst method, used in the program TRACE, [9] is based on the assumption of linear spacecharge forces. Linear space-charge forces allow the use of analytical and matrix methods resulting in a signiﬁcant simpliﬁcation, but imply a uniform density distribution, which is rarely the case for a real beam. However, it was discovered by Lapostolle [4] and Sacherer [3] that for ellipsoidal bunches, where the rms emittance is either constant or speciﬁed in advance, the evolution of the rms beam projections is nearly independent of the density proﬁle. This

9.6 Practical Methods for Numerical Space-Charge Calculations

means that for calculation of the rms dynamics, the actual distribution can be replaced by an equivalent uniform beam, which has the same rms values. The resulting space-charge dynamics of the rms sizes can be calculated by the ellipse transformation method, described in Chapter 7; instead of tracing individual particles, the rms beam envelope is traced. The matrix equation, σ2 = Rσ1 RT

(9.21)

˜ and γ˜ between points determines the transport of the ellipse parameters α, ˜ β, 1 and 2. These parameters are elements of the σ matrix, and in linear systems Eq. (9.21) can be solved by matrix multiplication if the transfer R matrix is given. To account for linear space-charge forces, the space-charge electric ﬁelds, given in Sections 9.8 and 9.9, are used. These ﬁelds depend on the rms beam size, which varies in the manner given by the ellipse β˜ parameter, an element of the σ matrix. To account for space charge, the code treats each space-charge impulse as a thin lens, which imparts a momentum impulse to each particle given by pi = qEsi

s , v

i = x, y, z,

(9.22)

where Esi is the space-charge ﬁeld component, and s is the step size over which the ﬁeld acts [10]. Because of the assumption of linearity, space-chargeinduced emittance growth cannot be calculated by this type of subroutine. However, this type of envelope tracking code is very useful for ﬁnding the beam-ellipse parameters that are matched to a periodic transport system in a linac, and for determining quadrupole gradients needed to produce a matched beam. To describe the evolution of the beam distribution and include spacecharge emittance growth effects, the space-charge calculation requires particle tracking. The simulation particles are known as macroparticles, because as individual sources of the space-charge ﬁeld, each macroparticle represents the total charge of many (typically 104 to 105 ) real beam particles. A frequently used macroparticle-tracking approach is called the particle-in-cell (PIC) method, in which at each step a mesh is superimposed on the bunch, which allows a smoothing of the ﬁelds to reduce the effects of artiﬁcially large forces that would otherwise be caused by binary encounters between macroparticles. The number of particles in each cell is counted, and the smoothed space-charge force acting on each particle is obtained by summing the ﬁelds from the charges in each cell. Finally, the forces are applied to deliver a momentum impulse to each particle. The time required for the calculation depends both on the number of macroparticles and the number of cells. A three-dimensional calculation needs a large number of cells, and some assumption of symmetry is usually made to speed up the calculation. The most commonly used computer subroutine for space-charge calculations is the SCHEFF subroutine used in the PARMILA codes, and written by K. R. Crandall [11]. SCHEFF, an acronym

293

294

9 Multiparticle Dynamics with Space Charge

for space-charge effect, uses a two-dimensional (r–z) PIC method, which calculates the radial and longitudinal space-charge forces for an azimuthally symmetric distribution, or for a distribution with elliptical symmetry by using a transformation to an equivalent azimuthally symmetric distribution. Given either the real or the transformed azimuthally symmetric distribution, each macroparticle is assumed to represent a ring of charge, centered on the axis. In r–z space, a rectangular mesh with Nr radial intervals, each of length r, and Nz axial intervals, each of length z, is superimposed on the bunch. The total charge in each cell is counted, and the radial and longitudinal electric ﬁelds Er and Ez resulting from the charge in each cell is computed at the nodes, located at the corners of each cell. Then, each macroparticle receives an impulse, using an electric ﬁeld obtained by a linear interpolation from the nodes to the coordinates of the macroparticle. Typical runs are made with 104 to 105 particles with a typical mesh given by Nr = 20 and Nz = 40. The advantage of allocating the particles to cells is to allow a smoothing method for eliminating the Coulomb singularities that occur when two particles are too close together. In SCHEFF there are two different options for providing this smoothing. The ﬁrst and simplest option is to assume that, for the purpose of calculating the ﬁelds, the total charge contained within each cell is placed at the location of the centroid of the particles within that cell. The second option is to replace the distribution of charged rings in each cell by n charged rings, whose positions are chosen, using a Gaussian quadrature formula, to approximate a uniform charge density within the cell. The number of rings per cell depends on the cell aspect ratio. If the aspect ratio is less than 2, the program chooses N = 4; if it is between 2 and 4, N = 8, and if it is greater than 4, N = 12. To reduce computer time, a Green’s function technique is used in which Er and Ez at every node due to a unit charge are precalculated and saved in a table. Then at each step the space-charge ﬁelds are obtained by simply scaling the ﬁelds from a unit source charge in proportion to the total charge contained in each cell. The electric ﬁeld components at position (r, z) due to a circular ring of charge centered at coordinates (0, zs ) with radius rs is Er (r, z, rs , zs ) =

Q 4π 2 ε0 r d2 + 4rrs

2 rs − r 2 + (z − zs )2 × K(α) − E(α) d2

(9.23)

Q(z − zs )E(α) (9.24) 2π 2 ε0 d2 d2 + 4rrs where d2 = (r − rs )2 + (z − zs )2 , α = 4rrs /(d2 + 4rrs ), and K(α) and E(α) are complete elliptical integrals of the ﬁrst and second kinds, given Ez (r, z, rs , zs ) =

9.6 Practical Methods for Numerical Space-Charge Calculations

by π/2 K(α) = 0

dθ 1 − α 2 sin2 θ

π/2 E(α) = 1 − α 2 sin2 θ dθ

(9.25)

(9.26)

The procedure for handling the elliptical beam is exact only for a uniformly charged elliptical cylinder, or a uniformly charge ellipsoid [12]. First the ellipticity of the beam is calculated from the parameter ε2 = ax /ay , where ax and ay are the x and y rmssizes of the beam. The effective radius of each particle is calculated as ρ = (x/ε)2 + (εy)2 . Each particle is assigned to a cell in this transformed space, and the electric ﬁelds Er and Ez are calculated in the transformed space. Finally the correction, applied to the Cartesian transverse electric-ﬁeld components to account for the elliptical cross section of the beam, is given by

2 x (9.27) Ex = Er 2 1+ε r

2ε2 y Ey = Er (9.28) 1 + ε2 r where Er is given by Eq. (9.23). Using Eqs. (9.27), (9.28), (9.23), and (9.24), each macroparticle receives an x, y, and z impulse. This feature that allows representation of an elliptical beam, albeit that the procedure is approximate and ignores any departures from transverse ellipticity, converts SCHEFF to a subroutine capable of handling bunched beams in quadrupole focusing systems. The effect of any number of adjacent beam bunches can also be included by assuming periodicity in the bunch train. The calculation for relativistic particles is corrected to include the magnetic effects by Lorentz transforming the particles to the bunch frame, where the space-charge electric ﬁelds are applied, followed by a transformation back to the laboratory frame. Another method for tracking interacting macroparticles is the straightforward particle-to-particle interaction (PPI) approach, in which the total three-dimensional space-charge force on a macroparticle is calculated as the vector sum of the individual forces from all the other macroparticles. To reduce the effects of artiﬁcially large macroparticle deﬂections, some method must be used to reduce the effect of the singularity in the Coulomb force, when two macroparticles get too close together. A typical approach [13] for treating the close encounters is to replace the point macroparticles by a spherical volume with uniform charge density, known as a cloud, which produces a linear force that falls to zero when the centers of the particles coincide. The choice of cloud radius Rc must be optimized, and a range of values of Rc near the average

295

296

9 Multiparticle Dynamics with Space Charge

interparticle spacing can usually be found, where the results are insensitive to the exact value. The main disadvantage of the PPI method is that the computer time required is proportional to the square of the number of particles, making large particle runs prohibitive. Finally, three-dimensional codes have been written, using analytic expressions that relate the space-charge ﬁeld to the particle distribution. The ﬁrst approach [14] of this type, developed at CERN, replaced the macroparticle distribution by an rms-equivalent Gaussian charge density with ellipsoidal symmetry, from which the space-charge ﬁelds were calculated by numerical integration. Although this approach leads to rapid three-dimensional computations, the restriction to a Gaussian proﬁle is in principle not compatible with realistic distributions for intense beams. Recently, other approaches have been proposed to generalize this method, the most recent of which is the proposal of Lapostolle, [15] which describes the macroparticle charge density without requiring any assumption of symmetry by using a sum of Hermite polynomials to describe the beam density. Because the lowest-order term is a Gaussian distribution, this method is a generalization of the original CERN method, which promises to provide improved accuracy for a relatively high-speed, truly three-dimensional calculation. 9.7 RMS Envelope Equation with Space Charge

Analytic expressions are presented in Sections 9.8 and 9.9 for the space-charge ﬁeld of a uniform continuous elliptical beam and a uniform ellipsoidal bunch. These expressions are useful for including space-charge forces in approximate equations of motion, for obtaining current-limit formulas, and for obtaining rms envelope equations. We begin with the latter application, following the early work of Lapostolle [4] and Sacherer [3]. Consider a beam moving in the s direction, where individual particles satisfy the equation of transverse motion x + κ(s)x − Fs = 0

(9.29)

The linear external force is given by −κ(s)x, and Fs is the space-charge force term, which in general is nonlinear and includes both the self-electric and the self-magnetic forces. The quantity Fs is related to the space-charge electric ﬁeld Es by Fs = qEs /γ 3 mv2 . For simplicity we assume that the mean displacement and divergence are zero, and we are interested in ﬁnding an equation of motion for the rms beam size. First, we write the equations of motion for the second moments of the distribution. Thus, dx2 = 2xx ds dxx = x2 + xx = x2 − κ(s)x2 + xFs ds

(9.30) (9.31)

9.8 Continuous Elliptical Beams

and dx2 = 2x x = −2κ(s)xx + 2xx + 2x Fs ds

(9.32)

where the averages are taken over the particle distribution. The ﬁrsttwo equations lead to the equation of motion for the rms beam size, a ≡ x2 . Using Eq. (9.30), we have aa = xx

(9.33)

Differentiating Eq. (9.33), and using Eq. (9.31) gives 2

a −

x2 x2 − xx xx − =0 3 a a

(9.34)

The numerator of the second term will be recognized as the square of the rms emittance, deﬁned by Eq. (9.6). Then, substituting Eq. (9.29) to eliminate x from Eq. (9.34) a + κ(s)a −

εr2 xFs − =0 a3 a

(9.35)

Equation (9.35) is the rms envelope equation, and it expresses the equation of motion of the rms beam size. The second term is the focusing term, and the third term is the emittance term. The emittance term is negative and is analogous to a repulsive pressure force acting on the rms beam size. The last term in Eq. (9.35) is the repulsive space-charge term. The equation of motion for the rms beam size is similar to the single-particle equation of motion, except for the presence of the additional emittance term. The third moment equation, Eq. (9.32), has not been used in this derivation. It can be shown that this equation affects the growth of rms emittance, [16] and to obtain a time-dependent solution, one would need to determine independently the evolution of the beam distribution.

9.8 Continuous Elliptical Beams

Sacherer and Lapostolle independently derived the envelope equations for continuous beams with arbitrary density proﬁles that have elliptical symmetry in x –y space. Although a linac beam is bunched, the continuous beam results are still useful for an approximate description of the transverse ﬁelds of a long bunch. The electric-ﬁeld components for the uniform-density distribution are Esx =

I I x y , and Esy = πε0 v(rx + ry ) rx πε0 v(rx + ry ) ry

(9.36)

297

298

9 Multiparticle Dynamics with Space Charge

where rx and ry are the semiaxes of the ellipse, related to the rms beam sizes by rx = 2ax , and ry = 2ay . Substituting Eq. (9.36) into Eq. (9.35) yields the rms envelope equations for a uniform density beam, which are given by ax + κx (s)ax −

2 εrx K − =0 a3x 2(ax + ay )

(9.37)

and ay + κy (s)ay −

2 εry

a3y

−

K =0 2(ax + ay )

(9.38)

The quantity K is called the generalized perveance given by K = qI/2πε0 mγ 3 v3 , where I = qN v is the current, expressed in terms of the number of particles per unit length N . These equations were ﬁrst derived by Kapchinsky and Vladimirsky [17] for a stationary uniform beam in a quadrupole-focusing channel, and are known as the K–V envelope equations. However, the remarkable result found by Lapostolle and Sacherer is that Eqs. (9.37) and (9.38) are valid not only for uniform density beams, but for all density distributions with elliptical symmetry. Thus the form of the envelope equation is independent of the density proﬁle of the beam. To calculate the rms beam trajectories, even in the presence of space-charge forces, we can replace the actual beam distribution, which may not be known in advance, with an equivalent uniform beam [18] having the same current and the same second moments as the real beam. It is convenient to work with an equivalent uniform beam because, as we have seen, the space-charge ﬁeld of a uniform beam, with elliptical or ellipsoidal symmetry, is easily calculated and is linear. The ratio of the space-charge to the emittance terms in Eqs. (9.37) and (9.38) can be used to determine when space-charge is important compared with emittance in determining the rms beam size. For a round beam, where a = ax = ay , we have an emittance-dominated beam when Ka2 /4ε2 1, and a space-charge dominated beam when Ka2 /4ε2 1. A space-chargedominated beam may be compared with cold plasma, where collective effects are dominant, whereas an emittance-dominated beam is dominated by random or thermal effects. Accelerators that are characterized as high-current machines can often be designed to avoid the space-charge-dominated regime by increasing the focusing force to reduce the rms beam size. √ For a round beam in an ideal uniform focusing channel, where k0 ≡ κx = √ κy is the zero-current phase-advance per unit length, a matched solution of Eqs. (9.37) and (9.38) can be obtained corresponding to ax = ay = 0 [19]. The phase advance per unit length including space charge for the equivalent uniform beam, is denoted by k, where k2 = k20 − K/4a2 . The matched beam size is given by a2 = ε/k, so that k plays the role of the inverse of the Courant–Snyder β function. The tune depression ratio is 1 k = √ k0 u + 1 + u2

(9.39)

9.9 Three-Dimensional Ellipsoidal Bunched Beams

where u = K/8εk0 is a space-charge parameter. An emittance-dominated beam corresponds to u 1, so that k/k0 ∼ = 1 − u, and the space-charge-dominated beam corresponds to k/k0 ∼ = 1/2u. The equivalent uniform beam for an arbitrary charge distribution can also be used to characterize an effective overall focusing strength in the presence of space charge. This can be done in a periodic-focusing system by comparing the phase advance per focusing period including the space charge, σ , for an equivalent uniform beam, with the phase advance per focusing period with zero space charge, σ0 . For a uniform density the quantities σ and σ0 can be obtained from the transfer matrix through one period, with and without space charge. The procedure assumes the beam is matched, so that the beam ellipse is geometrically similar to the trajectory ellipses, including the linear spacecharge ﬁeld. The ratio σ/σ0 , which is always in the range 0 ≤ σ/σ0 ≤ 1, is called the tune depression or the tune-depression ratio. When σ/σ0 ≈ 1, the beam is emittance dominated, whereas when σ/σ0 ≈ 0, the beam is space-charge dominated. 9.9 Three-Dimensional Ellipsoidal Bunched Beams

The electric-ﬁeld components for the three-dimensional uniform ellipsoid are of more interest for describing a typical linac bunch where the three semi-axes are comparable in length. The results are given by [20] Esx =

3Iλ(1 − f ) x 3Iλ(1 − f ) y , Esy = , 4πε0 c(rx + ry )rz rx 4πε0 c(rx + ry )rz r y

Esz =

3Iλf z 4πε0 crx ry r z

(9.40)

where rx , ry , and rz are the semiaxes of the ellipsoid in the laboratory frame, for N particles per bunch I = qNc/λ is the average current over an RF period, and λ is the √RF wavelength. The semiaxes ri are related to the rms beam sizes ai by ri = 5ai , i = x, y, z. The displacements x, y, and z are to be evaluated relative to the centroid of the bunch. The quantity f is an ellipsoid form factor √ and is a function of the parameter p = γ rz / rx ry . Values of f (p) for p < 1 and f (1/p) for p > 1 are obtained from Fig. 9.4. For a nearly spherical bunch where 0.8 < p < 5, a useful approximate form is f = 1/3p. Substituting the ﬁeld expressions into Eq. (9.35), the rms envelope equations are ax + κx (s)ax − ay + κy (s)ay −

2 εr,x 3K3 (1 − f ) − =0 3 ax (ax + ay )az 2 εr,y

a3y

−

3K3 (1 − f ) =0 (ax + ay )az

(9.41) (9.42)

299

9 Multiparticle Dynamics with Space Charge 1 0.9 0.8 0.7

f (p) 0.6

300

0.5 0.4 0.3 0.2

f (1/p)

0.1 0

0.2

0.4

0.6 0.8 p (OR 1/p)

Figure 9.4 Ellipsoid form factor f (p) versus p for p < 1 and f (1/p) for p > 1.

and az + κz (s)az −

2 εr,z 3K3 f − = 0. a3z ax ay

(9.43)

We have deﬁned a three-dimensional space-charge parameter K3 =

√

qIλ

20 5πε0 mc3 γ 3 β 2

(9.44)

Unlike the two-dimensional elliptical geometry discussed in Section 9.8, the space-charge terms in Eqs. (9.41) to (9.43) depend on the density proﬁle, even for different distributions with ellipsoidal symmetry. However the distribution dependence is very weak, and Eqs. (9.41) to (9.43) are still a good approximation for any distribution with ellipsoidal symmetry.

9.10 Beam Dynamics Including Linear Space-Charge Field

A longitudinal equation of motion can be written, assuming that the beam density is described by a uniform three-dimensional ellipsoid with linear space-charge-force components. We begin with the equation of longitudinal

9.10 Beam Dynamics Including Linear Space-Charge Field

motion, derived in Chapter 6, and add a longitudinal space-charge ﬁeld from Eq. (9.40). The single-particle equation of motion for the phase is

(φ − φs )2 3qIλf (φ − φs ) 2 =0 (9.45) φ + k0 (φ − φs ) − − 2 tan(−φs ) 4πε0 γ 3 β 2 mc2 rx ry rz where φs is assumed to be constant√ and the semiaxes of the uniform ellipsoid are related to the rms sizes by ri = 5ai , i = x, y, z. We deﬁne µ as the ratio of the longitudinal space-charge force to the linear part of the longitudinal focusing force. Then, Eq. (9.45) can be written as

(φ − φs )2 2 =0 (9.46) φ + k0 (1 − µ )(φ − φs ) − 2 tan(−φs ) where µ ≡

3qIλf 4πε0 γ 3 β 2 mc3 rx ry rz k20

(9.47)

If we deﬁne the phase advance per unit length, including space charge as k2 = k20 (1 − µ ), where k0 is given by Eq. (6.43), the linear equation of longitudinal motion, ignoring the quadratic term of Eq. (9.46) is φ + k2 (φ − φs ) = 0

(9.48)

A common ﬁgure of merit for evaluating the importance of the space-charge force is the longitudinal tune-depression ratio, deﬁned as k /k0 . This is 2 3qIλf k = 1 − µ = 1 − (9.49) 3 k0 4πε0 γ β 2 mc3 rx ry rz k20 An approximate transverse equation of motion can also be derived, again assuming that the beam density is described by a uniform threedimensional ellipsoid. We use an approximate treatment based on the smooth approximation for a periodic quadrupole focusing array, introduced in Section 7.17. The equation of motion in the smooth approximation has the form qEsx d2 x σ0 2 + x− 3 2 2 =0 (9.50) 2 ds P γ β mc where σ0 is the zero-current transverse phase advance per focusing period P, and Esx is the x component of the space-charge electric ﬁeld, given in Eq. (9.40). Identifying σ as the phase-advance per focusing period including space charge, we obtain 2 2 σ 3qIλ(1 − f ) P =1− (9.51) 3 3 2 σ0 4πε0 mc γ β (rx + ry )rx rz σ0

301

302

9 Multiparticle Dynamics with Space Charge

9.11 Beam-Current Limits from Space Charge

Using the uniform three-dimensional ellipsoid model, an approximate expression for the longitudinal current limit can be derived [21]. As the beam current increases, the repulsive space-charge forces reduce the size and depth of the longitudinal potential well that deﬁnes the bucket. For a given synchronous phase and a given accelerating ﬁeld, there will be some value of the current at which the space-charge force is just equal to the focusing force and the potential well vanishes. This limiting current deﬁnes the longitudinal space-charge limit, or the longitudinal current limit. If we return to the deﬁnition for µ given in Eq. (9.47), substitute k0 from Eq. (6.43), and solve for the current, we obtain I = µ

8π 2 ε0 crx ry rz E0 T sin(−φs ) 3βλ2 f

(9.52)

A current limit can be obtained from Eq. (9.52) that is consistent with numerical simulation results. We use the approximation for the ellipsoid form factor f = 1/3p, where p is deﬁned in Section 9.9. We assume that rz = |φs |βλ/2π, which implies that the separatrix width is unaffected by space charge, and that only ±|φs | of the phase width is available. It is also convenient to introduce a ﬂutter factor for the alternating transverse-focusing system, deﬁned by ψ ≡ rmax /rmin . At the point in each transverse focusing period, where rx is maximum and ry is minimum, we write ψ ≡ rx /ry = rx2 /rx ry . The longitudinal current limit is largest, when rx = a, where a is the aperture radius. Substituting all these into Eq. (9.52), we obtain an expression for the longitudinal current limit, which is generally in good agreement with numerical simulation results, I,max =

2µ βγ aE0 T sin(−φs )|φs |2 Z0 ψ 1/2

(9.53)

where Z0 = 1/ε0 c = 376.73 is the impedance of free space, and µ ≈ 1 for a space-charge dominated beam. Equation (9.53) implies that the longitudinal current limit increases with increasing velocity, increasing beam aperture, increasing accelerating ﬁeld, and for small angles where sin(−φs ) ≈ −φs , it increases approximately as the cube of the synchronous phase. Next we develop an expression for the transverse current limit. Physically, as the beam current increases, the matched beam size grows until it is equal to the aperture radius, which corresponds to the current limit. Any further increase of the current results in beam loss. Returning to Eq. (9.50), we deﬁne the ratio of the space-charge force to the focusing force as µT , and solving for the current, we obtain for the x motion I = µT

4πε0 γ 3 β 2 mc3 (rx + ry )rx rz σ0 2 qλ(1 − f ) P

(9.54)

9.12 Overview of Emittance Growth from Space Charge

Again, we use the approximation for the ellipsoid form factor f ∼ = 1/3p, we let rz = |φs |βλ/2π, and we introduce the ﬂutter factor ψ ≡ rx /ry = rx2 /rx ry . The transverse current limit will be largest, when rx = a. We write the transverse focusing period as P = Nβλ, and assume that we can equate the arithmetic and geometric mean of the beam size over a period. Substituting these relations into Eq. (9.54), we obtain an expression for the transverse current limit, IT,max = µT

4mc2 γ 3 β|φs |σ02 a 2 3Z0 qψN 2 (1 − f ) λ

(9.55)

In practice, the phase advance σ0 is limited to σ0 < 90◦ to avoid envelope instability, [22] which occurs for high-intensity beams in quadrupole channels. A useful rule of thumb is that if σ0 > 90◦ and the space-charge force reduces σ so that σ < 90◦ , the beam will be unstable. In some cases the value of σ0 may be limited instead by the quadrupole focusing strength (i.e., corresponding to the Gl product of the quadrupoles). Then, it is more convenient to substitute the expression for σ0 as a function of the quadrupole parameters. Using the results from Section 7.17, the zero-current phase advance per focusing period in a FODO channel in the smooth approximation, including both the quadrupole and the RF defocusing term, is

qGP 2 πqE0 T sin(φ)P2 σ02 ∼ + (9.56) = 2mcγβ mc2 λ(γβ)3 Expressing the focusing period as P = Nβλ, where N is an integer, and substituting Eq. (9.56) into Eq .(9.55), we obtain

4mc2 γ 3 β|φs |a2 qG 2 πqE0 T sin(−φs ) (9.57) IT,max = µT − 3Z0 qψ(1 − f ) 2mcγ mc2 γ 3 βλ For low-velocity beams the RF defocusing term should not be neglected. The maximum ratio of the transverse space-charge force to the focusing force is µT ≈ 1, which occurs for a space-charge-dominated beam. Typically in a proton linac, the smallest values of both the longitudinal and transverse current limits are found at the end of the gentle buncher section of the RFQ. The current limit formulas can be used as a basis for optimized RFQ-linac designs, by choosing the parameters to yield equal values of the longitudinal and transverse limits, which are chosen to be larger than the design current including a safety margin for errors.

9.12 Overview of Emittance Growth from Space Charge

Beams that are in equilibrium in the focusing channel of a linac experience no emittance growth. Unfortunately, beams observed in linac numerical

303

304

9 Multiparticle Dynamics with Space Charge

simulations are rarely in equilibrium, and when they appear to be near equilibrium, any changes that occur in the focusing system produce changes in the beam, usually accompanied by emittance growth. Nonlinear forces that act on a nonequilibrium beam will cause the rms emittance to increase. The space-charge force in high-current beams is typically the major cause of such emittance growth. Four different space-charge mechanisms can be identiﬁed, [23] and are described brieﬂy in this section. First, when a highcurrent, rms-matched beam is injected into the accelerator, the emittance can grow very rapidly as the charges redistribute to provide shielding of the external focusing ﬁeld. This mechanism, called charge redistribution, is the fastest known emittance-growth mechanism, producing growth in only one quarter plasma period, where the beam plasma frequency for a beam with particle density n is ωP = nq2 /ε0 m. The free energy for the emittance growth comes from the ﬁeld energy of the initial distribution [21, 24]. Second, if the injected beam is not rms matched, additional energy from the mismatch oscillations of the beam is available for emittance growth. Even for relatively small mismatches, this mechanism can become the largest contributor to emittance growth. Third, for nonsymmetric or anisotropic beams, there can be emittance transfer as a result of space-charge resonances that couple longitudinal and transverse oscillations, where in some cases the kinetic energies in the three planes may approach an approximate equalization, sometimes referred to as equipartitioning. In such cases, the emittance grows in a plane that receives energy, and decreases in a plane that looses energy. Finally, the periodicfocusing structure can resonantly excite density oscillations in the beam, the most serious of which is the envelope instability mentioned in Section 9.11. To avoid these resonances in real beams, it is sufﬁcient to keep σ0 < 90◦ . This is a more stringent requirement than for the stability of a zero-current beam in a periodic-focusing channel, for which the stability requirement is σ0 < 180◦ . The four mechanisms and some of their characteristics are summarized in Table 9.1. Numerical simulations of nonequilibrium linac beams, show that linac beams do evolve, although sometimes slowly, to quasiequilibrium distributions. When the focusing is a linear function of the displacement, but nonlinear space-charge forces are present, the equilibrium spatial distribution consists of a uniform central-density core, and at the edges the density falls to zero over a distance approximately equal to a Debye length [25]. For space-charge dominated beams, the Debye length is small compared with the beam size, and the distribution consists almost entirely of the uniform core, which falls sharply to zero at the edge. For emittance-dominated beams, the Debye length is large compared with the beam size, and the distribution is dominated by a broad Debye edge. The spatial density then has the appearance of a Gaussianlike distribution. It is also observed that the space-charge-induced emittance growth in nonequilibrium beams is often associated with the formation of an outer halo of low-density particles in phase space. The phase-space density of the core is reduced only slightly in the process, but the associated rms

For minimum growth

Distribution function sensitivity

Approximate timescale

Free- energy source

Avoid transitions toward stronger tune depression RMS match

Weakly dependent

Typically ≥10τplasma

≈

τplasma 4 Strongly dependent

Oscillation energy of excited mode

RMS mismatch

Nonlinear ﬁeld energy

Charge redistribution

Table 9.1 Properties of space-charge-induced emittance growth.

Avoid space-charge coupling resonances

Weakly dependent

Typically ≥10τplasma

Space-charge coupling resonances

Emittance transfer

Keep σ0 < 90◦

Strongly dependent

≈2τbetatron

Longitudinal energy

Structure resonance 9.12 Overview of Emittance Growth from Space Charge 305

306

9 Multiparticle Dynamics with Space Charge

emittance growth caused by the halo can be signiﬁcant. The halo is especially undesirable for high-duty-factor linacs, because it can result in particle loss in the accelerating structure, and radioactivation of the linac.

9.13 Emittance Growth for rms Matched Beams

First, we distinguish between internal matching and rms matching to a periodic-focusing structure. Internal matching corresponds to particle phasespace trajectories that coincide with the isodensity contours of the 6D phase-space density. For an internally matched beam, the distribution will be in equilibrium in the accelerator channel, and no emittance growth will occur, even though nonlinear forces may act on the beam. Such an equilibrium distribution has a constant beam radius if the focusing is uniform along the accelerator, or for a periodic focusing channel has a periodic radial proﬁle with the same period as the focusing channel. Examples of equilibrium distributions have been studied for beam transport channels with linear focusing forces: linear functions of the transverse beam displacement [26]. The most frequently studied is the Kapchinsky–Vladimirsky (K–V) distribution for transverse dynamics in 4D phase space, which is the only equilibrium distribution in a periodic transport channel for which an analytic description has been found [27]. For the K–V distribution the beam is distributed on the surface of a hyperellipsoid in four-dimensional phase space. No particles populate the central core of this 4D space, which is not the case in real beams. Nevertheless, the K–V distribution results in uniform ellipses for all 2D projections, which is convenient since it produces simple linear space-charge forces in the spatial projections. Given a beam that is not internally matched to a periodic structure, one may wish to transform it into an internally matched equilibrium distribution. We do not know any simple procedure for making such a transformation, and whether in principle such a transformation is possible without increasing the rms emittance in the process. Nevertheless, we do know how to match the rms beam sizes for each degree of freedom. This is accomplished by providing a beam-optics transformation to eliminate the mismatch oscillations of the rms beam sizes. In a uniform focusing channel the resulting rms sizes will be constant; in a periodic channel the rms sizes will undergo a periodic ﬂutter with the same period as that of the focusing lattice. An rms-matched beam is not generally internally matched, so the rms-matched beam is not generally in equilibrium. Therefore, the rms-matched beam can still relax to an equilibrium state with the possibility of irreversible emittance growth. Even so, injection of an rms-matched beam can be considered a ﬁrst approximation to the desired internally matched beam. Numerical simulations of nonequilibrium beams in uniform focusing channels show that such beams often evolve to quasiequilibrium distributions,

9.13 Emittance Growth for rms Matched Beams

which change only slowly as the beam is accelerated. The evolution of such beams is usually accompanied by rms-emittance change as a result of both nonlinearity and coupling between degrees of freedom, and the velocity distributions are Maxwellian-like. When the focusing is linear, the spatial distribution of a space-charge dominated beam has an approximately uniform density core of density n. The density decreases to zero at the edges of the beam over a distance approximately equal to the Debye length, λD , given nonrelativistically by λD =

ε0 kB T/nq2

(9.58)

where q is the charge per particle and ε0 is the free-space permittivity. In Eq. (9.58) the thermal energy is given by kB T = mc2 εn2 /a2 where mc2 is the particle rest energy, a is the rms beam size, and εn is the rms-normalized emittance, deﬁned without a factor of 4, but including the relativistic βγ factor. For space-charge dominated beams λD a, so the equilibrium spatial distribution is approximately uniform with a sharp falloff at the edges. For emittance dominated beams λD a and the Debye tail is dominant, resulting in a peaked Gaussian-like charge density. The K–V distribution is an exception because it always has uniform charge density, regardless of the relative importance of emittance and space charge. However, this distribution does not correspond to the ﬁnal equilibrium state of real beams. Emittance growth can occur in beams that are internally mismatched even if they are rms matched, through a mechanism referred to as charge redistribution. When a high-current rms-matched beam is injected into an accelerator, the charges redistribute rapidly to shield the interior of the beam from the linear external focusing force. Transverse energy conservation is valid in the smooth approximation, and the kinetic energy associated with the emittance growth comes from the difference of the ﬁeld energy of the initial distribution and that of the ﬁnal distribution. For linear external focusing ﬁelds in the extreme space-charge (zero emittance) limit, this implies a charge rearrangement to a uniform density to provide exact shielding. Finite emittance rms-matched beams in numerical simulations evolve toward an internally matched charge density with a central uniform core and a ﬁnite thickness boundary, whose width is approximately equal to the Debye length. The rms-emittance growth results from the nonlinear space-charge ﬁelds that are present while the beam has nonuniform density and is undergoing initial plasma oscillations. The emittance growth can also be described as the result of the decoherence of the plasma oscillations of particles with different amplitudes, which causes a phase mixing. In either case this is the fastest-known emittance-growth mechanism, producing full emittance growth in only one-quarter plasma period, where the plasma frequency of the beam with particle mass m and charge q, and with particle density n, is ωp = nq2 /ε0 m. The distance a beam particle travels

307

308

9 Multiparticle Dynamics with Space Charge

during one plasma period is given by 8π 2 a2 λp = Kγ 2

(9.59)

The initial growth in one-quarter plasma period is followed by damped oscillations for typically 10 or so plasma periods. This mechanism can be important when a beam that is internally matched to a strong-focusing emittance-dominated channel is injected after rms matching into a weakerfocusing space-charge-dominated channel. For ﬁxed rms beam size, the minimum space-charge ﬁeld energy corresponds to a uniform-density beam. If the rms-matched input beam has a peaked spatial proﬁle, characteristic of a channel where the space-charge force is small compared with the focusing force, the beam particles in the weaker-focusing channel will rapidly rearrange to produce a more uniform density, and in the process space-charge ﬁeld energy is converted to kinetic energy. If the ﬁeld energy difference is known, it can be used to calculate the emittance growth. To describe these effects more quantitatively, we introduce the concept of nonlinear ﬁeld energy U, a quantity that depends on the shape of the charge distribution and corresponds to the additional space-charge ﬁeld energy of beams with nonuniform charge density. The rate of change of unnormalized emittance ε is given by the expression [28] d U dε2 = −2a2 K (9.60) dz dz w0 where a is the rms beam size and K is the generalized perveance, which is proportional to the beam current K≡

qI 2πε0 mv3 γ 3

(9.61)

The nonlinear ﬁeld energy is deﬁned as U = W − Wu , where W is the self-electric-ﬁeld energy per unit length of the distribution, and Wu is the self-electric-ﬁeld energy per unit length of an equivalent uniform beam, which has the same current, emittance, and rms size as the actual beam. The quantity Wu is the ﬁeld energy per unit length contained within a radius b, which is equal to or larger than the beam radius, and is given by Wu = w0 (1 + 4 ln(b/2a)), b ≥ 2a

(9.62)

where w0 is the self-electric-ﬁeld energy per unit length within the boundary of an equivalent uniform beam. The electric ﬁeld and the ﬁeld energy outside the beam are independent of the charge density, and the contribution to the difference between W and Wu outside the beam is zero. The special role of the uniform distribution can be explained by its associated linear self-force,

9.13 Emittance Growth for rms Matched Beams

which causes no rms emittance growth. Thus, U is the residual self-electricﬁeld energy possessed by beams with nonuniform charge densities. Because nonuniform beams have nonlinear self ﬁelds, we call U the nonlinear ﬁeld energy. Equation (9.60) tells us that a decrease in nonlinear ﬁeld energy U corresponds to an increase in rms emittance. Both the electric and the magnetic ﬁeld contributions are contained in the equation by including the factor γ 3 in the deﬁnition of K (γ 2 accounts for the self-magnetic ﬁeld and γ accounts for the relativistic mass). RMS emittance growth induced by space charge is inherently a nonrelativistic effect; it is most important when γ is near unity. The quantity U/w0 , which affects the emittance growth, is dimensionless. It is zero for a uniform charge distribution, is positive for both peaked and hollow distributions, and increases as the distribution becomes more nonuniform. It is independent of the beam current and the rms beam size, and is a function only of the shape of the distribution. Table 9.2 shows U/w0 for some common unbunched beam distributions. If the initial distribution is known, the initial value of U/w0 can be calculated. In principle Eq. (9.60) could be solved to obtain the emittance growth if we knew the ﬁnal U/w0 . However, the ﬁnal distribution is usually not known, so an assumption of some kind is needed. To proceed further, Eq. (9.60) can be integrated if we assume the beam size of the rms-matched beam is approximately constant. Thus,

and

ε2 = −2a2 K(U/w0 )

(9.63)

ε2 2Ka2 U = 1 − 2 2 w0 εi εi

(9.64)

Next we rewrite the quantity Ka2 /εi2 in a more convenient way in terms of the space-charge tune depression ratio of an equivalent uniform beam as follows. The single-particle equation of motion for an equivalent uniform beam with a linear space-charge term is x + k20 x −

Kx =0 4a2

(9.65)

Table 9.2 U/w0 for some common unbunched beam distributions. Spatial Distribution Gaussian Waterbag Uniform Hollow (n = 2) Hollow (n = 10)

Charge Density ρ(r) exp (−r 2 /2a2 ) 1 − (r/R)2 1 r2 r 10

r r r r

≤R ≤R ≤R ≤R

U/w0 0.154 0.0224 0.0 0.0754 0.245

309

310

9 Multiparticle Dynamics with Space Charge

from which we identify the phase advance per unit length, including spacecharge, as k2 = k20 −

K 4a2

(9.66)

The matched envelope equation for a uniform focusing channel may be written as a = 0, or k20 a −

ε2 K − =0 4a 16a3

(9.67)

From these two equations we obtain k2 =

ε2 16a3

(9.68)

Equation (9.67) can be rewritten as k20 4Ka2 k20 = −1 − 1 = ε2 ε2 /(16a4 ) k2 and therefore Eq. (9.64) can be written as 2 ε2 k0 U − Ui = 1 − − 1 k2 2w0 εi2

(9.69)

(9.70)

If we knew the tune depression and the initial U/w0 , we could calculate the rms emittance growth if the ﬁnal value of U/w0 is known, but this would require knowledge of the ﬁnal distribution. There is one case, where we do know U/w0 . In the extreme space-charge limit we expect the beam to evolve toward a ﬁnal stationary state with a uniform charge density to obtain complete shielding of the linear applied focusing force. Let us assume that for all tune depression ratios, U/w0 = 0 in the ﬁnal state, which corresponds to assuming a ﬁnal uniform beam. This approximation should be very good for space-charge dominated beams, where emittance growth is largest. In the emittance-dominated limit, where the approximation would lead to a larger fractional error in emittance growth, the emittance growth is small anyway. In all cases the approximation will result in an upper limit to the emittance growth. With this approximation the upper limit to the ﬁnal emittance growth for an rms-matched beam is 2 ε2 k0 Ui (9.71) = 1 − − 1 k2 2w0 εi2 Equation (9.71) can also be written as εf2 = εi2 +

1 2

K k0

Ui w0

(9.72)

9.13 Emittance Growth for rms Matched Beams

This result tells us that as the initial emittance approaches zero, the ﬁnal emittance approaches a minimum value (the second term) that decreases with increased focusing strength (larger k0 ), increases with perveance K, and increases with the initial nonuniformity as measured by Ui /w0 . Figure 9.5 shows results of a numerical simulation for 10 plasma periods using an initial Gaussian charge density truncated at two standard deviations, and an initial space-charge tune depression of 0.02 (a space-charge-dominated beam). The abscissa is the distance along the beam line in plasma periods. The quantity ρ is a dimensionless radial distribution parameter deﬁned as ρ = 1 − r¯ /¯ru where r¯ is the average radius and r¯u is the average radius of an equivalent (same second moments) uniform beam. For common distributions the parameter ρ is positive for a peaked charge density, zero for a uniform density, and negative for a hollow beam. The parameter ρ in Fig. 9.5a indicates that the beam density undergoes damped radial plasma oscillations between peaked and hollow conﬁgurations. The quantity Ui /w0 in Fig. 9.5b is maximum at both the extreme peaked and hollow conﬁgurations and therefore oscillates at twice the frequency of the parameter ρ. Its minimum value is not exactly zero, expected for a uniform distribution partly because of the formation of some beam halo. The rms emittance-growth ratio, ε/εi in Fig. 9.5c shows the rapid initial growth during the ﬁrst quarter plasma oscillation followed by damped oscillations. By 10 plasma periods the beam has evolved to an approximately stationary state. A second example numerical example is shown in Fig. 9.6 using an initial semi-Gaussian charge density beam (uniform spatially and Gaussian in velocity space) truncated at four standard deviations in velocity space with an initial space-charge tune depression of 0.25 (a moderately space-charge-dominated beam). This example is interesting because the rms emittance decreases. The parameter ρ in Fig. 9.6a rises from its initial value of zero corresponding to uniform density and remains positive throughout indicating a peaked distribution. The radial density proﬁle (not shown) evolves from an initial hard-edged beam to one with a Debye tail. The quantity Ui /w0 in Fig. 9.6b increases while the rms emittance-growth ratio ε/εi in Fig. 9.6c decreases in accordance with Eq. (9.71). Figure 9.7 shows the emittance growth after 100 plasma periods versus the space-charge tune-depression ratio for an initial truncated Gaussian beam density. The agreement of the curve from Eq. (9.71) and the symbols from the second moments taken from the simulations is excellent. Figure 9.7 shows that if one tries to operate at very low tune depression ratio, there can be a risk of very large emittance growth. One might expect that the way to minimize the emittance growth is simply to produce a uniform density beam. However, this may be a little too simple. It may not be easy to maintain the uniformity because the charge redistribution process is so rapid. For moderate or weak tune depressions, a uniform beam will rapidly grow a Debye tail. After that happens, one needs to avoid introducing transitions that result in stronger tune depression, because that will surely lead to ﬁeld energy conversion to

311

9 Multiparticle Dynamics with Space Charge

0.05 0.00

(a) −0.05

U /w0

0.050 0.025

(b) 0.000 9.0 e/ei

312

5.0

(c)

1.0 0

5 z /lp

Figure 9.5 Numerical simulation results showing the evolution of three beam parameters for an initial Gaussian beam truncated at two standard deviations for a space-charge tune-depression ratio of k/k0 = 0.02 versus distance along the beam line in plasma periods. The lower curve

showing the emittance growth is actually two overlapping curves showing very good agreement, one from Eq. (9.71) and one from the second moments taken from the simulation.

kinetic energy accompanied by emittance growth. As a consequence of the risk of converting nonlinear ﬁeld energy into thermal energy accompanied by large emittance growth, control of emittance growth in the space-charge dominated regime is a signiﬁcant challenge. An experimental test of the theory was carried out at University of Maryland using a 5-keV electron beam with two solenoids used for rms beam matching followed by a 5-m long periodic channel with 36 solenoid lenses [29–31]. In the experiment the initial beam was highly nonuniform, consisting of ﬁve separate beamlets created using an aperture plate with ﬁve small holes. The beam was then rms matched, and the beam proﬁle was measured at six locations along the beam transport channel. The free energy was entirely due to the nonuniformity of the distribution. The relevant beam parameters were U/w0 = 0.2656, and initial tune depression ratio ki /k0 = σi /σ0 = 0.31. The ﬁnal emittance growth from theory was εf /εi = 1.56. The ﬁnal beam observed on a ﬂuorescent screen was hom*ogenized and showed no evidence for the initial ﬁve-beam beamlet structure. The emittance measurements using a slit-pinhole system showed good agreement with the theory.

9.13 Emittance Growth for rms Matched Beams

0.02 0.00

(a) −0.02

U /w0

0.02 0.01

(b)

0.00

e /ei

1.00

(c)

0.95 0

5 z /lp

ef /ei

Figure 9.6 Numerical simulation results showing the evolution of three beam parameters for an initial semi-Gaussian beam (uniform spatially and Gaussian in velocity space) truncated at four standard deviations for a space-charge tune-depression ratio of k/k0 = 0.25 versus

14 13 12 11 10 9 8 7 6 5 4 3 2 1

distance along the beam line in plasma periods. The lower curve showing the emittance growth is actually two overlapping curves showing very good agreement, one from Eq. (9.71) and one from the second moments taken from the simulation.

+ + + 0

0.2

+ 0.4

+ 0

k / k0

+ 0.8

Figure 9.7 Final emittance growth ratio versus initial space-charge tune depression ratio for an initial Gaussian beam truncated at four standard deviations. The curve is from Eq. (9.71) and the symbols are from the simulations after 100 plasma periods.

313

314

9 Multiparticle Dynamics with Space Charge

9.14 Model of Space-Charge-Induced Emittance Growth in a Linac

In this section we construct a nonrelativistic model of transverse emittance growth [32] for the charge-redistribution process, from which the scaling of emittance growth with respect to the beam parameters in a linac can be deduced. As a model for a long bunch, we consider a continuous beam with circular cross section, velocity v, current I, and zero initial emittance. As the beam propagates in a drift space, the space-charge force will change the transverse momentum components of each particle by an amount equal to the product of the space-charge force times the time over which the force acts. To simplify the calculation, we ignore the changes in the transverse positions, so that the spatial density is assumed to remain unchanged as the beam propagates. For simplicity we assume a parabolic density, which will contribute nonlinear space-charge ﬁelds that can produce emittance growth. The density distribution as a function of radius r is

r2 2qNl 1− 2 (9.73) ρ(r) = πR2 R where Nl is the number of particles per unit length and R is the beam radius. From Gauss’s Law, this corresponds to a nonlinear space-charge ﬁeld given by

qNl r3 Er (r) = r− 2 (9.74) πε0 R2 2R Over the drift length z, this ﬁeld delivers a radial momentum impulse to each particle given by pr (z) = qEr z/v. Suppose the initial phase-space distribution is represented by a straight line, corresponding to zero transverse emittance. As described in Section 9.2, the nonlinear ﬁeld changes an initial straight-line distribution into a line with curvature accompanied by rms emittance growth. Using the deﬁnition of rms emittance, it can be shown that the unnormalized rms emittance in x or y as a function of the drift distance z is ε(z) =

Kz √ 12 5

(9.75)

where K is the generalized perveance K, given nonrelativistically by K = qI/2πε0 mv3 . It may seem surprising that the emittance is independent of the radius R. This is because as R increases, although the momentum impulses from space charge are smaller, they extend over a larger radial extent. Because emittance is proportional to phase space area, these two effects cancel. When the injected beam is traced with numerical simulation codes for a longer distance, sufﬁcient to allow the particle positions to adjust, one ﬁnds that the beam approaches a steady-state distribution, and that most of the emittance growth occurs in roughly one quarter of a beam plasma period. To approximate the emittance growth as a function of position, suppose that

9.14 Model of Space-Charge-Induced Emittance Growth in a Linac

initially the emittance is assumed to grow linearly, at the rate given by√Eq. (9.75), until saturation occurs at the distance z = λp /4, where λp = 2πR/ 3K is the beam plasma period. Substituting the saturation distance z into Eq. (9.75), the saturated unnormalized rms emittance becomes Kλp πR K ε= √ = (9.76) 24 15 48 5 In contrast to the result of Eq. (9.75) in which the emittance growth rate was independent of R, Eq. (9.76) shows another surprising result that the saturated emittance increases with beam radius R. This result is a consequence of the proportionality of the plasma period to beam radius: for larger R the model predicts that the beam takes longer to relax to a steady state, and during the relaxation process, the emittance continues to grow. Equation (9.76) implies that to minimize rms emittance growth from space-charge forces, the linac should provide strong focusing to make the beam small. This result implies that to control transverse emittance growth, one should not design a highcurrent, high-brightness linac with weak transverse focusing to produce a large-radius beam, as one might be motivated to do to reduce the transverse space-charge force. Rather, one needs to increase the ratio of the focusing force to the space-charge force, and this requirement produces a reduced beam radius. Indeed, it can be shown that for a round beam in a uniform linear focusing channel, although weaker focusing increases the beam size and decreases the space-charge force, it also decreases the focusing force, such that the ratio of the space-charge force to the focusing force increases (see Problem 9.14). So far we have not expressed the result of Eq. (9.76) in a way that is suitable for a bunched beam in a linac. To do this, we write the peak current in terms of the number of particles per unit length Nl as I = qNl v. Furthermore, we assume that the linac accelerates a bunch train with bunch frequency f and a number of particles per bunch N. If we assume that a bunch can be represented by a cylinder of length 2b with a uniform longitudinal proﬁle, the peak bunch current I is expressed in terms of the average current I¯ = qNf ¯ over the bunch train, as I = Iv/2bf . Converting the unnormalized emittance of Eq. (9.76) to a normalized transverse rms emittance, and substituting the expression for K and for the current, we ﬁnd a nonrelativistic result for the saturated normalized emittance of an initially laminar beam, given by R πqI¯ (9.77) εn = εβ = 48 15ε0 mc2 bf Although the magnitude of the expression in Eq. (9.77) is only approximate, it predicts a dependence on the beam parameters that is in good agreement with numerical simulation. To minimize space-charge-induced transverseemittance growth for ﬁxed average current, one needs to design the linac with strong focusing to produce a small beam radius R, with high frequency f to

315

316

9 Multiparticle Dynamics with Space Charge

distribute the charge over more bunches, and to avoid focusing the beam too much longitudinally. 9.15 Emittance Growth for rms Mismatched Beams

A mismatched rms beam envelope oscillates with a period that depends on which envelope mode is excited. There are two linearly independent envelope modes, the antisymmetric mode where the x and y oscillations are out of phase, [33] whose wave number is given by ke1 = k0 + 3k2 (9.78) and the symmetric mode where they are in phase whose wave number is given by (9.79) ke2 = 2k20 + 2k2 For an rms-mismatched beam, excess potential energy associated with the mismatch oscillations is available for emittance growth. Emittance growth will occur when the beam, under the inﬂuence of nonlinear space-charge forces, relaxes toward an equilibrium state. For a uniform continuous beam with linear external focusing, Reiser has derived an equation for emittance growth of an initially rms-mismatched beam, assuming that the free energy of the initial oscillation is completely converted into thermal energy of the ﬁnal matched beam [34]. Numerical simulation studies [35] show that the emittance growth is associated with formation of beam halo surrounding the beam core. These studies suggest that rms mismatch may be the source of most beam halo in high-current beams. To calculate the emittance growth from rms mismatch, Reiser deﬁnes a free-energy parameter h as 1 k2 a2m 1 k2 am a2mm h= − 1 − (9.80) + 1 − ln 1 − 2 k20 a2mm 2 a2m amm k20 where amm is the initial mismatched beam radius, am is the initial beam radius if the beam was matched, and k/k0 is the initial space-charge tune depression ratio. It is assumed that the two rms beam radii are deﬁned at the location where the beam ellipse is upright. The three terms in Eq. (9.80) represent the contributions of the kinetic energy, the potential energy associated with the external focusing force, and the space-charge ﬁeld energy, respectively. After calculating h, the ratio af /am of ﬁnal beam radius to the initial matched beam radius can be calculated by solving numerically the transcendental equation a2f k2 af − 1 − =1+h (9.81) ln a2m am k20

9.15 Emittance Growth for rms Mismatched Beams

It is assumed that in the ﬁnal state the entire free energy from the mismatch oscillations is transferred to kinetic energy: there is no residual mismatch oscillation. Finally, the emittance growth may be calculated from εf af k2 a2f = 1 + 02 − 1 (9.82) εi am k a2m If residual mismatch oscillations are present in the ﬁnal beam state, the equation for emittance growth is an upper limit to the growth. Figure 9.8 shows the plots of emittance growth versus the free-energy parameter h, which varies between 0 and 1.0, as a function of space-charge tune depression, as calculated from Eq. (9.82). Figure 9.8 shows that for a given free energy, emittance growth from rms mismatch increases as the space-charge tune depression becomes stronger. Control of emittance growth in the space-charge-dominated regime is a signiﬁcant challenge, when the possibility of rms mismatch is present. In his book, Reiser presents the above material in a more general form which not only includes the rms mismatch effect, but also the nonlinear ﬁeld energy effect for an rms-matched beam, as well as the effect of an off-centered beam [36]. The same treatment can be used for all three effects, if the total free energy parameter h is the sum of contribution from all three effects. Thus, h = hs + hm + hc representing the contributions of nonlinear ﬁeld energy, mismatch, and off-center beams, respectively. Thus, h in Eq. (9.80) becomes hm , and Eq. (9.81) uses the total h value. Emittance growth for an off-center beam also requires some nonlinearity, since nonlinear space-charge does not affect the coherent oscillations. Otherwise the coherent beam oscillations for an off-center beam will continue indeﬁnitely and the free energy will not be thermalized. The predictions were tested experimentally at the University of Maryland from the same experiment in a 36-solenoid-lens channel that was discussed previously in Section 9.13 [29,30,37], where in this case the two solenoid matching lenses were set to produce an initial rms mismatched to matched beam-size ratio of amm /am = 0.5. The beam was highly nonuniform,

Emittance increase ef /ei

ki / k0 0.1

0.2

0.3 0.4

0.5

2 1.0 0

0.2

0.4

0.6

0.8

Figure 9.8 Emittance growth from rms mismatch versus the free-energy parameter h as the abscissa as a function of space-charge tune depression ki /k0 (courtesy of Martin Reiser).

317

318

9 Multiparticle Dynamics with Space Charge

consisting of ﬁve separate beamlets created using an aperture plate with ﬁve small holes. However, the main source of free energy was from the mismatch oscillations. The resulting free-energy parameter was h = 0.396, and the theoretical emittance growth ratio from mismatch was εf /εi = 3.72. The measured emittance growth value was εf /εi ∼ = 4.2.The experiment also showed the formation of beam halo which constituted the majority of the emittance growth. The explanation for the small discrepancy was attributed to greater free energy in the case of the asymmetry produced by the ﬁve beamlets.

9.16 Space-Charge Instabilities in RF Linacs from Periodic Focusing: Structure Resonances

An important mechanism that can lead to emittance growth in RF linacs is beam instabilities that occur in periodic focusing channels. The theoretical approach for the stability analysis is known as the linearized Vlasov theory in which perturbations in the form of electrostatic charge-density oscillation modes, described by their azimuthal symmetry such as quadrupole (n = 2), sextupole (n = 3), and so on, are added to known stationary solutions of the Vlasov equation. The theoretical objective is to determine whether the amplitudes for such modes grow or decay. For a beam transport channel comprising a periodic array of quadrupole or solenoid lenses, there is only one known stationary state available for such an analysis, which is the Kapchinsky–Vladimirsky (K–V) distribution. Consequently, the analysis is limited to the study of the K–V distribution. The stability problem for the K–V distribution was ﬁrst studied in 1970 for the case of a uniform or continuous-focusing channel by Gluckstern, [38] who found that for spacecharge tune-depression ratios above 0.4, all the oscillation modes are stable, whereas for tune depressions below 0.4 there existed some unstable modes. In simulation studies it was found that these unstable K–V oscillation modes led to charge-density redistribution with no emittance growth. In 1983, the stability analysis was carried out for a K–V distribution in a periodic-focusing channel, [39] where it was found that in the presence of periodic focusing some of the oscillation modes became unstable, driven by perturbations in the current distribution. The stability criterion depends on the zero-current phase advance per period σ0 and the space-charge tune depression ratio, σ/σ0 . The lowest order K–V modes, especially the quadrupole and sextupole modes, generate signiﬁcant emittance growth in simulations. The quadrupole mode is excited approximately when σ0 > 90◦ , and when σ lies within a stop band slightly under 90◦ , whereas the sextupole mode is excited when σ0 > 60◦ . For simulations with more realistic initial beam distributions, such as the Gaussian distribution, the emittance growth for the sextupole mode is not signiﬁcant, whereas the quadrupole mode still gives a large emittance growth.

9.17 Longitudinal-Transverse Coupling and Space-Charge Instabilities for Anisotropic Linac Beams

Thus, computer simulations for realistic beams predict that one can safely operate not only with σ0 < 60◦ , but also in the region with σ0 > 60◦ and σ0 < 90◦ . The region with σ0 > 90◦ is generally to be avoided. Next, we describe two experiments that tested the predictions of the theory. A single beam transport channel was constructed at Lawrence Berkeley National Laboratory using 82 electrostatic quadrupole lenses in a FODO conﬁguration, using a cesium beam, as part of the heavy-ion inertial-fusion program. Systematic experiments were conducted by Tiefenback and Keefe, [40] where the beam was matched in both transverse planes, and both σ0 and σ/σ0 were varied. The envelope instability predicted by K–V periodic-focusing beamtransport theory for a phase advance per period of σ0 > 90◦ led to major beam degradation with beam loss. No instability modes predicted by K–V theory, below the σ0 = 90◦ envelope instability were observed. Similarly, in a systematic experimental study carried out in a solenoid focusing lattice at University of Maryland, [41] the envelope instability was also observed with major beam loss. This was investigated systematically by varying σ/σ0 and changing σ0 from below to above 90◦ . Below 90◦ , no other instability predicted by the theory, including the third-order (sextupole) mode for σ0 > 60◦ , was found. The conclusion is that for real beams in periodic-focusing channels, the envelope instability predicted by theory for a phase advance per period of σ0 > 90◦ is the only instability of this theory that leads to emittance growth.

9.17 Longitudinal-Transverse Coupling and Space-Charge Instabilities for Anisotropic Linac Beams

Longitudinal-transverse coupling of the motion, and the behavior of anisotropic beams are subjects of particular importance for RF-linac beam dynamics. The transverse and longitudinal parameters of beams in RF linacs are usually different, which means that the beam bunches are anisotropic. Even without the space-charge forces, the transverse and longitudinal motions in linacs are coupled through nonlinear effects. The sources of the coupling are (1) the dependence of the transverse RF defocusing on the longitudinal phase, and (2) the dependence of the transit-time factor, which determines the longitudinal accelerating force, on the radial displacement. Weiss has calculated the resulting increase in the rms emittance of a bunched beam as it crosses an RF gap [42]. Furthermore, analyses of these nonlinear coupling effects on the transverse motion by Gluckstern [43] showed that substantial amplitude and emittance growth can occur, particularly at low velocities, from resonance when k = 2kt , where k and kt are the wave numbers of the longitudinal and transverse oscillations, respectively. The analyses showed that other zero-current coupling resonances also may exist, but are rarely signiﬁcant, with the possible exception of low-velocity resonant amplitude growth, when k = kt .

319

320

9 Multiparticle Dynamics with Space Charge

When space-charge is included, the physics of anisotropic beams is controlled by collective anisotropy resonances driven by space-charge forces [44–52]. By comparison with these collective effects, collisions between individual particles play no signiﬁcant role in the physics during the short time that the beam propagates in a linac. A 2D model of the physics by Hofmann [46] is based on analytical results obtained for an anisotropic K–V distribution in a linear uniform-focusing channel. The two dimensions are interpreted as transverse (x), and longitudinal (z) motion, coupled through the space-charge forces. The emittance growth is caused by density perturbations that excite the collective anisotropy resonant modes; some of these modes grow producing the nonlinear ﬁelds that cause the rms emittances to change. The resonant regions are conveniently displayed in ‘‘stability plots’’ that show the x spacecharge tune-depression ratio kx /k0x (ordinate), a measure of the importance of space charge, versus the tune ratio kz /kx (abscissa), which allow identiﬁcation of the anisotropy resonances. The ‘‘tunes’’ kx and kz are phase advances per transverse focusing period including space charge for an equivalent uniform beam, and the subscript zero refers to zero space charge. Some examples from Hofmann’s work are shown in the Fig. 9.9, [50] where in the ﬁgures the symbol k is replaced by the symbol ν. Plots are shown for constant values of the emittance ratio εz /εx , which is interpreted as the emittance ratio for the initial beam if emittances change. An energy anisotropy parameter Ta is deﬁned as the ratio of kinetic energies in the two degrees of freedom Ta ≡

εz kz εx kx

(9.83)

where energy equipartition corresponds to Ta = 1. To see this we consider the simple case of an rms-matched beam with uniform focusing in all three planes. Suppose we deﬁne a temperature for each degree of freedom by transforming to the center-of-momentum frame of the beam, which we call the beam frame, and deﬁning temperatures in the beam frame according to kB Tx ≡

p2bx , m

kB Ty ≡

p2by m

kB Tz ≡

p2bz m

(9.84)

Equipartitioning is deﬁned by the condition kB Tx = kB Ty = kB Tz . By performing a Lorentz transformation back to the laboratory frame, it can be shown [53] that equipartitioning implies 2 2 2 εn,y εn,z εn,x = = 2 2 xrms yrms γ 2 z2rms

(9.85)

Using the single-particle equations of motion for an equivalent uniform beam, one can also show that the rms beam sizes can be written as εn,y εn,x εn,z 2 2 xrms = , yrms = , z2rms = (9.86) βγ kx βγ ky βγ 3 kz

9.17 Longitudinal-Transverse Coupling and Space-Charge Instabilities for Anisotropic Linac Beams

where kx , ky , and kz are the phase advances per unit length of the equivalent uniform beam including space charge, and γ is the relativistic factor for the center of momentum frame. Substituting Eqs. (9.85) into Eqs. (9.86) to eliminate the rms beam sizes, yields a convenient expression for the equipartitioning condition εn,x kx = εn,y ky = εn,z kz .

(9.87)

The concept of equipartitioning is introduced here, not because linac beams always evolve to stationary states that are equipartitioned, which they do not, but because equipartitioned beams have no free energy available for energy and emittance transfer. Thus, if Ta = 1 there is no free energy available for emittance transfer. If Ta = 1, the beam is nonequipartitioned and free energy is available for emittance transfer from the degree of freedom that has the larger temperature, given by the larger product εi ki , to the degree of freedom having the smaller temperature. We are interested in the stability with respect to emittance transfer of beams that are rms matched but nonequipartitioned. The stability charts in Fig. 9.9 show contours for analytically calculated growth rates from the model. These contours identify the calculated stop bands (regions of exponential growth of a mode) that lie in the vicinity of tune ratios with integer tunes, for emittance ratios εz /εx ranging from 0.6 to 5.0. Growth rates in equidistant steps are deﬁned as the e-folding distance per zero-current transverse or betatron period. The most prominent stop bands in the plots are located near tune ratios νz /νx = 1/3, 1/2, 1, and 2. Note that there is no stop band at νz /νx = 1 for the emittance ratio εz /εx = 1, at νz /νx = 1/2 for the emittance ratio 2, and at νz /νx = 1/3 for the emittance ratio 3, all of which are equipartitioned cases. The tune ratios corresponding to equipartitioning at the given emittance ratio are marked as dashed vertical lines in the ﬁgures. The typical range for the transverse tune-depression ratio in high-current RF linacs is about 0.5 < νx /ν0x < 0.8. Energy and emittance transfer are conﬁned to the resonance stop bands, which are relatively narrow if the tune-depression ratio is moderate to weak. Simulations show that the integer stop band, νz /νx = 1, is usually the only signiﬁcant mode of concern for emittance transfer. Consequently, nonequipartitioned beams for RF linac designs are safe from emittance transfer, provided that the νz /νx = 1 stop band is avoided, or if the design trajectory does intersect this stop band that the emittance ratio and the parameter Ta are not far from unity, to limit the available free energy. As the space-charge tune depression becomes stronger (smaller tune depression ratios), Fig. 9.9 shows that the stop band widths increase and the structure of separated resonances seen at weak to moderate tune depressions, where most linacs are designed, vanishes as the resonance stop bands overlap. The thermodynamic picture that anisotropic beams approach energy equipartitioning for all tune ratios is applicable only very close to the spacecharge limit, where the stop bands completely overlap. Then emittance transfer

321

322

9 Multiparticle Dynamics with Space Charge ez /ex = 3

ez /ex = 0.6 0.03 1.18 0.33 0.48 0.63 0.78 0.93 1.08 1.23

0.6 0.4 0.2

0.5

1.5

0.8

0.6

0.4

0.2

0.4

n x / n0 x

0.8

0.2

0.5

ez /ex = 1.2

ez /ex = 5

0.8

0.6

0.4

0.2

0.5

1.5

nx /n0x

1.5 nz /nx

nz / nx

nz /nx

0.5

1.5

nz /nx

nx /n0x

ez /ex = 2 0.8

0.8

0.6

0.4

0.2

0.5

1.5

nz /nx

Figure 9.9 Stability charts with analytically calculated stop bands for arbitrary space-charge tune depressions νx /ν0x (ordinate), tune ratios νz /νx (abscissa), and emittance ratios εz /εx = 0.6, 1.2, 2, 3, and 5. Stop bands can be seen in the vicinity of νz /νx = 1/3, 1/2, 1, and 2. The growth rates

are deﬁned as the e-folding distance per zero-current betatron period. The dashed vertical lines indicate the tune ratios that correspond to energy equipartitioning, where emittance transfer would not occur (courtesy of I. Hofmann).

approaching equipartition can occur at all tune ratios rather than at only the discrete values; an initially equipartitioned case at very strong tune depressions would show no emittance transfer. While this might appear to avoid emittance transfer, at strong tune depressions emittance growth rather than emittance transfer could still occur in all three degrees of freedom with the free energy coming from nonlinear ﬁeld energy or from rms mismatch. Maintaining

9.17 Longitudinal-Transverse Coupling and Space-Charge Instabilities for Anisotropic Linac Beams

a bright beam at tune depressions near the space-charge limit remains a signiﬁcant challenge. To visualize better the risk of emittance transfer in a real high-current linac design, the design trajectory as a function of beam energy can be superimposed on the stability plot corresponding to the initial emittance ratio. To illustrate the effects discussed above for the parameter regime of a high-current linac, a study was made for three linac designs, the CERN-SPL design, the SNS linac, and the ESS linac [49]. The stability plots are shown in Figs. 9.10 and 9.11. For these two ﬁgures the phase advance per period is denoted by k rather than the symbol ν that was used in Fig. 9.9. Figure 9.10 shows the trajectory for the CERN SPL linac design on a stability plot for a nominal emittance ratio εz /εx = 2. Three cases are shown. Case 1 is the so-called short design that applies to the superconducting linac from 120 MeV to 2.2 GeV. Its trajectory stays outside the kz /kx = 1 stop band. Case 2 is a modiﬁed design with an increased tune ratio that overlaps with the kz /kx = 1 stop band. Case 3 corresponds to the full linac. For case 1, where the design trajectory does not enter the kz /kx = 1 stop band, no emittance exchange was observed. However, for case 2 where the beam trajectory enters the kz /kx = 1 stop band, signiﬁcant emittance transfer was observed; the longitudinal emittance decreased from about 0.75 to 0.56 mm-mrad and the transverse normalized emittance increased from about 0.4 to 0.5 mm-mrad. The authors comment that the transferred longitudinal energy is shared equally by both transverse degrees of freedom. Figure 9.11a shows the design trajectory for the full SNS linac on a stability plot for a normalized emittance ratio εz /εx = 1.4. In this design the tune ratio passes through three stopbands located near 1/3, 1/2, and 1. The tune ratio intercepts the kz /kx = 1 stop band at the low-energy end of the drift-tube linac (DTL) and decreases to about kz /kx = 1/3 at the end of the DTL. It increases

1 0.8

kx /k0x

0.73

SPL-short Case 2 SPL-full

0.63 0.53 0.43

0.6

0.33 0.23

0.4

0.13 0.03

0.2 0

0 0

0.2

0.4

0.6

0.8

1.2

1.4

kz /kx

Figure 9.10 Stability plot for εz /εx = 2 showing the superimposed CERN SPL design trajectory points as described in the text (courtesy of I. Hofmann).

323

9 Multiparticle Dynamics with Space Charge 1

0.73 0.63

kx /k0x

0.8

0.53 0.43

0.6

0.33 0.23

0.4

0.13 0.03

0.2 0

0 0

0.2

0.4

0.6

0.8

1.2

1.4

kz /kx

(a) 1

0.73 0.63

0.8

kx /k0x

324

0.53 0.43

0.6

0.33 0.23

0.4

0.13 0.03

0.2 0

0 0

(b)

0.2

0.4

0.6

0.8

1.2

1.4

kz /kx

Figure 9.11 Stability plot for εz /εx = 1.4(a) showing the SNS trajectory points, and for εz /εx = 1.3 (b) showing the ESS design. The design trajectory points are the + symbols (courtesy of I. Hofmann).

to near kz /kx = 1 in the superconducting linac. The simulation shows only a small effect of the stop band at kz /kx = 1, which is mainly in the DTL, and no signiﬁcant effects for the kz /kx = 1/3 and 1/2 stop bands. The net emittance change over the whole linac is +27% for the transverse case and +3% for the longitudinal case. The authors attribute some of the growth found in all directions to nonadiabatic behavior at some of the transitions. The design trajectory for the ESS linac design in Fig. 9.11b intersects the kz /kx = 1/3 and 1/2 stop bands, but not the kz /kx = 1 stopband, and no signiﬁcant emittance transfer was observed. To summarize, rms-emittance transfer is a small effect in nonequipartitioned beams, if the kz /kx = 1 stop band is avoided. If this is not possible, the

9.18 Beam Loss and Beam Halo

emittance ratio should not be large compared with unity to limit the available free energy for emittance transfer to the transverse degrees of freedom. A good example is the SNS design for which, even though the design trajectory intersects the stop band at kz /kx = 1, the emittance ratio εz /εx = 1.4 was still fairly close to unity. For moderate tune depressions such as in the examples shown, there are large safe regions in the tune depression versus tune-ratio space that are mostly resonance free, which means that nonequipartioned designs, such as that of SNS, can be safe from emittance transfer [54].

9.18 Beam Loss and Beam Halo

Beam halos are of concern for the operation of linacs with high-average beam intensity, because halos increase the risk of beam loss and radioactivation. An important design objective is to restrict beam losses to levels that will allow hands-on maintenance throughout the linac. The maximum tolerable beam-loss rate is conveniently expressed in terms of the lost beam power. If one adopts a hands-on-maintenance criterion that limits the activation level to 80 mrem/h at a distance of 1 ft from stainless steel at 4 h after shutdown of the accelerator, the loss rate for beam energies at 1 GeV must be limited to 1 W/m. Although not all emittance growth is necessarily associated with the generation of a halo, halo formation is always accompanied by emittance growth. It is known from computer simulations that a halo is produced from space-charge forces, especially from rms-mismatched beams. This is not unexpected physically, because mismatched beams produce time-varying space-charge forces that can transfer energy to some of the particles. Because there is no consensus about its deﬁnition, halo remains an imprecise term. In any given computer simulation one can unambiguously deﬁne an rms beam size, and a maximum particle displacement. Provided that the statistical precision is sufﬁciently adequate that the results are not sensitive to the motion of a few lone outer particles, the ratio of the maximum displacement to the rms size of the matched beam, which we call the maximum to rms ratio, is a useful ﬁgure of merit. Qualitatively, one can describe the evolution of the outer regions of the particle distribution for the case of an initial compact particle distribution (excluding the singular K–V distribution). A compact particle distribution might be arbitrarily deﬁned as having initial position and velocity coordinates that are contained within about 3σ . If this beam evolves in an rms-matched state, an equilibrium distribution develops in which the density at the beam edge falls off within about a Debye length. The Debye tail, whose size is a function of both the rms emittance and beam current, is a consequence of the propensity of the charges in a beam to provide shielding within the beam core. For a beam with a given current and emittance, the size of this tail relative to the rms size can be changed by changing the focusing

325

326

9 Multiparticle Dynamics with Space Charge

strength. Although there is no consensus about whether to call the Debye tail a halo, values of the maximum to rms ratio larger than about ﬁve are generally not observed in simulations for rms-matched beams, and such beams retain a very compact distribution. The outer region for a beam, with the same initial compact particle distribution in an rms-mismatched state, evolves differently. Many theoretical and numerical studies of halo formation in mismatched beams have been reported, showing larger amplitudes extending well beyond the Debye-tail of a matched beam. For practical estimates of expected mismatch in linacs, values for the maximum to rms ratio as large as 10 to 12 have been observed in simulations, and it is generally agreed that this is called halo. The principal halo-formation mechanism is a resonant interaction between the particles in the beam and a beam core that is oscillating because of mismatch. Numerical beam-dynamics simulations show that the largest transverse halo amplitudes are associated with mismatch excitation of the breathing mode of the beam, which is characterized by in-phase oscillations in x and y. The particle-core model of the breathing mode [56] provides understanding of the mechanism that transfers energy to halo particles, and provides a conceptual basis for the aperture choice. Suppose we represent a long bunch with a round, continuous beam in a uniform-focusing channel, and assume a uniform-density core. The core is assumed to be mismatched to excite the radial oscillation of the breathing mode. Halo formation can be studied by introducing single particles that oscillate through the core, driven by linear external focusing ﬁelds and the space-charge ﬁelds of the core, which are linear inside and nonlinear outside the core. Mathematically, the model may be expressed in dimensionless form as d2 X η2 I + X − − =0 ds2 X3 X and d2 x +x− ds2

Ix/X 2 , I/x,

x<X x≥X

(9.88)

(9.89)

where s is the product of the axial distance times the zero-current phaseadvance per unit length k0 , X is the ratio of the core radius to the matched core radius, η = k/k0 is the tune-depression ratio of the core, k is the single particle transverse phase advance per unit length including space charge, and I = 1 − η2 is a dimensionless parameter. Equation (9.88) is the rms envelope equation for the core, and Eq. (9.89) is the equation of motion of a single particle, where the normalized displacement x is deﬁned as the ratio of the particle displacement to the matched rms radius of the core. We note that since I depends on η, the two equations depend on a single dimensionless parameter η. It can be shown that η = 4ε/k0 R20 , where ε is the unnormalized rms emittance, and R0 is the matched core radius. It can also be shown that 1 − η2 = K/k20 R20 , where the K is the generalized perveance, deﬁned earlier.

9.18 Beam Loss and Beam Halo

Although the motion of a particle depends on the normalized core radius X, the dynamics of the core is independent of the motion of the particle, a reasonable approximation, if the halo particles represent only a small fraction of the total beam. The degree of mismatch is measured by the mismatch parameter µ, deﬁned as the ratio of the initial beam radius to the radius of the matched beam; an rms-matched beam has µ = 1. Equations (9.88) and (9.89) can be studied by systematically varying the initial coordinates of the particle. A stroboscopic phase-space plot, obtained by taking snapshots of many independent particle trajectories, once per coreoscillation cycle at the phase of the core oscillation that gives the minimum core radius, reveals a separatrix deﬁning three regions: (1) core-dominated, (2) resonance-dominated, and (3) focusing-dominated. The stroboscopic plot is shown in Fig. 9.12. The core-dominated trajectories are the innermost trajectories on the stroboscopic plot. The focusing-dominated trajectories are the outermost trajectories, which describe a peanutlike shape in the plot. The resonance-dominated trajectories are those centered about two ﬁxed points on the x axis, and are the trajectories most strongly affected by a parametric resonance, which occurs when the particle frequency is approximately half the core frequency. This resonance is responsible for the growth of the amplitudes that form a halo. For tune ratios below about η = 0.4, evidence for chaos is observed as the separatrix begins to spread, including the region of phase space very close to the core. This may provide a mechanism for including more

X ′/ k0R0

−2

−4 −4

−2

0 X /R0

Figure 9.12 Stroboscopic plot obtained by core oscillation that gives the minimum core radius. Initial particle coordinates were taking snapshots of many independent deﬁned on the x and x axes. particle trajectories, once per core-oscillation cycle at the phase of the

327

328

9 Multiparticle Dynamics with Space Charge

particles in the halo. The observation of chaos is consistent with numerical simulation of space-charge interactions, which often show extreme sensitivity to initial conditions. The nonlinearity of the resonance, associated with the decrease of the spacecharge force outside the core, results in a relationship between frequency and amplitude, which leads to a maximum amplitude for the resonantly driven halo particles, beyond which resonance is not sustained. This important feature is conﬁrmed by numerical simulations using the beam-dynamics codes, from which it is observed that the maximum halo amplitude does not grow without limit as the beam travels through an accelerator with linear external focusing forces. For given values of µ and η, the maximum amplitude for the resonantly driven particles is given by the location of the outermost point of the separatrix. The maximum amplitudes have been calculated from the numerical solution of Eqs. (9.88) and (9.89), as a function of µ between η = 0.5 and 0.9. The maximum amplitude is very insensitive to η, and can be described by an approximate empirical formula xmax /a = A + B| ln(µ)|

(9.90)

where xmax is the maximum resonant particle amplitude, and a is the matched rms beam size, which is identiﬁed in the model with the rms size of the core, and which can be expressed as a function of ε, k0 , and K, using the results in Section 9.8. The constants A and B are weak functions of the tune depression η. In the range 0.500 ≤ µ ≤ 0.952, and 1.05 ≤ µ ≤ 2.00, approximate results are A ≈ B ≈ 4. The ratio of the maximum amplitude to the matched rms beam size increases as a function of the deviation from unity of the mismatch parameter µ. If the main source of beam mismatch is at a single location, usually at a transition point, and if the maximum expected mismatch is known, Eq. (9.90) predicts the maximum halo amplitude. A choice of beam aperture can be made that is larger than the maximum, allowing an appropriate safety margin for a missteered beam centroid, and for trajectory ﬂutter in the quadrupoles. For the more realistic case where the mismatches originate from multiple sources, such as quadrupole gradient errors, the problem must be studied by numerical simulation. Modeling studies have shown that there are important differences between the transverse and the longitudinal dynamics responsible for beam halo. The transverse halo is caused primarily from particle interactions with the symmetric or breathing envelope mode of the bunch. This mode results in the largest density variations and therefore, the highest frequency, and interacts strongly with particles having high-frequency betatron orbits. The longitudinal dynamics are different in two respects [57]. First, the longitudinal particle frequencies are smaller because of the weaker longitudinal focusing, and second, the halo particles are subjected to relatively stronger nonlinear longitudinal forces. The combination of these effects, shifts the frequency scale for the resonant interaction and changes the primary mode that causes

9.19 Los Alamos Beam Halo Experiment

the longitudinal halo from the high-frequency breathing mode to the lowerfrequency antisymmetric envelope mode, where the longitudinal rms size oscillates out of phase with the two transverse rms sizes.

9.19 Los Alamos Beam Halo Experiment

A beam-halo experiment was carried out at the Low-Energy Demonstration Accelerator (LEDA) at Los Alamos [58, 59] to test two models of halo formation caused by beam mismatch in high-intensity beams: (1) the free-energy model for mismatched beams, which for a given mismatch strength, determines the maximum emittance growth resulting from the complete transfer of free energy into emittance (see Section 9.15), and (2) the particle-core model discussed in Section 9.18 in which beam mismatch excites a symmetric or antisymmetric (xrms and yrms in or out phase, respectively) oscillation of the core. A high-current, 75-mA, 6.7-MeV proton beam from an RFQ was injected into a 52-quadrupole focusing channel (see Fig. 9.13). The gradients of the ﬁrst four quadrupoles were independently adjusted to match or mismatch the injected beam. RMS emittances and beam widths were obtained from measured beam proﬁles for comparison with the theoretical models. The experimental results were also compared with multiparticle simulations. Beam scanners at nine stations, each located midway between pairs of quadrupoles, measured the horizontal and vertical beam distributions. These scanners were unique wire and scraper interceptive beam-proﬁle devices that measured beam proﬁles including the halo. Secondary electrons were produced in a 33-m carbon wire that was stepped through the beam core, allowing measurement of the beam density in the core. A pair of graphite scraper plates, in which the proton beam stopped, was stepped through the halo, allowing measurement of the density outside the core. Data from the wire and scraper plates were combined in the computer software to produce a single distribution. The scanners were labeled with numbers corresponding to the preceding quadrupole magnet number. The beam was matched, using a least-squares ﬁtting procedure that adjusted the ﬁrst four quadrupoles to produce equal rms sizes at the last eight scanner locations. Emittances and beam widths were obtained from the measured proﬁles. The major results are shown in Figs 9.14

Beam-profile diagnostic 52 quadrupole FODO lattice RFQ6.7 MeV

20 −26

45 −51

Figure 9.13 Block diagram of the LEDA halo experiment showing the 52-quadrupole-magnet lattice and the nine locations of beam-proﬁle scanners.

329

9 Multiparticle Dynamics with Space Charge

and 9.15. Figure 9.14 shows the x –y averaged rms-emittance growth results (points with error bars) versus the mismatch parameter µ (ratio of initial rms size to matched rms size) at scanner 20 for a 75-mA breathing-mode mismatch. The maximum emittance-growth curves from the free-energy model are shown for the two tune depression values that bracket the tune-depression values for the debunching beam. It can be seen that the theoretical maximum is insensitive to the tune depression over this range. The breathing-mode data in Fig. 9.14 are consistent at all µ values with the maximum emittance growth predicted by the model. The breathing-mode results at the downstream scanner 45 (not shown) show no signiﬁcant additional emittance growth, consistent with the upper limits from the model and with complete transfer of free energy. Overall, the data from both the breathing and the quadrupole mode indicate a rapid emittance growth with nearly complete transfer of free energy occurring in less than 10 mismatch oscillations. The particle-core model, discussed in Section 9.18, predicts the maximum resonant-particle amplitude as a function of the mismatch parameter µ. It was not possible to determine an experimental maximum amplitude for direct comparison with the model because of background halo in the input beam. Instead, the measured amplitudes (x –y averaged half widths of the beam) at three different fractional beam-proﬁle intensity levels (10, 1, and 0.1% of the peak) for breathing-mode mismatches were compared with the maximum amplitude predicted by the particle-core model. This comparison is shown in Fig. 9.15 at scanner 20 for 75 mA. The shapes of all three measured half-width curves are consistent with the shape and magnitude of the maximum amplitude curve from the particle-core model. Similar results are observed at scanner 51. Although the particle-core model based on a single mismatch mode is a simple

2.2 RMS emittance growth

330

Theory maximum−tune depression = 0.82 Theory maximum−tune depression = 0.95 Measurements

2 1.8 1.6 1.4 1.2 1 0.8 0.6

0.8

1.2

1.4

1.6

1.8

Mismatch parameter (µ)

Figure 9.14 Measured rms-emittance growth averaged over x and y for 75 mA at scanner 20 for a breathing-mode mismatch. The curves show maximum growth from the free-energy model.

9.20 Scaling of Emittance Growth and Halo

Half width/matched rms size

7 Maximum amplitude from particle-core model

6 5

0.1% level 4

1.0% level

10% level

2 1 0 0

0.2

0.4

0.6

0.8

1.2

1.4

1.6

1.8

Mismatch parameter (µ)

Figure 9.15 Measured beam half widths at scanner 20 (75 mA and a breathing-mode mismatch) at different fractional intensity levels versus mismatch strength µ for comparison with the maximum resonant amplitude of the particle-core model.

description of the beam dynamics, the agreement of the model for the curve shapes and for the consistency of the magnitudes supports the conclusion that the model describes the main physical mechanism responsible for the halo growth. The 6D-density distribution of the input beam, needed for multiparticle simulations, was not experimentally known. It was found that knowledge of the input Courant–Snyder parameters and emittances alone was not sufﬁcient to obtain agreement between numerical simulations and the observed halo distribution. Agreement between simulations and experiment for the growth rate of the halo will require knowledge of the input distribution, particularly of the tails of the input beam. The results strongly support both models for mismatched beams. This result is important because these models predict upper limits to emittance and halo-amplitude growth in high-current transport channels and linacs, and allow estimation of focusing strength and aperture requirements in new designs.

9.20 Scaling of Emittance Growth and Halo

Assuming that the bunch can be modeled as a long cylinder with a uniform longitudinal proﬁle, we obtain simple scaling formulas for transverse rms emittance growth and maximum particle amplitude for a mismatched beam in a linear focusing channel. The space-charge-induced rms-emittance growth

331

332

9 Multiparticle Dynamics with Space Charge

is a function of the tune-depression ratio k/k0 , where k and k0 are the transverse phase advances per unit length, with and without space charge. In the smooth approximation, where the focusing is represented by an equivalent uniform focusing channel, we may write k/k0 = 1 + ut2 − ut , where ut is a space-charge parameter given by 1 1 qI (9.91) ut = 16πε0 f γ 2 mc2 β εn k0 The parameters appearing in Eq. (9.91) are the charge q, rest energy mc2 , average beam current I, effective bunch length λ, bunch frequency f , relativistic mass factor γ , velocity (relative to that of light) β, and normalized rms emittance εn . The average current is related to the number of particles per bunch N and the bunch frequency, by I = qNf . The particle-core model for an rms-mismatched beam predicts that the halo particles created by the resonance have a maximum amplitude for a given core-oscillation amplitude. From Eq. (9.90), the numerical solution predicts that the maximum amplitude is proportional to the matched rms size a of the core, given by

εn a2 = 1 + ut2 + ut (9.92) k0 βγ In the space-charge-dominated limit, when ut 1, the rms beam size is 2εn ut qI 1 1 = (9.93) a2 ∼ = 2 2 2 k0 βγ 8πε0 f γ mc γβ k20 which is independent of the emittance. The emittance-dominated limit corresponds to ut 1, and we ﬁnd 1 qI 1 2 ∼ εn a = (9.94) 1+ k0 βγ 16πε0 f γ 2 mc2 β εn k0 In Eq. (9.90), the second term is much less than unity, and smaller emittance results in smaller a and smaller maximum amplitude. Higher frequency and longer bunch length produce smaller maximum amplitude. In Eqs. (9.93) and (9.94) stronger focusing (larger k0 ) also results in smaller maximum amplitude. These results must be taken with caution, because if the transverse focusing is too strong, the longitudinal space-charge forces will increase and eventually longitudinal halo could become a problem.

9.21 Longitudinal Beam Dynamics Constraint on the Accelerating Gradient

We introduce a simple model to describe a longitudinal envelope instability in a periodic array of cavities that provide longitudinal focusing of the beam.

9.21 Longitudinal Beam Dynamics Constraint on the Accelerating Gradient

The model illustrates the concept of a parametric resonance, which in this case occurs with periodic focusing, and shows how the accelerating gradient is limited by the instability. We write an equation for the longitudinal envelope Z with periodic longitudinal focusing from an array of RF cavities with period L, with space charge treated using the 3D ellipsoid model, and with s as the distance along the accelerator. Assuming a sinusoidal term with amplitude a for the periodic part of the longitudinal focusing, the envelope equation given by Eq. 9.43 is expressed as ε2 2πs 2 (9.95) Z + k 0 1 + a sin Z − 3 − K = 0 L Z where we have introduced k as the space-charge parameter. We write Z = Z0 as the matched solution where Z0 = 0, and introduce a small mismatch with amplitude ϕ, where Z = Z0 + ϕ and assume ϕ Z0 . Substituting Z = Z0 + ϕ into the envelope equation, and using the fact that the matched envelope Z0 also satisﬁes the longitudinal envelope equation, . we obtain an equation for the mismatch amplitude 3ε2 2πs ϕ+ 4φ=0 ϕ + k 20 1 + a sin (9.96) L Z or

3ε2 2πs ϕ=0 ϕ + k 20 + 4 ϕ = −k 20 a sin Z L

(9.97)

This is the equation for an oscillator with a driving term on the right side. Suppose that as a consequence of this driving term, the mismatch amplitude ϕ is driven sinusoidally. Then the driving term on the right side is the product of two sinusoidal factors, which constitutes the equation for a parametric resonance. The squared wave number of the envelope mismatch oscillation is the factor in the second term on the left: k2env = k 20 +

3ε2 Z4

(9.98)

Since the driving term is in general the product of two sinusoids with different wave numbers, it can be expressed as the sum of two terms, a ﬁrst term oscillating at the sum of the two wave numbers and the second term oscillating at the wave number difference. Resonance corresponds to the difference term, kenv =

2π − kenv L

(9.99)

or kenv L = π. Resonance corresponds to a phase advance of the envelope oscillation equal to π, which means that the period of the envelope oscillation is twice the period of the longitudinal focusing lattice.

333

334

9 Multiparticle Dynamics with Space Charge

Next we search for an approximate expression for the single particle wave numbers k and k0 , corresponding to the envelope being driven on resonance. These wave numbers will vary with the space-charge tune depression ratio. We begin by writing the single-particle equation of motion in a periodic focusing lattice for a particle whose longitudinal coordinate is z: 2πs K z z + k 20 1 + a sin z− =0 (9.100) L Z We will replace the second term using the smooth approximation from Section 8.7:

(ak0 L)2 K z =0 (9.101) z− z + k 20 1 + 2 8π Z Assuming a 1, we ignore the term that is second order in the amplitude a and obtain z + k 20 z −

K z ∼ =0 Z

(9.102)

The single particle wave number including space charge is K k2 ∼ = k20 − . Z

(9.103)

Recall that the wave number for the envelope is k2env = k 20 +

3ε2 Z4

(9.104)

In smooth approximation the matched envelope Z0 satisﬁes Z0 = 0. Then, k 20 Z0 −

ε2 − K = 0 Z03

(9.105)

K ε2 = 4 Z0 Z0

(9.106)

Then, k 2 = k 20 −

We can eliminate the emittance term and obtain a relationship between the wave number of the envelope mismatch oscillations and the wave numbers of single-particle motion with and without space charge, k2env = k 20 +

3ε2 = k 20 + 3k 2 Z04

(9.107)

The space-charge tune depression ratio is η = k /k0 . Then, k2env = k20 (1 + 3η2 ), or at resonance π σ0 = k0 L = 1 + 3η2

(9.108)

Problems

and σ = k L = ηk0 L =

πη 1 + 3η2

(9.109)

These two equations give the single particle tunes (phase advance per focusing period) σ and σ0 as a function of tune depression, when the longitudinal envelope is driven on resonance (kenv /L = π). At the space-charge dominated limit, η = 0, k = 0 and σ0 = k0 L = π. For the emittancedominated case, η = 1, σ = k L = π/2, and k0 L = π/2. Resonance is always avoided if σ0 = k0 L < π/2, and therefore we will conservatively adopt this as an upper limit to k0 . Further studies are required to determine whether this criterion is too conservative. Nevertheless, we will see that this constraint will limit the average accelerating gradient, and assuming this constraint and using Eq. 6.43 we write the squared zero-current longitudinal wave number as k20 =

σ 2 0

2π E0 T sin(−φ) π 2 ≤ mc2 γ 3 β 3 λ 2L

(9.110)

or E0 T ≤

πmc2 γ 3 β 3 λ 8q sin(−φ)L2

(9.111)

The longitudinal envelope instability constraint is important for high charge state, small mass, low velocities (cubic dependence on β), high frequencies, and for long focusing periods (quadratic dependence on L). It is a constraint on the spatial average gradient or real-estate gradient. Reducing the magnitude of the phase below about 30◦ does not help much because the phase width of the bucket shrinks, causing beam losses. Thus, this longitudinal envelope instability, a parametric resonance in a periodic focusing array, can limit the accelerating gradient at low velocities. It is generally not a problem for relativistic particles and is more of a problem for proton beams than for heavy ions.

Problems

9.1.

9.2.

9.3.

Given the transfer matrix of a drift space, derive the results for the change in the ellipse parameters. Show that the ellipse parameter γ˜ is invariant. What does this mean physically? Given the transfer matrix of a thin lens, derive the results for the change in the ellipse parameters. Show that the ellipse parameter β˜ is constant. Is this reasonable physically? Measurements of transverse beam distribution for a proton beam with velocity β = 0.06 show an upright ellipse in the x − x plane, with an

335

336

9 Multiparticle Dynamics with Space Charge

9.4.

9.5.

9.6.

9.7.

9.8.

rms size xrms = 3 mm and an rms divergence xrms = 0.06 mrad. (a) What is the unnormalized rms emittance? (b) What is the normalized rms emittance? (c) If the normalized rms emittance is conserved during acceleration to 100 MeV, what is the unnormalized rms emittance at 100 MeV? Measurements of the longitudinal beam distribution result in an upright ellipse in the φ –W plane with rms values φrms = 1◦ and Wrms = 100 keV. (a) What is the longitudinal emittance in deg-MeV? (b) If the RF frequency is 400 MHz, what is the longitudinal emittance in s-MeV? (c) If the beam is now injected into an 800 MHz linac, what is the emittance in deg-MeV and s-MeV? Assume a bunched proton beam at 100 keV with an average current of 100 mA and a frequency of 400 MHz, having a uniform spherical density with radius 2 mm. (a) Calculate the charge density and the number of particles per bunch. (b) Calculate the electric ﬁeld from space charge at the edge of the beam. (c) Calculate the space-charge parameter K3 . (d) Assume that the number of dimensions for which the distribution is uniform is n = 3. Calculate the projected rms size on the x axis. (e) If the rms unnormalized emittance in the x plane is εr π = 20π mm-mrad, what is the ratio of the space-charge term to the emittance term in the envelope equation? Is the beam space-charge or emittance dominated? Consider a 100-mA proton beam at 5 MeV in a 200-MHz DTL with parameters E0 T = 2 MV/m, and synchronous phase φs = −25◦ . Assume that the bunch is spherical and has an rms size of 2.24 mm in all three planes. (a) Calculate k0 , the zero-current wave number of small longitudinal oscillations. (b) For the equivalent uniform beam, what is the longitudinal tune-depression ratio k /k0 , the wave number of small longitudinal oscillations k including space charge, and the ratio of space-charge to focusing force µ ? For the same beam as in Problem 9.6, assume that the transverse focusing is provided by a FODO quadrupole array using quadrupoles with gradient-length product G = 3 T placed in every drift tube. (a) What is the period of the focusing lattice at 5 MeV? (b) Using the smoothapproximation formula given in Section 9.11 what is the zero-current phase advance per focusing period σ0 ? Does it meet the requirement σ0 < 90◦ to avoid envelope instability? (c) For the equivalent uniform beam, what is the transverse tune-depression ratio σ/σ0 , the phase advance per focusing period σ (including space charge), and the ratio of space-charge to focusing force µT ? (a) What is the space-charge electric ﬁeld in megavolt per meter at the edge of the beam bunch in Problem 9.6 assuming a uniform density bunch? (b) The longitudinal focusing force can be written as F = −mc2 k2o β 2 r. Calculate the effective longitudinal focusing ﬁeld at the edge of the beam in megavolt per meter for Problem 9.6 and compare to the axial accelerating ﬁeld Eo T cos ϕs .

Problems

9.9.

A 50-mA proton beam is injected into a 400-MHz DTL at a velocity β = 0.05. Assume the radial distribution is approximately parabolic with radius R = 4 mm, and bunch length 2b = 8 mm. Assume the emittance of the injected beam is essentially zero, and assume all buckets are ﬁlled. (a) Use the result of Section 9.14 to calculate the total normalized transverse emittance in units of mm-mrad after the beam is subjected to its own space-charge force over the ﬁxed distance of the ﬁrst cell. (b) Calculate the beam plasma period in meters at the DTL injection velocity assuming that the bunch parameters are constant during the ﬁrst beam-plasma oscillation. (c) Calculate the quasisteady state transverse normalized emittance after the beam has gone through one plasma period, using the result of Section 9.14. Express the answer in units of mm-mrad. (d) Suppose that the beam radius is decreased to R = 2 mm. How do the results of parts (a), (b), and (c) change as a result of the decreased beam radius. 9.10. (a) Consider a distribution of particles in phase space that lies on a straight line given by x = ax. Use the deﬁnition of rms emittance to show that the rms emittance is zero. (b) Suppose the particles lie on a curved line described by x = ax3 . This line still has zero area as in part (a), but is the rms emittance still zero? 9.11. A 100-mA dc proton beam with an initial parabolic charge density, and initial velocity β = 0.01 is injected from an ion source into an RFQ. Assume the RFQ electric ﬁelds will remove neutralizing electrons, so the beam sees the full space-charge force. The beam is matched and the focusing is always adjusted to give an average rms size R = 7.5 mm. (a) Calculate the generalized perveance K. (b) The beam charge will rapidly redistribute accompanied by emittance growth, and from the model in section 9.14 the ﬁnal unnormalized emittance ε for an initially zero emittance beam is given by Eq. 9.76. Assuming that for a beam

with initial emittance εi , the ﬁnal emittance is given by εf = εi2 + ε2 . Calculate and plot the ﬁnal normalized rms emittance βγ ε1 for initial normalized emittances βγ ε1 = 0, 0.1, 0.2, 0.3, and 0.4 mm-mrad. If an ion source breakthrough allows the input rms-normalized emittance to be reduced from 0.2 to 0.1 mm-mrad, would the RFQ be able to preserve this emittance reduction? (c) If the emittance growth and beam relaxation to the ﬁnal uniform density occurs in one quarter plasma period, what is this relaxation distance? 9.12. Show that the radial electric ﬁeld and potential from space charge at the edge of a round continuous uniform beam of radius R and velocity β can be written as E = Ib /2πε0 cβR = 60Ib /βR in V/m, and V = Ib /4πε0 cβ = 30Ib /β in V, where Ib = qN βc is the average current during the passage of the bunch, expressed in terms of the number of particles per unit length N . 9.13. If we choose a round continuous uniform beam of length as a cylinder model for a bunch, the current I averaged over an RF period can be

337

338

9 Multiparticle Dynamics with Space Charge

written as I = Ib /βλ, where Ib is the average current during the passage of the bunch, and λ is the RF wavelength. Use this and the results from Problem 9.12 to calculate the radial electric ﬁeld and potential from space charge at the edge of the bunch for the parameters β = 0.1, R = 0.5 mm, I = 0.1 A, and /βλ = 0.2. 9.14. Resolve the apparent paradox that the most effective approach for controlling space-charge effects is to apply strong focusing, which makes the beam-size small and increases the space-charge force. Assuming an rms-matched, round, continuous beam propagating in a uniformfocusing channel, write an expression for the ratio of the space-charge force to the focusing force. Using the analytic solution of the rms envelope equation, and assuming a ﬁxed generalized-perveance K and ﬁxed emittance ε, show that if the focusing strength k0 increases, the ratio of the space-charge force to focusing force decreases. Thus, although with increased focusing the space-charge force has increased, the focusing force has increased even more, making the ratio smaller.

References 1 For a more complete discussion of this

3 4 5 6

7 8

and other space-charge and emittance topics, the reader is referred to the excellent treatments in the books by Lawson, J.D., The Physics of Charged Particle Beams, 2nd ed., Oxford University Press, 1988; Reiser, M., Theory and Design of Charged Particle Beams, John Wiley & Sons, New York, 1994. Lawson, J.D., The Physics of Charged Particle Beams, 2nd ed., Oxford University Press, 1988, pp. 151–155. Sacherer, F., IEEE Trans. Nucl. Sci. 18, 1105 (1971). Lapostolle, P.M., IEEE Trans. Nucl. Sci. 18, 1101 (1971). Weiss, M., unpublished. Lawson, J.D., The Physics of Charged Particle Beams, 2nd ed., Oxford University Press, 1988, p. 197. Ibid. p. 246. For a more detailed treatment of numerical simulation of particle interactions see Hockney, R.W. and Eastwood, J.W., Computer Simulation Using Particles, Adam Hilger (IOP Publishing Ltd), 1988. Crandall, K.R., TRACE, An Interactive Beam Dynamics Code, Linear

11 12

Accelerator and Beam Optics Codes, ed. Charles R. Eminhizer, La Jolla Institute, 1988; AIP Conf. Proc. 177, 29 (1988). Crandall, K.R., TRACE 3-D Documentation, 2nd ed. Los Alamos Report LA-UR-90-4146, December 1990. Crandall, K.R., unpublished. Crandall, K.R., Los Alamos Group AT-1 Memorandum AT-1-237, September 2, 1980. Guy, F.W., 3-D Space-Charge Subroutine for PARMILA, Los Alamos Group AT-1 Report, AT-1:85–90, March 6, 1985. Martini M., and Prome, M., Computer studies of beam dynamics in a proton linear accelerator with space charge, Part. Accel. 2, 289 (1971). Lapostolle, P.M., Lombardi, A.M., Nath, S., Tanke, S., E., Valero, and Wangler, T.P., A New Approach to Space Charge for Linac Beam Dynamics Codes, Proc. 18th Int. Linac Conf., August 26–30, 1996, CERN, Geneva, 1996, p. 375. Wangler, T.P., Crandall, K.R., Mills, R.S. and Reiser, M., High current, high brightness, and high duty factor

References

20 21

ion injectors, AIP Conf. Proc. 139, 133 (1985). Kapchinsky I. and Vladimirsky, V., 2nd Conf. High Energy Accel., CERN, 1959, p. 274. Weiss, M., Proc. 1979 Linear Accel. Conf., September 10–14, 1979, Montauk, NY, Brookhaven National Laboratory Report, BNL 51134, p. 227. Wangler, T.P., Some Useful Elementary Relationships for a Continuous Beam with Space Charge, Space-Charge, Los Alamos Group AT-1 Report AT-1:85–71, February 20, 1985. Lapostolle, P., CERN Report AR/Int SG/65-15, 1965. Wangler, T.P., Space-Charge Limits in Linear Accelerators, Los Alamos Report LA-8388, December, 1980. Reiser, M., Theory and Design of Charged Particle Beams, John Wiley & Sons, New York, 1994, pp. 240–251. Wangler, T.P., High-Brightness Injectors for Hadron Colliders, 1990 Joint US-CERN School on Particle Accelerators, November 7–14, 1990, Hilton Head, SC. Wangler, T.P., Crandall, K.R. and Mills, R.S., Heavy ion inertial fusion, AIP Conf. Proc. 152, 166 (1986). The K–V equilibrium distribution is an exception that always has a uniform spatial density with no Debye tail. However, this distribution bears no resemblance to real beam proﬁles that have been measured or observed in numerical simulations. Kapchinsky, I.M., Theory of Resonance Linear Accelerators, Harwood Academic Publishers (1985), p. 273. Kapchinsky, I.M. and Vladimirsky, V.V., Limitations of Proton Beam Current in a Strong Focusing Linear Accelerator Associated wit the Beam Space Charge, Proc. Int. Conf. High-Energy Accel. Instrumentation CERN, Geneva, 1959, p. 274. Wangler, T.P., Crandall, K.R., Mills, R.S., and Wangler, T.P., Field Energy and RMS Emittance in Intense Particle Beams, High Current, High Brightness, and High Duty Factor Ion Injectors, AIP Conf. Proc., 139, La Jolla, CA, (1986), pp. 133–151.

29 Reiser, M., Chang, C.R., Kehne, D.,

30 31

Low, K., Shea, T., Rudd, H. and Haber, I., Phys, Rev. Lett. 61, 2933 (1988). Haber, I., Kehne, D., Reiser, M. and Rudd, H., Phys. Rev. A 44, 5194 (1991). Reiser, M., Theory and Design of Charged Particle Beams, 1st ed., John Wiley & Sons, 1994, pp. 482–488. Wangler, T.P., Lapostolle, Pierre and Lombardi, A., Proc. 1993 Part. Accel. Conf., Washington D.C., May 17–20, 1993, IEEE Catalog No. 93CH3279-7, p. 3606. Reiser, M., Theory and Design of Charged Particle Beams, 1st ed., John Wiley & Sons, 1994, p. 242. Reiser, M., Theory and Design of Charged Particle Beams, 1st ed., John Wiley & Sons, 1994, p. 476. Cucchetti, A., Reiser, M. and Wangler, T.P., Simulation Studies of Emittance Growth in RMS Mismatched Beams, Proc. 1991 Part. Accel. Conf., San Francisco, CA, May, 1991. Reiser, M., Theory and Design of Charged Particle Beams, 1st ed., John Wiley & Sons, 1994, pp. 470–479. Reiser, M., Theory and Design of Charged Particle Beams, 1st ed., John Wiley & Sons, 1994, pp. 488–491. Gluckstern, R.L., Proc. 1970 Linear Accel. Conf., Fermi National Accelerator Laboratory, Batavia, IL, 1970, Vol. 2, p. 811. Hofmann, I., Laslett, L.J., Smith, L. and Haber, I., Stability of the Kapchinsky-Vladimirsky (K–V) distributions in long periodic transport systems, Part. Accel. 13, 145 (1983). Tiefenback M.G. and Keefe, D., Measurements of stability limits for a space-charge-dominated ion beam in a long A.G. transport channel, IEEE Trans. Nucl. Sci. NS-32, 2483–2485 (1985). Brown, N. Thermal Equilibrium of Charged Particle Beams, Ph.D. Dissertation, Physics Department, University of Maryland, College Park, MD, 1995. Weiss, M. Bunching of Intense Proton Beams with Six-Dimensional Matching to the Linac Acceptance, CERN/MPS/LI report 73-2, Geneva, Switzerland

339

340

9 Multiparticle Dynamics with Space Charge

(1978); and Crandall, K.R. and Rusthoi, D.P., Appendix F, Emittance Increase in a Gap, 3rd ed., TRACE 3-D Documentation, Los Alamos Report LA-UR-97-886, May, 1997, pp. 78–80. Gluckstern, R.L., Transverse Beam Growth Due to Longitudinal Coupling in Linear Accelerators, Proc. of 1966 Linear Accel. Conf., LA Report LA-3609, Los Alamos, NM, October 3–7, 1966, pp. 207–213; and Gluckstern, R.L., Nonlinear effects, in Linear Accelerators, Lapostolle P.M. and Septier, A.L., John Wiley & Sons, New York, (1970) pp. 797–807. Hofmann, I. and Bozsik, I., Computer Simulation of Longitudinal-Transverse Space-Charge Effects in Bunched Beams, Proc. 1981 Linear Accel. Conf., Los Alamos National Laboratory Report LA-9234-C 1982, p. 116. Jameson, R.A., Equipartitioning in Linear Accelerators, Proc. 1981 Linear Accel. Conf., Los Alamos National Laboratory Report LA-9234-C (1982), p. 125. Hofmann, I., Stability of anisotropic beams with space charge, Phys. Rev. E 57, 4713 (1998). Hofmann, I., Qiang, J., and Ryne, R.D., Collective resonance model of energy exchange in 3D nonequipartitioned beams, Phys. Rev. Lett. 86, 2313 (2001). Hofmann, I., and Boine-Frankenheim, O., Resonant emittance transfer driven by space charge, Phys. Rev. Lett. 87, (034802-1) 2001. Hofmann, I., Franchetti, G., Qiang, J., Ryne, R., Gerigk,F., Jeon, D., and Pichoff, N. Review of Beam Dynamics and Space Charge Resonances in High Intensity Linacs, Proc. of EPAC 2002, Paris, France, p. 74. Hofmann, I., Franchetti, G., Boine-Frankenheim, O., Qiang, J. and Ryne, R.D., Space charge resonances in two and three dimensional anisotropic beams, Phys. Rev. ST-AB, 6, 024202 (2003). Reiser M., and Brown, N., Proposed high-current RF linear accelerators

with beams in thermal equilibrium, Phys. Rev. Lett. 74, 1111 (1995). Kishek, R.A., O’Shea, P.G. and Reiser, M., Energy transfer in nonequilibrium space-charge dominated beams, Phys. Rev. Lett. 85, 4514 (2000). Reiser, M., Theory and Design of Charged Particle Beams, John Wiley & Sons, New York, 1994, p. 573. Hofmann, I., Collective resonance model of energy exchange in 3D nonequipartitioned beams, Phys. Rev. Lett. 86, 2313 (2001). Gluckstern, R.L., Proceeding 1970 Linear Accelerator Conference, Fermi National Accelerator Laboratory, Batavia, IL, 1970, Vol. 2, p. 811. Wangler, T.P., Crandall, K.R., Ryne, R., and Wang, T.S., Particle-core model for transverse dynamics of beam halo, Phys. Rev. ST Accel. Beams 1, 084201 (1998); Gluckstern, R.L., Analytic model for halo formation in high current beams, Phys. Rev. Lett. 73, 1247 (1994). See these papers for additional references. Barnard J., and Lund, S., 1997 Part. Accel. Conf., Vancouver, Canada, May 12–16, 1997, to be published. Allen, C.K., Chan, K.C.D., Colestock, P.L., Crandall, K.R., Garnett, R.W., Gilpatrick, J.D., Lysenko, W., Qiang, J., Schneider, J.D., Schulze, M.E., Shefﬁeld, R.L., Smith, H.V. and Wangler, T.P., Beam-halo measurements in high-current proton beams, Phys. Rev. Lett. 89, 214802 (2002). Wangler, T.P., Allen, C.K., Chan, K.C.D., Colestock, P.L., Crandall, K.R., Garnett, R.W., Gilpatrick, J.D., Lysenko, W., Qiang, J., Schneider, J.D., Schulze, M.E., Shefﬁeld, R.L. and Smith, H.V., Beam Halo in Mismatched Proton Beams, 2002 Charged Particle Optics Conference, October 22–25, 2003, Greenbelt, MD, published in Nuclear Instruments and Methods in Physics Research A 519, p. 425–431 (2004).

341

10 Beam Loading

The beam is not simply a medium which absorbs radio frequency (RF) energy and adds an additional resistive load to the cavity, but is really equivalent to a generator, which can either absorb energy from the cavity modes or deliver energy to them. However, in the previous chapters we have assumed that the beam is affected by the cavity, but we have ignored the effects of the beam on the cavity ﬁelds. As the beam current increases it becomes important to treat the effects of the interaction between the beam and the cavity more carefully. The effects of the beam on the cavity ﬁelds in the accelerating mode are referred to as beam loading. They arise physically as a result of the charges induced in the walls of the cavity as the bunches, together with their comoving electromagnetic ﬁelds, pass through the cavity. The induced charges produce ﬁelds that act back on the particles in the bunch, and this interaction does work on the particles, which radiate energy to the cavity. In this chapter we introduce these effects through the fundamental theorem of beam loading, which relates the energy delivered to the cavity to the charge q of a single point charge passing through it. The proportionality factor between the energy transferred to the cavity and the squared charge is called the loss factor. The most important effects of beam-induced ﬁelds in modes other than the accelerating mode, are higher-order-mode power losses, and the beam-breakup instability, both of which are discussed in Chapter 11. In this chapter, we emphasize the accelerating mode, a special case in which an external generator is provided to establish the ﬁelds needed for acceleration of the beam, and in which the net ﬁeld is a superposition of the generatorinduced plus the beam-induced ﬁelds. When the beam-induced ﬁeld in the accelerating mode becomes comparable to the ﬁeld induced by the external generator, the net phase and amplitude will only be satisfactory for beam acceleration, if some means of compensation for the effect of the beam is provided. The conventional analysis of beam loading in a standing-wave cavity, including the conventional method used for compensation, is presented in this chapter. RF Linear Accelerators. 2nd, completely revised and enlarged edition. Thomas P. Wangler Copyright  2008 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-40680-7

342

10 Beam Loading

10.1 Fundamental Beam-Loading Theorem

Consider a point charge q moving through an initially unexcited loss-free cavity. The ﬁelds from the moving charge will induce surface charges on the cavity walls and a total beam-induced voltage in the cavity. After the charge has left the cavity, the induced surface charges and voltage will remain, and by energy conservation the moving charge must leave some fraction of its energy there. The energy that the charge looses must be the result of work done on the charge by the induced voltage, and that voltage must decelerate the charge to conserve energy. However, because the voltage is in the process of being established while the charge is in the cavity, it is not obvious what fraction of the ﬁnal induced voltage is actually seen by the exciting charge. We will apply energy conservation to deduce the answer to this question, using a proof that applies for a single arbitrary cavity mode [1]. We consider two equal charges q, each with kinetic energy W, that are spaced apart by half of a wavelength for the cavity mode of interest, the second charge following the ﬁrst through the initially unexcited cavity, as shown in Fig. 10.1. We assume that the velocity of each charge is constant while passing through the cavity. The ﬁrst charge excites the cavity, producing a retarding cavity voltage Vc = −Vb , shown in Fig. 10.2. We express the induced voltage seen by particle 1 as Vq1 = fVb , where f will be the unknown fraction of the total beam-induced voltage that acts on the charge that produces it. Thus the energy loss for charge 1 is W1 = −qfVb , and we express the induced stored energy by U = αVb 2 , where α is a constant of proportionality. From energy conservation U = αVb 2 = −W1 = qfVb . This implies that Vb = fq/α, so the induced cavity voltage is proportional to the charge that produced it. Half a period later, shown in Fig. 10.3, the second charge crosses the cavity. The voltage that was induced by particle 1 has changed its phase by π and, if the cavity is lossless, is now an accelerating voltage +Vb . The second charge, being identical to the ﬁrst, will also induce a retarding voltage −Vb . The two induced voltages add out of phase to give a net cavity voltage Vc = 0, so after the second particle has crossed the cavity, the cavity stored energy must be U = 0.

Cavity

W q

W q z l/ 2

Vc = 0 U=0

Figure 10.1 Initial conﬁguration showing two identical charges and an unexcited cavity.

10.1 Fundamental Beam-Loading Theorem

Cavity

q z

Vc = –Vb

W–qfV b

U = aV b2

Figure 10.2 Conﬁguration after the ﬁrst charge has crossed the cavity.

Cavity

q z

Vc = 0 U=0

W + qV b –qfV b

W – qfV b

Figure 10.3 Final conﬁguration with an empty cavity. The energy left in the cavity by the ﬁrst charge is absorbed by the second charge.

From energy conservation, the sum of the energy changes of the two charges must also be W1 + W2 = 0. The energy change of charge 2 is the sum of the energy gain qVb , from the voltage induced in the cavity by charge 1, which after half an RF period has changed its phase by π, plus the energy loss to the cavity qVb f . Adding the energy gain of charge 2, W2 = qVb (1 − f ), to the energy loss of charge 1, W1 = −qfVb , we obtain W1 + W2 = qVb (1 − 2f ) = 0, which implies that f = 1/2. Therefore, the energy loss of the ﬁrst charge to the initially unexcited cavity is given by W1 = qVb /2. In other words, the induced voltage that the ﬁrst charge sees is equal to half the induced voltage that the particle leaves in the cavity, or Vq = Vb /2

(10.1)

Equation (10.1) is called the fundamental theorem of beam loading. Note that no assumption was made regarding the velocity of the charges, so the result is valid for both relativistic and nonrelativistic charges. We notice also that the mechanism described above produces acceleration of a trailing charge, by using the cavity to transfer energy from a particle to a later particle.

343

344

10 Beam Loading

10.2 The Single-Bunch Loss Parameter

It is convenient to deﬁne the loss parameter k = 1/4α for each cavity mode, so that the energy loss of a point charge q into an empty cavity mode is U = αVb2 =

q2 = kq2 4α

(10.2)

the voltage induced in the cavity mode by the point charge q is Vb

= 2kq

(10.3)

and the induced voltage in that mode seen by the charge itself is Vq

= kq

(10.4)

Furthermore, because k = 1/4α, we can express the cavity stored energy induced by the point charge as U = αVb 2 = Vb 2 /4k, so that the loss parameter is k = Vb 2 /4U. It is convenient to express k in a form that allows it to be evaluated from the basic properties of the cavity mode, computed from a standard electromagnetic code such as SUPERFISH [2]. We identify the axial voltage Vb as the voltage V0 T that includes the transittime factor T. Then we introduce the effective shunt impedance r = Vb2 /P in ohms, where P is the cavity-power dissipation. Introducing the unloaded quality factor Q0 = ωU/P for the mode of interest, we express the loss factor, which describes the single-bunch beam-loading characteristics of a mode, as k=

ω r 4 Q0

(10.5)

10.3 Energy Loss to Higher-Order Cavity Modes

When a charge q traverses a cavity, it will induce a voltage Vbn = 2kn q for each mode n. The total induced voltage is the sum of the induced voltages for each mode. We identify the accelerating mode with a subscript 0. The other modes will be called higher-order modes and will be labeled by a subscript n > 0. When a charge q crosses the initially unexcited cavity, the total beam-induced voltage VbT is the sum of the fundamental plus the higher-order-mode voltages. Thus VbT =

n=0

Vbn = Vb0

Vbn n=0

Vb0

= Vb0

kn n=0

(10.6)

10.4 Beam Loading in the Accelerating Mode

where Vb0 = 2qk0 and Vbn = 2qkn . We deﬁne a higher-order-mode enhancement factor, as B=

∞ kn n=0

(10.7)

For the SLAC linac [3] with a typical RMS bunch length of 1 mm, the higherorder mode enhancement factor is about B = 4. The total energy left by the passage of the charge through the cavity is UT = q2

kn = k0 q2

n=0

kn n=0

= k0 q2 B

(10.8)

A point charge loses energy to all modes of an initially unexcited cavity. The unavoidable excitation of the higher-order modes results in beam-energy loss and cavity-power dissipation. The excess beam-energy loss to higher-order modes is Uhom = UT − U0 = k0 q2 (B − 1)

(10.9)

and the excess power loss is Phom =

ωn Un n>0

Q0n

= q2

ωn kn n>0

Q0n

(10.10)

10.4 Beam Loading in the Accelerating Mode

In this section we describe the effects of beam loading in the accelerating mode of a standing-wave cavity, [4] when an accelerating voltage is already present in the cavity before the bunch arrives. We need to add the generatorinduced and beam-induced voltages to obtain the total voltage. The result that the voltages from independent sources can be added together to give the total voltage follows from the principle of linear superposition, which is a consequence of the linearity of Maxwell’s equations. For the case of sinusoidally varying ﬁelds and voltages, one must add these, taking into account the relative phases. It is convenient to describe the voltages as vector quantities in the complex plane. Thus we write a voltage as V˜ = Vej(ωt+θ) where V is the magnitude and ωt + θ is the phase. This is a vector that rotates counterclockwise in the complex plane and is called a phasor. It is convenient to choose a frame of reference that is rotating with a frequency ω, so that the phasors remain ﬁxed in time. It is convenient to choose zero phase to correspond to the time of arrival of the center of the bunch at the electrical center of the cavity. Then the component of any voltage that contributes to acceleration of the bunch is the projection of the voltage onto

345

346

10 Beam Loading

the real axis, which corresponds to the phase of the beam current. In Fig. 10.4 we show a phasor diagram for a cavity that is operating on resonance in a steady state, and is excited both by a generator and a train of bunches. The generator-induced voltage is V˜ g , and it has a phase θ relative to the fundamental Fourier component of the bunch current. The beam-induced voltage, which maximally opposes the beam, is V˜ b . The net cavity voltage V˜ c is the vector sum of these two voltages, and it has a phase φ relative to the beam; φ is also the synchronous phase. Figure 10.4 shows that the net cavity voltage deviates from the generator-induced voltage, because of the beam-induced voltage. Beam-loading compensation is important, because it affects the RF system efﬁciency. In many high-energy electron linacs the bunches are accelerated at the peak of the accelerating wave to achieve the maximum energy gain. The cavity is driven at its resonant frequency by an RF generator through an input RF power coupler, which acts as a transformer that can be adjusted to match the real part of the cavity impedance to the input transmission line. A schematic drawing of the RF power system is shown in Fig. 10.5. The coupler geometry is ideally adjusted to minimize the reﬂected power when the beam is present. If the cavity is driven on resonance, the cavity impedance seen by the generator is a pure real number, and the coupler geometry can be chosen to give zero reﬂected power. The beam-induced and generator-induced voltages are 180◦ out of phase, and beam-loading compensation requires that the generator current must be increased to correct for the voltage reduction caused by the beam. If the bunches are injected off the crest of the accelerating wave, the reﬂected power can be adjusted to zero by employing the two-step approach of (1) compensating for the phase difference by detuning the cavity so that it is driven off resonance, and (2) adjusting the coupler for minimum reﬂected power. In the following section, we review the theory of a steady-state, beam-loaded cavity.

Im(V ) Re(V ) ∼ Vb

f ∼ Vc

q ∼ Vg

Figure 10.4 Voltage phasors for a steady-state, beam-loaded cavity operating at resonance. The quantities shown in the ﬁgure are deﬁned in the text.

10.5 Equations Describing a Beam-Loaded Cavity

Matched load Cavity

Power coupler

Circulator RF generator Figure 10.5 Simpliﬁed drawing of the RF system including the generator, circulator, RF input drive line or waveguide, power coupler, cavity, and the beam.

10.5 Equations Describing a Beam-Loaded Cavity

In this section, we present the theory by Wilson [5] for a beam-loaded cavity. The equivalent circuit shown in Figure 10.6 is the same as described in Section 5.4, except that an additional generator is added to represent the beam. We see immediately that the beam functions not as an additional load, represented by a resistance, but as a generator, which can excite the cavity. A consequence is that the cavity time constant and the loaded quality factor QL do not change when the beam is turned on. A physical argument for representing the beam by an equivalent generator rather than a resistive load is that if the generator is turned off for an operating accelerating structure, the ﬁelds do not simply decay to zero, but the structure continues to be excited by the beam to a new steady-state amplitude and phase. Figure 10.7 shows the equivalent circuit for the beam-loaded cavity transformed into the resonator circuit. The quantity Rc /β is the impedance seen looking back toward the generator from the cavity (i.e., the impedance seen by the reﬂected wave from the coupler or a wave emitted from the cavity).

Waveguide

1:n

Z0 C

Matched generator

Power coupler

Cavity

Beam

Figure 10.6 Equivalent circuit for the beam-loaded cavity. The cavity is represented by the LRC circuit. The RF generator and the beam are represented by current generators.

347

348

10 Beam Loading Vc

Rc /b

Figure 10.7 Equivalent circuit for the beam-loaded cavity transformed into the resonator circuit.

We introduce the following notation. The generator current is ig , the average beam current is I, and ib ≈ 2I is the harmonic component of the beam current at the bunch frequency, assuming that the bunches are short compared with their spacing, the beam image current −ib has opposite sign to the beam current ib , φ is the synchronous phase, which represents the phase of the net cavity voltage Vc relative to the phase of ib , θ is the phase of the generator current ig relative to the phase of ib , rs = 2Rc is the effective shunt impedance, deﬁned by rs = Vc2 /Pc , so that Vc = V0 T, where T is the transit-time factor, and Pc is the average power dissipation in the cavity walls. The average power delivered to the beam is Pb = IVc cos φ. As deﬁned in Chapter 5, the cavity Q or quality factor, unloaded by the external circuit is Q0 , and the quality factor loaded by the external circuit is QL = Q0 /(1 + β), where the waveguide-to-cavity coupling factor is β = Q0 /Qext . The external Q, or Qext , is a function of the geometry of the coupler. The cavity resonant frequency ω0 is not necessarily exactly equal to the driving frequency of the generator and beam ω. It is convenient to deﬁne the detuning angle ψ as tan ψ = −2QL δ

(10.11)

where δ≡

ω − ω0 , ω0

|δ| 1

(10.12)

General Results

The currents and voltages are conveniently expressed as phasors, which have a magnitude and phase, and can be represented in complex notation. By convention the beam current ib is taken along the positive real axis, which places the beam image current and the beam-induced voltage along the negative real axis. If the detuning angle ψ is positive, the generator and the beaminduced voltages lead then respective currents by ψ. The cavity impedance ZL , loaded by the external resistance, determines the voltage induced by any

10.5 Equations Describing a Beam-Loaded Cavity

driving current i according to V = iZL , and is given by ZL =

rs ejψ cos ψ 2(1 + β)

(10.13)

Then, we write the beam image current, the generator current, the beaminduced voltage, the generator-induced voltage, and the net cavity voltage in phasor notation as ˜ib = ib ejπ = −ib

(10.14)

˜ig = ig ejθ

(10.15)

Irs cos ψ j(π+ψ) V˜ b = Vb ej(π+ψ) = e 1+β 2 βrs Pg cos ψ j(θ+ψ) j(θ+ψ) ˜ = Vg = Vg e e 1+β

(10.16) (10.17)

and V˜ c = V˜ b + V˜ g

(10.18)

Including beam loading and cavity detuning, the general expression has been derived for the required generator power, (1 + β) 2 1 Pg = Pc 4β cos2 ψ

cos φ +

Vb cos ψ Vc

Vb sin ψ 2 + sin φ + Vc

(10.19)

where Vb is the magnitude of the beam-induced voltage in the cavity, given by Vb =

Irs cos ψ 1+β

(10.20)

From conservation of energy, Pr = Pg − Pc − Pb , from which we ﬁnd the general result for the reﬂected power Pr = Pc

(1 + β)2 1 4β cos2 ψ

cos φ +

Vb cos ψ Vc

Vb sin ψ 2 −1 + sin φ + Vc − IVc cos φ

(10.21)

349

350

10 Beam Loading

If the beam current is zero, Vb = 0, and the generator power and reﬂected power reduce to Pg = Pc and

(1 + β)2 4β cos2 ψ

Pr = Pc

(1 + β)2 −1 4β cos2 ψ

(10.22) (10.23)

From Section 5.4, we relate the generator current to the generator (forward) power as Pg = i2g rs /16β. The magnitudes of the beam-induced and generatorinduced voltages in the cavity are given by Vb = and Vg =

Irs cos ψ (1 + β) 2 βrs Pg 1+β

(10.24)

cos ψ

Optimum Detuning

To put the cavity voltage Vc in phase with the generator current ig , so that the cavity presents a real impedance to the generator, it is necessary that θ = φ. Then it is found that Vb sin ψ =− (10.25) Vc sin φ It is necessary to detune the cavity resonant frequency so that the detuning angle ψ is Irs sin φ Pb tan φ tan ψ = − =− (10.26) Vc (1 + β) Pc (1 + β) If φ is negative, the required choice to obtain longitudinal focusing, then ψ must be positive, and δ must be negative. Negative δ means that the resonant frequency must be tuned higher than the generator and beam frequency. To minimize the generator power with respect to β, it is necessary to choose β, and therefore Qext , so that Irs cos φ Pb =1+ (10.27) β0 = 1 + Vc Pc The corresponding generator power is Pg = Pc + Pb , and by conservation of energy the reﬂected power becomes Pr = Pg − Pc − Pb = 0, so that this choice of β both minimizes the generator power, and results in zero reﬂected power. The detuning angle for the optimum coupling factor β0 is β0 − 1 tan ψ0 = − tan φ (10.28) β0 + 1

10.5 Equations Describing a Beam-Loaded Cavity

Extreme Beam-Loaded Limit

To treat the problem of a very strongly beam-loaded cavity, we consider the extreme beam-loaded case, where Pb Pc . Then we obtain β0 ∼ = Pb /Pc , Pg ∼ = Pb ∼ = tan φ/2QL . The = β0 Pc , tan ψ0 ∼ = − tan φ, or ψ0 ∼ = −φ, Vb ∼ = Vc , and δ ∼ phasor voltages become Irs V˜ b = cos φej(π−φ) (10.29) β 2Irs cos2 φ V˜ g = β

(10.30)

Irs (10.31) V˜ c = cos φejφ β If the beam goes away, the generator power is related to the cavity power according to (1 + β0 )2 ∼ β0 Pg = Pc (10.32) = Pc 4β0 cos2 ψ 4 cos2 ψ and the corresponding reﬂected power is β0 Pr ∼ (10.33) = Pc 4 cos2 ψ Consider the case where the cavity is driven on resonance so ψ = 0. If the cavity voltage is maintained at a constant level by feedback control, the formulas imply that the generator power is reduced by a factor of 4 when the beam goes off. This would also result in forward and reﬂected traveling-wave ﬁelds in the input guide that are half of the ﬁeld in the forward wave when the beam is on. The maximum standing-wave electric ﬁeld in the input guide when the beam is off will be the sum of the ﬁelds of the two traveling waves, which gives the same maximum ﬁeld as when beam is on. Thus, if the cavity ﬁeld is maintained constant when the beam is off, the peak ﬁeld in the waveguide is not increased, and the RF window in the waveguide is not subjected to higher ﬁelds. Numerical Example of a Beam-Loaded Cavity

Consider the example of a beam-loaded normal-conducting cavity with bunches injected earlier than the crest of the accelerating wave. Consider typical numerical values: f = ω/2π = 700 MHz, I = 0.1 A, rs = 25 × 106 , Vc = 2 × 106 V, and φ = −35◦ . We ﬁnd that Pc = Vc2 /rs = 1.6 × 105 W, and Pb = IVc cos φ = 1.6 × 105 W. Then we ﬁnd that β0 = 1 + Pb /Pc = 2, tan ψ = tan 35◦ /3, or ψ = 13.1◦ determines the detuning of the cavity required for minimum reﬂected power. With beam present, there is zero reﬂected power, and Pg =Pc + Pb = 3.2 × 105 W, Vb = Irs cos ψ/(1 + β) = 0.81 × 106 V, and Vg = 2 βrs Pg cos ψ/(1 + β) = 2.6 × 106 V. If QL = 104 , we ﬁnd δ = − tan ψ/(2QL ) = −1.2 × 10−5 , or the resonant frequency must be detuned by f0 − f ∼ = 8.1 kHz. The resonant frequency must be detuned

351

352

10 Beam Loading Im (V ) Re (V ) ∼ Vb

ψ ∼ Vg ∼ Vc

Figure 10.8 Voltage phasors for a steady-state, beam-loaded, normal-conducting cavity, when bunches are injected at a phase φ, earlier than the crest of the wave. Im(V ) Re(V ) ~ Vb

~ Vc

~ Vg

Figure 10.9 Voltage phasors for a steady-state, strongly beam-loaded superconducting cavity, when bunches are injected at the crest of the wave.

higher than the drive frequency. For completeness the cavity time constant is τ = 2QL /ω0 = 4.5 µs, and the phasor plot is shown in Fig. 10.8. Example of a Heavily Beam - Loaded Superconducting Cavity with Bunches Injected on the Crest of the Accelerating Wave

Consider a strongly beam-loaded superconducting cavity with the following parameter values; f = ω/2π = 700 MHz, I = 0.1 A, rs = 2.5 × 1012 , Vc = 5 × 106 V, and φ = 0◦ . We ﬁnd that Pc = Vc2 /rs = 10 W, and Pb = IVc cos φ = 5.0 × 105 W, so Pb Pc . Then we ﬁnd that β0 = Pb /Pc = 5 × 104 , ψ = −φ = 0◦ (no detuning is required), Pg = Pb = 500 kW, Vb = Vc = 5 × 106 V, and Vg = 2Vc cos2 φ = 10 × 106 V. The magnitude of the beam-induced voltage is equal to the net cavity voltage, and the two voltages have opposite sign, as shown in Fig. 10.9. If the beam goes off and the cavity voltage is kept constant, the generator power would be reduced by a factor of 4.

10.6 Generator Power when the Beam Current is Less than Design Value

When a cavity is matched to the generator, under steady-state conditions all the generator power goes into the beam power plus the cavity wall–power dissipation, and no net power ﬂows back from the cavity.

10.6 Generator Power when the Beam Current is Less than Design Value

If the coupler geometry is not adjustable, it is usually designed for a matched condition at the maximum expected beam current. The coupler provides the matched or reﬂection-free condition only for this maximum beam current; it will not be matched for any lower beam currents. An alternative approach is to use an adjustable coupler that is capable of providing a matched condition at any beam current of interest. The advantage of an adjustable coupler is obvious, except that adjustable couplers are technically more challenging and generally more expensive than ﬁxed couplers. To compare the two options of ﬁxed versus variable couplers, we consider the case of an accelerator driven on resonance in the extreme beam-loaded case where the cavity wall–power losses are small compared with the beam power. We assume the bunches are injected on the crest of the accelerating wave, and ψ = φ = 0. We begin with the case of a ﬁxed coupler. From Eq. 10.19 we obtain

(1 + β)2 Vb 2 (10.34) Pg = Pc 1+ 4β Vc where from Eq. (10.20) the beam voltage is Vb = Irs /(1 + β), the beam power for an arbitrary beam current I is Pb = IVc , and the cavity power is Pc = Vc2 /rs . The waveguide-to-cavity coupling factor is given by Eq. (10.27), and in the extreme beam-loaded case, where the beam power is much larger than the cavity power, is given by β ∼ = Pb,m /Pc 1, where Pb,m is the matched beam power corresponding to a beam current Im chosen for a matched reﬂectionfree condition at a ﬁxed beam power. Using these equations one can obtain Vb /Vc ∼ = Pb /Pb,m , and the generator power as a function of beam power Pb is given by Pb,m Pg ∼ = 4

Pb 2 1+ Pb,m

(10.35)

For the matched case Pb = Pb,m , we obtain Pg = Pb,m so all the generator power goes into the beam in this approximation that the cavity power is negligible. When Pb = 0 the required generator power is one quarter of the power required for the matched case. The reﬂected power as a function of the beam power can also be derived in the same beam-loaded limit where β 1. We obtain

Pb,m β Pb 2 Pb 2 + (10.36) 1+ − 1 − Pb Pr = β 4 Pb,m Pb,m When Pb = 0, the equation gives Pr ∼ = Pb,m /4, so that in the extreme beamloaded approximation the generator power all goes into net reﬂected power. Also, for the matched case where Pb = Pb,m , the net reﬂected power is zero, as expected.

353

10 Beam Loading 1 0.8 Pg/Pb,m

354

0.6 0.4 0.2 0.0 0.2

0.4 0.6 Pb/Pb, m

Figure 10.10 Generator power versus beam power at ﬁxed cavity voltage. Both the generator and the beam power are scaled to the matched beam power Pb = Pb,m . Points on the solid curve (Pg = Pb ) corresponds to

0.8

a coupler that is always adjusted for a matched condition. Points on the dashed curve (Eq. 10.35) correspond to a ﬁxed coupler with a matched condition when Pb = Pb,m .

For the case of an adjustable coupler, the generator power is assumed to be set so that it always equals the beam power; thus Pg = Pb and there is zero net reﬂected power for all beam currents. The two functions for Pg are plotted in Fig. 10.10. The dashed curve shows that at zero current, the ﬁxed coupling conﬁguration requires the generator to supply one quarter of the beam power of the matched case, nearly all of which is reﬂected. However, the ﬁgure illustrates that not much power is lost with a ﬁxed coupler when operating at beam currents that are lower than the design matched current.

10.7 Transient Turn-On of a Beam-Loaded Cavity

The turn-on of the RF input power and the beam in a standing-wave accelerating cavity should be done in the proper order. It is usually desirable to reach the design ﬁeld level in a relatively short time to avoid wasting power. With the beam off, the RF generator is turned on at a sufﬁcient input power level that the cavity ﬁeld approaches an asymptotic value that exceeds the design value. The higher the input RF power level, the shorter the time to reach the design cavity ﬁeld. When the design cavity ﬁeld level is reached, the beam can be injected, provided that the ﬁeld is not still increasing. Different methods can be used to program the RF input power as a function of time. A simple method is based on the result that the cavity time constant and QL do not change when beam is turned on. Thus, the responses of the cavity to the RF generator and to the beam are both exponential functions with exactly the

10.7 Transient Turn-On of a Beam-Loaded Cavity

same time constant, τ = 2QL /ω, the cavity time constant when the cavity is loaded by the input power coupler. The parameters can be adjusted so that if the beam is turned on at the instant when the cavity reaches the design ﬁeld level, the ﬁeld remains constant at that value. Whether the input power from the generator needs to be adjusted when beam is injected is a matter of choice. Usually, the coupling is chosen so that the input waveguide is matched to the cavity with the beam on. With the beam off, some of the input power from the generator will generally be reﬂected from the input power coupler and some will be transmitted into the cavity. We describe a simple method, where the input power P+ from the generator is constant during the entire transient turn-on time. With no beam, the steady-state power dissipated in the cavity is Pcn = P+ (4β)/(1 + β)2 , the steady-state stored energy is Un = Q0 Pcn /ω = (Q0 P+ /ω)[(4β)/(1 + β)2 ], and the time-dependent cavity voltage, before the beam is injected, is √ V(t) = Vn (1 − e−t/τ ), where Vn ∝ Un . With beam on, and zero reﬂected power, P+ = Pcw + Pb = Pcw (1 + Pb /Pcw ), where Pcw is the steady-state power dissipated in the cavity, and Pb is the beam power. From Section 10.5, the coupling factor required for no reﬂected power with beam, is β = 1 + Pb /Pcw . Thus, the steady-state stored energy is Uw = Q√ 0 Pcw /ω = Q0 P+ /ωβ, and the steady-state cavity voltage with beam is Vw ∝ Uw . The time when beam injection should begin is when V(t) reaches the design voltage level of Vw, or tb = −τ ln(1 − Vw /Vn )

(10.37)

An example of the cavity voltage as a function of time for the case Pb /Pcw = 0.5 is shown in Fig. 10.11. Problems 10.5 and 10.6 have been chosen to illustrate the method.

Vn Vw 0.8 Steady state

0.6

V (t )

Beam turn on

0.4

0.2

RF turn on

t /t

Figure 10.11 Cavity voltage versus time, showing RF power and beam turned on.

355

356

10 Beam Loading

Problems [6]

10.1. Consider the following model shown in Fig. P10.1 of the beam-cavity interaction for a relativistic charge that crosses an initially unexcited cavity. The RF cavity is replaced by a plane, parallel-plate capacitor of gap g. The charge q is represented by an inﬁnitesimally thin, uniformly charged disk of radius a, which enters the gap through one capacitor plate at z = 0, crosses the gap moving at velocity v = c, and exits through the opposite plate. Each plate is a thin metallic foil through which the charge can pass. All electric ﬁelds are assumed to be parallel to the axial z direction. If g λ the charged disk crosses the gap in a time short compared to the RF period of the fundamental cavity mode, so that the problem is approximately electrostatic. (a) When the disk enters the gap, it induces images charges in the capacitor plates. To produce a net zero ﬁeld ahead of the charge q, as is required by causality (when the charge enters the gap its ﬁelds cannot move forward any faster than the velocity v = c of the charge itself), we assume that a uniform image charge distribution with total charge −q is induced at the entrance plane z = 0. Show from Gauss’s law that for any position of the charge q in the gap the induced electric ﬁeld within the gap is given by: E = −q/πε0 a2 (corresponding to a wakeﬁeld) behind the charge q and E = 0 ahead of the charge. Plot the ﬁeld as a function of z when the disk is at zd = g/2. (b) Assume the equation of motion for the moving disk is z = ct, where t = 0 is the time when the disk enters the gap. Calculate the stored electric energy UE (t) in the cavity as a function of time. Equate the total disk energy loss to the stored electric energy UE (t = g/c) left in the gap when the charge leaves the gap. From UE (t = g/c) what is the total effective retarding voltage Vq that has acted on the disk? What do you think the image charges are at the z = 0 and z = g conducting planes immediately after the disk leaves

−q

v=c

E=0

z a

z=0

Figure P10.1

z = ct

For problem 10.1.

z=g

Problems [6]

the gap. (c) Using the induced ﬁeld behind the charge from part a, calculate the retarding voltage Vb induced in the gap after the disk has left the gap. Show that Vq /Vb = 1/2, or the particle sees only half the voltage that it induces. This result is known as the fundamental theorem of beam loading. 10.2. To gain a better understanding for the fundamental beam-loading theorem, we replace the thin relativistic disk of Problem 10.1 with a relativistic, ﬁnite-thickness, uniform-density cylinder of charge q, length d, and radius a (see Fig. P10.2). Assume that the image charge −q is induced at the entrance plane as in Problem 10.1, when the cylinder of charge has entered the gap. Suppose the motion begins at t = −d/c with the head of the cylinder at the entrance plane z = 0, and describe the motion of the tail by zt = ct, and the head by zh = ct + d. At t = g/c the tail is at z = g and the cylinder has left the gap. Let the variable s, where 0 ≤ s ≤ d, represent an arbitrary position within the cylinder relative to the head of the cylinder. Thus s = 0 is at the head and s = d is at the tail. (a) Use Gauss’s law to show that for any position of the cylinder, when it is completely inside the gap, the induced electric ﬁeld is (1) behind the cylinder (wakeﬁeld), E = −q/πε0 a2 , 0 ≤ z ≤ zt , (2) within the cylinder at a point s = zh − z, E(s) = −qs/πε0 a2 d, 0 ≤ s ≤ d or zt ≤ z ≤ zh , and (3) from causality ahead of the cylinder E = 0, z ≥ zh . Plot the ﬁeld as a function of z at t = 0, t = g/2c, and t = g/c. (It can be shown that the ﬁeld at s within the cylinder is constant for all positions of s within the gap, even when the cylinder is entering through the z = 0 plane and leaving through the z = g plane.). (b) Using the result of part a, show that after the cylinder has moved completely through the gap, the total voltage seen by a point at s inside the cylinder is V(s) = −qsg/πε0 a2 d. (c) Multiplying the voltage V(s) by the inﬁnitesimal charge dq = q ds/d at s, and integrating over s −q

v=c

E=0 z a

z=0

Figure P10.2

For problem 10.2.

z=g

357

358

10 Beam Loading

s gives the total energy loss of the cylinder. Show that this energy loss of the cylinder to the gap ﬁelds is W = −q2 g/2πε0 a2 . Calculate from W the total effective retarding voltage Vq that is experienced by the cylindrical charge. (d) Another way to calculate W is to calculate the ﬁnal stored electric energy UE in the initially unexcited gap from the ﬁeld induced by the cylinder of charge. This must equal the energy lost by the cylinder. Show that this does give the same result as part (c). (e) Using the induced ﬁeld behind the charge given in part (a), calculate the retarding voltage Vb (t) induced in the gap over the time interval t. Show that Vq (t)/Vb (t) = 1/2 as was found in the previous problem. (f) Why are not the voltages Vq and Vb equal? (Hint: Explain by using the result that for a charge at position s in the cylinder, the voltage it sees as it crosses the gap is V(s) = −qsg/πε0 a2 d. Examine the behavior as a function of s. The plots from part a may also be helpful to visualize this.) If the cylinder represented a bunch of initially monoenergetic particles, would the gap induce a signiﬁcant energy spread in the bunch? 10.3. Suppose the capacitor gap of Problem 10.1 is charged with an axial accelerating ﬁeld E0 and voltage V0 = E0 g before a relativistic charged disk with charge q crosses the z = 0 conducting plane. Assume that the image charge is induced on the z = 0 plane just as in Problem 10.1, and that the induced ﬁelds are superimposed on the preexisting accelerating ﬁeld. (a) What is the initial stored electric energy within a cylinder of radius a before the disk enters. (b) After the disk crosses the gap, what is the stored energy? From energy conservation what is the energy change for the disk? Express the total voltage seen by the disk as the sum of the applied voltage V0 plus the induced voltage seen by the disk Vq . Deﬁne the beam-loading parameter µ = −Vb /V0 . Will the disk always gain energy and if not, what condition on µ must be satisﬁed for a net energy gain to the disk? (c) Express the total voltage after the disk has crossed the gap as the sum of the applied voltage V0 plus the induced voltage Vb . Is Vq still equal to half the induced voltage Vb as was true when the cavity was initially unexcited? (d) Deﬁne an energy extraction efﬁciency η as the fraction of the initial cavity energy that is extracted by the disk? Show that η = 2µ − µ2 . For what value of µ is the energy extraction maximum and what is η for this case? Can the charged disk remove all the initial gap stored energy? (e) If the disk is replaced with a ﬁnite-length cylinder, one can show the head of the cylinder sees zero induced voltage and the tail sees the full induced voltage Vb . The total voltage seen by the head and the tail includes the applied voltage V0 . The average voltage seen by the cylinder is the same as for the disk: V0 + Vq . Deﬁne the ratio δ as the full energy spread relative to the total energy gain of the cylinder as a function of the beam-loading parameter µ? Plot η and δ as functions of µ in

Problems [6]

the range 0 < µ < 2, where the beam extracts energy from the gap. As µ increases toward 1 to approach 100% transfer of energy from the gap to the beam, what happens to the energy spread parameter δ ? (f) For some typical numbers for a bunch in an electron linac, q = 5 × 10−9 C, a = 3 cm, and E0 = V0 /g = 20 MV/m, calculate µ, η, and δ. If the cavity length is g = 3 m, how much energy per bunch is removed from the cavity in joules? If the beam energy is 1 GeV, what is the induced energy spread as a percent of the beam energy? 10.4. Consider a beam-loaded superconducting cavity in a proton linac with the following parameter values; drive frequency f = ω/2π = 700 MHz, beam current I = 0.1 A, cavity shunt impedance rs = 2.5 × 1012 , quality factor loaded by the external circuit QL = 3 × 105 , synchronous phase φ = −35◦ , and cavity voltage (includes the transit-time factor) Vc = 5 × 106 V (see Fig P10.4). (a) Show that the power delivered to the beam is much larger than the cavity wall-power loss. (b) Calculate the detuning angle ψ, and the waveguide-to-cavity coupling factor β0 required to yield zero reﬂected power when the beam is on. How much and in what direction must the cavity resonant frequency be detuned with respect the driving frequency? (c) Assuming β = β0 and the detuning angle ψ from part (b), calculate the voltages induced by the beam, and the generator, and the required generator power. Conﬁrm that the magnitude of the beam-induced voltage is approximately equal to the magnitude of the net cavity voltage, and that the placement of vectors in Fig. P10.4 is qualitatively correct. (d) Suppose that the generator goes off while the beam remains on. Calculate the cavity voltage, the cavity power, and the power emitted from the cavity through the coupler after a steady state is reestablished. (e) Suppose that the voltage amplitude is ﬁxed by feedback control, and that the generator goes off while the beam is on. Calculate the generator power and the reﬂected power after a steady state is reestablished. 10.5. Consider the transient turn-on of a linac, designed with a ratio of beam power to structure-power loss equal to 0.50. Assume that the generator power is held constant during the entire turn-on. (a) Show that to maintain a constant ﬁeld value after the beam is injected, the beam should be injected at the approximate time t = 1.79τ , where τ is the Im (V ) Re(V ) ψ = −f

∼ Vb

Figure P10.4

∼ Vg

∼ Vc

For problem 10.4.

359

360

10 Beam Loading

cavity time constant (see Fig. 10.11). (b) Show that for this case, the steady-state ﬁeld level with beam on is about 83% of the asymptotic value with beam off. 10.6. A typical operating mode for a standing-wave electron linac is to inject the beam at the crest of the accelerating waveform, to drive the accelerating cavities at the resonant frequency, and to choose the coupling factor β for zero reﬂected power when the beam is on and at the design current I. For this case, the cavity voltages induced by the generator and by the beam are 180◦ out of phase. (a) Refer to the results of Sections 5.6 and 10.5 to conﬁrm that the cavity voltage during the turn-on transient has the form 1/2

Irs 2β V(t) = P+ rs (1 − e−t/τ ) − (1 − e−(t−tb )/τ ) 1+β 1+β where P+ is the constant value of the forward power from the generator, t = 0 is the time that the RF power is turned on, and t = tb is the time that the beam is injected into the cavity. Thus, I = 0 for t < tb , and the current has the constant value I for t > tb . The ﬁrst term in the expression for the cavity voltage is the generator-induced voltage, and the second term is the beamcurrent-induced voltage. Both terms have the same time constant τ = 2QL /ω. (b) Show that V(t) is constant for t > tb (see Fig. 10.11), when tb is

β −1 Irs = −τ ln tb = −τ ln √ 2β 2β 1/2 P+ rs where β is chosen to give zero reﬂected power when the beam is on and at the design-current value.

References 1 Wilson P.B., High energy electron

4 For a discussion of beam loading in

linacs: applications to storage ring RF systems and linear colliders, 1981 summer school on high energy accelerators, AIP Conf. Proc. 87, 450 (1981); Perry Wilson, SLAC-PUB-2884 (Revised) November, 1991. 2 Halbach K. and Holsinger R.F., SUPERFISH-a computer program for evaluation of Rf cavities with cylindrical symmetry, Part. Accel. 7, 213–222 (1976). 3 Wilson P., Wake ﬁelds and wake potentials, AIP Conf. Proc. 184, 526–564 (1989).

traveling-wave linacs see Leiss, J.E., in Linear Accelerators, Lapostolle P.M. and Septier A.L., John Wiley & Sons, New York, 1970, pp. 147–172. 5 Wilson P., 1981 summer school on high energy accelerators, ATP Conf. Proc. 87, 450 (1981); P. Wilson, SLAC-PUB-2884 (revised), November 1981. 6 Problems 10.1–10.3 are based on an unpublished wakeﬁeld model by J. Lawson.

361

11 Wakeﬁelds

The electric and magnetic ﬁelds of the bunch are modiﬁed by the conducting walls of the surrounding structure. In general, the ﬁelds carried along with the bunch induce surface charges and currents in the walls, and these become the sources of ﬁelds that act back on the particles in the bunch. At low velocities, the ﬁelds from the beam-induced surface charges are described by equivalent beam-induced image charges, and the effects of these ﬁelds on the beam are usually much smaller than the direct space-charge ﬁelds. At beam velocities near the speed of light the direct space-charge forces decrease because of the near cancellation of the electric and magnetic self ﬁelds. It is well known that the electric-ﬁeld distribution of a free relativistic charged particle is Lorentz contracted into a disk perpendicular to the direction of motion with a narrow angular spread of the order of 1/γ , and the longitudinal electric-ﬁeld component approaches zero. The ﬁeld compression is also present when the moving particles are placed within a smooth perfectly conducting pipe. However, scattered electromagnetic radiation is produced when the ﬁelds moving with the relativistic particles encounter geometric variations along the structure, such as radiofrequency (RF) accelerating cavities, vacuum bellows, and beam-diagnostic chambers. In the extreme relativistic limit, the scattered radiation cannot catch up to affect the source particles, but the radiation can and does act on trailing particles in the same or subsequent bunches. As is the convention for relativistic electron beams, the scattered radiation will be called wakeﬁelds, although strictly speaking for particles with v < c, the scattered radiation does not always travel behind the particles that are its source. The longitudinal compression of the ﬁelds traveling with the beam is an important effect that tends to increase the energy in the wakeﬁelds for short bunches, because of the constructive interference of the radiation from different induced surface charges. At lower velocities the ﬁeld carried by the beam becomes more isotropic, and when the induced surface charges radiate over a length scale that is comparable to or greater than the wavelength of a particular mode, destructive interference from the different sources of the radiation reduces the energy in the corresponding wakeﬁelds. RF Linear Accelerators. 2nd, completely revised and enlarged edition. Thomas P. Wangler Copyright  2008 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-40680-7

362

11 Wakeﬁelds

Wakeﬁelds generally exist as damped, oscillatory electromagnetic disturbances, which can be described equivalently as a sum of all the resonant modes excited within the structure. Wakeﬁelds include the higher-order modes and the beam-excited accelerating mode, whose main effect is usually described separately as beam loading (see Chapter 10). The highest frequency components comprising the wakeﬁelds do not remain localized, and for frequencies higher than the cutoff frequency of the pipe, propagate away along the beam pipe. The lower frequency modes, below the cut off frequency of the pipe, remain localized near the structures in which they were born, and may linger, depending on the time constants of the individual modes to act on particles in trailing bunches. The radiated ﬁelds are sometimes classiﬁed as either short-range or longrange wakeﬁelds. For ultrarelativistic bunches, the short-range wakeﬁelds generated by the particles at the head of the bunch affect trailing particles in the same bunch, causing energy loss, and for off-axis bunches, a transverse deﬂection of the particles in the tail. The most serious effect of the long-range wakeﬁelds is caused by the high-Q transverse deﬂecting modes that induce time-varying transverse deﬂections in trailing bunches. If the deﬂecting modes are strongly excited by the beam, they may cause the beam-breakup (BBU) instability, leading to effective emittance growth when averaged over time. Eventually the beam is lost on the walls. The wakeﬁelds usually extract only a small fraction of the beam energy, which can be replaced by additional RF power. The wakeﬁeld power extracted from the beam ultimately represents additional power dissipation, through Ohmic losses in the structure walls or power delivered to some external load. The wakeﬁelds can add signiﬁcantly to the refrigeration load in a superconducting linac, unless the power is coupled out to a load at higher temperature, by a specially designed higher-order mode coupler. This chapter is intended as an introduction to all those effects caused by the interaction of the beam with the surrounding walls of the accelerating structure. The reader who wishes to delve more deeply into this topic is referred to the book by A. Chao [1]. We begin the chapter with a description of image-charge forces induced by low-velocity beams in a smooth, round beam pipe. Then, we introduce the basic concepts for describing wakeﬁelds produced by high-current beams, and the general framework for describing the wakeﬁeld interaction with the beam, including wake potentials and beamcoupling impedances. Methods used to mitigate the effects of the wakeﬁelds are also discussed. Finally we introduce the topic of the BBU instability in linacs.

11.1 Image Force for Line Charge in Round Pipe

As an approximation for a long bunch, we consider a round, continuous, uniform-density beam with charge per unit length λ and radius R, inside a

11.1 Image Force for Line Charge in Round Pipe y

l −l

R x1

b 2 / x1

Figure 11.1 Conducting cylindrical pipe of radius b with image charge −λ induced by a beam with radius R and line density λ.

perfectly conducting cylindrical pipe of radius b. The geometry is shown in Fig. 11.1. Suppose that the beam centroid is displaced from the x-axis by an amount x1 . Charge will be induced on the surface of the pipe as required to make the pipe an equipotential surface. The electrostatic solution for the ﬁelds, valid for R b, is obtained placing an equivalent line-charge density −λ, the image charge, on the x-axis at x = b2 /x1 . The x component of the electric ﬁeld from the image charge is −λ(x − b2 /x1 ) x12 λx12 (x − x1 ) λx1 ∼ 1 + + Eim, x = = 2πε0 [(x − b2 /x1 )2 + y2 ] 2πε0 b2 b2 2πε0 b4 (11.1) and the y component is Eim, y =

−λx12 y −λy ∼ = 2πε0 [(x − b2 /x1 )2 + y2 ] 2πε0 b4

(11.2)

where we have assumed that (xx1 /b2 )2 1, and (yx1 /b2 )2 1, and have kept terms up the third order. Equation (11.1) contains the deﬂecting ﬁeld with both linear and cubic terms in the displacement x1 , and the defocusing term proportional to x − x1 . Equation (11.2) has a focusing term proportional to y. The space-charge electric ﬁeld inside the beam is Esc, x =

λ(x − x1 ) 2πε0 R2

(11.3)

Esc, y =

λy 2πε0 R2

(11.4)

and

Adding the image ﬁeld to the space-charge ﬁeld, and representing the focusing force by a smooth linear restoring force with wave number k0 , we

363

364

11 Wakeﬁelds

express the equation of motion for the centroid displacement as d2 x1 x1 x12 2 + k0 x1 − K 2 1 + 2 = 0 dz2 b b

(11.5)

where K = qI/2πε0 γ 3 mv3 is the generalized perveance, expressed in terms of the current I = λv, a factor of γ 2 to account for the magnetic forces, and a factor γ for the relativistic mass factor. Equation (11.5) describes a coherent oscillation that depends on x1 /b. The focusing in both x and y are described by the following equations of motion, K d2 (x − x1 ) x12 R2 2 + k0 (x − x1 ) − 2 1 + 4 (11.6) (x − x1 ) = 0 dz2 R b and K d2 y + k20 y − 2 dz2 R

1−

x12 R2 b4

y=0

(11.7)

The effect of the image charge on the focusing, compared with the direct space-charge force, depends on the square of the parameter x1 R/b2 . For a numerical example, suppose x1 = b/10, and R = b/2. Then, x1 R/b2 = 1/20, and x12 R2 /b4 = 1/400, which produces a very small correction compared with the space-charge force. Only if the beam displacement is large enough to put the edge of the beam close to the pipe do the transverse image-charge corrections in a smooth pipe become important in a linac. Generally, image effects decrease the transverse defocusing strength, but increase the longitudinal focusing strength. Image forces have been identiﬁed as important for linac beams that debunch in long high-energy beam-transport lines [2]. Additional discussion of transverse image effects and a thorough treatment of longitudinal image effects are given by Reiser. [3]

11.2 Fields from a Relativistic Point Charge and Introduction to Wakeﬁelds

We consider the electromagnetic ﬁelds in free space from a point charge q with velocity v = βc moving along the z-axis [4]. In spherical coordinates r, ψ, where r is the radius from the charge to the ﬁeld observation point, and ψ is the angle between the unit vector rˆ and the z-axis, the electric ﬁeld is E=

4πε0

r 2 γ 2 (1

qˆr − β 2 sin2 ψ)3/2

(11.8)

where rˆ points from the charge toward the observation point. The ﬁeld is directed radially, but when β ≈ 0, the magnitude becomes concentrated into the equatorial plane, as shown in Fig. 11.2. When ψ = 0 or π, corresponding

11.2 Fields from a Relativistic Point Charge and Introduction to Wakeﬁelds

v=0

v≈c

Figure 11.2 Illustration of the concentration of the electric ﬁeld of a moving charge into the equatorial plane.

to the longitudinal ﬁeld component, we have E = q/4πε0 r 2 γ 2 , and when ψ = π/2, corresponding to the transverse component, we have E = qγ /4πε0 r 2 . Relative to the ﬁeld of a charge at rest, the longitudinal component is reduced by a factor of γ −2 , and the transverse component is increased by a factor γ . In the ultrarelativistic limit [5] where the point charge q moves at the velocity of light c, we approximate λ(z–ct) = qδ(z–ct) where δ is the dirac δ-function. Then the radial electric ﬁeld, expressed in cylindrical coordinates is [6] Er =

qδ(z–ct) 2πε0 r

(11.9)

Ampere’s law can be applied, using a circular path centered on the charge at radius r, and writing the current as I = qcδ(z–ct), we obtain Bθ =

µ0 cqδ(z–ct) 2πr

(11.10)

Because all ﬁeld components are zero, ahead and behind the point charge, ultrarelativistic particles with different z coordinates cannot interact. The Lorentz force acting on a test charge q that is moving with velocity c at radius r with the same z coordinate as the original particle is Fr = q[Er − cBθ ] =

q qδ(z–ct) q qµ0 c2 δ(z–ct) − =0 2πε0 r 2πr

(11.11)

Therefore, in the ultrarelativistic limit there is no direct electromagnetic interaction between the particles in such a beam. The next question is how does the beam pipe affect the ﬁeld and the interaction of the two charges? If the source charge is centered in the pipe, one ﬁnds that the ﬁeld lines terminate on the pipe, but are otherwise unaffected. If the source charge is not on the axis of the pipe, the ﬁeld lines are distorted from their free-space conﬁguration, but are still compressed into the ﬂat disk, and the Lorentz force on any test charge is still zero. If the pipe has a ﬁnite conductivity, or if the pipe geometry varies, one ﬁnds that the passage of an ultrarelativistic charge is accompanied by the generation of electromagnetic ﬁelds behind the charge. The absence of ﬁeld ahead of

365

366

11 Wakeﬁelds

the charge is a consequence of causality, that is, no action occurring ahead of the v = c point charge can be causally related to it. For a linac the effect of ﬁnite resistivity is usually negligible compared with the effects of geometry variations [7]. It is convenient to think of the incident ﬁelds that move with the beam as inducing charges and currents in the conducting walls that radiate electromagnetic energy in directions other than that of the incident ﬁelds. The radiation cannot propagate any faster than the ultrarelativistic source charge, and therefore the radiation remain behind a moving ultrarelativistic source. One ﬁnds that these electromagnetic wakeﬁelds will often remain long after the passage of the source, and can affect trailing particles in the same bunch or in a trailing bunch. Figure 11.3 illustrates the wakeﬁelds excited by a bunch passing through a cavity. A frequency-dependent physical picture has been described by Chao [8]. The lowest-frequency bunch-induced excitations are associated with resonant excitation of the lowest cavity modes by the beam. Wakeﬁelds with ω < c/a, where a is the beam-pipe radius, are generated by scattering into each cavity the incident electromagnetic wave from the entrance edge of the cavity. After the bunch leaves each cavity, the wakeﬁelds at radius larger than the beam aperture are scraped off by the exit edges and remain behind. An estimate of the radiated electromagnetic energy trapped in the cavity is obtained by calculating the ﬁeld energy contained within an annular ring of length equal to the bunch length, and of radial thickness extending from the beam-aperture radius a to an effective outer cavity radius, assumed to be limited at 2a. An equal amount of energy from the high-frequency component of the incident ﬁelds with ω > c/a is diffracted into the beam pipes by the edges of the apertures, and these wakeﬁelds propagate along the pipe, because they are above the cutoff frequency of the pipe.

Figure 11.3 Numerically calculated electric wakeﬁelds generated by an ultrarelativistic Gaussian line-charge bunch in a PETRA cavity at DESY [9]. The beam moves along the dashed line on an axis of cylindrical symmetry. [From T. Weiland and R. Wanzenberg, Frontiers of Particla Beams: Intensity Limitations, Proc. Joint U.S.-CERN School on Particle Accelerators at Hilton Head Island, Lecture Notes Phys. 400, M. Dienes, M. Month, and S. Turner eds., 39–79 (1992). Copyright 1990 Springer–Verlag.]

11.3 Wake Potential from a Relativistic Point Charge

Consider an ultrarelativistic point charge moving through a linac, which contains typical structure-geometry variations associated with the RF cavities, beam pipes, bellows, diagnostic boxes, and so on. After the charge passes a perturbing element and induces surface charges that radiate electromagnetic energy, a local concentration of the electromagnetic ﬁelds may remain that exerts a force on trailing charges. We are interested in the integrated effect of these ﬁelds on trailing charges, as they pass through the same perturbing region. It is convenient to deﬁne a wake potential that characterizes the net impulse delivered to the trailing charges that are moving at the same velocity along the same or parallel paths. Assume that the source point charge q1 is moving parallel to a central axis of symmetry, but is displaced by an amount r1 , as shown in Fig. 11.4. Imagine that the source charge moves according to z1 = ct, and imagine a test charge q at a displacement r and at a ﬁxed distance s behind the source charge, that moves according to z = ct–s. We deﬁne the δ-function longitudinal and transverse wake potentials [10, 11] per unit source charge as 1 L wz (r, r1 , s) = − dz [Ez (r, z, t)]t = (z + s)/c (11.12) q1 0 1 L dz [E⊥ + c(ˆz × B)]t = (z + s)/c (11.13) w⊥ (r, r1 , s) = q1 0 The interval L must be large enough to include the entire ﬁeld that a test particle would see. Equations (11.12) and (11.13) are useful as Green’s functions from which the wake potentials for any arbitrary charge distribution can be calculated. Each wake potential determines the momentum kick delivered to a trailing test charge. Thus, for the longitudinal momentum

z = ct −s

z = ct r1

s L

Figure 11.4 An ultrarelativistic source charge q1 passes through a geometry perturbation and induces wakeﬁelds that extend over a length L, followed by an ultrarelativistic test charge q.

367

368

11 Wakeﬁelds

change pz or energy change U experienced by the charge q, we have c pz (r, r1 , s) = U(r, r1 , s) = −qq1 wz (r, r1 , s)

(11.14)

and for the transverse momentum change, we have c p⊥ (r, r1 , s) = qq1 w⊥ (r, r1 , s)

(11.15)

The corresponding ith component of the average wake force is Fi (r, r1 , s) = pi (r, r1 , s)c/L

(11.16)

Although the wakeﬁelds produced by a given geometric perturbation are complicated functions of position and time, the introduction of wake potentials helps to simplify the problem by providing time-independent functions of position in the bunch, which determine the net momentum impulse at that point. The δ-function wake potential is a characteristic property of the geometric perturbation, and as we will see, can be used as a Green’s function to determine the total wake potential of a ﬁxed arbitrary charge distribution.

11.4 Wake Potentials in Cylindrically Symmetric Structures

We now restrict the discussion to an ultrarelativistic beam in a loss-free cylindrically symmetric structure, which we will describe using cylindrical polar coordinates. We assume that the source charge is displaced at radius r1 along the x-axis where θ = 0, and a test charge is located at the coordinate (r, θ). The radial coordinates will be expressed as a fraction of the aperture radius of the beam tubes, which will be denoted by a. The structure itself will have modes that vary azimuthally as e jmθ , where m is an integer. For example, the m = 0, 1, and 2 modes are the monopole, dipole, and quadrupole modes, respectively. The total m-pole component of the longitudinal wake potential can be expressed as a separable function of the coordinates and as a sum over the longitudinal modes n of the particular structure as [12] wzm =

r m r m 1

cos mθ

∞ n=0

2kmn (r = a) cos

ωmn s c

s>0

(11.17)

Likewise, the transverse wake potential, which is nonzero for m > 0, is r m r m−1 1 w⊥m = m a a ∞ 2k (r = a) s ω mn mn s>0 × rˆ cos mθ –− θˆ sin mθ sin ωmn a/c c n=0 (11.18)

11.4 Wake Potentials in Cylindrically Symmetric Structures

These wake potentials are both zero for s < 0 from causality. The quantity kmn is the loss parameter for the mode, which was deﬁned in Chapter 10 as half the voltage per unit charge induced in an empty cavity immediately after a charge passes through the cavity. It is expressed in terms of the r/Q for a given mode as kmn = ωr/4Q0 . Thus, the loss parameter for any mode is a property of the structure. The total wake potentials are obtained by summing over m. If the radial displacements are small compared to a, the dominant terms will be the m = 0 monopole term for the longitudinal wake potential, and the m = 1 dipole term for the transverse wake potential. Then, the wake potentials are approximated by wz ∼ = wz0 =

∞

2k0n cos

n=0

ω0n s , c

s>0

(11.19)

and ∞

r1 2k1n ω1n s w⊥ ∼ x sin = w⊥1 = a n=0 ω1n a/c c

s>0

(11.20)

In this approximation, the longitudinal wake potential is independent of the transverse coordinates of both the source and test charges, and the transverse wake potential is linearly proportional to the source-charge radius, is independent of the transverse coordinates of the test charge, and is everywhere directed along the x-axis, the direction of the source displacement. Because each term in wz varies as cos (ωs/c), the wake potential immediately behind the point charge at s = 0+ is positive. From the deﬁnition of the wake potential given in Eq. (11.12), this means the wakeﬁeld is directed opposite the direction of motion, and test charges immediately behind the source charge are decelerated. The source charge itself at s = 0 sees half of the wake potential that is induced behind it (at s = 0+), as was discussed in Chapter 10. Each term in the transverse wake potential varies as sin (ωs/c). This implies that the transverse wake potential is zero immediately behind the source charge, and unlike the longitudinal case, the source charge does not see any of the transverse wake potential that it generates. As s increases, the sign of both wake potentials can change. Using the approximations given above, the situation can be summarized as follows. For the longitudinal wake potential   ∞ ω0n s     2k cos s > 0 (behind source) (11.21) 0n     c     n=0   ∞ wz (s) = (11.22) k0n s=0 (at source)         n=0       (11.23) 0 s < 0 (ahead of source)

369

370

11 Wakeﬁelds

and for the transverse wake potential  r1 ∞ 2k1n ω1n s  xˆ sin   c  a n=0 ω1n a/c  w⊥ (s) = 0      0

 s > 0     s = 0     s z, from the causality principle wz = 0 for these charges, because they cannot contribute to the wake at z. Therefore, the integrand is zero when z1 > z and we can replace the upper limit in the integral by inﬁnity. Thus we write ∞ 1 wz (z − z1 )ρ(z1 ) dz1 (11.40) Wz (z) = Qtot −∞ From Eq. 11.19, the δ-function longitudinal wake potential is jω0n (z−z1 ) ∞ e ω0n s 2k0n cos 2k0n Re wz (s) = = c c n=0 n=0 ∞

= Re

∞

2k0n e jω0n z/c e−jω0n z/c1

(11.41) (11.42)

n=0

Substituting Eq. (11.42) into Eq. (11.40), we obtain Wz (z) =

∞ ∞ 1 Re 2k0n ejω0n z/c e−jω0n z1 /c dz1 Qtot n=0 −∞

(11.43)

Suppose we choose the origin of the coordinates at the electrical center of the bunch, so ∞ z1 sin ω0n (11.44) ρ(z1 ) dz1 = 0 c −∞

11.8 Bunch Loss Parameters for a Gaussian Distribution

Then, we have Wz (z) =

1 Qtot

∞

2k0n cos(ω0n z/c)

n=0

∞ −∞

cos(ω0n z1 /c)ρ(z1 )dz1

(11.45) The bunch longitudinal wake potential in Eq. (11.45) for an arbitrary charge distribution has the same form as that of the δ-function longitudinal wake potential of Eq. (11.19), if we identify the bunch loss parameter as k0n ∞ cos(ω0n z1 /c)ρ(z1 ) dz1 (11.46) k0n,distr = Qtot −∞ One can write the same expression for the transverse bunch loss parameter. 11.8 Bunch Loss Parameters for a Gaussian Distribution

The Gaussian distribution is a good approximation for the distribution of a typical relativistic electron beam. Consequently, it is of interest to determine the bunch loss parameters for the Gaussian case. The line - charge density is Qtot −z2 /2σ 2 e 1 ρ(z1 ) = √ 2π σ

(11.47)

This is an even function and using Eq. (11.46), the Gaussian loss parameter is √ 2k0n ∞ 2 2 k0n,G = √ cos[ω0n z1 /c]e−Z1 /2σ dz1 (11.48) πσ 0 Evaluating the integral, one obtains k0n,G = k0n e−ω0n σ 2

/2c2

(11.49)

The exponential factor is called the Gaussian form factor. Equation (11.49) implies that the bunch loss parameter is reduced when the rms bunch length becomes a signiﬁcant fraction of the wavelength for that mode. Physically, a ﬁnite bunch length reduces the loss factor because the wakeﬁelds no longer add in phase. The total bunch loss parameter for the Gaussian beam is ktot =

k0n e−ω0n σ 2

/2c2

(11.50)

Figure 11.9 shows ktot for a Gaussian bunch versus the rms length σz . For the typical value of σz = 1 mm, ktot is reduced by the Gaussian form factor by nearly a factor of 2 compared with the σz = 0 case. One can compute the total energy loss of the particles in the bunch to the wakeﬁelds from Fig. 11.9 at σz = 1 mm, we have ktot = 2.1 V/pC per cell for

377

11 Wakeﬁelds 4 3

k tot (V / pC)

378

0 0

2 sz (mm)

Figure 11.9 Total loss parameter ktot versus rms bunch length from Ref. [10] for one cell of the SLAC linac structure. [Reprinted with permission from K. L. F. Bane, AIP Conf. Proc. 153, 972–1014 (1987). Copyright 1987 American Institute of Physics.]

the SLAC linac. We assume the number of particles per bunch is N = 5 × 1010 , and we calculate the number of cells from the ratio of the total length of 3000 m to the cell length of 3.5 cm. The average energy loss per electron for the whole linac is

U = e2 Nktot Ncells N

(11.51)

Substituting the numbers gives U/N = 1.44 GeV per electron, which is approximately 3% of the nominal ﬁnal energy of 50 GeV. The greater concern associated with the wakeﬁelds is not this energy loss, which is not very large, but the loss of beam quality in terms of increased energy spread and transverse emittance growth, as we will discuss later in the chapter.

11.9 Beam-Coupling Impedance

We have described the interaction between a beam and its environment in terms of wakeﬁelds and wake potentials, which are functions of distance or equivalently of the time behind a moving charge or distribution of charges. Description in the time domain is convenient for short times, but for longer times the wake potential rings at perhaps one or just a few discrete frequencies. For these problems it is more convenient to describe the problem in the frequency domain. When we studied the problem in the time domain, we found it convenient to treat the basic problem of the response of the beam environment to a δ-function source. We take a similar approach in developing a frequency-domain picture, and consider the response of the system to a single frequency source.

11.9 Beam-Coupling Impedance

The bunch longitudinal wake potential in terms of the δ-function wake potential for the ultrarelativistic case is Wz (z) =

1 Qtot

∞

wz (z)ρ(z − s) ds

(11.52)

Because wz (s) = 0 for s < 0, we can let the lower limit to the integral extend to minus inﬁnity. Introducing the source current I = ρc, Eq. (11.52) becomes Wz (z) =

1 Qtot c

∞ −∞

wz (s)I(z − s) ds

(11.53)

Now consider a source current consisting of a single frequency ω, so I(z1 ) = I0 ejωz1 /c , where z1 + s = z. Note that with positions measured backwards, increasing positions correspond to increasing times of arrival, and we can replace position with time. We write I(z − s) = I0 ejω(z – s)/c = I0 ejωz/c e−jωs/c = I(Z)e−jωs/c

(11.54)

Then Wz (z) =

I(z) Qtot c

∞ −∞

wz (s)e−jωs/c ds

(11.55)

Equation (11.55) may be interpreted as follows. Wz (z) is the wake-potential response to a sinusoidal source current I(z). We now consider frequency as the independent variable and make the following identiﬁcations. The source current is I(ω) = I0 ejωz/c . The voltage response is V(ω) = Qtot Wz (ω). The remaining factor is called the beam-coupling impedance and is given by Zz (ω) =

1 V(ω) = I(ω) c

∞ −∞

wz (s)e−jωs/c ds

(11.56)

The impedance is the Fourier transform of the δ-function wake potential. Consequently, the impedance function is the frequency spectrum of the δfunction wake potential. A typical impedance spectrum is shown in Fig. 11.10. The beam pipe at the ends of cavity is a waveguide with a cutoff frequency of approximately c/a, where a is the radial aperture. Below the cutoff frequency, the impedance spectrum shows the sharp peaks corresponding to the lower frequency cavity modes. Above the cutoff frequency, the modes can propagate out through the beam pipe, which lowers their Q and creates an overlapping spectrum and a continuous broadband distribution. Recalling that the decay time of the modes is given by τ = 2Q/ω, we have either low-frequency, high-Q modes with a long decay time τ , or high-frequency, low-Q modes with a short decay time.

379

380

11 Wakeﬁelds

11.10 Longitudinal- and Transverse-Impedance Deﬁnitions

Just as we have longitudinal and transverse wake potentials to describe the time domain, we deﬁne corresponding longitudinal and transverse impedances for the frequency-domain description. The longitudinal relationships are 1 ∞ wz (x, y, s) e−jωs/c ds (11.57) Zz (x, y, ω) = c −∞ wz (x, y, s) =

1 2π

∞ −∞

Zz (x, y, ω) ejωs/c dω

The transverse relationships are 1 ∞ Z⊥ (x, y, ω) = w⊥ (x, y, s) e−jωs/c ds jc −∞ j w⊥ (x, y, s) = 2π

∞ −∞

Z⊥ (x, y, s) ejωs/c dω

(11.58)

(11.59)

(11.60)

Because of causality, the real and imaginary parts of the impedances are not independent. For example, one can show that 1 ∞ wz (s) = dω Re{Zz (ω)} cos(ωs/c) (11.61) π −∞ Furthermore, the Panofsky–Wenzel theorem relates the transverse and longitudinal impedances according to ω Z⊥ (x, y, ω) = ∇Zz (x, y, ω) c

(11.62)

Re(Zz) Fundamental mode

Higher modes

Beam pipe

Figure 11.10 Typical frequency spectrum of the real part of an accelerator cavity impedance. [From T. Weiland and R. Wanzenberg, Frontiers of Particle Beams: Intensity Limitations, Proc. Joint U.S.-CERN School on Particle Accelerators at Hilton Head Island, Lecture Notes Phys. 400, M. Dienes, M. Month, and S. Turner (Eds.), 39–79 (1992).

11.11 Impedance and Wake Potential for a Single Cavity Mode

We apply the results of Section 11.10 to calculate the impedance for a single mode of an oscillator, and use the Fourier transform to calculate the wake potential. Consider the shunt resonant circuit of Fig. 11.11, which was discussed in Section 5.1 as a model of an RF cavity. If we identify the driving current with the beam current, and the shunt voltage with the induced axial cavity voltage, we can identify the circuit impedance with the beam-coupling impedance. The circuit impedance for a driving frequency ω is Z(ω) =

R ω ω0 1 + jQ − ω0 ω

(11.63)

where Q = RCω0 , and ω0 = (LC)−1 . The accelerator shunt impedance r is related to the circuit resistance R by r = 2R. The circuit impedance can be written to show the real and imaginary parts more clearly, as ω0 ω − R 1 − jQ ω0 ω (11.64) Z(ω) = ω0 ω 2 2 1+Q − ω ω0 The cavity wake potential is the Fourier transform of this circuit impedance, which is ω0 R −t/τ cos ω0 1 − 1/Q 2 t w(t) = e Q 1 − sin ω0 1 − 1/Q 2 t (11.65) 2Q 1 − 1/4Q 2 where the decay time is τ = 2Q/ω0 . Equation 11.65 is valid for t > 0; for t < 0, w(t) = 0. The real and imaginary parts of the impedance are shown in Fig. 11.12 and the wake potential is shown in Fig. 11.13. The quantity ω0 R/Q is equal to 2k, where k = ω0 r/4Q is the loss parameter for the mode. We see

I cos(wt)

Figure 11.11 LRC circuit used as a model of a single mode of a cavity.

381

382

11 Wakeﬁelds Re(Z )

1 0.8 0.6 0.4

Im(Z )

0.2

−2

w/w0

−1

−0.2 −0.4

Figure 11.12 The real and imaginary parts of the impedance of a shunt-resonant-circuit cavity model.

w (t) / 2k 1 Q = 10 000

0.5

Q=3 Q=0

wt / 2p

−0.5

−1

Figure 11.13 Wake potential corresponding to a shunt resonant circuit as a model for the δ-function wake potential of the cavity.

that w(0) = 2k. When Q 1, the wake potential simpliﬁes to 1 w(t) = 2ke−t/τ cos[ω0 t] − sin[ω0 t] 2Q

(11.66)

For a cavity with no ohmic losses, Q approaches inﬁnity and w(t) = 2k cos[ω0 t]

(11.67)

11.13 Short-Range Wakeﬁelds: Energy Spread

This expression for the wake potential will be recognized as the same result as the δ-function longitudinal wake potential in Section 11.4. 11.12 Short-Range Wakeﬁelds-Parasitic Losses

The wakeﬁelds do work on the beam causing what is called parasitic energy loss. The lost beam energy can be restored by increasing the RF applied voltage. Wakeﬁeld energy is dissipated through ohmic losses in the walls of the structure. A question is whether the heating effects of the wakeﬁelds are sufﬁcient to cause concern. The energy lost to the wakeﬁelds per cell from an electron beam with N particles per bunch is U = (eN)2 ktot , and for a linac section of length L and cell length lc , the energy loss to wakeﬁelds is

U(L) = (eN)2 ktot L/lc . For a given bunch repetition rate fR , the average power lost to wakeﬁeld is P = U(L)fR =

fR (eN)2 ktot L lc

(11.68)

Consider parameters that are close to those of the SLAC linac; N = 5 × 1010 electrons per bunch, lc = 3.5 cm, fR = 120 Hz, ktot = 2 × 1012 V/C, and L = 1 m. We obtain P = 0.44 W/m. This is a small parasitic heat load for a copper structure. For a superconducting structure, a heat load of this magnitude is still not large but is more signiﬁcant. To reduce the load, one would need to reduce the total loss parameter ktot . As might be expected from the result that k = ω0 r/4U, cavities with nose cones, which have higher shunt impedance r, generate more wakeﬁelds, than those without nose cones. At this point we introduce a note of caution. The contributions to the transverse wake potential from geometric perturbations other than the cavities, such as the vacuum bellows, is not always negligible, which means that for an accurate estimate of energy loss to the wakeﬁelds, one must include all the components, not only the RF cavities. 11.13 Short-Range Wakeﬁelds: Energy Spread

Energy spread must be minimized for applications such as linear colliders or free electron lasers. For a simple analysis suppose we consider a longitudinal wake potential that increases approximately linearly with respect to distance s along the bunch. We approximate the bunch wake potential as Wz (s) = Wz s, and the energy loss per cell for an electron with charge e varies with s as

W(s) = −eqWz s. For cell length lc , we ﬁnd the energy loss over a length L from the approximate expression

W(s, L) = −

eqWz sL lc

(11.69)

383

384

11 Wakeﬁelds

If the particles at the head at s = 0 experience a negligible energy loss from the wakeﬁelds, the total energy spread across the beam caused by the wakeﬁelds generated by the head that act on the tail is δW(L) =

eqWz stail L lc

(11.70)

where stail is the bunch length. This energy spread can produce undesirable chromatic aberrations in the linac output beam. 11.14 Short-Range Wakeﬁelds: Compensation of Longitudinal Wake Effect

The other effect causing energy spread is the phase dependence of the RF accelerating ﬁeld. The magnitude and sign of this effect depends on the operating phase of the beam with respect to the RF ﬁeld. If the phase of the beam centroid is chosen earlier than the peak, then late particles will gain more energy, which is in just the right direction to compensate for the energy spread from the wakeﬁelds. To explore this point further, it is convenient to change variables in Eq. 11.70 from s to phase difference along the bunch, using

φ = ωs/c. The energy loss to the wakeﬁelds as a function of phase is

W( φ, L) = −

eq[Wz c/ω] φL lc

(11.71)

The quantity in brackets is the derivative of the longitudinal wake potential with respect to phase. We compare Eq. (11.71) with the energy-gain difference from the RF ﬁelds as a function of the phase difference, which is

Wrf = [eEL cos φ] ∼ = −eEL sin φ0 φ

(11.72)

For cancellation we must set WRF = W( φ, L), which is satisﬁed for a beam-centroid phase given by sin φ0 = −

q(Wz c/ω) Elc

(11.73)

The larger the derivative of the bunch wake potential, the larger |φ0 | must be for compensation, and the cost in terms of operating off the peak of the RF waveform increases. 11.15 Short-Range Wakeﬁelds: Single-Bunch Beam Breakup

An important effect in high-intensity electron linacs is called single-bunch BBU, which is caused by the transverse dipole mode. To illustrate the effect, we use a simple two-particle model for the bunch, shown in Fig. 11.14, in which two

11.15 Short-Range Wakeﬁelds: Single-Bunch Beam Breakup N /2 N /2 x2

Figure 11.14 Simple two-particle model for single-bunch beam breakup. Each macroparticle contains half the particles in the bunch.

ultrarelativistic macroparticles, each containing N/2 particles, are separated by a distance s. As a result of the external focusing forces, an off-axis macroparticle at the head of the bunch will execute transverse (betatron) oscillations about the beam axis. Invoking the smooth approximation, in which the transverse motion is approximated as a simple harmonic oscillation with an angular frequency ωβ , we write for the head macroparticle x1 = X cos ωβ t. The tail macroparticle would do the same as the head, if it were not for the transverse wakeﬁelds induced at the head that affect the tail. The average force Fw caused by the wakeﬁelds from a charge eN/2 at the head, acting over a cell of length lc on an electron with charge e in the tail, is obtained from

p⊥ c = Fw lc = e2

N w⊥ (s) 2

We write the wake force as e2 N w⊥ (s) x1 Fw = 2lc x1 /a a

(11.74)

(11.75)

The equation of motion for a particle in the tail in the approximation of a constant-energy beam is mγ x¨ 2 = −mγ ωβ2 x2 + Fw or x¨ 2 + ωβ2 x2 =

e2 N w⊥ X cos ωβ t 2mγ lc a x1 /a

(11.76)

(11.77)

This is the familiar equation of motion for an oscillator driven on resonance. Reminding ourselves that γ is assumed to be constant, we write the general solution for a particle in the tail, which has the same initial transverse coordinates as the particle at the head, as x2 (t) w⊥ e2 Nt (11.78) = cos ωβ t + sin ωβ t X 4mγ lc aωβ x1 /a The ﬁrst term is the free oscillation solution to the hom*ogeneous equation. The second term is a particular solution to the inhom*ogeneous equation that represents the response to the wake force. Note the factor t in the amplitude of the second term, which implies continuous growth, as shown in Fig. 11.15. If the wakeﬁelds were not present we would have w⊥ = 0 and would have

385

386

11 Wakeﬁelds X2(t )

Figure 11.15 Resonant buildup of the displacement of the tail relative to the head of the bunch.

only the ﬁrst term with unit amplitude. The amplitude of the second term, evaluated at the end of the linac, becomes a measure of the importance of the instability. Over a distance L, the amplitude of the tail displacement relative to that of the head is w⊥ e2 NLc A(s) = (11.79) 4(γ mc2 )lc aωβ x1 /a The betatron frequency can be written in terms of a betatron wavelength using the relation ωβ = 2πc/λβ . The energy of the beam particle, γ mc2 , was assumed to be constant. Extending the model over a larger energy range, the energy should be evaluated as an average value over the length L. Figure 11.16 shows the bunch shape, at a particular time during the oscillation, when the tail has swung away from the path of the head. The transverse emittance, averaged over the length of the bunch, grows as the tail amplitude increases. If the amplitude grows large enough, the tail particles will be lost on the walls of the structure. Eq. (11.79) shows how the amplitude depends on the linac and beam parameters, and because of the ωβ factor in the denominator, it shows that the growth of the amplitude can be reduced by stronger focusing. 11.16 Short-Range Wakeﬁelds: BNS Damping of Beam Breakup

A method of compensation to inhibit the single-bunch BBU effect, and its associated emittance growth is called BNS damping [22], named for Balakin, Tail

Head

Axis

Figure 11.16 Bunch shape caused during beam breakup by the action of the dipole wake potential from the head to the tail.

11.16 Short-Range Wakeﬁelds: BNS Damping of Beam Breakup

Novokhatsky, and Smirnov. To introduce this method, we return to the same two-particle model presented in Section 11.15, and consider the case where the betatron frequency of a particle in the tail is not exactly the same as at the head. We will discuss later how this difference can occur. Referring to Section 11.15, the equation of motion of a particle in the tail, located at a distance s from the charge eN/2 at the head, and driven by the wakeﬁelds from the head, is 2 x¨ 2 + ωβ2 x2 =

e2 N w⊥ (s) X cos ωβ1 t 2mγ lc a x1 /a

(11.80)

where ωβ1 and ωβ2 are the betatron frequencies at the head and tail, respectively, and the trajectory of the charge at the head is x1 = X cos ωβ1 t. Equation (11.80) is the familiar equation of motion of an undamped oscillator, driven off resonance, and if we begin with the condition that at t = 0, x2 (0) = x1 (0) and x˙ 2 (0) = x˙ 1 (0), and ignore acceleration, the general solution is x2 (t) = X cos ωβ2 t + where

F0 (s) (cos ωβ1 t − cos ωβ2 t) 2 2 ωβ2 − ωβ1

e2 NX w⊥ (s) F0 (s) = 2(γ2 m)lc a x1 /a

(11.81)

(11.82)

To control the emittance it is necessary to minimize the difference x2 − x1 , which in the two-particle model is F0 (s) (11.83) − X [cos ωβ1 t − cos ωβ2 t] x2 (t) − x1 (t) = 2 2 (ωβ2 − ωβ1 ) or

x2 (t) − x1 (t) = 2X

F0 (s) −1 2 2 (ωβ2 − ωβ1 )X

× sin[(ωβ2 + ωβ1 )t/2] sin[(ωβ2 − ωβ1 )t/2]

(11.84)

This difference corresponds to a beat pattern between the driving frequency ωβ1 , the betatron frequency of the head, and the natural frequency ωβ2 , which is the betatron frequency of the tail. The frequency difference will generally be small and the resulting beat pattern is shown in Fig. 11.17. The amplitude of this oscillation is F0 (s) A(s) = 2X −1 (11.85) 2 2 (ωβ2 − ωβ1 )X The amplitude can be made zero if choose ωβ2 > ωβ1 , and 2 (ωβ2

F0 (s) =1 2 − ωβ1 )X

(11.86)

387

388

11 Wakeﬁelds X2–X1

Figure 11.17 Beat pattern describing the difference x2 (t) − x1 (t) between head and the tail in the two-particle model when the betatron frequencies at the head and the tail are not equal.

Assuming that the betatron frequency difference ωβ = ωβ2 − ωβ1 is small, 2 2 ∼ − ωβ1 we have ωβ2 = 2ωβ1 ωβ . Then, the condition for cancellation implies a betatron frequency difference between tail and head of

ωβ =

F0 (s) 2ωβ1 X

(11.87)

A natural way to produce a betatron-frequency difference is to use the property of magnetic lenses that the focusing force increases at lower momentum, which is true physically because the magnetic rigidity of a particle in a magnetic ﬁeld is less at lower momentum. One can introduce a correlation between energy and axial position along the bunch, by choosing the synchronous phase later than the crest, so that the particles in the tail see a smaller accelerating ﬁeld than those at the head. In a relativistic electron linac, the absence of phase oscillations means that the particles in the tail remain in the tail, and have lower momentum throughout the linac. Using the smooth approximation results from Section 7.17 for a quadrupole focusing channel, we obtain the result ωβ ∝ γ −1 , where we assume β ∼ = 1. Therefore, ωβ /ωβ1 = − γ /γ1 , and the BNS condition in the lowest-order approximation becomes e2 N w⊥ (s) ∼ (11.88)

W = − 4lc a(ωβ1 /c)2 x1 /a where W = (γ2 − γ1 )mc2 . We can also write this as a fractional energy Thus difference, relative to the average energy W. e2 N w⊥ (s)

W ∼ (11.89) =− c a(ωβ1 /c)2 x1 /a W 4Wl What phase of the accelerating waveform is required to produce this energy difference? Assuming that the energy at any position in the linac z = L is given by W = eE cos(φ1 )z, the energy difference W relative to the head is

W = −eE sin φ1 L φ = −eE sin φ1 Lωs/c. Then ωs

W = − tan φ1 W c

(11.90)

11.17 Long-Range Wakeﬁelds and Multibunch Beam Breakup

Equating Eqs. (11.89) and (11.90) gives the result for BNS compensation, e2 Nc/ωs

W c ∼ w⊥ (s) tan φ1 = − (11.91) = ¯ c a(ωβ1 /c)2 x1 /a W ωs 4Wl The right side of the equation is positive, so the phase φ1 is positive, and the bunch must arrive later than the RF voltage peak, as expected. Recall that for compensation of the energy loss from the longitudinal wake potential at the tail, it was necessary for the bunch to arrive earlier than the peak. The beam-phase requirements for energy-loss compensation and BNS damping are not compatible, so that they cannot be done simultaneously. Fortunately, they can be done sequentially [23]. The BNS compensation of the transverse wake effect for the SLC linac at SLAC is obtained by setting the klystron phases later than the crest at the low-energy end of the linac. At the high-energy end, the klystron phases are shifted to make the bunch arrive early for ﬁnal correction of the longitudinal energy spread. Typically the phase shifts used are about 15 to 20◦ relative to the crest. The ﬁnal energy spread is only about 0.3%.

11.17 Long-Range Wakeﬁelds and Multibunch Beam Breakup

The long-range wakeﬁelds generally consist of a few low-frequency modes. Of most concern are the deﬂecting or dipole modes, which can lead to the multibunch BBU instabilities that can limit the intensity in electron linacs. This section presents a brief introduction to the subject, which has been extensively studied by many workers. The classic introduction to the subject is the excellent article by Helm and Loew [24]. The ﬁrst observations of BBU were made in the late 1950s and early 1960s, and reported from SLAC in 1966. Since then, BBU has been a concern for the design of all high-current linacs, and especially for electron linacs. The most serious deﬂecting modes are similar to the pillbox-cavity TM110 mode. Since the accelerating mode is similar to the TM010 mode, one might expect that for a linac structure the ratio of the deﬂecting-mode frequency to the accelerating-mode frequency is about equal to the ratio of the corresponding zeros of the Bessel functions for these two modes, or 3.8/2.4 = 1.6. Indeed it is experimentally observed that the ratio of deﬂecting-mode frequency to the accelerating-mode frequency in linacs is near 3/2. The deﬂecting modes of real accelerating structures (with beam apertures) are generally neither pure TE nor pure TM, but are linear combinations, called HEM, because they are mixed or hybrid electromagnetic modes. Nevertheless, a mode such as the TM110 mode of a pillbox cavity without beam holes has many similar properties to the HEM110 mode in a cavity with beam holes, and it is useful to refer to the pillbox-cavity mode, which has a simple analytic solution. The nonzero ﬁeld components of the

389

390

11 Wakeﬁelds

TM110 mode are Ez = E0 J1 (kD r) cos φ Br = −j

E0 J1 (kD r) sin φ c kD r

Bφ = −j

E0 J (kD r) cos φ c 1

(11.92)

where the deﬂecting-mode wave number is kD = ωD /c, and the time dependence of each component is described by the factor e jωD t . There are really two modes with orthogonal polarities. The ﬁeld lines are shown in Fig. 11.18. The magnetic ﬁeld of the TM110 mode is transverse to the cavity axis and produces a transverse beam deﬂection that is perpendicular to both the beam axis and the magnetic-ﬁeld direction. The mode has a longitudinal electric ﬁeld, which vanishes and reverses direction on the cavity axis, as shown in the ﬁgure. Multibunch BBU can be initiated from noise, corresponding to a small transverse modulation of the beam bunches at a deﬂecting-mode frequency, which can excite a small excitation of the deﬂecting mode. Once the deﬂecting mode is excited, it acts on subsequent beam bunches to drive them further off axis, which further reinforces the excitation of the deﬂecting mode. There are two different mechanisms through which BBU can grow, known as regenerative and cumulative BBU. The regenerative mechanism occurs in a single multicell linac structure. A standing-wave linac typically has cells that are strongly coupled electromagnetically, and in this case the deﬂecting mode is a particular normal mode of the multicell array with a phase velocity that is at least approximately synchronous with the beam. The deﬂecting-mode excitation from the beam-cavity interaction is carried from one cell to the next by the strong electromagnetic coupling of the cells. The strongest excitation can occur if a HEM mode is excited with a phase velocity that is slightly lower than the beam velocity, such that the beam

Electric field

Magnetic field

Figure 11.18 Field pattern of a TM110 mode of a pillbox cavity.

11.17 Long-Range Wakeﬁelds and Multibunch Beam Breakup

can slip ahead in the structure by about 180◦ of the deﬂecting-mode cycle. To see how the regenerative effect can occur, suppose that in the front part of the structure the phase is such that the beam is deﬂected by the magnetic ﬁeld, and acquires a larger transverse displacement. One-quarter period later, the off-axis beam in the back part of the structure can be decelerated by the off-axis longitudinal electric ﬁeld. In this case the beam is both deﬂected by and transfers energy to the electromagnetic ﬁeld of the deﬂecting mode. The energy increment to the deﬂecting mode is rapidly distributed throughout the structure by the strong intercell coupling, and this results in an increased amplitude for the particular normal mode causing the deﬂection. Above a threshold current and pulse length, where the beam-induced growth rate exceeds the ohmic loss rate, the ﬁeld and beam displacement grow exponentially, and the beam will be eventually deﬂected so strongly that it is lost in the linac. A starting current for regenerative standing-wave oscillations for beam particles with charge q and rest energy mc2 can be estimated from [25] IS ≈

π 2 λ βγ (mc2 /q) d Fa (x) L 4Q(r⊥ /Q)

(11.93)

where γ and β are the usual relativistic factors corresponding to the ﬁnal beam energy, λD is the free-space wavelength of the deﬂecting mode, L is the structure length, and r⊥ is a transverse shunt impedance per unit length, deﬁned later in this section. The shunt impedance r⊥ measures the coupling of the beam to the deﬂecting mode, where a typical value is r⊥ /Q = 100 /λd . Fa (βγ /βi γi ) is a correction factor for acceleration from an initial momentum mcβi γi to a ﬁnal momentum mcβγ , given by Wilson [26]as Fa (x) =

6x[(x −

(x − 1)3 + x − 1 − x ln(x)]

1)2 /2

(11.94)

where for large arguments Fa (x) ≈ 1/3. A starting-current expression for regenerative traveling-wave oscillations is also given by Helm and Loew [27]. Growth rates for the instability are also given for both types of structures in this same reference. The starting current for pulsed beams can be larger than for the CW case, as discussed by Wilson [28]. The cumulative-BBU mechanism occurs in a linac consisting of an array of electromagnetically independent cavities, and again may be initiated from a small initial transverse modulation of the beam bunches at a deﬂecting-mode frequency [29]. The HEM deﬂecting mode is resonantly excited in each cavity by this modulation, and each cavity in turn ampliﬁes the transverse beammodulation amplitude. The growing beam displacements, resulting from the deﬂecting-mode excitation in the earlier cavities, accumulate to produce larger beam displacements in the later cavities, so that the instability is coupled from earlier to later cavities by the beam. The effect is cumulative both as a function of the length or number of cavities, and time or number of bunches. Cumulative BBU was important in the early operation of SLAC [30] and was

391

392

11 Wakeﬁelds

extensively studied there. Cumulative BBU effects are particularly serious in their transient behavior, where the amplitude growth can become very large. Our discussion of cumulative BBU will follow the deﬁnitions and results by Gluckstern et al. [31] but we have reexpressed the results in the notation used in many earlier references. An important parameter describing the transverse beam-cavity interaction is a transverse shunt impedance, or transverse coupling impedance, for which several deﬁnitions can be found in the literature. From the Panofsky–Wenzel theorem, the transverse momentum change of the particle passing through a cavity excited in a single mode is proportional to a parameter R⊥ /Q, the ratio of a transverse shunt impedance in ohms to its quality factor, deﬁned by [32] ∂Ez jω z/c 2 D dz e R⊥ c3 ∂x ≡ Z0 3 Q ωD E 2 dV

(11.95)

√ where Z0 = µ0 /ε0 is the impedance of free space and ωD is the deﬂectingmode frequency. By including the complex exponential factor, the transit-time effect is included. Usually ∂Ez /∂x is nearly an even function of z, and the cos(ωD z/c) term is dominant after the integration, because it is an even function. Also, frequently appearing in the literature is a transverse shunt impedance per unit length, used in the above discussion of regenerative BBU, deﬁned as r⊥ = R⊥ /, where is the active length per cavity for the deﬂecting mode. [33] Parameter values for the 2856-MHz (accelerating-mode frequency), 3-m SLAC structure with a deﬂecting-mode frequency about a factor of 1.5 larger than the accelerating mode, are R⊥ /Q ≈ 400 , and ≈ 0.25 m. The frequency scaling of R⊥ /Q depends on what is held constant. For ﬁxed ﬁeld amplitude in Eq. (11.92) ∂Ez /∂x ∝ ωD , and for ﬁxed accelerating structure length, the denominator is proportional to λ2D . Then R⊥ /Q ∝ λ−1 D , and useful rules of thumb are R⊥ /Q ≈ 400 [0.105m/λD (m)], and r⊥ /Q ≈ 100 /λD to 200 /λD . Returning to cumulative BBU, the transverse magnetic ﬁeld of the deﬂecting mode imparts a transverse momentum impulse to the bunch, and the transverse bunch modulation drives the deﬂecting mode in each cavity through the interaction with the off-axis axial electric ﬁeld. Between cavities the momentum impulse is converted to an increased displacement amplitude. When the bunch enters the next cavity, it has a larger displacement and interacts even more strongly with an electric ﬁeld that is larger at larger radius. The deﬂecting-mode amplitude in each cavity and the transverse amplitude of the beam are determined by competition between the growth induced by the beam-cavity interaction, and the decay caused by the ohmic power loss in the cavity walls. These amplitudes increase monotonically with length along the linac. As a function of time at a given cavity, the behavior is more complicated. As shown in Fig. 11.19, there is an exponentially growing

11.17 Long-Range Wakeﬁelds and Multibunch Beam Breakup

initial transient for times short or comparable with the deﬂecting-mode time constant. The growth of the displacement reaches a maximum when the sum of the growth plus decay exponents reach a maximum value, after which the amplitude decreases, a result of both the decay from the ﬁnite Q of the deﬂecting mode, and the driving forces from the bunches which get out of phase because the bunch frequency and the deﬂecting-mode frequency are not equal. For times much larger than the deﬂecting-mode time constant, the displacements approach an asymptotic limit, also called the steady-state regime. The large increase in displacement with increasing cavity number is evident. The steady-state displacement, observable in Fig. 11.19a, is not obvious in Fig. 11.19b because of the scale change from millimeters to centimeters in the plots. The net amplitude after the steady state has been reached is proportional to eFss , where ignoring acceleration and focusing, the total growth exponent is 1/2 33/4 πIR⊥ z2 Fss (z) = 3/2 (11.96) 2 V0 LλD Also, qV0 = mc2 γβ, L is the center-to-center cavity spacing, z is the position along the linac (z = Nc L, where Nc is the cavity number), λD is typically 2/3 of the free-space wavelength of the accelerating mode, and I is the average current (I = qN/τ , where N is the number of particles per bunch and τ −1 is the bunch frequency). For a relativistic beam, the effect of acceleration can be included by making the substitution, qV0 → mc2 γ z/4, where γ gives the acceleration rate [34]. Thus, for an accelerated beam with no focusing, the asymptotic solution is 1/2 33/4 πIR⊥ z Fss (z) = 1/2 (11.97) 2 (mc2 /q)γ LλD In the transient regime, the amplitude growth varies approximately as eFe (z,t)−ωD t/2Q with a growth and a decay term, and which for t < 2QFe /ωD is dominated by the growth term, Fe (z, t). The growth exponent for the transient case, assuming no focusing and no acceleration, is 1/3 33/2 π 2 I(R⊥ /Q)ctz2 Fe (z, t) = 5/3 (11.98) 2 V0 Lλ2D Using the prescription above to include the effect of acceleration, the transient solution for the growth exponent for a relativistic beam with no focusing, is 1/3 33/2 π 2 I(R⊥ /Q)ctz Fe (z, t) = (11.99) 2 (mc2 /q)γ Lλ2D Equations (11.96) to (11.99) can be used to estimate the threat of cumulative BBU, by comparing the computed e-folding factors with empirical values of

393

11 Wakeﬁelds

394

−1.4 mm

−1.4 mm (a)

0.0

200.0

400.0

600.0

800.0 1000.0 1200.0 1400.0 1600.0 1800.0 2000.0

15 cm

−15 cm (b)

0.0

200.0

400.0

600.0

800.0 1000.0 1200.0 1400.0 1600.0 1800.0 2000.0

Figure 11.19 Typical bunch-centroid displacement showing cumulative BBU at a ﬁxed cavity versus bunch number M (time is Mτ ), for the ﬁrst 2000 bunches, and for a beam in which all the bunches are initially offset at the ﬁrst cavity by 1 mm. The upper curve (a) is for cavity 15 and the lower curve

(b) is for cavity 30. The pattern in the plots is a consequence of the choice of an integer ratio or the deﬂecting-mode to accelerating-mode frequencies. [From R. L. Gluckstern. R. L. Cooper, and P. Channell, Part. Accel. 16, 125 (1985), Copyright 1985 Gordon and Breach].

11.17 Long-Range Wakeﬁelds and Multibunch Beam Breakup

about F = 10, where one may begin to observe emittance growth, and a value of about 15 to 20 where the beam strikes the walls, [35] the exact values depending on the size of the initial noise-induced transverse amplitude modulation. It was found in the early days at SLAC that the experimental threshold for BBU, when the beam begins to strike the accelerator walls, corresponds to a value of the e-folding factor for transient growth Fe ≈ 20 [36]. As pointed out by Wilson, with 960 cavities at SLAC, and if all cavities participate, the actual ampliﬁcation per cavity required to reach the beam-loss threshold was small, since the growth factor for the last cavity, (1 + ε)960 = eFe is satisﬁed for Fe ≈ 20, when ε ≈ 0.021. To estimate the value of Fe at the transient peak for any point z in the linac, one needs an expression for the time that the maximum occurs. An approximate result can be obtained from the formulas by ﬁnding the value of t that maximizes the total exponent eB (z, t) = Fe (z, t) − ωD t/2Q. If we express Fe = G(z)t1/3 , where G(z) ≡ Fe (z, t)/t1/3 , we can show that the time for the transient peak is tmax =

2G(z)Q 3 ωD

3/2 (11.100)

If tmax is substituted into Eq. (11.98) or (11.99), the growth exponent Fe (z, tmax ) at the transient maximum is obtained, which may be compared with a value in the range of 15 to 20, above which it is empirically known that beam loss may be expected. These results also predict that at the transient maximum, the decay exponent is one-third the size of the growth exponent. It has been found empirically that a simple modiﬁcation of Eqs. (11.96) to (11.99) includes the focusing in a ﬁrst approximation. In this method one simply replaces z with the average Courant–Snyder beta function [37]. For a ball-park estimate, in a focusing-defocusing (FODO) lattice, the average beta function is approximately equal to the focusing period at a phase advance per period of 60◦ . A space-charge-corrected beta function can be approximated by dividing the zero-current average β value by the tune depression. More accurate estimates for the effect of focusing in the transient regime of cumulative BBU can be obtained directly from the theory [38]. A dimensionless focusing parameter is deﬁned as 16 V0 sin2 σ0 ρ= √ 3 3 IR⊥ LkD

(11.101)

where σ0 is the zero-current transverse phase advance per cavity of the focusing lattice [39]. The focusing-dominated regime, where the BBU displacement is a small perturbation, corresponds to ρ > 1, whereas ρ < 1 is the BBU dominated regime, in which the expressions reduce to the previous results with no focusing when σ0 = ρ = 0. Assuming the parameters ρ and σ0 are constant along the accelerator, the total exponent for the focusing-dominated

395

396

11 Wakeﬁelds

regime with ρ > 1, is eB (ρ > 1) = Fe (z, t) −

MωD τ = 2Q

I(R⊥ /Q)k2D ctz 2V0 σ0

1/2 −

MωD τ 2Q (11.102)

which has a maximum value eB,max (ρ > 1) =

IR⊥ kD z 4V0 σ0

The maximum occurs at time tmax (ρ > 1) =

I(R⊥ /Q)k2D cτ 2 z 2V0 σ0

(11.103)

Q ωD τ

2 (11.104)

corresponding to bunch number Mmax = tmax /τ . The equations given above assume that the resonant frequency of the deﬂecting modes is the same for all cavities. Because of fabrication errors, this is generally not true. Although accelerating cavities must be tuned to give nearly the same accelerating-mode frequencies, the same accuracy is not obtained for the other modes. The presence of random frequency errors for the deﬂecting modes that are comparable to or greater than the cavity bandwidth is a beneﬁt because it inhibits the cooperative behavior of all the cavities in the BBU instability. It can be shown [40] that to evaluate the effect of deﬂectingmode frequency errors, the effective Q that should be used in the equations above is reduced to account for fabrication errors. The effective deﬂectingmode quality factor for an ensemble of cavities with random fabrication errors is approximately Q = fD /δf , where fD is the deﬂecting-mode frequency, and δf is the rms resonant-frequency error. For example, if the cavity radius is about R = 10 cm, and the rms fabrication error for the radius is δR = 0.1 mm, then Q = fD /δf ≈ R/δR = 1000. For the case where signiﬁcant random errors are present, cumulative BBU can only occur on a smaller scale involving a few cavities, which by chance happen to lie close in deﬂecting-mode frequency. If necessary, this restricted BBU effect can be eliminated by installing higherorder mode couplers to provide additional loading of the deﬂecting-mode Q values. In general, there are several effective cures for cumulative BBU: (1) good beam alignment, (2) strong focusing, (3) detuning of the deﬂecting-mode frequencies from cavity to cavity, a technique which might also result naturally from random construction errors, (4) choosing the accelerating gap to reduce the transit-time factor for the deﬂecting mode, thereby reducing the transverse shunt impedance, and (5) loading the deﬂecting modes by installing higherorder mode couplers that selectively propagate the electromagnetic energy to an absorber. Lower frequency also helps to reduce these effects. Finally, the general BBU problem has been studied [41] for a linac with coupled cavities, where the product of coupling strength k and the Q for the deﬂecting mode determines the BBU regime. Cumulative BBU corresponds to weak

11.18 Multipass BBU in Recirculating Electron Linacs

coupling, kQ 1, and regenerative BBU corresponds to strong coupling between cavities, kQ 1. This shows that regenerative and cumulative BBU are fundamentally the same phenomenon, representing two limiting cases of the cavity-coupling parameter kQ.

11.18 Multipass BBU in Recirculating Electron Linacs

In addition to single-pass BBU, which was discussed earlier for conventional single-pass linacs, recirculating electron linacs are susceptible to a regenerative form of a multipass transverse BBU instability. This is an instability caused by excitation of cavity higher-order modes in a recirculating linac, where the beam is reaccelerated in the same linac cavities one or more times. Transverse beam displacements on successive recirculations can excite cavity deﬂecting modes, which act to further deﬂect the beam. At sufﬁciently high currents, and with high deﬂecting-mode Qs especially for a superconducting cavity linac, the threshold for instability may be exceeded, resulting in undamped transverse beam oscillations. An analytic theory for multipass BBU was developed [42] to calculate the threshold current for instability and was applied to the CW CEBAF superconducting recirculating linac, showing that the instability threshold for that case was more than an order of magnitude greater than the design beam current. That treatment corresponds to the case where the beam phase is the same on each recirculation and energy is always being added to the beam. Subsequently, the multipass BBU theory was generalized to describe the case where the bunches do not necessarily have the same RF phase at each recirculation [43]. This generalization of the multipass BBU theory allows for the important application to the case of an energy-recovery linac (ERL). After an initial pass where the beam is accelerated and sent through a wiggler to convert some fraction of beam energy into FEL (free-electron laser) radiation, the beam bunches are reinjected into the same linac, but this time with a decelerating phase to convert the large fraction of unused beam energy into RF energy of the accelerating mode for subsequent acceleration of new bunches. Figure 11.20 shows a schematic drawing of the Jefferson Lab ERL system. A threshold-current formula has been derived for the case of a single higherorder mode (HOM) with angular frequency ωλ and wave number kλ , for one recirculation with recirculation time tr . The result is 2pc Ith = − R Qλ kλ M12 sin ωλ tr e Q

(11.105)

where p is the initial beam momentum, c is the speed of light, M12 is the transport matrix element for one circulation relating the initial transverse divergence before the cycle to displacement after the cycle, e is the electron

397

398

11 Wakeﬁelds tor

Injec

IR wiggler Linac Optical system Dump

Electron recirculation beam line

Figure 11.20 Schematic of the original FEL at Jefferson Lab with its photoinjected, superconducting RF linac with energy recovery. After passing through the

superconducting linac and the wiggler, the beam is transported back in a recirculation loop and reinjected in the linac at a deceleration phase.

charge, (R/Q)λ Qλ is the shunt impedance and Qλ is the quality factor of the higher-order deﬂecting mode. The formula applies for the case where M12 sin ωλ tr < 0. Several effects can be helpful for increasing the threshold current. Increasing the injection momentum helps, but an efﬁcient ERL requires a low injection energy to reduce the amount of beam energy that goes to the dump and is not recovered. Lowering the quality factor of the deﬂecting mode, which can be done by use of a HOM-coupler system, is important. Another approach for suppressing the BBU is through appropriate beam-optics design [44]. One such beam-optics solution that can help for regenerative multipass BBU when the deﬂection per pass is much larger in one transverse plane than the other, is to rotate the beam by 90◦ so that the large deﬂection is more equally shared by both transverse planes. It is observed that the threshold current for regenerative multipass BBU in a recirculating linac can be much lower than the threshold current for regenerative single-pass BBU. One reason for this lower threshold may be the fact that in single-pass regenerative BBU the relatively short length of an accelerating structure converts a deﬂection at the entrance into a relatively small displacement increase at the exit, contributing a relatively small amplitude increment in the deﬂecting mode. By contrast, for multipass BBU in a recirculating linac a much larger distance between kicks, nearly the entire orbit of the machine is available for converting a small deﬂection into a larger displacement, resulting in a larger deﬂecting mode increase. However, there is another important difference between single-pass and multipass regenerative BBU. In single-pass BBU the most dangerous modes are those that have a 90◦ phase slip between the particle and the wave within the structure. This allows a particle to see a magnetic ﬁeld for deﬂection at the entrance of the structure, and an electric ﬁeld producing energy exchange between the beam and the electric ﬁeld at the exit. In multipass BBU the most dangerous modes are those for which the phase slip between the particle

Problems

and the wave is zero. This allows the energy exchange that can build up the deﬂecting-mode amplitude to occur throughout the entire structure [45].

Problems

11.1.

11.2.

11.3.

The longitudinal wake potential wz (s) is deﬁned for a given mode as the induced voltage per unit source charge seen by a comoving particle that trails behind the source by a distance s. The loss factor k for the mode is related to the wake potential immediately behind the source charge wz (0) = 2k. (a) Calculate wz (0), the wake potential behind the source for the fundamental mode of a single cell of length g = λ/2 using the ﬁeld E = −q/πε0 a2 . Note that the charge q in this equation is equal to the source charge q1 . Assume that λ = 10 cm, a = 1 cm and obtain a numerical value in V/pC. (b) Using part (a) calculate the numerical value of the loss parameter k. (c) The loss parameter is a characteristic parameter of a cavity mode even in the absence of an exciting source charge, and is given by k = ωr/4Q0 , where r is the shunt resistance, and Q0 is the unloaded Q factor or quality factor. Calculate k in V/pC for a 1.3 GHz fundamental mode with r = 50 M and Q0 = 10, 000. Use the method from Section 11.8 to calculate the loss parameter form factor ff(d/λ) for a line-charge density, which is uniform over a distance d. Assume that the line-charge density is ζ (z1 ) = Q/d for −d/2 ≤ z1 ≤ d/2, and zero elsewhere. Write the line-charge loss parameter as kζ = k ff(d/l). Calculate the form factor for d/λ = 0, 1/4, 1/2, 3/4, and 1, and plot ff versus d/λ. Beyond what value of bunch length d/λ does the loss parameter for the uniform beam drop below 90% of the δ-function loss parameter? What is the physical cause of the reduction of the form factor? Consider a cylindrically symmetric linac structure containing a wakeﬁeld produced by a single longitudinal mode and a single transverse dipole mode. Assume that the cell length is 5 cm, the linac is 10-m long and the ﬁnal beam energy is 100 MeV. Assume that the longitudinal accelerating-mode frequency is ω0 /2π = 3 GHz, and its loss parameter k0 = 2V/pC per cell. Assume that the dipole-mode frequency is ω1 /2π = 4.5 GHz, its loss parameter is k1 = 1V/pC per cell, and the radial aperture of the iris and beam pipes is a = 1 cm. (a) Write the expression for the δ-function longitudinal wake potential per cell as a function of the position s, and plot it for −10 cm ≤ s ≤ 20 cm. (b) Write the expression for the transverse wake potential divided by r1 /a, and plot it versus the position s for the same range as in part (a). (c) Suppose the source charge is q1 = 5 nC, and the test charge is q = 1.6 × 10−19 C, located at a distance of 2 mm behind the source, and with a 1-mm displacement along the x-axis. Calculate the energy loss from the wakeﬁelds delivered to the test charge along the whole

399

400

11 Wakeﬁelds

11.4.

11.5.

11.6.

11.7.

accelerator in MeV. What fraction of the ﬁnal energy is this? Calculate the transverse momentum impulse in MeV/c. What fraction of the ﬁnal momentum is this? Beam energy spread from longitudinal wakeﬁelds must be corrected in linacs for linear colliders and free electron lasers. Suppose we approximate the bunch wake potential of the 50 GeV SLAC linac with the linear expression Wz (s) = Wz s. Suppose we choose the slope Wz = 1 V/pC/mm. Assume that stail = 2 mm, N = 5 × 1010 , lc = 3.5 cm, L = 3000 m, and ω/2π = 3 GHz. (a) Calculate the ﬁnal energy spread caused by wakeﬁelds that act from head to tail. What percentage of the ﬁnal energy is this energy spread? (b) By convention, zero phase refers to the point in the RF cycle of maximum energy gain, regardless of the sign of the particle’s electric charge. If the magnitude of the accelerating ﬁeld is 17 MV/m, what nominal phase angle is required for compensation of the energy spread caused by the wakeﬁelds to that caused by the sinusoidal accelerating voltage? Does this phase correspond to the beam being earlier or later than the point of peak energy gain? Single-bunch BBU can be a serious problem in a high-intensity electron linac unless it is corrected. Use the two-particle model to calculate the ratio of the amplitudes of the tail to the head oscillations at the end of the SLAC linac. Assume that the dipole wake potential for a test charge at s = 2 mm behind the source charge at the head is 1.2 V/pC/(r1 /a). Assume the number of particles per bunch is N = 5 × 1010 , and assume the betatron wavelength, representing the smoothed focusing force is λβ = 100 m. For the other parameters, use L = 3000 m for the length of the linac, lc = 3.5 cm for the cell length, a = 1.1 cm for the radial aperture, and assume that an average electron beam energy of γ mc2 = 25 GeV. Is single-bunch BBU a serious problem? Apply BNS damping for compensation of the transverse wake effect. Use the same parameter values as in problem 11.5 and two-particle model for BNS damping. (a) Calculate the required fractional energy difference between the tail and the head that provides cancellation of the dipole wake force at the tail. (b) What is the operating phase of the bunch head required to produce this energy difference from the RF accelerating ﬁeld? Does the bunch come earlier or later than the peak in the energy gain? Calculate the starting current for regenerative BBU in a 20-MeV standing-wave electron-linac structure, designed with no transverse focusing, with injection energy of 100 keV, length of 1 m, and accelerating-mode frequency equal to 3 GHz. Assume that the most dangerous deﬂecting mode lies at a frequency 1.5 times the acceleratingmode frequency, and has r⊥ /Q = 100 /λd , and Q = 10,000. Assume that the acceleration correction factor is Fa = 1/3. Based on the result, can this linac be expected to deliver 100 mA output current.

Problems

11.8.

11.9.

11.10.

11.11.

11.12.

11.13.

Consider a 3-GHz CW linac consisting of 36 independent cavities, each separated by L = 2 m, that accelerates an electron beam from 5 to 200 MeV with no focusing. Assume the following parameters: I = 100 mA, R⊥ /Q = 400 , and a deﬂecting mode at fD = 4.5 GHz with Q = 104 . (a) Assuming a uniform accelerating gradient, calculate the cumulative-BBU growth exponent for the steady state in the last linac cavity. (b) Assuming a value for Fss below 15 is required to avoid beam loss, is steady state beam loss from cumulative BBU expected for this linac, and if so where in the linac does it begin to occur? (c) Assuming a value for Fss below 10 is required to avoid emittance growth, is emittance growth from cumulative BBU expected for this linac, and if so where in the linac does it begin to occur? Consider the linac of problem 11.8, and assume that a FODO focusing lattice is added with a focusing period 2L, and an average Courant–Snyder ellipse parameter β˜ equal to the focusing period. Calculate the new cumulative-BBU growth exponent at the last linac cavity in the approximation that focusing can be included by replacing z → β˜ in the expression for the exponent. Using the thresholds for emittance growth and beam loss stated in problem 11.8, is emittance growth or steady-state beam loss from cumulative BBU expected when focusing has been added to this linac? Writing the total transient cumulative-BBU exponent as eB (z, t) = Fe (z, t) − ωD t/2Q, and expressing the growth exponent for transient cumulative BBU as Fe = G(z)t1/3 , where G(z) is a constant, shows that the value of time that maximizes the total exponent is given by Eq. (11.100). Show that at the time of the transient maximum, the decay exponent is one-third the size of the growth exponent. Consider the linac of problem 11.8 with no focusing. (a) Use the result from problem 11.10 to calculate at the last cavity the time for the maximum displacement, which occurs at the transient maximum. (b) Using the result from part (a), calculate the value of the maximum transient cumulative-BBU growth exponent Fe (zmax , tmax ) in the last cavity. Then calculate the decay term. Which term is larger? (c) Assuming a value for Fe below 15 is required to avoid beam loss, is beam loss expected from transient cumulative BBU? Repeat problem 11.11 assuming that the effect of focusing can be approximately included by replacing z → β˜ in all the expressions, and that the average Courant–Snyder ellipse parameter β˜ is equal to the focusing period 2L. Assuming that a value for Fe below 15 is required to avoid beam loss, is beam loss expected from transient cumulative BBU? Consider the linac of problem 11.8 with a FODO focusing lattice with a period 2L, a phase advance per period of 60◦ and a phase advance per cavity of σ0 = 30◦ . Calculate the focusing parameter ρ at 5 and

401

402

11 Wakeﬁelds

200 MeV, and decide whether focusing has helped by seeing if the problem is still BBU dominated or has become focusing dominated. 11.14. Consider the linac of problem 11.8 with no focusing and suppose that because of fabrication errors the RMS error in the cavity radius is 0.05 mm. (a) Assuming that the RMS frequency error for the deﬂecting mode is given by δf ≈ fD δR/R, where R = 38 mm, calculate the effective Q for the deﬂecting mode using the result from the statistical cumulative-BBU theory that the effective Q = fD /δf . (b) Using the effective Q from part (a), and assuming a uniform accelerating gradient, calculate the cumulative-BBU growth exponent in the steady state for the last cavity. Is steady-state beam loss from cumulative BBU expected to be a concern for this linac, assuming a value for Fss below 15 is required to avoid beam loss. (c) Using the result of Eq. (11.100), calculate the time for the transient maximum in the last cavity, and calculate the maximum transient cumulative-BBU growth exponent Fe (zmax , tmax ) in the last cavity. Assuming that a value for Fe below 15 is required to avoid beam loss, is beam loss expected from transient cumulative BBU? References 1 Chao, A. W., Physics of Collective Beam

Instabilities in High Energy Accelerators, John Wiley & Sons, 1993. Bongardt, K., Pabst, M. and Letchford, A. P., Space Charge Dominated Beams and Applications of High Brightness Beams, ed. Lee, S.Y., Bloomington, IN, October 10–13, 1995; AIP Conf. Proc. 377, 343 (1995). Reiser, M., Theory and Design of Charged Particle Beams, John Wiley & Sons, 1994, pp. 252–260, 267–273, 402–428. Jackson, J. D., Classical Electrodynamics, 2nd ed., John Wiley & Sons, 1975. An ultrarelativistic particle is an idealized concept of a particle that moves at velocity exactly c. Chao, A. W., Physics of Collective Beam Instabilities in High Energy Accelerators, John Wiley & Sons, 1993; and Coherent instabilities of a relativistic bunched beam, SLAC Summer School, AIP Conf. Proc. 105, 353 (1982). Weiland, T. and Wanzenberg, R., Wake Fields and Impedances, Frontiers

9 10

13 14

of Particle Beams: Intensity Limitations, Proceedings of the Joint US-CERN School on Particle Accelerators at Hilton Head Island, Lecture Notes Phys. 400, 39–79 1992. Chao, A. W., Physics of Collective Beam Instabilities in High Energy Accelerators, John Wiley & Sons, 1993, pp. 84–97. Weiland, T., Part. Accel. 15(4), 245–292 (1984). Bane, K. L. F., Wakeﬁeld effects in a linear collider, AIP Conf. Proc. 153, 972–1014 (1987). Wilson, P., Wakeﬁelds and wake potentials, AIP Conf. Proc. 184, 526–564 (1989). Bane, K. L. F., Wilson, P. B. and Weiland, T., Wake ﬁelds and wake ﬁeld acceleration, AIP Conf. Proc. 127, 875–928 (1985). The energy is eventually dissipated in Ohmic losses. The term higher-order modes to describe other modes than the accelerating mode may be a misnomer. For example, in an iris-coupled N-cell cavity, the π mode is usually the accelerating mode, and it

References

16 17 18

19 20 21

lies higher than the other N-1 normal modes in that family. Chao, A. W., Physics of Collective Beam Instabilities in High Energy Accelerators, John Wiley & Sons, 1993, pp. 110–117. Weiland, T., Nucl. Inst. Meth. 212, 13 (1983). Weiland, T., Part. Accel. 15(4), 245–292 (1984). Strictly speaking, the term wakeﬁelds may be misleading for the β < 1 case, where the scattered ﬁelds with v=c are able to advance ahead of the bunch. Weiland, T., Nucl. Inst. Meth. 212, 13 (1983). Chan, K. D. C., Krawczyk, F., private communication. Guignard, G., Space Charge Dominated Beams and Applications of High Brightness Beams, ed. S.Y. Lee, Bloomington, IN, October 10–13, 1995; AIP Conf. Proc. 377, 74 (1995). Balakin, V., Novokhatsky, A. and Smirnov, V., Proc. 12th Int. Conf. High Energy Accel., Fermilab 1983, p. 119. Seeman, J. T., Advances of Accelerator Physics and Technologies, ed. Schopper, Herwig, World Scientiﬁc Publishing, 1993, pp. 219–248. Helm, R. H. and Loew, G. A., Beam breakup, in Linear Accelerators, ed. Lapostolle, P. M. and Septier, A.L., John Wiley & Sons, 1970, pp. 173–221. Wilson, P. B., High Energy Electron Linacs: Applications to Storage Rings, RF Systems, and Linear Colliders, 1981 Summer School on High Energy Accelerators, 506. Wilson, P. B., A Study of Beam Blowup in Electron Linacs, HEPL-297 (Rev.A), High Energy Physics Laboratory, Stanford University, Stanford, CA, 1963. Helm, R. H. and Loew, G. A., Beam breakup, in Linear Accelerators, ed. Lapostolle, P. M. and Septier, A. L., John Wiley & Sons, 1970, p. 187. Wilson, P. B., High energy electron linacs: applications to storage rings, RF systems, and linear colliders, 1981 summer school on high energy accelerators, AIP Conf. Proc. 87, 506 (1981).

29 Wilson, P. B., A Simple Analysis of

30 31

34 35 36

37 38 39

Cumulative Beam Breakup for the Steady-State Case, HEPL-TN-67-8, High Energy Physics Laboratory, Stanford University, Stanford, CA, September, 1967. Panofsky, W. K. H. and Bander, M., Rev. Sci. Instrum. 39, 206 (1968). Gluckstern, R. L., Cooper, R. K. and Channell, P. J., Part. Accel. 16, 125 (1985). On page 142 of this paper the authors provide the substitutions required to compare the formulas in the paper with earlier work. Also provided in this reference are the basic difference equations that will allow the reader to do numerical calculations. This deﬁnition is equivalent to that given by Eq.(9) of the previous reference, converted to R⊥ /Q according to the substitution given on page 142 of that reference. Wilson, P. B., High energy electron linacs: applications to storage rings, RF systems, and linear colliders, 1981 summer school on high energy accelerators, AIP Conf. Proc. 87, 507 (1981). SLAC-PUB-2884 (Revised), November 1991, 56. Note that although the cavity end cells are tuned to produce a uniform distribution for the accelerating mode, the ﬁeld ﬂatness may be poor for other modes, and the deﬂecting mode may not effectively ﬁll all of the cells with electromagnetic energy. The active length may be chosen to take this into account. For the general case with β < 1, γ is replaced by βγ + β γ = γ 3 β . Wilson, P. B., private communication. Helm, R. and Loew, G., in Linear Accelerators, ed. Lapostolle, P. M. and Septier, A. L., John Wiley & Sons, 1970, p. 205. Wilson, P., private communication. Gluckstern, R. L., Neri, F. and Cooper, R. K., Part. Accel. 23, 53 (1988). Be careful here because σ0 is usually deﬁned as phase advance per focusing period. Gluckstern, R. L., Neri, F. and Cooper, R. K., Part. Accel. 23, 37 (1988).

403

404

11 Wakeﬁelds 41 Gluckstern, R. L. and Neri, F. Part.

Accel. 25, 11 (1989). 42 Bisognano, J. J. and Gluckstern, R. L., Multipass Beam Breakup in Recirculating Linacs, Proc. 1987 Part. Accel. Conf., Washington, DC, (IEEE Catalog No. 87CH2387-9), pp. 1078–1080. 43 Hoffstatter, G. H. and Bazarov, I. V., Beam-breakup instability theory for

energy recovery linacs, Phys. Rev. Special Topics Accel. Beams 7, 054401 (2004). 44 Rand, R. and Smith, T., Beam optical control of beam breakup in a recirculating electron accelerator, Part. Accel. 11, 1–13 (1980). 45 Smith, T., private communication.

405

12 Special Structures and Techniques

The history of linear accelerators is rich in ideas for improving the performance of linac structures and linac systems. In this chapter we will review some of these ideas, which are in various stages of development and some of which have been proposed only recently. We will discuss ideas for increasing the beam current, focusing strength, peak power, and electrical efﬁciency. It is probable that from some of these ideas, the linac ﬁeld will advance even further in the years ahead.

12.1 Alternating-Phase Focusing

In Chapter 7, we discussed Earnshaw’s theorem from which followed the incompatibility of simultaneous longitudinal and radial focusing from the radio frequency (RF) ﬁelds. The conventional solution is to overcome the radial defocusing by the use of external magnetic-focusing lenses. However, the general application of the alternating-focusing principle has led to two other methods that provide a net focusing effect in all three planes from the RF ﬁelds alone, without violating Earnshaw’s theorem. One method is the use of radiofrequency-quadrupole (RFQ) focusing, as described in Chapter 8. In the RFQ a time-varying alternating-focusing force is applied in the transverse directions x and y, and conventional nonalternating focusing is provided in the longitudinal z direction. In a second method, alternating-phase focusing (APF), [1, 2] the focusing force is alternated in r and z. Suppose we consider a nonrelativistic beam that is accelerated in a drifttube linac or DTL. If the synchronous phase is negative, corresponding to particles arriving at the center of an accelerating gap when the sinusoidal accelerating ﬁeld is increasing, the particles are focused longitudinally, but defocused transversely. If the synchronous phase is positive, corresponding to particles arriving at the center of a gap when the ﬁeld is decreasing, the particles are focused transversely, but defocused longitudinally. In either case the beam is unstable in at least one direction. However, both the transverse RF Linear Accelerators. 2nd, completely revised and enlarged edition. Thomas P. Wangler Copyright  2008 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-40680-7

406

12 Special Structures and Techniques

and the longitudinal motion can become stable, if the sign of the synchronous phase is changed periodically so that alternating-focusing forces are applied in all directions. Variation of the gap-to-gap synchronous phase in a multicell DTL tank, is obtained by suitable spacing of the gap centers. Instead of a strictly monotonic increase of the drift-tube lengths as in a conventional DTL, a periodic variation is also superimposed, so that the arrival time of the synchronous particle at the gap centers is programmed to provide the desired alternating-focusing pattern. An analysis of the APF linac has been given by Kapchinskiy [3]. Kapchinskiy found that when the synchronous phase alternates symmetrically about the peak, the longitudinal acceptance is small. If the center of the phase modulation is shifted to negative values, the longitudinal acceptance increases, while the transverse focusing strength decreases. APF is most effective at low particle velocities, and at low velocities electrostatic focusing effects, discussed in Chapter 7, become appreciable. Using typical numbers for the low-velocity end of a proton DTL, Kapchinskiy concluded that the optimum center of the phase modulation lies somewhere between −5◦ to −10◦ . In principle, the use of APF eliminates the need for installing quadrupoles in the drift tubes, and therefore removes restrictions on the geometry of the drift tubes, enabling higher shunt impedance for the DTL structure. It allows the DTL injection energy to be extended to lower velocities by providing the necessary focusing without quadrupoles in the drift tubes. These potential advantages have stimulated studies [4, 5] during the past two decades to explore the APF parameter space, including a substantial amount of work in the former Soviet Union [6]. Nevertheless, the longitudinal acceptance and the current limit achieved in these studies have generally been too small to be of interest for most high-current linac applications. The region of parameter space that is most advantageous for APF is the same as for the RFQ, and generally the RFQ has been found to give superior performance, providing larger longitudinal acceptance, and stronger longitudinal and transverse focusing. So far, accelerator designers have generally been unable to ﬁnd applications in which APF could win in the competition with the RFQ.

12.2 Accelerating Structures Using Electric Focusing

The use of electric rather than magnetic focusing is motivated by the need for stronger focusing at low velocities, where the current limit of a linac system is generally determined. The RFQ, the most well-known example of electric focusing, was discussed in Chapter 8. Other methods have also been proposed, but so far none of these has achieved the same popularity as the RFQ. Despite the unmatched advantages of the RFQ as a buncher, it may still be useful to consider the other electric-focusing concepts as possible alternatives for the more inefﬁcient higher energy accelerating section of the RFQ.

12.2 Accelerating Structures Using Electric Focusing

An early suggestion [7] for an electric-focusing structure was to construct the drift tubes with four conducting ﬁngers arranged in a quadrupole geometry within the accelerating gap, where two ﬁngers in one transverse plane are attached to the upstream drift tube and two in the orthogonal transverse plane are connected to the downstream drift tube. This arrangement produces an electric-quadrupole ﬁeld in the accelerating gap. This ﬁngerfocusing geometry led to the invention by Tepliakov of a π-mode electricfocusing structure, which alternate gaps without ﬁngers, optimized for acceleration, and gaps with ﬁngers, optimized for focusing [8]. The Tepliakov structure was built and operated as an injector linac at the IHEP laboratory in the former Soviet Union [9]. Another proposal is the Lapostolle matchbox geometry in which the drift tubes have the form of rectangular boxes with the long axis alternating between the x and y directions from one drift-tube to the next [10]. This geometry has the advantage of avoiding the use of ﬁngers that increase the peak surface electric ﬁeld. A related RFQ structure, used for acceleration of low-charge-state heavy ions, is the split coaxial structure [11], which was invented at the GSI laboratory in Darmstadt. The original GSI structure operated at 13 MHz [12]. Instead of vanes or rods, this structure is comprised of many individual small ring electrodes, which are concentric with the beam axis, and to which a pair of diametrically opposite focusing ﬁngers is attached. The rings are mounted on stems that are alternately connected to two pairs of long horizontal spears. Each pair of spears projects from one of the two end plates from which the spears and the attached electrodes are charged. The result is that adjacent electrodes are oppositely charged, and the ﬁngers provide both quadrupole focusing and acceleration in the gap between the rings. The axial potential at the end plates is not zero, which must be taken into account, especially at the low-energy end, where the beam is injected. This RFQ structure can provide efﬁcient acceleration, which is particularly desirable at the high-velocity end of the RFQ, after the beam is bunched. A recent proposal by Swenson is the RF-focused drift-tube linac or RFD [13]. In the RFD the drift tubes are split into two separate electrodes operating at different electrical potentials. Each electrode supports two ﬁngers which point inward toward the opposite electrode, instead of extending into the accelerating gap, and forms a four-ﬁnger quadrupole geometry within each drift-tube structure. These electrodes are electrically isolated from each other by the inductance of their separate water-cooled blade-type stems. During each period the beam traverses two separate regions, an accelerating gap between adjacent drift tubes, and a ﬁnger-focusing gap between the two electrodes that comprise each drift tube. As in a conventional linac, the synchronous phase for the accelerating gap corresponds to an accelerating ﬁeld that is rising in time, as required for longitudinal phase stability. The synchronous phase at the focusing gap corresponds to deceleration, but the longitudinal

407

408

12 Special Structures and Techniques

electric ﬁeld in this gap is small, since the main electric-ﬁeld component of this gap is a transverse quadrupole focusing ﬁeld, produced by the four ﬁngers. The orientation of the ﬁngers in the focusing gaps alternates from gap to gap to produce an alternating focusing-defocusing system. Because of its RF electric focusing, the RFD can provide strong focusing at lower velocities than for conventional magnetically focused DTLs, and consequently, the transition energy between an RFQ that normally bunches the beam and the following RFD can be lower, typically in the 0.5–1 MeV range for protons. Swenson has also proposed an RF-focused interdigital linac structure (RFI), which is an interdigital Wideroe linac with radiofrequency-quadrupole focusing. Some differences between the RFI and the RFD should be mentioned. For the RFD structure the period length is βλ/2 as compared with βλ for the Alvarez DTL. Unlike the DTL, the drift tubes of an interdigital structure alternate in potential. The electric ﬁelds between the drift tubes also alternate in direction from one gap to the next, and all drift tubes reverse sign every half of an RF period. The particles accelerated in one gap will arrive at the next gap in time for the ﬁelds to have reversed sign and produce beam acceleration in every accelerating gap. For the interdigital structure the drift tubes are supported alternately from above and below or from right to left, and the structure is excited in a TE110 -like mode that results in excitation of alternate drift tubes with opposite potential (see Fig. 12.1). As is the case for the RFD the drift tubes are split into two electrically independent electrodes, and each electrode has two diametrically opposite ﬁngers that point inward toward the other electrode instead of extending into the accelerating gap, and forms a four-ﬁnger quadrupole geometry within each drift-tube structure. However, because of the shorter period of the RFI, the focusing gap must be moved upstream closer to the accelerating gap. This results in an asymmetrical drift tube, with a minor electrode and a major electrode, as shown in Fig. 12.1. The minor electrode is supported on a pair of minor stems and the major electrode is supported on a major stem. To excite the quadrupole focusing ﬁelds two minor stems are offset symmetrically on both sides of the major stem, as seen in Fig. 12.2, to couple to the RF magnetic ﬁeld. As for the RFD, the RFI would be used as the linac structure that follows the RFQ. The main advantages of the RFI over the RFD are reduced size and signiﬁcantly increased shunt impedance. An interesting approach for obtaining very high currents of low-velocity ions is to combine both electric focusing and multiple beams. The maximum beam current in an ion-linac channel is usually limited either by the focusing ﬁelds that are attainable, or by the envelope instability in a periodic focusing channel. To obtain higher currents than are possible in single linac channels, multiple beams are required. For acceleration of high-current, low-velocity ion beams, Maschke proposed the multiple-beam electrostatic- quadrupole-array linear accelerator (MEQALAC) [14], a linac in

12.2 Accelerating Structures Using Electric Focusing

Figure 12.1 Two views of the RFI linac structure. The side view on the right shows asymmetrical split drift tubes comprised of minor and major electrodes which are

electrically independent because of inductive isolation. These electrodes contain the focusing gap (courtesy of D. A. Swenson).

Figure 12.2 Drawing of the RFI structure showing the minor stems that are offset symmetrically on both sides of the major stems to provide coupling to the RF magnetic ﬁeld (courtesy of D. A. Swenson).

which beamlets are accelerated in common RF gaps, and transported within individual channels using electrostatic quadrupoles. This idea was motivated by the desire to produce very bright beams for heavy-ion fusion. Maschke argued from linac current-limit formulas that the brightest linac beams are obtained by ﬁlling the available aperture at the lowest practicable velocity and

409

410

12 Special Structures and Techniques

the highest possible RF frequency, which implies using small apertures. The effective emittance, including all beams, is enlarged as a result of the geometrical dilution caused by the ﬁnite separation of the beamlets. Maschke argued that brighter beams are still possible by sufﬁcient reduction in the bore sizes and beamlet spacings. He proposed the construction of compact electrostatic-quadrupole arrays that contain thousands of individual channels. The ﬁrst MEQALAC prototype was built and tested at Brookhaven in 1979 [15]. It accelerated nine beams of Xe+1 from 17 to 73 keV, and operated at a very low 4-MHz frequency. The bore diameters were 0.79 cm. The measured average output current for nine beams was 2.8 mA. At the FOM Institute, a MEQALAC was constructed [16] which accelerated four He+ beams from 40 to 120 keV using an interdigital cavity resonator that operated at 40 MHz (see Fig. 12.3). The bore diameters were 0.6 cm, and a beam current of 2.2 mA per beam was accelerated. Although the MEQALAC can potentially produce high total beam currents, no one has yet demonstrated the design with drastically scaled-down channel dimensions, as was proposed by Maschke. 12.3 Coupled-Cavity Drift-Tube Linac

The coupled-cavity drift-tube linac (CCDTL) [17] combines features of both’ the Alvarez DTL and the π-mode CCL. The CCDTL resembles the side-coupled linac that was discussed in Chapter 4, except that the accelerating cavity is a short zero-mode DTL. Figure 12.4 shows a CCDTL structure where the accelerating cavity is a two-gap DTL. The chain of cavities operates in a π/2-like structure mode with adjacent accelerating cavities having oppositely directed electric ﬁelds. For large enough bore radius, the CCDTL can have

12.4 Beam Funneling

a larger shunt impedance than the DTL, because of the boundary condition between accelerating cells in a π mode, where it forces the accelerating ﬁeld to zero, providing longitudinal ﬁeld concentration to maintain a high transit-time factor (see Section 2.4). Focusing can be provided by replacing an accelerating cavity by a drift space in which an external quadrupole can be installed. If a quadrupole space is inserted, the two end cells on each side of the drift space may still be coupled through a coupling cavity, whose TM010 -like mode frequency is equal to that of the accelerating cells. The idea is similar to an excited bridge-coupler cavity, used for the CCL, except that for the CCDTL, the bridge is a nominally unexcited coupling cell. If the distance between the gap centers of the end cells of adjacent accelerating cavities is an odd number times βλ/2, the coupling cavities must be oriented so that the axes of the coupling-cell posts are parallel to the beam axis, as shown in Fig. 12.4. If the distance is an even number times βλ/2, the coupling cavities must be rotated by 90◦ , so that the axes of the posts are radially directed. This ﬂexibility makes it easy to insert quadrupoles into the chain wherever they are needed, with minimal disruption to the electromagnetic structure mode. 12.4 Beam Funneling

The concept of funneling was proposed at the heavy-ion fusion workshops in the late 1970s [18]. In its simplest form, funneling combines two beams with frequency f and current I, by longitudinal interlacing of the bunches, into a single beam of current 2I, suitable for injection into a linac of frequency 2f .

bl bl 2 3bl 2

Figure 12.4 A coupled-cavity drift-tube linac (CCDTL) structure with a single drift tube in each accelerating cavity.

411

412

12 Special Structures and Techniques

After a frequency doubling, the number of available buckets per unit time doubles and all the buckets can be ﬁlled. Funneling of high-current beams in RF linacs can be advantageous when the current limits, which increase with increasing energy and decrease with increasing frequency, allow a frequency doubling for the subsequent linac sections. Another motivation for funneling two low-energy beams is that a single ion source may have insufﬁcient output current to supply the required total current for the linac. A generic beam funnel, shown in Fig.12.5a, contains bending magnets to merge the two beams, and quadrupole lenses, and rebuncher cavities to focus the beam radially and longitudinally. An RF cavity completes the merging and interlacing of the two beams (Fig. 12.5b). If the bunches from the two low-energy injector linacs are 180◦ out of phase, they can be interlaced by the action of an RF deﬂecting cavity with frequency f that produces a time-varying transverse electric ﬁeld. Ideally, funneling would double the beam current per channel with no increase in transverse emittance. However, some emittance growth can be expected because of nonlinear, chromatic, and time-varying effects associated with the funnel-line elements. It should be noted that, because funneling increases the current by interlacing bunches, funneling does not increase the charge per bunch.

Merging sections

425 MHz

Matching sections

850 MHz DTL

RFQ Transport sections 425 MHz (a)

RF Deflector

RFQ

(b)

merged and interlaced. The transverse scale Figure 12.5 (a) Schematic diagram of a is magniﬁed compared with the longitudinal beam funnel. The output beams from two RFQs are funneled and the beam is injected scale. into a DTL with twice the frequency. (b) The two bunched beams are shown being

12.5 RF Pulse Compression

Simulation studies of funneling lines have been reported [19, 20] and an experiment has been performed to test the funnel concept using a beam in one of the two transport lines [21]. Although funneling has been included in some recent proposals for high-power linacs, [22, 23] no complete two-beam funnel demonstration has been reported in the published literature. A novel proposal for a beam funnel in the form of an RFQ structure for highcurrent, very low-velocity beams was proposed by Stokes and Minerbo [24]. The structure accepts two bunched ion beams and combines them into a single beam with interlaced microstructure pulses, while providing uninterrupted periodic RF electric-quadrupole focusing.

12.5 RF Pulse Compression

When the SLAC linac was designed it was anticipated that at a later date, the output energy could be increased by adding new klystrons to increase the RF power and the corresponding accelerating ﬁeld. However, the plan to purchase and install new klystrons for the upgrade would have been very expensive, and an alternative method was devised using the existing klystrons, in which the electromagnetic energy from the klystron pulse would be stored in a high-Q cavity, which could then be discharged into the linac structure. The technique can be generally used for obtaining very short pulses in traveling-wave electron accelerators for linear colliders. The SLAC approach is known as SLAC energy development (SLED) [25], and the concept is illustrated in a simpliﬁed form in Fig.12.6. The system consists of a klystron, a fast phase shifter, a circulator, a high-Q storage cavity, and an accelerating structure. The klystron is pulsed on with a pulse length that exceeds the ﬁlling time of the accelerating structure. Over most of the klystron pulse, the energy builds up in the storage cavity. No power is reﬂected back to the klystron because of the circulator, and because

HighQ cavity

Klystron

Accelerator 180° Phase shifter

Circulator

Figure 12.6 Schematic diagram of the SLED pulse compression.

413

414

12 Special Structures and Techniques

the structure is matched to the waveguide. At a time equal to one structure ﬁlling time before the end of the klystron pulse, the phase of the klystron drive signal is rapidly shifted by 180◦ . The phase reversal allows a discharge of the energy-storage cavity in which the emitted wave from the storage cavity adds in phase with the forward wave coming directly from the klystron. This results in an enhanced RF power delivered to the accelerating structure over the ﬁnal part of the pulse. At the end of the RF pulse, the accelerating structure is ﬁlled with the enhanced electromagnetic stored energy, and beam is then injected into the structure. The implementation of SLED is somewhat different than is indicated by the ﬁgure. The major components in the real system are: the fast 180◦ phase shifter installed on the low-level input side of the klystron, two high-Q cavities used for energy storage, and a 3-dB coupler to replace the circulator. Typical numerical values are 2.5 µs for the full kystron pulse, 0.8 µs for the ﬁlling time, and an effective power enhancement of about a factor of 2 [26].

12.6 Superconducting RF Linacs

One of the disadvantages of conventional copper-cavity linacs is the large RF power required because of ohmic power dissipation in the cavity walls. As a result, the RF equipment and ac power for accelerator operations are major if not dominant costs for a copper linac. Also, the design choices for cavity geometry and accelerating ﬁelds are dominated by considerations of RF power efﬁciency. Reducing the ﬁelds, to reduce the RF power costs, results in more real estate needed for the linac. For high-duty-factor linacs, the cooling of the copper cavities becomes an important engineering constraint, restricting the ﬁelds even further. Because the mean free path for an electron moving in a conductor increases with decreasing temperature, one might hope to improve matters by cooling the copper cavities to cryogenic temperatures. However, because of the anomalous skin effect, which determines the RF surface resistance when the mean free path exceeds the classical skin depth, the surface resistance of cryogenically cooled copper is reduced by less than a factor of 10 compared with room-temperature copper. Unfortunately, the gain in reduced surface resistance is more than offset by the inefﬁciency of cryogenic refrigeration, so that more total power is required to operate a cryogenically cooled copper linac compared with a room-temperature copper linac. However, the use of superconducting niobium cavities allows a reduction in RF surface resistance by about 105 , compared with room-temperature copper, which results in a net operating power savings for the superconducting case of one to two orders of magnitude. Thus, the application of superconductivity to linacs has been long recognized as an important step toward better performance and lower costs.

12.6 Superconducting RF Linacs

Throughout this book, we have discussed many issues related to superconducting linacs. The superconducting RF surface resistance was introduced in Chapter 1. In Chapter 2, we presented a model for the transit-time factor for large bore π-mode elliptical cavities, we discussed design issues for superconducting cavities, and we discussed frequency scaling for superconducting linacs. In Chapter 4, we discussed independent-cavity linacs and presented results for π-mode coupled-cavity systems. In Chapter 5, we discussed electron-loading effects, including multipacting and ﬁeld emission. Then, in Chapter 10 we discussed operation of heavily beam-loaded linac cavities. In this section, we present general issues regarding superconducting linacs, emphasizing the basic physics. Electric-ﬁeld limitations arise from ﬁeldemission-induced electron loading, and magnetic-ﬁeld limitations are caused by the presence of normal-conducting impurities on the niobium cavity walls. High performance requires attention to cleanliness during the fabrication and assembly of the cavities and cryomodules. Signiﬁcant progress has been made, and the applications of superconducting linacs can be expected to increase in the future.

Brief History

Superconductivity was discovered in 1911 by H. Kamerlingh Onnes [27]. Theoretical understanding of the phenomenon of superconductivity developed gradually during the next 50 years, eventually leading to the theory of Bardeen, Cooper, and Schrieffer, known as the BCS theory [28]. The application of superconducting technology for RF accelerator cavities began in 1965 with the acceleration of electrons in a lead-plated resonator at Stanford [29]. This initial demonstration was followed in the 1970s by projects using superconducting niobium cavities at Stanford, University of Illinois, CERN and Karlsruhe, Cornell, and Argonne [30]. These early projects were motivated by the promise of high accelerating ﬁelds with high-power efﬁciency. The early applications were for relativistic electron accelerators and storage rings, a particle separator, and a heavy-ion post accelerator for a tandem van de Graaff. For these early projects the accelerating ﬁeld levels were typically only about 2–3 MV/m. During the late 1970s and the decade of the 1980s, several important advances led to higher operating ﬁeld levels in the range of about 5–7 MV/m. The solution of multipacting problems by proper choice of the cavity geometry, combined with improved clean-room techniques and increases in the thermal conductivity of commercially available niobium, made it possible to obtain peak surface electric ﬁelds in superconducting cavities of 15–20 MV/m routinely in test facilities. At these levels, ﬁeld-emission-induced electron loading became an important ﬁeld limitation. Among the accelerator projects that closely followed these developments were two superconducting linac projects in the United States. First was the ATLAS superconducting linac at Argonne for

415

416

12 Special Structures and Techniques

the acceleration of heavy ions. Second was the superconducting recirculating electron linac at Thomas Jefferson National Accelerator Facility (TJNAF), formerly known as the continuous wave electron-beam accelerator facility or CEBAF. Also, in Europe were the electron-positron storage ring facilities, LEP at CERN, and HERA at DESY, while Japan had the TRISTAN facility at KEK. The spallation neutron source (SNS) superconducting linac at Oak Ridge National Laboratory accelerated a negative hydrogen ion beam to 1 GeV in 2007. It contains 11 medium-beta cryomodules with three 6-cell β = 0.61 cavities per cryomodule (186–379 MeV) followed by 12 high-beta cryomodules with four 6-cell β = 0.81 cavities per cryomodule (to 1 GeV).

Introduction to the Physics and Technology of RF Superconductivity

Next we summarize the physics and technology of RF superconductivity. A more complete discussion, relevant to accelerator applications, is found in book by Padamsee, et al. [31], and the article by Piel [32]. The present theoretical basis for understanding the phenomenon of superconductivity is provided by the BCS theory. Ordinarily, the interaction of the conduction electrons with the crystal-lattice vibrations results in ohmic energy dissipation producing heat. According to the BCS theory, superconductivity is the result of an attractive interaction between the conduction electrons through the exchange of virtual phonons, which leads to the formation of correlated electron pairs at temperatures below a critical temperature Tc , and below a critical magnetic ﬁeld. The correlated electron pairs, known as Cooper pairs, occupy the lowest energy state, which is separated from the lowest conduction electron state by a ﬁnite energy gap. The energy required to break up a Cooper pair and raise both electrons from the ground state is twice the energy gap, which is about 3 eV. Unless this amount of energy is provided, electrons bound in a Cooper pair cannot be put into a different energy state. Consequently, these electrons are essentially locked into the paired state, and behave like a superﬂuid, thus producing the phenomenon of superconductivity. As the temperature decreases below the critical temperature, the fraction of electrons that condense into Cooper pairs increases. At T = Tc , none of the electrons are paired, while at T = 0 all the electrons are paired. Between these temperatures two ﬂuids coexist, the superﬂuid of Cooper pairs, and the normal ﬂuid of conduction electrons. Superconductors exhibit zero dc resistance. However, for ac applications a superconductor is not a perfect conductor. A superconductor still experiences ohmic losses for time-dependent ﬁelds, because the Cooper pairs that are responsible for the superconducting behavior do not have inﬁnite mobility, and are not able to respond instantly to time-varying ﬁelds. Consequently, the shielding is not perfect for time-dependent ﬁelds. The ﬁelds in the superconductor attenuate with distance from the surface; the characteristic attenuation length is called the London penetration depth, which is of order

12.6 Superconducting RF Linacs

10−8 m in niobium. Then, the unpaired normal electrons that are always present whenever the temperature T > 0, are accelerated by the residual electric ﬁelds, and dissipate energy through their interaction with the crystal lattice. For the RF surface resistance of superconducting niobium, we use the approximate formula f 2 (GHz) Tc (12.1) + Rres Rs () = 9 × 10−5 exp −α T(K) T where α = 1.83, and Tc = 9.2 K is the critical temperature. The ﬁrst term is the BCS term, although the constant factor in the formula is really a function of many different variables. Rres is known as the residual resistance; it represents anomalous losses associated with imperfections in the surface, and typically lies in the range 10−9 − 10−8 . The superconducting surface resistance of niobium is of order 10−5 lower than that of copper. For accelerating ﬁelds in the megavolt per meter range, the power dissipated on the cavity walls is typically only a few Watts per meter. At low frequencies, usually below about 500 MHz, a convenient operating temperature is 4.2 K, the temperature of liquid helium at atmospheric pressure. At higher frequencies, the quadratic dependence of Rs with frequency in the BCS term makes it attractive to lower the operating temperature to typical values of 1.8–2.0 K, which lie below the λ point, where the helium is a superﬂuid. The most commonly used material for RF superconducting applications has been niobium. Niobium has the highest critical temperature (Tc = 9.2 K) of any element, can be machined and deep drawn similar to copper, and is commercially available in bulk and sheet metal forms. Because it getters gases easily, welding needs to be performed under vacuum, and electron-beam welding is primarily used. Superconducting cavities must be operated within a cryostat, where their cryogenic temperature is maintained, and the small RF power dissipated on the cavity walls is absorbed at low temperatures by a liquid-helium refrigeration system. The ac power required to operate the refrigerator must include the Carnot efﬁciency for an ideal refrigerator, plus an additional mechanical efﬁciency factor to represent irreversible effects in the real system. The total required operating power amounts to nearly 1000 times the power absorbed at low temperature. However, even including the refrigeration with its relatively poor efﬁciency, less total power is required compared with a normal-conducting accelerator, because of the large reduction in the superconducting surface resistance. The superconducting state in the presence of dc ﬁelds will exist if the dc magnetic ﬁeld is less than the critical ﬁeld, which at T = 0 is Hc = 2000 G for niobium. In the presence of RF ﬁelds, superconductivity will exist as long as the RF magnetic ﬁeld is below a critical superheating ﬁeld, which at T = 0 is Hsh = 2400 G in niobium. This ﬁeld level has not yet been attained in superconducting cavities. Typical maximum values range from about 500–1000 G. There is no known fundamental reason for the discrepancy, but the practical ﬁeld limitation is mostly the result of three effects, which

417

418

12 Special Structures and Techniques

are unrelated to the physics of RF superconductivity; electron multipacting, normal-conducting defects, and electron ﬁeld emission. Electron multipacting had been an early limiting factor in achieving high ﬁelds. Analysis showed that electron multipacting could be eliminated in the TM010 -pill-box-like resonators by using a geometry with a rounded shape, sometimes called the elliptical geometry [33, 34, 35, 36]. Another limiting factor in achieving higher ﬁelds is thermal instabilities produced by normalconducting surface defects. When such defects are heated in the RF ﬁeld, the temperature gradient produced across the cavity wall, between the liquidhelium-cooled outer surface and the inner cavity RF surface, can raise the temperature near the defect to a level that exceeds the critical temperature of niobium. Cavity stored energy will be dissipated near the defect. Increasing the thermal conductivity will raise the threshold ﬁeld for which these thermal instabilities will occur. The thermal conductivity is determined by the mobility of the normal-conducting electrons, which at these temperatures is limited by interstitial gas impurities from O, N, C, and H. Thus, improving the purity of the niobium is the way to increase the thermal conductivity. By the Wiedemann–Franz law the thermal and electrical conductivity of metals are related because the electron mobility affects both. It is customary to measure the purity of the niobium in terms of the residual resistivity ratio (RRR), deﬁned as RRR = electrical resistivity at room temperature divided by the resistance at 4.2 K in the normal-conducting state. The normal-conducting state in 4.2 K niobium can be obtained by applying a dc magnetic ﬁeld larger than the critical value. Increasing the thermal conductivity of niobium has been achieved through improvements in manufacturing procedures [37, 38]. Reﬁnements in the manufacturing process have produced factors of about 10 increase in RRR of commercially available niobium, corresponding to values today that are typically 250 and higher. These advances have resulted in many cavities that are no longer limited by defect-induced thermal instabilities, but are now predominantly limited by electron ﬁeld emission. At CERN, a parallel development of plated niobium on copper offers another approach to higher thermal conductivity. The third limiting factor has been electron ﬁeld emission, which produces electron loading at high ﬁelds. The onset of the ﬁeld emission occurs at lower surface ﬁelds than would be expected even after accounting for metallic protrusions with static electric-ﬁeld enhancements. Instead, most strong emission sites are believed to be associated with micron-size dust particles on the surface [39]. This means that to achieve high surface ﬁelds, it is important to maintain clean working conditions. Furthermore, it has been found that the number of ﬁeld-emitting sites can be reduced considerably after hightemperature bake-out to temperatures of about 1400 ◦ C. In general, chemical etching, electropolishing, rinsing with pure water, and attention to fabrication and assembly under clean conditions are characteristic requirements for mitigating the effects of ﬁeld emission and achieving high performance with superconducting cavities.

12.7 Examples of Operating Superconducting Linacs

The ﬁrst linac to use superconducting cavities was the electron recirculating linear accelerator at Stanford in 1972 [40]. A short time later, RF superconducting booster linacs for heavy-ion nuclear physics applications, were successfully developed [41]. These linacs typically accelerate heavy ions within the velocity range from about β = 0.01–0.20. Ions with a large range of charge-to-mass ratio are injected into the linac either by a tandem van de Graaff electrostatic accelerator or an ion source on a high-voltage platform. The ion beams have been accelerated up to energies of 25 MeV/nucleon.

Atlas

A well-known example is the ATLAS continuous-wave or CW superconducting linac booster at Argonne National Laboratory, which accelerates heavy ions with mass as large as uranium [42]. The linac consists of independently phased superconducting cavities, and uses superconducting solenoid magnets for focusing. The independent phasing of the cavities provides the ﬂexibility needed to accelerate beams with a wide range of charge-to-mass ratio. The low frequencies (48.5, 72.75, and 97 MHz) provide linear longitudinal focusing across the bunch, allowing the linac to attain the longitudinal beam quality required for precision nuclear physics experiments, producing either very lowenergy spread or short subnanosecond pulses for time-of-ﬂight measurements. To keep the cavity sizes manageable at these frequencies, the cavities are loaded capacitively and inductively to resemble lumped circuits. The cavities include quarter-wave resonators loaded with drift tubes for acceleration of the lowest velocity particles, and split-ring resonators for acceleration of the highervelocity particles. Cavity design velocities include β = 0.06, 0.1, and 0.16. The split-ring resonators consist of hollow niobium drift tubes mounted onto the ends of a split ring, which is installed in a cylindrical outer housing made of niobium that is explosively bonded to copper. Because the beam current is very low, the loaded Q can be very large, and the RF power requirement is small. The small bandwidth of the high-Q cavities means that the resonators are sensitive to mechanical vibrations. To achieve good phase control the resonators must be mechanically stable, and must use a fast feedback system to compensate for the vibrations. The properties of the ATLAS linac are summarized in Table 12.1.

CEBAF

CEBAF is the CW recirculating linac at the Thomas Jefferson National Accelerator Facility (TJNAF) [44]. The beam is circulated through the linac

419

420

12 Special Structures and Techniques Table 12.1 Parameters of ATLAS and CEBAF superconducting Linacs [43] Facility Application First beam Species Average beam current Duty factor Output energy

Beam microstructure Accelerating structures

Operating temperature Length

ATLAS at Argonne

CEBAF at TJNAF

Heavy-ion linac for nuclear and atomic physics 1978 Lithium through uranium 50–500 nA (electrical) CW 5–17 MeV/nucleon in normal operation

Recirculating electron linac for nuclear physics 1994 Electrons 50 µA CW 2 antiparallel 400-MeV linacs used as a recirculating linac to achieve a nominal 4 GeV in 5 passes. 1497 MHz 320 5-cell elliptical cavities, each driven by a 5-kW klystron.

12.125 MHz 18 quarter-wave resonators, 48.5 and 72.75 MHz, 2.7-kW RF power; 44 split-ring resonators, 97.0 MHz, 6-kW RF power. 4.2 K 34.5 m

2K 320 cavities, each with 0.5 m active length.

ﬁve times to achieve a nominal output energy of 4 GeV. The recirculation option is attractive because it provides substantial savings in capital and operating costs for RF power. Recirculation is an option for accelerators of relativistic particles, because the beam velocity is essentially ﬁxed at almost the speed of light, and therefore the effective transit-time factor of the multicell cavities is large for every cycle. The subject of recirculating accelerators is not treated in this book. For those who are interested, we recommend the book by R.E. Rand, Recirculating Electron Accelerators (Harwood Academic Publishers, 1984). The choice of a CW superconducting linac, compared with the most attractive normal-conducting option consisting of a pulsed linac followed by a storage ring for pulse stretching, enjoys advantages of lower capital and operating costs, higher average currents, fully continuous beams, and much higher beam quality. The injector consists of a 100-keV electron source followed by a 45-MeV superconducting linac, using 18 superconducting cavities. The superconducting cavity design chosen was the proven 5-cell elliptical-geometry design produced by Cornell, and shown in Fig.12.7. The design speciﬁcations, 5-MV/m accelerating gradient, and Q0 = 2.4 × 109 at 2 K, were easily met. The average accelerating gradient of the installed cavities is about 8 MV/m (mostly limited by

12.7 Examples of Operating Superconducting Linacs

Figure 12.7 Five-cell 1.497-GHz niobium elliptical cavity for the CEBAF recirculating linac at TJNAF showing the waveguide coupling port on the left and two

higher-order mode couplers on the right. [Photo Courtesy of P. Kneisel, Thomas Jefferson National Accelerator Facility.]

ﬁeld emission), and the average Q0 = 5 × 109 . Some properties of the CEBAF linac are summarized in Table 12.1. Plans call for an upgrade to 12 GeV, which requires performance improvements in new 7-cell RF cavities [45].

Spallation Neutron Source

The SNS is an accelerator based neutron source that was designed by six national laboratories (Fig. 12.8), and constructed at Oak Ridge National Laboratory for the U.S. Department of Energy. At its full beam power of

Front-end systems (lawrence berkeley)

Accumulator ring (Brookhaven) Target (Oak ridge)

Linac (Los alamos and jefferson) Instrument systems (Argonne and oak ridge)

00-04492D/arb

Figure 12.8 Drawing of the SNS facility showing the individual systems and the laboratories that designed them.

421

422

12 Special Structures and Techniques

1.4 MW, SNS will provide the world’s most intense pulsed neutron beams for scientiﬁc research and industrial development. In February, 2007 the SNS accelerated beam to its design energy of 1 GeV, a new world energy record for a proton linear accelerator. The SNS accelerator system consists of a negative hydrogen (H− ) ion source, a low-energy beam-transport system with a ﬁrst stage beam chopper, followed by a normal-conducting room-temperature copper linac. The function of the beam chopper is to provide a beam-free gap in the accumulator ring to reduce beam loss during beam extraction from the ring. The normal-conducting linac consists of a four-vane RFQ (Fig. 12.9) for acceleration to 2.5 MeV, a medium-energy beam-transport system with a second-stage chopper, a 6-tank Alvarez DTL (Fig. 12.10) for acceleration to 87 MeV, and a four-module coupled-cavity linac (Fig. 12.11) for acceleration to 186 MeV. This is followed by the superconducting linac comprised of 11 medium-beta cryomodules with three 6-cell β = 0.61 cavities per cryomodule (to 379 MeV) and 12 high-beta cryomodules with four 6-cell β = 0.81 cavities per cryomodule (to 1 GeV) shown in Fig. 12.12. The superconducting niobium linac cavities are cooled with liquid helium to an operating temperature near 2 K (Fig. 12.13). The SNS linac delivers a 1-ms-long 38-mA peak current chopped beam pulse at 60 Hz to the 248-m diameter accumulator ring injecting the beam in multiple turns. In this way the accumulator ring compresses the 1-ms pulse to ∼ 700 ns for delivery by a high-energy beam transport line to the mercury target, where neutrons are produced by spallation. The neutron energies are moderated to useable levels by supercritical hydrogen and water moderators before delivery to 24 beam lines for experiments.

Figure 12.9 402.5 MHz four-vane RFQ that bunches and accelerates the beam from 75 keV to 2.5 MeV.

12.8 Future Superconducting Linac Facilities

Figure 12.10 402.5-MHz Alvarez drift-tube linac that accelerates the beam from 2.5 MeV to 87 MeV.

Figure 12.11 805-MHz coupled-cavity linac that accelerates the beam from 87 MeV to 186 MeV.

12.8 Future Superconducting Linac Facilities

Superconducting-linac technology has matured sufﬁciently in recent years that all new RF linac projects are now examined for consideration of a superconducting option. We discuss three applications of superconducting linacs projects currently under study.

International Linear Collider

The International Linear Collider (ILC) is a proposed future international particle-accelerator facility that would produce high-energy particle collisions

423

424

12 Special Structures and Techniques

Figure 12.12 805-MHz SNS superconducting linac operates near 2 K and accelerates the beam from 186 MeV to 1000 MeV.

Figure 12.13 Six-cell 805-MHz medium-beta (β = 0.61) and high-beta (β = 0.81) superconducting niobium cavities designed for the SNS.

between electron and positron beams at center-of-mass energies from 200 to 500 GeV [46]. The machine would also be upgradable in a second phase to 1 TeV. The present design is based on two 11-km-long superconducting linacs operating at 1.3 GHz at an operating temperature near 2 K, and with an average accelerating gradient of 31.5 MV/m (Fig. 12.14). The linacs will be comprised of 9-cell superconducting niobium cavities similar to that shown in Fig. 12.15. The total machine length in the ﬁrst stage is 31 km long, and the machine would produce a peak luminosity of about 2 × 1034 / cm2 /s at a center-of-mass energy of 500 GeV. The site length would have to be extended in the second stage to reach 1 TeV in the center of mass. The design achieves this high luminosity through a combination of small emittance beams and

Main linac

Positron DR Main linac

Main linac

Second stage 5-GeV Electron linac DR Undulator

Positron

Main linac

5-GeV linac

First stage 5-GeV Electron linac DR Undulator

5-GeV linac

12.8 Future Superconducting Linac Facilities

Figure 12.14 Block diagram of the proposed ILC facility showing the two stages of the project.

Figure 12.15 The ILC will use 9-cell superconducting niobium cavities similar to this one. It will operate near 2 K at a 1.3-GHz RF frequency.

high beam power. The beam pulse length is 1.6 ms, the repetition rate is 5 Hz, the duty factor is 0.5%, and the availability goal is 75%. The electrons are injected from an electron source using a photocathode gun, and the positrons are produced by photoproduction using an undulator-based photon source, driven by a 150-GeV electron beam. The particles will be injected into 5-GeV electron and positron damping rings before being injected into the main linacs. A 4.5-km beam-transport system at the ends of the linacs will bring the two beams into collision at a 14-mrad crossing angle. The R&D effort must focus on achieving high accelerating gradients for the 9-cell superconducting cavities. To ensure an average operating gradient of 31.5 MV/m, the designers wish to achieve 35 MV/m with Q greater than 0.8 × 1010 in acceptance tests during mass production of the cavities. The choice of the operating frequency is a balance between the high cavity cost due to greater cavity size at lower frequency, and a lower sustainable gradient at higher frequency resulting from the increased BCS surface resistance, which causes heating and a global thermal instability that occurs below the RF critical ﬁeld [47]. In addition, the availability of klystrons limits the frequency choice; 1.3 GHz satisﬁes all the criteria. The choice of average accelerating gradient of 31.5 MV/m assumes at the present time that further progress will be made in achieving improved cavity performance. The total number of linac components includes 14,460 superconducting cavities and 560 superconducting quadrupoles.

425

426

12 Special Structures and Techniques

Next-Generation Rare Isotope Facility

A schematic drawing of a next-generation facility for rare isotope beams, previously called the rare isotope accelerator (RIA) is shown in Fig. 12.16. A superconducting heavy-ion driver linac would accelerate high-intensity beams of stable heavy ions for the production of rare isotopes. The driver linac could deliver beams from protons to uranium at energies of several hundred MeV per amu to one or more targets where the rare isotopes are produced. These rare isotopes would be extracted from the target and could be accelerated in a second superconducting linac called the post accelerator. The facility would enable a wide range of studies including measurements of nuclear reactions at astrophysical energies and searches for new heavy elements with long lifetimes. The driver linac is a CW superconducting linac designed for simultaneous acceleration of stable multiple charge states of heavy-ion elements with masses up to 240 amu. The linac is comprised of independently phased superconducting cavities that provide a wide velocity acceptance as required for acceleration of a wide range of ions with a range of charge states for any ion mass. The beams from an electron cyclotron resonance or ECR ion source are injected into a low frequency RFQ ( 50 MeV/u)

RIA concept

In-flight fragment separation

Production targets Driver linac 400 MeV/u U, 900 MeV p

Ion sources/ preacceleration

1 Structure and reactions E < 15 MeV/u

Secondary linac 2 Astrophysics E < 1 MeV/u

Gas stopping

3 No acceleration: traps, laser spec., and so on

Isotope recovery

Reaccelerated beams

Figure 12.16 The proposed rare isotope accelerator (RIA) shown in this schematic drawing is an example of a next-generation nuclear physics facility for rare isotope beams. Shown are the high-power superconducting driver linacs, which accelerate stable beams from protons to

uranium to the target region where the rare isotopes are produced in one or more targets. The rare isotopes are extracted from the target and can be accelerated to higher energies in a second superconducting linac for a variety of nuclear physics and astrophysics experimental studies.

12.8 Future Superconducting Linac Facilities

10 MeV/amu and one near 85 MeV/amu. The net voltage gain of the driver linac was 1.4 GeV. The nominal ﬁnal beam energy was 400 MeV/amu and the beam power was 400 kW. Different types of resonators available for use in different velocity regions of the driver linac are shown in Fig. 12.17.

Free-Electron Lasers

This topic could have been included in Section 12.7 as an example of operating superconducting linacs, but I have decided to place it in Section 12.8 as future

(a)

(b)

0 (c )

9 10

Scale (in.)

Figure 12.17 Example of superconducting cavities that have been proposed for use in the driver linac including low velocity (a), medium velocity (b), and high-velocity (c) structures.

427

428

12 Special Structures and Techniques

superconducting linac facilities, because of the great likelihood of further signiﬁcant expansion of this ﬁeld. The free-electron laser (FEL) produces intense beams of coherent electromagnetic radiation, and unlike other types of lasers its wavelength is tunable over a wide range. The FEL uses a relativistic electron beam from an accelerator, usually a linac. The electron beam is transported through a periodic alternating-polarity permanent-magnet array called a wiggler or undulator, which induces a transverse sinusoidal oscillation in the electron beam, causing it to emit synchrotron radiation in a forward cone (Fig. 12.18) at a resonant wavelength that depends on the period and magnetic ﬁeld of the wiggler, and the electron-beam energy. The photon wavelength of the FEL radiation is given by a resonance condition λph

λu = 2γ 2

K2 1+ 2

(12.2)

where λu is the undulator period, γ is the usual relativistic factor, K = eBu λu /2πmc is the undulator parameter, e and m are the electron charge and mass, c is the speed of light, and Bu is the peak magnetic ﬁeld of the magnets in the undulator. Equation 12.2 describes the condition that the electrons, while traveling through one full period of the undulator, slip by one wavelength of the radiation relative to the faster electromagnetic wave that is propagating parallel to the undulator axis. Equation 12.2 shows that in principle the resonance condition can be satisﬁed for arbitrarily small photon wavelength by increasing the electron-beam energy. The radiation can build up to saturation by inserting the undulator between a pair of normal-incidence mirrors (not shown) to create an optical resonator in which the radiation level builds up from multiple reﬂections with a properly synchronized electron beam to add more radiation in each pass. This works well over a wide range of wavelengths from infrared to ultraviolet and X-ray

N S S N N S N S N S S N N S S N N S S N N S N S N S S N N S S N

N S S N N S N S S N S N N S S N N S N S S N S N

Figure 12.18 A schematic drawing showing an electron beam from an accelerator passing through a wiggler or undulator comprised of an alternating-polarity permanent-magnet array. The wiggler

induces an oscillation of the electron beam causing it to emit coherent electromagnetic radiation that propagates in the forward direction and adds to the radiation from previous cycles as shown in the insert.

12.8 Future Superconducting Linac Facilities

wavelengths. The electron beam must be produced with a small emittance, and the emittance must be preserved both in the linac and in the transport line to the wiggler. This is because efﬁcient lasing requires high overlap between the electron and photon beams, which is expressed by the condition that the unnormalized emittance ε of the electron beam must satisfy ε < λph /4π. The two main directions for future FEL development are (1) shorter wavelengths into the vacuum ultraviolet or VUV and X-ray regimes, and (2) higher average power. At shorter wavelengths below about 200 nm, high reﬂectivity mirrors are not available. In this case a single – pass method is used based on the principle called self-ampliﬁed spontaneous emission (SASE). For a SASE FEL a long undulator is used, in which the spontaneous radiation emitted in the ﬁrst part of the undulator interacts with the electron beam in second part causing ampliﬁcation by stimulated emission at the resonant wavelength. The SASE method does not require mirrors, does not require an external input signal, and is an attractive solution for the short wavelength VUV and X-ray regimes. Development of the photocathode electron gun for production of highbrightness electron beams represents a signiﬁcant technical advance. In this case, the photocathode is placed in an RF cavity with a high electric ﬁeld and the electrons produced by photoproduction are rapidly accelerated within the cavity to relativistic energies, greatly reducing the effect of the nonlinear space-charge forces on the beam emittance. In addition, a beam-dynamics design approach called space-charge compensation [48] is effective in cancellation of much of the remaining space-charge-induced emittance growth. The highest priority for the design of the main accelerator is the control of emittance growth particularly from wakeﬁelds, coherent synchrotron radiation, RF input-coupler forces on the beam, and misalignments. The need for high accelerating gradients is of less importance for FELs than for the linear collider application. At the present time two major SASE FEL projects are (1) the Linac Coherent Light Source (LCLS) at SLAC that uses the ﬁnal third of the two-mile SLAC linac, and (2) the European XFEL at DESY that will use a 20-GeV superconducting linac. A list of three key design parameters for these two facilities is given in Table 12.2. For producing higher average radiated FEL power, the most important development has been that of the energy recover linac (ERL) combined with the use of a superconducting linac. This development is motivated by the fact that only a small fraction of the electron-beam energy is converted to Table 12.2 Some Design Parameters for SASE FEL Projects Parameter Wavelength Electron-beam energy Normalized RMS emittance

LCLS

XFEL

1.5 A˚ 14.35 GeV 1.2 mm-mrad

1.0 A˚ 20 GeV 1.4 mm-mrad

429

430

12 Special Structures and Techniques tor

Injec

IR wiggler Linac Optical system Dump

Electron recirculation beam line

Figure 12.19 Schematic diagram of the original FEL at Jefferson laboratory with its photoinjected, superconducting RF linac with energy recovery. After passing through

the superconducting linac and the wiggler, the beam is transported back in a recirculation loop and reinjected in the linac at a deceleration phase.

radiation as the electron beam passes through the wiggler. At high-power levels, dumping the output electron beam after it passes through the wiggler with nearly its full power would result in high levels of radioactivity and severe cooling problems. For a practical FEL device it is desirable to recover as much of the unused electron-beam energy as possible rather than covert all this energy to heat. In the ERL approach the electron beam, after passing through the wiggler, is reinjected into the same linac using an injection phase that decelerates the beam. An example of the FEL at Jefferson laboratory is shown in Fig. 12.19, where most of the electron-beam energy is converted to microwave energy that can be used simultaneously for acceleration of new bunches of electrons [49]. The used electron beam emerges from the linac with only a small fraction of its full energy, and is then transported to the beam dump, where the radioactivity and cooling problems are reduced and are now more manageable. Because of the high beam current, a potential beamphysics problem is the multipass beam breakup or BBU instability caused by higher-order modes that deﬂect the beam (see Section 11.18). Higher-order mode couplers combined with dampers to absorb the energy in these modes can be used to lower the Q of the modes to raise the instability thresholds. There are also beam optics solutions that can mitigate the BBU effect. The Jefferson laboratory infrared FEL has demonstrated ERL operation in CW mode at high power [50]. In July 2004, 10 kW of radiated power was produced at 6 µm wavelength and in October 2006, 14.2 kW CW was achieved at 1.6 µm. References 1 Good, M.L., Phys. Rev. 92, 538 (1953). 2 Fainberg, Ya.B., Proc. CERN Symp.

High Energy Accel. Pion Phys. Geneva 1956. 3 Kapchinskiy, I.M., Theory of Resonance Linear Accelerators, Harwood

Academic Publishers, New York, 1985, pp. 184–199. 4 Swenson, D.A., Part. Accel. 7, 61 (1979). 5 Cheng, W.H. et al., Alternating Phase Focusing Including Space Charge, 1996

References

7 8

9 10 11

17 18

Linac Conf., Chalk River, Ontario, Canada, August, 1992. Wells, N., Radio Frequency Quadrupole and Alternating Phase Focusing Methods used in Proton Linear Accelerator Technology, prepared for DARPA, published by Rand Corporation, January, 1985. Vladimirskii, V.V., Prib. Tekh. Eksp. 3, 35 (1956). Tepliakov, V.A., Proc. 1992 Int. Linac Conf., Ottawa, Ontario, Canada, August 24–28, 1992, p. 21. Gorshkov, B.M. et al., Sov. Phys. Tech. Phys. 22(11), 1348 (1977). Lapostolle, P.M., Compt. Rend. 256, 5294 (1963). Muller, R.W., 1979 Linear Accel. Conf., Brookhaven National Laboratory Report BNL-51143 1979, p. 148. Muller, R.W. et al., 1984 Linear Accel. Conf., GSI Report GSI 84-84-11 1984, p. 77. Swenson, D.A., RF-Focused Drift-Tube Linac Structure, Proc. 1994 Int. Linac Conf., Tsukuba, Japan. Maschke, A.W., Brookhaven National Laboratory Report BNL 51029, 1 June, 1979. Adams, R. et al., Brookhaven National Laboratory Report BNL 27128, December, 1979. Thomae, R.W. et al., High Current, High Brightness, and High-Duty Factor Ion Injectors, Gillespie, G.H., Kuo, Y.Y., Keefe, D. and Wangler, T.P, AIP Conf. Proc. 139, 95 (1986). Billen, J.H., Proc. 1994 Int. Linac Conf., Tsukuba, (1994) p. 341. Young, D.E., Proc. Heavy Ion Fusion Workshop, Brookhaven National Laboratory Report BNL 50769, 1977, p. 17; also Custom R., et al.ibid. 94; also Swenson, D.A., Proc. Heavy Ion Fusion Workshop, Lawrence Berkeley Laboratory Report LBL-10301, 1979, p. 239. Bongardt, K., Proc. 1984 Linac Conf., Gesellschat fur Schwerionforschung report GSI-84-11, 1984, p. 389. Guy, F.W., in High Current, High Brightness, and High-Duty Factor Ion Injectors, ed. Gillespie, G.H., Kuo,

Y.Y., Keefe, D., and Wangler, T.P; AIP Conf. Proc. 139, 207 (1986); also Guy, F.W. and Wangler, T.P., ibid. 185. Johnson, K.F. et al., Proc. 1990 Int. Linac Conf., Albuquerque, NM, 1990 p. 701. Lawrence, G.P. et al., Conventional and Superconducting RF Linac Designs for the APT Project, Proc. 1996 Int. Linac Conf., Geneva, Switzerland, August 26–30, 1996; also Wangler, T.P., et al., Proc. 1990 Int. Linac Conf., Albuquerque, NM, 1990, p. 548. Gardner, I.S.K., Lengeler, H. and Rees, G.H., Outline Design of the European Spallation Neutron Source, ESS 95-30-M, September, 1995. Stokes, R.H. and Minero, G.N., High Current, High Brightness, and High-Duty Factor Ion Injectors, ed. Gillespie, G.H., Kuo, Y.Y., Keefe, D., and Wangler, T.P; AIP Conf. Proc. 139, 79 (1986). Farkas, Z.D., Hogg, H.A., Loew, G.A., and Wilson, P.E., Proc. 9th Int. Conf. High Energy Accel., SLAC, 1974, p. 576. Loew, G.A., and Talman, R., Elementary Principles of Linear Accelerators, AIP Conf. Proc. No. 105, American Institute of Physics, New York, 1983, p. 89. Kamerlingh, H., Onnes, Comm. Phys. Lab., Univ. Leyden, Nos. 199, 120, 122 (1911). Bardeen, J., Cooper, L.N., and Schrieffer, J.R., Phys. Rev. 108, 1175 (1957). Pierce, J.M., Schwettman, H.A., Fairbank, W.M., Wilson, P.B., Proc. 9th Int. Conf. on Low Temp. Phys., Part A Plenum Press, New York, 1965, p. 36. For a complete list of references for the application of rf superconductivity to particle accelerators, see the excellent articles by H. Piel and by H. Lengeler in the CERN Accelerator School for Superconductivity in Particles Accelerators, ed. S. Turner, CERN 89-04, 10 March, 1989, pp. 149–229. Paadamsee, Hasan, Knobloch, Jens, and Hays, Thomas, RF

431

432

12 Special Structures and Techniques

36 37

40 41

Superconductivity for Accelerators, John Wiley & Sons, New York, 1975, 1998. Piel, H., CERN Accelerator School for Superconductivity in Particles Accelerators, ed. Turner, S., CERN 89-04, 10 March, 1989, pp. 158–165. Lyneis, C.M., Schwettman, H.A., and Turneaure, J.P., Appl. Phys. Lett. 31, 541 (1977). Lagomarsino, V., Manuzio, G., Parodi, R., and Vaccarone, R., IEEE Trans. MAG-15, 25 (1979). Klein, U., and Proch, D., Proc. Conf. on Future Possibilities for Electron Accel. Charlottesville Va, 1979, McCarthy, J.S., and Whitney, R.R., University of Virginia, Charlottesville, 1979, pp. 1–17. Kneisel, P., Vincon, R., and Halbritter, J., Nucl. Instr. Meth. 188, 669 (1981). Kneisel, P., Amato, J., Kirchgessner, J., Nakajima, K., Padamsee, H., Phillips, H.L., Reece, C., Sundelin, R., and Tigner, M., IEEE Trans. MAG-21, 1000 (1985), see also p. 149, 1007; and Kneisel, P., Cornell University, SRF 840702 1984. Lengeler,H., Weingarten, W., Muller, G., and Piel, H., IEEE Trans. MAG-21, 1014 (1985). Piel, H., CERN Accelerator School for Superconductivity in Particles Accelerators, CERN 89-04, 10 March, 1989, ed. Turner, S., pp. 177–179. McAshan, M.S. et al., Appl. Phys. Lett. 22, 605 (1973). Bollinger, L., Nucl. Instum. Meth. A244, 247 (1986); Proc. 1987 Part. Accel. Conf., IEEE Cat. No. 87CH2387-9 (1987). Shepard, K.W., Proc. 4rd Workshop on RF Superconductivity (KEK,

Tsukuba, Japan), Y. Kojima, ed. KEK Report o. 89-21 1990. Grunder, H.A., Bisognano, J.J., Diamond, W.I., Hartline, B.K., Leemann, C.W., Mougey, J., Sundelin, R.M., and York, R.C., Proc.1987 Part. Accel. Conf., Washington, D.C., IEEE Cat. No. 87CH2387-9, 1987, p. 13; Dunham, B.M., Jefferson Lab, A Status Report, Proc. 18th Int. Linac Conf., Geneva, Switzerland, 26–30 August, 1996, eds. \C. Hill and M. Vretenar, CERN 96-07, November, 15 1996, p. 17. Delayen, J.R., Upgrade of the CEBAF Acceleration System, Proc. 1999 Part. Accel.Conf., New York, 1999, pp. 3498–3500. Clendenin, J., Rinolﬁ, L., Takata, K., and Warner, D.J., Compendium of Scientiﬁc Linacs, 18th Int. Linac Conf., Geneva, Switzerland, August 26–30, 1996, CERN/PS 96-32 (DI), November, 1996. ‘‘International Linear Collider Reference Design Report 2007,’’ ILC-Report 2007-01, April, 2007. Padamsee, Hasan, Knobloch, Jens, and Hays, Tom, RF Superconductivity for Accelerators, Wiley, New York, 1998, p. 223. Carlsten, B.E., Nucl. Instrum. Meth. Phys. Res., A285, 313 (1989); Carlsten, B.E., Part. Accel., 49, 27 (1995). Tigner, M., A possible apparatus for electron-clashing experiments, Nuovo Cimento, 37, 1228–1231 (1965). Bogacz, A. et al.CEBAF Energy Recovery Experiment, Proc. 2003 Part. Accel. Conf., Portland, Oregon, May 12–16, 2003, pp. 195–197.

433

Index

a Accelerating cell array of 176 RFQ transition cell 266–272 Accelerating ﬁeld ion linac, single particle acceleration 175–199 Accelerating gap coordinate transformation 208–209 radial impulse near axis 204–207 standing-wave electric-ﬁeld solution 205 Accelerating gradient 34, 401 constraint on 333 in multicell cavity 34 longitudinal beam dynamics constraint 333–335 two competing effects 87 Accelerating mode free-space wavelength 393 generator-induced, and beam-induced ﬁelds 341 Accelerating structure higher-frequency Alvarez DTL 89 principle wave, with largest Fourier amplitude 57 quarter-wave resonator 24, 88 resonant coupled 65 subtraction technique 375 TEM mode, for superconducting linacs 121 using electric focusing 406–410 with periodic geometries, and phase velocity 53 Accelerating structures, heavy-ion 46 Accelerating system and accelerators 1–31 Accelerating waveform energy difference 388

Acceleration 5 efﬁciency of, bunches 4 space-charge effects 247 synchronous, radial position and particle velocity 241–242 Acceleration cavity power and acceleration efﬁciency, ﬁgures of merit 42–44 Accelerator 2, 5, 14 current per RF bucket 10 driven ﬁssion-reactor concepts 13 linear accelerator 1–2 linear focusing forces, trajectory ellipse 284 little machine, the ‘linac’ 1 normalized emittance 287 parameters, cavity tuning and RF system 13 particle accelerator 2 Accelerator cavity schematic drawing, RF system 158 Accelerator cavity impedance typical frequency spectrum 380 Accelerator production of tritium (APT) 13 Adiabatic bunching beam-current capacity 233 Adiabatic phase damping phase-space ellipses 187–188 Alternating-phase focusing (APF) 405 Alvarez drift-tube linac 7, 91–96 DTL cavity 92 ﬁeld-free regions 91 medium-velocity ion, acceleration of 8 Amp´ere’s law 39 Anisotropic beam physics of 320

RF Linear Accelerators. 2nd, completely revised and enlarged edition. Thomas P. Wangler Copyright  2008 Wiley-VCH Verlag GmbH & Co. KGaA, Weinheim ISBN: 978-3-527-40680-7

434

Index Anisotropic linac beams longitudinal-transverse coupling 319–325 APF see Alternating-phase focusing 405 APT see Accelerator production of tritium 13 Asymmetrical drift tube quadrupole focusing ﬁelds 408 Axial electric ﬁeld 41 average 93 Axial energy gain 39 Axial ﬁeld distribution 41 Axial transit-time factor aperture-dependent factor 40 gap, independent 42 gap-dependent factor 40 Azimuthal magnetic ﬁeld pattern in four-vane RFQ 254

b BBU see Beam-breakup 362 BCS theory phenomenon of, superconductivity 416 Beam 7, 23, 24 acceleration by, RF electromagnetic ﬁelds 2 and emittance 12 and longitudinal focusing 11 and vacuum system 6 beam-loading compensation 13 energy gain 23 isodensity contours of 284 lighter protons and electrons 4 low velocity, image-charge forces 362 multiparticle effects 12–13 net energy gain 4 phase space 12 Beam axis 251 and magnetic-ﬁeld direction 390 nominal gap-to-gap spacing 90 Beam breakup bunch shape 386 Beam bunches 6, 8 Beam cavity conﬁguration after ﬁrst charge 343 ﬁnal conﬁguration 343 Beam current beam image current, and beam-induced voltage 348 Beam current, acceleration 10 Beam displacement 364 Beam distribution proper matching 12

Beam dynamics aperture radius 45 linear space-charge ﬁeld 300–301 Beam ellipse and beam matching 222–223 in RFQ 272–274 Beam envelope phase-space plots 262 Beam frame transverse deﬂection, for accelerated relativistic beam 229 Beam funnel schematic diagram 412 simulation studies, of funneling lines 413 Beam funneling concept of 411–413 Beam halos 283 and beam loss 325–329 Beam holes 44 Beam loading 12, 341–360 and cavity detuning 349 and wake ﬁelds 282 fundamental theorem 343 in accelerating mode 345–346 standing-wave cavity 341 Beam loss 12 Beam matching current-independent beam matching 224 Beam particle axial-ﬁeld pattern 236 transverse defocusing 12 transverse focusing 12 Beam pipe decoupling of cavities 116–117 ﬁeld lines, terminating on pipe 365 Beam velocity increased cavity, and cryostat lengths 87 Beam, continuous 3 Beam, ideal laminar beam 283 Beam-breakup (BBU) 362 Beam-cavity interaction 356–360 interaction problems 356 Beam-centroid phase 384 Beam-coupling impedance 378–380 single frequency source 378 Beam-dynamics codes numerical space-charge calculations 292–296

Index Beam-dynamics design approach, space-charge compensation 429 Beam-halo experiment at Los Alamos 329–331 Beam-induced ﬁeld 12 Beam-loaded cavity equations 347–352 equivalent circuit 347 extreme, beam-loaded case 351 Wilson theory 347 with bunches 351 Beam-loading compensation 346 Beam-loading ratio 43 Beam-loading theorem fundamental 342–343 initial conﬁguration 342 surface charge, and beam-induced voltage 342 Beam-particle energy accelerator-physics units unit conversions, and physical constants 16 Beam-plasma frequency space-charge-dominated beams 291 Beat pattern describing difference 388 Bessel functions 40, 38, 56 and modiﬁed Bessel function, zero order 38 Bessel’s equation 24, 37 Bethe’s theory coupling of cavities, through apertures 53 BNS damping 387 Boltzmann–Ehrenfest theorem adiabatic invariants, of oscillator 135 Boron–neutron capture therapy 13 Bunch ﬁelds carrying, surface induced charges 361 Bunch loss parameter arbitrary charge distribution 377 Bunch wake potential arbitrary line-charge density 372 in arbitrary charge distribution 371–376 loss parameters, particular charge distribution 376–377 Bunch-centroid displacement center-to-center cavity spacing 393 showing cumulative BBU 394 Buncher cavity phase-dependent kick 195

Bunching 4 longitudinal phase-space ellipse, of incident bunch 195 Bunching effect 194

c Capital cost in accelerator design 47 Cartesian coordinates 209 Cavity electrical center 345 resonance driven, forward and reﬂected traveling-wave ﬁelds 351 Cavity array large apertures 86 Cavity design cross section of cavity, nose cones and spherical outer wall 44 issues of 44–46 Cavity ﬁeld and loss factor 341 beam effect 341 circular aperture, and aperture radius 70 transient build up 153 Cavity geometry adiabatic invariant of, oscillator 164–165 and accelerating ﬁelds 414 ﬁeld-emission-induced electron loading 415 Cavity loading with lumped elements 46 Cavity mode 12 deﬂecting modes 12 Cavity parameters designing linac, and operating frequency 46 frequency scaling 46–47 Cavity parameters, conventional 154 Cavity power dissipation 154 Cavity resonant frequency detuning of 350 Cavity resonant-frequency shift 258 Cavity resonator circuit model of 136 coupled, two external circuits 144 transverse-magnetic mode of, circular cylindrical cavity 24 Cavity shunt impedance 51 Cavity stored energy SUPERFISH 344

435

436

Index Cavity system wave description, waveguide-to-cavity coupling 148–156 Cavity time constant heavily, beam-loaded superconducting cavity with bunches 352 Cavity voltage 342 versus time 355 Cavity wake potential Fourier transform 381 Cavity walls ohmic losses 138 Cavity-power dissipation 344 point charge, loses energy 345 CCDTL see Coupled-cavity drift-tube linac 410 CCL see Coupled-cavity linacs structure 83 CEBAF recirculating linac 419–420 ﬁve-cell 1.497-GHz niobium elliptical cavity 421 CEBAF see Continuous wave electron beam accelerator facility 416 Cell length synchronous acceleration 41 Cell resonant-frequency perturbation ﬁrst-order deviations, from nominally ﬂat ﬁeld amplitude 107 CERN SPL design trajectory points 323 Circuit with resonant coupling element 64 Circuit impedance for driving frequency 381 Circuit model dispersion relation 65 dispersion relation for 64 of single mode of cavity 381 periodic array of cavities, with resonant coupling 66 Circular accelerator 10 relativistic electrons 13 Coaxial half-wave resonator and single loading element 123 Coaxial quarter-wave resonator 122 Coaxial resonator 22, 23 and accelerating cavities 22–24 coaxial line application 23 with voltage, and current standing waves 24 Colliding beams 2 Computed transverse δ-function wake potential 372

Conductor boundary conditions, and conducting walls 19–20 resistivity 19 surface charge density 19 Constant-energy beam 385 Constant-impedance structure constant-gradient 74–76 Continuous wave (CW) 5, 50 ATLAS superconducting linac booster 419–420 pulsed or continuous operation 10 reducing, space-charge 10 Continuous wave electron beam accelerator facility (CEBAF) 416 Cooper pairs correlated electron pairs 416 Coordinate system relative to reference plane 373 Copper-cavity technology superconducting niobium cavities 10 Cosine functions orthogonality of 37 Coulomb ﬁeld collisional regime, and space-charge regime 290 Coupled cavity periodic arrays of 58 Coupled circuits electrically, periodic array 62–63 magnetically, periodic array 63–64 Coupled multicell TM cavities Alvarez DTL and CCL 122–125 Coupled oscillators 58 and time dependence 105 closest-mode spacing 106 identical internal oscillators and half-cell 105–108 Coupled three-cavity system effects of, power losses 132 Coupled-cavity array 7 Coupled-cavity drift-tube linac (CCDTL) 410–411 with single drift tube 411 Coupled-cavity linac (CCL) 8, 83 design of, biperiodic CCL structures 111–114 dispersion curve of, biperiodic structure 112 resonant cavities, and multicavity accelerating structure 98–99 side-coupled structure 111 Coupled-cavity linac (CCL) structure 98 DYNAC beam-dynamics code 191

Index Coupled-cavity system resonant-frequency errors 132 Coupler geometry matched or reﬂection-free condition 353 Couplers ﬁxed versus variable 353 Coupling circuit problem, between generator and resonator circuits 142 electromagnetic energy, to cavities 138–139 Coupling constant k coupling between, adjacent oscillators 105 Coupling factor backward power, values of 147 Courant–Snyder ellipse parameter maximum value 220 Crandall derivation zero-slope condition 269 Critical coupling corresponding to, outgoing-wave amplitude 153 round-trip ﬁeld attenuation factor 153 Cryogenic refrigeration 46 CW neutron sources 13 CW superconducting linac driver linac 426 CW see Continuous-wave 5, 419 Cylindrical cavity 6 transverse-magnetic resonant mode 6 Cylindrical pillbox synchronous acceleration problems 49–52 Cylindrical pipe image force for line charge 362–364 radius b, with image charge 363 Cylindrical resonator transverse electric modes 28–31

d Decay time, of modes 379 Design parameters for SASE FEL Projects 429 Design particle equation of motion 191 Disk-loaded traveling-wave structure electric ﬁeld acceleration 8 Disk-loaded waveguide 5 Dispersion curve cutoff frequency 255 for uniform waveguide 21

measurement of (a) π mode, resonant frequencies 67 measurement, in periodic structures 65–67 Dispersion curve (Brillouin diagram) for uniform waveguide 54 Dispersion relation 27 Drift space transfer matrix of 335–338 Drift tube 3, 4, 8, 9 and accelerating gaps 176 axial bore holes 39 charges and currents, in DTL 92 compact 90 constructing, with nose cones 35 eddy currents, and power dissipation 92 ﬁeld penetration 39 ﬁrst accelerating system 1 gap between 35 loading 89 longitudinal phase space 198 transverse focusing, with magnetic quadrupole lenses 94 Drift tube linac (DTL) adiabatic bunching, in RFQ 246 electric-ﬁeld pattern 93 for accelerating protons and heavier ions 96 Drift-kick-drift method 97 Drift-kick-drift treatment 185 Drift-tube linac with post couplers for, ﬁeld stabilization 96 Drift-tube linac (DTL) 4, 83, 210, 324 accelerating gaps 8 cooling difﬁculties 45 design of 96–98 nonrelativistic beam 405 sinusoidal accelerating ﬁeld 405 Drift-tube support stems 210 Drift–kick–drift momentum kick 204 Drive cell intercell coupling strength k 107 power-ﬂow phase shift 107 Drive line wave-interference effects 135 Driven oscillator 138 DTL design code, PARMILA 97 DTL tank longitudinal matching 225 matched phase-space conﬁgurations 224

437

438

Index DTL see Drift-tube linac 4, 210, 324 DYNAC beam-dynamics code 191

Electrode potential of 4 Electrode geometry 238 electric-quadrupole, with unequal e electrode spacing 234 Earnshaw’s theorem 202 two-term potential function 238–240 ECR see Electron cyclotron resonance 426 Electrode, isopotential 4 Eigenvectors Electromagnetic ﬁeld ﬁrst-order corrections 102 deﬂecting-mode cycle 391 Electric ﬁeld 4, 6–8, 10, 11, 17, 22, 23, 25, ﬁelds, from relativistic point charge 28, 30 364–366 accelerating gap, and energy gain time-dependent, quasistatic 33–36 approximation 167–168 and charged-particle beam 2 Electromagnetic ﬁeld–solver codes 44 and energy gain in each cell 97 Electromagnetic mode 3 average axial electric-ﬁeld amplitude Electromagnetic standing wave 34 in RF cavity 32 beam, maximal energy transfer 5 Electromagnetic stored energy 23 determination of, particle dynamics Electromagnetic traveling wave 240–241 and electromagnetically coupled cells 7 electric-quadrupole, cross section 234 Electromagnetic wave localized into, bunches 3 phase velocity 7 longitudinal and radial, in RF gap 206 uniform waveguide 7 moving charge into, equatorial plane Electromagnetic-cavity structure 365 beams accelerated with 3 pattern of disk, and washer modes 120 Electromagnetically coupled cells 7 peak surface, pulsed operation 10 Electron square proﬁle of 36 multipacting 159–160 tngential 22 Electron cyclotron resonance (ECR) 426 Electric focusing 9 Electron ﬁeld emission Electric wakeﬁelds and superconducting cavities 162–163 ultrarelativistic, Gaussian line-charge limiting factor 418 bunch 366 Electron linac 5 Electric- and magnetic-ﬁeld patterns 7 and beam injection 11 Electric-ﬁeld line applications of, modern RF linacs in accelerating gap, longitudinal 13–15 focusing 11 derivation, pillbox cavity 24 in RF gap 202 drift tubes and supporting stems 4 projections of DTL cells 94 high-energy 346 radial RF electric, and magnetic forces in cancer therapy 13 202 resonant antennas, with high power transverse RF focusing and defocusing losses 4 201–202 Electron linear accelerator Electric-ﬁeld vectors and quarks 1 x − z plane 236 Electron loading 125 Electric-focusing structure Electron multipacting quadrupole geometry 407 limiting factor 418 Electrical cell Electron traverse gap periodic chain of 59 two-point multipacting 159 Electrical center 35 Electron-linac application calculating, transit-time factor 52 free-electron lasers, and linear colliders gap, geometric center 35 289 standard electromagnetic ﬁeld solver Electron-linac peak current 290 code SUPERFISH 191 Electron-positron colliders 13

Index Electrostatic accelerator 2 electric breakdown 2 limitation of 2 Electrostatic ﬁeld electron volt (eV) 2 Electrostatic focusing gap, effects of position and velocity changes 207–208 Ellipse general trajectory ellipse, and its parameters 214 Ellipse parameters time-independent set 261 Ellipse transformation transporting beam ellipses, between two locations 221–222 Ellipsoid form factor 300 Elliptical beam continuous 297–299 RMS envelope equation, with space charge 296–297 Emittance conventions rms emittance 288–289 Emittance growth for rms mismatched beams 316 from rms mismatch 317 initial semi-Gaussian beam 313 Energy conservation and energy loss 343 Energy gain 10, 29 common trigonometric identity 34 limited by, ﬁxed potential drop 3 maximum, for relativistic electrons 34 radial dependence, and synchronous wave 39 Energy loss higher-order cavity modes 344–345 higher-order-mode enhancement factor 345 Energy recover linac (ERL) and superconducting linac 429 Energy spread short-range wakeﬁelds 383–384 Energy velocity velocity of, electromagnetic energy ﬂow 22 Energy-gain process 32 Energy-gain, result standing-wave ﬁeld pattern 39 Energy-recovery linac (ERL) 397 Envelope mode emittance growth, for mismatched rms beam 316–318 Equivalent circuit

containing generator 145 generator circuit 141 into resonator circuit 143 Equivalent circuit model for periodic structures 59–61 resonator-to-guide coupling factors 145 ERL see Energy-recovery linac 397, 429 ESS see European spallation source 13 Europe electron-positron storage ring facilities 416 European spallation source (ESS) 13

f FEL see Free-electron laser 397 Field distribution equal and opposite δ-function errors 261 periodic iris-loaded structure 55 Field distribution, perturbed, —δ-function error 260 (a) a δ-function error 259 Field emission 45 Field errors shift in, ﬁnal energy 182 Field pattern TM mode of, pillbox cavity 390 Field tilt and coupling 108 fabrication errors, and distortions 97 Field-free drift spaces 212 Fields and currents skin depth 19 Finger-focusing geometry Tepliakov, π -mode electric-focusing structure 407 First accelerator and ground potential 4 Floquet theorem 53 and space harmonics 54–57 loss-free periodic structure 55 stopbands, intervals of 55 Focusing thin drift tubes, without quadrupoles 89 Focusing-defocusing (FODO) 395 FODO lattice structure thin-lens 215 FODO quadrupole lattice with accelerating gaps 211 FODO see Focusing-defocusing 395 FODO-lattice structure thin-lens focusing structure 215–217

439

440

Index Four vane, periodic with windows RFQ structure 279 Four-rod RFQ 276 Four-vane cavity 248 cloverleaf geometry 249 vane-tip patterns 248 Four-vane cloverleaf cavity 252 Four-vane RFQ ﬁeld-tilt effects 254 idealized mode spectrum 253 plotted against, longitudinal-mode number 254 Four-vane RFQ structure 240 Fourier integral accelerating gap 36 and particle velocity 37 longitudinal electric ﬁeld 36–39 nonzero ﬁeld region 37 Fourier integrals in exponential form 38 Fractional-ﬁeld error 258 Free-electron laser (FEL) 397 Free-electron lasers pulsed neutron sources 13 Free-energy model maximum emittance-growth curves 330 Full cell RFQ linac vanes 266

g Gap factor for ﬁnite chamber radius 41 gap and particle velocity 86 Gap geometry and ﬁeld distribution 33 Gaussian longitudinal wake potential 373 Gaussian beam ﬁnal emittance growth ratio 313 three beam parameters, for initial 312 total bunch loss parameter 377 Gaussian bunch ktot versus the rms length 377 Gaussian distribution bunch loss parameters 377–378 Gaussian form factor 377 Generator power beam current, less than design value 352–354 versus beam power, ﬁxed cavity voltage 354 zero reﬂected power 350 Generator-induced voltage 346

Geometry factor QRs versus design velocity β 127 Golf club see Separatrix, shape 182 Group velocity 20 stored energy and ﬁelds 69 total attenuation parameter 73 velocity of, amplitude-modulation envelope 21 Gustav Ising ﬁrst accelerator 3

h H-mode accelerating structures transverse-electric (TE) 89–91 H-mode linac structure longitudinal beam dynamics 196–198 H-mode structure high shunt impedance, in high RF power efﬁciency 90 H-mode structures IH structure crossbar H-mode 90 Half-wave resonator 23 Halo 12 Halo see Beam halo 304 Hamiltonian and Liouville’s theorem 182 Hamiltonian theorem motion ignoring acceleration 182 Heavy-ion acceleration quasi-Alvarez DTL, and quasi-Alvarez approach 94 Heavy-ion beam buncher for 30 Heavy-ion inertial-fusion program Lawrence Berkeley National Laboratory 319 Heavy-ion linacs 13 ﬂexibility of 190 HEM deﬂecting mode transverse beam-modulation amplitude 391 HEM see Hybrid electromagnetic modes 389 High-energy linac CW spallation neutron sources 13 nuclear wastes, transmutation of 13 Hill’s equation linear second-order differential equation 212 solution, phase-amplitude form 213 transfer-matrix solution 211–213 Hybrid electromagnetic modes (HEM) 389

Index

i ILC see International linear collider 423 Impedance function frequency spectrum, δ-function wake potential 379 Independent-cavity ion linac longitudinal dynamics 190–191 Independent-cavity linacs and longitudinal beam dynamics 87 computational approach 191 Inductance calculation idealized quadrant shape 251 Input coupler 44 Input guide reﬂection coefﬁcient 145 Input waveguide and traveling waves 148 or transmission line, and power losses 44 Intercell coupling constant electric and magnetic dipoles 114–116 Interdigital four-beam MEQALAC 410 Interdigital H-mode (IH) structure 89 sequence of electrodes, for acceleration 90 International linear collider (ILC) 423–425 9-cell superconducting niobium cavities 425 International linear collider (ILC) facility showing two stages 425 Ion implantation semiconductor fabrication 13 Ion linac beam matching, current-independent 224 independent cavities 11 longitudinal dynamics, in coupled-cavity linacs 189–190 longitudinal stability 201 repulsive space-charge forces 12 synchronous traveling wave 204 Ion-linac channel maximum beam current 408 Iris(disk)-loaded traveling-wave structure 68 Iris-loaded traveling-wave accelerating structure 70 Iris-loaded structure periodic, analysis of 69–72 traveling-wave accelerator, for relativistic electrons 80–81

Iris-loaded waveguide see Disk-loaded waveguide 5 Isotope beams next-generation rare isotope facility 426–427

j Jacobian transformation increases, phase-space area 184 Jacobian determinant 182 Japan TRISTAN facility at KEK 416 Jefferson laboratory infrared FEL, ERL operation in CW mode 430

k K–T potential function 237 Kapchinsky and Vladimirsky quantity K, generalized perveance 298 Kilpatrick criterion 45 choosing, design ﬁeld level 163 Kilpatrick formula and Kilpatrick criterion 163 KONUS scheme ten drift-tube, IH drift-tube array 198

l LANSCE DTL 94 LANSCE linac highest average proton beam 14 or LAMPF 14 LANSCE proton accelerator DTL parameters 95 LANSE see Los Alamos Neutron Science Center 9 Lapostolle convention and rms emittance 288 Lapostolle matchbox geometry 407 LCLS see Linac coherent light source 429 LEDA halo experiment 52-quadrupole-magnet lattice 329 LEDA see Low-energy demonstration accelerator 329 Linac 4–7, 14 800-MeV LANSCE proton linac 15 acceleration method 7 and linac structure 6–10 application 135 BBU instability 362 beam bunch, beam quality phase space, and emittance 283–285

441

442

Index Linac (continued ) beams, paraxial approximation 38 betatron frequency, betatron wavelength 386 better beam quality 2 biperiodic structures 108–111 block diagram 6 bunched beams, space-charged dynamics 289–292 Coulomb effects 282 designed waveguides, or high-Q resonant cavities 5 emittance growth, space-charge-induced 314–316 high-power vacuum-tube ampliﬁers 156 independent-cavity 83–87 length, and ohmic power consumption 10 linear accelerator 1 modular array of, accelerating structures 5 multiparticle dynamics with space charge 282–338 parameters of, SLC and LANSCE linacs 15 pulsed machine 289 quadrupole focusing 209–211 risk of, emittance transfer 323 rms emittance 285 single-pass device 10 strong focusing 5 superconducting cavities, accelerator in Stanford 419–422 superconducting linacs 6 synchronous acceleration 182 synchronous-velocity proﬁle 181 tandem van de Graaff electrostatic accelerator 419 transverse stability plot 217–218 wake potential, from relativistic point charge 367–368 Linac accelerating structures 83–132 Linac beam 283 dynamics, multicell ion linac 10–12 equipartitioning 321 stable phase 10 synchronous phase 10 Linac bunch three-dimensional, ellipsoidal bunched beams 299–300 Linac coherent light source (LCLS) 429 European XFEL at DESY 429 Linac cost model 47

Linac economics impact of, design choices 47–49 resistive losses 47 Linac injector neutron-spallation source 289 Linac particle dynamics ﬁrst approximation for, computer simulations 184 Linac periodic focusing smooth approximation 226–227 Linac structure and linac systems 405 bunches 6 longitudinal dynamics, of low-energy electron beam 193 mode concept 107 traveling-wave, disk-loaded or iris-loaded waveguide structure 68–69 Linac technology group velocity, and energy velocity 20 Linac terminology synchronous-energy gain 241 Linac see Accelerator little machine, the ‘linac’ 1 Linac, efﬁcient construction of 7 Linac-cavity ﬁeld maintaining phase and amplitude 157 Lincac radio frequency (RF) acceleration 32–52 Linear accelerator 5 alternating-phase focusing (APF) 405–406 and high-power beams 2 beams, high energy 1 historical perspective 2–5 new type 232 special structures and techniques 405–430 spin-off, nuclear physics research 1 straight-line trajectory, power loss avoidance 2 Linear colliders and circular colliders 13 Linear focusing ideally matched beam with 285 Linearized Vlasov theory theoretical approach, for stability analysis 318 Liouville’s theorem 183, 284 and dashed parallelogram 186 phase and position impulses 209

Index Long cavity ﬁgure of merit, ﬁeld level and cavity length 43 Long-range wakeﬁelds and multibunch beam breakup 389–397 Longitudinal beam dynamics 175 Longitudinal mode 27 Longitudinal motion accelerating cells 176 differential equations 178 equations of, Hamiltonian and Liouville’s Theorem 182–186 small, acceleration rate 178–182 Longitudinal phase space and longitudinal potential well 180 Longitudinal phase-space ellipse at phase focus 196 buncher cavity, after kick 195 Lorentz condition 18 Lorentz force 29 Los Alamos beam halo experiment 329 Los Alamos Neutron Science Center (LANSCE) 9 Low-energy beams longitudinal dynamics of 192–194 Low-energy demonstration accelerator (LEDA) 329 Low-frequency structures 46 Low-pass ﬁlter periodic array 61–62 periodic, basic cell of 61 Low-pass ﬁlter, periodic dispersion curve of 61 Lumped-circuit model four-vane cavity 249–252 four-vane RFQ, in cloverleaf geometry 249 lumped-circuit picture 93 numerical, electromagnetic-ﬁeld-solver code, SUPERFISH 93

predominant RF longitudinal 89 standard perturbation techniques 91 typical(upper curve) and electric ﬁeld (lower curve) 93 Magnetic ﬁeld line in quadrupole mode at end 253 Magnetic ﬁeld, axial two-cavity klystron 156 Magnetic ﬂux voltage divider, behavior of 92 Magnetic mode cylindrical resonator transverse-magnetic modes 27–28 Magnetic quadrupole lens 233 Magnetic-quadrupole 9 Matched phase-space drawings 273 Mathieu equation 244 betatron frequency 244 Maximum acceleration 241 Maxwell’s equations 17–19, 39 accelerating cavity, approach of Condon 137 electric and magnetic ﬁelds 17 quasistatic approximation 135 MEQALAC see Multiple-beam electrostaticquadrupole-array linear accelerator 409 MHz four-vane RFQ 422 Micropulses 9 Microwave and microwave systems 135–174 Microwave power system for linacs 156–158 Mismatched beams halo-formation mechanism 326 Mode 12, 23, 30 beam-breakup, or instability 12 higher-order mode 12 π -mode 41 π -mode boundary conditions 41 Mode, electromagnetic 3 Modulation parameter 237 Multicell, superconducting elliptical cavity 41 m Multigigaelectron volt linac 402.5-MHz Alvarez drift-tube linac 423 for heavy-ion-driven 805-MHz SNS superconducting linac 424 inertial-conﬁnement fusion 13 805-MHz coupled-cavity linac 423 Multipacting Macropulses 9 characteristics of 160–162 Magnet lower electric ﬁeld levels 160 quadrupole 89 Multiparticle system Magnetic ﬁeld Coulomb interactions 282 calculating, effective loop area Multiple-beam electrostatic171–174 quadrupole-array linear accelerator longitudinal, and solenoid focusing 225–226 (MEQALAC) 409

443

444

Index

n Net cavity voltage 346 Nonequipartitioned beam rms-emittance transfer, having small effect 325 Nonlinear space-charge ﬁeld 314 Normal-mode spectrum of coupled oscillator system 106 Nose cone capacitance 44 elliptical cavity shape 45 Nose cones 35 nth space harmonic wavenumber, and phase velocity 56

o Ohmic power dissipation, effects of 103–104 Ohmic power loss 5 Operating temperatures 45 Optimum gap geometry risk of, RF electric breakdown 36 Oscillation small amplitude, and synchronous particle 186–187 Oscillator impedance and wake potential, single cavity mode 381–383 Oscillator error 106 Outgoing wave reﬂected wave, and emitted wave 154

p Panofsky equation 35 Panofsky–Wenzel theorem 392 beam-breakup instability 168–171 Particle peak energy gain 43 Particle acceleration normal mode characteristics 76–79 Particle interaction high-frequency betatron orbits 328 Particle velocity electrode spacing, increase of 89 transit-time factor, and energy gain 37 Particle-in-cell (PIC) macroparticle-tracking approach 293 Particle-to-particle interaction 295 Peak electric stored energy and electromagnetic stored energy 25 Peak magnetic ﬁeld 45 Peak surface electric ﬁeld 125, 240

accelerating-gradient ratio versus design velocity 126 Peak surface electric-ﬁeld normal-conducting cavities 45 Peak surface ﬁeld 44, 45 Peak surface magnetic ﬁeld 45 versus design velocity 126 Periodic accelerating structures 53–81 dispersion curve 59 general description 57–58 Periodic focusing channels space-charge instabilities 318 Periodic structure phase velocity 54 Periodic structures designing of 67 Permanent magnet quadrupoles, compact 91 Perturbation theory effects of resonant-frequency errors 101–103 Perturbation-theory, general 107 Phase damping longitudinal beam ellipse 188 Phase oscillation 11 and phase damping 188 ﬁnal, synchronous energy 11 synchronous particle 11 Phase-space area trace-space, and unnormalized phase-space 283 Phase-space ellipse rms emittance 285–287 transformation, between two locations 222 Phase-space trajectory in coupled-cavity linac structure 190 Phase-volume 12 Phasor currents and voltages 348 Physical constants 16 Pillbox-cavity 5, 24 array of 53 Pillbox cavity, unperturbed 71 Pipe geometry ﬁnite conductivity 365 Potential 18, 29 Power densities 45 Power loss 46 PPI see Particle-to-particle interaction 295 Program TRACE linear space-charge forces 292 Proton drift-tube linac parameters of 198–199

Index Proton linac and modern applications 13 cross section, elliptical cavity 45 superconducting 83 Proton linear accelerator spallation neutron source (SNS) 422 Pulsed linac micropulses and macropulses 9

principles of operation 232–236 Radiofrequency–Quadrupole (RFQ) 90 Random error effects, for periodic or quasiperiodic accelerator systems 218 Random quadrupole misalignment effects of errors 218–221 Rare isotope accelerator (RIA) 426 Recirculating electron linacs multipass BBU 397–399 q Reference particle Quadrupole maximum phase excursion 189 misaligned, centroid oscillation Regions, kinematic conditions amplitude 221 two parallel plates 160 Quadrupole focusing system Relativistic electron beams 361 dynamics of 210 Relativistic electron linac 12 Quadrupole lens Relativistic mechanics periodic lattice of 201 basic formulas 16 Quadrupole magnet Lorentz force, particle with charge and focusing 5-MeV proton beam, velocity 17 229–230 useful relationships 16–17 showing four poles, coils, and Residual resistance 20 magnetic-ﬁeld pattern 210 Residual resistivity ratio (RRR) 418 Quadrupole spectrum Resonance accelerator 5 transmission-line model 255–261 linac, cyclotron, and synchrotron 3 Quadrupole-mode dispersion curve Resonance-detection electronics four-vane cavity eigenmodes 252–254 RF pickup signal 158 Quality factor 23 Resonant buildup waveguide-to-cavity coupling, through of displacement of tail 386 iris 154 Resonant cavities 5 Resonant cavity r theory of 137–138 Radial ﬁeld distribution Resonant coupling mode, resonant frequency of 25 coupling elements 117–121 Radial impulse post-coupled drift-tube linac structure RF-defocusing impulse 203 119 Radial momentum 203 Resonant coupling element Radial motion cavities, periodic array 64–65 unfocused relativistic beams 227–229 shunt resonant admittance 64 Radiated ﬁeld Resonant mode 24 short-range or long-range wakeﬁelds Resonant-cavity system 362 equivalent circuit 139–144 Radiation methods of, coupling to cavities 140 synchrotron 5 transient behavior of 146–148 Radio frequency (RF) 2, 282 Resonator Radio-frequency quadrupole (RFQ) 2 quality factor of 42 electric ﬁeld 9 time dependence of 148 transverse RF electric-restoring force 9 Resonator circuit Radiofrequency (RFQ) equivalent circuit 348 transverse dynamics 243–245 Resonator ﬁeld Radiofrequency quadrupole (RFQ) turn-on transient 146 radial-matching 261–265 RF accelerating ﬁeld Radiofrequency quadrupole linac (RFQ) compensation of, longitudinal wake new linear accelerator 232–281 effect 384

445

446

Index RF accelerator adiabatic bunching 245–248 RF accelerator cavity application of, superconducting technology 415 RF accelerators 2, 3, 13 RF bunching RF linac input beam, and bunches 194–196 RF cavity periodic or quasiperiodic, array of cells 32 pillbox cavity 33 RF cycle 9 bucket, stable region 9 electric ﬁeld, longitudinal 9 RF efﬁciency 10, 51 RF electric breakdown Kilpatrick criterion 163–164 Rf electric breakdown risk of 10 RF electric ﬁeld 9 beam, sustained energy transfer 32 RF electrode four-vane cavity 248–249 RF ﬁeld energy-gain difference 384 RF ﬁelds alternating-phase focusing approach 201 focusing difﬁculties 201 RF generator efﬁciency 44 RF input power transient turn-on, beam-loaded cavity 354–355 RF linacs beam bunches of 6 funneling of, high-current beams 412 space-charge instabilities 318–319 stability analysis, and linearized Vlasov theory 318 RF linear accelerator 3, 5 and resonance accelerator 5 RF pickup probes for vacuum pumping 30 RF power 2, 30 RF power costs and operational costs 49 RF power losses RF power dissipation scales 46 RF power system and coupler geometry 346 RF pulse compression klystron pulse 413

RF quadrupole structure general potential function 236–237 quasistatic approximation 236 RF superconductivity 2 Padamsee, accelerator applications 416 physics and technology 416–418 RF surface resistance 19 superconducting niobium cavities 414 RF system 13 block diagram, and equivalent circuit 141 circulator, RF input drive line 347 RF see Radio frequency 2, 282 RF-focused drift-tube linac (RFD) 407 RF-focused interdigital linac structure (RFI) 408 RF-power 6 RF-power tubes klystrons 5 RFD see RF-focused drift-tube linac 407 RFI linac structure, views of 409 RFI structure showing minor stems 409 RFI see RF-focused interdigital linac structure 408 RFQ and bunched beam 262 RFQ accelerator four electrode arrangement 235 RFQ adiabatic bunching 246 radial-matching (RM) 248 RFQ bunches 9 RFQ cell parameters 279–281 RFQ electric ﬁeld transverse electric focusing 9 RFQ electrode geometry and longitudinal dynamics 242–243 RFQ four-vane cavity azimuthal modes 253 RFQ operating mode 257 squared-frequency-error distribution 258 RFQ operation four-vane RFQ accelerator 233 principles 232 RFQ quadrupole mode four-vane cavity 255 RFQ quadrupole-focusing geometry 280 RFQ structure electrode geometry 233 four-rod cavity 275–276 windows RFQ, four vane 277–279

Index RFQ tuning in four-vane resonator 278 RFQ see Radio-frequency quadrupole 2 RIA see Rare isotope accelerator 426 RM section designs 263 RM section geometry Crandall, four-term potential function 263 RMS ellipse 286 Courant–Snyder parameters 285 RMS emittance transverse and longitudinal 287–288 RMS emittance growth 315 RMS matched beam emittance growth 306–312 RMS-emittance growth measured 330 measured, beam half widths 331 RMS-mismatched beam 316 particle-core model 332 Reiser’s, free-energy parameter 316 r over Q 43 Rolf Wider concept of, Wider¨oe drift-tube linac 3 Rolf Wider¨oe ﬁrst experimental test 2 RRR see Residual resistivity ratio 418

s SASE FEL projects three key design parameters 429 SASE see Self-ampliﬁed spontaneous emission 429 Scalar potential 18 Scaling formulas for λ/4, superconducting structures 131 stored energy, for TM 130 Scattering matrix for n-port device 149 reciprocity 150 Schematic diagram SLED pulse compression 413 SCL see Side-coupled linac 9 Self-ampliﬁed spontaneous emission (SASE) 429 Separatrix ﬁsh 181 golf club, longitudinal phase-space trajectories 183 Separatrix see Golf club 179 Short-range ﬁeld binary, small impact-parameter Coulomb collisions 282

Short-range wakeﬁeld single-bunch beam breakup 384–386 Short-range wakeﬁelds 362, 376 Shunt impedance 5 and ﬁgure of merit 42 electromagnetic-ﬁeld-solver codes, SUPERFISH 97 normal-conducting cavities 43 per unit length 43 TM and λ/2 superconducting structures 128–130 unit, megohms per meter 43 Shunt impedance, transverse transverse beam-cavity interaction 392 Shunt impedance 43 Shunt resonant circuit model parallel resonant circuit 135–137 Shunt-resonant-circuit cavity model impedance of 382 Side-coupled linac (SCL) 9 Simple two-particle model for single-bunch beam breakup 385 Single gap multicell transit-time factor 85 multigap periodic structure 40 Single wave same phase velocity, and net energy 39 Single-bunch BBU effect BNS damping of beam breakup 386–389 Single-bunch loss parameter 344 Single-point multipacting 161 limiting, pillbox-like superconducting cavities 162 Sinusoidal voltage distribution 235 Skin depth see Fields and currents skin depth 19 Skin effect RF electric and magnetic ﬁelds 19 Skin effect see Fields and currents skin effect 19 SLAC 5 SLAC electron accelerator disk-loaded waveguide 370 SLAC linac RF pulse compression 413–414 typical RMS bunch 345 SLAC linac structure 371 total loss parameter ktot versus rms bunch length 378 SLAC see Stanford Linear Accelerating Center 5 Slater perturbation method ﬁeld-measuring apparatus 167

447

448

Index Slater perturbation theorem 53 resonant frequency shift 165–167 resonant-frequency change 69 SLC electron-positron linear collider 14 SLC linac BNS compensation, of transverse wake effect 389 Smooth-approximation solution 244 SNS facility individual systems 421 SNS see Spallation neutron source 13, 416, 421 Space charge beam current limits 302–303 Space harmonics traveling waves, inﬁnite number 56 Space-charge calculation 294 SCHEFF subroutine, in PARMILA codes 293 Space-charge ﬁeld charge-density uniformity 283 coupled Vlasov–Maxwell equations 282 image ﬁeld 363 smoothed ﬁeld distribution 282 Space-charge force 282 Space-charge forces 12 extended halo, of beam 12 Space-charge mechanism emittance growth 303–306 rms-matched beam, charge redistribution 304 Space-charge-induced emittance growth 283, 305 Space-charge-induced rms-emittance growth scaling of 332 Spallation neutron source (SNS) 13, 416, 421–422 Split-ring resonator concept for 115-MHz β = 0.13 resonator, 122 Square waves transit-time factor 35 Square-wave transit-time factor 36 Stability chart for transverse motion, Smith and Gluckstern 219 Smith and Gluckstern, for DTL 218 stop bands 322 Stability plot showing ESS design 324 Stable operating point 10

Standing waves forward and backward traveling waves 105 Standing-wave linacs difference equations, of longitudinal motion 177 Stanford Linear Accelerating Center (SLAC) 5 Stroboscopic plot independent particle trajectories 327 Structure efﬁciency 51 Superconducting cavities quench, critical magnetic ﬁeld 45 RF power efﬁciency 44 Superconducting linac at Oak Ridge National Laboratory 416 Superconducting linac facilities free-electron lasers 427–429 Superconducting linacs 83 electric-ﬁeld limitations 415 Superconducting niobium cavities 424 Superconducting RF linac schematic drawing, Jefferson Lab ERL system 398 Superconducting RF Linacs large RF power 414–418 Superconducting structures half-wave resonators 124 Superconducting structures, λ/2 half-wave resonators 121–124 Superconducting surface resistance 20 Superconducting-linac technology future facilities 423–430 Superconductivity discovery of 1 phenomenon of 415–416 Superconductors London penetration depth 416 zero dc resistance 416 SUPERFISH see Electrical center standard electromagnetic ﬁeld solver code SUPERFISH 191 Surface-resistance and power loss scaling, in superconducting RF resistance 46 Synchronism factor 86 Synchronous acceleration 49 and periodic structures 53 Synchronous particle accelerated reference particle 189 and synchronous phase 32 energy gain 242

Index longitudinal focusing 175–176 stable phase 176 Synchrotron radiation 2, 13

Traveling and standing-wave structures, physics regimes 79–80 Traveling wave constant-impedance structure 72–74 t frequencies and wave numbers 20 TE mode 28 Traveling wave, synchronous TEM see Transverse electromagnetic 84 radial impulse 203–204 Thin-lens approximation and lattice properties, comparison 227 Traveling-wave accelerator principles of operation 68 Thin-lens quadrupole Traveling-wave power 72, 147 random transverse misalignments 219 Traveling-wave structure Thomas Jefferson National Accelerator 50-GeV electron linac 7 Facility (TJNAF) 416 Tune-depression ratio Three coupled oscillators 99–100 space-charge-induced rms-emittance TJNAF see Thomas Jefferson National growth 332 Accelerator Facility 416 Tuning stubs TM and TEM structures tuning for, the desired ﬁeld distribution RF properties and scaling laws in RFQ 274–275 125–127 Two-conductor TM cell array two-point multipacting 159 dispersion relation, for electrically Two-gap cavities coupled periodic 62 schematic drawing 85 TM circuits basic cell of, periodic array 63 TM circuits, electrically coupled u basic cell of, periodic array 62 Ultrarelativistic beam TM mode wake potentials, cylindrically symmetric cylindrical cavity resonator 25 structures 368–372 TM superconducting structures Ultrarelativistic bunches 362 or elliptical cavities 122 Ultrarelativistic source charge 367 TM011 cavity mode Uniform waveguide node, and energy gain 36 cutoff wavenumber 54 Transfer matrix transfer-matrix elements, through one v period 214 Vane geometry Transfer-matrix formalism 221 potential function 232 Transit-time effect Vane voltage and transit-time factor 35 ratio of surface magnetic ﬁeld 252 Transit-time factor 35 Vane-tip proﬁle and synchronous velocity 40 in accelerating cell 269 Transit-time-factor models Vane-tip proﬁles, periodic 268 penetration of ﬁeld 39–42 Vector potential 18 Transition cell, Crandall 266 Vectors, transformed 52 Transition-cell application 272 Velocity change Transmission lines 3, 5, 20 and transit-time factor 35 Transverse coupling impedance 392 Velocity modulation 4 Transverse electromagnetic (TEM) 84 Voltage resonant standing-wave mode 22 axial radio frequency 34 Transverse emittance Voltage distribution 256 unnormalized emittance 287 Voltage phasor Transverse magnetic wave 53 beam-loaded superconducting cavity Transverse particle dynamics 201–230 352 off-axis particles 201 for beam-loaded cavity 346, 352 Transverse-magnetic mode 27

449

450

Index

w Wake force free oscillation solution 385 Wake potential arbitrary charge distribution, and Green’s functions 367 computed Gaussian wake potential per cell 374 corresponding to, shunt resonant circuit 382 frequency scaling of 370–371 longitudinal 399–402 and transverse-impedance deﬁnitions 380 Wake potentialδ-function wake potential 368 Wakeﬁelds bunch, electric and magnetic ﬁelds 361–402 damped, oscillatory electromagnetic disturbances 362 equivalent beam-induced image charges 361 higher-order modes, and beam-excited accelerating mode 362 parasitic energy loss, short-range 383 scattered radiation 12 Wave phase velocity, matching beam velocity 32

Wave attenuation and decibels (db) 31 Wave equations 18 in cylindrical coordinates 37 Wave group and wave packet 21 Wave packet 21 Wave power 22 Waveguide and cavity section, conducting plate with iris 150 cavities, coupling of 135 equivalent circuit, for cavity coupled to 144–146 ﬁlling time 76 particle acceleration, in RF Field 32–33 Waveguide mode 30 Waveguide-to-cavity coupling parameter 155 Wider¨oe linac, forerunner to, modern RF accelerators 3 Wider¨oe Linac 4 acceleration of, single particle species 87–89 Wider¨oe-type linac and drift tubes 4