Sei sulla pagina 1di 1121

cover next page >

title : Foundations of Vacuum Science and Technology


author : Lafferty, J. M.
publisher : John Wiley & Sons, Inc. (US)
isbn10 | asin : 0471175935
print isbn13 : 9780471175933
ebook isbn13 : 9780585339368
language : English
subject Vacuum, Kinetic theory of gases.
publication date : 1998
lcc : QC166.F68 1998eb
ddc : 621.5/5
subject : Vacuum, Kinetic theory of gases.
cover next page >
< previous page page_iii next page >
Page iii

Foundations of Vacuum Science and Technology

Edited By
James M. Lafferty

< previous page page_iii next page >


< previous page page_iv next page >
Page iv

This book is printed on acid-free paper.

Copyright © 1998 by John Wiley & Sons, Inc. All rights reserved.
Published simultaneously in Canada.

No part of this publication may be reproduced, stored in a retrieval system or transmitted in any form or by any means,
electronic, mechanical, photocopying, recording, scanning or otherwise, except as permitted under Sections 107 or 108
of the 1976 United States Copyright Act, without either the prior written permission of the Publisher, or authorization
through payment of the appropriate per-copy fee to the Copyright Clearance Center, 222 Rosewood Drive, Danvers,
MA 01923, (508) 750-8400, fax (508) 750-4744. Requests to the Publisher for permission should be addressed to the
Permissions Department, John Wiley & Sons, Inc., 605 Third Avenue, New York, NY 10158-0012, (212) 850-6011, fax
(212) 850-6008, E-mail: PERMREQ @ WILEY.COM.

Library of Congress Cataloging in Publication Data:


Foundations of vacuum science and technology/edited by J. M. Lafferty.
p. cm.
"A Wiley-Interscience publication."
Includes bibliographical references and index.
ISBN 0-471-17593-5
1. Vacuum. 2. Kinetic theory of gases. I. Lafferty, J. M.
(James Martin), 1916.
QC166.F68 1997
621.5′5dc21 96-29895

Printed in the United States of America

10 9 8 7 6 5 4 3 2 1

< previous page page_iv next page >


< previous page page_v next page >
Page v

Preface

The industrial and scientific importance of vacuum technique has continued to grow during the past 30 years, even with
the demise of television and radio receiving tubes and many gas-filled tubes. The replacement of vacuum and low-
pressure gas discharge tubes by semiconductors and integrated circuits has placed an even greater demand and more
stringent conditions on vacuum technique for the processing and manufacture of these devices. This has led to the
development of a number of "dry" vacuum pumps to produce a "clean vacuum" free of hydrocarbons. The space
programs, high-energy accelerators, analytical instruments, freeze-drying of foods and drugs, and the manufacture of
color television picture tubes, incandescent and metal vapor lamps, high-power vacuum, and x-ray tubes all continue to
require the need for vacuum technique.

While the material in this book is totally new, it follows in the tradition set by Scientific Foundations of Vacuum
Technique by Saul Dushman, published in 1949. That book enjoyed unprecedented success and is now a classic in its
field. By 1960 it was badly in need of revision. This editor had the privilege of participating as editor of the revised
edition, which was published in 1962. This second edition was brought up to date by a number of contributors with
specialized knowledge in the disciplines involved. An attempt was made to introduce the new developments made in
vacuum technique but keep the original plan of the book and retain much of the material that was still of current interest.

The editor was encouraged by Leonard Beavis of the American Vacuum Society Education Committee and the
publishers of the previous editions to undertake the publication of the present volume. The advances made in vacuum
science and technology during the past three decades has required a complete reworking of the material in the previous
volume. However, every effort has been made to follow the unique style of the original bookthat is, to present a survey
of fundamental ideas in physics and chemistry that would be useful to both scientists and engineers dealing with
problems associated with the use, production, and measurement of high vacuums. This volume is a critical survey of
important developments in vacuum technique with many references for those who seek a better understanding and more
detailed information in the field. It is not a vacuum handbook, many of which are listed in the Appendix of this book.

Every effort was made to select on a worldwide basis a number of outstanding vacuum specialists who were willing to
take time to contribute to this volume. With

< previous page page_v next page >


< previous page page_vi next page >
Page vi

the curtailment of vacuum research in the major industrial laboratories, one now only finds vacuum experts as
independent consultants or in companies manufacturing vaccum systems and components, a few educational
institutions, and government laboratories.

While the basic laws of kinetic theory of gases have not changed over the years, a better understanding of gas flow over
a wide range of pressures has necessitated an expanded chapter on the subject. It now encompasses all flow regimes
from free molecular flow to atmospheric pressure. It treats compressible flow through tubes and orifices under choked
and nonchoked conditions as well as turbulent flow in ducts of any cross section.

Many topics that were only mentioned in the second edition of Scientific Foundations of Vacuum Technique now have
full chapters devoted to them. The progress made in vacuum pumps over the past three decades, for example, is
remarkable. Three chapters are now devoted to this subject. Detailed information is given for the first time on liquid
ring pumps, dry pumps, turbo pumps, getter pumps, and cryopumps.

The subject of leak detection, which had only a few paragraphs devoted to it in the old edition, now has a full chapter
describing leak detectors as a rugged industrial tool for everyday use capable of quantitative measurements.

Information on the design of high-vacuum systems has been expanded to help the reader in selecting pump sets for
various system applications and in predicting their performance.

Pressure measurements continue to be important on every vacuum system. This subject is fully covered in the chapter
on vacuum gauges. While the ionization gauge continues to be the principal pressure sensor for measuring total pressure
in high- and ultrahigh-vacuum systems, modern solid-state electronics has simplified its use. The accuracy of this
device in measuring pressure depends on a knowledge of the composition of the gas being measured. The partial
pressure analyzer has become a far more sophisticated way to measure pressure and give the vacuum system operator an
insight of what is occurring within the system. The invention of the quadrupole mass spectrometer with solid-state
electronics has done much to make partial pressure measurements relatively simple and inexpensive. A full chapter is
devoted to this subject.

In discussing pressure measurements, a word about pressure units seems appropriate. While use of the pascal, the ISO
unit of pressure, has been encouraged in this book, many of the European contributors strongly preferred using the
millibar (mbar). The mbar falls in a class of units that are temporarily accepted for use by the ISO. You will find both
units in this book. The advantage of the mbar is that it is nearly equal in magnitude to the Torr or mmHg found in earlier
publications and is familiar to many readers (1 mbar = 0.75 Torr). When several orders of magnitude of pressure are
plotted on a log scale, the mbar and Torr plots are nearly indistinguishable. Some of the figures in this book that have
been copied from earlier publications may still have the pressure plotted in Torr.

Ultrahigh-vacuum technique had its infancy in midcentury. Today it is a matured procedure used in a great variety of
applications and in commercially available equipment. Researchers do not appear to have reached a limit yet in their
quest to produce and measure a perfect vacuum. This work is described in the chapter devoted to ultrahigh and extreme
high vacuum. Considerable progress has been made in this area by pushing vacuum techniques to their limit and gaining
a better

< previous page page_vi next page >


< previous page page_vii next page >
Page vii

understanding of the gassurface interactions and diffusion in solids as described in the chapter on this subject. While
pressures as low as 1011 Pa have been measured in the laboratory, this is still at least three orders of magnitude higher
than that in interstellar space.

The final chapter is devoted to calibration and standards. It describes the physical background and state of the art of
today's primary vacuum standards in the various national laboratories. It should be useful reading not only for those
involved in calibration and quality control but for those interested in the accuracy limitations of various vacuum
instruments.

The editor is indebted to several people for suggestions concerning this volume. Special mention is made of the late
Hermann Adam for helpful discussions and for suggesting a number of German contributors for the book. John Weed, a
member of the American Vacuum Society Education Committee, solicited suggestions for the volume from a number of
A.V.S. members. Nigel Dennis coordinated chapters three and four on vacuum pumps, and Benjamin Dayton and Paul
Redhead have made many helpful suggestions.

J. M. LAFFERTY
SCHENECTADY, NEW YORK

< previous page page_vii next page >


< previous page page_ix next page >
Page ix

Contributors

Helmut Bannwarth, LEDERLE GmbH, Gundelfingen, Germany

Benjamin B. Dayton, Consultant, East Flatrock, North Carolina, USA

Nigel T. M. Dennis, Edwards High Vacuum International, Crawley, West Sussex, England

Johan E. de Rijke, Vacuum Technical Services, Morgan Hill, California, USA

Robert E. Ellefson, Leybold Inficon, Inc., East Syracuse, New York, USA

Bruno Ferrario, SAES Getters S.p.A., Lainate (Milano), Italy

Werner Grosse Bley, Leybold Vakuum GmbH, Cologne, Germany

Hinrich Henning, Leybold Vakuum GmbH, Cologne, Germany

Jörgen Henning, intervac Henning, GmbH, Kreuzwertheim, Germany

John B. Hudson, Materials Science and Engineering Department, Rensselaer Polytechnic Institute, Troy, New York,
USA

Karl Jousten, Physikalisch-Technische Bundesanstalt, Berlin, Germany

R. Gordon Livesey, Edwards High Vacuum International, Crawley, West Sussex, England

R. Norman Peacock, MKS Instruments, HPS Division, Boulder, Colorado, USA

Paul A. Redhead, Institute for Microstructural Sciences, National Research Council, Ottawa, Ontario, Canada

Wolfgang Schwarz, Leybold Systems GmbH, Hanau, Germany

< previous page page_ix next page >


< previous page page_xi next page >
Page xi

Contents

Preface v

Contributors ix

Acronyms xxiii

1. Kinetic Theory of Gases 1


Benjamin B. Dayton

2
1.1. Ideal Gas Law

6
1.2. Avogadro's Number

8
1.3. Molecular Collisions; Mean Free Path; MaxwellBoltzmann Distribution Laws

16
1.3.1. Relation Between Molecular Velocities and Velocity of Sound

17
1.3.2. Determination of Avogadro's Constant from Distribution of Particles in Brownian Motion

18
1.4. Gas Pressure and Rate at Which Molecules Strike a Surface

22
1.5. Rate of Evaporation and Vapor Pressure

26
1.6. Free Paths of Molecules

29
1.7. Relation Between Coefficient of Viscosity, Mean Free Path, and Molecular Diameter

37
1.7.1. Viscosity at Low Pressures

39
1.7.2. Molecular Diameters
39
1.7.3. Application of the van der Waals Equation

40
1.7.4. From the Density of the Solid or Liquid

41
1.7.5. Cross Section for Collision with Electrons

41
1.8. Heat Conductivity of Gases

44
1.9. Thermal Conductivity at Low Pressures

46
1.9.1. Free-Molecule Conductivity (Knudsen)

50
1.9.2. Temperature Discontinuity (Smoluchowski)

< previous page page_xi next page >


< previous page page_xii next page >
Page xii

53
1.10. Thermal Transpiration (Thermomolecular Flow)

57
1.11. Thermal Diffusion

62
1.12. Theory of Diffusion of Gases

65
1.12.1. MaxwellLoschmidt Method for Determination of Diffusion Coefficients

67
1.12.2. Effect of Pressure of Gas on Rates of Evaporation of Metals

69
1.13. Random Motions and Fluctuations

71
1.14. Scattering of Particle Beams at Low Gas Pressures

73
References and Notes

2. Flow of Gases Through Tubes and Orifices 81


R. Gordon Livesey

83
2.1. Flow Conductance, Impedance, and Gas Throughput

85
2.2. Molecular Flow

86
2.2.1. Conductance of an Aperture

87
2.2.2. General Considerations for Long Ducts

87
2.2.3. General Considerations for Short Ducts

88
2.2.4. Uniform Circular Cross Section

90
2.2.5. Duct of Uniform Rectangular Cross Section
92
2.2.6. Tube of Uniform Elliptical Cross Section

93
2.2.7. Cylindrical Annulus (Flow Between Concentric Cylinders)

94
2.2.8. Uniform Triangular Section (Equilateral)

94
2.2.9. Other Shapes

96
2.2.10. Combinations of Components

102
2.2.11. Cases of Unsteady Flow

105
2.3. Continuum Flow

108
2.3.1. Viscous Laminar Flow

112
2.3.2. Turbulent Flow

116
2.3.3. Compressible Flow

119
2.3.3.1. Flow through an Aperture or Short Duct

121
2.3.3.2. Approximation for Flow Through an Aperture

121
2.3.4. Corrections for Flow Obstructions

122
2.3.5. ApproximationsEntrance Correction Model

124
2.3.6. ApproximationsKinetic Energy Model

126
2.3.7. Long Duct Criteria

128
2.4. Transitional Flow

129
2.4.1. Transitional Flow in Long Ducts
134
2.4.2. Long Duct Criterion in Transitional Flow

135
2.4.3. Transitional Flow through Apertures and Short Ducts

137
Symbols

139
References

< previous page page_xii next page >


< previous page page_xiii next page >
Page xiii

3. Positive Displacement Vacuum Pumps 141

Part I. Oil-Sealed Vacuum Pumps 143


Nigel T. M. Dennis

143
3.1. Oil-Sealed Vacuum Pumps

143
3.1.1. Pump Design

144
3.1.2. Gas Ballast

147
3.1.3. Pump Oil

148
3.1.4. Oil Suckback

148
3.1.5. Power Requirements and System Protection

149
3.1.6. Accessories

Part II. Liquid Ring Pumps 151


Helmut Bannwarth

151
3.2. Liquid Ring Pumps

151
3.2.1. Mechanism

152
3.2.2. Single-Stage Liquid Ring Vacuum Pumps

153
3.2.3. Two-Stage Liquid Ring Vacuum Pumps

154
3.2.4. The Operating Liquid

154
3.2.5. Operating Ranges of Liquid Ring Gas Pumps
154
3.2.6. Cavitation and Protection Against Cavitation

156
3.2.7. Types of Operation; Conveyance of Operating Liquid

157
3.2.8. Materials of Construction

157
3.2.9. Sealing

157
3.2.10. Drives

158
3.2.11. Accessories

Part III. Dry Vacuum Pumps 159


Nigel T. M. Dennis

159
3.3. Dry Vaccum Pumps

159
3.3.1. Roots Pump

162
3.3.2. Claw Pump

164
3.3.3. Screw Pump

167
3.3.4. Scroll Pump

169
3.3.5. Piston and Diaphragm Pumps

170
References

171
General References

4. Kinetic Vacuum Pumps 173

Part I. Diffusion and Diffusion-Ejector Pumps 175


Benjamin B. Dayton

176
4.1. Diffusion Pumps
176
4.1.1. History of Development

181
4.1.2. Diffusion Pump Design

< previous page page_xiii next page >


< previous page page_xiv next page >
Page xiv

183
4.2. Diffusion-Ejector Pumps

185
4.3. Performance of Vapor-Jet Pumps

185
4.3.1. Pumping Speed

187
4.3.2. Limiting Forepressure for Maximum Speed

190
4.3.3. Influence of Nozzle and Entrance Chamber Design on Speed

192
4.3.4. Ultimate Pressure

194
4.3.5. Backstreaming and Back Migration of Pump Fluid

198
4.3.6. Throughput

202
4.4. Theory of Pump Performance

202
4.4.1. Speed

204
4.4.2. Limiting Forepressure

205
4.4.3. Vapor-Jet Flow Pattern

221
4.4.4. Ultimate Pressure

Part II. Molecular Drag and Turbomolecular Pumps 233


Jörgen Henning

233
4.5. Molecular Drag Pumps

234
4.5.1. Theoretical Considerations and Performance Data
237
4.5.2. Design Considerations

237
4.5.3. Typical Performance Data of Commerical Pumps

238
4.5.3.1. Compression

238
4.5.3.2. Pumping Speed

238
4.5.3.3. Ultimate Pressure

238
4.6. Turbomolecular Pumps

239
4.6.1. Theoretical Considerations and Performance Data

241
4.6.2. Design Considerations

241
4.6.2.1. Rotor and Stator Geometry

242
4.6.2.2. Rotor Suspension

242
4.6.2.3. Lubrication of Mechanical Bearings

242
4.6.2.4. Magnetic Rotor Suspension

243
4.6.2.5. Balancing and Vibration

243
4.6.2.6. Rotor Materials

243
4.6.2.7. Drive Systems

243
4.6.3. Applicational Considerations

243
4.6.3.1. Venting

244
4.6.3.2. Baking
244
4.6.3.3. Cooling

244
4.6.3.4. Operation in Magnetic Fields

245
4.6.3.5. Pumping Corrosive Gases

245
4.6.3.6. Pumping Toxic or Radioactive Gases

245
4.6.3.7. Turbomolecular Pumps in Combination with Other Pumps

< previous page page_xiv next page >


< previous page page_xv next page >
Page xv

245
4.6.4. Performance Data of Commercial Pumps

246
4.6.4.1. Compression

246
4.6.4.2. Pumping Speed

247
4.6.4.3. Ultimate Pressure

247
4.7. Combined Molecular Drag and Turbomolecular Pumps

248
4.7.1. Design Considerations

248
4.7.2. Typical Performance Data for Commericial Combined Molecular Drag and Turbomolecular
Pumps

248
4.7.2.1. Compression

248
4.7.2.2. Pumping Speed

248
4.7.2.3. Ultimate Pressure

248
4.8. Backing Pumps

Part III. Regenerative Drag Pumps 251


Nigel T. M. Dennis

251
4.9. Regenerative Drag Pumps

251
4.9.1. Mechanism

254
References

5. Capture Vacuum Pumps 259


Part I. Getters and Getter Pumps 261
Bruno Ferrario

261
5.1. Types of Gas Surface Interactions

262
5.2. Basic Concepts of Getter Materials

263
5.3. Adsorption and Desorption

265
5.4. Bulk Phenomena

265
5.4.1. Diffusion

267
5.4.2. Solubility

268
5.5. Equilibrium Pressures

269
5.6. Getter Materials

269
5.6.1. Basic Characteristics of Getter Materials

269
5.6.2. Sorption Speed and Sorption Capacity

271
5.6.3. Principal Types of Getter Materials and Their General Working Conditions

275
5.6.4. Interaction of Getters with Common Residual Gases

275
5.6.5. Evaporable Getters

276
5.6.5.1. Ba Getters

291
5.6.5.2. Titanium Sublimation Getter Pumps

297
5.6.6. Nonevaporable Getters
305
5.6.6.1. Ternary Alloys

310
5.6.6.2. Other Ternary and Multicomponent Alloys

< previous page page_xv next page >


< previous page page_xvi next page >
Page xvi

310
5.7. Getter Configurations

313
5.8. Getter Applications

314
5.8.1. Nonevaporable getters Versus evaporable getters

315
5.8.2. Start-up and Working Conditions of Getters

Part II. Sputter Ion Pumps 317


Hinrich Henning

317
5.9. Gas Discharge Vacuum Pumps

319
5.10. The Penning Discharge

321
5.10.1. Pump Sensitivity

323
5.10.2. Ion Motion

324
5.10.3. Electron Cloud

326
5.10.4. Secondary Electrons

327
5.10.5. Transition from HMF Mode to HP Mode

328
5.10.6. Transition from LMF Mode to HMF Mode

329
5.10.7. Sputtering

329
5.11. SIP Characteristics

329
5.11.1. Gettering
330
5.11.2. Ion Burial

331
5.11.3. Volume Throughput

335
5.11.4. Pumping Mechanism

338
5.11.5. Bakeout

338
5.11.6. Types of SIPs

342
5.11.7. Starting Properties

343
5.11.8. Memory Effect

343
5.11.9. Ultimate Pressure

345
5.11.10. Magnets

Part III. Cryopumps 347


Johan E. de Rijke

348
5.12. AdsorptionDesorption

352
5.13. Cryotrapping

353
5.14. Pumping Speed and Ultimate Pressure

355
5.15. Capacity

357
5.16. Refrigeration Technology

359
5.17. Pump Configuration

363
5.18. Regeneration
364
5.19. Partial Regeneration

364
5.20. Sorption Roughing Pumps

368
References

373
General References

< previous page page_xvi next page >


< previous page page_xvii next page >
Page xvii

6. Vacuum Gauges 375


R. Norman Peacock

377
6.1. Pressure Units Used in Vacuum Measurements

378
6.2. Liquid Manometers

379
6.3. McLeod Gauge

381
6.4. Piston Pressure Balance Gauge

382
6.5. Bourdon Gauge

384
6.6. Capacitance Diaphragm Gauges

385
6.6.1. Sensitivity of the Capacitance Method

386
6.6.2. Deflection of a Thin Tensioned Membrane

387
6.6.3. Accuracy of Commercial Gauges

388
6.6.4. Thermal Transpiration

388
6.6.5. Conclusions

389
6.7. Viscosity Gauges

391
6.7.1. Spinning Rotor Gauge

391
6.7.1.1. Theory

394
6.7.1.2. Commercial Gauges
397
6.7.1.3. Stability

399
6.7.1.4. Secondary or Transfer Standard

401
6.7.1.5. Use Precautions

401
6.7.1.6. Advantages and Disadvantages

402
6.7.2. Oscillating Quartz Crystal Viscosity Gauge

402
6.7.2.1. Advantages and Disadvantages

403
6.8. Thermal Conductivity Gauges

404
6.8.1. Theory

406
6.8.2. Calibration

408
6.8.3. Lowest Useful Pressure

409
6.8.4. Constant Pressure Pirani

410
6.8.5. Calibration Dependence Upon the Gas

410
6.8.6. Upper Pressure Limit

411
6.8.7. Ambient Temperature Compensation

412
6.8.8. Comparison of Pirani and Thermocouple Gauges

412
6.8.9. Stability

412
6.8.10. Thermistor Pirani Gauges and Integrated Transducers

413
6.8.11. Commercial Gauges and Applications
414
6.9. Ionization Gauges

414
6.9.1. Hot-Cathode Gauge Equation

419
6.9.2. Geometric Variations in the BayardAlpert Gauge

421
6.9.3. Modulated BayardAlpert Gauge

422
6.9.4. Extractor Gauge

423
6.9.5. Helmer Gauge

424
6.9.6. Long Electron Path Length Gauges

425
6.9.7. Secondary Standard Hot-Cathode Gauges

426
6.9.8. High-Pressure Ionization Gauges

427
6.9.9. Cold-Cathode Gauges

435
6.9.10. Ionization Gauge Accuracy

< previous page page_xvii next page >


< previous page page_xviii next page >
Page xviii

438
6.9.11. Gauge Constant Ratios for Different Gases

439
6.9.12. Ionization Gauge Controllers

441
References

7. Partial Pressure Analysis 447


Robert E. Ellefson

447
7.1. Ion Sources

448
7.1.1. Electron-Impact Ionization Process

449
7.1.2. Open Ion Source

452
7.1.3. Closed Ion Source

454
7.2. Ion Detection

454
7.2.1. Faraday Cup Ion Detection

454
7.2.2. Secondary Electron Multiplier Detection

456
7.2.3. Microchannel Plate Detector

456
7.3. Mass Analysis

456
7.3.1. Quadrupole Mass Analyzer

460
7.3.2. Magnetic Sector Analyzer

464
7.3.3. Time-of-Flight Mass Analyzer
465
7.3.4. Trochoidal (Cycloid) Mass Analyzer

466
7.3.5. Omegatron

467
7.4. Optical Measurement of Partial Pressures

468
7.4.1. Photoionization Measurement of Partial Pressure

469
7.4.2. Infrared Absorption Measurement of Partial Pressure

470
7.5. Computer Control, Data Acquisition, and Presentation

471
7.6. Residual Gas Analysis

474
7.7. Pressure Reduction Sampling Methods for Vacuum Process Analysis

475
7.8. Calibration of Partial Pressure Analyzers

477
References

8. Leak Detection and Leak Detectors 481


Werner Grosse Bley

482
8.1. Principles of Vacuum Leak Detection

482
8.1.1. Types of Leaks and Leak Rate Units

484
8.2. Total Pressure Measurements

486
8.3. Partial Pressure Measurements

486
8.4. Measurement of Leakage Rates with Helium Leak Detectors

487
8.5. Helium Leak Detection of Vacuum Components
490
8.6. Helium Leak Detection of Vacuum Systems

493
8.7. Special Methods and Other Tracer Gases

493
8.8. Mass Spectrometer Leak Detectors

494
8.8.1. Mass Spectrometer System for Helium Leak Detection

494
8.8.2. Direct-Flow Helium Leak Detectors

496
8.8.3. Simple Counterflow Helium Leak Detectors

< previous page page_xviii next page >


< previous page page_xix next page >
Page xix

498
8.8.4. Advanced Counterflow Helium Leak Detectors

499
8.8.5. Oil-Free and Dry Helium Leak Detectors

500
8.9. Specifications of Mass Spectrometer Leak Detectors

502
8.10. Quantitive Leakage Rate Measurements

504
8.11. Mass Spectrometer Leak Detectors for Other Tracer Gases and Future Developments in Leak
Detection

505
References

9. High-Vacuum System Design 507


Wolfgang Schwarz

507
9.1. Calculations of Vacuum Systems

508
9.1.1. Basic Pumpdown Equations

511
9.1.2. Process Pressure

513
9.2. Gas Loads in High-Vacuum Systems

513
9.2.1. Outgassing

516
9.2.2. Leaks

516
9.2.3. Permeation

518
9.2.4. Process Gas

519
9.3. Design of High-Vacuum Pump Sets
519
9.3.1. Forepump sets

519
9.3.1.1. Fore-Vacuum Pumps

520
9.3.1.2. Roots Combinations

524
9.3.2. High-Vacuum Pump Sets

526
9.3.2.1. Turbomolecular Pump Sets

528
9.3.2.2. Diffusion Pump Sets

531
9.3.2.3. Pump Sets with Cryosurfaces

535
9.3.2.4. Cryopump Sets

537
9.4. Calculation Methods for Vacuum Systems

538
9.4.1. Analytical Approximations

541
9.4.2. Numerical Methods

541
9.4.2.1. Dedicated Software

542
9.4.2.2. Network Approach

546
General References

10. GasSurface Interactions and Diffusion 547


John B. Hudson

548
10.1. Adsorption

548
10.1.1. Basic Equations
551
10.1.2. Adsorption Isotherms

567
10.1.3. Heat of Adsorption

568
10.1.4. Observed Behavior

572
10.1.5. Adsorption Kinetics

575
10.1.6. Chemisorption Kinetics

< previous page page_xix next page >


< previous page page_xx next page >
Page xx

582
10.1.7. Kinetic Measurements

584
10.1.8. Capillarity Effects

589
10.2. Absorption

590
10.2.1. Equilibrium Solubility

590
10.2.2. Diffusion Rates

591
10.2.3. Kinetics of Absorption and Permeation

592
10.2.4. Steady-State Permeation

595
10.2.5. Transient Permeation

600
10.2.6. Effect of Desorption Kinetics on Permeation

606
10.3. Surface Chemical Reactions

614
10.4. Outgassing Behavior

614
10.4.1. Desorption of Adsorbed Gases

616
10.4.2. Dissolved Gases

616
10.4.3. Overall Pumpdown Curves

619
10.4.4. Mitigation of Outgassing

619
10.4.5. Surface Treatments During Construction
620
10.4.6. In Situ Surface Treatments

620
10.4.7. Bakeout Processes

622
References

11. Ultrahigh and Extreme High Vacuum 625


Paul A. Redhead

628
11.1. Limits to the Measurement of UHV/XHV

629
11.1.1. Residual Currents

636
11.1.2. Effects at Hot Cathodes

639
11.1.3. Gauges with Long Electron Paths

641
11.1.4. Comparison of UHV/XHV Gauges

642
11.2. Limits to Pumps at UHV/XHV

642
11.2.1. Kinetic Pumps

643
11.2.2. Capture Pumps

646
11.2.3. Comparisons of Pumps for UHV/XHV

647
11.3. Leak Detection at UHV/XHV

648
11.4. Outgassing

648
11.4.1. Reduction of Outgassing Rates

652
11.5. UHV/XHV Hardware
652
References

12. Calibration and Standards 657


Karl Jousten

658
12.1. Primary Standards

659
12.1.1. Liquid Manometers and Piston Gauges

661
12.1.2. Static Expansion

< previous page page_xx next page >


< previous page page_xxi next page >
Page xxi

665
12.1.3. Continuous Expansion

670
12.1.4. Molecular Beam Expansion

673
12.2. Calibration by the Comparison Method

676
12.3. Calibration of Vacuum Gauges and Mass Spectrometers

676
12.3.1. Capacitance Diaphragm Gauges

680
12.3.2. Spinning Rotor Gauges

683
12.3.3. Ionization Gauges

686
12.3.4. Mass Spectrometers

689
12.4. Calibration of Test Leaks

692
12.5. Measurement of Pumping Speeds

695
References

Appendix 701

701
Graphic Symbols for Vacuum Components

708
Conversion Factors for Pressure Units

709
Vapor Pressure of Common Gases

711
Vapor Pressure of Solid and Liquid Elements

714
General Reference Books on Vacuum Science and Technology
Index 715

< previous page page_xxi next page >


< previous page page_xxiii next page >
Page xxiii

Acronyms

Acronyms used in the text are listed under the chapter where they occur.

Chapter 5
Part I. Getters and Getter Pumps

bcc Body-centered cubic (crystal structure)


CCRT Color cathode ray tube
CRT Cathode ray tube
fcc Face-centered cubic (crystal structure)
FED Field emission display
hcp Hexagonal close-packed (crystal structure)
HPTF High-porosity thick film (getter)
HT High temperature
HV High vacuum
LN2 Liquid nitrogen
PDP Plasma display panel
rf Radio frequency
RT Room temperature
SEM Scanning electron microscope
UHV Ultrahigh vacuum

Chapter 5
Part II. Sputter Ion Pumps

BA BayardAlpert (gauge)
DI Diode sputter ion pump with two cathode materials
FEM Finite element method
HMF High magnetic field (mode)
HP High pressure (mode)
LMF Low magnetic field (mode)
SIP Sputter ion pump
TM Transition (mode)
< previous page page_xxiii next page >
< previous page page_xxiv next page >
Page xxiv

UHV Ultrahigh vacuum


VTP Volume throughput

Chapter 5
Part III. Cryopumps

BET BrunauerEmmettTeller (adsorption model)


GM GiffordMcMahon (thermodynamic cycle)

Chapter 6
Vacuum Gauges

BAG BayardAlpert gauge


CCD Cold-cathode gauge
CDG Capacitance diaphragm gauge
ESD Electron-stimulated desorption
FS Full scale
HCG Hot-cathode gauge
JHP Jauge haut pression (Choumoff gauge)
NIST National Institute of Standards and Technology (USA)
PTB Physikalisch-Technische Bundesanstalt (Germany)
QBG Quartz helix Bourdon gauge
SCR Silicon controlled rectifier
SRG Spinning rotor gauge
UHV Ultrahigh vacuum
XHV Extreme high vacuum

Chapter 7
Partial Pressure Analysis
AP Appearance potential
CRDS Cavity ringdown spectroscope
IP Ionization potential
IR Infrared
MCP Microchannel plate
MS Mass spectrometer
PPA Partial pressure analyzer
QMS Quadrupole mass spectrometer
rf Radio frequency
RGA Residual gas analyzer
SEM Secondary electron multiplier
TOF Time of flight
TOFMS Time-of-flight mass spectrometer
UHV Ultrahigh vacuum
XHV Extreme high vacuum

< previous page page_xxiv next page >


< previous page page_xxv next page >
Page xxv

Chapter 11
Ultrahigh and Extreme High Vacuum

BA BayardAlpert (gauge)
BAG BayardAlpert gauge
ESD Electron stimulated descorption
EXB Cycloidal mass spectrometer
GIP Getter-ion pump
MG Magnetron gauge
IMG Inverted magnetron gauge
MS Mass spectrometer
MBAG Modulated BayardAlpert gauge
NEG Nonevaporable getter
RGA Residual gas analyzer
SIP Sputter-ion pump
TMP Turbomolecular pump
TSP Titanium sublimation pump
UHV Ultrahigh vacuum
XHV Extreme high vacuum

Chapter 12
Calibration and Standards

AVS American Vacuum Society


CDG Capacitance diaphragm gauge
DIN German industry standard
DKG German calibration service
IG Ionization gauge
IMGC Institutodi Metrologia ''G. Colonnetti" (Italy)
ISO International Organization for Standards
NIST National Institute of Standards and Technology (USA)
NPL National Physical Laboratory (England)
NPL National Physical Laboratory (India)
PTB Physikalisch-Technische Bundesanstalt (Germany)
QBS Quartz Bourdon spiral manometer
SI Système Internationale
SRG Spinning rotor gauge
UHV Ultrahigh vacuum
XHV Extreme high vacuum

< previous page page_xxv next page >


< previous page page_1 next page >
Page 1

1
Kinetic Theory of Gases

Benjamin B. Dayton

For a proper understanding of phenomena in gases, more especially at low pressures, it is essential to consider these
phenomena from the point of view of the kinetic theory of gases [1]. This theory rests essentially upon two fundamental
assumptions. The first of these postulates is that matter is made up of extremely small particles, which in the gaseous
state at moderate temperatures are monatomic or polyatomic molecules and at higher temperatures may be entirely
dissociated into atoms or even into positive ions and electrons to form a "plasma." The second postulate is that the
molecules of a gas are in constant motion, and this motion is intimately related to macroscopic properties known as the
temperature and pressure, which characterize the state of the gas in a given small region.

The center of mass of the molecule is assumed to move in a straight line (neglecting the force of gravity) with a constant
velocity between collisions with other molecules, and the forces between molecules are negligible except when the
molecular centers approach within a distance known as the mean molecular diameter. Velocity is a vector which must
be measured with respect to some "fixed" reference frame. The magnitude of the velocity vector is called the speed of
the molecule. In the "laboratory reference frame" the motion of the particles may be divided into bulk or fluid motion
due to pressure or concentration gradients which generate mass flow or diffusive flow and the random velocity
components associated with the concept of temperature as measured in a reference frame moving with the fluid flow
velocity. Kinetic energy is proportional to the square of the molecular velocity and is a scalar quantity.

Foundations of Vacuum Science and Technology, Edited by James M. Lafferty.


ISBN 0-471-17593-5 © 1998 John Wiley & Sons, Inc.

< previous page page_1 next page >


< previous page page_2 next page >
Page 2

Temperature is defined as a quantity proportional to the average (kinetic) energy of translation of the particles in a
reference frame moving with any fluid flow at a given small region. In the case of monatomic molecules (such as those
of the rare gases and the vapors of most metals), the effect of increased temperature is evidenced by increased
translational energy of the molecules. In the case of diatomic and polyatomic molecules, an increase in temperature also
increases through intermolecular collisions the rotational energy of the molecule about one or more axes, as well as
vibrational energy of the constituent atoms with respect to mean positions of equilibrium. However, in the following
discussion, only the effect on translational energy will be considered.

1.1
Ideal Gas Law

According to the kinetic theory, a gas exerts pressure on the enclosing walls because of the impact of molecules on
these walls. Since the gas suffers no loss of energy through exerting pressure on the stationary solid wall of its
enclosure, it follows that each molecule is thrown back from the wall with the same speed as that with which it
impinges, but in the reverse direction with respect to the normal; that is, the impacts are perfectly elastic.

Suppose a molecule of mass m to approach a flat wall surface lying in the x, y plane with velocity component vz
perpendicular to the wall. Since the molecule rebounds with the same speed, the change of momentum per impact is
2mvz. If ν molecules strike unit area in unit time with velocity component vz, the total impulse exerted on the unit area
per unit time is 2mvzν. But the pressure, P, on a wall is defined as the rate at which momentum is imparted to a unit area
of surface. Hence,

where the summation is over all values of vzν and it is assumed that all molecules have the same mass m.

It now remains to calculate ν. Of all the molecules within a volume ∆V extending outward from a small area, ∆x∆y, of
the wall by a distance |vzdt|, where dt is a short interval of time, at equilibrium only one half of the molecules will be
moving with velocity components vz toward the wall. Let nz denote the number of molecules per unit volume in the gas
within ∆V that have a velocity component of either vz or vz. Then the flux rate against the wall will be ν = nzυz/2, and
Eq. (1.1) becomes

The total speed of a molecule with velocity components vx, vy, and vz is the square root of the quantity

Defining the averages

< previous page page_2 next page >


< previous page page_3 next page >
Page 3

where n is the total number of molecules per unit volume regardless of velocity components, we have by symmetry
and

Then from Eq. (1.2) the basic equation for the pressure is

where ρ denotes the gas density, Eq. (1.6) can be expressed in the form

which shows that, at constant temperature, the pressure varies directly as the density, or inversely as the volume. This is
known as Boyle's law.

Now it is a fact that no change in temperature occurs if two different gases, originally at the same temperature, are
mixed. This result is valid independently of the relative volumes. Consequently, the average kinetic energy of the
molecules must be the same for all gases at any given temperature, and the rate of increase with temperature must be the
same for all gases. We may therefore define temperature in terms of the average kinetic energy per molecule, and this
suggestion leads to the relation

for each of the three degrees of freedom of translational motion, where T is the absolute temperature (degrees Kelvin),
defined by the relation T = 273.15 + t (t = degrees Centigrade), and k is a universal constant, known as the Boltzmann
constant. The total mean translational energy is then

where υr is known as the root-mean-square velocity. The total kinetic energy of the molecules in a volume V will be

Combining Eq. (1.8) with (1.5) it follows that Boyle's law can be expressed in the form

where the units chosen must be consistent with either the cgs system or the SI system. When n is expressed as
molecules/cm3 and k is in erg/K, then P will be the pressure in
< previous page page_3 next page >
< previous page page_4 next page >
Page 4

dyne/cm2. Also, from Eq. (1.6) and Eq. (1.11), it follows that

which is known as Charles' law or Gay-Lussac's law.

Lastly, let us consider equal volumes of any two different gases at the same values of P and T. Since P and V are
respectively the same for each gas, and is constant at constant value of T, it follows from Eqs. (1.10) and (1.11)
that n must be the same for both gases. That is, equal volumes of all gases at any given values of temperature and
pressure contain an equal number of molecules. This was enunciated as a fundamental principle by Avogadro in 1811,
but it took about 50 years for chemists to understand its full significance.

On the basis of Avogadro's law the molecular mass, M, of any gas or vapor is defined as that mass in grams, calculated
for an ideal gas, which occupies, at 0°C and 1 atmosphere, a volume [2]

V0 = 22,414.10 cm3.

This is therefore designated the molar volume, and the equation of state for an ideal gas can be written in the form

where W is the mass in grams, M is the molecular mass in grams, and R0 is a universal constant in units which depend
on the choice of units for the volume V, the pressure P, and the absolute temperature T. It is then convenient to express
Eq. (1.14) in the form

where nM denotes the number of moles (corresponding to M in grams) in the volume V under the given conditions of
temperature and pressure.

Excellent summaries of the various proposals prior to 1967 regarding the unit of pressure to be used in vacuum science
and technology have been given by Thomas and Leyniers [3]. When the units of length [l], mass [m], and time [t] are
chosen for the three basic units of a coherent system, the pressure unit is expressed by

[p] = [l]1 [m][t]2

or by

[p] = [F][l]2,

where F is the force with dimensions [l][m][t]2. In the cgs system the unit of force is the dyne and the pressure unit is 1
dyne/cm2. In the MKS or SI system the unit of

< previous page page_4 next page >


< previous page page_5 next page >
Page 5

force is the newton (abbreviated N), defined as 1 kg · m · s2 and the pressure unit is 1 N · m2. In 1962 the French
government approved the term "pascal" (abbreviated Pa) for the newton per square meter [4], and this term was
officially adopted by the International Standards Organization in 1978 while deprecating the use of older pressure units
such as mmHg and the torr (or Torr) [5]. Decimal multiples and submultiples of the Pa are accepted, and some have
special nomenclature such as

1 bar = 105 Pa,

1 millibar (mbar) = 100 Pa.

The previously used unit of 1 microbar (µbar) = 106 bar for the dyne/cm2 equals 101 Pa or 7.5 × 104 Torr. The main
reason for now using the mbar in vacuum technology is that its value is close to that of the Torr and the mmHg, which
were used for many years in the vacuum industry and the scientific literature. In 1954 both the British [6] and American
Committees on Vacuum Nomenclature [7] recommended replacing the mmHg by the Torr, defined as exactly
1,013,250/760 dyne/cm2, thus making it independent of the changing values for the measured density of mercury and
the acceleration of gravity. The term torr had previously been used in the German literature as a substitute for mmHg,
and at the First International Congress on Vacuum Techniques in Belgium in 1958 the newly defined Torr was favored
by the Germans and the Americans while the pascal was favored by the Belgians and the French representatives [8].
The Torr then became widely used throughout the world except in France until 1978, when the International Standards
Organization deprecated the use of Torr and defined the standard atmosphere as exactly 101,325 Pa and recommended
the pascal. Thus

In practical use the conventions of 1 Torr = 133.322 Pa and 1 Torr = 1333.22 µbar are sufficient. It may be noted that
the Torr is not identical to the unit mmHg used prior to 1955 which is based on the equation

where ρ = 13.5951 g/cm3 (at 0°C) is the density of mercury, g = 980.665 cm/s2 was the standard acceleration of gravity
accepted at that time, and h is the height (in mm) of a mercury column in a tube above some reference point such as the
surface of the mercury in the reference tube of a U-tube manometer or a McLeod gauge. One bar = 750.06 mmHg.
Another unit no longer used is the micron = 103 mmHg.

The basic unit of volume in the cgs system is 1 cm3, and that in the MKS or SI system is 1 m3. Following the Twelfth
Conférence Général des Poids et Mesures in 1964 the liter (or litre) is defined as 1 dm3 or 103 cm3 or 103 m3. The
preferred abbreviation for the liter is L, but l and the script l have been used. In the United States the volume unit
associated with industrial vacuum chambers and gas flow through mechanical pumps is commonly the cubic foot, which
equals 28.316847 liter. The basic unit of mass in the cgs system is the gram (g), and that in the MKS system is

< previous page page_5 next page >


< previous page page_6 next page >
Page 6

the kilogram (kg). The pound avoirdupois is seldom used for mass, but the pound force (poundf) is used in industrial
engineering in connection with the pressure unit pound per square inch (lb/in2.). 1 lb/in2. = 0.0689476 bar or 6.89476
kPa.

Units of temperature are the Celsius degree (°C) (also known as the degree centigrade) and the degree Kelvin (K) on the
absolute or thermodynamic scale. The zero of the Kelvin scale is 273.15°C, so that temperature T (in K) = temperature t
(in °C) + 273.15. The Fahrenheit temperature scale is seldom used in vacuum physics.

From the equation of state for an ideal gas we obtain the following values for the universal gas constant R0:

R0 = 760 · 22.41383/273.15 = 62.3632 Torr · L · K1 · mol1


= 1013250 · 22414.10/273.15 = 8.314511 · 107 erg · K1 · mol1
= 8.314511 joule · K1 · mol1
= 8.314511/4.1840 = 1.9872 cal (thermochemical) K1 · mol1.

In dealing with gases at low pressures, it is convenient to express the quantity of gas (Q = PV) in Pa · m3 at standard
room temperature 23°C. Note that 1 Torr · L = 0.133322 Pa · m3. The quantity of gas thus defined flowing per unit time
through a given cross section of a pipe has been called the throughput in the British and American Glossaries of terms
used in vacuum technology [9].

The unit of throughput is then

1 Pa · m3 · s1 = 1 N · m · s1 = 1 J · s1 = 1 W.

Objections have been raised against the use of pressurevolume products for quantity of gas on the basis that the
temperature of the walls of a vacuum system is not always clearly specified, and the suggestion is made that flow of gas
should preferably be expressed in mol/s to avoid any ambiguity due to unspecified temperatures [10]. The term
throughput should be reserved for flow in Pa · m3 · s1 at some specified temperature, such as a "standard" room
temperature of 23°C, and a term such as molar flow rate should be used for flow in mol · s1. Then

1 mol · s1 = 8.31451 × 296.15 Pa · m3 · s1 = 2462.342 Pa · m3 · s1 at 23°C.

1.2
Avogadro's Number

Avogadro's law states that the number of molecules per gram-molecular mass is a constant, which is designated NA.
Although a number of different methods have been used for the determination of this constant, the most accurate
method depends upon the determination of both the Faraday (F) and the charge of the electron (e).

For the deposition of one gram-equivalent the most accurate value [2] is

F = 96,485.309 absolute coulombs.

< previous page page_6 next page >


< previous page page_7 next page >
Page 7

Assuming that each univalent ion has a charge equal in magnitude to that of one electron,

NA = F/e.

The value of the charge of the electron is

e = 4.8032068 · 1010 absolute esu


= 1.60217733 · 1019 absolute Coulomb

since 1 absolute Coulomb = 1/10c times the charge in esu absolute coulombs, where c = velocity of light = 2.99792458
· 108 m · s1. Hence

NA = 6.0221367 × 1023 mol1.

From Eqs. (1.11) and (1.15) it follows that for one mole of gas in the volume V, nM = 1 and n = NA/V so that

From Eq. (1.9) it also follows that the average kinetic energy per molecule is given by

The mass per molecule is evidently

From Eq. (1.11) it follows that the number of molecules per cubic centimeter is given by

where Pµb is the pressure in microbars and PPa is the pressure in pascal units, while

where Pτ is the pressure in Torr.

Table 1.1 gives values of n for a series of values of T, Pµb, PPa and Pτ (Torr). The value n = 2.687 × 1019 cm3 is
known as the Loschmidt number.
< previous page page_7 next page >
< previous page page_8 next page >
Page 8

Table 1.1. Number of Molecules per Cubic Centimeter


T(°K) Pµb PPa Pτ n
273.15
1.01325 × 106 1.01325 × 105 760 2.687 × 1019
298.15
1.01325 × 106 1.01325 × 105 760 2.461 × 1019
273.15
1.333 × 103 1.333 × 102 1 3.535 × 1016
298.15
1.333 × 103 1.333 × 102 1 3.239 × 1016
273.15
10 1 7.50 × 103 2.652 × 1014
298.15
10 1 7.50 × 103 2.429 × 1014
273.15
1 0.1 7.50 × 104 2.652 × 1013
298.15
1 0.1 7.50 × 104 2.429 × 1013

1.3
Molecular Collisions; Mean Free Path; MaxwellBoltzmann Distribution Laws

It is evident that there must be a nonuniform distribution of velocities among all the molecules in a given volume because of the constant
occurrence of collisions. In an elementary treatment of the collision process the two molecules are assumed to be solid spheres of mass m1
and a well-defined diameter δ1 for one molecule and m2 and δ2 for the second molecule. The collisions are assumed to be elastic; that is, no
translational kinetic energy is lost by excitation of molecular rotation or atomic vibrations or by excitation of molecular vibrations when one
of the molecules is part of the wall of the enclosure. While the molecular diameters on impact can only be defined in terms of the fields of
attractive and repulsive force around each molecule and the relative momenta, it is assumed for simplicity that δ1 and δ2 have fixed values.
The distance between centers on impact will then be

and the mutual collision cross section is defined as

If we consider a single molecule of diameter δ1 (cm) moving at high speed, v, through a gas composed only of molecules of diameter δ2
(cm) moving relatively slowly and having a concentration of n molecules per cm3, then it is obvious that the average distance (in cm)
between collisions, known as the mean free path, will be given by

and the collision rate will be v/Lc = nσcv.

In an actual gas at equilibrium where all molecules have random velocities with an average speed va, the mean speed of a first molecule
relative to one of the other molecules depends on the angle between their respective directions of motion and the distribution law for
molecular velocities and will be equal to 21/2va in a homogeneous

< previous page page_8 next page >


< previous page page_9 next page >
Page 9

gas as calculated by Maxwell [11]. The collision rate is therefore

and the mean free path in a homogeneous maxwellian gas is

where in this case δ1 = δ2 = δ and σc = πδ2. This quantity is of fundamental importance in vacuum physics since the
Knudsen number defined by the ratio

where Dc is a characteristic length in the vacuum system, such as the diameter of a cylindrical tube, determines the
physics of the gas flow through tubes and ducts. From Eq. (1.11) and Eq. (1.26) in cgs units we obtain

and for nitrogen [12] we have δ = 3.78 × 108 cm, giving

At T = 298 K and Pτ = 103 Torr, this gives a mean free path for nitrogen L = 4.92 cm. At this temperature and Pµb = 1
µbar = 0.1 Pa = 7.5 × 104 Torr, L = 6.56 cm. This pressure is regarded as the upper limit of the high vacuum region [13]
in which the molecules may collide with the walls of the vacuum device more frequently than with other molecules in
the gas phase. We therefore must also consider the case in which a molecule of the gas collides with a molecule which
is part of the wall.

In the laboratory reference frame the velocity vector for the first molecule before the collision may be represented as

and for the second molecule it may be expressed as

where u1 and u2 are the velocity components perpendicular to the line of centers at the time of impact while v1 and v2
are the velocity components along this line of centers before collision as shown in Fig. 1.1.

Similarly, after the collision the velocity vectors for the first and second molecules, respectively, may be represented by
< previous page page_9 next page >
< previous page page_10 next page >
Page 10

Fig. 1.1
Velocity components
in elastic collision
between spheres.

where V1 and V2 are the respective velocity components along this line of centers after the collision, while U1 and U2
are the respective velocity components perpendicular to this line of centers after the moment of collision. For a collision
to take place, v1 and v2 must be antiparallel with vectors directed toward each other; but if they are parallel, one
velocity component must be greater than the other, and the relative positions, must be such that the molecule with the
larger component overtakes the other molecule in this direction.

The conservation of linear momentum in elastic collisions requires

Also, for the momentum components we require

The conservation of translational energy requires


since u1 and v1 are orthogonal components. Similarly,

< previous page page_10 next page >


< previous page page_11 next page >
Page 11

From Eqs. (1.36), (1.37), (1.38), and (1.39) we have

Also, from Eqs. (1.34) and (1.36) we obtain

Dividing Eq. (1.40) by Eq. (1.41) gives

Combining Eq. (1.42) and Eq. (1.43), we obtain

and similar calculations give

Equations (1.44) and (1.45) together with Eq. (1.36) give the vector components after the collision in terms of the
components before the collision. If m1 = m2, then V1 = v2 and V2 = v1 so that the velocity components along the line of
centers are interchanged. The resultant speeds after the collision are then

If the initial velocity component v2 was greater than v1, then the collision will cause the first molecule to increase in
velocity and the second molecule will suffer a decrease in velocity.

To illustrate how some molecules can acquire very high velocities while others are brought to near zero velocity as a
result of collisions, we consider a special class of orthogonal collisions between like molecules in which the first
molecule has an initial velocity u1 orthogonal to the line of centers at impact while the second molecule has an initial
velocity v2 parallel to this line of centers. In this special case we have v1 = 0, u2 = 0, V1 = v2, V2 = 0, U2 = 0 and the
resultant velocity of the second molecule is zero while the speed of the first molecule after the collision has been
increased from u1 to W1 as given by Eq. (1.46). Note that we could have assumed the second molecule to have any
value of a velocity component u2 other than zero orthogonal to
< previous page page_11 next page >
< previous page page_12 next page >
Page 12

the line of centers, but this component would have no effect on the final velocity of the first molecule as given by Eq.
(1.46).

If we consider only orthogonal collisions in which the first molecule is repeatedly struck in successive collisions by one
of the other molecules with a velocity component v2 which is a constant equal to the most probable velocity, vm, and
assume that the initial u1 = vm also, then N successive orthogonal collisions would give a final speed

By a process of reasoning similar to that for calculating the rate of absorbtion of a molecular beam by scattering in a gas
[14], it can be shown that the probability of N + 1 or more successive orthogonal collisions, without collisions of
another type, is

which is the probability that the speed of m1 is equal to or greater than W1(0 < W1 < ∞), where b is a constant to be
determined. When N is large, the velocity W1 can be much greater than vm but the probability is quite small. This
resembles the actual distribution law for molecular speeds as determined by J. C. Maxwell and L. Boltzmann.

The form of the distribution law differs according to the particular type of velocity distribution of interest. If we let
designate the components, along the three coordinate axes, of the randomly directed velocity v, then

and the distribution function, with respect to, say, is given by the relation

where N = number of molecules in the volume under consideration.

The distribution function for all three components of v has the form

With respect to the polar coordinates θ and φ, we have

The most important distribution function is that with respect to v in a random direction, which is given by the relation
Differentiating fv with respect to v, it is observed that the maximum value occurs for

< previous page page_12 next page >


< previous page page_13 next page >
Page 13

Hence α corresponds to the value of the most probable velocity, which is given by the relation

In terms of c = v/α, we can express Eq. (1.53) in the form

where c varies from 0 to ∞, and dy (= fcdc) corresponds to the fraction of the total number of molecules which have
values of c ranging between c and c + dc. Hence

From Eq. (1.57) we derive the value of the arithmetical average velocity, va = αca, where

and, using Eq. (1.55), we obtain

The root-mean-square velocity, vr, corresponds to the square root of the average value of v2 as derived from Eq. (1.9)
and is therefore given by the relation

The second column in Table 1.2 gives values of fc for a series of values of c, and Fig. 1.2 shows a plot of these data, on
which the values of fc are indicated for the values c = 1, 1.1284, and 1.2247, respectively.
< previous page page_13 next page >
< previous page page_14 next page >
Page 14

Fig. 1.2
Plots illustrating Maxwell-Boltzmann distribution laws. Plot fc shows distribution
function for random velocity, c expressed in terms of the most probable velocity α;
plot fx shows distribution function for energy, E, in terms of x = E/(kT); y corresponds
to the fraction of the total number of molecules for which the random velocity
(expressed in terms of α) is less than or equal to a given value c.

gives the fraction of the total number of molecules which have a random velocity equal to or less than that
corresponding to the value c, or to v = αc. The third and fourth columns in Table 1.2 show values of y and of ∆y, where
∆y gives the fraction of the total number which have velocities (in terms of α as a unit) ranging between c and the
immediately preceding value of c. Thus, 8.35% of the molecules have velocities between c = 1 and c = 1.1, and 42.76%
have velocities equal to or less than the most probable value. The values in parentheses are those of (1 y). A plot of y
versus c is shown in Fig. 1.2. It is evident that y corresponds to the area under the curve for fc from the origin to the
given value of c.

From Eq. (1.53) the distribution formula for translational energy (E) can be derived. It has the form

Substituting the variable x = E/(kT), Eq. (1.61) becomes


< previous page page_14 next page >
< previous page page_15 next page >
Page 15

Table 1.2. Values of fc, y, and fx, Illustrating Application of Distribution Laws
c fc y ∆y x fx
0
0 0 0 0
0.1
0.0223 0.0008 0.0008 0.05 0.2401
0.2
0.0867 0.0059 0.0051 0.1 0.3229
0.3
0.1856 0.0193 0.0134 0.2 0.4131
0.4
0.3077 0.0438 0.0245 0.3 0.4578
0.5
0.4393 0.0812 0.0374 0.4 0.4785
0.6
0.5668 0.1316 0.0504 0.5 0.4839
0.7
0.6775 0.1939 0.0623 0.6 0.4797
0.8
0.7613 0.2663 0.0724 0.7 0.4688
0.9
0.8129 0.3453 0.0790 0.8 0.4535
1.0
0.8302 0.4276 0.0823 0.9 0.4352
1.1
0.8142 0.5101 0.0835 1.0 0.4152
1.2
0.7697 0.5896 0.0795 1.2 0.3722
1.3
0.7036 0.6634 0.0738 1.4 0.3294
1.4
0.6232 0.7286 0.0642 1.6 0.2882
1.5
0.5350 0.7878 0.0602 1.8 0.2502
1.6
0.4464 0.8369 0.0491 2.0 0.2160
1.7
0.3624 0.8772 0.0403 2.2 0.1855
1.8
0.2862 0.9096 0.0324 2.5 0.1464
1.9
0.2204 0.9348 0.0252 3.0 0.0973
2.0
0.1652 0.9540 0.0192 3.5 0.0637
2.2
0.0864 0.9784 0.0244 4.0 0.0413
2.5
0.0272 0.9941 0.0157 4.5 0.0266
3.0 (4.2 × 104)
0.0024 0.0055 5.0 0.0170
4.0 4.1 × 106 (5.1 × 107)
6.0 0.0069
5.0 7.8 × 1010 (7.9 × 1011)
7.0 0.0027
6.0 1.9 × 1014 (4.4 × 1016)
8.0 0.0011

The last two columns in Table 1.2 give values of fx as a function of x, and Fig. 1.2 shows a plot of this function. By differentiating fx with
respect to x and equating the result to zero, it is readily shown that fx has a maximum value for x = 0.5; that is, fE has a maximum value for
E = 1/2k T. On the other hand, as stated in Eq. (1.9), Eav = 3/2kT.

Since

it is possible, from the plot for y in Fig. 1.2, to determine the fraction of the total number of molecules which have an energy equal to or less
than that corresponding to a given value of E.

It follows from the equations above that the value of v for which fv is a maximum increases with T½, while that of E for which fE is a
maximum increases with T.

Values of va, at 0°C and 25°C, for a number of gases and vapors are given in Table 1.3.

< previous page page_15 next page >


< previous page page_16 next page >
Page 16

Table 1.3. Masses, Velocities, and Rates of Incidence of Molecules


104 · νa

Gas or Vapor M 1023m 1010 0°C 25°C 1017ν1 1020 105G1 102
H2
2.016 0.3347 0.8878 16.93 17.70 11.23 14.97 0.3759 0.5012
He
4.003 0.6646 1.7631 12.01 12.56 7.969 10.63 0.5297 0.7062
CH4
16.04 2.663 7.063 6.005 6.273 3.981 5.308 1.060 1.414

NH3
17.03 2.827 7.498 5.829 6.089 3.865 5.152 1.092 1.456
H2O
18.02 2.992 7.936 5.665 5.919 3.756 5.007 1.124 1.498
Ne
20.18 3.351 8.886 5.355 5.594 3.550 4.733 1.190 1.586

CO 12.34
28.01 4.651 4.543 4.746 3.012 4.016 1.402 1.868
N2 12.34
28.02 4.652 4.542 4.745 3.011 4.015 1.402 1.868
Air 12.77
28.98a 4.811 4.468 4.668 2.962 3.950 1.425 1.900

O2 14.09
32.00 5.313 4.252 4.442 2.819 3.758 1.497 1.996
Ar 17.59
39.94 6.631 3.805 3.976 2.523 3.363 1.675 2.230
CO2 19.38
44.01 7.308 3.624 3.787 2.403 3.204 1.756 2.342

CH3Cl 22.23
50.49 8.38 3.385 3.356 2.244 2.991 1.881 2.508
SO2 10.64 28.21
64.06 3.004 3.139 1.992 2.656 2.118 2.825
Cl2 11.77 31.23
70.91 2.856 2.984 1.893 2.524 2.229 2.973

Kr 13.90 36.85
83.7 2.629 2.747 1.743 2.324 2.422 3.229
C7H16 16.63 44.12
100.2 2.403 2.510 1.593 2.123 2.650 3.533
Xe 21.80 57.82
131.3 2.099 2.193 1.392 1.856 3.034 4.044
CCl4 25.54 67.72
153.8 1.939 2.026 1.286 1.714 3.283 4.377
Hgb 33.31 (88.33)
200.6 1.698 1.774 (1.126 1.501 3.750 4.998)
Note: ν1 = rate of incidence of molecules per square centimeter per second, at 0°C and 1 µbar.

= rate of incidence of molecules per square centimeter per second, at 0°C and 1 Torr.
G1 = mass of gas corresponding to ν1 (g · cm2 · s1).

= mass of gas corresponding to (g · cm2 · s1).


m = mass of molecule (g); = density of gas at 0°C and 1 µbar (g · cm3).
va = average velocity (cm · s1).
a Calculated from the value ρ = 1.293 × 103 at 0°C and 760 Torr.
b Since the vapor pressure of mercury at 0°C is 1.85 × 104 Torr (= 0.247 µbar), the values given in parentheses have no physical
significance. Actual values at 0°C, corresponding to saturation pressure, are as follows: ρ = 21.79 × 1010; ν = 2.777 × 1016; G = 9.249 ×
106.

1.3.1
Relation between Molecular Velocities and Velocity of Sound

It is of interest to note that the relations for α, va, and vr can also be expressed in terms of the velocity of sound, which we shall designate by u.

Since

< previous page page_16 next page >


< previous page page_17 next page >
Page 17

where ρ1 = density at 1 µbar, and P = Pµb, we can write the relations for molecular velocities in the forms

On the other hand,

where γ = Cp/Cv = ratio of specific heats (per gram-mole) at constant pressure and constant volume. Hence

For mercury and other monatomic gases, γ = 1.667; for diatomic gases (such as H2, N2, and O2), γ = 1.40
(approximately). Hence,

va/u = 1.236 for monatomic gases


= 1.349 for diatomic gases
and
u/α = 0.9124 for monatomic gas.

Thus the velocity of sound in a gas approaches molecular velocities very closely.

1.3.2
Determination of Avogadro's Constant from Distribution of Particles in Brownian Motion

Under high magnification, all suspensions of very fine particles in gases or liquids exhibit ''Brownian" motions. Einstein
(1905) suggested that the motion of these particles is essentially that to be expected, on the basis of the kinetic theory of
gases, of "large molecules" and therefore subject to the same laws as gas molecules. That is, the average energy per
particle at any given temperature T is 3/2kT, and the average velocity of the particles is given by the relation

where m = mass of particle.

< previous page page_17 next page >


< previous page page_18 next page >
Page 18

Application of the BoltzmannMaxwell laws leads to the following relation for the distribution of particles at different levels in
a gravitational field:

where n0 = number of particles per cubic centimeter at h = 0,


n = number of particles per cubic centimeter at height h (centimeters),
g = 981 dynes,
m′ = apparent mass of particles, which is different from the actual mass because of the buoyancy of the
medium.

Let m = actual mass of particle. Then

where ρ′ = density of the medium and ρ = density of the particles.

Actually m′ is determined from the rate of settling of the particles, by application of Stokes' law.

Thus it is possible to determine k (and consequently the value of NA = R0/k) from observations on the value of m′ and the
relation between n/n0 and h. Using a fine suspension of gum arabic in water, Perrin obtained the value NA = 6.8 × 1023.

Equation (1.71) has been applied to the determination of the variation with altitude of the density of the atmosphere.

Assuming an average temperature of T = 230 K at higher altitudes, we obtain

m′/k = M/R0 = (29 × 107)/8.315,

and hence

where Pmm, is the pressure in mmHg (Torr) at the altitude H (in meters) above sea level.

1.4
Gas Pressure and Rate at Which Molecules Strike a Surface

The pressure in a gas is a tensor quantity and has to be defined with respect to an imaginary stationary plane surface passing
through a point in the gas. The pressure is defined as the net rate at which momentum normal to this surface is transmitted
across it per unit area in the positive direction, momentum transmitted in the opposite direction being counted as negative
[15]. Let ∆S represent a small surface element in this imaginary plane surface through which gas molecules included within a
hemispherical surface of radius L centered on the center of ∆S can pass directly without collision, where L is the mean free
path at the prevailing pressure as shown in Fig. 1.3.

< previous page page_18 next page >


< previous page page_19 next page >
Page 19

Fig. 1.3
Diagram illustrating the
calculation of the pressure in
a gas. L=the mean free path.
v=a particular molecular
velocity.

Consider the gas coming from within the solid angle dσ/4πL2, where

is a surface element on the hemisphere. For a gas at rest, momentum normal to the surface will be mv cos θ for such
molecules where v is the speed of the molecule within the range v and v + dv, and the rate at which this momentum is
transmitted across the surface is mv2 cos2θ. Then for a gas at rest with respect to this imaginary surface the total
momentum transmitted will be 2mv2 cos2θ per molecule since the momentum transmitted through ∆S from the opposite
side is counted as negative. The total momentum transmitted per unit solid angle will be 2nL · ∆S · mv2 cos2θ. Dividing
by L · ∆S to obtain force per unit area of surface, the quantity L · ∆S cancels out. If the molecular velocities are
distributed according to Eq. (1.53), the pressure is given by

This reduces to

using Eq. (1.9).

In a gas at rest with molecular velocities distributed according to the MaxwellBoltzmann equilibrium distribution, the
molecular flux across a plane surface element A due to all molecules having velocity vectors with directions within a
small solid angle dω whose axis makes an angle θ with the normal to A is given by the cosine law [16] formula,
< previous page page_19 next page >
< previous page page_20 next page >
Page 20

where is the average molecular velocity. While collisions occurring within the solid angle dω may scatter
molecules out of the region, in an equilibrium gas the collision processes must result in other molecules entering this
region and having the same direction. If we consider only those molecules crossing an imaginary plane surface in the
gas in the direction of the positive normal to the surface, or alternatively only those molecules which strike a plane solid
or liquid surface and are not scattered backward, then the total molecular flux across or against this surface is

Some of the molecules striking a solid surface will be adsorbed and the remainder are scattered back in various
directions depending on the surface roughness and the intermolecular forces during close approach. The adsorbed
molecules eventually reach accommodation with the prevailing temperature of the material at the surface; and because
they must acquire velocity components perpendicular to the surface due to molecular vibrations which have maximum
values ranging from zero to very high values by processes similar to those considered above for orthogonal collisions
within a gas, some of the molecules will acquire sufficient velocity after a certain time known as the adsorption lifetime
to escape the attractive force fields at the surface.

It has been shown by Comsa and coworkers [17] that the velocity distribution in the gas from these molecules which are
desorbed is not necessarily that of a Maxwellian gas. Comsa and several other investigators [18] have shown
experimentally that the angular distribution does not necessarily obey the cosine law. The experiments show that the
evaporated or desorbed flux is peaked in the direction of the normal to the surface varying as cosnθ with n greater than
1 and as high as 9 for strongly peaked emissions. However, as deduced by Clausing and confirmed by Comsa, under
equilibrium conditions the sum of the distribution of the molecules leaving the surface due to various processes
(adsorptiondesorption, reflection, diffraction, and inelastic scattering) has to obey the cosine law.

Epstein [19] has presented a model of the wall boundary condition in terms of the relation between the distribution
functions of the incident and reflected particles, assuming that a certain fraction reflect diffusely (uncorrelated with the
incident conditions) while the remainder reflect specularly depending on the velocity of the incident particles.

The molecular flux of vapor, as measured in the laboratory frame, in the beam of gas issuing into a vacuum from a
Knudsen cell, comprising a source of vapor at a uniform temperature T enclosed in a box with a small thin-edged orifice
in one wall, obeys the cosine law approximately, but the mean translational energy per molecule in the beam at the exit
is 2kT rather than (3/2)kT because the faster molecules have a higher probability of exiting the orifice and the mean
energy per molecule involves averaging the translational energy with a velocity distribution function which

< previous page page_20 next page >


< previous page page_21 next page >
Page 21

contains the factor v3 rather than v2. In fact, this value 2kT is characteristic of the mean translational energy in the flux
of molecules in an equilibrium gas against any surface and must be considered when calculating heat transfer.

At very low pressures where the mean free path is much greater than the linear dimensions of the vacuum vessel, the
velocity of the molecules in the gas phase is entirely determined by collisions with the walls and the temperature of the
walls. Since such collisions are far less frequent than collisions in a dense gas where the m.f.p is only a small fraction of
1 cm, the relaxation time, or time to restore equilibrium in a gas which has been disturbed from equilibrium by
transitory pressure or temperature gradients, is relatively long. Because of outgassing and readsorption at various
surfaces within a vacuum system and the removal of molecules by vacuum pumps, equilibrium conditions seldom
prevail [20], and the pressure measured by a tubulated vacuum gauge will depend on the orientation of the plane of the
entrance of the tubulation with respect to mass flow vectors [21]. Pressures measured by an ionization gauge, which has
the envelope and tubulation removed down to the base mounted inside the vacuum chamber, are less dependent on
orientation except for the shielding effect of the base.

Muntz [22] has described a method of making localized measurements of the molecular velocity distribution function in
rarefied gas flows.

From Eq. (1.78) the incident molecular flux per unit area is

Substituting for n and va, from Eqs. (1.20), (1.21), and (1.58), we obtain the relations

The volume which strikes unit area per unit time is given by

and is therefore a constant at all pressures, but varies with (T/M)1/2.

In the literature, especially that originating in Germany, Eq. (1.87) is expressed in the form

where ρ1 is the density at 1 µbar.


< previous page page_21 next page >
< previous page page_22 next page >
Page 22

The last four columns of Table 1.3 give values of v1 and G1, the values calculated for a pressure of 1 microbar, and of
the values calculated for a pressure of 1 Torrall at 0°C.

The equations given above for v and G are also applicable to the effusion of gases at low pressures through small holes
in very thin plates. The requisite condition for the application of Meyer's relation to effusion is that the diameter of the
opening should be small compared with the mean free path.

A comparison of relative values of v or G for different gases or vapors streaming through such a hole makes it possible
to obtain relative values of M, since, for constant values of P and T, v varies inversely as M1/2, and G varies directly as
M1/2.

A good check on the above equations was obtained by Knudsen [23] in some experiments in which hydrogen, oxygen,
and carbon dioxide, at pressures ranging from 100 to 0.01 Torr, were made to flow into a vacuum through a 0.025-mm
hole in a 0.0025-mm-thick platinum strip.

Equation (1.87) shows that the volume per unit area per unit time, measured at the pressure P, is always the same. It
follows that FPµb corresponds to the volume at 1 µbar. Hence, if Pµb1 and Pµb2 (> Pµb1) denote the pressures on the
two sides of a very thin-walled orifice of area A, the net quantity of gas (Q) flowing through the orifice per unit time is
given by

where Q = volume in cubic centimeters per second, measured at 1 µbar. That is, Q denotes microbars × cubic
centimeters per unit time.

1.5
Rate of Evaporation and Vapor Pressure

An interesting application of Eq. (1.84) was first made by Langmuir [24] to the determination of vapor pressure from
rates of evaporation in high vacua. Quoting from Langmuir's original paper, The Vapor Pressure of Metallic Tungsten:

Let us consider a surface of metal in equilibrium with its saturated vapor. According to the kinetic theory we
look upon the equilibrium as a balance between the rate of evaporation and rate of condensation. That is, we
conceive of these two processes going on simultaneously at equal rates.

At temperatures so low that the vapor pressure of a substance does not exceed a millimeter, we may consider
that the actual rate of evaporation of a substance is independent of the presence of vapor around it. That is, the
rate of evaporation in a high vacuum is the same as the rate of evaporation in presence of saturated vapor.
Similarly we may consider that the rate of condensation is determined only by the pressure of the vapor.

The rate at which molecules will, in general, condense on a surface is given by αv, where α is known as the
condensation coefficient or sticking coefficient. It represents the ratio between the rate at which molecules actually
condense on the surface and the rate at which they strike the surface. If we let µ denote the rate at which molecules
evaporate from the surface, then, at equilibrium, we obtain

< previous page page_22 next page >


< previous page page_23 next page >
Page 23

Langmuir [25] has shown that for metal atoms condensing on the surface of a metal the value of α may be assumed to
be equal to 1. In a later paper on the vapor pressures of high-boiling point organic liquids, Verhoek and Marshall [26]
showed that the same assumption is justified in respect to these liquids. Hence, we may, in practically all cases of
evaporation from the bulk phase, express the relation for rate of evaporation in the form

For the purpose of calculating the vapor pressure of a metal from a determination of loss of weight per unit area per unit
time, it is convenient to express Eqs. (1.85) and (1.86) in the forms

where G = rate of evaporation in grams per square centimeter per second.

As an illustration of the application of these equations, Table 1.4 gives values of G for tungsten [27] and tantalum [28]
at a series of temperatures (degrees K) together with calculated values of Pµb.

For the evaporation from a wire of diameter d′ (in mils) the loss in weight per second per centimeter length is given by

G1 = 2.54 × 103πd′G.

Hence Eqs. (1.92) and (1.93) assume the forms

Table 1.4. Rates of Evaporation and Vapor Pressures of Tungsten and Tantalum
Metal T(K) G Pµb
Tungsten 2600 8.41×109
M = 183.92 7.23×104
2800 1.10×107
9.81×103
3000 9.95×107
9.18×102
3200 6.38×106
6.08×101
3400 3.47×105
3.41

Tantalum 2400 3.04×109


M = 180.88 2.58×104
2600 5.54×108
4.90×103
2800 6.61×107
6.07×102
3000 5.79×106
5.40×101
3200 3.82×105
3.77

< previous page page_23 next page >


< previous page page_24 next page >
Page 24

Equations (1.92) and (1.93) have also been applied by Knudsen and subsequent investigators to the determination of
vapor pressures from rates of effusion through a small orifice.

Thus let us consider the case in which molecules evaporating from a hot surface pass through a small orifice into
another chamber in which they are condensed. If the pressure of residual gas in this "cool" compartment is extremely
low and the radius of the opening is less than L (the mean free path of the evaporating molecules in the "hot"
compartment), then the rate at which molecules pass through the hole is equal to the rate at which they strike this
opening. Consequently, the vapor pressure for any given temperature will be given by Eq. (1.92) or Eq. (1.93), where G
represents the weight passing through the orifice per unit area, per unit time.

These equations are, however, strictly applicable only if the thickness (l) of the wall, in which the orifice of area πa2 is
located, is vanishingly small compared to a. If the orifice consists of a short tube for which l/a is appreciable, then a
correction factor has to be applied, and instead of Eq. (1.82) we have the relation

where K is a function of l/a which is less than 1 for l/a > 0. The manner in which the value of K varies with l/a is
discussed subsequently in Chapter 2. Hence, if G′ denotes the actual loss in weight, at temperature T, of material of
molecular mass M, through an opening of area A, over a period of t seconds, then we obtain

These equations have been applied by a number of investigators for the determination of vapor pressure at low
temperatures, where the values are of the order of a few pascals. The method has been used, for instance, by Egerton for
such metals as zinc, cadmium, mercury [29], and lead [30].

As an illustration let us consider one such determination made for mercury vapor. In this case the area of the opening
was A = 0.0335 cm2. At 33.7°C, the loss of mercury through this orifice was 0.7867 g over a period of 2370 min. To
correct for the fact that l/a was not negligible, the value of K was found (by means of the relations given in Chapter 2) to
be 0.93.

Hence the corrected value of G is given by

Since T = 306.9 and M = 200.6, it follows from Eq. (1.93) that Pτ = 3.77×103 Torr.

Another interesting application of the above relations, and one which is of increasing importance in industrial
chemistry, is provided by the development of

< previous page page_24 next page >


< previous page page_25 next page >
Page 25
high-vacuum distillation for the separation of certain organic compounds in the pure state from naturally occurring oils. The
great advantage of this process arises from the fact that these organic compounds are unstable at higher temperatures and
therefore they can be distilled only at lower temperatures, at which the vapor pressures are in the range of 104 to 106 atm [31].

In this operation, evaporation takes place from a very thin film of liquid, which is renewed continuously, and condensation
occurs on an adjacent cooled surface. In a sufficiently high vacuum (pressure of residual gas less than 1 Pa) the rate of transfer
of distilland is in accordance with Eqs. (1.85) and (1.86). As the pressure of residual gas is increased, however, the rate of
distillation is decreased because of collisions between the molecules of the distilland and those of the gas. This is illustrated by
the data shown in Table 1.5, taken from Hickman's paper.

The distilland used was Octoil, which has the chemical formula C6H4 (COOC8H17)2 and molecular weight M = 390.3. From
Eq. (1.86), it follows that the rate of evaporation, W, in grams per second per square meter, is

where T is the absolute temperature corresponding to the vapor pressure of Pτ in Torr or Ppa in pascals.

It is evident from these observations, as well as from observations of a similar nature mentioned in the next section, that
molecules leaving the surface of the distilland are prevented from reaching the surface of condensation because of collisions
with the molecules of the residual gas. As a result of such collisions, many of the molecules leaving the hot film are driven
back, the number of such molecules increasing with the magnitude of residual pressure.

Table 1.5. Variation of Rate of Distillation with Pressure (Hickman)a


Pressure (mTorr) of Residual Gas W (g · s1 · m2)
(Air)
Pτ = 103 Pτ = 3×103 Pτ = 102
T = 368 K T = 383 K T = 393 K
0.3 0.6 1.85 6.4
4.0 0.46 1.59 5.7
7.0 0.38 1.37 5.2
10.0 0.32 1.18 4.6
15.0 0.25 0.95 3.8
25.0 0.21 0.70 2.1
50.0 0.12 0.40 1.67
aThe values given for T were taken from a plot of log Pτ versus 1/T and are therefore only approximate, which
accounts for the fact that values of W calculated by means of Eq. (1.99) for extremely low pressure are slightly less
for 3×103 and 102 Torr than those given in the first row.

< previous page page_25 next page >


< previous page page_26 next page >
Page 26

1.6
Free Paths of Molecules

Although the individual molecules in a gas at rest possess very high velocities, as shown previously, it is a matter of
ordinary observation that gases diffuse into one another very slowly. This is explained on the kinetic point of view by
assuming that the molecules do not travel continuously in straight lines, but undergo frequent collisions. The term
"collision" naturally leads to the notion of free path. This may be defined as the distance traversed by a molecule
between successive collisions. Since, manifestly, the magnitude of this distance is a function of the velocities of the
molecules, we are further led to use the expression "mean free path" (denoted by L), which is defined as the average
distance traversed by all the molecules between successive collisions.

However, this definition assumes that the molecules actually collide like billiard balls; that is, the molecules are
assumed to be rigid elastic spheres possessing definite dimensions and exerting no attractive or repulsive forces on one
another. But this concept can certainly not be in accord with the facts. It is probably impossible to state definitely the
diameter of a hydrogen atom or molecule, much less that of a poly-atomic molecule. Also there is no doubt that the
molecules exert attractive forces on one another for certain distances and repulsive forces when they approach
exceptionally close. Otherwise, how could we explain surface tension, discrepancies from Boyle's law, and a host of
related phenomena? To speak of collisions among molecules, such as these, is impossible. What meaning, therefore,
shall we assign to the free path under these conditions?

Let us consider at t = 0 a group of N0 "tagged" molecules moving in a given direction. As time goes on, these molecules
will suffer random collisions and a number will disappear from the original group. Let N denote the number which, after
a period t, are still identified with the original group, and let ω denote the collision frequency. Then

Integrating this equation, we obtain the result,

If we let l (= vat) designate the path that has been traversed by a molecule without suffering collision during the interval
t, then Eq. (1.101) can be written in the form

Furthermore, we can write

where L is a distance covered between collisions. Then it follows that Eq. (1.102) assumes the form

< previous page page_26 next page >


< previous page page_27 next page >
Page 27

That is, φ(l) is the fraction of the original group of molecules that are still traveling without having suffered a collision
in the distance l.

Furthermore, it follows from Eq. (1.102) that

represents the fraction of all the free paths that have a length between l and l + dl. (Hence the omission of the negative
sign in the differentiation.)

It follows from Eq. (1.105) that the average value of the free path is

For l = L, φ(l) = εl = 0.3679. This result shows that 63.21% of the molecules collide with other molecules in a distance
equal to or less than L. Furthermore, it is seen from Eq. (1.101) that this 63.21% of collisions occur in the interval τ = 1/
ω. Thus 1/ω is a constant of the same nature as the ''decay" constant in radioactive disintegrations, while 1/L may be
regarded as an "absorption" coefficient similar to the coefficient that measures the decrease in intensity of a beam of
light in passing through a medium.

Equation (1.104) indicates an experimental method for the determination of L′ which has been used by Born [32] and
Bielz [33] and which is described by Fraser [34]. A beam of silver atoms is sent into nitrogen or air, and a determination
is made of the amount of silver deposited by the beam in a given time t on a surface distant l from the source. If we let
I0 designate the intensity of the beam at the source, then the intensity at the collector is

where LP is the mean free path of silver atoms in the gas at the pressure in the collecting chamber.

Measurements of the mean free path of potassium in nitrogen have also been reported by Weigle and Plesset [35].

As Fraser [34] points out:

With noncondensable gases, it is not possible to measure I0 directly. We assign therefore to I0 a different
meaning: namely, the intensity which the beam would have if it were not, as is actually the case, weakened
through scattering by the alien molecules present in the collimator chamber. Now, clearly I0 is directly
proportional to the quantity of gas issuing from the source slit per second; but so also is the pressure P in the
collimator chamber, if a constant pump speed is assumed. We can therefore set I0 = c·P. On the other hand,
LP is inversely proportional to P; that is, LP = L/P, where if P is measured say in mTorr, L is the mean free
path at a pressure of 1 mTorr. I can therefore be expressed as a function of the pressure P; thus if l is the
distance between source slit and image slit,

I = c · P eP · l/L,

< previous page page_27 next page >


< previous page page_28 next page >
Page 28

it being assumed that the pressure in the observation chamber is negligibly small. I is a maximum for that
value of P which makes P · l/L = 1. To make a measurement, the intensity is plotted as a function of the
pressure in the collimator chamber, and the value of the latter at the maximum intensity is observed. Then

LP = L/P = l.

At this value of P, I = 0.3679I0.

It is of interest to observe, as Fraser emphasizes, that the requisite condition for obtaining a directional effect of the
molecules passing through the slit is that LP must not be less than d, the width of the slit.

The determination of mean free path for hydrogen has been carried out by Knauer and Stern [36]. The value they
obtained, however, is only about 0.44 times that derived from viscosity relations (see discussion in the following
section). The reason, as Fraser [34] points out, is that

the standard methods require an intimate encounter in order that the molecules may exchange energy and
momentum in amounts capable of affecting the viscosity or heat conductivity of the gas. The molecular ray
method on the other hand counts as a collision an approach of two molecules sufficiently close to deflect them
very slightly out of their paths; with narrow slits angular deflections of less than 104 are detectable.

In this connection the reader will find an interesting description of the many uses of molecular beams in a paper by
Taylor [37]. As he states, "Molecular beams, narrow rays of molecules formed by a slit system and moving in one
direction in an evacuated apparatus, may be used to advantage in many types of research." Among these are
determinations of molecular velocities (involving experimental tests of the validity of the MaxwellBoltzmann
distribution law), mean free paths, vapor pressures, accommodation coefficients, and mechanism of chemical reactions
and of adsorption.

Evidently the mean free path must depend upon the molecular diameter, and simple considerations indicate that the
length of the mean free path must vary inversely as the total cross-sectional area of the molecules per unit volume.
Again, the magnitudes of the coefficients of viscosity, heat conductivity, and diffusivity of gases are intimately bound
up with the length of the free path; whether it be transference of momentum from one layer to another as in viscosity, or
transference of increased kinetic energy of the molecules as in heat conductivity, the rate of this transference must
depend upon the number of collisions that each molecule experiences as it passes from point to point. It is therefore to
be expected that there should exist very similar relations between the values of the mean free path and those of the
coefficients of viscosity, heat conductivity, and diffusion. However, in attempting to deduce such relations, the
theoretical physicist has found himself confronted with the problem regarding the laws governing the variation with
distance of attractive and repulsive forces between molecules. As a result of successive attacks on this problem, by a
number of investigators, the exact forms of these relations have been modified from time to time. The reader is referred
to Chapman and Cowling [1d] for a detailed discussion of the whole problem.

< previous page page_28 next page >


< previous page page_29 next page >
Page 29

1.7
Relation between Coefficient of Viscosity, Mean Free Path, and Molecular Diameter

A gas streaming through a narrow-bore tube experiences a resistance to flow, so that the velocity of this flow decreases
uniformly from the center outwards until it reaches zero at the walls. Each layer of gas parallel to the direction of flow
exerts a tangential force on the adjacent layer, tending to decrease the velocity of the faster-moving and to increase that
of the slower-moving layers. The property of a gas (or liquid) by virtue of which it exhibits this phenomenon is known
as internal viscosity.

As a simple working hypothesis we may assume, as Newton did, that the internal viscosity is directly proportional to the
velocity gradient in the gas. Furthermore, the viscosity must depend upon the nature of the fluid, so that in a more
viscous fluid the tangential force between adjacent layers, for constant velocity gradient, will be greater than in a less
viscous fluid. We thus arrive at the following definition of the coefficient of viscosity:

The coefficient of viscosity is the tangential force per unit area for unit rate of decrease of velocity with distance (i.e.,
per unit velocity gradient).

With this definition we are in a position to deduce the approximate form of the relation between the coefficient of
viscosity and the free path.

Let u denote the velocity of flow of the gas at a distance d from a stationary surface. In uniform flow along a surface,
the velocity will decrease uniformly to zero as the surface is approached. We can therefore represent (see Fig. 1.4) the
velocity at distance OA = d by the ordinate AB = u and represent velocities at intermediate distances by the
corresponding ordinates of the line OB.

We shall imagine the gas divided into layers parallel to the surface, each having a depth equal to the free path, L.

Fig. 1.4
Diagram illustrating the derivation of simple
relation between the coefficient of viscosity (η)
of gas and the molecular mean free path (L).
< previous page page_29 next page >
< previous page page_30 next page >
Page 30

Let us denote the tangential force per unit area between adjacent layers by B. By definition:

where η denotes the coefficient of internal viscosity.

But, according to the kinetic theory, the tangential force per unit area is measured by the rate at which momentum is
transferred per unit area between adjacent layers. Because of the relative motion of the layers, the molecules moving
from a faster- into a slower-moving layer possess more momentum in the direction of flow than those moving in the
opposite direction.

Let us consider any layer, CE or EH, of thickness equal to L. We have chosen this particular value of the thickness so
that we may be justified, as a first approximation, in assuming that the molecules starting at either of the planes CD and
HK reach the plane EF without suffering collisionthat is, without change of momentum.

The momentum, parallel to the surface, of any molecule reaching the plane EF from the plane CD is m(u′ + v), where u′
denotes the velocity of flow at the plane CD and v is the mean velocity of the molecules.

The momentum, parallel to the surface, of a molecule reaching the plane EF from the plane HK is

m(u′ + v + 2uL/d).

The number of molecules that cross unit area per unit time in any direction in a gas at rest is equal to (1/6) nv; and this
must be the same for the molecules traveling in a direction perpendicular to the plane EF, because the velocity of flow
is assumed to be so small that the density remains constant throughout the different layers.

Hence the net rate of transference of momentum across unit area of the plane EF is equal to

From Eqs. (1.108) and (1.109) it follows that

The dimensions of η are evidently ml1t1, and in the cgs system the unit of viscosity is 1 poise = 1 g·cm1·s1 = 1
dyne·s·cm2. This is the unit of coefficient of viscosity used in this volume.

In deducing Eq. (1.110) it has been assumed that all the molecules possess the same velocity v and the same free path L.
Introducing the law of distribution of velocities, Boltzmann (1881) deduced the relation

where va = average velocity, and LB is defined as the average free path.

< previous page page_30 next page >


< previous page page_31 next page >
Page 31

O. E. Meyer, in his Kinetic Theory of Gases, used a different method of calculation and derived a relation of the form

where LM is also defined as the average free path.

From these equations an interesting conclusion may be deduced regarding the dependence of viscosity on pressure. As
has been mentioned, it is evident from very simple considerations that L must vary inversely as the number of molecules
present per unit volume. Consequently the product pL is constant and independent of the pressure. The velocity, v,
depends only upon the temperature and molecular weight. It therefore follows that, for any gas at constant temperature,
the viscosity is independent of the pressure and must increase with the temperature. The confirmation of these two
deductions has been justly regarded as one of the most signal triumphs of the kinetic theory of gases. As is well known,
the viscosity of all ordinary liquids decreases with increase in temperature. That the viscosity of gases must increase
with temperature was therefore regarded as a remarkable conclusion.

At both extremely low pressures and very high pressures, the conclusion that the viscosity is independent of the pressure
is not in accord with the observations, but this is due to the fact that the same derivation as has been presented above is
not valid under those conditions where either attractive forces between the molecules come into play or the pressure is
so low that a molecule can travel over the whole distance between the walls of the enclosure without suffering collision.

Both Eqs. (1.111) and (1.112) have been used by physicists, until comparatively recently, for deriving values of L from
η. However, the work of S. Chapman and D. Enskog, since about 1911, has led to the following relations, which are
discussed by Chapman and Cowling [1d] in their treatise.

To derive a more exact relation between η and L, it is necessary to introduce the relation between L and δ, the molecular
diameter, which has been shown to be of the form

where n = number of molecules per cubic centimeter.

For smooth rigid elastic spherical molecules, it is shown that

Substituting in this equation the relations va = (8kT/pm)½ and ρ = mn, it follows that

A further approximation leads to the conclusion that the right-hand side of the last equation should be multiplied by the
factor 1.016, and consequently

< previous page page_31 next page >


< previous page page_32 next page >
Page 32

Combining this with Eq. (1.113), the result is the relation used in the following discussion:

It follows directly from Eq. (1.114) that, for two gases having approximately equal values of δ, the viscosities should be
in the same ratio as the square roots of the molecular masses. This conclusion has been confirmed by observations on
the relative viscosities of H2 and D2 (deuterium) [38]. At room temperature we have ηD2/ηH2 = 1.39, which is
approximately equal to 21/2.

The difference between Eq. (1.116) and the equations of Boltzmann and Meyer arise from the fact that the two latter
investigators failed to take into account the existence of forces of attraction and repulsion between molecules. The
accurate form of the relation between η and δ has been a topic of considerable discussion by theoretical physicists [39].

One interesting contribution to this subject, to which reference is made in a subsequent section, is the model suggested
by W. Sutherland [40]. Let us assume that the molecules are "smooth rigid elastic spheres surrounded by fields of
attractive force."

The relation between coefficient of viscosity and molecular diameter then has the form

where the constant C is a measure of the strength of the attractive forces between the molecules. In this equation the
notation δm has been introduced to indicate that the value of the molecular diameter derived in this manner is different
from the value δ used in Eq. (1.114). From a comparison with Eq. (1.114) it follows that Eq. (1.117) is equivalent to the
relation

Equation (1.117) leads to the relation, used for calculating C, which has the form

< previous page page_32 next page >


< previous page page_33 next page >
Page 33

where ηT and η0 are the values of the coefficient of viscosity measured at T and T0 respectively. Equation (1.122) is
also often expressed in the form

By combining Eqs. (1.123) and (1.118) it is seen that the value of δm thus deduced is independent of T and actually
corresponds to the value of the molecular diameter at infinitely high temperature, whereas, as pointed out below, the
value of δ decreases with increase in T. For this reason, Eqs. (1.118) and (1.119) were used formerly by many writers on
this subject. However, following the procedure of Chapman and Cowling and of Kennard, Eqs. (1.115) and (1.116) are
used in the following discussion for the calculations of the values of δ and L, respectively. The conversion to values
deduced by application of Sutherland's theory may then be made by means of Eqs. (1.120) and (1.121).

From Eq. (1.116) the following numerical relations are derived:

where Pµb and Pτ denote the pressure in microbars and Torr, respectively, and η is expressed in poises. These relations
have been used to calculate values of L and, from these, values of δ given in Tables 1.6 and 1.7.

From Eq. (1.115) we obtain the following relations for δ2 and Sc = πδ2, the latter being defined as the mean equivalent
cross section for viscosity [41]:

Where δ has been determined by some other method [42], an approximate calculation of L may be made by means of
the relations

Another useful magnitude is the collision frequency per molecule, which is given by the relation
Values of ω are shown in the last row in Table 1.6.

< previous page page_33 next page >


< previous page page_34 next page >
Page 34

Table 1.6. Mean Free Paths, Molecular Diameters, and Related Data for a Number of Gases
Gas: H2 He Ne Air O2 Ar CO2 Kr Xe
107η15
871 1943 3095 1796 2003 2196 1448 2431 2236
x
0.69 0.64 0.67 0.79 0.81 0.86 0.95 0.85 0.92
107η0
839 1878 2986 1722 1918 2097 1377 2372 2129
107η25
892 1986 3166 1845 2059 2261 1496 2502 2308

103
8.39 13.32 9.44 4.54 4.81 4.71 2.95 3.69 2.64

11.04 17.53 12.42 5.98 6.33 6.20 3.88 4.85 3.47

9.31 14.72 10.45 5.09 5.40 5.31 3.34 4.06 2.98

12.26 19.36 13.75 6.69 7.10 6.67 4.40 5.34 3.93


108δ
2.75 2.18 2.60 3.74 3.64 3.67 4.65 4.15 4.91
C
84.4 80 56 112 125 142 254 188 252
1014Ns
15.22 24.16 17.12 8.24 8.71 8.54 5.34 6.69 4.78
109ω
14.45 7.16 1.68 6.98 6.26 5.70 8.61 6.48 5.71

Table 1.7. Mean Free Paths, Molecular Diameters, and Related Data for Water and Mercury Vapors

t(°C) Pt 105η 108δt 1014Ns


H2O 2.90 6.34 × 104 4.68 5.27
0 4.58 8.69
15 12.79 9.26
3.37 1.42 × 104
25 23.76 9.64

Hg 6.28 1.99 × 104 4.27 6.32


219.4 31.57 46.66
4.87 1.74 × 103 4.50 5.70
150.0 2.807 39.04
3.93 1.44 × 102 4.70 5.22
100.0 0.2729 33.56
2.66 1.45 5.11 4.42
25.0 0.0018 25.40
6.26(J)
0.0 16.2(J)
Note: In Tables 1.6 and 1.7 the following notation has been used:
Pτ = vapor pressure at t (°C) in Torr.

= mean free path in centimeters at 0°C and 1 Torr.

= mean free path in centimeters at 25°C and 1 Torr.


= mean free path in centimeters at t°C and P Torr.
ω = collision frequency (per second) at 25°C and 760 Torr.

Although Sutherland's equation (1.123) has been used most frequently to express the temperature variation of viscosity, several other relations
have been derived in the literature. The simplest of these is the exponential relation

where a and x are constants characteristic of each gas. This relation has been recommended especially for relatively small ranges of temperature.

Equations (1.123) and (1.133), as well as some other relations, have been tested by Licht and Stechert [43] for a number of typical gases and
vapors, using for this purpose data published in the LandoltBornstein tables [44].

< previous page page_34 next page >


< previous page page_35 next page >
Page 35
According to these investigators, "For twenty-four representative gases and vapors at atmospheric pressure, Sutherland's equation has been found
to fit extensive experimental data with an average error of less than 1 percent." Values of the constants C and K in Eq. (1.123) and of a and x in
Eq. (1.133), taken from the original discussion, are shown in Table 1.8. These constants were used to derive values of η25 shown in the fifth and
eighth columns of the table.

The values of C given in Table 1.6 are those deduced by Schuil [45], and it is of interest to compare them with those given in Table 1.8.

In deriving the values of L and δ shown in Tables 1.6 and 1.7, the values of η used are those given by Kennard [46] for 15°C. Values for 0°C and
25°C (η0 and η25, respectively) were derived from the values for 15°C by means of Eq. (1.133), using the values of x, taken from Kennard's
table, which are given in the second row of Table 1.6. The values of δ were derived from those of η0 and therefore apply strictly only at 0°C.

Combining Eq. (1.133) with Eq. (1.114), it follows that at room temperature, δ varies as T0.5(x0.5). Since x is greater than 0.5, it also follows
that the calculated value of δ must decrease with increase in temperature.

An interesting method of calculating values of L for air as a function of the temperature has been used by Tsien [47]. From Eqs. (1.116) and (1.67)
it follows that

L = 1.255ηγ1/2/ρu.

Thus L may be expressed in terms of the "kinematic viscosity," η/ρ, and the velocity of sound, u. By means of this relation, Tsien [47] has
calculated values of the mean free path for air for the range 0500°C. A few of the values thus deduced (in centimeters) are as follows:

t (°C): 0 20 40 100 200 300 400 500


106L(P/P0): 5.89 6.48 7.06 8.78 11.73 14.86 17.78 20.83

Table 1.8. Characteristic Constants for ViscosityTemperature Functions


Substance Temperature Sutherland Equation Exponential Equation
Range (°C)
C 106 K 105η25 106a x 105η25
Ammonia
77441 472 15.42 10.30 0.274 1.041 10.30
Argon
183827 133 19.00 22.67 2.782 0.766 21.83
Benzene
0313 403 10.33 7.58 0.299 0.974 7.71
Carbon dioxide
981052 233 15.52 15.03 1.057 0.868 14.86
Helium
258817 97.6 15.13 19.66 4.894 0.653 20.18
Hydrogen
258825 70.6 6.48 9.04 1.860 0.678 8.85
Mercury
218610 996 63.00 25.03 0.573 1.082 27.20
Methane
18499 155 9.82 11.14 1.360 0.770 10.92
Nitrogen
191825 102 13.85 17.80 3.213 0.702 17.60
Oxygen
191829 110 16.49 20.78 3.355 0.721 20.42
Water vapor
0407 659 18.31 9.84 0.170 1.116 9.82

< previous page page_35 next page >


< previous page page_36 next page >
Page 36

In the lower row, P0 denotes the standard pressure, 1 atmosphere, and P denotes any other value of the pressure in
atmospheres. The value 5.89 for 0°C is to be compared with the value 5.98 given for in Table 1.6.

In the case of H2O, for which values of L and δ for a series of temperatures are given in Table 1.7, the Sutherland
relation was used with C = 650 and η15 = 9.26 × 105 cgs unit.

In the case of Hg (see Table 1.7), the values of η used were based on that given by Braune, Basch, and Wentzel [48] for
t = 219.4°C. Values at other temperatures were derived by means of Sutherland's relation, with C = 942.2 [49]. It should
be observed that in their publication these authors used Eq. (1.118) to calculate values of δm.

The value of η for 0°C is quoted by Jeans (from Kaye and Laby, Physical Constants, 1936 edition) in his book (p. 183).

In addition, Tables 1.6 and 1.7 give values of Ns, the number of molecules per square centimeter, to form a
monomolecular layer at 0°C. On the assumption that the spacing is that of a close-packed (face-centered) lattice, we
have

Formulas for the viscosity of mixed gases are given by Kennard [50]. As he points out, the viscosity of a binary mixture
does not necessarily lie between the values for the pure components; it may be below or above both these values.

The relations for collision frequency per unit volume between molecules are of importance in many problems of
interaction between molecules. Let ZAA and ZAB denote the number of collisions between like and unlike molecules,
respectively, per cubic centimeter, per second. Then,

where nA = number of molecules per cubic centimeter of A at pressure PA (microbars), with similar definition for nB
and δAB = ½ (δA + δB).

For instance, for nitrogen (M = 28.02, δ = 3.62 × 108) at T = 298.2 K, we have

Z = 8.157 × 1016 cm3·s1 at 1 microbar


= 8.357 × 1028 cm3·s1 at 1 atmosphere.
< previous page page_36 next page >
< previous page page_37 next page >
Page 37

For the mean free paths of the molecules A and B in a mixture, the following general relations have been derived:

where vA and vB refer to the average velocity of each type of molecule.

For TA = TB, Eq. (1.138) becomes

and similarly for 1/LB.

For and we obtain

For we obtain

For and TA not identical with TB, we obtain

The last equation has been used by Gaede [51] to calculate the mean free path of nitrogen (A) in the blast of a mercury-
vapor pump. In this case we may assume the following: TA = 300 K; and the temperature of the mercury vapor, TB, is
400 K. At this temperature the pressure of mercury vapor is about 1 Torr. Hence, nB = 2.414 × 1016.

From the data in Tables 1.6 and 1.7, we have

108 · δAB = ½(3.78 + 4.70) = 4.24.

Substituting these values and those for MA and MB in Eq. (1.143), the result is LA = 6.73 × 103 cm, whereas from the
data for in Table 1.7 the value derived for the mean free path of mercury molecules in the saturated vapor at 400 K
is L = 4.4 × 103 cm.

1.7.1
Viscosity at Low Pressures
As mentioned previously, the coefficient of viscosity is not independent of pressure when the pressure decreases to a
low value. Under those conditions it was observed by Kundt and Warburg (1875) that the damping of a vibrating
surface by the surrounding gas is decreased, as if the gas were slipping over the surface. If one surface is at rest and
another surface, at a distance d, is moving parallel to the first surface with

< previous page page_37 next page >


< previous page page_38 next page >
Page 38

a uniform velocity, u, the viscous drag upon each surface at normal pressures is given in accordance with the definition
of η by the relation

At low pressures, however, the observed value of the tangential force B is less than that given by Eq. (1.144),
corresponding to an increase in the value of d. That is, the equation assumes the form

where ζ is known as the coefficient of slip.

Considerations based on the kinetic theory of gases lead to the relation

which, combined with Eq. (1.116), leads to the relation

where f is a numerical coefficient which has a maximum value of 1. It was introduced by Maxwell with an interpretation
given by Kennard [52] as follows:

The value of f, the transfer ratio for momentum, will presumably depend upon the character of the interaction
between the gas molecules and the surface; it may vary with the temperature. We can imagine a surface that is
absolutely smooth and reflects the molecules ''specularly" with no change in their tangential velocities; in such
a case f = 0 and ζ = ∞, viscosity being unable to get a grip upon the wall at all. On the other hand, we can
imagine the molecules to be reflected without regard to their directions of incidence and therefore with
complete loss of their initial average tangential velocity.

In this case, f = 1 and ζ = L [53].

More generally, ζ = βL, where β is of the order of unity; and at very low pressures, where Eq. (1.145) becomes

Thus, at very low pressures the rate of transference of momentum from a moving surface to another surface adjacent
and parallel is directly proportional to the pressure and to the velocity of the moving surface. This conclusion was
applied by Langmuir [54] to the design of a molecular gauge for the measurement of low pressures.

The quantity ρva/4, which, as noted previously, corresponds to the mass of gas incident on unit area per unit time, has
been designated by Kennard as the free-molecule viscosity of the gas between the plates.
< previous page page_38 next page >
< previous page page_39 next page >
Page 39

The concept of coefficient of slip has also been used to interpret observations on the flow of gases at very low pressures
through capillaries.

1.7.2
Molecular Diameters

Mention has been made in the previous section of the two relations for deducing values of δ and of δm from viscosity
measurements. These relations are as follows:

As illustrated in Table 1.7 by the values of δ for mercury, these values exhibit a considerable decrease with increase in
temperature, and in this case it is found that δm = 2.50 × 108. The variation with T is obviously greater for those
molecules for which the Sutherland constant C, has a large value.

In spite of the fact that the values of δ thus deduced exhibit a variation with T, Chapman and Cowling [1d] have been
followed in this discussion in choosing Eq. (1.150) rather than Eq. (1.151). While, as pointed out by Chapman and
Cowling, such a variation in the value of δ with T "receives a simple explanation on the hypothesis that the molecules
are centers of repulsive forces, not hard spheres," it is of interest to compare the results obtained by means of Eq.
(1.150) with those obtained by means of Eq. (1.151) and also by other methods.

There are several such methods, and only a few of the more important ones can be mentioned briefly, together with
some of the results deduced.

1.7.3
Application of the van der Waals Equation

Near the critical temperature and pressure the behavior of gases can be described very satisfactorily by a modified form
of Eq. (1.14), deduced by van der Waals, which is as follows:

In this equation, V is the volume per mole, the term a/V2 is a correction term which takes into account the attractive
forces between the molecules, and the constant b is a measure of the actual volume of the total number of molecules in
accordance with the relation

< previous page page_39 next page >


< previous page page_40 next page >
Page 40

The values [55] of the constant b may be determined for any given gas from the values of the critical temperature (Tc) and
critical pressure (Pc) by means of the relation

1.7.4
From the Density of the Solid or Liquid

Assuming that the molecules are closely packed, as in a face-centered cubic lattice, the projected area per molecule is given
by the relation [56]

where m = mass of molecule = M/NA and ρ = density of condensed phase.

But, according to Eq. (1.134),

Equation (1.156) has been used to calculate the number of molecules per unit area required to form a unimolecular layer (or
monolayer). (See last column of Table 1.9.)

Table 1.9. Values of Molecular Diameter (cm·108)


Gas From η From From Electron Collision 1014·Ns
b ρ from ρ
δ0 δm
H2 2.75 2.10 2.76 4.19 2.2
6.58
He 2.18 1.69 2.66 4.21 1.7
6.49
Ne 2.60 2.16 2.38 3.40 2.2
9.98
Ar 3.67 2.42 2.94 4.15 3.6
6.71
O2 3.64 2.50 2.93 3.73 3.4
8.30
Hg 6.26 2.50 2.38 3.26
10.86
CO2 4.65 3.32 3.24 4.05 4.4
7.04
H2O 4.68 2.45 2.89 3.48 3.8
9.53
C6H6 7.65 4.71 4.51 5.89
3.33
CH4 4.19 3.31 3.24 4.49
5.73
C2H6 5.37 3.87 3.70 5.01
4.60
C3H8 6.32 4.45 4.06 5.61
3.67
n-C4H10 7.06 4.84 4.60 6.10
3.18
n-C5H12 7.82 5.05 4.89 6.45
2.78
n-C6H14 8.42 5.22 5.16 6.74
2.54

< previous page page_40 next page >


< previous page page_41 next page >
Page 41

1.7.5
Cross Section for Collision with Electrons [57]

A cathode-ray beam of initial intensity I0 is decreased to intensity I, after passing through a layer of the gas of thickness
x, in accordance with the relation

where, as shown previously in Eq. (1.104), α is a measure of 1/Le, where Le = mean free path for electrons = 4(2)½L.
Hence it follows from Eq. (1.113) that

Thus the collision cross section is given by α/n. However, as has been observed experimentally, the value of α varies in
a rather complex manner with the potential used to accelerate the electrons, with the result that it is actually impossible
to assign a definite value to δ as derived from electron-collision measurements.

The structure of molecules has also been determined from measurements of dipole moments and from electron-
diffraction experiments, all of which are discussed at length by Stuart [57] in his book.

Table 1.9 gives, for comparison, values of δ for a number of gases and vapors as deduced by at least four different
relations. The second and third columns give values of δ0 and δm as deduced by means of Eqs. (1.126) and (1.120),
respectively, from the values of η (extrapolated to 0°C) and the values of C given by Schuil [45]. The fourth column
gives values of δ calculated by means of Eq. (1.154) from the values of the van der Waals constant, b; the fifth column
gives values deduced by means of Eq. (1.158) from values of ρ at extremely low temperatures (in general for the solid
state). The sixth column gives values, derived from observed values of α by means of Eq. (1.160), for 36-volt electrons
[57], and the last column gives values of Ns calculated by means of Eq. (1.157) from the values of δ in the fifth column
[58]. The values thus derived are to be compared with those deduced in Tables 1.6 and 1.7 from kinetic-theory values of
δ.

1.8
Heat Conductivity of Gases

The kinetic theory of gases achieved a great triumph when it led to the conclusion that the viscosity is independent of
the pressure. It led to still further important results when it predicted the existence of simple relations between the
properties of viscosity, heat conductivity, and diffusivity.

From the kinetic point of view it is the same whether the molecules transfer momentum or translational energy from one
layer to another. The equations are quite analogous.

As in the case of viscosity, we consider any two layers CE, EH (Fig. 1.4), each of thickness L, between two plates
whose temperatures are T1 and T2 and distance apart d. Let cv denote the heat capacity per unit mass at constant
volume. The relative temperature drop between the planes CD and HK is equal to

2(T1 T2)L/d.

< previous page page_41 next page >


< previous page page_42 next page >
Page 42

Hence the heat transferred per unit area is

Therefore the coefficient of heat conductivity

If cv is expressed in calories per gram, the unit of λ is 1 cal · cm1 · s1 · deg1 · Comparing the last equation with Eq.
(1.110), it follows that

As in the case of the relation for η, a more careful consideration of the mechanism of energy transfer by means of the
molecules leads to the relation

where, according to Eucken [59, 60], we have

and γ = ratio of specific heat at constant pressure to that at constant volume.

For monatomic gases, γ = 5; and for polyatomic gases, γ tends to approach the value 1, with increase in total number of
atoms per molecule.

Hence

1 ≤ε≤ 2.5.

Table 1.10 gives data published by Kannuluik and Martin [61, 62] on values of λ0 (conductivity at 0°C) and of ε, as
derived from observation by means of Eq. (1.164) and as calculated by means of Eq. (1.165). Similar data for these
gases and a number of others are given in the treatise by Chapman and Cowling [1d].

One important conclusion that follows from Eq. (1.164) is that the thermal conductivity of a gas is independent of
pressure, which is valid as long as the pressure is higher than the range in which molecular flow occurs. (See the
following section).

With regard to the variation in λ with T the following remarks may be made. To a first approximation the variation in
value of λ follows that in the value of η, since ε exhibits only a slight variation with T. However, for larger ranges of T,
account must be taken of the increase with T in the value of cv. Denoting the molecular specific heat, at constant
volume, by Cv, Partington and Shilling [63] give the relations shown in Table 1.11 for different gases:

Since the variation in η with T is given by the Sutherland relation, Eq. (1.123), it follows that, in terms of λ0 in calories
per centimeter per second per degree,
< previous page page_42 next page >
< previous page page_43 next page >
Page 43

Table 1.10. Values of Heat Conductivity Compared with Coefficients of Viscositya


Gas cv (cal·g1) ε (obs) ε (calc)
105λ0 105η0
He
34.3 18.76 0.746 2.45 2.44
Ne
11.12 29.81 0.150 2.50 2.44
Ar
3.82 21.02 0.0745 2.44 2.44

H2
41.3 8.50 2.43 2.00 1.90
Air
5.76 17.22 0.171 1.96 1.91
O2
5.83 19.31 0.157 1.92 1.95
CO
5.37 16.65 0.178 1.81 1.91

CO2
3.43 13.74 0.153 1.64 1.72
N2O
3.61 13.66 0.155 1.71 1.73
a To convert values of λ0 from cal · cm1 · s1 · deg1 to watts·cm1 · deg1 multiply values in Table 1.10 by 4.186.

Table 1.11. Molecular Specific Heat and Specific Heat Ratio (Partington and Shilling)
Gas γ = Cp/Cv Cv
Air
1.4034 4.924 + 1.7 × 104T + 3.1 × 107T2
N2
1.405 4.924 + 1.7 × 104T + 3.1 × 107T2
O2
1.396 4.924 + 1.7 × 104T + 3.1 × 107T2
CO
1.404 4.924 + 1.7 × 104T + 3.1 × 107T2
H2
1.408 4.659 + 7.0 × 104T
CO2
1.302 5.547 + 4.5 × 103T 1.02 × 106T2
H2O
... 6.901 1.19 × 103T + 2.34 × 106T2
Hg, Ar, etc.
1.667 2.990

where α and β are determined from the expression for Cv as a function of T, and lT is expressed in watts per centimeter per degree. For
hydrogen,
< previous page page_43 next page >
< previous page page_44 next page >
Page 44

Similar expressions for gases for which λ has not been determined can be derived, as is evident, from determinations of
η, using the observed values of C and values of ε calculated from those of γ by means of Eq. (1.165).

It will be observed that the heat conductivities of hydrogen and helium are much greater than those of heavier gases,
such as oxygen and carbon dioxide.

For the case of a wire of radius a and length l, suspended along the axis of a cylinder of radius r, the energy loss per unit
time due to thermal conduction by the gas is [64]

where T T0 is the difference in temperature and r/a is not "excessively" large [65] while lm is the average conductivity
over the temperature range T T0.

The units in which E is usually expressed are watts per square centimeter.

The energy loss per unit area per unit time is

and the energy loss per unit length of wire per unit time is

Since the total energy loss from a heated wire is the sum of that lost by radiation (which varies as ) and that
lost by conduction of the gas, the former has to be subtracted from the total energy loss in order to obtain the amount
due to the latter. Furthermore, in the case of short wires especially, a correction has to be made for the loss by
conduction at the ends.

For wires of low emissivity, such as platinum, operating at a temperature below about 500°C, the loss due to radiation is
negligible compared to that due to conduction.

Since the thermal-conduction loss in the case of mixtures varies with both the nature of the gas and the composition, this
fact has been applied to the analysis of gases [66]. An instrument devised for this purpose by Shakespear [67], known as
a katharometer, consists of a platinum spiral filament in a copper block. This instrument has been used extensively by
English investigators for determinations of rates in gaseous reactions and for experiments on thermal transpiration
(which is discussed in Section 1.10) [68].

1.9
Thermal Conductivity at Low Pressures

As mentioned in the previous section, the heat conductivity of gases should theoretically be independent of pressure.
That this is at least approximately confirmed
< previous page page_44 next page >
< previous page page_45 next page >
Page 45
by observation is illustrated by the data shown in Table 1.12, obtained by Dickins [69].

The values under W represent the total heat loss (in calories) by conduction, from a platinum wire (a = 3.765 × 103 cm, l = 20.09
cm) suspended along the axis of a Pyrex glass tube (inside radius, r = 0.3346 cm). The tube was maintained at about 0°C by
external cooling, and ∆t is the temperature differential between the wire and the wall. The pressure of the gas in Torr is given in
the third column, and the value indicated by W∞ represents the heat loss extrapolated for P = ∞. The values under λt give the
thermal heat conductivity at the mean temperature indicated in the last column. The values of λt (in calories) were derived by
means of the relation, deduced from Eq. (1.167),

It will be observed that over a range of pressures the values of W did not exhibit any considerable decrease.

That the heat conductivity is practically constant over a large range of pressures is also shown by the plots of the energy loss (in
watts) at constant temperature (about 99°C) from a platinum filament (Fig. 1.5). The filament was a 14-cm length of 3-mil wire
located along the axis of a glass tube 2.54 cm in diameter. The wall temperature was 0°C.

As will be observed, the heat loss in hydrogen was about 10 times that in argon. The loss at a pressure of about 1 m Torr was
0.006 W. A comparison of the relative losses in the three gases at 760 Torr yields values which agree, within a few percent, with
the values of λ given in Table 1.10.

Table 1.12. Variation in Thermal Conduction with Pressure (Dickins)


Gas ∆t Pτ W W∞ 105λt t (°C)
H2
17.866 625.0 0.21323 0.21494 42.77 9.04
442.0 0.21258
302.3 0.21145
229.9 0.21038
168.2 0.20875
129.8 0.20692

Air
23.832 91.7 0.04021 0.04052 6.044 11.94
52.1 0.03997
31.3 0.03961
22.3 0.03926
17.1 0.03891
11.9 0.03825

CO2
23.80 83.3 0.02461 0.02473 3.694 11.91
40.5 0.02447
19.2 0.02417
11.1 0.02379
< previous page page_45 next page >
< previous page page_46 next page >
Page 46

Thus, over the range 100760 Torr, the heat loss in hydrogen increased only about 7%, whereas in the range below 50
Torr the decrease was nearly 100%. Careful measurements show that at very low pressures the thermal conductivity
decreases linearly with the pressure.

The theory of heat conduction at these pressures has been developed from two different points of view. The first of
these, due to Knudsen [70], involves a consideration of the mechanism of energy transfer by individual molecules
incident on the hot surface.

The second point of veiw, due to Smoluchowski [71], is based upon the concept of a temperature discontinuity which is
the thermal analogue of the phenomenon of "slip" discussed in Section 1.7.

1.9.1
Free-Molecule Conductivity (Knudsen)

When molecules originally at a temperature Ti strike a hot surface at temperature Ts (> Ti), complete interchange of
energy does not occur at the first collision. In fact
Fig. 1.5
Plots illustrating the variation in thermal conductivity
with pressure, for nitrogen, argon, and hydrogen. Ordinates
give values of total watts conducted from a platinum
filament located along the axis of a cylindrical glass
tube. Scale of watts for hydrogen should be multiplied
by 10. Abscissas give pressures in centimeters
of mercury (10 Torr).

< previous page page_46 next page >


< previous page page_47 next page >
Page 47

it may often require many collisions for this to occur. Therefore Knudsen introduced a constant, known as the accommodation coefficient,
designated by α, which "can be defined as standing for the fractional extent to which those molecules that fall on the surface and are reflected or
re-emitted from it, have their mean energy adjusted or 'accommodated' towards what it would be if the returning molecules were issuing as a
stream out of a mass of gas at the temperature of the wall [72].

The molecules re-emitted or reflected from the hot surface consequently possess a mean energy which corresponds to a temperature lower than
Ts, which we shall designate as Tr, and the accommodation coefficient is defined by the relation

For α = 1, Tr = Ts; for α < 1, Ts > Tr > Ti. It should be noted that the temperature Tr is not clearly defined, as Blodgett and Langmuir [73]
have pointed out, unless the molecules leaving the surface have a Maxwellian distribution of velocities.

Because the faster molecules emitted from a surface carry more energy than the average, the Maxwell distribution function must be replaced by

= 2(m/2kT)2v3 exp( mv2/2kT)

so that the mean translational energy transferred from a surface at temperature T is given by

(instead of 3/2kT, which is the average energy of the molecules in a volume).

Let us now consider heat transfer in a monatomic gas at low pressure:

E0 = energy transfer from hot to cold surface (at temperature Ti) per square centimeter of hot surface per second

where vi = average velocity at Ti.

Thus the rate of energy transfer at low pressure is proportional to the pressure and the temperature difference.

For diatomic and polyatomic gases, the molecules striking the hot surface acquire not only increased translational energy but also increased
amounts of both rotational and vibrational energy. The amount of the vibrational energy possessed by molecules as compared with the amount
of translational energy is measured by the value of γ. In these cases, a detailed calculation leads to the relation

which for γ = 5/3 (case of monatomic gases) becomes identical with Eq. (1.173).

< previous page page_47 next page >


< previous page page_48 next page >
Page 48

In this equation, α has the value [74]

where α1 and α2 are the values of the accommodation coefficient for the two surfaces.

Substituting for vi from the relation in terms of Ti and M, Eq. (1.174) assumes the form

where Λ0 = free-molecule heat conductivity at 0°C

(In terms of calories per second, Λ0 should be multiplied by 0.2389.)

Table 1.13 gives values of Λ0 in watts per square centimeter per degree per mTorr for a number of gases and vapours
[75], and, for comparison, values of 4.186 λ0 (derived from the values in Table 1.10 and from other sources) which
correspond to the conductivity at 0°C in terms of watts per centimeter per degree.

In terms of calories (15°C) per mole, we have

R0 = 8.3145 × 107/4.1855 × 107 = 1.9865 cal · deg1 · mol1·


Table 1.13. Values of Molecular Heat Conductivity
Gas M γ 106Λ0 4.186 × 104λ0
H2
2.016 1.41 60.72 17.30
He
4.003 1.67 29.35 14.36
H2O
18.016 1.30 26.49
Ne
20.18 1.67 13.07 4.66
N2
28.02 1.40 16.63 2.38
O2
32.00 1.40 15.57 2.44
Ar
39.94 1.67 9.29 1.60
CO2
44.01 1.30 16.96 1.44
Hg
200.6 1.67 4.15

< previous page page_48 next page >


< previous page page_49 next page >
Page 49

Since γ + 1 = (2Cv + R0)/Cv and γ 1 = R0/Cv, where Cv = molar specific heat in calories (at constant volume) and R0
has the above value, in all of the above expressions we can set

Hence Eq. (1.177) can be expressed in the form

For coaxial cylinders of radii r1 and r2 (r2 > r1), the rate of energy transfer per unit area from the inner cylinder or
wire, at temperature Ts and at low pressures, is given by the relation

and when α1 = α2 = α this becomes [76]

Jeans [77] has pointed out that, according to general dynamical considerations, the constant α should be determined by
a relation of the form

where m′ = mass for molecules striking the surface for which the molecular mass is m.

Morrison and Tuzi [78] have used a molecular beam method with mass spectrometer detector to measure the thermal
accommodation coefficient for translational energy of a number of vapors on Pyrex glass.

For completely roughened surfaces, α = 1. Values of α which have been observed for different surfaces are given in
Table 1.14 [79].

Keesom and Schmidt [80] have measured the thermal accommodation coefficients for He, Ne, H2, and N2 on
Thuringian glass at 0°C at pressures from 0.04 to 0.2 Torr. Average values were 0.335 for He, 0.286 for H2, 0.853 for
N2, and 0.670 for Ne.
Klett and Irey [81] have measured the thermal accommodation coefficients for air, nitrogen, and helium on commercial
copper surfaces at pressures from 2 × 104 Torr to 4 × 103 Torr, obtaining 0.799 for N2 at 77.4 K, 0.760 for N2 at 243
K, 0.823 for air at 77.4 K, 0.698 for air at 243 K, 0.564 for He at 77.4 K, and 0.407 for He at 243 K.

< previous page page_49 next page >


< previous page page_50 next page >
Page 50

Table 1.14. Values of the Accommodation Coefficient for Several Gases


Surface H2 O2 CO2
Polished Pt
0.358 0.835 0.868
Pt slightly coated with black
0.556 0.927 0.945
Pt heavily coated with black
0.712 0.956 0.975

H2 N2 Ar Hg Air
Tungsten
0.20 0.57 0.85 0.95
Ordinary Pt
0.36 0.89 0.89 0.90

1.9.2
Temperature Discontinuity (Smoluchowski)

In Section 1.7, mention was made of the phenomenon of ''slip" which occurs in a gas at moderately low pressures. In
Eq. (1.145) we introduced a coefficient of slip ζ = βL to account for the apparent decrease in viscosity at low pressures.

An analogous phenomenon was observed by Smoluchowski in the case of thermal conduction at low pressures. For two
parallel plane surfaces separated by a distance d, the heat loss per unit area per unit time may be represented at these
pressures by a relation of the form

where λm is the mean conductivity over the range T1 T0.

In this equation, g is a coefficient defined by the relation

where ∆T represents the temperature discontinuity at any one surface, and dT/dx is the temperature gradient normal to
the surface.

As shown by Kennard,

where the numerical constant ε has a value, according to Eq. (1.165), which varies between 1 and 2.5, and β′ is a
numerical constant of the order of unity.
Hence Eq. (1.183) may be expressed in the form

where α is assumed to be the same for both surfaces, L = c/P, and the symbol E0P indicates that the conductivity per
unit area is a function of P.

< previous page page_50 next page >


< previous page page_51 next page >
Page 51

A comprehensive treatment of the heat transport between parallel plates in a gas of rigid sphere molecules for the whole
range of rarefaction conditions from a dense Clausius gas to the extreme rarefaction of a gas with a Knudsen number
has been published by Frankowski, Alterman, and Pekeris [82]. Willis [83], using a relaxation-type model for
intermolecular collisions, has derived equations based on the Krook [84] kinetic model for the heat transfer in a rarefied
gas between parallel plates at large temperature ratios. He has also considered heat transfer and shear between coaxial
cylinders for large Knudsen numbers [85].

The problem of heat transfer between concentric spheres in the near free-molecule region has been treated by Havekotte
and Springer [86] and other authors cited in their article.

For a hot wire of radius a situated along the axis of a cylindrical tube of radius r, Eq. (1.186) must be replaced by the
relation

where it is assumed that α is the same for both the wire and inside surface of the cylinder.

For very low values of P, Eq. (1.186) becomes

and Eq. (1.187) becomes

That is, at very low pressures E0P is independent of d and varies linearly with P. This conclusion is identical with that
deduced above by Knudsen for the case of free-molecule flow. That Eq. (1.188) is identical with Eq. (1.176) may be
shown as follows.

Substituting for β′ from Eq. (1.185), substituting for λm from Eq. (1.164), and utilizing the relation η = 0.5ρvaL, we
obtain the result
where α/(2 α) takes the place of α used in Eq. (1.176).

< previous page page_51 next page >


< previous page page_52 next page >
Page 52

Equations (1.186) and (1.187) can be expressed in a form which is very convenient for calculation. We shall consider
especially Eq. (1.187), which is more important in practical measurements of heat conductivity. This equation may
obviously be written in the form

That is, if 1/E0P is plotted against 1/P, a straight line is obtained. From the intercept for 1/P = 0, which we shall
designate by 1/E0 ∞, the value of λm may be obtained, while the slope is given by the relation

where X is a constant determined by the relation

Equation (1.190) thus applies to the transition range of pressures in which E0 changes from being independent of
pressure (at higher pressures) to varying linearly with the pressure (at very low pressures).

In a very interesting paper published in 1912 [87, also see 88], Langmuir formulated a theory of the mechanism of heat
conduction from wires (as well as plane surfaces). Observations on the heat losses in hydrogen from tungsten wires
heated to incandescence showed that whether the wire was in a horizontal or vertical position made a difference of only
a few percent (at constant temperature) in the heat loss. "This," states Langmuir, "was strong indication that the heat
loss was dependent practically only on heat conduction very close to the filament and that the convection currents had
practically no effect except to carry the heat away after it passed out through the film of adhering gas."

In the case of a wire of radius a, Langmuir assumed that the energy loss will occur through a "stationary" film of
adherent gas, the outer edge of which is at a distance b from the axis of the wire. As shown by Langmuir, the value of b
then satisfies the relation

and the energy loss per centimeter length of wire is given by the relation

< previous page page_52 next page >


< previous page page_53 next page >
Page 53

in which λ(T) is given by Eq. (1.166).

Introducing into Eq. (1.195) the relation expressed by Eq. (1.192), it follows that

From measurements of the energy losses from platinum wires in air [89], Langmuir derived the conclusions that, for air
at 20°C and 760 Torr, B = 0.043 cm and is "surprisingly independent" of T. He also concluded that B should be
proportional to the viscosity and inversely proportional to the density, "for it is the viscosity that causes the existence of
the film and it is the difference of density between hot and cold gas (proportional to the density itself) that keeps the
film from becoming indefinitely large."

The problem of heat conduction at low pressures from a wire of radius a, suspended coaxially in a glass tube of radius r,
where r is very much greater than a, has been discussed from the point of view of Langmuir's film theory by Jones [88].
Brody and Körösy [90] have shown that although 90% of the wattage lost can be accounted for by assuming pure
conduction, the film is not "stationary," but convection currents exist with the velocity of convection increasing as the
surface of the filament is approached up to the limit (0.25 mm) of their measurements for incandescent tungsten
filaments in nitrogen.

1.10
Thermal Transpiration (Thermomolecular Flow)

From Eq. (1.82),

it follows that the rate of efflux of a given gas through a small opening varies as ρ(T)½. If we have two chambers A and
B, separated by a small orifice, at two different temperatures, TA and TB, transpiration will occur until a steady state is
established at which

the limiting ratio of pressure is

It was expected that the same result would be obtained when the chambers are separated by a wall containing a porous
plug or by a single long tube of diameter 2a
< previous page page_53 next page >
< previous page page_54 next page >
Page 54

such that the Knudsen number because under this "molecular flow" condition the probability of
passage of a molecule through a long tube is the same in both directions when the molecules enter the ends with a
cosine law distribution. Also, when the effect should be independent of the wall temperature of the
connecting tube. However, Hobson et al. [91] found that, while Eq. (1.199) is obeyed for apertures, this lower limit is
not necessarily reached when the two chambers are separated by a long tube in which the pressure is so low that
. Their data showed that the measured lower limit was aRm with a ranging from 1.1 to 1.3, with lower values
of a as the tube diameter is reduced and the molecular weight M increased. Miller and Buice [92] attempted to explain
the need for the correction factor a based on a model in which the faster moving molecules have a higher probability of
being specularly reflected. They applied a Monte Carlo method to calculate the flow through the tube with a special law
of scattering at the wall, but their results were not in good agreement with the experimental data of Hobson and
coworkers.

Thermal transpiration is of importance in the application of vacuum gauges and vacuum microbalances at low pressures.
For instance, if the chamber A is a part of a system at liquidair temperature (TA = 90) and the pressure is measured by
means of a gauge at room temperature (TB = 300), then the real value of PA is given by the relation

PA = PB(90/300)1/2 = 0.55PB.

Because the envelope of a tubulated ionization gauge is usually at a temperature much higher than room temperature, if
the gauge is not used under the same temperature conditions as those during its calibration, a correction based on Eq.
(1.199) may be required [93]. Poulis et al. [94] have shown that longitudinal Knudsen forces associated with the thermal
transpiration effect can give spurious mass variations in vacuum microbalances when temperature gradients exist along
the balance case.

When the mean free path L is very much smaller than 2a, so that collision between molecules become predominant over
collisions against the walls, then the condition of equilibrium is

There exists therefore a range of pressures in which the pressure ratio PA/PB changes from that given by Eq. (1.200) to
that given by Eq. (1.199) in the case of an aperture, and to a(TA/TB)1/2 in the case of narrow tubes, with a being an
experimentally determined factor between 1.0 and 1.3.

The theory of the phenomenon was first discussed by Maxwell (1879) and subsequently by Knudsen [95], who derived
equations for the pressures gradient along the connecting tube as a function of the tube diameter and pressure in the hot
chamber. For the pressures and temperatures in the two vessels, Knudsen proposed

< previous page page_54 next page >


< previous page page_55 next page >
Page 55

the semiempirical relation

which for reduces to Eq. (1.199), and for reduces to the equilibrium condition PA = PB. For a temperature
gradient along the connecting tube results in a boundary layer flow of gas (thermal creep) along the wall of the tube from the cold to the
hot side balanced by a return flow through the interior from the hot side to the cold side [96], and complicated theoretical equations for
the thermal transpiration effect in this case have been proposed by various authors.

Bennett and Tompkins [97] have recommended that an equation proposed by Liang [98, 99] be used. The equation is supported by a
kinetic argument, and Bennett and Tompkins found it to fit experimental data over a wide range. It should be noted, however, that these
workers used glass systems. Los and Fergusson [100] observed that the nature of the wall affects the magnitude of the correction, as
Liang had predicted [99].

The form of Liang's equation given by Bennett and Tompkins can be expressed as follows:

in which

R = the ratio of the pressures in the two regions at different temperatures, TA and TB,
b = 1 for connecting tubes of internal diameter, d < 1 cm
= 1.22 for connecting tubes of internal diameter, d > 1 cm,
X = PBd,
αHe = 3.70 (1.70 2.6 × 103∆T)2 (∆T = TB TA),
βHe = 7.88[1 (TA/TB)½] for (TA/TB)½≤ 1,

and values of φg are given in Table 1.15.

Bennett and Tompkins also point out that φg values may be calculated from the collision diameters, r0, given by Hirschfelder, Bird, and
Spotz [101], by making use of

Table 1.15. Experimental φg Values for Glass Systems (Bennett and Tompkins)
Gas: He Ne Ar Kr Xe H2 O2 N2 CO CO2 C2H4
φg: 1.00 1.30 2.70 3.90 6.41 1.44 2.87 3.53 3.31 4.52 6.72

< previous page page_55 next page >


< previous page page_56 next page >
Page 56

the equation

Arney and Bailey [102] studied the thermal transpiration effect for air, argon, and helium for Knudsen numbers from
0.01 up to 100 and for temperature ratios from 1.5 to 3.8 by an unusual technique involving two tubes, one with large
diameter and the other with small diameter, joined together by an annular junction at which the temperature could be
raised by an electrical resistance heater while the other ends of the tubes were maintained at a fixed temperatures by
water cooling. The pressure at the cold ends of the two tubes was measured for a series of temperature ratios, and the
pressure at the junction was assumed to be that of the cold end of the larger tube as long as the Knudsen number Kn was
≤ 0.01 for the larger tube. Then when Kn reached higher values in the larger tube, the pressure ratio between the cold
end and the junction could be estimated from the experimentally determined variation of the ratio of pressure at the cold
end of the smaller tube to that at the junction with the Knudsen number for the smaller tube which was determined when
Kn for the larger tube was still less then 0.01. Comparison of their results with the predictions of Knudsen's equation
[103] for dp/dT showed agreement at the upper and lower ends of the range of Kn but somewhat lower values of the
ratio of the pressure at the cold end to that at the hot end for Kn in the middle range from 0.1 to 10. They present a chart
of curves, based on their measurements, of pressure ratios versus temperature ratios for a series of Knudsen numbers
from 0.03 to 10.

Hobson [104] compared the experimental data obtained by Edmonds and Hobson [105] for helium and neon effusing
through an aperture with 2a = 20 mm, TA = 77.4 K, and TB = 295 K with the values predicted by Eq. (1.201) and by Eq.
(1.202). For the mean free path in Eq. (1.201), Hobson used the relation of Weber and Schmidt [106]:

where (PL)0 = 0.1339 Torr · mm and n = 0.147 for helium while (PL)0 = 0.08841 Torr · mm and n = 0.20 for neon, and
T was chosen as 295 K. An appropriate choice of the constants αHe, βHe, and φg in Liang's equation gave a curve of
pressure ratio R versus PB in very good agreement with the experimental data, while the curve based on the simple Eq.
(1.201) agreed with the experimental data at the lower limit 0.512 and upper limit 1.0 but was slightly higher at values
of PB in the middle range.

Podgurski and Davis [107] obtained results that agreed with Bennett and Tompkins' equation [Eq. (1.202)] for argon
and xenon but deviated appreciably from this equation for neon and hydrogen at low pressures.

Sharipov [108] has made extensivse calculations based on a modified Boltzmann equation of the exponent γ in the
equation

as a function of the Knudsen number (which he defines as Kn = L/a) and other factors, for long capillary tubes of radius
a under conditions of a balance between the

< previous page page_56 next page >


< previous page page_57 next page >
Page 57

flow due to the pressure gradient and the flow due to the temperature gradient. He tried to compare his calculations with
the experimental results of Edmonds and Hobson as recalculated using a rarefaction factor proportional to the inverse
Knudsen number at PB and TB but found a disagreement for different capillaries.

Dietz [109] studied the thermal transpiration effect in the mass spectrometer inlet system for argon in the leak tube and
found that sensitivity response was not affected if gas mixtures and calibrating gases are run at the same pressure.

Variable capacitance diaphragm transducers used as transfer standards in an intercomparison of pressure standards in
the low vacuum region of 1 to 10 Torr are operated at a stabilized elevated temperature to reduce inherent sensitivity to
ambient temperature changes, and thermal transpiration causes a pressure differential between the transducer and the
vacuum system which must be determined in order to achieve maximum accuracy of calibration comparisons. Jitschin
and Röhl [110] have studied this problem quantitatively and applied correction procedures based on Eq. (1.202) above
using both the Liang expression for f(X) and the more recent formula of Takaishi and Sensui [111]

where the parameters A*, B*, and C* depend on the gas species and are determined experimentally. The correction
according to the Liang formula was in poor agreement with their experimental data, but the correction using the formula
of Takaishi and Sensui reproduced the experimental data within about 4%. The degree of agreement seemed to depend
on the material used for the connecting tube, and by adopting an effective temperature (TA)eff and an effective tube
diameter they were able to obtain agreement to better than 0.1%.

Williams [112] derived equations for thermal transpiration in the near-continuum regime starting with the NavierStokes
equations and using the slip boundary condition at the wall including thermal creep. Comparison of his equations with
the experimental data of various authors showed good agreement for low Knudsen numbers but gave too small values as
the Knudsen number increased.

1.11
Thermal Diffusion

The thermal-diffusion effect has been described by Ibbs [113] as follows:

If a temperature gradient is applied to a mixture of two gases of uniform concentration, there is a tendency for the
heavier and large molecules (mass m1, diameter δ1) to move to the cold side, and for the lighter and smaller molecules
(mass m2 and diameter δ2) to move to the hot side. The separating effect of thermal diffusion (coefficient DT) is
ultimately balanced by the mixing effect of ordinary diffusion (coefficient D12) so that finally a steady state is reached
and a concentration gradient is associated with the temperature gradient. The amount of thermal separation thus
produced by a given difference in temperature depends upon the ratios m1/m2 and δ1/δ2 and upon the proportion by
volume of the heavier gas f1 and of the lighter gas

< previous page page_57 next page >


< previous page page_58 next page >
Page 58

f2 (where f1 + f2 = 1) and also upon the nature of the field of force operating between the unlike molecules.

The effect was predicted by Enskog (1917) and, independently, by Chapman and Dootson [114]. Following the earlier
experimental investigations of Ibbs and others, Clusius and Dickel [115] devised a continuous method for separating
mixtures of gases and isotopes which has been applied extensively by subsequent investigators.

In this section the discussion will be limited largely to the experimental data, since the mathematical theory is quite
complex and beyond the scope of the objectives of this chapter.

The coefficient of thermal separation is defined by the relation

where DT and D12 have been defined above. Now

If kT is a constant, then we obtain from Eq. (1.207) the relation

for the amount of separation, where T1 and T2 denote the hot and cold temperatures, respectively.

Figures 1.6 and 1.7 illustrate results obtained by Ibbs [116] with mixtures of H2 and N2. The concentrations of H2 in
the different mixtures varied from 4.7% for No. 2
Fig. 1.6
Relation between degree of separation by thermal diffusion
and log(T1/T2) for mixtures of H2 and N2.

< previous page page_58 next page >


< previous page page_59 next page >
Page 59

Fig. 1.7
Relation between composition of H2N2 mixtures
and degree of separation for log(T1/T2) = 0.2.

to 50.5% for No. 7. It will be observed that these observations are in accord with Eq. (1.208), since ∆f is found to be a
linear function of log(T1/T2). Figure 1.7 shows the variation in value of ∆f with composition for a constant value of the
ratio T1/T2.

Chapman has shown that kT can be expressed in the form [117]

where the coefficients A to D are functions of the molecular weights and of the ratio δ1/δ2. These can be calculated
from the force constants s12 in the repulsive force relation F = c/rv, where v = s12 for unlike molecules and r = 1/2(δ1 +
δ2).

The separation in a mixture of H2 and CO2 was investigated by Ibbs and Wakeman [118] over a range of 600°C. For
any given mixture it was observed that the amount of separation was in accordance with Eq. (1.208). For a constant
difference in temperature, ∆f increased from 0 for pure H2 to a maximum value of 0.0365 (for ∆T = 600°C), for a
mixture containing 61% H2.

According to Chapman, kT in this case should be given by a relation of the form

where f1 refers to CO2 and f2 refers to H2. Since this relation is deduced on the assumption that the molecules behave
as rigid elastic spheres with a force relation in which s12 has a definite value for the interaction between the two types
of molecules, it was expected that the observed values of kT would differ from those calculated. Table 1.16 gives the
results of a series of measurements. The last column gives values of a constant defined by the relation
< previous page page_59 next page >
< previous page page_60 next page >
Page 60

Table 1.16. Separation Measurements in a Mixture of H2 and CO2


f1 Temperature Range (°C) kT (calc) kT (obs) RT
0.469 0.1575 0.0665 0.422
Below 144
0.1575 0.0939 0.596
Above 144
0.1575 0.1045 0.663
At 470

0.34 0.1685 0.0695 0.415


Below 145
0.1685 0.0929 0.552
Above 145

For CO2N2 mixture, Chapman's expression for kT is of the form

where f1 again refers to the heavier gas. (In all the following equations, f1 has the same significance.)

For f1 = 0.494, the value of kT calculated is 0.0564, whereas the value of RT observed varied from 0.247 for temperatures below 144°C to 0.441
for temperatures above 144°C.

For the mixtures ArHe, NeH2, and N2He, relations similar to Eq. (1.210) and Eq. (1.212) for kT have been published by Ibbs and Grew [119].
All the observed values of RT showed a decrease with decrease in temperature down to 190°C.

According to Chapman and Cowling,

Hence observations on the values of RT should yield values of s12. In fact the observations on thermal diffusion of gases should reveal
interesting information on the forces between molecules. Table 1.17 gives values of the repulsive force constant s12, as determined from
observed values of RT, and, for comparison, values of s1 and s2 for the individual gases, as determined from other phenomena.

In 1938, as mentioned previously, Clusius and Dickel [115] devised an arrangement of hot and cold surfaces that could be utilized practically for
the separation of mixtures of different gases and of isotopes. They used a long vertical tube with a hot wire along the axis. Because of thermal
diffusion, the relative concentration of the heavier molecules is greater at the cold wall. Convection causes the gas at the hot surface to rise to
the top, where it is deflected to the cold wall. There the gas sinks to the bottom and the cycle is then repeated. As a result, the heavier
components concentrate at the bottom, whereas the lighter ones concentrate at the top.

For instance, in a chamber 1 meter in height, for a temperature difference of 600°C, it was possible to separate a mixture containing 40%
(by volume) CO2 and 60% H2, with a yield of practically 100% CO2 at the bottom and 100% H2 at the top. In

< previous page page_60 next page >


< previous page page_61 next page >
Page 61

Table 1.17. Values of ''Force" Constants for Interaction of Molecules


Gases RT s12 s1 s2
O2H2 0.48
7.3 7.6 11.3
N2Ar 0.47
7.2 8.8 7.35
H2N2 0.58
8.2 11.3 8.8
HeAr 0.65
9.0 14.6 7.35
H2Ne 0.74
11.4 11.3 14.5
H2CO2 0.47
7.2 11.3 5.6
NeAr 0.54
7.9 14.5 7.35
HeNe 0.80
11.4 14.6 14.5

a chamber 2.9 meters in height, for a temperature difference of 600°C, air (21% O2, 78% N2) could be separated, yielding 85% O2 at the
bottom. Finally, from a mixture of 23% HCl37 and 77% HCl35, a mixture containing 40% of the heavier isotope was obtained at the bottom of
the diffusion chamber.

Brewer and Bramley [120] used a heated inner tube 1 cm in diameter and an outer cooled concentric tube 2 cm in diameter, each 1 meter long.
With a 350°C difference in temperature, He and Br2 could be separated, so that after 15 min no Br2 could be detected at the top. Under
similar conditions a 5050 mixture of CH4 and NH3 was enriched 25% in NH3 at the bottom. Brewer and Bramley deduced the following
relation: "If l denotes the cylinder length, r the radius of the outer tube, and d the difference in radii, then, to a first approximation, the rate of
separation varies as rd and the purity as l/d."

In a thermal-diffusion column developed by Nier [121] for the concentration of C13H4, a Nichrome heating wire threaded through porcelain
insulators was inserted along the axis of a steel tube 3/4 in. in outside diameter, which was concentric with a steel tube 1¾ in. in outside
diameter having a wall thickness of 0.035 in. This tube constituted the inner wall of the annular space containing the gas, and the outer wall
consisted of a 2-in.-outside-diameter brass tube, which was water-cooled. The column, which was 24 ft in length, is used as a return lead for the
current flowing through the Nichrome wire. At the lower end, the two steel tubes were brazed concentrically to a circular steel plate. The
average temperature of the outer wall was 27°C, and that of the inner wall was approximately 375°C. The samples of methane taken at the
bottom and top were analyzed by means of a mass spectrometer for the ratio of C13H4 to C12H4. At a pressure of 656 Torr and after operating
for 23.5 hours, the values of this ratio at the bottom and top were 0.0215 and 0.0054 respectively, so that the ratio of C13H4 at the bottom to that
at the top was 3.99.

The results obtained were found to be in agreement with conclusions deduced from theoretical considerations by Furry, Jones, and Onsager
[122]. They derived the relation

< previous page page_61 next page >


< previous page page_62 next page >
Page 62

where

represent equilibrium concentrations of C13H4 in the lower and upper ends of the column, respectively,
a and b are constants for the gas used and the dimensions of the column, and p denotes the pressure. Thus εx
corresponds to the separation factor. In the investigation under consideration, this factor was observed to pass through a
maximum at p = 0.6 atm, approximately.

The separation of C13H4 and C12H4 has also been investigated, in a column 40 ft in length, by Taylor and Glockler
[123], who found that the separation factor could be represented, in agreement with the theory, by the above equations.

The rate at which equilibrium is approached in such a column has been discussed by Bardeen [124] and the conclusions
reached have been found to be in good agreement with experimental results obtained by Nier.

Although the above discussion has dealt largely with the separation of gases which differ in value of M, it follows from
the theory of thermal diffusion that it should be possible by this method to effect a certain degree of separation in a
mixture of gases which have the same value of M but differ with respect to the magnitude of the molecular diameter.
Results in agreement with this prediction have been obtained by Wall and Holley [125].

1.12
Theory of Diffusion of Gases

As in the case of viscosity and heat conduction, approximate kinetic-theory considerations lead to the conclusion that
the coefficient of self-diffusion, D, is given by a relation of the form

which indicates that, at constant temperature, D varies directly as the mean free path or inversely as the pressure.

Combining Eq. (1.216) with the relation

Introducing the correction for persistence of velocities and that for Maxwellian distribution of velocities, it is found that
the more accurate relation is

According to Chapman and Enskog [126], the coefficient in Eq. (1.217) should have a value lying between 1.200 for
hard spheres and 1.543 for inverse-fifth-power repulsion.

< previous page page_62 next page >


< previous page page_63 next page >
Page 63

Actually observed values of D yield the following values of the coefficients [127]:

CO, 1.30; H2, 1.37; O2, 1.40; CO2, 1.58.

For the derivation of a relation for the coefficient of interdiffusion, D12 (corresponding to the fact that molecules 1 and
2 are involved), we shall follow the StefanMaxwell derivation [128], which, as will be evident, is based on the same
considerations as those used in deriving Eq. (1.216) above.

Letting n1 and n2 represent the number of molecules of each kind per unit volume at a location (x, y, z) and assuming
the concentration gradients to occur only with respect to the x-axis, then for a gas at rest we obtain

dn1/dx + dn2/dx = 0.

The rate of flow, J1, of molecules of the first type countercurrent to the flow of molecules of the second type at the rate
J2 is defined by

where D12 is the mutual diffusion coefficient or coefficient of interdiffusion. The Meyer formula [129] for D12 is

where v1 and v2 are average velocities.

The StefanMaxwell formula for the mean free path, L1, in a mixture of two gases, designated 1 and 2, according to Eq.
(1.140), is

However, as Stefan and Maxwell have pointed out, the collisions between like molecules do not influence diffusion, and
hence for this case the mean free paths are given by the relations

Since for thermal equilibrium


< previous page page_63 next page >
< previous page page_64 next page >
Page 64
and

On this basis it is found that the relation for the coefficient of interdiffusion is

where n = n1 + n2.

It is evident that for like molecules this relation becomes

Combining this with the relation

η = 0.499ρva/(21/2πnδ2),

the result is

which is in substantial agreement with Eq. (1.217).

Table 1.18 [130] gives the values of D12 (cm2 · s1) observed for several pairs of gases, at 0°C and 1 atm, values of δ12 calculated
from the values of D12 by means of Eq. (1.224), and, for comparison, values of δ12 calculated from viscosity measurements (see
Table 1.6) for each of the two gases.

For the first six pairs, which are comprised of "hard" molecules (with a value for the repulsive force constant greater than 7.4), the
agreement is very satisfactory. Even in the case of "softer" molecules the agreement is fair, as shown by the values for the second
set of five pairs.

Table 1.18. Coefficients of Interdiffusion and Average Molecular Diameters


Gases D12 (obs) 108δ12 (calc from D12) 108δ12 (calc from η)
H2air
0.661 3.23 3.23
H2O2
0.679 3.18 3.17
O2air
0.1775 3.69 3.68
O2N2
0.174 3.74 3.70
COH2
0.642 3.28 3.25
COO2
0.183 3.65 3.70

CO2H2
0.538 3.56 3.69
CO2air
0.138 4.03 4.20
CO2CO
0.136 4.09 4.22
N2OH2
0.535 3.57 3.69
N2OCO2
0.0983 4.53 4.66

< previous page page_64 next page >


< previous page page_65 next page >
Page 65

Summerhays [131] has used the katharometer (see Section 1.8) to measure the diffusion coefficient of water vapor in
air. The value observed was D = 0.282 cm2 · s1 at 16.1°C.

As will be noted, Eq. (1.224) indicates that the interdiffusion coefficient should be independent of composition.
However, a series of experiments carried out at Halle [132] to test this conclusion show that there is a variation of not
more than a few percent with variation in the ratio n1/n2.

Mention should be made in this connection of a relation for the coefficient of interdiffusion, given by Loeb [133], which
is of the form

in which β is a numerical factor, the value of which is between 1.00 and 1.50, and n = n1 + n2.

From the equations for D it follows that to a first approximation D should vary as T3/2 and as P1. Since, however, η
varies with T in accordance with Sutherland's relation, Eq. (1.122), it is expected that the exponent of T would exceed
3/2. Lonius [132] gives the following relation for D as a function of T:

where DT = value at T (in degrees Kelvin) and Pt (in Torr) and D = value at 15°C and 760 Torr. The values of x used by
Lonius for several binary mixtures are all in the neighborhood of 1.75.

Ivakin and Suetin [134] have measured the temperature dependence of the diffusion coefficients of 18 pairs of gases in
the temperature range 290470 K. Using their experimental data they calculated the interaction potential parameters for
potential functions of the Lennard-Jones type and of the modified Buckingham type and also for point center repulsion,
φ(r) = d/rδ. For the latter case they used the formula

D12 = D0TN,

where D0 is given by a complicated formula taken from the treatise by Hirschfelder, Curtiss, and Bird [135], while N =
3/2 + 2/δ. The calculated δ values ranged from 4.33 for H2Ar to 14.29 for Heair. For comparison they also present
potential parameters obtained from viscosity measurements and the second virial coefficient. The potential parameters
obtained from the diffusion equations gave the best fit to the temperature behaviour.

1.12.1
MaxwellLoschmidt Method for Determination of Diffusion Coefficients

In view of the fact that in vacuum technique a problem arises occasionally regarding the time required for gases to mix,
it is of interest to review briefly one method which has been used for the determination of D for gases.

< previous page page_65 next page >


< previous page page_66 next page >
Page 66

A tube of length l is divided by means of a wide-bore stopcock into two parts of equal length. This is mounted in a vertical position; the
heavier gas (A) is put in the lower part, and the lighter gas (B) is put in the upper part. At time t = 0 the stopcock is opened and diffusion is
allowed to occur for a definite period, t. The stopcock is then closed, and the proportions of A and B are measured in each compartment.

Let U = amount of A expressed as the fraction of the total number of moles in the lower compartment,
Q = amount of A in the upper compartment, expressed similarly.

Then, on the basis of the equations for diffusion,

For instance, in the case of H2O2 mixture, for l = 99.93 cm and t = 1800 s we obtain U = 0.5982 and Q = 0.4018 at 14.8°C and 749.3 Torr.
Hence, at 15°C and 760 Torr, D = 0.788.

Table 1.19 gives values of UQ for two different values of l as a function of Dt, along with values of t, for l = 100 cm, assuming D = 0.2 cm2
· s1.

For l = 100, only the first term in the series on the right-hand side of Eq. (1.229) is of importance for values of Dt > 500. For l = 25, only the
first term is of importance for Dt > 100. Equation (1.230) shows that for values of l and Dt for which the first term only is sufficiently
accurate, t varies as l2 for constant value of (U Q).

Ziering and Sheinblatt [136] have considered diffusion theory for both thermal gradients and density gradients in rarefied gases valid in the
entire pressure range from the continuum domain to the free-molecular-flow domain as applied to

Table 1.19. Illustrating the Application of Eq. (1.229)


Dt l = 100 cm l = 25 cm t (min) for D = 0.2 cm2 · s1
UQ UQ
100 0.9550 0.2063 8.33
200 0.8401 0.0425 16.67
300 0.7514 0.0088 25.00
400 0.6771 33.34
500 0.6118 41.67
700 0.5014 58.67
1000 0.3728 83.83
1500 0.2276 125.00
2000 0.1390 166.67
3000 0.0518 250.00
5000 0.0071 416.67

< previous page page_66 next page >


< previous page page_67 next page >
Page 67

molecules which obey a center of repulsive force F = K12/rv, where v = 5 (Maxwell molecules) and in the extreme of
rigid-sphere molecules, v = ∞.

Cunningham and Geankoplis [137] have presented equations for diffusion in three-component mixtures in the transition
region between Knudsen and molecular diffusion in an open system. Solution of the equations is, by trial and error,
aided by computer.

1.12.2
Effect of Pressure of Gas on Rates of Evaporation of Metals

In Section 1.5, equations were derived for rates of evaporation in a vacuum. These equations have been applied, as
illustrated in the above connection, to the calculation of vapor pressures of high-melting-point metals from observations
on rate of loss in weight as a function of the temperature. However, in the presence of a gas which does not react
chemically with the metal, the observed rate of evaporation is lower; as is well known, this fact was utilized by
Langmuir in the invention and development of the gas-filled tungsten-filament lamp. This phenomenon has been
explained by Langmuir by assuming the existence of a film adjacent to the evaporating surface through which the atoms
evaporated from the surface diffuse.

Fonda [138] has shown that this theory accounts very satisfactorily for his observations on the rate of evaporation of
tungsten in the presence of argon at various pressures, and the following remarks on the mechanism of evaporation
under these conditions are quoted from his paper:

The filament is considered to be surrounded by tungsten vapor at the same pressure as would be present in a
vacuum. The atoms of this vapor, however, instead of being projected directly from the filament, as in a
vacuum, are pictured as diffusing through the stationary film of gas. Once an atom reaches the outer boundary
of the film, it would be carried away in the convection current of gas and would be lost to the filament; but the
path within the film is so irregular that an atom may in fact return to the filament and be deposited on it, thus
leading to a reduced evaporation as compared with that in a vacuum.

Let dc/dr denote the concentration gradient at the surface of the wire, where r is the distance from the axis, and let D
denote the diffusion constant. Assuming uniform distribution over the surface of the wire, the rate of diffusion per unit
length of the wire is

where D is given by Eq. (1.224).

Since the value of n for tungsten atoms is negligible compared with that of n for argon gas, the expression for D
becomes identical with Eq. (1.216), where L and va refer to the values of the mean free path and average velocity,
respectively, of the tungsten atoms at the temperature of the filament. Thus D varies inversely as the pressure, P of
argon, and Eq. (1.231) becomes

where A is a proportionality constant.

< previous page page_67 next page >


< previous page page_68 next page >
Page 68

For a wire of diameter 2a and gas film of diameter 2b, it follows, from the same considerations as those that lead to Eq. (1.192) for the heat loss, that

Let m denote the rate of evaporation in grams per square centimeter per second. Then it follows from Eqs. (1.232) and (1.233) that

where ca and cb are the concentrations of the tungsten atoms at r = a and r = b, respectively.

Table 1.20 shows results obtained by Fonda for the rate of evaporation from a filament in a mixture of argon and nitrogen, such as is used in
gas-filled lamps. The value m = 230 × 109 g · cm2 · s1 observed for P = 0 is in agreement with the value 250 × 109 observed by Langmuir,
Jones, and Mackay [139].

As Fonda states: "The expression maPcm log (b/a) is sufficiently constant at pressures above 10 cm (100 Torr) to allow of credence being given
the hypothesis developed above."

The constancy of the expression has a further significance, because, as is evident from its derivation, it denotes a constant difference between
the vapor pressure of tungsten at the surface of the filament and at the surface of the film of gas. For constant filament temperature this implies
that the concentration of tungsten vapor at the surface of the stationary film of gas should be constant for all pressures above 10 cm. It is this
vapor which constitutes that which effectively evaporates from the filament, because the rising convection currents of gas carry it off to be
deposited on the bulb. The rate of evaporation at different pressures should be determined therefore by the area of the assumed film of gas if
the concentration of vapor at its border is in

Table 1.20. Rate of Evaporation (m) of Tungsten in 86% Argon, 14% Nitrogen at 2870 Ka
Pcm 109m b (cm) 109maPcm log(b/a) 109m/b
0 230
1 57.5 9.68 3.9
5 23.5 2.42 6.3
10 20.5 1.31 9.8 15.6
25 10.3 0.63 10.4 16.2
50 5.4 0.36 9.6 14.8
70 4.2 0.28 9.6 14.8
165 2.0 0.15 8.8 13.5
aDiameter of filament (2a) = 0.00978 cm; Pcm = pressure in cm of Hg.

< previous page page_68 next page >


< previous page page_69 next page >
Page 69

fact a constant. The constancy of the ratio m/b is therefore a further confirmation of the validity of the theory. Other
evidence, based on the appearance of the surface of the filament, has also been shown by Fonda to be in agreement with
the views expressed above.

From Eq. (1.234) it also follows that at constant pressure and for filaments of different diameters the expression ma log
(b/a) should be constant. The validity of this conclusion was confirmed by Fonda in a series of observations made with
both pure nitrogen and a mixture of nitrogen and argon [140]. In these cases the value actually used for B (the thickness
of the film for a plane surface in gas at atmospheric pressure) was the same as that obtained from data on heat
conduction in nitrogen.

1.13
Random Motions and Fluctuations [141]

Let us consider the case in which a group of N molecules start from the plane z = 0, at the instant t = 0, and diffuse
through the gas. The differential equation for the diffusion is

At any instant, t, the distribution of these molecules along the z axis is given by the relation

which satisfies the condition

Hence, the average value of z is

And similarly it can be shown that the root-mean-square value is given by the relation

That is, the total net displacement in any given direction varies as t1/2.

These equations are of special importance in the determination of diffusion coefficients from observations on the
Brownian motion of small particles.

< previous page page_69 next page >


< previous page page_70 next page >
Page 70

We can also express Eqs. (1.237) and (1.238) in terms of the mean free path L, on the basis of the relation

Substituting in the above equations we obtain the relations

The collision frequency per molecule is

ω = va/L,

while the total length of path actually traversed by a molecule in time t is

That is, the total length of path varies as the square of the net displacement from the point of origin at t = 0.

As an illustration of the above equations let us consider the case of molecules in air at 25°C and atmospheric pressure.
Since L = 6.69 × 106 and va = 4.67 × 104, it follows from Eq. (1.239) that D = 0.104 cm2 · s1, and for t = 60 s we
obtain zr = 3.53 cm, while l = 6.98 × 109 t = 4.19 × 1011 cm.

That l = 1.19 × 1011zr is obviously due to the fact that after each collision the probability of a displacement toward
lower (or more negative) values of z is just as great as that of a displacement toward more positive values of z.

The equations for random motion have been applied to determine the value of Avogadro's constant from observations
on rates of diffusion of Brownian particles. It may be demonstrated that the rate of diffusion of spherical particles is
given by the relation
where a = radius of particle and η = viscosity of medium.

< previous page page_70 next page >


< previous page page_71 next page >
Page 71

Also, the mean square of the displacement per second is expressed as

Values of NA obtained by application of these relations, though not nearly as accurate as those obtained from
determinations of the unit electric charge (the method used by Millikan), are in good agreement with them.

1.14
Scattering of Particle Beams at Low Gas Pressures

As calculated on the basis of classical collision theory of smooth elastic spheres, a neutral projectile particle of mass m1
traveling with uniform velocity w1 before scattering by a target molecule of mass m2 moving with average speed, w2,
much smaller than the magnitude of the projectile velocity so that we can set w2 = 0, will suffer a fractional loss of
energy (see Section 1.3.).

so that, using Eqs. (1.39) and (1.44),

where θ is the angle of incidence between w1 and the line of centers at collision in the laboratory system.

When the target particle is at rest in the laboratory system, the center of mass moves in this system with the constant
velocity

In the center-of-mass coordinate system, CM, the velocity of the projectile particle is w1 w12 and the velocity of the
target particle is w12. By combining the various vectors in both systems in one diagram it can be shown [142] that

where β is the angle of scattering in the center-of-mass system. Then the fractional energy loss can be written as

where βL is the angle of scattering in the laboratory system.

< previous page page_71 next page >


< previous page page_72 next page >
Page 72

For a parallel monoenergetic beam of projectile particles uniformly distributed in a plane normal to the direction of
motion, the probability of a particle being scattered in the angle between βL and dβL is

so that the integral of this probability over all angles from 0 to π equals 1. There is no backwards scattering from an
initially stationary target particle in the laboratory reference frame.

When as in the case of elastic scattering of an electron, the scattering angle βL is approximately equal to β
and the laboratory system is almost the same as the center-of-mass system.

For a molecular beam emerging into a ''perfect vacuum" there will still be some scattering as the faster molecules
overtake the molecules moving more slowly. By dividing the beam molecules into two groups, each with a narrow
range of differing velocities but following a Maxwellian law of distribution, Troitskii [143] has calculated that the mean
free path for collision of molecules within a unidirectional beam (far from the source opening) is about three times the
mean free path in a stationary gas of the same particle density as in the beam, neglecting second collisions with particles
diffusely scattered by a first collision.

When the projectile particles are charged and moving with very high speeds, the target particle becomes the nucleus of
an atom in the gas, and the scattering is designated as Coulomb scattering. At pressures of 106 Torr in the vacuum
chamber of a synchrotron, or the beam tube of particle accelerators in general, the mean free path for a hydrogen atom
in nitrogen might typically be of the order of 104 cm, but the "mean free path" for collision of a hydrogen nucleus
(proton) with the nucleus of a nitrogen atom might be more like 107 cm, depending on the velocity. When the projectile
particle is charged, the effect of scattering is compensated to a large degree by focusing magnetic fields which redirect
the particle away from the walls while being accelerated by electromagnetic fields over a total path length of nearly cta,
where ta is the acceleration time. For ta = 1 s, this path length is 3 × 1010 cm, so that many collisions with the nuclei in
gas molecules at 106 Torr will occur. Without magnetic focusing, even deflection through a small scattering angle β
would be sufficient to scatter the projectile particle into the wall of the vacuum chamber.

In the presence of the magnetic fields, elastic scattering gives rise to "betatron" oscillations with amplitudes which
increase slowly with time until the particle is captured by the wall of the vacuum chamber. Blachman and Courant [144]
in 1948 published formulas which make possible an estimate of the fraction of the original particles in a circular
synchrotron which is scattered to the wall. They begin with the Rutherford formula for the scattering cross section, dσ,
of a particle of charge e by the nucleus of a target particle with charge Ze, into any solid angle, dω, at the angle β,

where m is the relativistic mass of the projectile particle and v0 is its velocity. Because of the finite size of the nucleus,
there is an upper limit on β given by λ/b, where

< previous page page_72 next page >


< previous page page_73 next page >
Page 73

λ = h/mv0 is the de Broglie wavelength and b is the "radius" of the nucleus. These authors give an equation for the
allowable pressure in Torr inside the vacuum chamber of a synchrotron which is directly proportional to the parameter

when the effects of damping are included, where is the mean square maximum amplitude of the vertical betatron
oscillations and A is the vertical semiaperture of the chamber cross section. If no more than a 10% loss through vertical
scattering is to be permitted, then η must be no more than 0.089 and the calculated upper limit of pressure is of the order
of 1.3 × 104 Pa (106 Torr) in typical synchrotrons. In 1949 Blachman and Courant [145] derived new equations to apply
to the "racetrack" modification of the synchrotron by the insertion of straight sections.

In 1951 Greenberg and Berlin [146] considered the gas scattering problem for protons and for electrons in great detail
and estimated the limiting pressures at which 50% of the protons are lost and at which 10% of the protons are lost in the
Brookhaven and Birmingham Synchrotrons. The values depend on the method of particle injection, but in general
indicate that pressures less than 1.3 × 104 Pa (106 Torr) are required to avoid loss of more than 10% in these machines.
Similar calculations were made in 1955 by Moravcsik and Sellen [147] for the strong focusing electron synchrotron at
Cornell.

In 1953 Courant [148] revised the previous gas scattering theory to include effects of scattering through angles large
enough for immediate particle loss. Kheifets [149] has used scattering formulas to estimate the lifetime in a storage
device of a beam of electrons. Various methods for computing particle loss by gas scattering in cyclical accelerators
were compared by Didenko and Serdyutskii [150] in 1963. Orlov and Kheifets [151] have used revised formulas with
appropriate boundary conditions for estimating the particle loss due to multiple Coulomb scattering in a cyclical
accelerator.

References and Notes

1. In connection with this chapter, use has been made of the following treatises:

a. E. H. Kennard, Kinetic Theory of Gases with an Introduction to Statistical Mechanics. McGraw-Hill, New York,
1938.

b. L. B. Loeb, Kinetic Theory of Gases. McGraw-Hill, New York, 1934.

c. J. H. Jeans, An Introduction to the Kinetic Theory of Gases. Macmillan, New York, 1940; Cambridge University
Press, Cambridge, UK, 1960.

d. S. Chapman and T. G. Cowling, The Mathematical Theory of Non-Uniform Gases. University Press, Cambridge,
UK, 1939; 3rd ed., 1970.

Some other treatises which are recommended for further details are:

G. A. Bird, Molecular Gas Dynamics. Oxford University Press (Clarendon), Oxford, 1978.

C. Cercignani, Mathematical Methods in Kinetic Theory. Plenum, New York, 1969.

E. A. Guggenheim, Elements of the Kinetic Theory of Gases. Pergamon, New York, 1960.

M. N. Kogan, Rarefied Gas Dynamics. Plenum, New York, 1969.

L. C. Woods, An Introduction to the Kinetic Theory of Gases and Magnetoplasmas. Oxford University Press,
Oxford, 1993.
T. Wu, Kinetic Equations of Gases and Plasmas. Addison-Wesley, Reading, MA, 1966.

< previous page page_73 next page >


< previous page page_74 next page >
Page 74

2. The values given for V0 and other constants in this and subsequent sections are those recommended by CODATA: E.
R. Cohen and B. N. Taylor, Rev. Mod. Phys. 59, 11211148 (1987); Phys. Today, August, BG9 (1996).

3. E. Thomas and R. Leyniers, in: Adam (ed.) Trans. 3rd Int. Congr., 1965, Vol. 2, Part 1, pp. 279285 Pergamon,
London (1966); Trans. Czech. Conf. Electron. Vac. Phys., 3rd, Prague, 1965, pp. 457470 (1967).

4. E. Thomas, Vacuum 13, 376 (1963).

5. Document ISO 31/111-1978(E). International Standards Organization, London.

6. Glossary of Terms Used in Vacuum Technology, British Standard 2951. Part 1. British Standards Institution, London,
1969 (first published in 1958).

7. B. B. Dayton, Trans. Natl. Vac. Symp., 1st, Asbury Park, NJ, 1954, pp. 132147 (1955).

8. E. Thomas, ed., Advances in Vacuum Science and Technology (Proc. 1st Int. Congr., Namur, Belg., 1958), Spec.
Meet. Pressure Units, pp. 155158. Pergamon, London, 1960.

9. M. S. Kaminsky and J. M. Lafferty, eds., Dictionary of Terms for Vacuum Science and Technology, Surface Science,
Thin Film Technology and Vacuum Metallurgy. AIP Press, Woodbury, NY, 1980.

10. C. D. Ehrlich, J. Vac. Sci. Technol. A 4, 2384 (1986); C. D. Ehrlich and J. A. Basford, ibid. 10, 1 (1992); G. M.
Solomon, ibid. 4, 327 (1986); G. Lewin, J. Vac. Sci. Technol. 5, 75 (1968).

11. E. H. Kennard, Kinetic Theory of Gases with an Introduction to Statistical Mechanics, p. 110. McGraw-Hill, New
York, 1938.

12. A. Roth, Vacuum Technology, p. 45. North-Holland Publ., Amsterdam, 1976; 3rd ed., 1990.

13. M. S. Kaminsky and J. M. Lafferty, eds., Dictionary of Terms for Vacuum Science and Technology, Surface
Science, Thin Film Technology and Vacuum Metallurgy, p. 71. AIP Press, Woodbury, NY, 1980.

14. E. H. Kennard, Kinetic Theory of Gases with an Introduction to Statistical Mechanics, p. 101. McGraw-Hill, New
York, 1938.

15. E. H. Kennard, Kinetic Theory of Gases with an Introduction to Statistical Mechanics. p. 6. McGraw-Hill, New
York, 1938.

16. M. Knudsen, Ann. Phys. (Leipzig) [4] 48, 1113 (1915).

17. G. Comsa, R. David and K. D. Rendulic, Phys. Rev. Lett. 38, 775 (1977); G. Comsa, Surf. Sci. 299/300, 77 (1994).

18. W. Van Willigen, Phys. Lett. 28A, 80 (1968); R. L. Palmer, J. N. Smith, Jr., H. Saltsburg and D. R. O'Keefe, J.
Chem. Phys. 53, 1666 (1970); T. L. Bradley, A. E. Dabiri and R. E. Stickney, Surf. Sci. 29, 590 (1972); 38, 313 (1973);
26, 522 (1971); F. C. Hurlbut and F. S. Sherman, Phys. Fluids 11, 486 (1968); S. Nocilla, in Rarefied Gas Dynamics (J.
H. de Leeuw, ed.), Vol. 1, p. 315. Academic Press, New York, 1966.

19. M. Epstein, AIAA J. 5 (October issue), 1797 (1967).

20. D. Da and X. Da, J. Vac. Sci. Technol. A 5, 2484 (1987); R. A. Haefer, Vak.-Tech. 16, 149, 185 (1967); M. Lunc, in
Rarefied Gas Dynamics (J. A. Laurmann, ed.), Vol. 1, pp. 94101. Academic Press, New York, 1953.
21. B. B. Dayton, Ind. Eng. Chem. 40, 795 (1948); Vacuum 15, 53 (1965); W. W. Stickney and B. B. Dayton, Trans.
Natl. Vac. Symp. 10, 105116 (1963); R. A. Haefer, Vak. Tech. 16, 149, 185, 210 (1967); W. Steckelmacher and D.
Turner, J. Sci. Instrum. 43, 893 (1966); S. Komiya and T. Ikeda, in Adv. Vac. Sci. Technol., Proc. Int. Congr., 1st,
Namur, Belg., 1958, E. Thomas (ed.), pp. 323328 Pergamon, London (1960); D. H. Holkeboer, 10th Natl. Vac. Symp.,
292, Macmillan Co., New York (1963).

22. E. P. Muntz, Phys. Fluids 11, 64 (1968).

23. M. Knudsen, Ann. Phys. (Leipzig) [4] 28, 75 [4] (1909).

< previous page page_74 next page >


< previous page page_75 next page >
Page 75

24. I. Langmuir, Phys. Rev. 2, 329 (1913).

25. I. Langmuir, Phys. Rev. 2, 329 (1913), and subsequent papers.

26. F. H. Verhoek and A. L. Marshall, J. Am. Chem. Soc. 61, 2737 (1939).

27. H. A. Jones and I. Langmuir, Gen. Electr. Rev. 30, 310, 354, 408 (1927).

28. D. B. Langmuir and L. Malter, Phys. Rev. 55, 748 (1939).

29. A. C. Egerton, Philos. Mag. [6] 33, 33 (1917).

30. A. C. Egerton, Proc. R. Soc. London, Ser. A 103, 469 (1923).

31. A comprehensive review of this topic has been published by K. C. D. Hickman, one of the pioneers in this field, in
Chem. Rev. 34, 51 (1944).

32. M. Born, Phys. Z. 21, 578 (1920).

33. F. Bielz, Z. Phys. 32, 81 (1925).

34. R. G. J. Fraser, Molecular Rays. Cambridge University Press, Cambridge, UK, 1931; also Molecular Beams.
Methuen, London, 1937.

35. J. J. Weigle and M. S. Plesset, Phys. Rev. 36, 373 (1930).

36. F. Knauer and O. Stern, Z. Phys. 53, 766 (1929).

37. J. B. Taylor, Ind. Eng. Chem. 23, 1228 (1931); also see W. H. Bessey and O. C. Simpson, Chem. Rev. 30, 239
(1942).

38. A. C. Torrey, Phys. Rev. 47, 644 (1935); A. B. Van Cleave and O. Maass, Can. J. Res., Sect. B 12, 57 (1935); 13,
384 (1935).

39. This topic is discussed at length by Chapman and Cowling [1d] in their treatise on this subject. It will be observed
that LM = 1.611L = 1.131LB and LB = 1.425L.

40. S. Chapman and T. G. Cowling, The Mathematical Theory of Non-Uniform Gases, Sect. 10.41. University Press,
Cambridge, UK, 3rd ed., 1970.

41. E. H. Kennard, Kinetic Theory of Gases with an Introduction to Statistical Mechanics, p. 147. McGraw-Hill, New
York, 1938.

42. See sub-section 1.7.2 on molecular diameters in Section 1.7.

43. W. Licht, Jr. and D. G. Stechert, J. Phys. Chem. 48, 23 (1944).

44. LandoltBornstein Physikalisch-Chemische Tabellen, 5th ed., Springer, Berlin, 19231935.

45. A. E. Schuil, Philos. Mag. [7] 28, 679 (1939). This paper gives data on values of η for gases and vapors over the
range 0250°C.

46. E. H. Kennard, Kinetic Theory of Gases with an Introduction to Statistical Mechanics, p. 149. McGraw-Hill, New
York, 1938.
47. H.-S. Tsien, J. Aeronaut. Sci. 13, 653 (1946).

48. H. Braune, R. Basch, and W. Wentzel, Z. Phys. Chem. Abt. A 137, 447 (1928).

49. Value given in Ref. [48], which differs somewhat from that given in Table 1.8.

50. E. H. Kennard, Kinetic Theory of Gases with an Introduction to Statistical Mechanics, pp. 160162. McGraw-Hill,
New York, 1938.

51. W. Gaede, Ann. Phys. (Leipzig) [4] 46, 357 (1915).

52. E. H. Kennard, Kinetic Theory of Gases with an Introduction to Statistical Mechanics, p. 297. McGraw-Hill, New
York, 1938.

53. A. Timiriazeff [Ann. Phys. (Leipzig) [4] 40, 971 (1913)] assumed that f, which may be regarded as an
accommodation coefficient for transfer of momentum, has the same value for any gas-surface combination as α the
accommodation coefficient for heat transfer (discussed in Section 1.9.1); but B. Baule [Ann. Phys. (Leipzig) [4] 44, 145
(1914)], disagreed with this assumption and concluded that the value of the ratio between ζ and L [denoted by β in Eqs.
(1.148) and (1.149)] must be a complicated function of the diameters of the molecules in the gas and those constituting
the surface. The theoretical

< previous page page_75 next page >


< previous page page_76 next page >
Page 76

investigations on this topic are discussed by Kennard and Loeb. Loeb quotes results obtained by R. A. Millikan
[Phys. Rev. 21, 217 (1923)], which lead to values of f ranging from 0.87 to 1.00.

54. I. Langmuir, Phys. Rev. 1, 337 (1913).

55. Values of b for a large number of gases and vapors are given in the Handbook of Chemistry and Physics, published
by CRC Press, Boca Raton, FL. Since these values are expressed in terms of the volume at 0°C and 1 atm as unity, they
should be multiplied by 22,414 to give values for use in Eq. (1.154).

56. H. K. Livingston, J. Am. Chem. Soc. 66, 569 (1944); also S. Brunauer, The Adsorption of Gases and Vapors, p. 287.
Princeton University Press, Princeton, NJ, 1942.

57. H. A. Stuart, Molekülstruktur, pp. 49 et seq. Springer, Berlin, 1934.

58. See also P. H. Emmett and S. Brunauer, J. Am. Chem. Soc. 59, 1553 (1937).

59. A. Eucken, Phys. Z. 14, 324 (1913).

60. L. B. Loeb, Kinetic Theory of Gases, pp. 234252. McGraw-Hill, New York, 1934.

61. W. C. Kannuluik and L. K. Martin, Proc. R. Soc. London, Ser. A 144, 496, (1934).

62. See also the following references on this topic: H. Gregory and C. T. Archer, Proc. R. Soc. London, Ser. A 110, 91
(1926); 121, 284 (1928); H. Gregory and S. Marshall, ibid. 114, 354 (1927); 118, 594 (1928); B. G. Dickins, ibid. 143,
517 (1934); H. A. Daynes, Gas Analysis by Measurement of Thermal Conductivity. Cambridge University Press,
Cambridge, UK, 1933, gives a very comprehensive table of relative thermal conductivities (air = 1) for a large number
of gases, including hydrocarbons.

63. J. R. Partington and W. G. Shilling, The Specific Heat of Gases. Benn, London, UK 1924; see also G. N. Lewis and
M. Randall, Thermodynamics, p. 80. McGraw-Hill, New York, 1923; Bulletin 30, Cornell University, Engineering
Experiment Station, October, 1942, gives specific heats of a number of gases over a wide range of pressures and
temperatures.

64. ln = loge = 2.303 log10.

65. The exact meaning of this requirement will appear in the subsequent discussion.

66. See A. Farkas and H. W. Melville, Experimental Methods in Gas Reactions, p. 190. Cambridge University Press,
Cambridge, UK, 1939; also A. Farkas, Orthohydrogen, Parahydrogen, and Heavy Hydrogen. Cambridge University
Press, Cambridge, UK, 1935, for illustration of the application of this method.

67. G. A. Shakespear, Proc. R. Soc. London, Ser. A 97, 273 (1920); see also the comprehensive discussion of this
instrument by H. A. Daynes, Gas Analysis by Measurement of Thermal Conductivity. Cambridge University Press,
Cambridge, UK, 1933.

68. See especially a description of a modified construction by T. L. Ibbs, Proc. R. Soc. London, Ser. A 99, 385 (1921);
also 107, 470 (1925); W. E. Summerhays [Proc. R. Soc. London 42, 218 (1930)] has used the katharometer to measure
the coefficient of diffusion of water vapor.

69. B. G. Dickins, Proc. R. Soc. London, Ser. A 143, 517 (1934).

70. M. Knudsen, Ann. Phys. (Leipzig) [4] 31, 205 (1910); 32, 809 (1910); 33, 1435 (1910); 34, 593 (1911); [5] 6, 129
(1930); see also Loeb [1b, pp. 310325] and Kennard [1a, pp. 311320].
71. M. von Smoluchowski, a number of papers published before 1911 and Ann. Phys. (Leipzig) [4] 35, 983 (1911); also
discussed by Loeb [1b] and Kennard [1a].

72. E. H. Kennard, Kinetic Theory of Gases with an Introduction to Statistical Mechanics, pp. 311312. McGraw-Hill,
New York, 1938.

73. K. B. Blodgett and I. Langmuir, Phys. Rev. 40, 78 (1932).

74. E. H. Kennard, Kinetic Theory of Gases with an Introduction to Statistical Mechanics, p. 316. McGraw-Hill, New
York, 1938.

< previous page page_76 next page >


< previous page page_77 next page >
Page 77

75. A table of values of γ is given by Loeb [1b, p. 445]; also see Partington and Shilling [63].

76. M. von Smoluchowski, Ann. Phys. (Leipzig) [4] 35, 983 (1911).

77. J. H. Jeans, An Introduction to the Kinetic Theory of Gases, p. 193. MacMillan, New York, 1940.

78. J. A. Morrison and Y. Tuzi, J. Vac. Sci. Technol. 2, 109 (1965).

79. The data for platinum are from Loeb [1b, p. 321]. The data for tungsten are from H. A. Jones and I. Langmuir [Gen.
Electr. Rev. 30, 354 (1927)], while those for "ordinary" platinum are from B. G. Dickins [proc. R. Soc. London, Ser. A
143, 517 (1934)]. That the value of the accommodation coefficient is greatly affected by the nature of the surface film
has been shown by K. B. Blodgett and I. Langmuir [Phys. Rev. 40, 78 (1932)]. In the case of a tungsten wire in
hydrogen at 0.20 Torr, the value of α changes from 0.534 for bare tungsten to 0.094 for tungsten with an adsorbed film
of oxygen. The general problem of the interaction between molecules and solid surfaces, which is obviously involved in
the interpretation of α, has received considerable attention from a number of investigators. For detailed discussion,
especially with regard to the role of α in adsorption phenomena, see Loeb [1b, p. 311] and also J. K. Roberts (Some
Problems in Adsorption. Cambridge University Press, Cambridge, UK, 1939). Values of α for air and a number of
different metals having etched, polished, and machined surfaces have been determined by M. L. Wiedemann [Trans.
ASME 68, 57 (1946)]. These values range from 0.87 to 0.97.

80. W. H. Keesom and G. Schmidt, Physica (Amsterdam) 3, 590, 1085 (1936).

81. D. E. Klett and R. K. Irey, Adv. Cryog. Eng. 14, 217 (1969).

82. K. Frankowski, Z. Alterman, and C. L. Pekeris, Phys. Fluids 8, 245 (1965).

83. D. R. Willis, in Rarefied Gas Dynamics (J. A. Lauermann, ed.), Vol. 1, pp. 209225. Academic Press, New York,
1963.

84. P. L. Bhatnager, E. P. Gross, and M. Krook, Phys. Rev. 94, 511 (1954).

85. D. R. Willis, Phys. Fluids 8, 1908 (1965).

86. J. C. Havekotte and G. S. Springer, AIAA J. 7, 782 (1969).

87. I. Langmuir, Phys. Rev. 34, 401 (1912).

88. H. A. Jones, Gen. Elec. Rev. 28, 650 (1925).

89. Of course the loss due to radiation was subtracted from the total observed energy loss.

90. I. Brody and F. Körösy, J. Appl. Phys. 10, 584 (1939).

91. J. P. Hobson, T. Edmonds, and R. Verreault, Can. J. Phys. 41, 983 (1963); T. Edmonds and J. P. Hobson, J. Vac.
Sci. Technol. 2, 182 (1965).

92. G. A. Miller and R. E. Buice, J. Phys. Chem. 70, 3874 (1966).

93. Tentative Recommended Practice for Ionization Gauge Application to Space Simulators, ASTM E296-66T,
Appendix A2. Am. Soc. Test. Mater., Philadelphia.

94. J. A. Poulis, B. Pelupessy, C. H. Massen, and J. M. Thomas, J. Sci. Instrum. 41, 295 (1964).

95. M. Knudsen, Ann. Phys. (Leipzig) [4] 31, 205, 633 (1910); 83, 797 (1927).
96. L. B. Loeb, Kinetic Theory of Gases, p. 358. McGraw-Hill, New York, 1934.

97. M. J. Bennett and F. C. Tompkins, Trans. Faraday Soc. 53, 185 (1957).

98. S. C. Liang, Can. J. Chem. 33, 279 (1955); J. Phys. Chem. 57, 910 (1953).

99. S. C. Liang, J. Appl. Phys. 22, 148 (1951). See also J. P. Hobson, J. Vac. Sci. Technol. 7, 351 (1970) for measured
effect of surface roughness.

100. J. M. Los and R. R. Ferguson, Trans. Faraday Soc. 48, 730 (1952).

101. J. O. Hirschfelder, R. B. Bird, and E. L. Spotz, J. Chem. Phys. 16, 968 (1948); Chem. Rev. 44, 205 (1949).

< previous page page_77 next page >


< previous page page_78 next page >
Page 78

102. G. D. Arney and A. B. Bailey, AIAA J. 1, 3863 (1963).

103. M. Knudsen, Kinetic Theory of Gases, p. 37. Methuen, London.

104. J. P. Hobson, Vacuum 15, 543 (1965).

105. T. Edmonds and J. P. Hobson, J. Vac. Sci. Technol. 2, 182 (1965).

106. S. Weber and G. Schmidt, Kamerlingh Onnes Laboratory, Commun. No. 246C. University of Leiden, 1936.

107. H. H. Podgursky and F. N. Davis, J. Phys. Chem. 65, 1343 (1961).

108. F. Sharipov, J. Vac. Sci. Technol. A 14, 2627 (1996).

109. L. A. Dietz, Rev. Sci. Instrum. 27, 817 (1956).

110. W. Jitschin and P. Röhl, J. Vac. Sci. Technol. A 5, 372 (1987).

111. T. Takaishi and Y. Sensui, Trans. Faraday Soc. 59, 2503 (1963).

112. J. C. Williams, III, J. Vac. Sci. Technol. 8, 446 (1971).

113. T. L. Ibbs, Physica (Amsterdam) 4, 1135 (1937).

114. S. Chapman and F. W. Dootson, Philos. Mag. [6] 33, 248 (1917); see also Jeans [1c, p. 251]. A mathematical
treatment of the subject is given by Chapman and Cowling [1d, pp. 140147, 351], as well as by W. H. Furry, R. Clark
Jones, and L. Onsager [Phys. Rev. 55, 1083 (1939)] and R. Clark Jones and W. H. Furry [Rev. Mod. Phys. 18, 151
(1946)]. A "simple theory" has been published by L. J. Gillespie [J. Chem. Phys. 7, 530 (1939)], and a much more
elaborate mathematical discussion by S. Chapman [Proc. R. Soc. London, Ser. A 177, 38 (19401941)].

115. K. Clusius and G. Dickel, Naturwissenschaften 26, 546 (1938); 27, 148 (1939).

116. T. L. Ibbs, Proc. R. Soc. London, Ser. A 107, 470 (1925).

117. See L. J. Gillespie, J. Chem. Phys. 7, 530 (1939), for a discussion of the derivation of this equation.

118. T. L. Ibbs and A. C. R. Wakeman, Proc. R. Soc. London, Ser. A 134, 613 (1932).

119. T. L. Ibbs and K. E. Grew, Proc. R. Soc. London, Ser. A 43, 142 (1931).

120. A. K. Brewer and A. Bramley, Phys. Rev. 55, 590 (1939).

121. A. O. Nier, Phys. Rev. 57, 30 (1940).

122. W. Furry, R. Clark Jones, and L. Onsager, Phys. Rev. 55, 1083 (1939).

123. T. I. Taylor and G. Glockler, J. Chem. Phys. 8, 843 (1940).

124. J. Bardeen, Phys. Rev. 57, 35 (1940).

125. F. T. Wall and C. E. Holley, J. Chem. Phys. 8, 949 (1940).

126. E. H. Kennard, Kinetic Theory of Gases with an Introduction to Statistical Mechanics, pp. 195196. McGraw-Hill,
New York, 1938.
127. J. H. Jeans, An Introduction to the Kinetic Theory of Gases, p. 216. MacMillan, New York, 1940.

128. J. H. Jeans, An Introduction to the Kinetic Theory of Gases, pp. 207210. MacMillan, New York, 1940.

129. E. H. Kennard, Kinetic Theory of Gases with an Introduction to Statistical Mechanics, p. 189. McGraw-Hill, New
York, 1938.

130. J. H. Jeans, An Introduction to the Kinetic Theory of Gases, pp. 217218. MacMillan, New York, 1940.

131. W. E. Summerhays, Proc. Phys. Soc., London, 42, 218 (1930).

132. The results are summarized by A. Lonius, Ann. Phys. (Leipzig) [4] 29, 664 (1909).

133. L. B. Loeb, Kinetic Theory of Gases. McGraw-Hill, New York, 272 (1934).

134. B. A. Ivakin and P. E. Suetin, Sov. Phys.Tech. Phys. (Engl. Transl.) 9, 866 (1964).

< previous page page_78 next page >


< previous page page_79 next page >
Page 79

135. J. Hirschfelder, C. Curtiss, and R. Bird, Molecular Theory of Gases and Liquids. Wiley, New York, 1954.

136. S. Ziering and M. Sheinblatt, Phys. Fluids 9, 1674 (1966).

137. R. S. Cunningham and C. J. Geankoplis, Ind. Eng. Chem., Fundam. 7 (August issue), 429 (1968).

138. G. R. Fonda, Phys. Rev. 21, 343 (1923).

139. I. Langmuir, H. A. Jones, and G. M. J. Mackay, Phys. Rev. 30, 211 (1927).

140. G. R. Fonda, Phys. Rev. 21, 343 (1923).

141. E. H. Kennard, Kinetic Theory of Gases with an Introduction to Statistical Mechanics, Chapter 7. McGraw-Hill,
New York, 1938.

142. E. W. McDaniel, Collision Phenomena in Ionized Gases. Wiley, New York, 1964.

143. V. S. Troitskii, Sov. Phys.JETP (Engl. Transl.) 14, 281 (1962).

144. N. M. Blachman and E. D. Courant, Phys. Rev. 74, 140 (1948).

145. N. M. Blachman and E. D. Courant, Rev. Sci. Instrum. 20, 596 (1949).

146. J. M. Greenberg and T. H. Berlin, Rev. Sci. Instrum. 22, 293 (1951).

147. M. J. Moravcsik and J. M. Sellen, Rev. Sci. Instrum. 26, 1158 (1955).

148. E. D. Courant, Rev. Sci. Instrum. 24, 836 (1953).

149. S. A. Kheifets, Instrum. Exp. Tech. (Engl. Transl.) 6 (June), 873 (1961).

150. A. N. Didenko and V. A. Serdyutskii, Sov. Phys.Tech. Phys. (Engl. Transl.) 7, 679 (1963).

151. Y. F. Orlov and S. A. Kheifets, Sov. Phys.Tech. Phys. (Engl. Transl.) 7, 671 (1963).

< previous page page_79 next page >


< previous page page_81 next page >
Page 81

2
Flow of Gases through Tubes and Orifices

R. Gordon Livesey

The nature of gas flow in pipes and ducts changes with the gas pressure and its description is generally divided into
three parts or regimes. The flow dynamics are characterized by λ, the molecular mean free path, in relation to some
characteristic dimension such as the diameter of a pipe. The flow regime cannot be determined from the mean free path
alone but only from the relation of this parameter to the characteristic dimension. The relation is known as the Knudsen
number, defined as*

Three regimes are generally identified:

1. Free Molecular Flow. The mean free path is of the same order as, or greater than, the characteristic dimension (the
range of relatively large Knudsen numbers), and gas dynamics are dominated by molecular collisions with the walls of
the retaining vessel or pipe.

2. Continuum Flow. The mean free path is small compared with the characteristic dimension (the range of small
Knudsen numbers), and intermolecular collisions are much more frequent than wall collisions. In this regime the
properties of the gas

* In the literature the Knudsen number may be variously defined, for a cylindrical tube, as λ/d, d/λ, λ/R, or R/
λ.
Foundations of Vacuum Science and Technology, Edited by James M. Lafferty.
ISBN 0-471-17593-5 © 1998 John Wiley & Sons, Inc.

< previous page page_81 next page >


< previous page page_82 next page >
Page 82
(temperature, density, flow velocity) do not vary significantly over several mean free paths and the gas can be considered a continuous medium.
The gas dynamics are therefore described and analysed hydrodynamically. Flow in this regime is often referred to as viscous flow, although there
are circumstances (such as flow through short ducts) in which. viscosity plays no part.

3. Transitional Flow. The transition between continuum and free molecular flow occurs at intermediate values of the Knudsen number where
both wall collisions and intermolecular collisions are influential in determining the flow characteristics.

Expressed in terms of pressure and characteristic dimension the Knudsen number is

For air at 20°C, with d in mm and P in mbar we obtain

Table 2.1 shows the generally accepted range of Knudsen numbers for the three regimes. There is no sharp transition between the regimes, and
somewhat different values may be quoted by different authors. The gas factor Fg in the table, used to correct for different gases, is the ratio of
the mean free path for air to that of the gas under consideration (at the same pressure and temperature) and can be calculated from

Values of the gas factor for a number of common gases are shown in Table 2.2.

Applications of vacuum technology range from the lowest pressures attainable (< 1014 mbar) through to around atmospheric pressure, so that all
of the regimes described are of interest to workers in this field. One of the main aims of this chapter is to enable calculation of flow under as wide
a range of conditions as possible, so that a large number of flow equations is presented. The widespread availability of scientific calculators,
personal computers, and mathematical software means that some of the more cumbersome formulae are considerably less daunting than in times
past. However, the "back of an envelope" is still a much favored tool of scientists and engineers and rough calculations are often sufficient, so
that approximations will be given wherever possible. Equations for the molecular, continuum and transitional

Table 2.1. Flow Regimes versus Knudsen Number and Pressure


Regime Kn (λ/d) P (mbar), d (mm)
Molecular
Kn > 0.5 PdFg < 0.133
Transitional
0.5 > Kn > 0.01 0.133 < PdFg < 6.6
Continuum
Kn < 0.01 PdFg > 6.6

< previous page page_82 next page >


< previous page page_83 next page >
Page 83

Table 2.2. Properties of Some Common Gases at 20°C


Gas Relative Molecular Mass Viscosity Viscosity Ratio (Air/Gas) Fg
Pa·s × 106
H2
2 8.8 2.07 0.543
He
4 19.6 0.929 0.345
H2O (vapor)
18 9.7 1.88 1.48
N2
28 17.6 1.03 1.02
Air
29 18.2 1 1
O2
32 20.4 0.892 0.937
Ar
40 22.3 0.818 0.959
CO2
44 14.7 1.24 1.53

regimes are discussed in Sections 2.2, 2.3, and 2.4, respectively. For the most part, derivations are not given since there are textbooks and
papers where the basic theories are discussed extensively; several references are listed in each section for the reader interested in studying the
subject in more detail.

All equations are written at least once in the text in SI units; where numerical coefficients are given, the units used are stated.

Except for the discussion of adiabatic compressible flow, it is generally assumed throughout this chapter that isothermal conditions apply.

2.1
Flow Conductance, Impedance, and Gas Throughput

In the field of vacuum science and technology it is common practice to express gas flow rate as throughput in pressurevolume units. The symbol
is normally used and the throughput of gas at a particular pressure is then

If the volumetric flow rate is due to a pump

where S is the speed (or volumetric rate) of the pump at the pressure P.

The pumping speed available at a chamber will be affected by restriction due to connecting pipework. One of the most common problems in
vacuum technology is to estimate the loss in speed due to such restrictions (system design is covered in Chapter 9).

Knudsen [1] first introduced the notion of a pipe as an impedance or resistance in the electrical sense and Dushman [2] introduced the concept
of conductance, which is defined by the relation

< previous page page_83 next page >


< previous page page_84 next page >
Page 84

where Pu is the upstream pressure and Pd is the downstream pressure. These pressures normally refer to values in
(perhaps notional) plenums at the entrance and exit of a duct or a system fitting such as a valve. Gas flow conductance
is thus analogous to electrical conductance, with pressure difference being the analogue of voltage difference and the
analogue of current. The reciprocal of conductance (resistance or impedance, Z = 1/C) could equally well be used;
however, conductance has come into common usage in vacuum technology mainly because of its intuitive relation to
volume flow rate and pumping speed.

Applying this concept to a set of pipes or components in series, the net conductance is found from

The net speed of a pump in series with a component or pipe is found in a similar way:

In practice it is seldom quite so easy as these equations imply. Some care is needed with combinations of components,
and this is discussed in Section 2.2.10 in relation to molecular flow. In continuum flow, conductance depends in a
complicated way on the flow conditions, and a better approach is to calculate the pressure ratio (Kp) across a component
or series of components.

It is usually assumed that continuity applies through a system; that is, the throughput is the same through all sections.
This will be the case as long as sufficient time has elapsed (from opening a valve or starting a pump for example) and
there are no temperature differences between the points of interest. In many common situations, steady conditions are a
reasonable assumption. (Some cases of unsteady molecular flow are covered in Section 2.2.11.) If Pd is the inlet
pressure to a pump of speed S which is connected via a pipeline or component to a chamber, then, assuming steady
conditions, the speed (Sn) and pressure (Pu) at the chamber are simply related by

In this way, the net pumping speed can be found if the pressure ratio can be calculated.

From the definition of conductance

Dividing through by the downstream pressure Pd and rearranging gives


< previous page page_84 next page >
< previous page page_85 next page >
Page 85

This calculation is often easier than taking the reciprocal of a sum of reciprocals, and Kp is the factor by which the
pumping speed is reduced.

2.2
Molecular Flow

One of the fundamental assumptions in the derivations of molecular flow conductance is that molecules scatter from a
surface according to a cosine distribution. This is also referred to as diffuse or random scattering and means that there
are no favored directions. A scattered molecule has the same probability of emerging in any direction, and this is
unrelated to its direction of incidence. There are special circumstances in which nondiffuse scattering may occur, but for
microscopically rough surfaces the diffuse scattering law is well established theoretically and experimentally [35].

Figure 2.1 shows the distribution of molecules emerging from an aperture and from tubes of various lengths. The
lengths of the vectors are proportional to the number of molecules emerging in that direction. It is noticeable that the
longer the tube, the more heavily weighted is the emerging flux to the tube axis. This is an indication of what occurs
inside a tube. An observer close to the entrance (looking upstream) will see the entrance plane as a diffuse source. Deep
inside the tube an observer will see a perturbed flux which is peaked toward the axis. This is often referred to as the
beaming effect of a tube.

The aperture in Fig. 2.1 acts as a plane cosine emitter, and the molecular flux shows a spherical distribution. At small
angles to the plane the flux of molecules is reduced,

Fig. 2.1
Angular distribution of molecules
exiting tubes of various length-to-diameter
ratios. Reproduced with permission from
L. Valyi, Atom and Ion Sources, p. 86.
Copyright 1977, Akadémial Kiadó, Budapest.
< previous page page_85 next page >
< previous page page_86 next page >
Page 86

compared with larger angles. This is not because there is a preferred direction but is a consequence of the angle of view:
The emitting plane does not appear ''dimmer," simply smaller.

In the molecular regime, solution of gas flow problems can be reduced to finding the conductance of the elements
involved since conductance is independent of pressure or flow conditions. The derivation of conductance, by theoretical
or analyticostatistical methods, assumes that molecules arrive at the entrance plane of a duct from a chaotic gas, so that
the entrance plane effectively behaves as a diffuse (cosine) emitter. When vacuum components are connected in series,
this may not be the case and some correction is needed; this will be covered in Section 2.2.10.

Clausing [6] first introduced the concept of transmission probability, denoted by α. If N2 molecules arrive at the
entrance plane of a duct then the number of these which reach the exit plane is N2α, and N2(1 α) return to the entrance.
Similarly, of N1 molecules striking the exit plane (from a downstream chamber), N1α reach the entrance. The net flux
of molecules from entrance to exit is then (N2 N1)α. Although proportional to the pressure difference across the duct,
the net flux is not driven by a pressure difference; it actually consists of two independent fluxes, and there is not a flow
in the usual sense of the word. Some of the molecules which enter the duct will return to the entry plane after one or
more wall collisions. The flow dynamics are thus very different to the continuum flow case in which all molecules
crossing the entrance plane will leave the exit (apart from the possibility of back diffusion which can occur in some
circumstances).

Expressions for conductance are usefully formulated in terms of transmission probability, so that the conductance of a
duct (or other component) is given by the entrance aperture conductance multiplied by the transmission probability

2.2.1
Conductance of an Aperture

The molecular flow conductance of a thin aperture is directly related to the rate of impingement of molecules over the
aperture area A (discussed in Chapter 1):

where R0 is the universal gas constant, T is the thermodynamic (or absolute) temperature, and Mm is the molar mass (e.
g., 0.028 kg/mole for nitrogen).

This gives the familiar result that the molecular flow conductance of an aperture for air at 20°C is 11.6 liters per second
per square centimeter.

For air at 20°C, Eq. (2.15) can be written in the convenient form

and Table 2.3 gives values of the constant ka for several combinations of commonly used units.

< previous page page_86 next page >


< previous page page_87 next page >
Page 87

Table 2.3. Aperture Conductance for Air at 20°C in Various Units


Ca = kaA
Ca ka A
m3·s1
115.6 m2
liter·s1
11.56 cm2
liter·s1
0.1156 mm2
m3·h1
0.4163 mm2
cfm
0.245 mm2
cfm
158.1 in.2
liter·s1
74.62 in.2

2.2.2
General Considerations for Long Ducts

Molecular flow in long ducts was first studied experimentally and theoretically by Knudsen [7]. He deduced a general
relationship for a long duct of length l, varying cross-sectional area A and perimeter B, which can be written as

where va is the mean thermal velocity of molecules.

However, as discussed by Steckelmacher [8, 9], Eq. (2.17) gives the correct result only in the case of a long cylindrical
tube and leads to erroneous results for all other cross sections.

A correct expression was derived by Smoluchowski [10] which may be written

where ρ is a chord making an angle θ with the normal to the perimeter s. Expressions for a number of different cross
sections have been derived from Eq. (2.18).

2.2.3
General Considerations for Short Ducts
For small values of l it is clear that the long duct relation will give values for conductance which are too high. As the
length tends to zero the conductance apparently tends to infinity. Reasoning from the point of view that the entrance of a
duct can be considered as a vacuum circuit element with resistance Za = 1/Ca in series with the duct proper (regarded as
a "long" duct), of resistance Zml = 1/Cml, the net conductance is [applying Eq. (2.8)]

< previous page page_87 next page >


< previous page page_88 next page >
Page 88

Using this principle, the net transmission probability of a short duct then becomes

where al is the long duct transmission probability.

This principle was originally applied by Dushman [2] to short circular cross-sectional tubes as an approximate method
of correcting for the end effect. A similar logic is often applied to short ducts of other cross sections. The maximum
error for a cylindrical tube is about 12% (too high). Errors of this order are expected for blocky cross sections, but for
other cross sections the errors may be more serious. In the case of narrow rectangular ducts the errors can be greater
than 50%. The most accurate results are obtained using transmission probabilities which have been derived for a
number of different shapes either theoretically or via Monte-Carlo methods.

Transmission probability data for cylindrical tubes, from the results of Cole [11], are shown in Table 2.5. It is apparent
that the transmission probability for a unit length increases as the length of the duct increases. Consider a unit length l/d
= 1 for which α = 0.514. For two unit lengths (l/d = 2) the transmission probability, expected from the shorter length,
would be 0.257, whereas the actual value is 0.357. At 10 unit lengths the transmission probability is almost twice the
expected value. This reflects the effect of the random molecular distribution near the entrance compared with the
beamed distribution which evolves further down the tube.

2.2.4
Tube of Uniform Circular Cross Section

Long Tubes. The familiar expression for the conductance of a long cylindrical tube of diameter d was first presented by
Knudsen [7] in 1909:

It can be seen from Eq. (2.21) that the transmission probability for a long cylindrical tube is

Berman [12] derived the solution as an asymptotic expansion, the first four terms of which are

For this reduces to Eq. (2.22) (which needs significant correction for l < 50 d).

Table 2.4 lists values of long tube and aperture conductance for a number of gases.

< previous page page_88 next page >


< previous page page_89 next page >
Page 89

Table 2.4. Conductance of Long Cylindrical Tubes and Apertures for Air at 20°C
Gas Relative Molecular Cml (l/d3) liter · s1 Ca/A liter · s1 Ca/d2 liter · s1
Mass (mm) (mm) (mm)
H2
2 0.461 0.440 0.346
He
4 0.326 0.311 0.245
Air
29 0.121 0.116 0.0908
Ar
40 0.103 0.0985 0.0773

Short Tubes. Transmission probabilities for short tubes derived by Cole [11] are shown in Table 2.5. Berman [12] presented equations for the
direct calculation of α for any length.

Table 2.5. Transmission Probabilities for Cylindrical Tubes


α (Cole [11]) α [Eq. (2.20)] % Error
l/d
0.952399 0.963855
0.05 1.20
0.869928 0.898876
0.15 3.33
0.801271 0.842105
0.25 5.10
0.743410 0.792079
0.35 6.55
0.694044 0.747664
0.45 7.73
0.671984 0.727273
0.5 8.23
0.632228 0.689655
0.6 9.08
0.597364 0.655738
0.7 9.77
0.566507 0.625000
0.8 10.33
0.538975 0.597015
0.9 10.77
0.514231 0.571429
1 11.12
0.420055 0.470588
1.5 12.03
0.356572 0.400000
2 12.18
0.310525 0.347826
2.5 12.01
0.275438 0.307692
3 11.71
0.247735 0.275862
3.5 11.35
0.225263 0.250000
4 10.98
0.206641 0.228571
4.5 10.61
0.190941 0.210526
5 10.26
0.109304 0.117647
10 7.63
0.076912 0.081633
15 6.14
0.059422 0.062500
20 5.18
0.048448 0.050633
25 4.51
0.040913 0.042553
30 4.01
0.035415 0.036697
35 3.62
0.031225 0.032258
40 3.31
0.027925 0.028777
45 3.05
0.025258 0.025974
50 2.83
0.002646 0.002660
500 0.51

Defining

< previous page page_89 next page >


< previous page page_90 next page >
Page 90

we obtain

This gives results which agree with the Cole [11] data to within 0.13%.

Santeler [13] devised a simpler and more convenient formulation and calculates the transmission probability as

where le is an "equivalent length" and

This gives transmission probabilities with a maximum error of less than 0.7% relative to the Cole data.

2.2.5
Duct of Uniform Rectangular Cross Section

The convention used to denote dimensions of rectangular (and elliptical) ducts is as follows: a and b are the cross-
sectional dimensions, with b ≥ a, and l is the length in the direction of gas flow. Thus the cross-sectional area is A = ab.
To avoid any confusion, equations quoted from various authors have been recast to conform with this convention.

Long Ducts. The transmission probability, due to Smoluchowski [10] is

where δ = a/b and

A useful approximation is

This is accurate to better than 1% for aspect ratios (b/a) up to almost 100. The error increases slowly with aspect ratio
but is still only 1.9% and 2.4% for aspect ratios of 1000 and 10,000, respectively.

< previous page page_90 next page >


< previous page page_91 next page >
Page 91

Short Ducts. There appear to be no general expressions which cover the whole range of lengths and aspect ratios. Data are available from the
Monte Carlo calculations of Levenson et al. [14]. The results of Santeler and Boeckmann [15], which cover a greater range of lengths and
aspect ratios, are shown in Table 2.6 (for short ducts the original data are listed to six significant figures). Also shown for comparison in
Table 2.7 are some of the results of Cole [16], derived using a complementary variational method which gave upper and lower bounds for α;
the values listed are the means and are accurate to 1.2% or better.

Table 2.6. Transmission Probabilities for Rectangular Ductsa


l/a b/a
1 1.5 2 3 4 6 8 12 16 24
0.9902 0.9918 0.9926 0.9934 0.9938 0.9942 0.9944 0.9946 0.9947 0.9948
0.01
0.9807 0.9839 0.9854 0.9870 0.9878 0.9885 0.9889 0.9893 0.9895 0.9897
0.02
0.9626 0.9685 0.9715 0.9744 0.9759 0.9774 0.9782 0.9789 0.9793 0.9797
0.04
0.9370 0.9467 0.9515 0.9564 0.9589 0.9613 0.9625 0.9638 0.9635 0.9650
0.07
0.9131 0.9260 0.9326 0.9392 0.9425 0.9458 0.9475 0.9491 0.9500 0.9508
0.1
0.8428 0.8645 0.8757 0.8869 0.8926 0.8982 0.9011 0.9039 0.9053 0.9067
0.2
0.7334 0.7659 0.7829 0.8004 0.8093 0.8182 0.8227 0.8272 0.8295 0.8317
0.4
0.6178 0.6575 0.6793 0.7022 0.7140 0.7260 0.7321 0.7381 0.7411 0.7442
0.7
0.5363 0.5786 0.6026 0.6285 0.6421 0.6560 0.6631 0.6702 0.6737 0.6773
1
0.3780 0.4192 0.4444 0.4733 0.4893 0.5063 0.5150 0.5240 0.5285 0.5330
2
0.2424 0.2759 0.2977 0.3245 0.3404 0.3583 0.3679 0.3781 0.3833 0.3885
4
0.1596 0.1848 0.2020 0.2242 0.2380 0.2545 0.2639 0.2742 0.2796 0.2852
7
0.1195 0.1397 0.1537 0.1723 0.1843 0.1991 0.2078 0.2177 0.2230 0.2287
10
0.0655 0.0776 0.0864 0.0984 0.1066 0.1171 0.1238 0.1319 0.1366 0.1419
20
0.0346 0.041 0.0464 0.053 0.058 0.0652 0.0695 0.075 0.078 0.083
40
0.020 0.024 0.0275 0.032 0.035 0.039 0.042 0.046 0.048 0.052
70
0.014 0.017 0.019 0.023 0.025 0.028 0.030 0.033 0.035 0.038
100
a From Santeler and Boeckmann [15].
Table 2.7. Transmission Probabilities for Rectangular Ductsa
l/a b/a
1 5 10 20 50 100 1000 10000
0.53619 0.66722 0.68266
1
0.24233
4
0.11930 0.21280 0.2372 0.2400
10
0.11207
20
0.07234
40
0.04464
80
0.01438 0.0320 0.0438 0.0464 0.0468
100
0.0224
200
0.01295
400
a From Cole [16].

< previous page page_91 next page >


< previous page page_92 next page >
Page 92

For short slits (or large flat plates): Values of transmission probability were first calculated by Clausing [6]. Berman
[12] devised equations for the direct calculation of transmission probability to a greater accuracy than the tabulated
Clausing values.

For and and putting x = l/a we obtain

If (long, closely spaced slot, again with ), this equation simplifies to

2.2.6
Uniform Elliptical Cross Section

a and b are the minor and major axes,

Long Ducts. The expression derived from Eq. (2.18) by Steckelmacher [8] is

is the complete elliptic integral, and

Steckelmacher [8] has shown that, for the same aspect ratio and cross-sectional area, the expressions for a rectangular
duct (Eq. (2.26)] and an elliptical duct are in close agreement. The rectangular duct approximation was derived on this
basis, so that a similar approximation is available for the elliptical duct:
The constant is chosen to give the correct result for b = a (circular) and minimises the errors for practical aspect ratios
up to 10 (although the errors are less than 1% even for unrealistically large aspect ratios up to 1000).

< previous page page_92 next page >


< previous page page_93 next page >
Page 93

Generally for ducts which have cross sections intermediate between rectangular and elliptical we have

where αr is the rectangular duct transmission probability and

Short Ducts. No data are available for short ducts of elliptical or similar cross section. However, it is expected that the
similarity between long elliptical and rectangular ducts (of the same cross-sectional area and aspect ratio) will also
apply to short ducts. It is suggested that approximate transmission probabilities can be found from

For very short ducts this reduces to αs = αr since the transmission probability of an aperture is independent of shape.

2.2.7
Cylindrical Annulus (Flow between Concentric Cylinders)

Long Ducts.

where K(e2) and E(e2) are the complete elliptic integrals of the first and second kinds. X(e) is listed for a range of
values of e in Table 2.8.

Short Ducts. Table 2.9 presents the results of Berman [17], who calculated transmission probabilities over a large range
using the variational method. Berman also obtained an empirical expression which is more convenient for computer
calculation.

< previous page page_93 next page >


< previous page page_94 next page >
Page 94

Table 2.8. The Function X(e) for a Long Cylindrical Annulus [Eq. (2.31)]
e = d1/d2:
0 0.1 0.2 0.3 0.4 0.5
X(e):
1.3333 1.231 1.1238 1.0116 0.8942 0.7711
e = d1/d2:
0.6 0.7 0.8 0.9 0.95 1.0
X(e):
0.6416 0.5044 0.3576 0.1966 0.1071 0

Defining x = l/(d2 d1), we obtain

The expression is valid in the range 0 ≤ x ≤ 50 and 0 ≤ e ≤ 0.9.

2.2.8
Uniform Triangular Section (Equilateral)

Long Ducts.

Short Ducts. Approximate transmission probabilities can be obtained using the entrance correction principle [Eq. (2.20)].

2.2.9
Other Shapes
Transmission probabilities for a number of geometries are shown in graphical form in Figs. 2.2 [18], 2.3, and 2.4. It is worth
noting that the transmission probability for an

< previous page page_94 next page >


< previous page page_95 next page >
Document
Page 95

Table 2.9. Transmission Probabilities (× 104) for Cylindrical Annulusa


y = l/(R2 R1) R1/R2
0.25 0.5 0.75 0.9
0.1 0.2 0.4 0.6 0.8 0.95
0.5 8017 8022 8030 8037 8043 8046
1.0 6737 6754 6783 6808 6829 6842
1.5 5842 5867 5915 5958 5997 6020
5295 5365
2.0 5175 5206 5266 5324 5378 5413
2.5 4655 4690 4758 4826 4894 4926 4940
3.0 4237 4274 4348 4423 4501 4558
3.5 3893 3931 4007 4087 4174 4241
3661 3761 3872
4.0 3604 3642 3720 3804 3896 3972
5.0 3123 3181 3260 3347 3448 3507 3538
2948 3071
6.0 2791 2828 2906 2994 3100 3201
7.0 2513 2548 2625 2712 2820 2929
2339 2436 2559
8.0 2286 2321 2395 2481 2589 2704
9.0 2099 2132 2204 2288 2496 2515
2081 2200
10.0 1914 1973 2042 2124 2230 2304 2352
1819 1933
12.0
1617
14.0
15.0 1414 1440 1499 1569 1666 1740 1792
1381 1456 1559
16.0
1325
18.0
1216 1310
20.0 1404
25.0 921.7 941.1 984.5 1038 1116 1180 1230
30.0 1019
35 897
40 802
50 496.0 507.6 533.9 567.4 618.2 700.4
100 258.9 265.4 280.1 299.2 328.9 380.5
200 132.7 136.1 144.0 154.3 170.6 200.1
500 53.97 55.4 58.69 63.04 70 82.99
1000 27.15 27.88 29.56 31.77 35.34 42.09
104 2.733 2.807 2.978 3.204 3.57 4.273
105 0.2735 0.2809 0.298 0.3207 0.3575 0.4282
a From Berman [17].

< previous page page_95 next page >


< previous page page_96 next page >
Page 96

Fig. 2.2
Molecular transmission probabilities of an elbow,
from the results of Davies [18].

elbow is almost the same as two short tubes (with length measured at the inside of the elbow) connected by a large
volume. The effect of the elbow is to randomize, at least partly, the molecular distribution.

2.2.10
Combinations of Components

If two components, with transmission probabilities α1 and α2, are connected in series, then the usual method of
determining the net transmission probability is
< previous page page_96 next page >
< previous page page_97 next page >
Page 97

Fig. 2.3
Molecular transmission probabilities of a cylindrical
tube with restricted openings, from the results of Davies [18].

Consider two identical, short tubes, of transmission probability α, connected via a large volume V as shown in Fig. 2.5a.
The inlet of tube 2 and the outlet of tube 1 are also connected to large volumes and, for convenience, the downstream
pressure is taken to be zero. It is supposed that there is no beaming between the tubes, and the effect of the large volume
is to randomize the molecular distribution between the tubes.
The net flux of molecules through tube 2 is (N2 N1)α, and the net flux through tube 1 is N1α. Under steady conditions,
these must be the same, so that N1 = N2/2 and the number of molecules transmitted is N2α/2. The overall transmission
probability of the system is then α/2. The conclusion is the same if the transmission probabilities are combined as in Eq.
(2.38).

< previous page page_97 next page >


< previous page page_98 next page >
Page 98

Fig. 2.4
Molecular transmission probabilities of a chevron baffle.
Reproduced with permission from Levenson et al. [14]
Copyright 1963, Société Française d'Engenieurs
et Techniciens du Vide.

Now consider Fig. 2.5b, in which the two tubes (each with l/d = 1) have been brought together. The transmission
probability of each tube separately is (from Table 2.5) 0.514. However, the transmission probability for the joined tubes
(l/d = 2) is 0.357 and not 0.514/2 = 0.257.

Clearly the method of combining transmission probabilities for the joined tubes is incorrect. In this case every molecule
which crosses plane AA also crosses plane BB and vice versa, but this is not so when the tubes are separated by a large
volume.

Oatley [19] discussed the correct method of combining transmission probabilities, α1 and α2, for two joined tubes of
the same cross section and showed that the net transmission probability is given by
In the example above, this gives the overall transmission probability as 0.346, which is much closer to the correct result.
The Oatley method gives results with a maximum error of 5% or 6% for l/d ~ 2. It is, perhaps, surprising that the
method gives such good results, since the derivation assumes random gas entry into the second tube and thus ignores the
beaming effect. However, in short ducts the molecular distribution is

< previous page page_98 next page >


< previous page page_99 next page >
Page 99

Fig. 2.5
Combination of two short tubes.

not too seriously perturbed from a chaotic distribution, and in long ducts the entrance effect is relatively small.

The transmission probabilities for each of the two tubes, in effect, includes an entrance correction. Thus, when Eq.
(2.38) is applied, the overall transmission probability includes two entrance effects. This is correct when the two tubes
are separated by a large volume but not when they are joined and the Oatley method is equivalent to removing one of
the two corrections.

A typical case is illustrated in Fig. 2.6, in which a pump is connected to a chamber via a tube of the same size as the
pump inlet. The pump speed (S) has been measured, so that any entrance effects are already accounted for (at least in
principle). A pump can be regarded as a conductance with a transmission probability aH equal to its Ho coefficient (the
ratio of the pump speed to the conductance of the pump inlet aperture). If Ca is the pump and tube aperture conductance
and α is the tube transmission probability, then S = aHCa and the tube conductance C equals aCa. The net speed at the
chamber is then Sn = anCa, where

Alternatively,

The effect of this procedure is to remove an entrance correction.


< previous page page_99 next page >
< previous page page_100 next page >
Page 100

Fig. 2.6
Pump connected via a tube
of the same diameter.

If the pump speed is 300 liter · s1 and the connecting tube is 200 mm long and 100 mm in diameter, then Ca = 908 liter
· s1 (for air), α (for l/d = 2) = 0.357, and

which gives Sn = 188 liter · s1 [instead of 155 liter · s1 using Eq. (2.34)].

An addition theorem developed by Haefer [20] enables the calculation of multiple components of differing diameters.
The overall transmission probability of n elements α1n is related to the individual transmission probabilities ai and inlet
areas Ai by

The overall transmission probability α1n is expressed in terms of the inlet tube aperture.

Some cases of the application of this theorem will be discussed with reference to Fig. 2.7.

(a)
Series Arrangement of Tubes of Different Diameters (Fig. 2.7a). The overall transmission probability is
< previous page page_100 next page >
< previous page page_101 next page >
Page 101

Fig. 2.7
Combinations of components.

The diameter increases from tube 1 to tube 2 so that δ2, 1 = 0, but δ3, 2 = 1 since the diameter decreases from tubes 2 to
3. If l/d = 2 for each tube and the diameters are 15 mm, 25 mm, and 20 mm for tubes, 1, 2, and 3 respectively, then α1n
evaluates to 0.214. The aperture conductance of tube 1 (15-mm diameter) is 20.4 liter·1, so the conductance of the
arrangement is 0.214Ca1 = 4.38 liter·1. If the order of the tubes is reversed (20 mm, 25 mm, 15 mm), α1n now
evaluates to 0.121. The inlet aperture conductance (for 20-mm diameter) is now 36.3 liter·s1, so the overall conductance
is 0.121 × 36.3 = 4.38 liter·s1 as expected. The conductance of an arrangement cannot be changed by reversing the
order of components; but note that if the tubes were rearranged in ascending order of size, then α1n = 0.224. It is
generally the case that the highest conductance for a series of components is achieved when they are physically
arranged in order of size. In a calculation, the temptation to reorder the components for mathematical convenience
should be resisted since this can lead to incorrect results.

Equation (2.42) relating transmission probabilities can be expressed in terms of conductances:

which is more convenient if conductance values are given for components. δ has the same meaning as in Eq. (2.42).

The first summation in Eq. (2.44) contains the ''tube only" conductance Cmt discussed by Holland et al. [21]:

that is, a tube with its entrance correction subtracted. If the conductance quoted for a component is the "tube only" value
then the entrance correction should not be subtracted. A typical example would be a quarter swing valve which is not
designed for direct connection to a large chamber but is normally connected via a manifold. Similar remarks apply to
Eq. (2.42) if the tube only transmission probability is given.

< previous page page_101 next page >


< previous page page_102 next page >
Page 102

Either of Eqs. (2.42) or (2.44) can be used to find the net speed of a pump in series with a set of components.

(b)
Pump Connected to a Chamber via Two Tubes or Components (Fig. 2.7b).

Since Cm3 = αHCa3 = S, this reduces to

which can be written

Written in terms of "tube only" conductances, this becomes

The second term on the right-hand side of this equation can be seen as the total correction required to account for the
differing sections (remembering that a correction for A3 is inherent in the pump speed).

If, in Fig. 2.7b, tubes 1 and 2 and the pump all have the same aperture size of conductance Ca, then Eq. (2.48) becomes

Each tube has its entrance correction subtracted and only one correction, inherent in the pump speed, is applied.

(c)
Pump Connected to a Chamber Via a Second Chamber (Fig. 2.7c). This case is equivalent to making Ca2 and Cm2 very
large in Eq. (2.47), giving

In this case the entrance correction for tube 1 is retained.


2.2.11
Cases of Unsteady Flow

Consider a system in which a pump is connected to a vessel via a valve and a pipe of some significant length. It is clear
that, at the instant that the valve is opened, the throughputs at the pump and vessel must be

< previous page page_102 next page >


< previous page page_103 next page >
Page 103

different. Some time is required for continuity to be established. Usually this is very short compared with the timescale
for exhaust of the vessel, especially at continuum flow pressures where the pipe conductance is relatively high.
However, if the "vessel" is a long pipe, then steady conditions are never achieved.

Unsteady flow is of most interest under molecular flow conditions and is particularly relevant to filling, exhaust, or leak
testing of long pipelines. Mathematically, unsteady molecular flow in a constant section duct is analogous to heat
conduction in an infinite slab. Both can be treated as one-dimensional since there is no transverse flow. The cases
described here are intended only as a small sample to illustrate the kind of unsteady flow problems that can be solved.
Solutions for many heat conduction cases, which have practical parallels in molecular flow, are available in the
literature; Carslaw and Jaeger [22] is a good source of reference.

In the equations which follow, V is the volume and C the conductance of the pipe, assumed to be long. Given a relation
for pressure distribution, the throughput (from or into the pipe) can be obtained from the pressure gradient at the end of
a pipe:

Case 1. Pipe with uniform initial pressure P0, closed at x = 0. x = l opened to an environment maintained at constant
pressure Pe at t = 0.

This covers both exhaust and filling of a pipe. Lawson [23] discussed the application of this and similar cases to
pumping of trapped volumes and leakage.

The time constant for the first (slowest) term is

Contrast this with the time constant of chamber of volume V pumped at speed S or through a restrictive conductance C
(covered in Chapter 9).

For air at 20°C (l and d in meters) we have

and for helium we have


< previous page page_103 next page >
< previous page page_104 next page >
Page 104

A 1-meter-long pipe, 10 mm in diameter would have a time constant for helium of ~100 msec. For a pipe length of 10
meters, this increases to ~10 sec. Leak testing can be difficult with a detector at the downstream end of such long lines
since any leak will be located at the position of the helium probe several seconds prior to the indication on the leak
detector gauge.

The series converges quite rapidly and taking only the first term is a good approximation for t > 0.3τ0. Thus, taking
only the first term we obtain, at x = 0

and the throughput

Case 2. Pipe, with constant initial pressure P0, closed at x = 0 and pumped at the other end by a pump of constant speed
S, for t > 0.

where r = S/C and φn are the roots of

A good approximation for the first root is

accurate to within 3% in the worst case.

Taking only the first term of the series for throughput, we obtain

For the time constant becomes V/S and the case reduces to the simple pumpout of a large chamber.
Case 3. Pipe with zero initial pressure. Constant leak into the pipe at x = 0 and constant pumping speed S at x = l
for t > 0.

< previous page page_104 next page >


< previous page page_105 next page >
Page 105

This is again relevant to leak test of a long pipeline since the initial partial pressure of He will be zero.

where r = S/C and φn has the same meaning as in Case 2.

Taking only the first term, the throughput from the pipe is

2.3
Continuum Flow

In the continuum regime, calculation of gas flow through ducts is complicated by the different types of flow which can
occur. Flow may be broadly distinguished into two major types, referred to as viscous laminar flow and turbulent flow.
Since flow through a duct is driven by a pressure difference, all gas flow is compressible. There are circumstances in
which gas can be treated as incompressible, and this leads to considerable simplification of the equations describing
flow. However, there are also many circumstances in which compressibility cannot be ignored, so compressible flow
will also be discussed.

Continuum (or viscous) flow is often thought of as occurring at relatively high pressures. But consider air flowing
through a 100 mm (4 inch) diameter pipe. From Eq. (2.2) the Knudsen number Kn is < 0.01 (and hence the flow
continuum) down to pressures of about 0.1 mbar. Thus many vacuum processes will operate at pressures where
continuum flow conditions prevail.

At relatively low velocities, gas flows smoothly in stream lines, generally parallel to the duct walls, and the flow is said
to be laminar. In long ducts, viscosity of the gas is a controlling factor in the flow rate; this is not the case in short ducts,
although the flow may still be laminar. As the flow velocity is increased, there comes a critical point at which the flow
breaks up into turbulent eddies. These two types of flow, viscous laminar and turbulent, are described by different
equations. It is important to distinguish these flow types in calculation of flow rates; failure to do so can lead to wildly
inaccurate results. In short ducts, or in longer ducts at high flow velocities, compressibility becomes important and use
of incompressible flow formulae can also lead to serious errors.

The primary controlling parameter in the viscous behavior of Newtonian* fluids is the dimensionless Reynolds number

* Newtonian fluids are those in which the shear stress is proportional to the transverse velocity gradient. Most
common fluids (water, oils, and gases) are newtonian.

< previous page page_105 next page >


< previous page page_106 next page >
Page 106

where ρ is the density of the fluid, η is its viscosity, and u is the flow velocity. Dh is the hydraulic diameter* of the duct
(applicable to any cross section) defined by

where A is the cross-sectional area and B is the perimeter.

For a circular cross-sectional tube we obtain

so that the hydraulic diameter is simply the actual tube diameter.

For a concentric annulus we have

Note that, in calculating B, the perimeter of all "wetted" surfaces must be included.

It is useful to write Re in terms of the throughput:

In the case of a circular cross-sectional tube, B = πd. For air at 20°C, Eq. (2.67) can conveniently be written as

and Table 2.10 lists values for the units conversion kR for a variety of commonly used units.

As shown by the work of Senecal and Rothfus [24], the transition from viscous laminar to turbulent flow begins at
Recrit ~2000, although there is no sharp boundary, and flow is normally fully turbulent at a Reynolds number of 3500.
In very smooth pipes with well-rounded entrances, the transition to turbulence may be delayed to higher values of Recrit.

Using the value Recrit = 2000, the value of throughput for the onset of turbulent flow is

* Re defined in terms of diameter is the most commonly used convention. Some texts define Re in terms of
radius and make use of hydraulic radius rather than diameter. When making comparisons, note that hydraulic
diameter = 4 × hydraulic radius (yes, 4!).
< previous page page_106 next page >
< previous page page_107 next page >
Page 107

So for air at 20°C the flow will become turbulent if

Values of kT for a variety of units are given in Table 2.11.

For a cylindrical tube for example, with QT in mbar·liter·s1 and tube diameter in mm, we obtain

For example, in a 25-mm-diameter tube the conditions will be turbulent if the throughput exceeds 600 mbar · liter · s1.
If this tube size were used with a 20 liter · s1 pump, then turbulent flow would persist down to about 30 mbar.

In the continuum flow regime, conductance may be a function of pressure, pressure ratio, or rate of flow depending on
the type of flow and particular circumstances. The term conductance is useful as a conceptual toolto say, for example,
that the conductance of some tubulation is too small does provide a succinct means of characterizing a flow
conditionbut has limited practical value. Calculation of conductance is generally a means to an endcalculation of
throughput (when the end pressures are known) or calculation of pressure difference (when throughput is known), for
example. Thus, relations will be presented for throughput and pressure ratio where these can be expressed explicitly.

Table 2.10. Units Conversions for Reynolds Number and Throughput [Eq. (2.68)] for Air at 20°C

B kR
Pa · m3 · s1
m 2.615
mbar · m3 · h1
mm 72.64
mbar · liter · s1
mm 261.5
torr · liter · s1
mm 348.6
torr · liter · s1
in 13.73
torr · cfm
in 6.478
Table 2.11. Units Conversions for Turbulent Throughput [Eq. (2.70)] for Air at 20°C

B kT
Pa · m3 · s1
m 764.8
mbar · m3 · h1
mm 27.53
mbar · liter · s1
mm 7.648
torr · liter · s1
mm 5.736
torr · liter · s1
in 145.7
torr · cfm
in 308.7

< previous page page_107 next page >


< previous page page_108 next page >
Page 108

Before proceeding to discuss the equations of continuum flow, it is convenient to define

This collection of terms occurs frequently in equations of continuum flow. As is evident, the term is closely related to
Ca, the molecular flow conductance of an aperture [Eq. (2.15)], differing only by a numerical factor. The term therefore
has units of volumetric flow rate and provides a succinct means of expressing many continuum flow relations.

The long tube equations which will be presented for laminar and turbulent flow are of limited practical application.
There are few circumstances in which straight tubulation is employed of sufficient length to meet the long tube criteria
(to be discussed later). In vacuum technology, "short and wide" is beautiful! The long tube equations will be discussed
first since they form the basis for expressions which apply to ducts of any length. This will be followed by discussion of
the equations for compressible flow and then of approximations which are sufficiently accurate for many practical
purposes.

2.3.1
Viscous Laminar Flow

The simplest solutions for viscous flow, in ducts of constant cross section, are based on four assumptions: (1) The fluid
is incompressible; (2) the flow is fully developedthat is, the flow velocity profile is constant throughout the length; (3)
the flow is laminarthat is, in one direction only parallel to the duct axis and there are no turbulent motions; (4) the flow
velocity at the walls is zero.

The assumptions may appear restrictive but are true for newtonian fluids flowing in long ducts at relatively low
velocities. Except for the assumption of incompressibility, this also includes gases, but even for gases it can be shown
that compressibility can be ignored if

where Ma is the Mach number of the flow, defined as the ratio of the flow velocity to the local velocity of sound. The
generally accepted criterion is Ma < 0.3.

When fluid flows into the entrance of a duct, the flow velocity is approximately uniform over the entrance area. As the
fluid moves down the duct, shear stress due to viscous friction retards the flow near the walls and the fluid develops a
velocity profile which, after some entry length, becomes constant and the flow is said to be fully developed. In the case
of a circular cross section, for example, the profile is a parabola of revolution with the greatest fluid velocity at the tube
axis. The pressure drop over the entrance length is greater than that for fully developed flow because of the increased
shear stress and the kinetic energy needed to accelerate the flow. Shah and London [25] have given the following
correlation for entry length:

< previous page page_108 next page >


< previous page page_109 next page >
Page 109

The first term is significant only at very low Reynolds numbers, and the condition is usually expressed as

At the transition to turbulent flow, with Re = 2000, this gives an entry length of 112 diameters for a circular pipe. Since
the viscous laminar flow equations are valid only for fully developed flow, some correction is required unless the duct
length is much greater than the entry length. Corrections for entrance effects and long tube criteria are discussed in
Sections 2.3.4 and 2.3.7.

Viscous laminar flow has proved irresistibly attractive to theorists since the fundamental equations of flow can be
solved analytically; a wealth of solutions is available for an amazing variety of cross-sectional shapes. A few of the
more practical shapes will be discussed.

For incompressible fluids, equations are usually presented for volumetric flow rate which is proportional to the pressure
difference across the pipe. For gases, density varies along the duct, so equations are normally given for throughput in
terms of the average pressure which reflects the average density.

If Cv is written for the viscous flow conductance, then

where the conductance is proportional to the average pressure

and kV is a constant for the particular duct, containing the numerical factors, viscosity and geometric terms.

The throughput is then

Given a relationship for kV, it is no particular problem to determine the throughput if both end pressures are known. If
the throughput is given (because, for example, the downstream pumping speed at a given pressure is known), it is not so
easy to calculate the pressure drop because both pressures are required by Eq. (2.76). However, the equation can be
expressed as, for example,

where KP = Pu/Pd is the pressure ratio across the duct.

The term kVPd has the same units as conductance (i.e., volumetric flow rate); it can be thought of as the conductance of
the duct calculated at the downstream pressure

< previous page page_109 next page >


< previous page page_110 next page >
Page 110

(instead of the average pressure). In order to express flow equations in a form convenient for calculation, it is useful to
define two artificial conductances:

Cvu = viscous flow conductance calculated at the upstream pressure,


Cvd = viscous flow conductance calculated at the downstream pressure.

This artifice enables the equations of flow to be expressed in terms of quantities which are known.

When both end pressures are known, the throughput can be written as

If the downstream pressure and pumping speed (i.e., the volumetric flow rate) are known, then the pressure ratio across
the duct can be found from

If the upstream pressure and downstream pumping speed are known, we obtain

Cv, Cvu, and Cvd can be calculated, for different geometries, from the equations that follow, and Eqs. (2.78) to (2.81)
can be used to calculate throughput or pressure ratio.

Circular Cross-Sectional Tube. The HagenPoiseuille equation for a long circular pipe is probably the best-known
equation in viscous flow. The throughput is

hence the conductance is

For any gas, this can be written as


< previous page page_110 next page >
< previous page page_111 next page >
Page 111

Values of the units conversion constant kv are given in Table 2.12 for a variety of commonly used combinations of units.

It should be apparent from the previous discussion that, for a circular pipe for example, the artificial conductance Cvd is

Rectangular Cross Section.

Although only the first three terms are needed for good accuracy, this expression is rather cumbersome. A good
approximation is [21]

This expression, valid for all aspect ratios, is accurate for a = b and and it shows a maximum error of <3% at
intermediate aspect ratios.

Concentric Cylindrical Annulus.

Eccentric Cylindrical Annulus. It is worth noting that flow rate (for the same pressure difference) increases significantly if
the inner cylinder is offset from the axis. In the extreme case of a narrow annulus with maximum offset,
the flow rate is increased by a factor of 2.5. This explains the observation that flow rate can be very difficult to adjust with
an almost closed needle valve.

Table 2.12. Units Conversions for Viscous Flow Referenced to Air at 20°C [Eq. (2.84)]
Cv d l P kv
m3·s1
m m Pa 54,963
liter·s1
mm m mbar 5.496 × 103
liter·s1
mm mm mbar 5.496
liter·s1
mm m torr 7.326 × 103
liter·s1
mm mm torr 7.326
m3·h1
mm m mbar 0.01978
m3·h1
mm mm mbar 19.78
cfm
in ft torr 21,195
cfm
in in torr 254,363

< previous page page_111 next page >


< previous page page_112 next page >
Page 112

2.3.2
Turbulent Flow

Although particular solutions of the fundamental equations of fluid flow are known (viscous laminar flow, for example),
no general analysis of fluid motion has been successfully developed. The reason is the dramatic change in fluid behavior
which occurs at a critical Reynolds number when the flow becomes turbulent. The chaotic, fluctuating nature of
turbulent flow has defied theoretical analysis since it was first observed by Hagen in 1839. The flow equations which
have been developed are semiempirical, describing the gross mean properties of the fluid and ignoring small-scale
fluctuations. The DarcyWeisbach (1850) equation for head loss in a pipe can be expressed as

where ∆P is the pressure difference, u is the fluid velocity, and fD is the Darcy* friction factor. The friction factor fD is
not a constant (would that life were so easy!) but varies with the Reynolds number and is also a function of the cross-
sectional shape.

The Blasius relation holds for smooth pipes:

However, most real pipes are not smooth. Haaland [26] devised a useful general relation which takes surface roughness
into account. Jones [27] showed that the optimum friction factor correlation with Re for rectangular ducts is based on an
effective hydraulic diameter of Deff = (Dh/SF) which leads to an effective Reynolds number of Re/SF to be used in
calculating the friction factor. This approach is recommended by White [28] for all noncircular ducts. Haaland's
expression, modified to incorporate this proposal, is

where ε is the surface roughness, typically 0.0015 mm for the drawn tubing commonly used for system pipework.

SF is the shape factor, given by

and Ge is the numerical constant and cross-sectional geometric terms in the viscous flow conductance equations of
Section 2.3.1.

* The Fanning friction factor is also in common use in the literature on fluid flow. Darcy friction factor = 4 ×
(Fanning friction factor).

< previous page page_112 next page >


< previous page page_113 next page >
Page 113

Circular cross section:

Rectangular cross section:

Concentric annulus:

The shape factor for some cross sections is shown in Fig. 2.8.

Figure 2.9 shows (for circular cross section) the friction factor plotted against Re for three pipe diameters with ε =
0.0015 mm. Also shown for comparison is the Blasius smooth pipe relation. Larger-diameter pipes (d > 25 mm)
approach the smooth pipe curve for values of Re up to ~ 105, but there is significant deviation for smaller pipe sizes.

For viscous laminar flow the friction factor is a simple function of the Reynolds number:

For a circular tube we have SF = 1, so that fD = 64/Re.


Fig. 2.8
Shape factor for some cross sections.

< previous page page_113 next page >


< previous page page_114 next page >
Page 114

Fig. 2.9
Friction factor for a cylindrical tube.

< previous page page_114 next page >


< previous page page_115 next page >
Page 115

For a very narrow rectangular duct we have SF = 3/2, so that fD = 96/Re.

If the relation for fD is substituted into the turbulent flow Eq. (2.88), then the viscous laminar flow equations appear. It
would be convenient if fD were independent of cross-sectional geometry, because then laminar flow solutions for any
shape could be derived from Eq. (2.88). Unfortunately, the laminar flow solutions must be known in order to find the
shape factor.

For turbulent flow in a long duct, Eq. (2.88) can be rearranged to give the throughput as

Written in terms of Cz [defined by Eq. (2.72)], we obtain

Expressed in this latter form, it is immediately apparent that throughput equates to the product of a volumetric flow rate
term (Cz) and a pressure. In calculating throughput from this relation, it is only necessary to calculate Cz and express
the pressure in whatever units the user finds convenient. The terms under the square root sign are either dimensionless
or dimensionless ratios.

If throughput and downstream pressure are known (and hence the pumping speed or volumetric flow rate), then the
pressure ratio across the duct can be found from

Calculation is straightforward in this case; since the throughput is known, Re and hence fD are readily found. However,
the nature of the variation in friction factor with Re complicates calculation of throughput. A value of fD must be
chosen and calculated, Re can then be calculated, and a new value of fD obtained from Eq. (2.90). The process is
then repeated to obtain to the desired degree of accuracy. This procedure is painful for hand calculation, although it
is a simple matter to program on a personal computer. Often a single iteration will be sufficient and, if only a rough
estimate is needed, a constant value fD ~ 0.03 can be used. In vacuum pumping lines, Re seldom exceeds 105 and will
usually be less. Vacuum system pumping lines are often sized to avoid significant performance loss at the lowest
operating pressure. This usually means that losses are small at pressures where turbulent flow occurs, and accurate
calculation may be unnecessary. If accurate calculation is needed, then one simply has to go through the pain (or speak
to a friendly programmer). This is not the

< previous page page_115 next page >


< previous page page_116 next page >
Page 116

least of calculation problems that may be faced: It should be emphasized that the equations described thus far apply only
to long pipes without obstructions to the flow (such as bends and pipeline components) and to flow velocities of Ma <
0.3.

The entry length in turbulent flow is

At Re = 3000 for example, this gives an entry length of 17 diameters for a circular pipe. The criterion for a long pipe
will be discussed in Section 2.3.7.

2.3.3
Compressible Flow

Unlike other common fluids, such as water and oils, the flow of gases can involve significant changes in density. As
noted in the discussion on flow in long pipes, the effects can be ignored for small flow velocities (Ma < 0.3). At higher
flow velocities and in short pipes the effects of density changes become significant. Density is related to both pressure
and temperature, so that in analysis of gas flow the laws of thermodynamics must be considered in addition to the laws
of motion and continuity.

The flow relations will be discussed with reference to Fig. 2.10, which relates to an adiabatic process. When gas enters a
duct (from a large volume) the flow accelerates across the entrance, and this is associated with a fall in both temperature
and pressure. As the gas proceeds down the duct, pressure and temperature continue to fall and the gas continues to
accelerate, reaching its maximum velocity at the duct exit. If the pressure upstream of a duct is kept fixed and the
downstream pressure is reduced, then the flow rate will increase. As the pressure is reduced further, the flow rate
becomes constant when the pressure ratio across the duct reaches a critical value. Further reduction in the downstream
pressure will produce no further increase in rate of flow. This is due to the limiting velocity which can be achieved by a
gas flowing in a duct of constant cross sectionMach 1, the speed of sound.* In this condition the flow is said to be
choked (or blocked). The flow rate can be changed by changing the upstream pressure; but once the critical pressure
ratio is reached, the flow rate becomes completely independent of the pressure on the downstream side of the duct. The
term choked flow is used because the mass flow rate has reached the maximum value that it is physically possible to
achieve for a given upstream pressure and the duct behaves as though something is preventing, or choking off, any
further increase in flow. The phenomenon is particularly apparent with a small orifice restricting the inlet of a pump.
Once the pumping speed exceeds a certain value, the throughput will become constant and cannot be increased no
matter how large a pumping speed is used. Orifices, nozzles, and short or long ducts can become choked; the necessary
condition is that the flow velocity of the gas reaches the speed of sound. For any particular duct there is a specific value
of pressure ratio at which choking will occur; this critical pressure ratio is generally referred to as the choked pressure
ratio. In the case of an orifice or very short duct, the choked pressure ratio is independent of pressure or flow

* Supersonic velocity can be achieved downstream of the throat of a convergingdiverging duct.

< previous page page_116 next page >


< previous page page_117 next page >
Page 117

Fig. 2.10
Compressible flow through a duct.

conditions. In longer ducts the choked pressure ratio, because of its dependence on the friction factor, will change with
the pressure and type of flow (laminar or turbulent). Note that if the gas flow velocity in a duct does reach March 1, it
will do so at the duct exit. It is not physically possible for an initially subsonic flow (in a constant section duct) to reach
March 1 at any other point.

In the flow of gases through apertures or short ducts, or through longer ducts at high flow velocities, there is little or no
heat exchange between the gas and its surroundings and the process is approximately adiabatic. In longer ducts there
may be significant heat exchange and the flow may be approximately isothermal.

The thermodynamic theory of high-speed flow shows that isothermal conditions cannot be maintained at flow velocities
close to the limiting value because the required rate of energy input to the gas tends to infinity. Approximately
isothermal conditions may prevail for part of the flow through a duct but cannot be sustained if the flow becomes
choked. In long ducts the results of isothermal analysis and adiabatic analysis converge; in short ducts, adiabatic flow is
the most reasonable assumption. Some of the results of compressible flow analysis will be quoted without proof; the
basic theory is covered in a number of textbooks on fluid dynamics [2830].

Using results of thermodynamic analysis, it can be shown that throughput, velocity, and pressure at any point a along
the duct are related by

where Cz is defined by Eq. (2.72). The throughput is referenced to the stationary (or stagnation) conditions in the
upstream chamber (because of temperature changes, throughput is not constant through the duct).

For subsonic flow, the duct exit pressure (Px) must be the same as the downstream chamber pressure (Pd); so,
rearranging Eq. (2.100)

Since Max≤ 1, it is apparent that the flow will be choked if


< previous page page_117 next page >
< previous page page_118 next page >
Page 118

This is a general relation, true for a duct of any length. If the pumping speed S at the downstream end of a duct is
known, it can be immediately determined if the flow is choked. β is simply used as a convenient way of writing the
terms in γ.

It can be shown that the relationship between throughput and entry velocity is

and it can be shown further that the relationship between entry and exit velocity is given by

Choking is a property of this equation in that the variation of flow rate with Mach number exhibits a maximum at Max =
1.

For a finite length duct, the equations of compressible flow cannot be rearranged to give an explicit relation between
throughput and pressure. They can be solved for choked and nonchoked, laminar and turbulent flow, but complex
iterative procedures are needed. Sadly, envelopes (no matter how large) must be discarded and resort made to computer
programming. However, an approximate treatment, giving explicit equations accurate to a few percent, is covered in
Section 2.3.6.

It is instructive to examine how the rate of flow varies with duct length and an example is shown in Fig. 2.11. The
throughput (normalized to the throughput of an aperture) is plotted for a 10% pressure difference across a 12.5-mm-
diameter tube; flow conditions were arranged to be viscous laminar and turbulent. The normalized molecular flow
throughput is also shown for comparison. The striking difference

Fig. 2.11
Variation of normalized throughput with tube length for a
12.5-mm-diameter tube and a pressure difference of 10%.
< previous page page_118 next page >
< previous page page_119 next page >
Page 119

between the molecular and continuum flow cases illustrates the relative insensitivity to length of continuum flow in
short ducts. For larger pressure differences the reduction in continuum flow throughput with length is even smaller.

2.3.3.1
Flow through an Aperture or Short Duct

Since there is no frictional work and the process is adiabatic, flow through an aperture is often referred to as isentropic
flow.

For a duct of zero length, the right-hand side of Eq. (2.104) is zero and an obvious solution is Man = Max.

The critical pressure ratio for choked flow can be found by combining Eqs. (2.101) and (2.103) along with Man = Max
= 1:

The maximum throughput for choked flow, from Eq. (2.103), is

Typical values of Kpca, γ, and the function G(γ) are shown in Table 2.13 for several gas types. The values of γ are
typical of monatomic, diatomic, and so on, gases; but more accurate values can be found in Kaye and Laby [31], for
example.

For sharp-edged apertures the throughput is reduced by a factor of ~ 0.85 because the flow narrows to a vena contracta
which has a cross section smaller than the duct inlet area. This entry loss can be reduced to negligible proportions by a
radius on the entrance edge of ~ 0.2 diameters (or 0.2 × smallest dimension for noncircular ducts).

It is apparent from Eq. (2.106) that a choked aperture has a constant speed, given by

A choked aperture (or short nozzle) is often used a convenient means of providing a constant volumetric flow rate, or
speed, which is independent of pressure. Such a nozzle might be used to provide a 'soft start' pumpdown, to avoid
disturbances due to turbulence or pressure fluctuations in a process chamber during the initial stages of roughing. Any
disturbances occurring in the inlet ducting to the pump (downstream of the choke) will travel at the speed of sound; they
cannot travel upstream through the sonic choke and hence cannot be communicated to the chamber.
Table 2.13. Some Thermodynamic Properties of Gases
Gas Type γ Kpca G(γ)
Monatomic 0.7252
1.66 2.049
Diatomic 0.6847
1.4 1.893
Triatomic 0.6673
1.3 1.832
Polyatomic 0.6284
1.1 1.71

< previous page page_119 next page >


< previous page page_120 next page >
Page 120

For air at 20°C, Sa = 20A liter·s1, with A in cm2. Compare this with 11.6A liter·s1 for the molecular flow conductance
of an aperture.

The downstream pumping speed needed to choke an aperture is [cf. Eq. (2.102)]

For air at 20°C, Sca = 37.5A liter·s1, with A in cm2. Once the pump speed exceeds this value, the pumping speed at the
chamber will remain constant at Sa no matter how large the pump.

At pressure ratios smaller than the critical value, the throughput is given by

or, if the downstream pressure and pumping speed are known, the pressure ratio can be found from

The net speed at the inlet of the aperture will then be

There is no particular value in expressing the conductance of an aperture, since this varies with the pressure ratio. For
example, an aperture which is just choked (pressure ratio 1.89 for air) has a conductance of 42A liter·s1 (with A in cm2)
reducing to 20A liter·s1 for a very large pressure ratio The conductance C increases as the pressure ratio
is reduced and C →∞ as Kp→ 1.

For a choked aperture, it is interesting to note that for a given pressure difference the throughput is reduced by only 19%
with two apertures (of the same area) in series compared with a single aperture. In molecular flow the throughput would
be reduced by 50%

Temperature Changes. In adiabatic flow, the temperature at any point along a duct is

which indicates that there can be significant temperature changes in high-speed flow. For example, in the case of a
diatomic gas such as air, the temperature will fall from 20°C to 29°C if the gas reaches sonic velocity. This can cause
freezing up of choked nozzles due to condensation of water vapor initially present in the source air.

Time to Vent a Chamber Through an Aperture or Short Duct. This useful result can be derived from the choked and
nonchoked throughput relations given above. The
< previous page page_120 next page >
< previous page page_121 next page >
Page 121

source pressure (e.g., atmospheric) is assumed constant and much greater than the initial chamber pressure. The time for
the chamber pressure to equalize with the source pressure is then

where Cz is defined by Eq. (2.72) and Ca is the molecular flow conductance of an aperture [Eq. (2.15)].

The derivation assumes that the temperature of the gas, after entering the chamber, is the same as the source gas
temperature. In fact the gas temperature in the chamber will be greater, and this will reduce the equalization time.
Although the entering gas cools due to adiabatic expansion at the throat of the inlet aperture or nozzle, reheating occurs
as the gas comes to rest in the chamber. If the process were entirely adiabatic (i.e., no exchange of heat between the gas
and walls of the chamber), then the temperature rise (above the external ambient air temperature of T0 Kelvin) would be
T0(γ 1). For an ambient temperature of 293 K, this implies a temperature rise of about 100 K.

Practically, such large temperature rises are not observed due to heat exchange, especially since the thermal capacity of
the gas is much lower than that of the chamber. As an example, rapid venting (< 1 s) of a 22-liter chamber led to an
observed temperature rise of about 40°C. Greater temperature increases may be observed in larger chambers which are
vented rapidly.

2.3.3.2
Approximation for Flow through an Aperture

A simple approximation for nonchoked flow through an aperture can be obtained from Bernoulli's equation:

Taking downstream values for ρ and u leads to

or, if the downstream pumping speed is known,

The maximum errors occur at the choked pressure ratio and are 2.5% (γ = 1.66), 3.1% (γ = 1.4), 5.5% (γ = 1.3), and
10.9% (γ = 1.1). At lower pressure ratios the errors are less.

2.3.4
Corrections for Flow Obstructions

In most practical circumstances, gas and vacuum lines include obstructions such as bends, valves, or other components.
The basis for corrections is Bernoulli's equation

< previous page page_121 next page >


< previous page page_122 next page >
Page 122

and the observation that pressure head losses are proportional to the square of the mean flow velocity over the cross
section.

where nc is the loss coefficient, or number of corrections. This relation is commonly used to estimate head losses in
incompressible flow. The same principle is applied here for gases, with the velocity taken to be the mean value over the
length of the flow path. nc can apply to more than one component, and the corrections are simply summed for all the
components in the pipeline. The implication of taking the mean velocity is that the separate obstructions are regarded as
evenly distributed through the pipeline. It appears that the positioning of flow obstructions makes little difference to the
overall pressure loss. For example, with nc = 4 and maximum flow rate the variation in pressure loss is only about 2%
wherever the obstructions are located. This is likely to be smaller than the uncertainties in applying corrections.

For ducts longer than the entry length, the loss coefficient [32] is nc ~ 0.7 (flat plates) and 1.25 (cylindrical tube), for
viscous laminar flow. About half of this is due to excess shear over the entry length and the remainder due to the
additional pressure difference needed to accelerate a uniform flow into the developed velocity profile. This loss is in
addition to the kinetic energy needed to accelerate the flow from zero velocity. In ducts shorter than the entrance length,
correction is more complex; the subject is discussed in some depth by Shah and London [25]. The additional entrance
loss appears to be less important in turbulent flow. It is suggested to take nc ~ 1 to account for entrance losses. Note that
at high velocities in short ducts the kinetic energy needed to accelerate the flow is much more important than viscous
drag.

Table 2.14 lists suggested corrections to be used in the equations that follow. These are in addition to the kinetic energy
allowance built into the equations which thus assume that gas enters a duct from a large volume.

2.3.5
ApproximationsEntrance Correction Model

In a similar fashion to Dushman's original treatment of short ducts in molecular flow, entrance correction models
imagine a real duct to consist of an aperture in series with an ideal duct.

The total pressure drop is then taken to be the sum of the pressure drop across the entrance plus viscous losses in the
duct:

Pu Pd = (Pu Pi) + (Pi Pd) = entrance loss + viscous loss.

Table 2.14. Loss Coefficients Due to Flow Obstructions


Obstruction nc
Sharp-edged entrance
0.5
Mitred (90°) elbow
1
Standard (90°) elbow
0.8
Tee (used as elbow)
1
Tee (in-line flow)
0.25
Right-angle valve (fully open)
3
< previous page page_122 next page >
< previous page page_123 next page >
Page 123

Santeler [33] discussed a similar model but with the loss considered as an exit loss. For nonchoked flow, as shown
previously, it makes little difference where the correction is placed. Applying the isentropic equation for an aperture to
this model, the variation in calculated pressure difference between aperture-at-entrance and aperture-at-exit is
(for any length of duct). For a duct with a choked exit the Santeler model makes more intuitive sense since an entrance
cannot be choked.

In continuum flow this model is equivalent to allowing for the kinetic energy required to accelerate the gas to the mean
flow velocity:

As discussed previously, this can be generalized for any number of corrections:

Using this expression the following equations can be derived.

Viscous Laminar Flow. When the two end pressures are known the throughput can be found from

When the throughput and downstream pressure are known we obtain

Turbulent Flow. When the two end pressures are known we obtain

When the throughput and downstream pumping speed are known we obtain

These equations will be found useful in many practical circumstances. Strictly speaking, they should not be used for
choked flow or for very short pipes. In this case errors in calculated throughput can be up to around 50% (for an
aperture), although even this may be acceptable for a rough estimate. These equations do not show
< previous page page_123 next page >
< previous page page_124 next page >
Page 124

a maximum with pressure ratio; that is, the equations do not exhibit a choke property unlike the thermodynamic
equations and those which follow.

2.3.6
ApproximationsKinetic Energy Model

The main deficiency of the previous approximations is the failure to take proper account of the kinetic energy acquired
by the gas. The entrance correction approach is valid for an incompressible fluid which must reach its final velocity at
the duct entrance. If the fluid density is constant and the mass flow rate through the length of the duct is constant, there
can be no change in velocity. In contrast, a gas flowing in a duct reaches its final velocity at the duct exit. Thus, the
kinetic energy accounting must relate to the exit velocity.

In this case, the viscous loss term relates to mean density and velocity, but the kinetic energy correction applies to
density and velocity at the duct exit.

The flow equations derived from this expression show a maximum in the throughput with pressure ratio. In other words,
the equations exhibit the property of choking in the same way as the compressible flow relation. Unfortunately, an
explicit expression cannot be found for the critical pressure ratio, and the value of approximate equations is nullified if
iterative methods are required for solution. However, with a little analytical trickery, a solution is found by borrowing a
result from the thermodynamic theory, namely

obtained from Eqs. (2.101) and (2.103) with Ma = 1.

A set of equations can then be derived covering both nonchoked and choked flow including allowance for flow
obstructions. Iteration is needed to find the friction factor for turbulent flow when the two end pressures are known and
the throughput is to be calculated. Otherwise, all the relations are explicit and iteration procedures are not required.
These equations give calculated flow rates and pressure differences to within 7% of values predicted by the
thermodynamic equations.

Viscous Laminar Flow. (a) With both end pressures known, the choked pressure ratio is given by

If Kp > Kpc, so the flow is choked, then the throughput can be found from Eq. (2.123).

< previous page page_124 next page >


< previous page page_125 next page >
Page 125

If the flow is not choked, then

(b) If the throughput and downstream pressure are known, then the test for choked flow is [Eq. (2.102)]

S ≥βCz.

If the flow is choked, then the choked pressure ratio in terms of the known downstream conditions is

and the upstream pressure can then be found from

If the flow is not choked, then the pressure ratio is

Turbulent Flow. The choked pressure ratio is

(a) Both end pressures known: If the flow is choked, then the throughput is given by [Eq. (2.123)]

If the flow is not choked, then the throughput is


< previous page page_125 next page >
< previous page page_126 next page >
Page 126

(b) Throughput and downstream pressure known: The flow will be choked if [Eq. (2.102)]

If the flow is choked, then the upstream pressure can be found by rearranging Eq. (2.123):

If the flow is not choked, then the pressure ratio is

It is interesting to consider the case of flow from atmospheric pressure through a hole, of diameter 0.1 mm and length
10 mm, in a vacuum chamber. This might be, for example, a pinhole leak in a weld. With a pressure of 103 mbar in the
chamber, it is generally assumed that the flow changes from viscous through transitional to molecular at the vacuum end
of the hole. The flow is most likely to be viscous laminar, so Eq. (2.124) can be used to calculate the choked pressure
ratio. In this case, Cz = 2.276 × 103 liter·s1 [Eq. (2.72)], Cvu = 1.349 × 103 liter·s1 [Cv of Eq. (2.83) but calculated at
the upstream instead of the average pressure], and (for γ = 1.4) β = 1.296 [Eq. (2.102)]. Putting these values into Eq.
(2.124) gives Kc = 5.55, so the pressure at the hole exit is 160 mbar and the flow is continuum through the whole
length. The throughput [from Eq. (2.123)] is calculated to be 0.532 mbar·liter·s1; turbulent flow [from Eq. (2.71)]
would require mbar·liter·s1, so the flow conditions are viscous laminar as assumed. An almost identical
result is obtained from solution of the thermodynamic equations.

2.3.7
Long Duct Criteria

In setting out the simple equations for turbulent and viscous laminar flow, it was said that these were applicable only to
long ducts. This raises the obvious question as to what constitutes ''long." This is most easily answered by considering
the approximate equations derived from the kinetic energy model. Equation (2.130) is quite general, covering both
turbulent and viscous laminar flow (with appropriate choice of the friction factor). A duct can be considered long if
viscous drag is the dominant effectin other words, if the viscous drag term greatly exceeds the kinetic energy term so
that kinetic energy losses can be ignored. The long duct condition is then

The maximum pressure ratio is achieved when the flow is choked. Substituting the expression for the choked pressure
ratio [Eq. (2.129)] in this inequality leads to

< previous page page_126 next page >


< previous page page_127 next page >
Page 127

Interestingly, the right-hand term of this inequality is the (approximate) choked pressure ratio for an aperture.
Depending on the value of γ, this then gives

Low-speed flows imply a small pressure difference, Kp ~ 1, so that

Substituting for the friction factor fD in Eq. (2.134) gives

For example, with Re = 250 the criterion gives l 7.8d, meaning that the tube should be at least 78 diameters long
(taking "much greater than" to mean a factor of 10). The case is illustrated in Fig. 2.12, for which a set of pressures was
chosen (using the thermodynamic equations) to maintain constant throughput in a 10-mm-diameter tube under choked
viscous laminar flow conditions. Throughputs were then calculated using the entrance correction [Eq. (2.118)] and
kinetic energy [Eq. (2.123)] models and the uncorrected Poisseuille equation [Eq. (2.82)]. Even at l = 78d the
Fig. 2.12
Comparison of throughput calculated using various equations
and illustrating the long duct criterion.

< previous page page_127 next page >


< previous page page_128 next page >
Page 128

error in the Poisseuille values is still 19%, so the criterion is, if anything, rather conservative. The entrance correction
model gives improved accuracy and does not give the wildly inaccurate values for smaller lengths. The kinetic energy
model gives values within a few percent of the correct value over the whole range.

Thus, for viscous laminar, high-speed flows the long duct criterion is

For air at 20°C, flowing in a cylindrical tube the criterion can be expressed as

For high-speed flows, the likely error is still about 20% and the length would need to be increased by a factor of 2 to
achieve errors of less than 10%. For low-speed flows which also meet the incompressible criterion (Ma < 0.3), the error
in calculated values should be within 10%.

For high-speed, turbulent flow the long tube criterion is

The friction factor varies from 0.04 at Re = 4000 to 0.02 at Re = 50,000, giving l > 500Dh to 1000Dh. For low-speed
flows the values are halved; that is, l > 250Dh to 500Dh. With these criteria the error in calculated values should be
within 10%.

The conclusions to be drawn from this analysis are that a "long" duct is much longer than is commonly thought and that
is generally unsafe to use the simple uncorrected viscous and turbulent flow equations.

2.4
Transitional Flow

In the transition regime, gas flow dynamics are intermediate between free molecular flow and continuum flow. In very
short ducts transition occurs between molecular and isentropic flow and in long ducts between molecular and viscous
laminar flow. The flow cannot be turbulent at or near transitional flow pressures as is easily shown.

The maximum flow velocity possible is equal to the speed of sound, and the Reynolds number at this velocity is

Combining this with Eq. (2.2) for the Knudsen number Kn gives

< previous page page_128 next page >


< previous page page_129 next page >
Page 129

Kn = 0.01 marks the high-pressure boundary of transitional flow, so the maximum possible value of Re is ~ 160, much
less than the value (~ 2000) at which turbulent flow occurs.

Molecular flow analysis is concerned with the effect of bounding walls on the free flight of individual molecules, while
continuum flow analysis is based on hydrodynamic and thermodynamic considerations. It is clear that an analytical
unification between the regimes presents considerable difficulty, but a number of attempts have been made. A number
of analyses based on numerical solution of the Boltzmann equation have been described [3436]. Since, at the most
fundamental level, gas flow dynamics is determined by molecular interactions, attempts [37, 38] have been made to
extend Monte Carlo methods into the transition regime by taking account of moleculemolecule collisions. However,
these methods require considerable computing power even for the simplest geometries. Scherer-Abreu and Abreu [39]
developed a probabilistic three-dimensional model requiring more modest computing power and obtained good
agreement with published results.

2.4.1
Transitional Flow in Long Ducts

In spite of the significant amount of work, there have been no general derivations of flow equations which are entirely
based on first principles. The state of theory was reviewed by Thompson and Owens [40] who discuss, in particular, slip
theory and empirical methods of obtaining an equation for the total flow regime. In viscous laminar flow, a stationary
layer of fluid is assumed to exist adjacent to the duct walls. Slip theory supposes that the velocity of this layer is
nonzero, essentially due to a degree of specular reflection of molecules at the surface. One of the equations for a long
tube, discussed by Thompson and Owens, derived from slip theory can be written as

where Cm is the long-tube molecular flow conductance [Eq. (2.14) or (2.21)] and δ is the fraction of molecules diffusely
scattered at a surface. For large values of Kn the conductance tends to the molecular flow value, and for small values of
Kn the first term inside the brackets becomes dominant and the conductance becomes equal to the Poiseuille, viscous
flow conductance [Eq. (2.83)]. The best correlation with experimental data (for glass and copper tubes) is achieved with
δ = 0.84.

There is considerable evidence that δ ~ 1 under molecular flow conditions, and there is no evidence of slip in the
viscous regime (the Poiseiulle equation has been verified to a high degree of accuracy). Equation (2.141) has the
character of a linear combination of viscous and molecular flows with a suitable weighting function to achieve
agreement with observation. The theory of slip flow is not entirely consistent, and there is some question whether slip
actually occurs; the concept was introduced primarily as a means of extending solution of the NavierStokes equation
into transitional flow. Slip theory does not appear to give any better results than empirical methods.

< previous page page_129 next page >


< previous page page_130 next page >
Page 130

Knudsen [7] was the first to develop an empirical expression for transitional flow in a long circular cross-sectional tube,
which can be written as

where Cv is the Poiseuille long-tube conductance [Eq. (2.83)], Cm is the molecular flow conductance [Eq. (2.21)], and
XK is given by

Expressed in terms of the Knudsen number, we have

With pressure in millibars and diameter in millimeters, Eq. (2.142) becomes

where the first term inside the brackets is Cv/Cm and Fg is the gas factor defined by Eq. (2.4). At low and high
pressures the conductance tends to the molecular and viscous conductances respectively, as expected.

Table 2.15 shows values of the conductance (as a ratio to the molecular flow conductance) and the term Z1 for a range
of values of Kn together with the average

Table 2.15. Ratio of Conductance of Cylindrical Tube (Ct) for That for Molecular Flow (Cm) as a
Function of Knudsen Number
Pmean (mbar) Z1 Ct/Cm
Kn for d = 25 mm
0.809
104 26.44 737.1
0.810
103 2.644 74.44
0.811
0.01 0.264 8.174
0.821
0.1 0.0264 1.557
0.831
0.2 0.0132 1.199
0.856
0.5 5.289 × 103 1.003
0.884
1 2.644 × 103 0.958
0.905
1.55 1.706 × 103 0.952
0.917
2 1.322 × 103 0.954
0.955
5 5.289 × 104 0.970
0.974
10 2.644 × 104 0.982
0.986
20 1.322 × 104 0.990
0.994
50 5.289 × 105 0.996
0.997
100 2.644 × 105 0.998

< previous page page_130 next page >


< previous page page_131 next page >
Page 131

pressure (for air at 20°C) in a 25-mm-diameter tube. For Kn < 0.01 the flow is almost entirely viscous, whereas for Kn
= 0.5 the conductance has fallen to the molecular flow value. This is the justification for the values of Kn, delimiting the
flow regimes, discussed in the introduction.

The pressure versus conductance results are also plotted in Fig. 2.13 for a 1-meter-long, 25-mm-diameter tube. At low
pressures the conductance passes through a shallow minimum; the slope then rises through the transition region,
increasing to 45° in the viscous laminar flow region when conductance becomes proportional to pressure. At pressures
above 10 mbar the conductance has increased to more than 1000 liter·s1 and it might be thought that this size of pipe
could be used with a 50- or 100-liter·s1 pump with little loss in speed. However, Fig. 2.13 also shows the effect of using
this pipe with a 50-liter·s1 pump. At pressures ~ 1 mbar, the conductance begins to deviate from the long tube
conductance and flattens sharply at around 10 mbar as the flow becomes turbulent. At 100 mbar the conductance is ~
380 liter·s1 rather than ~ 5000 liter·s1. If a 100 liter·s1 pump were used, the conductance would be reduced to ~ 200
liter·s1. Even at a length of 40 diameters the pipe is not "long," as discussed in Section 2.3.7, and it cannot be assumed
that conductance always continues to increase with pressure.

As an illustration of the application of the Knudsen equation, it is of interest to calculate the rate of leakage of air from
atmosphere through a small hole into a vacuum chamber at 20°C. The pressure in the chamber is assumed to be much
lower than atmosphere (say, 1000 mbar), so the mean pressure is 500 mbar. Consider two sizes of holes:

1. l = 1 mm, d = 2 × 103 mm. The Knudsen number is 0.066, so flow is in the transition regime. The aperture
conductance (from Table 2.3) is 3.63 × 107 liter·s1 and α = 0.00265 [Eq.(2.25)], so Cm = 9.64 × 1010 liter·s1.
Applying Eq. (2.145) gives the conductance as Cm × 1.93 = 1.86 × 109 liter·s1, so the throughput is 1.86 × 106
mbar·liter·s1.

2. l = 1 mm, d = 0.02 mm. The Knudsen number is now 0.0066, so the flow is still transitional. Cm = 9.24 × 107
liter·s1. Applying Eq. (2.145) gives the conductance as Cm × 11.95 = 1.1 × 105 liter·s1, so the throughput is 0.011
mbar·liter·s1. To maintain a pressure of 104 mbar with this leak rate would require a pumping speed of 110 liter·s1.

As will be observed, the ratio of transitional to molecular flow conductance is greater for the larger-bore capillary.
These examples serve to illustrate the fact that, even for a very fine hole, the rate of leakage can be substantial.

The conductance minimum observed by Knudsen at intermediate pressures (Kn ~ 1.6) may not always occur at this
value and may be entirely absent. Pollard and Present [41] have offered a qualitative explanation for the minimum and
have shown that it should depend on the length of the tube. When the pressure is sufficiently low that the mean free path
is much greater than both the diameter and the length of the tube, subpopulations of molecules occur which have near
axial velocity components on entry to the duct or after scattering from a wall. These molecules, which can travel great
distances before a further wall collision, make a disproportionately large contribution to the transmitted flux. As the
pressure is increased, the mean free path

< previous page page_131 next page >


< previous page page_132 next page >
Page 132

Fig. 2.13
Variation of long-tube conductance with pressure and effect of
50 liter·s1 pump for a 25-mm-diameter, 1-meter-long tube.

< previous page page_132 next page >


< previous page page_133 next page >
Page 133

becomes smaller relative to the tube length, and the paths of these molecules are disrupted by intermolecular collisions.
At the same time, the effect of the increased number of intermolecular collisions is to initiate an overall drift velocity.
Pollard and Present reasoned that, since the development of continuum properties (and hence a drift velocity) depends
on λ/d, the decrease in flow due to the shortened mean free path will outweigh the increase due to a drift velocity until
the mean free path is more closely comparable with the tube diameter than with its length. The implication of this model
is that a conductance minimum is likely to occur for long tubes but may not occur for short tubes or, at least, will be less
pronounced.

The conductance minimum is evident in the computed results of Sharipov and Seleznev [42] who presented tabulated
values of normalized flow rate for a range of Knudsen numbers, based on solution of the Boltzmann equation. The
tabulated values can be used in calculation of conductance or throughput over the transition region and give results
which are a maximum of about 6% lower (at round Kn ~ 1) than those given by the Knudsen equation.

The term Z1 in the Knudsen equation varies between 1 and 0.81, depending on the pressure. A simple approximation to
the Knudsen equation takes Z1 = 1, so the equation becomes a straightforward summation of viscous and molecular
flows:

Calculation of throughput is then straightforward, using

If the downstream pumping speed and pressure are known, then the pressure ratio can be found from

There appears to be little information on transitional flow in ducts of noncircular cross section. Dong and Bromley [43]
discussed transition and slip flow and developed empirical equations for rectangular and annular cross sections,
although it is difficult to know how to interpret their results or to say whether the empirical relations are applicable to
ducts with lengths and aspect ratios different to those used in their experiments.

It is therefore suggested that the Knudsen Eq. (2.142) [or the simple approximation of Eqs. (2.146) to (2.148)] be used
as a rough approximation for transitional flow in other cross sections, taking Cv as the viscous conductance for the
particular duct shape.

An alternative to the Knudsen equation is

Like the Knudsen equation, this also exhibits a minimum in the transition region, although significantly deeper. This
may be more appropriate for narrow rectangular

< previous page page_133 next page >


< previous page page_134 next page >
Page 134

ducts which are known to show a more pronounced minimum than circular cross sections [43]. The basis of this
formulation will be covered in Section 2.4.3.

2.4.2
Long Duct Criterion in Transitional Flow

For continuum, high-speed flow, it was shown previously [Eq. (2.132), Section 2.3] that the worst-case condition for a
long tube is

so that, for a circular cross-sectional tube in laminar flow we have

Combining this with the result of Eq. (2.140) for the maximum Reynolds number leads to (for high speed flows)

At the high-pressure limit of transitional flow (Kn ~ 0.01, at the duct exit), this gives

depending on the value of γ. As discussed under continuum flow, this suggests llong ~ 40d even at sufficiently low
pressures that conditions at the duct exit are close to transitional. Since the Knudsen equation is a combination of
viscous and molecular flows, it is expected that a similar criterion is applicable through the regime. This implies that
llong becomes smaller as the pressure is reduced and flow conditions become transitional (Kn > 0.01). At molecular
flow pressures the kinetic energy term must vanish, so the criterion becomes irrelevant.

Sharipov and Seleznev [42] showed that the conditions of applicability of their solution of the Boltzmann equation are

where Kn is the Knudsen number at the mean pressure and ∆P is the pressure difference across the tube. The results of
this solution agree well with those calculated using the Knudsen equation. Thus, these conditions should also apply to
the Knudsen

< previous page page_134 next page >


< previous page page_135 next page >
Page 135

equation and to other empirical transitional flow formulations which are based on a combination of viscous and
molecular flows. These conditions imply that solutions are valid for relatively short tubes if the pressure difference is
small. However, one of the main assumptions underlying solutions of the Boltzmann equation is that the tube is
sufficiently long that end effects can be neglected and the flow considered as one-dimensional. This suggests that l
should not be less than about 20d even well into the transitional flow regime. It is difficult to be more specific because
of the dearth of experimental and theoretical information on transitional flow in short ducts.

2.4.3
Transitional Flow through Apertures and Short Ducts

Transitional flow through a thin slit was studied by Kieser and Grundner [44], who gave an empirical fit to data for air
at 20°C, which can be written as

where Ca is the molecular flow aperture conductance, a is the short dimension of the slit, λ is the mean free path at the
mean pressure, k1 = 0.5, and k2 = 0.3412. In their experiments, ; thus the equation only describes the transition
from molecular to continuum, choked flow. The authors noted fluctuations of the measuring points in the high-pressure
range, which they attributed to small alterations of flow pattern as a function of pressure. The equation gives the
maximum flow rate as 85% of the theoretical isentropic flow rate through an aperture [Eq. (2.106)], which is typical of
thin, sharp edged apertures.

As noted earlier (Section 2.3.3.1), the theoretical flow rate can be achieved with a small radius on the inlet edge of finite-
length ducts. It is suggested that the equation can be generalized for any gas and for apertures with radiused edges, with
k1 = 0.5 and

so that, at high pressures, Eq. (2.155) will give the theoretical value for an orifice.

Kieser and Grundner [44] also studied flow, at large pressure ratios, through rectangular ducts. They regard the duct as a
series combination of entrance aperture and the duct itself. In the modified form suggested by O'Hanlon [45] the flow
rate can be found from

< previous page page_135 next page >


< previous page page_136 next page >
Page 136

where Ct is essentially the Knudsen transitional flow conductance, given by Eq. (2.142) with Cv as the viscous flow
conductance for a rectangular duct. As for the aperture, this equation is valid only for the case when .

Santeler [33, 46] suggested the following formulation for transitional flow through an aperture (again valid only for
large pressure ratios):

where the subscripts ma and 0c refer to molecular flow [Eq. (2.15)] and isentropic flow [Eq. (2.106)] respectively, and θ
is a weighting function which can be written, in terms of Kn at the mean pressure, as

The best fit to the results of Eq. (2.155) is obtained with ks ~ 12 (Santeler used ks = 28). In a fashion similar to that of
Eq. (2.19), short ducts are considered to consist of an ideal duct in series with an exit aperture. The total pressure drop is
then due to purely viscous losses in the duct plus the pressure drop across the transitional/isentropic aperture.

If the formulation of Eq. (2.158) is applied to a finite length duct, then

where are the molecular and viscous flow throughputs. For a long circular section tube, it is interesting to
note that

If ks = 128/3π is taken as the basis for the weighting factor θ, the result is Eq. (2.149).

An expression, due to DeMuth and Watson [47], for nonchoked transitional flow through orifices is

where [Eq. (2.109)] is the isentropic and ma the molecular flow throughput of an aperture. The Knudsen number
is calculated at the mean pressure. The equation is based on studies of air flowing through apertures at pressure ratios
ranging from 1.1 to 1.4. Since the authors appear to define the Knudsen number as R/λ, their equation has been
modified to conform with the definition in this chapter.

For large pressure ratios, the duct + isentropic exit (or entrance) models appear to give good results for short ducts. This
is principally because high-speed flow through short ducts is very insensitive to length (as illustrated in Fig. 2.11) and
the isentropic aperture conductance is independent of pressure. But this type of model breaks down at lower pressure
ratios. As the pressure ratio is reduced, the isentropic aperture conductance increases and tends to infinity at vanishingly
small pressure ratios. The
< previous page page_136 next page >
< previous page page_137 next page >
Page 137
isentropic term, in Eq. (2.158) for example, then makes a disproportionately high contribution to the total flow. For example,
consider a pressure ratio Kp across an aperture. Using the approximation of Eq. (2.114) for continuum flow through an aperture and
dividing Eq. (2.158) by the pressure difference gives the conductance as

For a small pressure difference (say Kp = 1.01) and with Kn = 1 (molecular flow) so that θ = 0.93 [Eq. (2.159) with ks = 13], this
gives

The enhancement factor from Eq. (2.162) is 1.83 for the same conditions. It seems unlikely that the aperture conductance would be
enhanced by such a large a factor under essentially free molecular conditions.

Clearly, simple combinations of isentropic and molecular flow and aperture + entrance models are not a good basis for general
empirical equations applying to short ducts. As discussed under continuum flow, in short ducts a term must be included to account
for the kinetic energy required to accelerate the gas. It is suggested that approaches to empirical equations in the transition regime
could be based on formulations which force the kinetic energy term to vanish under molecular flow conditions.

Symbols

Symbol Meaning Section

a Duct cross-section dimension, minor axis of ellipse 2.2.5


A Cross-sectional area 2.2.2
b Duct cross-sectional dimension, major axis of ellipse, b ≥ a 2.2.5
B Cross-sectional perimeter (includes all wetted surfaces) 2.2.2
C Conductance 2.1
Ca Molecular flow conductance of an aperture 2.2
Cm Molecular flow conductance of a duct 2.2
Cn Net value conductances in series 2.1
Ct Transitional flow conductance of a duct 2.4.1
Cv Viscous flow conductance 2.3.1
Cvd Viscous flow conductance, calculated at downstream (lower) pressure 2.3.1
Cvu Viscous flow conductance, calculated at upstream (higher) pressure 2.3.1

Cz The term 2.3


d Diameter of a cylindrical tube Introduction
Dh Hydraulic diameter 2.3

(table continued on next page)

< previous page page_137 next page >


< previous page page_138 next page >
Page 138

(table continued from previous page)

Symbol Meaning Section

fD Darcy friction factor 2.3.2


Fg Gas factor Introduction
Ge Numerical constant and cross-sectional geometry terms in viscous conductance equations 2.3.2
G(γ) Function of γ defined by Eq. (2.106) 2.3.3.1
k Generally used for constants and units conversions
Kn Knudsen number Introduction
Kp Pressure ratio: (higher pressure)/(lower pressure) 2.1
Kpc Choked pressure ratio 2.3.6
Kpca Choked pressure ratio for an aperture 2.3.3.1
l Length of a duct (in direction of gas flow) 2.2.2
le Equivalent length used in Eq. (2.25) 2.2.4
lentry Entry length 2.3.1
Ma Mach number (ratio of flow velocity to local velocity of sound) 2.3.1
Man Mach number at duct entrance 2.3.3
Max Mach number at duct exit 2.3.3
Mm Molar mass (e.g., 0.028 kg·mole1 for nitrogen) Introduction
n Number density of molecules Fig. 2.5
nc Loss coefficient or number of corrections for flow obstructions 2.3.4
N Number of molecules striking an area 2.2

Mean pressure 2.3.1


P Pressure Introduction
Pd Downstream (lower) pressure 2.1
Pu Upstream (higher) pressure 2.1

Throughput (in pressure × volume units) 2.1

Choked throughput for an aperture 2.3.6

Value of throughput at onset of turbulent flow 2.3


r Generally used for ratios
R Radius of a cylindrical tube 2.2.4
Re Reynolds number 2.3
R0 Universal gas constant (8.314 J·mole1·K1) Introduction
S Pumping speed 2.1
Sa Speed of a choked aperture 2.3.3.1
Sca Pumping speed required to choke an aperture 2.3.3.1
SF Shape factor 2.3.2
Sn Net speed of a pump in series with conductances 2.1
t Time 2.2.11
T Thermodynamic temperature Introduction
u Flow velocity [and dimensionless terms in Eq. (2.36)] 2.3
va Mean thermal velocity of molecules 2.2.2
V Volume 2.1
Z Gas flow impedance 2.1
Z1 Term in Knudsen transitional flow Eq. (2.142) 2.4.1
α Transmission probability 2.2
αn Net transmission probability of components in series 2.2.10

(table continued on next page)

< previous page page_138 next page >


< previous page page_139 next page >
Page 139

(table continued from previous page)

Symbol Meaning Section

αH Ho coefficient of a pump 2.2.10

β The term 2.3.3


γ Ratio of the principal specific heats for a gas 2.3.3
δ Dimension ratio used in Eq. (2.26), diffuse scattering fraction in Eq. (2.141), and factor in Eq.
(2.42)
ε Wall roughness 2.3.2
λ Mean free path Introduction
η Viscosity of a gas Introduction
ρ Gas density 2.3
θ Angle in Eq. (2.18) and weighting factor in Eq. (2.158)
τ Time constant 2.2.11

References

1. M. Knudsen, Ann. Phys. (Leipzig) [4] 28, 9991016 (1909).

2. S. Dushman, Scientific Foundations of Vacuum Technique, Chapter 2. Wiley, New York, 1949.

3. W. Steckelmacher, Vacuum 16, 561584 (1966).

4. W. Steckelmacher, Proc. Int. Vac. Congr., 6th, Kyoto Japan, 1974; J. Appl. Phys., Suppl. 2 (Part 1), 117125 (1974).

5. W. Steckelmacher and M. W. Lucas, J. Phys. D 16, 1453 (1983).

6. P. Clausing, Ann. Phys. (Leipzig) [5] 12, 961 (1932); J. Vac. Sci. Technol. 8, 636646 (1971).

7. M. Knudsen, Ann. Phys. (Leipzig) [4] 28, 75 (1909).

8. W. Steckelmacher, Vacuum 28, 269275 (1978).

9. W. Steckelmacher, Rep. Prog. Phys. 49, 1083 (1986).

10. M. Smoluchowski, Ann. Phys. (Leipzig) [4] 33, 1559 (1910).

11. R. J. Cole, Rarified Gas Dyn. 10 (Part 1), 261272 (1976).

12. A. S. Berman, J. Appl. Phys. 10, 3356 (1965); erratum, ibid. 37, 2930 (1966).

13. D. J. Santeler, J. Vac. Sci. Technol. A 4(3), 338 (1986).

14. L. L. Levenson, N. Milleron and D. H. Davies, Vide 103, 42 (1963).

15. D. J. Santeler and M. D. Boeckmann, J. Vac. Sci. Technol. A 9(4), 2378 (1991).

16. R. J. Cole, Proc. R. Soc. Edinburgh 82A, 211223 (1979).


17. A. S. Berman, J. Appl. Phys. 40, 4991 (1969).

18. D. H. Davies, J. Appl. Phys. 31, 1169 (1960).

19. C. W. Oatley, Br. J. Appl. Phys. 8, 15 (1957).

20. R. A. Haefer, Vacuum 30, 217 (1979).

21. L. Holland, W. Steckelmacher and J. Yarwood, Vacuum Manual. Spon, London, 1974.

22. H. S. Carslaw and J. C. Jaeger, Conduction of Heat in Solids. Oxford University Press, London, 1959.

23. J. D. Lawson, J. Sci. Instrum. 43, 565 (1966).

24. V. E. Senecal and R. R. Rothfus, Chem. Eng. Prog. 49, 533 (1953).

< previous page page_139 next page >


< previous page page_140 next page >
Page 140

25. R. K. Shah and A. L. London, Laminar Flow Forced Convection in Ducts. Academic Press, New York, 1978.

26. S. E. Haaland, J. Fluids Eng. 105, 89 (1983).

27. O. C. Jones, J. Fluids Eng. 98, 173 (1976).

28. F. M. White, Fluid Mechanics, 2nd ed. McGraw-Hill, New York, 1986.

29. B. S. Massey, Mechanics of Fluids, 5th ed. Van Nostrand-Reinhold, Wokingham, Berkshire, England, 1983.

30. A. H. Shapiro, The Dynamics and Thermodynamics of Compressible Fluid Flow, Vols. 1 and 2. Ronald Press, New
York, 1953.

31. G. W. C. Kaye and T. H. Laby, Tables of Physical and Chemical Constants, 15th ed. Longman, New York, 1986.

32. F. M. White, Viscous Fluid Flow. McGraw-Hill, New York, 1991.

33. D. J. Santeler, J. Vac. Sci. Technol. A 4(3), 348 (1986).

34. C. Cercignani and F. Sernagiotto, Phys. Fluids 9, 40 (1966).

35. C. Cercignani, Theory and Application of the Boltzman Equation. Scottish Academic Press, Edinburgh, 1975.

36. S. K. Loyalka, T. S. Storvick and H. S. Park, J. Vac. Sci. Technol. 13(6), 1188 (1976).

37. Masahiro Ota and Hiroyoshi Taniguchi, Vacuum 44, 685 (1993).

38. L. Fustoss, Vacuum 31, 243 (1981).

39. G. Scherer-Abreu and R. A. Abreu, Vacuum 46, 863 (1995).

40. S. L. Thompson and W. R. Owens, Vacuum 25, 151 (1975).

41. W. G. Pollard and R. D. Present, Phys. Rev. 73, 762 (1948).

42. F. M. Sharipov and V. D. Seleznev, J. Vac. Sci. Technol. A 12, 2993 (1994).

43. W. Dong and L. A. Bromley, Trans. Natl. Vac. Symp. 8, 1116 (1961).

44. J. Kieser and M. Grundner, Vide 201, 376 (1980).

45. J. F. O'Hanlon, J. Vac. Sci. Technol. A 5, 98 (1987).

46. D. J. Santeler, J. Vac. Sci. Technol. A 12, 1744 (1994).

47. S. F. DeMuth and J. S. Watson, J. Vac. Sci. Technol. A 4, 344 (1986).

< previous page page_140 next page >


< previous page page_141 next page >
Page 141

3
Positive Displacement Vacuum Pumps

A positive displacement pump is a vacuum pump in which a volume filled with gas is cyclically isolated from the inlet,
the gas being then transferred to an outlet. The first pumps of this type were piston pumps, but in 1905 Gaede [1]
introduced the modern rotary vane (blade) vacuum pump. It has developed in many ways over the last 90 years: First
and very early on, it was oil-sealed and then a number of improvements were incorporated, including reduction of noise
and vibration, handling of condensibles and corrosive materials, elimination of oil being sucked back when stopped,
direct drive (running at 1450 or 1800 rev/min), and so on.

Another type of pump which has been widely used, primarily in the chemical industry, is the liquid ring pump wherein
the pumping action is developed by a rotating liquid. Pumps of this type produce nearly isothermal compression and can
handle dry gases or vapor gas mixtures. By reason of the isothermal compression, it is possible to handle explosive
gases or gases subject to polymerization. With the pumping of vapor-laden gases, condensation occurs in the pump,
resulting in an enhanced capacity.

In the early 1980s it became clear that the pumping of corrosive and inflammable materials, combined sometimes with
abrasive dust, was complicating the operation of the oil-sealed rotary pump and much reduced its normal long life and
performance. It was also becoming necessary to take precautions against backstreaming of the oil into some vacuum
systems to reduce system contamination. These two requirements led to the search for pumps which did not use liquid
sealing in their pumping mechanisms. The resultant pumps were referred to as ''dry vacuum pumps" because of the
absence of liquid sealing. The first pumps were based on a number of stages working either on the Roots Principle
(circa 1861) or on the more recent Claw Principle (circa 1930), but there are now a number of other mechanisms in use
and these are described in a later

Foundations of Vacuum Science and Technology, Edited by James M. Lafferty.


ISBN 0-471-17593-5 © 1998 John Wiley & Sons, Inc.

< previous page page_141 next page >


< previous page page_142 next page >
Page 142

section. An attempt is made to indicate the relative merits of the various mechanisms, but due to the rapid development
in this field they are likely to be modified with time by design changes.

When a positive displacement pump is used to back a secondary pump which has high pumping speed, but in a lower
pressure region, the forepressure requirement of the secondary pumps must be satisfied. In recent years the introduction
of turbo pumps with integral molecular drag stages has resulted in pumps requiring a forepressure of around 10 mbar, so
a two-stage diaphragm pump is sometimes used. When choosing a forepump it is always necessary to ensure that the
effect on the performance of the secondary pump is taken into account (see Chapter 9).

< previous page page_142 next page >


< previous page page_143 next page >
Page 143

Part I
Oil-Sealed Vacuum Pumps

Nigel T. M. Dennis

3.1
Oil-Sealed Vacuum Pumps

Capacities available: 1 to 1500 m3/hr

Operating pressure range:

Single-stage: 1000- to 5 × 102-mbar total and partial pressure


Two-stage: 1000- to 103-mbar total pressure
1000- to 104-mbar partial pressure

Total pressure is influenced mainly by the type and quality of the oil charge, while the partial pressure is due to internal
transfer of gas within the pump.

3.1.1
Pump Design

There are two basic designs of oil-sealed rotary pumps, and these are illustrated in Fig. 3.1. Figure 3.1a shows the two-
vane pump where gas is trapped between the vanes and the stator before it is swept out through the outlet valve. Figure
3.1b illustrates the rotary piston pump where a single vane is slotted into the stator by the use of a hinge pin and is part
of a sleeve which fits around the rotor. The vane is hollow and acts as an inlet valve, closing off the pumping chamber
from the inlet when the rotor is at top center. Smaller pumps tend to be of the two-vane design, while larger pumps are
more frequently of the rotary piston design.

Foundations of Vacuum Science and Technology, Edited by James M. Lafferty.


ISBN 0-471-17593-5 © 1998 John Wiley & Sons, Inc.

< previous page page_143 next page >


< previous page page_144 next page >
Page 144

Fig. 3.1
Two basic designs of rotary pump mechanism. (a) Two-vane pump.
(b) Rotary piston pump. A, rotor; B, stator casing; C, rotor sleeve and
single vane; D, vane; E, hinge pin; F, gas ballast port; G, outlet valve;
H, gas entering the swept volume; J, the gas being compressed.

In the two-vane design the main path of leakage between the inlet and the outlet is the rotor to stator gap at the point
between these two ports. To reduce this leakage to a minimum, it is usual to use a long path seal which is provided by a
groove in the stator which is of the same diameter and on the same center as the rotor. This gives a long path clearance
which is sealed with oil between the inlet and the outlet and reduces carryover of gas between the two ports to a
minimum.

In a two-stage pump it is normal to feed outgassed oil from the outlet stage to seal and lubricate the inlet stage, thereby
ensuring that the best ultimate can be achieved.

3.1.2
Gas Ballast [2]

This is a feature useful in reducing the extent of vapor contamination of the oil. Atmospheric air or, if required, a dry or
inert gas is admitted to the pump during the compression stage (see Fig. 3.2) to increase the proportion of
noncondensable gas in the pump by the time compression has progressed to a point (about 1200 mbar) when the outlet
valve lifts. By this means the partial pressure of the vapor being pumped, at the time when the outlet valve lifts, does not
exceed its saturated vapor pressure at pump temperature, so that the vapor is discharged without liquefaction.
Furthermore, due to the extra work done in compressing the gas introduced as gas ballast, the pump temperature rises
and further assists in preventing vapor condensing within the pump. Gas ballasting is a useful technique for purging the
pump oil of condensed or dissolved vapor; it should proceed for a minimum of 20 minutes to allow the pump to warm
up completely. Gas ballasting is very useful for pumping vapors that do not dissolve in the pump oil and is still of some
value for vapors that do dissolve.

< previous page page_144 next page >


< previous page page_145 next page >
Page 145

Fig. 3.2
Method of introducing gas ballast to a bladed rotary pump.

The gas ballast flow is generally about 10% of the free air displacement of the pump. Because the effectiveness of gas
ballast is very dependent on pump temperature, it is desirable to run the pump as hot as possible. Typically, running
temperatures are 60°C to 90°C.

Condensation will occur when the temperature in the pump outlet line reaches the saturated vapor pressure of the vapor
that is being pumped. It is, therefore, important that a catch pot is fitted immediately adjacent to the outlet port so that
fluid condensed in the outlet line does not run back into the pump oil box.

The ultimate pressures attainable when gas ballast is used are as follows:
Single-stage pump: about 0.5-mbar total pressure
Two-stage pump: about 102-mbar total pressure

< previous page page_145 next page >


< previous page page_146 next page >
Page 146

To calculate the minimum (maximum) safe pumping rate it is assumed that the ballast gas just becomes fully saturated
with vapor at the end of the compression as it is forced through the outlet valves and that the vapors are insoluble in the
pump oil [3]. It is therefore necessary to know the volume flow rate, the partial pressure, and the absolute temperature
of the vapor under these conditions (this may be assumed to be the temperature of the pump itself). The partial vapor is
then the saturation vapor pressure at the pump temperature (T, K), and its value is available from vapor pressure data
(Ps, mbar). If the total pressure reached under the outlet valve is PT (mbar), the partial ballast gas pressure is PT Ps.

The volume rate of flow for the vapor is the same as for the ballast gas and is therefore obtained by calculating the
volume flow rate for the ballast gas, in terms of its measured flow rate (V, m3/h) into the pump at atmospheric pressure
and temperature (PA and t, K).

Thus we have the volume/hour of emerging gas:

If the maximum pumping rate is W (kg/h) and if the vapor density of the saturated vapor at T (K) is ρs (kg/m3), then

The ordinary perfect gas equation is used to obtain ρs with adequate accuracy in this pressure range; thus for water
vapor we obtain

Taking PA as 1013 mbar and assuming that the pressure PT under the outlet valve when the mixture is expelled is 1200
mbar and t = 293 K, then for water vapor we obtain

Knowing Ps (the vapor pressure for water) at the pump's temperature T and the volume rate of flow of the ballast gas, it
is now possible to calculate the maximum safe water vapor pumping rate in kg/h. Knowing the speed of the pump (S,
m3/h), it is then possible to calculate the maximum safe inlet pressure of water vapor:

< previous page page_146 next page >


< previous page page_147 next page >
Page 147

If the ballast air is humid, then Ph (mbar) is the partial water vapor pressure, the true mass flow rate of air is reduced in
the ratio (PA Ph)/PA, and water vapor is also entering the pumps with the ballast air at a rate of

Equation (3.1) must be changed to allow for the reduced true ballast air flow and the additional water vapor entering
with the ballast air; thus

This is only applicable when pumping water vapor. If other vapors are pumped, the moisture contributes useful ballast.

The effect of ballast humidity is small at the normal gas-ballasted pump temperature of 70°C or higher and is generally
disregarded.

As the vapor pressure of water is rapidly increasing at these higher temperatures, the pump temperature is very
important in determining its water vapor capacity. In the standard on determining this part of the pump performance, it
is indicated that an initial test is carried out allowing air to be substituted for water vapor at the likely safe inlet pressure,
so that the pump temperature T is adjusted.

High or low ambient temperature has an effect which can be appreciable if it is more than a few degrees away from 20°
C.

An alternative to gas ballast is the use of extremely small air bubbles passing through the oil to carry the contamination
away. This has the advantage of minimal loss of pump oil and a normal ultimate vacuum. The amount of contamination
that can be removed by this method is much less than by normal gas ballasting and does require a small air pump to
produce the bubbles.

3.1.3
Pump Oil [4]

The type most widely used is normally highly refined hydrocarbon oil. For very-heavyduty applications, synthetic
organic fluids are used, and in highly corrosive or flammable situations a perfluoropolyether fluid is used. The oil used
should be of the viscosity recommended by the manufacturer; lower-viscosity oil normally results in the pump being
noisier, while higher-viscosity oil can give difficulty in starting at low ambient temperature and occasionally leads to
seizure.

1. Hydrocarbon Oils: The general fluid used in most applications is a highly refined mineral based oil; it normally has a
vapor pressure around 106 mbar at room temperature. Some oils have additives to help reduce the corrosive effects of
any vapors being pumped, but their addition results in a higher ultimate total pressure. They are also gradually
neutralized, so their effective life is limited. Hydrocarbon oils are slightly hygroscopic, so the pump should be filled
from

< previous page page_147 next page >


< previous page page_148 next page >
Page 148
a bottle or can which has only recently been opened. If the pump is not used for a lengthy period, it is advisable to gas ballast for a while to clear
the oil of any dissolved water vapor.

2. Synthetic Organic-Type Fluids: These are used on heavy-duty applications because certain types have improved resistance to oxidation at the
high pump temperature associated with these applications.

3. Fluorinated Fluids [5]: These are rotary pump grades which are used in applications where their corrosion-resistant properties and
chemical inertness result in greatly increased operational times between pump maintenance. The basic properties of perfluoropolyether fluids
and their advantages for use in vacuum systems are as follows:

Chemical inertness. Ideal for pumping aggressive materials, particularly in semiconductor processing.
Noninflammable. No fire risk.
High thermal No residual "tars" formed by overheating.
resistance. Eventually reduced to gaseous products.
Oxygen-compatible. Allows safe pumping of oxygen. Refer to manufacturers literature for maximum recommended
service temperature and pressure with oxygen.
Immiscible with most common solvents. Allows pumping of solvent in some cases without gas ballast.

The pump must have no traces of hydrocarbon oil when charged with perfluoro-polyether fluid; otherwise, some or all of the above advantages
are adversely affected. With this type of fluid the ultimate vacuum of the pump is not normally as good as with the normal oils. This is due to oil
box gases being dissolved in the fluid and evolving when the fluid reaches the high-vacuum stage of the pump. Note: Perfluropolyether fluids
decompose at 250300°C, and the resultant products are toxic.

3.1.4
Oil Suckback [6]

This occurs if the pump is stopped for a period of time under vacuum. Oil suckback is prevented by incorporating into the pump inlet line a
nonreturn valve. When this is an integral part of the pump, it is operated in a number of different ways. When this arrangement is used, it is
normal for the pump mechanism to be allowed to reach atmospheric pressure so that there is no chance of oil being sucked back due to faulty
sealing of the valve. An alternative way is to keep the pump under vacuum and stop any oil entering the pump by an oil valve and a sealing
outlet valve. The oil and outlet valves are designed so that only a small amount of oil can be sucked into the pump mechanism if the valves fail
to seal.

3.1.5
Power Requirements and System Protection

The power to the pump rises at a maximum of 100300 mbar and then falls to a low level. It is mainly caused by friction within the pump and
motor losses, when the pump reaches 10 mbar and below.

< previous page page_148 next page >


< previous page page_149 next page >
Page 149

Contamination of the system occurs at low gas flows toward the pump, normally when the inlet pressure is below 1.0
mbar. This is due to condensation of both the oil vapor from the pump and its breakdown products in the system which
is at a lower temperature than the pump. Therefore, it is inadvisable to leave a rotary pump on a system at or near its
ultimate vacuum for a prolonged period without a trap or another type of pump between it and the system. The traps
used contain either an activated alumina or a molecular sieve trapping medium. Both these materials absorb water
vapor, and it is therefore best to bypass them during pumpdown because otherwise it becomes difficult to achieve a
good ultimate vacuum.

3.1.6
Accessories [7]

Besides the inlet trap mentioned above, there is a wide range of accessories mainly aimed at preventing dust from
getting into the pump mechanism, decontaminating the pump oil, and preventing oil from being lost from the pump.
Inlet filters are often used on dusty systems, but these reduce the pumping speed considerably especially at the lower
pressures. The filter element is normally removable and of a pleated construction to give a high conductance. In
applications where dust that is too fine to be readily stopped by inlet filters or is being continuously formed at the pump
inlet, an oil circulation system is fitted to the pump oil box. This system includes an oil filter and a circulating oil pump
(sometimes an integral part of the vacuum pump). An oil pressure gauge or switch gives an indication of when the filter
element has to be changed.

On a system where large quantities of vapor are to be pumped, it is normal to use an inlet condenser. The cooling
medium of this condenser is normally water, but it can be cooled down to as low as 80°C depending on the application.
For details see "Sorption Roughing Pumps," Section 5.20. Outlet mist filters are used to capture the extremely fine mist
that is evolved from a pump. Some pumps include a preliminary oil filter directly above the outlet valve, but to
completely eliminate mist under all operating conditions it is still necessary to fit an external filter. In smaller systems
this filter can also include a charcoal deodorizer. In cases where the pumps are running at reasonably high pressure
(about 10 mbar), it is possible to suck the collected oil back from the outlet filter into the inlet of the pump or through
the gas ballast connection; otherwise, it is necessary to periodically empty the oil from the filter and return it to the
pump. If vapors are being pumped, neither of these systems is possible and the oil should not be returned to the pump.

< previous page page_149 next page >


< previous page page_151 next page >
Page 151

Part II
Liquid Ring Pumps

Helmut Bannwarth

3.2
Liquid Ring Pumps

Capacities available: 1 m3/h to 27,000 m3/h vacuum pumps [8]


Operating pressure range (minimum): 1013 to approximately 33 mbar total pressure extended to 5 mbar with gas ejector

3.2.1
Mechanism

With regard to the arrangement of the impeller in the housing, it is necessary to distinguish between the concentrically mounted and the
eccentrically mounted vane-impeller. Machines with cylindrical housings are designated as single-acting or single-chamber liquid ring
pumps. According to their method of operation, machines with oval-shaped or elliptical housing are designated as double-acting or two-
chamber liquid ring pumps. A further classification is made with regard to the direction of the gas inlet and outlet stream in relation to the
impeller. Usually, pump assemblies have axial or radial flow. Liquid ring vacuum pumps are, as a rule, of single-stage or two-stage
design. The functioning of a liquid ring machinefor example, a vacuum pump with eccentrically arranged impelleris described as follows.
The liquid ring vacuum pump with eccentric impeller, arranged in a circular housing, belongs to the group of single-acting or single-
chamber liquid ring pumps. The housing must not be exactly cylindrical, but should have an elliptical profile. Through varying the profile
of the housing, it is possible to attain the optimal adjustment of the machine for the operating requirement. In these pumps, the driving
force from the impeller is transferred through the liquid ring to the pumped medium. The liquid ring will accelerate on the

Foundations of Vacuum Science and Technology, Edited by James M. Lafferty.


ISBN 0-471-17593-5 © 1998 John Wiley & Sons, Inc.

< previous page page_151 next page >


< previous page page_152 next page >
Page 152

Fig. 3.3
Cross section of a one-stage liquid ring vacuum pump
with canned motor (Lederle-Hermetic GmbH).

suction side and, therefore, has augmented energy. On the discharge side, the ring liquid enters the impeller
compartments, whereupon a portion of the kinetic energy is transformed into static energy of compression. On the
dissipation of the static energy, the losses in the gas and liquid streams are overcome, so that the gas is compressed and
ejected.

Assuming that the vane-impeller is arranged eccentrically in the properly adjusted housing, the moving impeller gives
impetus to the rotating liquid ring. Because of the eccentricity of the impeller to the housing, a crescent shaped cavity is
formed between the impeller hub and the liquid ring as the impeller rotates. The impeller vanes split up this cavity into
many segments of differing volume. On rotation of the impeller, the segments increase in size in the region of the intake
port (Fig. 3.3) and gas or gas vapor mixture is sucked in.

In the region of the discharge port opposite, the pumped medium in the segments (which are here becoming smaller)
compresses and is ejected. The intake and discharge ports are to be found with control plates on the side of axial-intake
machines. Liquid ring pumps are rotary machines with the characteristics of piston pumps. The liquid ring assumes the
function of the piston.

3.2.2
Single-Stage Liquid Ring Vacuum Pumps [9]

With the single-stage liquid ring vacuum pump, it is possible to attain various pressure ratios through the design and
arrangement of ports. Liquid ring vacuum pumps with control plates and venting holes are operated economically at
compression ratios of up to 1:7 for suction pressures between 1013 mbar and about 150 mbar

< previous page page_152 next page >


< previous page page_153 next page >
Page 153

Fig. 3.4
Cross section of a two-stage liquid ring vacuum pump (Sihi GmbH). 1, Shaft;
2, impellers; 3, control plates; 4, middle section of housing; 5, end cover;
6, shaft seal; 7, shaft bearing; 8, suction port; 9, discharge port.

with water at 15°C as the ring liquid. The single-stage liquid ring vacuum pump can operate at compression ratios
greater than 1:7 but at considerably diminished capacities, provided that the discharge port is reduced, the venting holes
are suitably arranged, and an automatic valve is provided. Disc valves are usually used. The compression ratio is
accommodated by means of this valve, which is contiguous with the discharge opening in the control plate. The flexible
plate in use covers or uncovers the venting holes automatically over the entire pressure range and prevents
overcompression and back flow. Using water at 15°C as the ring liquid, inlet pressures in the range between 1013 mbar
and about 33 mbar are achieved with a corresponding compression ratio between inlet and discharge pressure of 1:30.7.
If other ring liquidsfor example, oilare used, inlet pressure is limited to between about 10 mbar and 30 mbar due to
outgassing.

3.2.3
Two-Stage Liquid Ring Vacuum Pumps [8,9]

For inlet pressures which lie in the region below 150 mbar, instead of using single-stage machines with flexible
discharge ports, it is possible to use a two-stage liquid ring vacuum pump (Fig. 3.4). This construction has gas discharge
openings without valves. The discharge from the first stage becomes the suction to the second stage. The partial
compression ratio per stage is variable in regard to the inlet pressure. Through
< previous page page_153 next page >
< previous page page_154 next page >
Page 154

a fixed outlet port, a compression ratio of less than 1:7 per stage is maintained. The overall compression ratio is the
product of the two partial compression ratios. With water at 15°C as the ring liquid, two-stage liquid ring vacuum
pumps may also be used to accommodate suction pressures in the range between 1013 mbar and 33 mbar.

3.2.4
The Operating Liquid

The operating liquid serves both to transfer energy and to seal the vane-impeller and spaces between the impeller,
control plate, and housing, and it also absorbs and removes quantities of heat generated in the pump [10]. Water as well
as other ring fluids, dependent upon the nature of the process, may be usedfor example, hydrocarbons, solvents, and
synthetic light oils. In addition to the heat of compression, additional heating in liquid ring pumps may be generated by
condensation of vapors and absorption of gases, through chemical reactions occurring between process gas and the ring
liquid or through the cooling of the principal gas at elevated temperatures.

3.2.5
Operating Ranges of Liquid Ring Gas Pumps

To obtain the lowest achievable pressure with the liquid ring vacuum pump, it is necessary to prevent boiling of the
operating liquid. Using water at 15°C as the ring fluid, the lowest possible inlet pressure which is practical as well as
possible is approximately 33 mbar [11]. Lower inlet pressures may be reached through the use of a liquid having a
higher boiling point. An improvement of the inlet pressure can also be attained through the use of an additional vacuum
pump operating on a different principle connected on the inlet side of the liquid ring pump. Liquid ring vacuum pumps
are often used in combination with gas ejector pumps; and here it is possible, using water at 15°C as the ring liquid, to
accommodate inlet pressures in the range between 33 mbar and approximately 5 mbar. In the combination, the gas
ejector is placed directly at the intake of the vacuum pump (Fig. 3.5). The special curve over the operating range of a
vacuum pump with gas ejector, using water at 15°C as the ring liquid, is shown in Fig. 3.6. As the driving medium for
the gas ejector, we may use atmospheric air, the process gas itself, or any suitable noncondensable gas at atmospheric
pressure.

The liquid ring vacuum pump can also be adapted for use at inlet pressures below 5 mbar. In these cases, one or more
vacuum pumps utilizing a different pumping principle are placed at the inlet side of the liquid ring pump. The choice of
vacuum pumps is to be such that the pump combination will operate synchronously over the required suction pressure
range. Since over wide pressure ranges there are changes in specific gas densities and types of flow (viscous, Knudsen,
and molecular flow), different pumping principles are used for operation over wide pressure ranges. A condenser is
often used in a vacuum pump combination for handling vapors and often improves the overall efficiency of the pumping
equipment [12].

3.2.6
Cavitation and Protection against Cavitation

The application range of the liquid ring vacuum pump for lower inlet pressures is limited by the vapor pressure of the
operating liquid. If the physical properties of the

< previous page page_154 next page >


< previous page page_155 next page >
Page 155

Fig. 3.5
Liquid ring vacuum pump with gas
ejector. A, intake gas; B, atmospheric
air as motive gas; P0, intake pressure;
P, back pressure of the gas ejector =
intake pressure for the liquid ring
vacuum pump; C, exit of the
intake gas and motive gas.
Fig. 3.6
Operating range of a liquid ring vacuum pump with gas ejector.

< previous page page_155 next page >


< previous page page_156 next page >
Page 156

operating liquid are unsatisfactory, it will boil and the impeller vanes will fill up with vapor. The vapor transportation
by the movement of the impeller vanes toward the pump outlet side results in collapse of the vapor bubbles. This
implosion, known as cavitation, causes a banging noise and noticeable shaking of the pump. Cavitation will not occur in
the liquid ring vacuum pumps if the pump operation is always limited to a minimum inlet pressure and corresponding
volumetric flow and noncondensable gases are handled. The pump can also be provided with gas from a liquid separator
on the inlet side operating in conjunction with a regulating valve toward the inlet. If a gas ejector is used in a pump
combination and there is always a gas flow through the open driving gas inlet, cavitation will not occur [11]. To prevent
cavitation damage, a liquid ring vacuum pump can be provided with a built-in cavitation protection. This is in the form
of a gas ballast in the working space on the discharge side. This has only a small effect on the pumps operating
characteristics.

3.2.7
Types of Operation; Conveyance of Operating Liquid

Three principal methods of operation may be differentiated: operation without back flow of liquid (fresh liquid injected,
cooling circuit); operation with back flow of liquid (combination operation, economical circuit); and closed liquid
circulation (return circulation, closed circulation) [12]. The principle of operation and disposal of the operating liquid
will depend upon the process technology, the suction pressure, the available cooling medium and its temperature, the
situation with regard to corrosion of materials, and the nature of the process gas and operating liquid. The installation of
these machines and correspondingly optimized vacuum systems should be done in an economically favorable manner so
that environmental safety precautions are observed. Figure 3.7 shows the principal circuitry for operation with closed
liquid circulation.

In this method of operation, the operating liquid in the liquid separator is separated from gas as well as vapor and passes
back into the pump through a heat exchanger wherein it is indirectly cooled. In the heat exchanger, the heat is removed
from the system. The special advantage of this method of operation is that the intake gases and

Fig. 3.7
Principal circuit for operation with
closed liquid circulation.

< previous page page_156 next page >


< previous page page_157 next page >
Page 157

vapors as well as the operating liquid do not come into contact with the cooling medium. There is no problem of
disposal from the cooling liquid side. By reason of the isothermal compression at low gas temperatures, the gases or
vapors discharged into the atmosphere contain only small quantities of noxious materials.

The closed circulation method of operation is used in the chemical industry and in other situations wherein the
requirement for freedom from leakage and protection of the environment exists. The contaminated operating liquid may
require special disposal procedures.

3.2.8
Materials of Construction

In the choice of materials of construction, resistance to corrosion normally plays the major role. In process technology,
the gases to be compressed are very often chemically active or process conditions require the use of alkaline or acidic
operating liquids. Liquid ring pumps are robust machines whose moving and fixed internal parts are not subject to
failure. It is practical to use all materials in the construction of these pumps which are usual in machine building and in
process technology. Liquid ring machines may be constructed using such materials as gray cast iron, spheroidal cast
iron, cast steel, chemically resistant chrome nickel steels, high nickel alloys, titanium bronze polypropylene, polyvinyl
plastics, Noryl (polyphenylen oxide), Ryton (PPS), ceramics, and stoneware. Combinations of different materials may
also be used. Parts for these pumps may also be coated with hard rubber, plastics, enamel, or special lacquer to protect
against corrosion and erosion.

3.2.9
Sealing

In liquid ring vacuum pumps, one differentiates between static and dynamic sealing. Fluid seals based on
polyvinylacetate, polyisobutylene, and epoxyresin are used as static seals for the housings. Nitrile rubber, fluorinated
elastomers, polytetrafluorethylene, ethylenepropylene rubber and chloroprene rubber are used for O rings and flat
gaskets. Where dynamic sealing is required, such as for shafts, conventional seal elements such as stuffing boxes or
single- or double-acting mechanical seals are used. Stuffing boxes with double packing and lantern rings are usual. In
vacuum technology the demand exists that there be zero leakage when machinery is operating and when it is stopped.
Liquid ring machines in hermetic construction are presented as an alternative to vacuum pumps of the usual design with
dynamic seals.

There exist two alternative types of hermetic machines: those using a drive with permanent magnet coupling and those
with canned motors (Fig. 3.3). In choosing a drive system, the transfer of torque and the requirement for shaft sealing
exist concurrently. The hermetically encapsulated construction is recommended in cases where an absolutely leak free
and maintenance free operation is demanded. With machines of this type, a zero rate of leakage is achieved.

3.2.10
Drives

Two-, four-, or six-pole low-voltage asynchronous squirrel-cage motors are usually used as drives. For larger machines,
having drive requirements above about 150 kW, high-voltage squirrel-cage motors are used. For large pumps operating
at low speeds,

< previous page page_157 next page >


< previous page page_158 next page >
Page 158

gear reducers of V-belt reducers may be placed between motor and pump, thus allowing for speed and adjustment such
that specific operating conditions may be optimally met. Especially for large machines it is useful to place a
hydrodynamic coupling between the motor and pump in order to take care of starting resistance and frequency control
on start-up of induction motors.

Hermetic drive systems may be used as alternatives to conventional drives. The torque is transferred through either
permanent magnets or electromagnets. The complete leak tightness of these pumps is especially necessary when toxic,
carcinogenic, or radioactive gases or vapors are handled and where no leakage of the operating liquid can be tolerated.
Liquid ring vacuum pumps with hermetic drive systems are presently in use in the process industries; these pumps have
maximum capacities of about 3000 m3/h and can be provided in special construction for higher capacities.

With canned motors, the noise emissions of a pump may be reduced by an average of 810 dB(A).

3.2.11
Accessories

A heat exchanger and liquid separator are needed for the operation of a liquid ring vacuum pump with closed circulation
(Fig. 3.7). Dependent upon the operating conditions, various other accessories such as gas ejector, check valves,
emission cooler, control and surveillance instrumentation, and so on, may be required.

< previous page page_158 next page >


< previous page page_159 next page >
Page 159

Part III
Dry Vacuum Pumps

Nigel T. M. Dennis

3.3
Dry Vacuum Pumps

3.3.1
Roots Pump

Roots Pump (mechanical booster) plus oil-sealed rotary forepump

Capacities available: 75 to 30,000 m3/h


Operating pressure range: 10 to 103 mbar total pressure, inlet pressure is sensitive to fore pressure

Multistage Roots Pump

Capacities available: 25 to 1000 m3/h


Operating pressure range: (4-stage) 1000 to 5 × 102 mbar total pressure
(5-stage) 1000 to 2 × 102 mbar total pressure
(6-stage) 1000 to 103 mbar total pressure

A typical Roots pump is illustrated in Fig. 3.8. It consists of two figure-of-eight-shaped rotors, although for higher pressure duty three-lobe rotors
are sometimes used. These rotors are synchronized by external gears and rotate in opposite direction within the stator. The gears and rotor
bearings are oil-lubricated; but they are external to the pumping chamber, so the swept volume of the pump is dry. A small clearance, generally
between 0.05 and 0.25 mm, is maintained between the rotors and between each rotor and the stator wall. As the rotors run dry, back leakage
occurs through these clearances at a rate dependent on the pressure difference across the pump and

Foundations of Vacuum Science and Technology, Edited by James M. Lafferty.


ISBN 0-471-17593-5 © 1998 John Wiley & Sons, Inc.

< previous page page_159 next page >


< previous page page_160 next page >
Page 160

Fig. 3.8
Cross section through a Roots (mechanical booster)
pump and its operating cycle.

the gas being pumped. The pressure ratio generally achieved by this mechanism is shown in Fig. 3.11, indicating a high
compression at low pressures and a low compression at high pressure.

A mechanical booster is basically a Roots pump modified for high-vacuum applications. This generally involves
improved leak tightness, different stator/rotary seals, and the connection of the gear box to the outlet of the pump so the
same pressure exists in the gear box as in the foreline.

As indicated in Fig. 3.11, the ultimate pressure achieved by a mechanical booster is very dependent on the ultimate of
the fore pump [13], and where a liquid ring pump [14] is used it is usually necessary to use two mechanical boosters in
series to obtain pressure below 1 mbar.

Rotational speeds vary between 500 and 3440 rev/min and depend on the size of pump; the upper limit is dependent on
the supply frequency.

Forepump speed is normally between one-fourth and one-tenth the displacement of the mechanical booster so as to give
a relatively small pressure differential across the pump.
Overheating is a potential source of trouble with a mechanical booster causing loss of clearance which may result in
seizure. Operation up to about 90°C is generally permissible with cooling only applied to the shaft seal, bearings, and
gear box. The cause of overheating is that the gas in the foreline, heated as a result of compression,

< previous page page_160 next page >


< previous page page_161 next page >
Page 161

Fig. 3.9
Speed curve and power consumption of a 100-m3 · h1
multistage Roots pump at 50 Hz. The ultimate vacuum
improves when the pump is run at 60 Hz, and the
speed curve increases by about 20%. The speed
and power curves are without gas purge.

can flow back into the pump because there is no exhaust valve; it then undergoes further compression and heating [15].
This normally only becomes a problem when the inlet pressure is held between 2 and 40 mbar for long periods. With
larger-size pumps, an after cooler consisting of a water-cooled finned tube is inserted in the outlet as close as possible to
the rotor. Because the forepump is generally much smaller, start-up of the mechanical booster is normally at about 20
mbar, but with the use of a spring/gravity loaded bypass valve or a fluid coupling [16] it is possible to start the
mechanical booster at atmospheric pressure. In the case of a fluid coupling which varies the rotational speed of the
rotors, assistance is given during the roughing cycle and no overheating occurs when pumping a large volume from
atmospheric pressure.

System contamination is still possible from either the forepump or the gear box oil which is pumped via the outlet of the
mechanical booster [17]. This latter source of contamination is often overlooked. If an oil-free forepump is used to back
a mechanical booster, then to largely eliminate system contamination the gear box should be pumped separately.

The mechanical booster with a forepump provides a group capable of high pumping speed in the range 5 to 102 mbar.
In conjunction with a water ring pump it can handle fine powders, although coarse powders can cause wear.

Multistage roots pumps normally have three, four, or five stages of Roots mechanisms in series and are driven off the
same shaft. In the case of three-stage pumps it is normal to back them with a further three-stage pump, thereby giving
six in total.

Rotational speed is normally that of a two-pole electric motorthat is, 2850 rev/min (50 Hz) or 3440 rev/min (60 Hz).

Stage ratios are important to ensure efficient running at the normal inlet pressures. The inlet stage is the largest and the
outlet stage the smallest. A typical speed curve for a five-stage pump is shown in Fig. 3.9 with the typical power
requirements. The power starts low and then reaches a level which remains constant as the pressure drops.

< previous page page_161 next page >


< previous page page_162 next page >
Page 162

Overheating is a potential problem in the high-pressure stages when the pump is working at low pressure, and for this
reason the higher pressure stages are frequently fitted with water-cooled heat exchanges in their outlet port. Gas ballast
or purging is often used on a number of the stages to ensure that vapors being pumped do not condense. Frequently
there are gas purges at the shaft seals to keep them clean so as to improve their reliability.

System contamination is minimal because the gear box, which is at atmospheric pressure, is adjacent to the outlet of the
pump, or in the case of six-stage pumps it is at least three stages from the pump inlet. The bearings at the inlet stage of
the pump are normally the only source of contamination and are generally packed with a very-low-pressure
perfluoropolyether grease.

Dry pumping group controls often include thermocouples placed at critical points, outlet pressure and motor current
monitor, water flow switch, and so on. Frequently all these functions are monitored via a computer-compatible
controller so that a large number of units may be run on one in-house computer.

Dry pump outlet management [18] is an important factor in systems where aggressive toxic or explosive gases are being
pumped. The use of water scrubbers on their own is not generally advisable because water vapor and water can enter the
pump and, by combining with the pumped gases, can cause extensive damage. With the many gases used in the
semiconductor industry [e.g., halogen compounds (i.e., fluorine, chlorine, and bromine)], it is now becoming more
common to react and inert the pumped gases by the use of a number of different technologies, some of which are listed
below:

Burning/Oxidation: This needs to be combined with an absorber or wet scrubber to remove the by-products. It should
not be used for chlorinated materials due to risk of the formation of toxic by-products (e.g., dioxins).

Adsorption/Chemisorption: Possible explosive risk from adsorbed gases and hazardous waste disposal issues.

Hot Bed Reactors: Gases are passed into a heated granular bed, forming solid by-products, or are catalytically converted
by the bed into other gases or solids. By-products are nonhazardous or totally recyclable.

Application: Pumping of highly chemically active gases and the handling of fine dustthat is, semiconductor manufacture
plus where system contamination is a concern.

3.3.2
Claw Pump

Multistage Claw

Capacities Available: 25 to 500 m3/h (also including pumps with inlet Roots stage increasing to 1200 m3/h when used
with a mechanical booster)

Operating Pressure Range: 1000 to 104 mbar

With Additional Mechanical Booster: 1000 to 7 × 104 mbar

A typical claw mechanism is illustrated in Fig. 3.10. It consists of two claw-shaped rotors. These are synchronized by
external gears and rotate in opposite directions

< previous page page_162 next page >


< previous page page_163 next page >
Page 163

Fig. 3.10
The claw mechanism and its operating cycle.

within the stator. The gears and rotor bearings are oil-lubricated but are external to the pumping chamber, so the swept
volume is dry. The clearance between the rotors and the stator wall is normally 0.1 to 0.2 mm. As the rotors run dry
back leakage occurs through these clearances. The shape of the rotors and of the inlet and outlet ports is designed to act
as a valve, thereby almost eliminating back flow. This results in a much higher compression ratio at high pressure than
can be achieved with a Roots pump mechanism as indicated in Fig. 3.11. A further advantage is that the heat generation
between stages and at the outlet is much smaller than with the Roots-type mechanism, and interstage cooling on the
higher pressure stages is not required. The power consumption is also lower when the pump is operating at low
pressures.
Multistage claw pumps [19, 20] normally have three or four claw stages, although in some cases the inlet stage is of the
Roots design. The advantage of using a Roots

< previous page page_163 next page >


< previous page page_164 next page >
Page 164

Fig. 3.11
Attainable pressure ratio (for air) with a single claw and a single Roots
mechanism as a function of outlet pressure with no gas throughput.

inlet stage is that the inlet to a claw pump is more restrictive at low pressure than a Roots stage, so the use of the latter as an
inlet stage improves the speed of the pump at low pressures where the compression ratio of the Roots stage is somewhat higher
than that of a claw-type stage.

Rotational speed is normally that of a two-pole motorthat is, 2850 (50 Hz) or 3440 (60 Hz).

A typical speed curve of a multistage claw pump is shown in Fig. 3.12, along with a curve of its power consumption.

As with most other dry pumps, system contamination is minimal because the drive gears and oil box are at the atmospheric end
of the pump [21].

For gas ballast, system contamination, pumping group controls, pump outlet management, and applications, see previous
section.

3.3.3
Screw Pump

Capacities Available: 24 to 2700 m3/h


Operating Pressure Range: 1000 to 5 × 103 mbar

The mechanism of a screw pump consists of two intermeshing screw rotors enclosed in a close-fitting stator. There is a very
small clearance between the screw form of the two rotors, the rotors, and the stator wall; the stator has specially shaped ports for
the inlet and outlet. The rotors are normally coated to improve resistance to chemical attack and also to reduce the clearances
within the pumps to a minimum.

Two forms of screw are currently used; one type is of a rectangular form (see Fig. 3.13) and runs at about 10,000 rev/min, and
the other type is illustrated in Fig. 3.14 and runs at two-pole motor speed. Figure 3.15 shows a trapped volume of gas and its
transfer from the inlet to the outlet which is at atmospheric pressure. A typical speed curve is shown in Fig. 3.16 which also
indicates the power requirements.

< previous page page_164 next page >


< previous page page_165 next page >
Page 165

Fig. 3.12
Speed curve and power consumption of a four-stage
claw pump at 50 Hz. The ultimate vacuum improves when
the pump is run at 60 Hz and speed curve increases
by about 20%. The speed and power
curves are without gas purge.

Fig. 3.13
Screw pump with a square thread form.

This type of pump is particularly suitable for applications where slugs of liquid from the process vessel are likely to
reach the pump.
For gas ballast (or purging), system contamination, pumping group controls, pump outlet management, and applications,
see Section 3.3.1 on the Roots pump.

< previous page page_165 next page >


< previous page page_166 next page >
Page 166

Fig. 3.14
Alternative screw profile which is normally run at two-pole motor speed.
Fig. 3.15
Transport of gas by the square thread mechanism.

< previous page page_166 next page >


< previous page page_167 next page >
Page 167

Fig. 3.16
Speed curve and power consumption of a typical pump using a
screw form similar to that in Fig. 3.13 and without gas purge.
Obtainable pressure of pump type with the different screw form
(Fig. 3.14) and using two pole motor is about one decade higher.

3.3.4
Scroll Pump

Capacities Available: 20 to 50 m3/h


Operating Pressure Range: 1000 to 102 mbar

The two main components of this type of pump is a stationary scroll and an identical moving scroll. The moving scroll is
mounted on a spigot, or spigots, so that the rotation of the shaft produces an orbital motion of this member. The moving
scroll is constrained from rotating by an arrangement which ensures that it maintains the same angular position during the
orbiting motion. Such an arrangement is shown in Fig. 3.17, which indicates how gas is taken in at the inlet and transferred
to the outlet. The main problem in achieving a good vacuum is leakage across the edge of the moving scroll and the
stationary scroll. A spring-loaded gasket is normally fitted to ensure good sealing along the moving scroll to overcome this
problem [22]. However, this gasket can lead to the formation of a small amount of dust. So it is important to ensure that this
is not at any stage blown back into the system.

A typical speed curve and power consumption of a pump is shown in Fig. 3.18.

This pump is generally not suitable for severe or chemically active gas pumping applications but gives almost no
contamination when pumping clean systems or if it is used as a forepump (e.g., with a turbomolecular pump).

A special version of the pump has been used for many years, mainly in the nuclear industry. In this version the control of
the motion is made completely external to the system by the use of metal bellows. This ensures that there are no organic
components in the system that could be contaminated by the radioactive materials being pumped. Pumps of this type are
available in capacities ranging from 15 to 600 m3/h and with ultimate vacuum of 5 × 102 to 103 mbar.

< previous page page_167 next page >


< previous page page_168 next page >
Page 168

Fig. 3.17
Transport of gas in a scroll mechanism.
Fig. 3.18
Speed curve and power consumption of a 30-m3 · h1
scroll pump without gas purge.

< previous page page_168 next page >


< previous page page_169 next page >
Page 169

3.3.5
Piston and Diaphragm Pumps

Multistage Piston Pumps


Capacities Available: 12 m3/h
Operating Pressure Range: 1000 to 3 × 102 mbar

Fig. 3.19
Three-stage piston pump.

Two-Stage Diaphragm Pumps


Capacities Available: 0.8 to 5 m3/h
Operating Pressure Range: 1000 to 2 × 102 mbar
Fig. 3.20
Speed curve of a two-stage diaphragm pump.

< previous page page_169 next page >


< previous page page_170 next page >
Page 170

These two types of pump are basically of similar design. In the case of the diaphragm pump the diaphragm is used to
seal the piston while the piston in the piston pump either is a close fit in the piston housing or uses piston rings. This
results in the diaphragm design being more applicable to small capacities while the normal piston is used for the larger
sizes.

In vacuum applications the piston sealing is often obtained by a long path with a close clearance. The inlet valve is a
slot in the cylinder wall opened by the piston and the outlet valve is pushed open by the piston [23]. A three-stage pump
[24] is shown in Fig. 3.19. Three or four stages are used to achieve a vacuum of less than 0.1 mbar.

The diaphragm pump used in vacuum is normally a two-stage pump. The inlet and outlet valve are opened and sealed
by the diaphragm. The use of a molded diaphragm with strengthening ribs has greatly improved the reliability of this
component.

Gas ballast, pumping group controls, and pump outlet arrangements are not normally required. The main application of
the piston pump is for use on reasonably clean systems or as a forepump for a turbo pump. The diaphragm pump is
becoming more widely used, outside its normal laboratory application, as a forepump for a turbomolecular drag pump.

A typical speed curve for a two-stage pump is shown in Fig. 3.20.

References

1. F. A. Flecken, Gaede's influence on the development of mechanical vacuum pumps. Vacuum 13, 583 (1963).

2. W. Gaede, Gas Ballast Pumpen. Z. Naturforsch A. 2A, 233 (1947).

3. B. D. Power and R. A. Kenna, Vapour pumping characteristics of gas ballast pumps. Vacuum 5, 35 (1955).

4. L. Laurenson, Technology and application of pumping fluids. J. Vac. Sci. Technol. 20(4), 989 (1982).

5. L. Laurenson and G. Caporiccio, PerfluoropolyethersUniversal vacuum fluid. Proc. Int. Vac. Congr., 7th, 1977, Vol.
1, p. 263 (1977).

6. N. S. Harris and L. Budgen, Design and manufacture of modern mechanical vacuum pumps. Vacuum 26(12), 525
(1976).

7. D. Balfour, L. Budgen, P. Connock and D. Phillips, Pumping systems for corrosive and dirty duties. Vacuum 34, 771
(1984).

8. U. Segebrecht, Liquid Ring Vacuum Pumps and Liquid Ring Compressors, 1st ed. Expert-Verlag, Renningen-
Malmsheim, 1994.

9. H. Bannwarth, Liquid Ring Vacuum Pumps, Compressors and Installations, 2nd ed. VCH-Verlagsges., Weinheim.

10. H. E. Adams, Thermodynamic Characteristics of Nash Compressors. Nash Engineering Company, South Norwalk,
CT, 1953.

11. M. Wutz, H. Adam and W. Walcher, Theory and Practice of Vacuum Technology. Vieweg Verlagsges.,
Braunschweig, 1989.

12. J. L. Ryans and D. L. Roper, Process Vacuum System Design and Operation. McGraw-Hill, New York, 1986.

13. M. Bussard, La Technologie des pompes à vide système roots. Le Vide 12, 425 (1957).
< previous page page_170 next page >
< previous page page_171 next page >
Page 171

14. W. Armbatruster and A. Lorenz, Die Kombination Roots PumpeWasserring Pumpe. Vak. Tech. 7(2), 85 (1958).

15. C. M. Van Atta, Theory and performance characteristics of a positive displacement rotary compressor as a
mechanical booster pump. Trans. Natl. Symp. Vac. 3, 62 (1957).

16. H. Wycliffe and A. Salmon, The application of hydrokentic drives to high vacuum mechanical boosters (Roots
pumps). Vacuum 21, 223 (1971).

17. N. T. M. Dennis, L. J. Budgen and L. Laurenson, Mechanical boosters on clean and corrosive applications. J. Vac.
Sci. Technol. 18, 1030 (1981).

18. P. Mawle, Exhaust management. Eur. Semicond. May, p. 17 (1995).

19. H. Wycliffe, U.S. Pat. 4,504,201 (1985).

20. H. Wycliffe, Mechanical high vacuum pumps with an oil-free swept volume. J. Vac. Sci. Technol. A 5, 2608 (1987).

21. W. Wong, L. Laurenson, R. G. Livesey and A. P. Troup, An evaluation of the composition of the residual
atmosphere above a commercial dry pump. J. Vac. Sci. Technol. A 6, 1183 (1988).

22. D. Arthur, Little Inc., U.S. Pat. 1,507,254 (1976).

23. M. H. Hablanian, E. Bez and J. L. Farrant, Elimination of backstreaming from mechanical vacuum pumps. J. Vac.
Sci. Technol. A 5(4), 2612 (1987).

24. E. Bez, A compact oil-free rough vacuum pump. Solid State Technol., February (1995).

General References

L. Cummings, Liquid ring vacuum pumps optimise power generation. World pumps, January pp. 1415 (1987).

P. Duval, Will tomorrow's high vacuum pumps be universal or highly specialized. J. Vac. Sci. Technol. A 5(4), 2548
(1987).

R. G. P. Kusay, Vacuum equipment for chemical processes. Br. Chem. Eng. 16, 29 (1971).

J. F. Lennon, Vacuum process in drying of power transformer insulation. J. Vac. Sci. Technol., 19, April (1982); Proc.
Natl. Symp. of Am. Vac. Soc., Pt. 2, p. 1039 (1981).

P. A. O'Neill, Industrial Compressors: Theory & Equipment. Butterworth-Heinemann, London, 1993.

B. D. Power, High Vacuum Pumping Equipment. Chapman & Hall, London, 1966.

W. F. Ravette, Paper machine vacuum systems. TAPPI Wet End Oper., Semin. Notes, Seattle, WA, 1980, p. 97, TAPPI
Press, Atlanta, GA, 1980.

G. F. Smith, Machine vacuum systemhow to get the most out of this versatile papermaker's tool. TAPPI Eng. Conf.,
Pap., Houston, TX., 1976, Book 1, p. 243, TAPPI Press, Atlanta, GA, 1976.

A. P. Troup and N. T. M. Dennis, Six years of dry pumpinga review of experiences and issues. J. Vac. Sci. Technol. A 9
(3), 20482052 (1991).

M. Wutz, H. Adam and W. Walcher, Theory & Practice of Vacuum Technology. Vieweg Verlagsges., Braunschweig,
1989.
< previous page page_171 next page >
< previous page page_173 next page >
Page 173

4
Kinetic Vacuum Pumps

A kinetic vacuum pump is a pump that imparts momentum to the gas being pumped in such a way that the gas is
transferred continuously from the inlet of the pump to the outlet. This distinguishes it from the positive displacement
vacuum pumps described in Chapter 3 in which a volume filled with the gas being pumped is cyclically isolated from
the pump inlet and then transferred to the outlet. In the capture vacuum pumps described in Chapter 5 the gas molecules
being pumped are trapped on internal surfaces within the pump by sorption or condensation.

In the kinetic vacuum, pump momentum may be transferred to the gas being pumped in the direction of the pump outlet
by mechanical moving parts or by entrainment in a high-speed vapor stream. Both types of pumps are described in this
chapter.

Foundations of Vacuum Science and Technology, Edited by James M. Lafferty.


ISBN 0-471-17593-5 © 1988 John Wiley & Sons, Inc.

< previous page page_173 next page >


< previous page page_175 next page >
Page 175

Part I
Diffusion and Diffusion-Ejector Pumps

Benjamin B. Dayton

When a liquid such as water, mercury, or petroleum fluid is vaporized in a boiler and the vapor is conducted through a
diverging nozzle exiting into an evacuated chamber, the vapor expands through the nozzle and continues to expand
beyond the nozzle exit, acquiring a high forward mass velocity as the random molecular motion is converted into a
directed stream of molecules known as a vapor jet. If some gas is present in the chamber, the expansion beyond the
nozzle exit is limited to a certain extent and a boundary region is formed where gas mixes with the vapor stream. This
boundary region may take the form of a turbulent layer of mixed gas and vapor when the pressure is high enough, or at
lower pressures it may be a diffuse layer of vapor mixed with penetrating gas molecules moving in laminar flow. In
either case the entrained gas is moved forward in the direction of the vapor jet; and if some means can be found to
separate the gas and vapor without allowing the gas to find its way back into the region near the nozzle, then a pumping
action has occurred.

One means of separating the entrained gas from the vapor stream is to design the pump housing or chamber wall so that
the cross section narrows gradually to a throat in front of the vapor jet and then expands rapidly beyond this throat. Such
a structure is called a Venturi tube or ''diffuser," although no gas diffusion is necessarily involved. If the cross-sectional
area of the throat of the diffuser is about equal to the exit area of the nozzle and the gas pressure beyond the diffuser in
the region known as the "fore-vacuum" is not too high, the vapor stream will converge and accelerate through the
diffuser throat at such a density and mass velocity that gas molecules cannot

Foundations of Vacuum Science and Technology, Edited by James M. Lafferty.


ISBN 0-471-17593-5 © 1988 John Wiley & Sons, Inc.

< previous page page_175 next page >


< previous page page_176 next page >
Page 176

readily penetrate back through the diffuser throat. The vapor expanding beyond the throat is cooled and condensed,
thereby releasing the entrained gas which must then be discharged into another (backing) pump or into the atmosphere.
This is the principle of action of vapor-jet ejector pumps. While multistage steam-jet ejector pumping systems [1] have
been used in the past to produce pressure as low as 7 Pa while discharging the gas directly into the atmosphere, they are
seldom used today because of the availability of mechanical booster or blower pumps of various sizes.

Another method of separating the entrained gas from the vapor stream which can be employed when the gas pressure in
the fore-vacuum is sufficiently low and the vapor is readily condensable, as in the case of mercury or oil vapor, is to
cool the pump housing walls in the region where the main vapor stream encounters the wall. The cross section of the
pump at this point does not necessarily have to be about equal to the nozzle exit area since at low pressures the vapor
expands freely and fills the pump cross section at the level where condensation occurs. Since the vapor must move a
certain distance along the pump axis before it is entirely condensed, a vapor barrier is formed by forward-moving vapor
molecules through which only a few gas molecules can diffuse back from the fore-vacuum. This back diffusion is not
always negligible and may limit the performance of the pump when the gas molecules have a low molecular weight and
collision cross section, as for hydrogen and helium. Vapor pumps employing a cooled casing and low-vapor-pressure
working fluids are termed "diffusion" or "condensation" pumps. They require backing pumps or ''forepumps" to produce
the necessary fore-vacuum and can only operate when the gas pressure in the chamber to be pumped has already been
reduced to less than about 70 Pa.

Ejector pumps will not be considered here, but some hybrid pumps are still in use which combine oilvapor ejector
stages with diffusion pumping stages. The latter are known as "diffusion-ejector" or "vapor booster" pumps and are
described briefly in Section 4.2.

4.1
Diffusion Pumps

4.1.1
History of Development

The invention of the diffusion pump arose out of studies made by W. Gaede [2] in Germany in 1913 on the counterflow
of mercury vapor and air in a vacuum system pumped by his rotary mercury-sealed mechanical pump with a cold trap to
condense mercury vapor. During this study to determine the lowest partial pressure of gas that could be produced
beyond the trap, he discovered the effect of the counterdiffusion of air and mercury vapor on the reading of a McCleod
gauge and conceived of the pumping effect which might be obtained by allowing gas to diffuse through a narrow slit
into a rapidly moving stream of mercury vapor. Figure 4.1 shows his first embodiment of this concept. Note that the slit
region is surrounded by a water-cooling jacket to condense mercury vapor escaping through the slit.

In 1916 Irving Langmuir [3] showed that the slit or diffusion diaphragm through which the gas was pumped in the
Gaede pump [4] could be widened if the mercury vapor was directed away from the slit by a suitable nozzle as shown in
Fig. 4.2. He emphasized the importance of adequate condensation of the mercury vapor and referred to his pumps as
condensation pumps. The pumping speed was thereby greatly increased, and Langmuir was granted U.S. Patent
1,393,550 in 1921.

< previous page page_176 next page >


< previous page page_177 next page >
Page 177
Fig. 4.1
Gaede's slit-jet mercury vapor pump. a, return
tube; b, steel cylinder with slit e; c, condenser;
d, mercury-filled gutter; f, gas inlet; f1, connection
to roughing pump; g, connection to fore-pump;
h, thermometer; V, mercury cut-off valve; Q, boiler.

In 1916 Langmuir also applied for a patent on the inverted or "mushroom cap" nozzle design which was issued in 1919
as U.S. Patent 1,320,874 as shown in Fig. 4.3.

The detailed history of the development of mercury diffusion pumps is given in earlier editions and will not be repeated
here. Mercury vapor pumps became widely

< previous page page_177 next page >


< previous page page_178 next page >
Page 178

Fig. 4.2
Langmuir's mercury condensation pump with cylindrical nozzle.

used in the electronic tube industry in the United States from 1920 to 1940, but they required the use of refrigerated
traps to keep the vapor out of the tubes (except for mercury rectifier tubes). In 1928 in connection with experiments on
the vacuum impregnation of pressboard with transformer oils to improve the dielectric strength of insulators, C. R.
Burch [5] at Metropolitan-Vickers in England became interested in high-vacuum distillation at moderate temperatures,
known as "molecular distillation." He succeeded in obtaining some very-low-vapor-pressure fractions from the
petroleum oil used in rotary mechanical pumps. It then occurred to him to try these oil fractions as the pump fluid in a
diffusion pump in place of mercury. He was able to obtain a pressure reading on an ionization gauge of 104 Pa without
a cold trap and thus began a revolution in the application of diffusion pumps (British patent, 303,078).

After visiting England, Dr. Kenneth Mees, Research Director of the Eastman Kodak Laboratories, brought the work of
Burch to the attention of Dr. K. C. D. Hickman, who had been working with low-vapor-pressure synthetic organic esters
in place of mercury in special manometers for measuring the pressure in apparatus for
< previous page page_178 next page >
< previous page page_179 next page >
Page 179

Fig. 4.3
Langmuir's metal condensation pump
with umbrella nozzle.

the drying of photographic film where mercury vapor was harmful. In 1929 Dr. Hickman found that these esters gave
good results when used as the pump fluid in small glass diffusion pumps of simplified design which he constructed
himself [6]. He applied for a patent on the use of butyl phthalate, and similar esters, in place of mercury in these pumps
(U.S. patent 1,857,506 issued in 1932). Many articles were published in the period from 1932 to 1940 on the conversion
of mercury vapor pumps to oil diffusion pumps and the use of baffles and charcoal traps to reduce oil vapor
contamination.

During the period from 1929 to 1937 the factors involved in designing diffusion pumps for use with the new synthetic
oils were studied by Dr. Hickman and his co-workers in the Research Laboratories of the Eastman Kodak Company,
resulting in the development of multiboiler "self-fractionating" oil diffusion pumps [7] which extended the lowest
pressure attainable without cold traps from 103 Pa to about 5 × 106 Pa. Figure 4.4 shows a water-cooled three-stage
fractionating pump constructed of Pyrex glass with coiled nichrome heater wire in each of the boilers which in use are
covered with glass-wool-lined insulation bags. The total length is 25 in. Two alembic heads, or catchment lobes, are
provided in the tubing leading to the forepump connection for the purpose of retaining the most volatile fluids distilled
from the boiler.
< previous page page_179 next page >
< previous page page_180 next page >
Page 180

Fig. 4.4
Hickman's three-stage glass oil fractionating diffusion pump.
Fig. 4.5
Vertical multistage oil fractionating
pump design of Malter and Markuvitz [8].

< previous page page_180 next page >


< previous page page_181 next page >
Page 181

In 1937 L. Malter [8] of the Radio Corporation of America applied for a patent on an all-metal vertical multistage oil
diffusion pump, as shown in Fig. 4.5, which incorporated the Hickman self-fractionating principle by having separate
boiler compartments feeding vapor through concentric cylindrical tubes to the various nozzles with small openings
between compartments at the bottom of the cylinders so that returning condensed oil first enters the outer compartment
feeding vapor to the lowest stage where the more volatile components are removed and the purged oil then enters the
next compartment, where more of the volatile constituents are removed, and finally passes to the innermost
compartment feeding vapor to the nozzle nearest the pump inlet, so that the oil condensed on the walls near this top
stage has a lower vapor pressure than the original unpurified oil mixture. An alembic head is also shown in the forearm
to catch the more volatile components of the pump fluid.

Much of the history of diffusion pumps involves investigation of various synthetic oils for use as diffusion pump fluids
and attempts to minimize the backstreaming of the vapor of these oils from the first stage into the vacuum chamber.

4.1.2
Diffusion Pump Design

The first practical diffusion pumps were single-stage with a boiler at the bottom and a vapor conduit leading to a
cylindrical or expanding conical nozzle, or an inverted (umbrella-type) nozzle to form a high-speed vapor jet. In early
pumps the vapor jet was directed vertically upward into a condensing region with an alembic at the bottom to catch the
distilled mercury or oil and return the fluid to the boiler. Later designs employed nozzles which directed the jet slightly
downward, or vertically downward, so that the condensed pump fluid ran down the pump wall and returned to the boiler
through a connecting tube or through a narrow passage between the pump wall and a skirt on the vapor conduit so that
the head of oil prevented the escape of vapor from inside the conduit. Single-stage pumps with parallel multiple nozzles
were tried and gave improved speed and required less power input, but multiple nozzle designs are not practical in
multistage pumps.

Multiple in-line stages were introduced when it was realized that the compression ratio between forepressure and
maximum inlet pressure before jet breakdown had to be limited for each stage of an oil diffusion pump because of the
decomposition of organic pump fluids at elevated temperatures and pressures. Since the pumping speed of any stage
depends on the nozzle clearance area (the annular gap between the rim of the nozzle exit and the wall of the pump
housing) while the forepressure at which the jet breaks down depends on the boiler pressure (or temperature) and the
ratio of the nozzle throat area (narrowest cross section) to the "body clearance area" (the annular gap between the wall
of the pump housing and the vapor conduit leading from the boiler to the nozzle), by arranging a series of nozzles along
a common axis with stepped increases in the nozzle clearance area and body clearance area from the stage nearest the
primary pump to the stage nearest the vessel to be pumped, the overall speed of the pump could be maximized while
still maintaining operation against relatively high forepressures as produced by the primary pump without requiring an
excessively high boiler temperature.

The design of the first stage (nearest the pump inlet) was particularly important since some vapor from this stage
scatters backward into the high vacuum and influences the speed and ultimate pressure obtainable. In 1937, Embree [9,
10] applied

< previous page page_181 next page >


< previous page page_182 next page >
Page 182

Fig. 4.6
Design of Embree jet (b) compared with straight-sided jet (a) [10].

for a patent on a streamlined inverted nozzle design (as shown in Fig. 4.6b) which greatly improved the pumping speed
over prior designs of the type shown in Fig. 4.6a.

Since for multistage pumps the speed depends on the nozzle clearance area of the first stage (nearest the inlet), one
method of increasing the speed without changing the size of the flange connection to the vacuum chamber or the pump
length is to widen the pump housing in the neighborhood of the first- and second-stage nozzles (as shown in Fig. 4.7) of
a pump designed by Hablanian and Maliakal [11]. Also shown in Fig. 4.7 is a cowl or cap, cooled by conduction
through the supports mounted on the water-cooled pump casing, placed over the first-stage umbrella nozzle which
serves to intercept and condense vapor scattered backwards from the outer edge of the jet or evaporating from the rim of
the hot umbrella nozzle. This "cold cap," which may also be cooled directly by a loop of water-cooled copper tubing,
invented by Power [12] of the Edwards High Vacuum Ltd., greatly reduces the back-streaming of pump fluid into the
high-vacuum region.

A similar design, including a built-in butterfly valve at the inlet, is shown in Fig. 4.8. Both Figs. 4.7 and 4.8 show a
boiler filler and drain pipe at the side of the boiler and last-stage nozzle directing the vapor jet into a water cooled side-
arm in the form of an ejector diffuser. Water cooling of metal pumps is preferably through copper tubing wrapped
around the casing and brazed to the wall since double-wall cooling jackets as shown in Fig. 4.5 are less efficient and
subject to corrosion. Coiled tubing may also be

< previous page page_182 next page >


< previous page page_183 next page >
Page 183

Fig. 4.7
Metal multistage pump with enlarged nozzle clearance
area and cold cap [Varian Vacuum Products].

placed around the boiler region through which water can be circulated during the cool-down of a pump on a system
without a high-vacuum valve.

Proper design of the pump boiler region involves several details, such as (1) the size and spacing of boiler
compartments for fractionating pumps, (2) having a "skirt" at the lower end of the outer vapor conduit close to the pump
wall and of sufficient height to maintain a head of pump fluid (from the condensed fluid returning down the wall) to
balance the vapor pressure inside the boiler, and (3) the method of conducting heat from the electric heating units to the
oil in the boiler in such a manner that superheating and "bumping" is reduced.

4.2
Diffusion-Ejector Pumps

Before the development of efficient mechanical booster (rotary blower) pumps with adequate pumping speed down to
10 Pa, there was a problem in providing an adequate fore-vacuum for the diffusion pumps used on large molecular stills.
K. C. D. Hickman decided to develop a booster pump for the 5-Pa to 200-Pa pumping range based on the design of
steam ejectors but using a stable hydrocarbon oil at high boiler pressures instead of steam. This resulted in the
production of an oilvapor ejector or 'kerosene booster" pump which was then used on the molecular stills and in the

< previous page page_183 next page >


< previous page page_184 next page >
Page 184

Fig. 4.8
Cross section of three-stage oil-vapor diffusion pump
with integral water cooled baffle and high-vacuum
valve [Edwards High Vacuum International].
drying of frozen foods and blood plasma. Hickman was unable to obtain a patent on the basic idea of an oil-ejector
because of a prior invention by W. K. Lewis of an oil-ejector pump for the petroleum industry. However, U.S. patent
2,279,436 was issued to K. C. D. Hickman and G. Kuipers for an improved oil-ejector design, and a series of single-
stage and multistage oil-ejector pumps was developed and applied to use in the vacuum metallurgy industry. The
development of mechanical boosters (of the RootsConnersville blower type with rotating lobes) eventually obviated the
need for these oil-ejector pumps. However, hybrid ejector-diffusion pumps with the high-pressure stage in the form of
an oil-ejector were developed in the United States and in England and found use in various industrial applications where
the gas load involved relatively high pressures in the fore-vacuum produced by mechanical pumps.

< previous page page_184 next page >


< previous page page_185 next page >
Page 185

Fig. 4.9
Diffusion-ejector pump [Edwards
High Vacuum International].

Figure 4.9 is a diagram of such a diffusion-ejector pump [13] with a 38.7-cm-i.d. inlet which can operate against a 200-
Pa backing pressure and which has a maximum pumping speed of 4000 liter·s1 at inlet pressures up to 1 Pa using
Apiezon A201 booster pump fluid and 6.0-kW heater input. The boiler pressure can be as high as 4000 Pa.

Theoretical and experimental studies of the ejector stages of vapor pumps were performed by Jaeckel, Nöller, and
Kutscher [14], by Kutscher [15], and by Bulgakova et al. [16].

4.3
Performance of Vapor-Jet Pumps

4.3.1
Pumping Speed

The rate at which gas flows across the inlet of a vapor-jet pump can be expressed in at least four ways: (1) the number
of molecules crossing the inlet in unit time, (2) the mass of gas in grams per unit time flowing across the inlet, (3) the
volume of gas in volumetric units (liters, m3, cubic feet, etc.) flowing through the inlet in unit time at the prevailing
pressure and temperature at the inlet, and (4) the throughput (usually
< previous page page_185 next page >
< previous page page_186 next page >
Page 186

Fig. 4.10
Typical speedpressure curve of diffusion pump
as measured with a total pressure gauge.

represented by the letter Q) of gas across the inlet area in pressurevolume units per unit time at the prevailing gas
temperature. Historically, the precedent set by W. Gaede was to express pump performance in terms of the volumetric
flow rate or ''pumping speed" (usually represented by the letter S). The flow rates (1) and (2) are almost impossible to
measure directly. Flow rates (3) and (4) can easily be measured by procedures given in the next section, but differences
arise in the method of measuring the pressure (represented in the following by the letter p) depending on the type of
vacuum gauge used, the location and orientation of the gauge tubulation relative to the pump inlet, and the method of
connecting the pump to the test system. In any case, the speed at the cross section where the pressure is measured is
defined by

for a given gas at the specified temperature.

The equation of continuity for gas flow through the diffusion pump into the forepump is

where C is the "capacity" or volumetric speed of the forepump at the prevailing forepressure F.

Measurements of the speed of diffusion pumps using the test dome recommended by the American Vacuum Society
[17] (or by the equivalent ISO standard), along with a total pressure gauge, show that the speed varies with the intake
pressure as shown in Fig. 4.10. The speedpressure curve can be divided into three regions or zones of pressure: (1) the
declining speed in the "jet breakdown region," or forepressure breakdown region, from the breakdown value of the
intake pressure, pb, (corresponding to a forepressure F equal to the limiting forepressure, or breakdown forepressure,
Fb), to the crossover pressure, pc, where the inlet pressure equals the forepressure Fc, (2) the region of nearly constant
maximum speed, Sm, from pb to pd, the lower limit of pressure at which the speed is near Sm, and (3) the declining
speed in the ultimate pressure region from pd to the ultimate pressure, pu, at which the speed S becomes zero due to
back migration of pump fluid vapor and fragments of decomposed pump fluid, and back-diffusion of gas in the fore-
vacuum through the vapor jets.

< previous page page_186 next page >


< previous page page_187 next page >
Page 187

When the forepressure is maintained at a sufficiently low value to eliminate appreciable back-diffusion of gas from the
forevacuum and the outgassing of the measuring dome is negligible, the ultimate pressure, pu, is determined mainly by
the vapor pressure of the pump fluid on the walls near the pump inlet or on a cooled baffle or trap between the pump
and the test dome. In the region from pb to pu the pump speed is given approximately by

with apparent deviations from this formula during measurement being due to the varying response of the pressure gauge
(usually a tubulated ionization gauge) to the changing mixture of air and pump fluid vapor and the change in the
outgassing rate of the ion gauge itself.

In the region from pc to pb, when no gas is leaked into the fore-vacuum while the intake pressure p is increased by
admitting gas to the test dome, the pump speed is given approximately by

where C is the volumetric speed of the forepump at the prevailing forepressure and can be assumed to be nearly
constant over the forepressures prevailing during jet breakdown.

4.3.2
Limiting Forepressure for Maximum Speed

The limiting forepressure for maximum speed depends on the amount of gas admitted through the pump inlet and the
speed of the forepump. It is convenient to distinguish between (a) static breakdown, which occurs when the forepressure
has been raised by throttling the forepump at a given gas load or by admitting gas to the fore-vacuum until the inlet
pressure (high vacuum) has risen by about 2 × 102 Pa, and (b) dynamic breakdown, which occurs when gas is admitted
only through the pump inlet until the forepressure maintained by a given forepump has risen to the limiting value, Fb,
marked by a noticeable decrease in the pump speed (knee of the pump speed curve).

When the forepressure, F, is increased by opening an adjustable leak in the fore-vacuum while the high-vacuum
pressure, p, is held constant by adjusting the rate of air admitted through a leak into the test dome, the variation of speed
as a function of forepressure for a fixed value of p can be obtained and plotted for different heater inputs. For a given
heater input the speed remains constant until the forepressure reaches the limiting value and then decreases rapidly. The
maximum speed varies with heater input because of the vapor-backstreaming factor α and the fore-vacuum gas back-
diffusion factor β as explained in Section 4.4.

When the heater input is constant and the throughput is also maintained constant by controlling the leak into the test
dome and reading the intake pressure, p, while the forepressure, F, is slowly raised by admitting gas into the fore-
vacuum, a curve of inlet pressure versus forepressure at constant throughput is obtained. Two extreme values

< previous page page_187 next page >


< previous page page_188 next page >
Page 188

Fig. 4.11
High vacuum versus forepressure with constant
throughput and constant heater input.

of throughput are of special interest: the no-load value, or Q = 0, and the maximum load value, Q = Qm. Figure 4.11
shows such curves for a two-stage pump.

For a single-stage oil-vapor diffusion pump, as explained later, it can be shown that as an approximation, when no gas is
leaked into the fore-vacuum while gas is admitted to the pump inlet until p = pb at which dynamic jet breakdown
begins, then

where At is the nozzle throat area, B is the body clearance area (the open cross section of the pump casing in front of
forward-directed nozzles, or area between the inside wall of the pump casing and the outside wall of the vapor conduit
leading to an inverted or "umbrella"-type nozzle), Kn is a constant with value in the range from 0.5 to 1.1 depending on
the design of the nozzle as explained later, and Pc is the vapor pressure inside the vapor conduit leading to the nozzle,
which is approximately equal to the vapor pressure in the boiler. Since, as shown in Section 4.3.6, the boiler pressure
increases more or less linearly with the watt input to the heater, we note that pc, Fc, pb, and the limiting forepressure,
Fb, all increase linearly with the watt input. The knee of the speed versus intake-pressure curve for a single stage pump
can be moved to higher pressures, pb, by increasing the speed C of the forepump.

The static forepressure breakdown of single-stage pumps is given by


where Kn is nearly independent of the gas load (throughput) and Pc is approximately equal to the boiler pressure, but
for multistage pumps the static limiting forepressure depends on the gas load. When curves of p versus F for a two-stage
pump are plotted

< previous page page_188 next page >


< previous page page_189 next page >
Page 189

Fig. 4.12
Intake pressure versus forepressure for each stage in
two-stage pump and combined stages at no load.

together with the corresponding curves for each stage separately, as in Fig. 4.12, using the fact that the forepressure for
the first (top) stage, F1, is the same as the intake pressure, p2, for the second stage while the inlet pressure, p, equals p1,
it can be seen from the figure that the no-load static limiting forepressure for a two-stage pump is given approximately
by

where F1b is the no-load limiting forepressure for the first stage alone, while F2b is the no-load limiting forepressure
and F2c is the crossover forepressure for the second stage alone. At the upper end of the forepressure breakdown curve
for any individual stage the crossover pressure is given by

where At/Ae is the ratio of nozzle throat area to nozzle exit area. Hence, from Eqs. (4.6) and (4.8) we obtain
< previous page page_189 next page >
< previous page page_190 next page >
Page 190

where A2e/B2 is the ratio of nozzle throat area to the body clearance area for the second stage. Usually, for the stage
nearest the forepump, Ae/B = 0.3 0.5. Thus, for a two-stage pump, Eq. (4.7) gives Fb = 0.6F1b + F2b.

Similarly, it can be shown that for a three-stage pump the no-load static limiting forepressure is

where F3b is the static limiting forepressure and F3c is the crossover forepressure for the third stage (nearest the
forepump) alone. Again, F3b/F3c = A3e/B3, the ratio of nozzle throat area to body clearance area for the third stage.

The compression ratio of a single pumping stage is defined as the ratio of forepressure to inlet pressure when pumping a
gas that is not absorbed within the pump. This ratio is a variable quantity which depends on the ratio of the pumping
speed of a given stage to the net speed of the next stage in series. However, for a given throughput there is a limiting
compression ratio which can be maintained at a given power input. At maximum throughput of air the limiting
compression ratio for single-stage ejector pumps is about 10, and for the individual stages of multistage diffusion pumps
the limiting ratio is about 4. Thus a four-stage diffusion pump can compress air from 0.2 Pa to 51 Pa, corresponding to a
total compression ratio of 44 or 256. For a single-stage diffusion pump the limiting compression ratio as obtained from
Eq. (4.5) will be

4.3.3
Influence of Nozzle and Entrance Chamber Design on Speed

In order to simulate the gas flow pattern into the pump from a pipe connecting the pump to the vessel to be evacuated,
pump speed is always measured using a test dome over the inlet with an inside diameter either equal to the inlet
diameter of the pump housing or else equal to the inside diameter of the pipe fitting which is normally used to connect
the pump to the vessel. The American Vacuum Society has established recommended configurations for the test dome
[17] as described in Chapter 12. The location and orientation of the tubulation of the total pressure gauge installed on
the dome influence the measured speed [1820]. The nozzle design of the first stage of pumping (nearest the high-
vacuum inlet) determines the pump speed. For this stage the exit of the nozzle should not be beveled to deflect the vapor
jet toward the pump wall, but rather should direct the jet along the axis of the pump as much as possible by allowing
some inward expansion as in the Embree nozzle design. The "nozzle clearance area," Anc, between the rim of the vapor
nozzle exit (or the rim of a cold cap over an umbrella nozzle) and the inside wall of the pump housing in the entrance
chamber region is the limiting factor determining pump speed. The theoretical maximum speed across this area
considered as an opening in a thin plate would be

< previous page page_190 next page >


< previous page page_191 next page >
Page 191

where fn is the "effusion law" factor in liter · s1 · cm2 as given by

in which R0 is the molar gas constant, T is the absolute temperature, and Mn is the molecular weight of the gas.
However, because the nozzle clearance area is the exit of a tube or chamber of varying cross section consisting of the
walls of the pump casing and the nozzle assembly between the nozzle clearance area, Anc, and the pump inlet area, Ap,
gas does not cross the area Anc according to the cosine law but is beamed forward across this area by scattering from
the walls of this entrance chamber. If gas at pressure p (in Pa) were to enter the pump inlet from a large chamber
directly connected to the pump so that the molecules crossed the inlet area with a cosine law distribution, then there
would be a transition probability w less than 1 that a molecule crossing the pump inlet would also cross the nozzle
clearance area into the vapor jet. The forward flow in Pa · liter · s1 across the pump inlet would then be wfnApp. For gas
at a partial pressure pj scattered back from the vapor jet and crossing the nozzle clearance area (with an approximate
cosine law distribution) and continuing on out of the pump inlet, the backward flow in Pa · liter · s1 out of the pump
would be wfnAncpj. When the back-diffusion of gas from the fore-vacuum is negligible, the pump speed as measured in
the large chamber would then be

Theoretically, w could be calculated from the known geometry of the entrance chamber by applying various formulas
due to Clausing [21], Oatley [22], Harries [23], and others, or w could be calculated by Monte Carlo methods, but pj
would not be known. However, it is not necessary to know w and pj exactly, but merely to recognize that the entrance
chamber impedance, Rc (in seconds per liter), to gas flow should be made as small as possible while the nozzle
clearance area of the first-stage nozzle should be as large as practical to obtain maximum speed. The maximum
measured speed, Sm, can then be expressed as

where H is the "speed factor" or "Ho coefficient" which has been defined by T. L. Ho [24] as

where α is Gaede's factor for the fraction of the gas molecules from the high vacuum that succeed in diffusing through
the back-scattered vapor and reaching the nozzle clearance area which depends on the ratio of the mean free path of gas
molecules through the back-scattered vapor in the nozzle clearance area and the width of the nozzle clearance, while γ is
the fraction of the gas molecules which reach the nozzle clearance area and pass on to be carried away by the vapor jet
and 1 γ is the fraction which return through the nozzle clearance area. It should be noted that for mercury vapor pumps
the vapor pressure of mercury in the entrance chamber is not negligible,

< previous page page_191 next page >


< previous page page_192 next page >
Page 192

and Ho considered the chamber impedance Rc to include a factor which increased with increasing vapor density. In
deriving Eq. (4.16), Ho used the net speed relation

which is not strictly correct unless the molecules cross the various sections involved according to the cosine law (i.e.,
with no beaming).

Gaede [2] applied kinetic theory to the interaction of gas and vapor molecules in the neighborhood of the entrance slit
(or an equivalent orifice in a thin plate separating the mercury vapor stream from the gas in the high vacuum) of his first
mercury vapor pump and derived an equation for the speed at the slit area as shown in Section 4.4.1.

Speed for Various Gases. As can be seen from Eq. (4.12) and Eq. (4.13), theoretically, the speed of a diffusion pump
varies inversely as the square root of the molecular weight of the gas flowing across the inlet. However, back-diffusion
of light gases, such as hydrogen and helium, from the fore-vacuum to the high vacuum may decrease the pumping speed
when the vapor stream does not have sufficient velocity and density to reduce this back diffusion to a negligible value.
Theoretical equations for this back-diffusion are given in Section 4.4.4.

4.3.4
Ultimate Pressure

The ultimate pressure of a diffusion pump without a cold trap, in the absence of leaks through the system walls, is
determined primarily by the vapor pressure of the pump fluid at the temperature of the walls of the baffle or connecting
tube between the first stage and the vacuum chamber. Since organic pump fluids are a mixture of molecules of varying
molecular weight, fractionating diffusion pumps were developed by Hickman [7] and by Malter and Markuvitz [8]
which circulate the pump fluid through a series of boiler compartments from a compartment feeding the last-stage
nozzle to the compartment feeding the first-stage nozzle (nearest the pump inlet) with means of separating the vapor
supply to the nozzles. The more volatile constituents tend to be vaporized first in each compartment so that the last
compartment feeding the first stage receives oil with the lowest vapor pressure. Coe [25] studied the fractionating
effectiveness of concentric cylinder vapor conduits forming separate compartments in the boiler by raising the nozzle
assembly by small distances from the base of the pump casing and also demonstrated the value of a catchment lobe in
the fore-vacuum line to prevent volatile fractions from returning to the boiler. The variation of ultimate pressure with
ambient temperature for a fractionating pump with Octoil S, DC 703, and Apiezon C was also measured by Coe and by
Latham, Power, and Dennis [26].

When the forepressure is not sufficiently low, back-diffusion of light gases from the fore-vacuum may also influence
the ultimate pressure as explained in Section 4.4.4.

Vapor Pressure of Pump Fluids. In the first diffusion pumps, mercury was employed as the working fluid. The vapor
pressure of clean mercury at 25°C is about 0.27 Pa, and at 196°C the vapor pressure of liquid nitrogen is below 1031 Pa.

< previous page page_192 next page >


< previous page page_193 next page >
Page 193

Gallium and fusible metal alloys have been suggested as pump fluids, but are not considered practical. P. Alexander [26.
a] found glycerol as the pump fluid to give advantages over mercury vapor pumps, but glycerol has a vapor pressure of
about 102 Pa at room temperature.

Table 4.1 gives vapor-pressure data for oils used as pump fluids as obtained by various authors using tensimeters over
the pressure range from 0.1 to 1000 Pa. Since most commercially available pump fluids are mixtures of organic
molecules having a range of molecular weights and isomeric forms, the measurement of the vapor pressure is
complicated by the tendency of fractionation to occur during the measuring process. Therefore it is not surprising that
different methods give slightly different results and that the data reported by different authors for the same material are
not in close agreement as shown by this table. Parameters A and B are those of the vapor-pressure relation

where P is the vapor pressure in Pa and Tc is the temperature in °C. This equation is a good approximation to most
vapor-pressure curves over the range of temperatures involved in diffusion pumps, but plots of log10(7.5 × 103P)
versus the reciprocal of the absolute temperature obtained by actual measurements usually are not perfectly straight
lines but show a slight downward curvature. Vapor pressure measurements in the range from 1 to 103 Pa by molecular
beam [27] and dew-point [28] methods give higher values of B. The latent heat of vaporization of the liquid in kcal/
mole as calculated from

is listed in the last column. The values of temperature listed under the vapor pressure equal to 67 Pa are of especial
interest since this vapor pressure is about the optimum value for the boiler pressure in oil diffusion pumps. The ultimate
pressure of fractionating pumps will be about the same as the pressure listed in the column labeled P (Pa), 25°C, but, as
proved experimentally by Blears [29], the reading of a tubulated ionization gauge may be somewhat lower than the true
vapor pressure as measured by a nude gauge because of clean-up or sorption of the vapor inside the envelope of the
tubulated gauge, resulting in a pressure drop across the tubulation. In interpreting ultimate pressure readings by nude
ionization gauges using the "nitrogen equivalent" calibration factor, it must be borne in mind that the equivalent
nitrogen pressures of organic vapors may be 5 to 10 times higher than the true vapor pressure depending on the
ionization probability for the molecules of the vapor and its decomposition products within the gauge.

Octoil® is vacuum-distilled commercial grade di-ethyl-hexyl phthalate (molecular weight 390.3). Octoil-S® is vacuum-
distilled commercial grade di-ethyl-hexyl-sebacate (molecular weight 426.3). Apiezon B and Apiezon C are molecularly
distilled petroleum fractions. DC 704 and DC 705 are silicone fluids. Chemically, DC 705 is a pentaphenyl trimethyl
trisiloxane with molecular weight 546 g/mol [30]. DC 704 is a tetraphenyl tetramethyl trisiloxane with molecular weight
484 g/mol [31]. Convalex-10® is a vacuum-distilled polyphenylether of average molecular weight 454 g/mol. Santovac-
5 is a mixed penta-phenyl ether of average molecular weight 447.

< previous page page_193 next page >


< previous page page_194 next page >
Page 194

Table 4.1. Vapor-Pressure Data for Diffusion Pump Oils


Degrees C for P in Pa P (Pa) 25°C ∆H (kcal/mol)
Name Ref. A B 133 66.7 13.3 1.33 0.133 0.013
Octoil 1 5590 188 177 153 123 4.1 × 105 25.6
12.12 97 74
2 4808 193 180 152 117 2.0 × 104 22.0
10.31 88 63
4 5540 204 192 166 134 1.5 × 105 25.3
11.62 106 82
Octoil-S 3 5514 215 204 177 142 2.7 × 106 25.2
11.26 114 89
2 4878 214 199 169 133 6.1 × 105 22.3
10.02 102 75
4 5780 213 201 175 143 4.4 × 106 26.4
11.90 115 90
Convoil-20 7 4800 210 196 166 129 9.2 × 105 22.0
9.94 98 71
Convalex-10 2 5525 281 265 230 188 3.6 × 107 25.3
9.97 156 122
Santovac 5 6 5450 290 273 237 193 3.3 × 107 24.9
9.68 157 125
Lion A 2 4950 210 196 167 131 5.7 × 105 22.6
10.24 101 74
DC 704 2 5376 215 202 174 140 1.3 × 105 24.6
11.02 110 85
5 5700 223 210 183 149 3.2 × 106 26.1
11.49 149 94
DC 705 2 6098 250 237 209 174 2.1 × 107 27.9
11.65 143 116
5 6490 254 241 214 180 4.7 × 108 29.7
12.31 151 125
Apiezon B 2 4831 214 200 170 132 6.8 × 105 22.1
9.91 101 74
Apiezon C 2 4808 246 229 195 153 1.9 × 105 22.0
9.27 119 89
Cellulube 2 5780 225 212 185 152 1.7 × 106 26.4
11.61 122 97

References

1. K. Hickman, N. Embree and J. Hecker, Ind. Eng. Chem., Anal. Ed. 9, 264 (1937).
2. K. Nakayama, J. Vac. Soc. Jpn. 8, 333337 (1965).
3. Metropolitan-Vickers, S. Dushman, personal communication (1946).
4. E. Perry and W. Weber, J. Am. Chem. Soc. 7, 3726 (1949).
5. D. Crawley, E. Tolmie and A. Huntress, Trans. 9th Natl. Vac. Symp. 9, 399, Macmillan (1962).
6. K. Hickman, Trans. 8th Natl. Vac. Symp., 2nd Intern. Congr., 1961, Vol. 1, p. 307, Pergamon (1962).
7. B. Dayton, calculated from vapor pressure data of CVC.

< previous page page_194 next page >


< previous page page_195 next page >
Page 195

The application of polyphenyl ethers as diffusion pump fluids was introduced by K. C. D. Hickman [32]. Cellulube 220
is a triarylphosphate. Lion A is an alkylnapthalene oil. Convoil-20 is a vacuum-distilled petroleum fraction and has an
average molecular weight 388 g/mol.

The stability of the various pump fluids against exposure while hot to air at atmospheric pressure has been investigated
by several authors [30, 31, 3336]. In general, the silicone fluids are more resistant to thermal decomposition and
oxidation than the phthalate and sebacate esters and the paraffin hydrocarbon oils, but have the disadvantage that
silicone vapor migrating back into a vacuum system can decompose on hot filaments and electrodes bombarded by
charged particles to form silica-like insulating deposits which disturb the operation of electrical devices. The polyphenyl
ethers have much higher thermal stability than the ester fluids in addition to being resistant to damage by radiation.

The open-cup flash point of the ester pump fluids is about 200°C, and thermal decomposition results in formation of
crystals of the acid anhydride and tarry deposits on the nozzle assembly whereas the nozzle assemblies remain relatively
clean when silicone fluid is employed. If the exposure time is brief, fractionating pumps can usually recover in a short
time of normal operation, so that the speed and ultimate pressure are not seriously affected. However, dark deposits on
the nozzle assembly increase the radiation loss to the water-cooled walls and require an increase in heater input to
maintain the normal boiler temperature. Jaeckel [34] found that thermal dissociation of organic oils is favored by metal
surfaces, so that all-glass pumps tend to give lower ultimate pressures than metal pumps. In any case, it is recommended
that for ester and hydrocarbon pump fluids the oil in the boiler be allowed to cool below 100°Cand for polyphenyl
ethers, to below 150°Cbefore exposure to the atmosphere.

Perfluoropolyether has been investigated as a diffusion pump fluid and has the advantage of comparable or greater
stability but the disadvantage of somewhat lower speeds as compared to polyphenyl ether and silicone pump fluids [37,
38]. It has been used under the trade name Fomblin as the fluid in rotary mechanical pumps [39].

4.3.5
Backstreaming and Back Migration of Pump Fluid

Vapor molecules, originating from the pump fluid and its decomposition products, can escape back through the pump-
inlet cross section into the components of the vacuum system attached to the inlet. In 1957 Ruf and Winkler [40] and
also Power and Crawley [41] showed that these vapor molecules can have four sources: (1) back migration due to
evaporation of fluid condensed on the pump wall or on any baffles in the region between the first-stage nozzle clearance
area and the pump inlet; (2) backstreaming due to backward curving streamlines and scattered vapor molecules from the
vapor jet between the rim of the nozzle exit and the pump wall; (3) backstreaming due to evaporation from hot liquid
pump fluid clinging to the outside of the first stage nozzle; (4) backstreaming due to leaks around the nut holding the
top (umbrella) cap to the nozzle assembly. The invention of the cold cap by B. Power [12] greatly reduced the
backstreaming from sources (3) and (4). Examples of a cold cap installed over the top nozzle are shown in Figs. 4.7, 4.8,
and 4.9. In general, it is common practice to place water-cooled baffles and refrigerated or liquid-nitrogen-cooled traps

< previous page page_195 next page >


< previous page page_196 next page >
Page 196

between the diffusion pump and the vacuum chamber to reduce the backstreaming and back migration to an acceptable
level.

When the backstreaming rate is high, it can be measured during pump speed measurements with a test dome by having
the top of the dome dished convex or sloping slightly to one side so that condensed oil runs down the side walls into a
gutter at the bottom of the dome from which the oil can drain into a burette as described in the AVS Standards
Committee document T.S. 4.5 (1963). When the rate is very small, other techniques must be used, such as weighing the
oil collected on a thin metal foil clamped to a liquid-nitrogen-cooled plate above the pump and any baffles employed
[42]. Measurements of the backstreaming rate, B, (in cm3/h of oil) obtained by the author on the MC series of three-
stage vertical metal diffusion pumps having a top-stage nozzle with 3-mm throat width (without cold cap) at normal
heater input and ultimate pressure with either Octoil or Convoil-20 for casing diameters 12, 14, 16, 20, 32, and 48 in.
When plotted as a loglog graph gave a straight line corresponding to

where D is the inside diameter (in inches) of the straight cylindrical pump casing. The specific gravity of Convoil-20 is
0.90, so that 1.0 cm3/h = 15 mg/min. For the 12-in. pump we obtain B = 0.52 cm3/h, which corresponds to 0.011 mg ·
cm2 · min1. Power and Crawley [41] obtained a rate of 8.2 × 103 mg · cm2 · min1 on a 6-in. pump with Apiezon C and
no cold cap, while addition of the cold cap reduced this rate by a factor of 10. Duval [43] measured the backstreaming
from various source points within a 600-mm-i.d. pump and showed that 99.8% came from the lip of the top nozzle. He
also showed that the reduction in backstreaming produced by a cold cap depended on the position of the lower lip of the
cold cap relative to the lip of the top umbrella nozzle. Baker [44] has described a cooled quartz crystal microbalance
method for measuring backstreaming and showed that the peak backstreaming rate is as much as ten times the normal
running rate during warmup of the pump and approximately four times as high during cooldown. These high rates
during start-up and shut-down can be largely avoided by using the gas purge technique [45, 46]. Hablanian, using the
foil and liquid-nitrogen-cooled plate technique, obtained 1.6 × 103 mg · cm2 · min1 for a 6-in. pump with DC-705
pump fluid and a water-cooled cold cap. Addition of a liquid-nitrogen-cooled trap with no creep barrier above the pump
reduced the rate to about 6 × 106 mg · cm2 · min1. Interposing a water-cooled baffle between the pump and the liquid
nitrogen baffle gave a rate of 2.8 × 107 mg · cm2 · min1, and adding a creep barrier to this combination reduced the rate
to about 1 × 107 mg · cm2 · min1.

Design of Baffles and Traps to Condense Backstreaming Vapor. When a baffle or trap is installed directly over the
nozzle assembly in a water-cooled pump casing, the baffle plates can be cooled by conduction through copper, brass, or
aluminum support fixtures making thermal contact with the pump wall. However, such baffles must be removable so
that the nozzle assembly can be inserted and removed for cleaning. External baffles and traps are to be preferred since
they can be cooled independently of the pump casing to any desired low temperature.

< previous page page_196 next page >


< previous page page_197 next page >
Page 197

When backstreaming is relatively high, it is important that most of the pump fluid condensed on the first baffle nearest
the pump inlet drain back into the diffusion pump. This baffle is therefore usually air- or water-cooled, and the housing
is fitted to the diffusion pump in a manner that allows the condensed fluid to return. A second refrigerated baffle can
then be installed beyond the first baffle and operated at a temperature so low that the condensate does not drain away
except when this baffle trap is warmed again. The design of an effective baffle or refrigerated trap requires that there be
no line of sight through the baffle plates or the cold-trap reservoirs and housing (which is usually not refrigerated),
while the conductance for air should be as large as possible consistent with the ''optically tight" requirement. However,
even optically tight baffles or traps may be a source of vapor contamination since condensed pump oil tends to "creep"
along the inner surface of the baffle or trap housing and may reach regions at warmer temperatures where the oil can
reevaporate directly into the vacuum chamber. Thus, anticreep barriers are sometimes incorporated [47, 48].

The efficiency, Be, of a baffle in condensing backstreaming vapor depends on the number of collisions, n, made by a
vapor molecule on the cooled baffle plates and the condensation coefficient, αc, according to the equation

where for baffles with short passages between bends in the plates or between entrance and exit cross sections the
average number of hits with the wall for molecules passing through can be estimated from

in which L is the length (in cm) of the shortest passage from entrance to exit along a path equidistant from adjacent
baffle surfaces, b is the number of bends (usually right angle) in the passage, and Re is the equivalent radius (in cm) of a
straight cylindrical tube of length L having the same probability of transmission for noncondensing molecules as the
baffle. For long straight cylindrical tubes of length L and radius R, the number of collisions with the wall in free
molecular flow is [49]

and the transmission probability is [21]

For mercury vapor molecules colliding with water-cooled steel or aluminum baffle plates the condensation coefficient,
αc, is less than 0.5, but for the large molecules of organic or silicone pump fluids αc will be nearly unity, and most oil
molecules will be condensed within the baffle. The effectiveness of the baffle then depends on the rate of reevaporation
from the baffle surfacethat is, on the sorption lifetime, ta, on the surface. Assuming ta constant for the whole baffle
surface, the mean transit time through the baffle for vapor molecules will be

< previous page page_197 next page >


< previous page page_198 next page >
Page 198

where

in which R0 is the molar gas constant, and Tn is the mean absolute temperature of the vapor in transit which depends on
the temperature in the entrance chamber, the thermal accommodation coefficient, the temperature of the baffle plates,
and the number of hits with the wall after entering the baffle. For oil molecules at the usual baffle temperatures,
When the vacuum system is operated for times longer than tn, the flow of vapor molecules through the
baffle will reach a steady state and the baffle has no effect. The quantity of vapor passing through the baffle will then be
equal to the backstreaming flux from the pump into the baffle times the probability of passage for any molecule
(without regard to the time spent in the condensed state) in molecular flow. For long narrow passages the time to reach
the steady state can be several hours, which is the principle of the Alpert copper-foil trap for oil vapor [50].

Conductance of Baffles and Cold Traps. The conductance of a baffle or cold trap can be measured directly by admitting
gas from a throughput meter into a test dome, similar to that used for measuring diffusion pump speed, over the end
which will be furthest from the diffusion pump and placing a short pipe section (collar) with inside diameter equal to
that of the exit of the baffle, and having a gauge attached to a tube joining the pipe at right angles to the axis so as to
measure static pressure between the baffle and a diffusion pump. If the maximum speed, Sd, of the diffusion pump has
already been measured without the baffle over a wide pressure range, the pipe collar can be omitted and the net speed,
Sn, of pump and baffle measured over a wide pressure range with the standard test dome, the baffle conductance, U,
then being obtained from the net speed formula

The conductance may be calculated with or without entrance and exit corrections. The conductance with these
corrections corresponds to the throughput divided by the difference in pressure between two very large vacuum
chambers placed over the entrance and exit of the baffle, as measured in the interior of the chamber far from the
entrance and exit. Since baffles are seldom used in this configuration, it is the conductance without entrance and exit
corrections, corresponding to measurement of the static pressure exactly at the entrance and exit cross sections, which is
more relevant. Because of the location of the gauge in the AVS or ISO standard test domes, an entrance correction is
included in the measurement of pump speed or net speed of pump and baffle. As an approximation, this correction term
to the impedance can be assumed to cancel out when using Eq. (4.27), so that U is the conductance without entrance and
exit corrections. The entrance correction to a straight pipe section is about equal to one-half the impedance of a hole in a
thin plate of the same area and cross section as the entrance to the pipe as measured between two large chambers. The
actual calculation is complicated by beaming of gas from one pipe section across the entrance to an adjacent section and
the scattering of gas from any baffle plates or right angle bends [18, 51].

< previous page page_198 next page >


< previous page page_199 next page >
Page 199

Among the various types of baffle design which have been tried are (1) circular disk sandwiched between two flat
annular rings or located at the center of a cylindrical housing of about twice the diameter of the disk, (2) straight
chevrons, (3) concentric circular chevrons, (4) stepped flat annular rings, (5) stepped concentric circular chevrons, (6)
alternating semicircular plates inside cylindrical housing, and (7) split-chevron or louvre baffles. When backstreaming
is high, some alteration in the design of the baffle plates is needed to avoid condensed pump fluid dripping off a baffle
plate on to the top of the hot nozzle assembly in the diffusion pump. Zinsmeister [52] has treated very thoroughly the
characteristic properties of various baffle designs, including cold caps over the top nozzle. He defines a specific
conductivity as the total conductivity of the baffle divided by the area of the cross section of the baffle housing alone
and obtains 3.6 liter · s1 · cm2 for typical single-chevron baffles. The specific conductivity of optically tight baffles is
relatively independent of the diameter of the baffle housing. Davis, Levenson, and Milleron [53] have investigated the
optimum dihedral angle for the chevrons in a straight chevron baffle for maximum conductance both by experiment and
by Monte-Carlo calculation. Their results indicate an optimum angle of 120°.

4.3.6
Throughput

In vacuum engineering calculations the quantity of gas flowing through the pumping system is usually expressed as the
throughput (symbol Q), defined as the product of pumping speed, S, and the static pressure, p, at the prevailing gas
temperature at a specified cross section. It is recommended that speed be expressed in either liters per second (liter·s1)
or cubic meters per second (m3·s1). The ISO unit of pressure is the Pascal (abbreviated Pa); but current practice also
allows use of the millibar (abbreviated mbar), which equals 100 Pa. The unit Torr is no longer recommended. One
millibar equals 0.75 Torr. Suitable units of throughput are mbar·liter·s1, Pa·liter·s1, or Pa·m3·s1 at a specified
temperature (such as 23°C).

As pointed out by Lewin [54] and by Ehrlich [55], the above definition of throughput leads to complications when the
gas temperature is not the same in all parts of the vacuum system. However, by correcting all measurements to a
standard room temperature of 23°C and using the continuity relations,

where U12 is the conductance of the flow path between the points 1 and 2 at which the static pressure is measured,
vacuum system calculations are simplified. The conductance, U12, in the molecular flow region is independent of the
wall temperature [49, 56, 57]; but the entrance and exit pressures, p1 and p2, depend on the gas temperature and, when
measured by hot-filament ionization gauges, also depend on the wall temperature of the gauge [58, 59].

Each stage of a multistage vapor-stream pump has a maximum throughput which can be handled without jet breakdown
determined by the heater input and the speed of the next pumping stage in series. From Eq. (4.5) the maximum
throughput is

< previous page page_199 next page >


< previous page page_200 next page >
Page 200

where the subscript n refers to the number of the stage (the first stage being nearest the pump inlet and the last stage
nearest the forepump connection). For the last stage of the vapor-stream pump, the next pumping stage is that of the
forepump; and Sm(n + 1) = C, the speed of the forepump. When the nozzle assembly has optimum design, the value of
Qm for a given boiler pressure, P0, should be approximately the same for each stage in series. The required speed, C, of
the forepump then depends on the characteristics of the last stage of the vapor pump as given by Eq. (4.5). When the
speed of the forepump is sufficiently high, the limiting forepressure, Fmb, at this maximum throughput can be
determined by measuring the forepressure, F, as gas is admitted to the forevacuum region until jet breakdown begins, as
indicated by a rapid rise in the high-vacuum pressure, p, when the leak rate of gas on the high-vacuum side is
maintained constant at a series of increasing values.

Heater Input and Water Cooling. Since according to Eq. (4.29) the maximum throughput is proportional to the boiler
pressure, it is of interest to determine the dependence of the boiler pressure on the watt input to the pump heaters, the
heat carried away by the cooling water, and the radiation and conduction losses from the pump casing. The heat carried
away by the cooling water per second (in watts) can be estimated from

where we have used Eq. (4.53) below and

in which P0 is the vapor pressure (in Pa) and T0 the absolute temperature in the boiler, M is the mean molecular weight
of the pump fluid, R0 = 8.314 × 107 erg · K1 · mol1, Σ Ant is the sum of all the nozzle throat areas in an n-stage pump,
Lv is the latent heat of vaporization per gram of pump fluid, s is the specific heat of the liquid pump fluid, T0 is the
stagnation temperature of the vapor from the nozzles, Tw is the lowest temperature reached by the cooled condensed oil
on its way back to the boiler, D is the casing diameter (in cm), Lbw is the distance (in cm) between the level of fluid in
the boiler and the lowest turn of the cooling coil, kc is the thermal conductivity of the casing material, and t is the
thickness of the casing wall (in cm), Wr is the rate of heat transfer (in watts) by radiation from the hot nozzle assembly
to the cooled casing in the region of the cooling coils, Aj is the area (in cm2) of the outside surface of the hot nozzle
assembly directly exposed to the casing in the region of the cooling coils, ej is the emissivity and Tj is the average
temperature (in K) of this surface which will be slightly less than the vapor temperature, T0, in the boiler, and Tac is the
average temperature (in K) of the inside of the pump casing in the region of the cooling coils which will be somewhat
higher than Tw and Wc is the rate of heat transfer by conduction along the casing wall from the boiler to the lowest turn
of the cooling coil. Heat transfer from the ambient air to the cooling coil is neglected. The

< previous page page_200 next page >


< previous page page_201 next page >
Page 201

heat exchange between the casing wall and the returning pump fluid over the distance Lbw is also neglected.

For organic pump oils, Lv is in the neighborhood of 6070 cal·g1 and s is about 0.2 cal·g1·deg1. A more exact treatment
would use the emissivity factors for concentric cylinders and consider the heat conduction through the film of pump
fluid on the casing wall; but since emissivity of the pump fluid is nearly equal to 1 and the film is thin, the above
equation is sufficient for a rough estimate. The value of ej will vary from about 0.2 to 1.0, depending on the degree of
discoloration of the surface of the nozzles assembly. It has been found that the radiation loss is somewhat less than 10%
of the heater input when the nozzle assembly is clean.

Finally, the required heater input to the boiler will be

where Wb is the rate of heat loss by radiation and air convection to the surroundings of the boiler region up to the lowest
cooling coil. The boiler region in some pumps is thermally insulated with fireproof insulating material. The electric
heating units may be installed as hot plates in direct contact with the flat bottom of the boiler or may take the form of
cartridge units inserted in cavities which extend into the pump fluid. The latter arrangement reduces the "bumping"
which sometimes occurs as the oil becomes superheated and flashes into vapor. This bumping can cause vibrations
which create problems in apparatus attached to the pump, such as electron microscopes.

The heat transferred to the cooling water has been measured to be between 55% and 75% of the heater input W for
external heaters and between 65% and 80% of W for immersion heaters. The rate of flow of cooling water through the
coils should be adjusted so that the discharge temperature is 3545°C to maintain a relatively warm temperature of the oil
descending along the wall from the last coil turn to the boiler. This helps to prevent condensation and return to the
boiler of light ends of the pump fluid which normally undergoes some decomposition during pump operation. As a rule-
of-thumb, for pumps with external heaters the cooling water should circulate through the coils from the high-vacuum
toward the fore-vacuum side at a rate (in cm3/min) about equal to the heater input in watts.

While the vapor pressure, P0, during operation of diffusion pumps may vary over the range from about 10 to 100 Pa,
depending on the heater input, W, the absolute temperature, T0, of the vapor in the boiler varies only from about 440 to
480 K for typical pump oils as shown by Table 4.1. The variation in in Eq. (4.30) is therefore only from 21.0 to
21.9, and for all practical purposes it can be assumed that the boiler pressure, P0, is a linear function of the heater input
W. This is confirmed by the experimental data.

To ensure a supply of vapor to the nozzles at a pressure near that in the boiler and to avoid overheating the fluid in the
boiler, the area of the free surface of the pump fluid in the boiler should be at least five times the total nozzle throat
area, ΣAnt. In the forepressure breakdown region from pb to pc the vapor jet is deformed and cooling is less efficient, so
that the boiler temperature tends to rise to dangerous levels. Also, the deformation of the first-stage vapor jet will
increase the backstreaming rate. Oil

< previous page page_201 next page >


< previous page page_202 next page >
Page 202

diffusion pumps should not be allowed to operate in the forepressure breakdown region for any length of time, or else
the heater power should be turned off.

4.4
Theory of Pump Performance

The theoretical analysis of the pumping action of diffusion and ejector pumps has engaged the attention of many
authors. Kinetic theory can be applied to analyze the counterdiffusion of gas from the high-vacuum side and vapor
spreading sideward and backward from the nozzle exit as well as the counterdiffusion of gas from the fore-vacuum and
vapor discharged in the forward direction from the nozzle. Hydrodynamic theory, as developed in connection with
steam turbines, can be applied to the expansion of the vapor through the nozzle and beyond the nozzle exit. Theoretical
equations have been derived for the following aspects of pump performance: speed, limiting forepressure, ultimate
pressure, and backstreaming.

4.4.1
Speed

In 1915 Gaede derived a theoretical equation for the speed of a diffusion pump by analyzing the counterdiffusion of gas
in a mercury-vapor stream as follows. First, he considered the case of a stream of mercury vapor flowing in the axial
direction (x-axis) through a cylindrical tube of radius a with average velocity v and a density high enough to be in the
viscous flow region so that entrapped gas molecules diffusing countercurrent to the vapor stream tend to be carried
along by the vapor stream. In the steady state the concentration n of gas molecules at x varies linearly from a very small
value n0 at x = 0 to a maximum nm at x = L because of a reservoir of gas beyond L from which gas molecules can
penetrate into the vapor stream and diffuse counter to the stream velocity with diffusion coefficient D. For L/a < 10 the
velocity u will be approximately constant over any cross section of the tube. Since n is proportional to the pressure p of
the gas and D varies inversely with the pressure P (in mmHg) of the mercury vapor, Gaede obtains

where D is the diffusion coefficient at 760 mmHg (101,308 Pa), and dp/dx is the pressure gradient of gas along the tube,
which is equal and opposite to the pressure gradient, dP/dx, in the vapor. Integrating Eq. (4.33) from p = p0 to p = pm
gives

The counterdiffusion of gas and vapor results in an actual pressure drop in the gas from x = L to x = 0 given by

< previous page page_202 next page >


< previous page page_203 next page >
Page 203

As Gaede pointed out, this introduces an error in the reading of gas pressure with a McLeod gauge. For a mercury vapor
stream of sufficiently high pressure P and velocity u the gas pressure at x = 0 can be reduced to a very low value, p0 =
βpm.

Gaede then considered the case of mercury vapor flowing past a small hole of diameter d in a thin wall separating the
main body of vapor from a region containing gas at the pressure p. From kinetic theory the number of vapor molecules
crossing a small surface element σ normal to the radius from the hole and located at the angle ϕ with the normal to the
plane of the hole will be

where N is the number of vapor molecules per unit volume in the vapor stream near the hole coming from the source
side and ua is the average molecular velocity at the temperature T. Since the emerging vapor molecules do not actually
start at a point at the center of the hole, Gaede introduces a spherical surface of radius x0 at which the concentration is
N. Since the total number of vapor molecules crossing the hemisphere of radius x0 per second will be equal to the
number Z crossing the hole of diameter d, it follows that the concentration N will be equal to Z divided by the volume
or

so that The number of vapor molecules per unit volume, N(x, ϕ), at σ will be

The mean free path, λ(x, ϕ), of the gas molecules through the vapor at x and ϕ can be assumed to be inversely
proportional to the concentration N(x, ϕ) of the vapor molecules since the gas pressure is small compared to the vapor
pressure, or

where λ0 is the mean free path of a gas molecule through the vapor at the hole, where x = x0. Gas molecules diffuse
through this mercury vapor toward the orifice and are scattered so that only a fraction, α, of the molecules finally reach
the orifice depending on the mean free path, λ(x, ϕ), for the gas molecules through the vapor as given by

By integrating over all angles, Gaede obtains

For λ0 = d, this gives α = 0.793.

< previous page page_203 next page >


< previous page page_204 next page >
Page 204

Then Gaede's complete formula for the speed at the entrance slit is

where A is the slit area, and β is the back-diffusion coefficient for gas from the fore-vacuum at the forepressure F which
penetrates back through the vapor jet by counterdiffusion and generates a partial pressure

of this gas at the entrance slit area, while α is the fraction of the gas molecules from the high-vacuum side that penetrate
through the backstreaming vapor and reach the slit area as given by Gaede's Eq. (4.41), and γ (using Ho's γ notation in
place of Gaede's k) is the fraction of the gas molecules which have reached the slit area that actually pass through and
are carried away by the vapor stream. The coefficient β is given by Eq. (4.34) when P, v, and L are constant, but for
vapor jets diverging from a nozzle in modern pumps these parameters are not constant and integration must be
performed over a path through the vapor jet from the fore-vacuum to the high-vacuum side. Experimentally, β
corresponds to the slope of the p versus F curve for no load in Fig. 4.11.

As explained below, Molthan [60] modified Gaede's equation for α by taking account of the effect of the stream
velocity, v, of the vapor molecules on the pattern of flow of vapor molecules emerging through the hole, or slit.

4.4.2
Limiting Forepressure

As the forepressure is increased by admitting a gas leak into the fore-vacuum with a constant gas throughput on the inlet
side, a shock boundary is formed between the supersonic vapor jet and the gas in the fore-vacuum. Increasing the fore-
pressure further pushes this shock boundary back toward the nozzle exit as shown by Dayton [61] in 1948 in
experiments with a glass diffusion pump using pump fluid colored with a red dye and observing a glow discharge in the
gas and vapor in front of the nozzle. When the shock boundary is pushed back to the point at which the jet no longer
makes contact with the cooled pump casing, the vapor jet again shoots forward as a narrow beam into the fore-vacuum
but has a side boundary with the gas at the elevated pressure and eventually bends to one side to condense on the pump
wall.

This behavior is also shown for the case of a metal pump by the isobar diagrams obtained by a pressure probe moved up
and down through the jet of a mercury vapor pump by P. Alexander [62] (see Fig. 4.13) and by the glow discharge
photographs and line drawings in a 1955 article by Kutscher [15]. Similar diagrams were prepared in 1960 by N.
Florescu [63].

The gas density distribution around the upper stage of a 6-in. vertical metal pump at an inlet pressure of 5.3 × 102 Pa
was explored by Hablanian and Landfors [64, 65] using an ionization gauge probe. They also measured the distribution
of vapor arriving at the wall from the top nozzle (both with and without a cold cap) by using a series of gutters and drain
tubes.

< previous page page_204 next page >


< previous page page_205 next page >
Page 205

Fig. 4.13
Isobars of air pressure in mercury vapor pump [62].
4.4.3
Vapor-Jet Flow Pattern

When the average stream velocity is parallel to the nozzle axis and both the throat area, At, and the exit area, Ae, are
orthogonal to this axis and no condensation occurs

< previous page page_205 next page >


< previous page page_206 next page >
Page 206

Fig. 4.14
Expanding conical nozzle showing various
defined cross sections.

in the nozzle, then the mass of vapor discharged by an expanding conical nozzle is given (in cgs units) by

where ρt is the vapor density, Pt is the static vapor pressure in dyne·cm2, vt is the stream velocity in cm/s, Tt is the
absolute vapor temperature at the nozzle throat of cross-sectional area At, and the subscript e indicates the
corresponding quantities at the nozzle exit (see Fig. 4.14). It is known that for adiabaticisentropic expansion of vapor
through the nozzle when the forepressure is below the limiting value, the velocity vt at the throat equals the acoustic
velocity,

where k is the ratio of specific heats. Since any change in downstream pressure cannot influence the throat pressure
when the flow is supersonic, the discharge rate is a maximum.

In passing adiabatically from the cross-sectional area Ac of the vapor conduit from the boiler to the throat area At the
vapor density and pressure changes according to

where ρc is the density of the vapor in the conduit from the boiler. For sufficiently large Ac relative to the throat area,
At, it can be assumed that Pc is approximately equal to the boiler pressure, P0, and ρc is approximately equal to the
vapor density, ρ0, in the boiler.

When the forepressure is below the limiting value, the discharge rate is a maximum and the static vapor pressure at the
throat cannot be increased beyond the critical value,
where Ptc will be in Pa if P0 is in Pa.

< previous page page_206 next page >


< previous page page_207 next page >
Page 207

Combining the above equations gives for the maximum discharge rate (in cgs units) of short nozzles

For adiabaticisentropic expansion through a short nozzle the static pressure, Px, the vapor density, ρx, and the absolute
vapor temperature, Tx, at any cross section x are related to the initial pressure Pc, density ρc, and temperature Tc in the
vapor conduit by

There are no experimental data on the value of k for the high-molecular-weight organic and silicone molecules of
typical diffusion pump fluids, but from the equipartition theorem as modified by quantum mechanics it can be assumed
that for all temperatures above 300 K the specific heat ratio can be estimated from

where Nv is the number of vibrational degrees of freedom which are fully excited at the temperature of the vapor in the
boiler [66]. A study of the available data on the value of k for large organic molecules reveals that in the temperature
range 30°C to 300°C the value of k will be about 1.5 times the number of atoms in the molecule [67]. The theoretical
maximum value is 3n 6, where n is the number of atoms in a molecule (with n > 3), but this value cannot be reached
except at extremely high temperatures where in most cases the molecule decomposes. The pump fluid Octoil (dioctyl
phthalate) contains 66 atoms, so that Nv = 99 and k = 1.019. For typical hydrocarbon pump fluids with molecular weight
about 390, the molecules have about 28 carbon atoms and 54 hydrogen atoms and Eq. (4.50) using the 1.5 factor gives k
= 1.015. It can be assumed, as an approximation, that k = 1.02 for organic pump fluids. The specific heat ratio for
mercury vapor is k = 1.666.

The above equations involve the assumption that the initial stream velocity, vc, in the conduit of cross-section area Ac is
practically zero. This will be true when but for some diffusion pump designs Ac is less than 5At and a factor

should be included in Eq. (4.48). Neglecting this correction factor for the initial velocity and assuming that P0 at the
nozzle entrance is about equal to the boiler pressure and using k = 1.02, Eq. (4.47) gives the critical throat pressure Ptc
= 0.60P0 and Eq. (4.48) gives

< previous page page_207 next page >


< previous page page_208 next page >
Page 208

for oil diffusion pumps with forepressure below the limiting value corresponding to a maximum discharge rate. Since
the vapor density, ρ0 (in cgs units), is not measured directly but calculated from

where M is the average molecular weight of the pump fluid, R0 is the molar gas constant, and T0 is the absolute
temperature of the vapor in the boiler region, a more convenient form of Eq. (4.51) is

As the vapor passes from the throat to the nozzle exit of area Ae > At, it expands and acquires supersonic stream
velocities. When the forepressure is below the limiting value and the initial velocity, v0, is negligible,
adiabaticisentropic flow through a short expanding conical nozzle results in an average stream velocity (parallel to the
nozzle axis) at the nozzle exit given by

where Pe is the static vapor pressure at the exit. For organic pump fluids with values of k approaching 1, it can be
shown by an application of L'Hôpital's Rule [67] that the nozzle discharge rate is given by

Differentiating G with respect to Ptc and setting the result equal to zero gives the critical pressure at which G is a
maximum,

which is approximately the value found above. Substituting in Eq. (4.55) then gives

which is the same as Eq. (4.53). Similarly, applying L'Hôpital's Rule to Eq. (4.54) as k → 1, we obtain

Then, using Eq. (4.45) and Eq. (4.63) below with k = 1, the ratio of exit to throat velocity is

< previous page page_208 next page >


< previous page page_209 next page >
Page 209

Combining Eq. (4.44), Eq. (4.56) and Eq. (4.59) using gives

Since this equation cannot be solved explicitly for Pe, but Pe/P0 < 1, then by successive approximations we obtain for
the static vapor pressure at the exit of a short expanding conical nozzle:

where e is the base of the natural logarithms and the last P0/Pe has been replaced by as an approximation. For a
cylindrical nozzle, At = Ae and the static pressure at the exit is

where Pt is the throat pressure as given by Eq. (4.56). In this case the exit velocity, ve, equals vt as given by Eq. (4.59).

When the forepressure is below the limiting value, the vapor temperature at the throat is

and for organic vapors with k = 1.02 one can assume Tt = T0, the boiler temperature, so that the acoustic velocity, vs, at
the throat is essentially the same as the acoustic velocity at the boiler temperature,

When the forepressure is below the limiting value, the vapor diverges from the nozzle exit at supersonic velocities. The
central portion of the vapor jet can be assumed to follow the hydrodynamic equations for streamline flow developed by
Prandtl and others for steam jets [68]. However, the outer regions of the vapor jet expand to such low pressures that
continuum physics cannot be applied, and the vapor expands freely according to the laws of kinetic theory so that the jet
pattern in the outer fringe region can be estimated by superimposing a downward stream velocity on the random
direction and thermal velocity distribution according to Maxwell's equation.

Figure 4.15 is a modification of a diagram first presented by Molthan [60] which shows that a hemispherical surface of
radius u, corresponding to some fixed value of the thermal molecular velocity for molecules escaping through a hole c
in a wall between the vapor source and a high vacuum with zero initial stream velocity, becomes displaced in the
direction of a superimposed vapor stream velocity, v, of the vapor on the other side of the wall. Choosing, as Gaede did,
a circular hole with

< previous page page_209 next page >


< previous page page_210 next page >
Page 210

Fig. 4.15
Displacement of vapor stream pattern
diverging from slit c for fixed thermal velocity,
u, by superimposing a stream velocity, v.

diameter d, then when the mean free path of a vapor molecule in the vapor at the hole is less than 10 times the diameter,
d, Molthan assumed that the vapor does not emerge from the hole according to the cosine law but rather emerges with
equal probability in all directions so that the density of molecules at any point on the hemispherical surface of radius u
will be a constant.

The magnitude of the thermal velocity, u, varies from 0 to ∞ with the probability, w(u), that the velocity lies between u
and u + du being given by Maxwell's distribution law:

where the unit of the velocity u is chosen as the most probable velocity

As a result of the translation of the surface of a given radius u in the direction v, the flux per unit solid angle in different
directions is no longer the same. Let a line (which will be termed the x-axis) be drawn on the surface of the wall through
the center of the hole and parallel to the stream velocity v, then any plane through this line cuts out two semicircles on
the two hemispheres as shown in Fig. 4.15. The intersection of this plane (which will be called the ''diagram plane")
with a plane through the center of the hole and perpendicular to the stream direction determines the line . To
calculate the new density distribution on the displaced hemisphere, consider a small arc on the initial semicircle
which is displaced to the position of equal arc length, on the displaced semicircle. A plane passed through the
line orthogonal to the diagram plane will determine a dihedral angle θ with the plane through perpendicular to
the stream direction, and a plane passed through orthogonal to the diagram plane will form the dihedral angle θ'
with the plane through . Now let the arc represent a small surface element, AB, cut from the hemisphere by the
planes forming the dihedral angle and planes through x-axis. The solid angle determined by the surface element AB will
now be greater than the solid angle at c determined by the surface element A'B' of area equal to that of AB. As shown in
the diagram, let the latter solid angle also determine the surface element ab on the hemisphere of radius u. The vapor

< previous page page_210 next page >


< previous page page_211 next page >
Page 211

stream flux per unit solid angle will then be greater in the direction of A'B' (or ab) than in the direction of AB by the
ratio A'B' : ab. To find this ratio, form the intersection of all the rays within the solid angle A'B' and a sphere of radius
which cuts the ray at B". Then the spherical surface element of area A'B" determines the solid angle

while the element AB determines the solid angle AB/u2 which is numerically equal to A'B'/u2. For infinitesimal surface
elements we have

where ψ is the dihedral angle B'A'B" which is numerically equal to the dihedral angle AcA' = θ' θ, where angles to the
right of cB are considered positive and angles to the left are considered negative. Then

For surface elements on the portion of the displaced hemisphere to the left of the plane through , this ratio will be
less than unity corresponding to reduced flux per unit solid angle in directions with x-components opposite to the stream
velocity vector. For θ' = 0, corresponding to a surface element of the displaced hemisphere intersecting the line , the
flux is unchanged because ab = AB. The probability that a molecule has the velocity u now depends on the direction of
flight, θ', so that Eq. (4.65) must be replaced by

in which β = θ + π/2 is the angle between v and u, and

Only those molecules having a backward component of thermal velocity greater than the downward stream velocity will
be able to travel back through the nozzle clearance area. For a cylindrical nozzle the vapor velocity at the center of the
nozzle exit will be near the acoustic velocity, which for mercury is only 0.8 times the mean thermal velocity, so that a
considerable fraction of the vapor molecules will be scattered back through the clearance area. For these molecules, θ' is
negative. To

< previous page page_211 next page >


< previous page page_212 next page >
Page 212

determine this fraction it is necessary to find the value of the integral

where θ' has been eliminated using Eq. (4.72).

For vapor molecules with positive θ' the lower limit of integration is replaced by

because then molecules with thermal velocities, u, smaller than the stream velocity, v, contribute to the flow, depending
on θ'.

Molthan evaluates the integral equations for Z(θ) by a series expansion of the integrand in different quadrants and by
use of the Error Function when appropriate. He then uses the relation

to convert the distribution Z(θ) to a function of θ', which can be denoted D(θ'), and which Molthan presents in tabular
form. For a stream velocity v = vs, the velocity of sound in the vapor, his calculated values are given in Table 4.2.

Neglecting friction at the nozzle wall, Molthan obtains the vapor jet pattern shown in Fig. 4.16.

The outer curved dashed line is orthogonal to all streamlines. Along this dashed line the distance between two
streamlines is proportional to the mean free path of gas through the vapor. In the region a where vapor travels backward
from the edge of the nozzle exit, Molthan calculates that the mean free path is 4.742/0.063 = 75 times greater than in the
region b in the center of the vapor jet. For mercury vapor at 133 Pa and 400 K the mean free path of nitrogen through
the vapor is about 7 × 103 cm, and if this vapor pressure and temperature occurs in the region b, then the mean free path

Table 4.2. Molthan Angular Distribution Function


θ' D(θ') θ' D(θ')
90° 0.063 0.422
0°10'
73°20' 0.070 0.757
15°50'
60°10' 0.081 1.309
30°40'
44°40' 0.111 2.139
45°40'
28°40' 0.169 3.177
60°30'
16°20' 0.247 4.742
90°

< previous page page_212 next page >


< previous page page_213 next page >
Page 213

Fig. 4.16
Scattering of mercury-vapor stream at
the exit of a cylindrical nozzle [60].

for nitrogen through the vapor in the region a will be about 0.53 cm, which is comparable to the slit width in Langmuir's
cylindrical nozzle pump.

Molthan did his work under Gaede's direction, and one purpose of his mathematical analysis was to confirm Gaede's
hypothesis that the speed of all diffusion pumps, including Langmuir's pump, will become small when the mean free
path of gas through the vapor at the pumping annulus (nozzle clearance area) becomes smaller than the "slit
width" (distance from nozzle exit rim to pump wall). Alexander [62] applied an analysis similar to Molthan's
(apparently without having read Molthan's papers) and came to the conclusion that Gaede's diffusion principle cannot
explain the action of vapor-stream pumps. By using a mercury vapor pump with an expanding nozzle, he showed that
the vapor stream is mainly directed downward away from the nozzle clearance area and very little vapor is scattered
backward. Using a long vertical tube to probe the vapor jet, Alexander obtained the isobars of gas pressure shown in
Fig. 4.13. However, Jaeckel [69] considering the same problem disagrees with Alexander's conclusion. Also, Avery and
Witty [70] showed that Alexander's criticism of Gaedes' theory was based on a misconception of Gaede's mean-free-
path rule.

Molthan obtained a corrected form of Gaede's α factor by dividing the original hemispherical surface centered on the
hole c into zones of area
< previous page page_213 next page >
< previous page page_214 next page >
Page 214

However, one can also use differentials,

Then, instead of Gaede's Eq. (4.38), one now has for the vapor concentration at dσ,

where D(θ') is Molthan's vapor dispersion factor. Molthan omits the factor cos ϕ for simplicity since it is difficult to
integrate this factor over a zone σ (or dσ). Gaede's α formula is then modified to

where K is of the order of unity and depends on the method of correcting for the variation over the angle ϕ with the
normal to the plane of the hole. Since

the general effect of K is to introduce a factor of 1/2. Molthan's failure to account for the angle ϕ leads him to the
incorrect result α = 2 when λ0→∞ and hence a maximum flow of 734 cm3 · s1 through a hole whose area is 0.0314
cm2.

Oyama [71] criticized Molthan's assumption of spherical symmetry for the velocity directions of the thermal motion on
the grounds that the gas is expanding after leaving the nozzle and not in thermal equilibrium and in fact may incur some
condensation into liquid droplets. He also questioned Molthan's assumption that the cosine law did not apply to the
vapor crossing the hole, although Molthan may have been justified in this assumption for the case that the mean free
path is less than d/10. However, the cosine factor should be included for the gas molecules crossing the hole since their
mean free path is greater than d, but Molthan omitted this factor. Oyama considered the problem in two dimensions and
applied the PrandtlMeyer theory of the supersonic flow around the edge of a flat wall where the streamlines curve
outward until the vapor is expanded to the point at which the mean free path was 1 mm, after which he considered the
vapor to continue in the same direction in a straight line without further change in velocity. By integrating over all
possible values of the thermal velocity u he calculates the angular distribution of the molecular flux from a
nonexpanding (two-dimensional) nozzle as a function of the angle θv between the molecular velocity w and the stream
velocity v for mercury vapor with k = 5/3 and for a vapor with k = 20/19 = 1.053 for three different boiler pressures.

The PrandtlMeyer formula [68] for the bending of the streamlines of a vapor expanding adiabatically with specific heat
ratio k around the edge of a horizontal flat plate in two dimensions into a vacuum is

< previous page page_214 next page >


< previous page page_215 next page >
Page 215

where r is the radial distance from the edge as a function of ϕ, the angle of turning measured from the starting ray
through the edge to the initial point on the streamline at which the velocity component along the ray direction is zero so
that the ray makes the angle 90° with the stream velocity. For a hypothetical nonexpanding nozzle with a rectangular
cross section of width d and length so great that the flow can be considered in two dimensions only, the vapor stream
velocity at the exit plane will equal the acoustic velocity and be directed parallel to the nozzle axis so that the radial
component toward the nozzle edge is zero and no lateral expansion occurs until the vapor has crossed this exit plane.
Then for streamlines starting near one edge of this nozzle the angle ϕ equals 0 for the ray lying in the exit plane and r0
is the vertical distance to the streamline as it crosses the exit. For mercury vapor, k = 5/3 and the expression in the
brackets becomes cos(ϕ/2); thus the maximum angle of turning is 180°, at which r = ∞. For organic vapors, such as
Octoil, k = 1.02, and theoretically for hydrodynamic flow with adiabatic expansion the maximum angle of turning

would be so that the vapor streamline closest to the wall of the flat nozzle could curl right around the edge
and flow backward on the underside. However, when the vapor has expanded to the point that the mean free path is no
longer small compared to the width (of order d) of the vapor jet, the path of the molecules is no longer influenced by a
gradient in the vapor pressure but is determined by the local stream velocity and the random thermal motion at the local
temperature as calculated by Molthan. A plot of Eq. (4.81) for the case k = 1.02 is shown in Fig. 4.17.

Fig. 4.17
Two-dimensional streamline flow pattern around edge of
flat plate into vacuum for organic pump fluid vapor.

< previous page page_215 next page >


< previous page page_216 next page >
Page 216

Fig. 4.18
Expansion ratio for k = 1.02 as a function of the angle
of turning. Nozzle exit plane horizontal.

The static vapor pressure decreases along a streamline as the vapor expands, so that the pressure P at (r, ϕ) is a certain
fraction of the initial vapor pressure, P0, before any expansion occurs as given by the PrandtlMeyer formula [68]:

where k is the specific heat ratio and P0 can be considered to be the boiler pressure. A plot of this equation for the case
k = 1.02 is given in Fig. 4.18, where P0 is the boiler pressure and Pe = 0.602 P0 is the static vapor pressure at the exit
(represented by the horizontal line), where ϕ = 0, of the above hypothetical flat rectangular nozzle or of an equivalent
nonexpanding annular nozzle formed by concentric cylinders of very large diameter. At ϕ = 180° the ratio is 3.6 × 103.

Table 4.3 lists numerical values for the PrandtlMeyer ratios r/r0 and P/P0 for Octoil (k = 1.02) and mercury (k = 5/3).

By an application of L'Hôpital's rule for the evaluation of indeterminate forms it can be shown that for organic pump
fluids with k → 1, Eq. (4.81) reduces to the very simple form

which is equivalent to the isothermal solution of the problem of steady flow around a corner as given in Stodola and
Loewenstein [68, p. 983]. Similarly, as k → 1, Eq. (4.82) reduces to
< previous page page_216 next page >
< previous page page_217 next page >
Page 217

Table 4.3. PrandtlMeyer Ratios


ϕ k = 1.02 k = 1.67 ϕ k = 1.02 k = 1.67
r/r0 P/P0 r/r0 P/P0 r/r0 P/P0 r/r0 P/P0
1.0039 0.602 1.0049 0.487 0.170
0 90 3.4507 4.000 0.086
1.0153 0.600 1.0154 0.485 0.126
5 100 4.6232 5.8582 0.053
1.0349 0.593 1.0350 0.478 0.091
10 110 6.3857 9.2421 0.030
1.0629 0.581 1.0632 0.467 0.063
15 120 9.1158 16.000 0.015
1.1000 0.567 1.1007 0.451 0.0345
20 135 16.477 46.642 0.004
1.1471 0.523 1.1488 0.410 0.0175
30 150 32.031 88.731 0.0006
1.2762 0.469 1.2824 0.357 3.6 × 103
40 180 151.15 ∞ 0
1.4641 0.408 1.4821 0.298 1.5 × 103
50 195 367.3
1.7322 0.344 1.7778 0.237 5.4 × 104
60 210 965
2.1128 0.281 2.2212 0.180 5.9 × 105
70 240 8450
2.6589 0.222 2.9036 0.129 102800 4.7 × 106
80 270

and using Eq. (4.62) this becomes

where Pe is the static vapor pressure at the exit of a short cylindrical nozzle of large diameter or of a concentric cylinder nozzle of large mean
radius so that the flow pattern is approximately two-dimensional. It is evident that the static pressure is assumed to be constant along any ray
with fixed angle ϕ. Combining Eqs. (4.83) and (4.85) gives

thus for organic vapors issuing from the type of nozzle specified above, the static vapor pressure in the jet is inversely proportional to the length
of the ray from the edge of the nozzle to the intersection with the streamline starting at r0. For concentric cylinder nozzles with finite mean
radius, one might expect the vapor pressure to decrease as some higher power of 1/r.

Using Eqs. (4.86) and (4.83) the mean free path of a vapor molecule through the vapor at (r, ϕ) will be given by

where T is the local vapor temperature, σv is the mean molecular diameter (in cm), and the pressure is expressed in Pa. As an example,
consider Octoil with σv = 1.2 × 107 cm at a local temperature of 160°C (433 K) and boiler pressure P0 = 67 Pa, for which λv = 2.33 × 103 exp
(ϕ2/2). If we assume that the boundary between hydrodynamic theory (continuum) flow and kinetic theory scattering occurs when the mean
free path is of the order of 0.1 cm, then the boundary angle will be

< previous page page_217 next page >


< previous page page_218 next page >
Page 218

ϕ = 2.742 rad = 157.1°, which from Fig. 4.17 occurs in front of the nozzle so that kinetic theory must be used for the
vapor molecules scattered back through the nozzle clearance area.

As the vapor expands the flow velocity, v(r, ϕ), becomes greater than the acoustic velocity, vs, and the Mach number is
given by v(r, ϕ)/vs. Stodola defines a Mach angle, ψ, by

where vr is the component of the stream velocity v(r, ϕ) directed downward along the ray from the turning edge. Along
the starting ray we have ϕ = 0, ψ = 0, and vr = 0, so that no lateral expansion has begun and the ray is perpendicular to
the streamline.

For Octoil with k = 1.02 this becomes

or for k → 1 by L'Hôpital's rule we have tan ψ = ϕ.

The acute angle between a ray from the nozzle rim and the tangent line to a streamline is the complement of Mach's
angle, or ψc = (π/2) ψ, as defined by Stodola, and this angle is constant for all streamlines cutting a given ray as shown
in Fig. 4.19, which is adapted from Fig. 954 in Stodola and Loewenstein.
Fig. 4.19
Prandtl's two-dimensional flow around a corner into gas at pressure p2 [68].

< previous page page_218 next page >


< previous page page_219 next page >
Page 219

Nöller [72] has modified Oyama's approach by calculating the angular distribution function from the state of mercury
vapor at the discontinuity surface between the PrandtlMeyer expansion and molecular flow and by assuming that an
observer moving with the preferred velocity will observe a normal Maxwellian velocity distribution at that point. At the
discontinuity surface the mean free path is not kept constant but rather the Knudsen number λ/x is kept constant, x being
the distance from the edge of the nozzle. He applies this approach to a nozzle for which the mercury vapor velocity at
the exit is 2.2 times the acoustic velocity; thus from Eq. (4.88) the angle ϕ in the Prandtl diagram increases from 0 at the
starting ray inside the nozzle to ϕ = 95.45° at the ray corresponding to Mach 2.2, as compared to the angle ϕ = 53.13°
for the Mach 1.0 case above. For an exit velocity of Mach 2.2 the Mach angle at the exit is ψ = 65.56° as compared to ψ
= 45° for the Mach 1.0 case. The starting ray for the Mach 2.2 case is thus rotated counterclockwise by about 21° with
respect to the starting ray for the Mach 1.0 case. This rotates the ray pattern for a nozzle with the acoustic velocity at the
exist counterclockwise by about 21°, so that the streamlines are less divergent.

In the above cases the normal to the exit plane of the nozzle was parallel to the axis of symmetry of the nozzle.
However, many diffusion pump designs involve nozzles for which this normal makes an angle with the axis of
symmetry for which we use the term beveled exit. Nozzles of this type are used in steam turbines, and Zerkowitz [73]
has analyzed the deflection of the jet pattern away from the nozzle axis. Applying his equations for the case of oil vapor
with specific heat ratio k = 1.02, we [67] obtain for a nonexpanding nozzle with exit plane making the angle ϕ = 65°
with a plane orthogonal to the axis of symmetry a jet deflection of the middle streamline by 36° as shown in Fig. 4.20.

Fig. 4.20
Oil-vapor jet deflection
from nonexpanding vertical
nozzle with beveled exit.

< previous page page_219 next page >


< previous page page_220 next page >
Page 220

In moving from the plane CD orthogonal to the nozzle axis at the outer rim to the exit plane CE the static vapor pressure
decreases from P1 to P2 and the vapor velocity increases. The ratio P2/P1 will be a function of the angle ϕ, and
Zerkowitz assumes that this ratio will be given by the PrandtlMeyer equation [Eq. (4.82)]. The velocity vector for the
vapor crossing the middle of the beveled exit CE can be shown to have the magnitude

where vs is the acoustic velocity at the throat and makes the angle with the nozzle axis as given by

For a bevel angle ϕ = 65° this gives v2 = 1.6vs and But there is a further deflection after leaving the exit
plane CE as shown by Fig. 7 in the Zerkowitz article. As the vapor crosses the exit plane CE it is free to expand on both
sides so that the pressure beyond the exit, P3 is lower than P2 and there is a driving pressure, P2 P3, acting in a
direction perpendicular to CE. The derivation of the equation [67] for this second deflection will not be given here, but
the result for oil vapor with k → 1 is the additional angle ε as given by

When the throughput is small so that the inlet pressure and forepressure are low, one can assume P3 small compared to
P2 so that ε = 16° and the total jet deflection along the middle streamline is as shown in Fig. 4.20.

Iliasova et al. [74] computed the vapor flow from the top-stage nozzle of a diffusion pump by a marching procedure in
the frames of parabolized NavierStokes equations. These authors also computed the gas flow through the pump inlet
and into the vapor stream by the Monte Carlo method and computed the net pumping speed to be expected for a given
inlet chamber design and vapor stagnation density. Sadykov and Figurov [75] investigated the vapor flow inside heated
nozzles experimentally with the help of Pitot tubes and in the free jet beyond the nozzle exit by pressure and

< previous page page_220 next page >


< previous page page_221 next page >
Page 221

temperature measuring instruments. They also calculated theoretically the vapor flow allowing for a boundary layer
using the equations of aerodynamics and rarefied gas dynamics. The mixing of the pumped gas with the vapor jet in the
backstreaming region and the forward pumping region was measured with pressure probes. Backstreaming rate from the
inlet stage was measured by weighing. Heating the nozzle to 25 K higher than the vapor conduit decreased the
backstreaming by 20%.

Attempts to explain the operation of diffusion pumps by kinetic theory considerations of the collisions of gas molecules
and vapor molecules were made by Riddiford and Coe [76], by Reichelt [77], by Florescu [63], and by Tóth [78].

4.4.4
Ultimate Pressure

Many authors have discussed the ultimate pressure obtainable with diffusion pumps which depends mainly on four
factors: (1) the back-diffusion [61] of gas from the fore-vacuum through the vapor jets to the high-vacuum side; (2) the
vapor pressure of the pump fluid [7, 79, 80] condensed on the walls of the inlet chamber or on baffles over the inlet; (3)
the evolution from the oil in the boiler of dissolved gas and low-molecular-weight (cracking) fragments [33, 34] from
decomposition of the pump oil; and (4) the outgassing [81] of the vacuum system beyond the pump. In addition, back
migration of oil vapor from a mechanical forepump may contaminate the system when the diffusion pump is not
operating and no valves are closed. This source of contamination can be reduced by the use of a purge gas during
shutdown [45].

Back-Diffusion of Gases from the Fore-Vacuum. Several authors have derived a first approximation to a more explicit
theoretical value of the back-diffusion coefficient β than that given by Gaede's simple Eq. (4.34). Matricon [82]
assumed that the vapor jet from a cylindrical nozzle in a single-stage mercury vapor pump could be simulated by
straight lines of flow originating at a point inside the nozzle exit at a distance from the exit plane of about one-third of
the exit diameter. However, to derive a formula for the β factor, as a first approximation, he assumed that in the region
of counterdiffusion of vapor and gas from the fore-vacuum the pressure gradient from the axis to the wall at any
distance x along the axis from the nozzle exit was negligible and that no condensation of mercury vapor occurred over
the distance x = 0 to x = L, where L is the useful length of the vapor jet when the forepressure is so low that the vapor
stream is not obstructed by the gas in the fore-vacuum. Then he obtains

where N is the concentration and v is the stream velocity of the vapor molecules independent of x, while D0 is the value
(in cm1 · s1) of the gas-vapor diffusion coefficient at unit vapor concentration. Using the Meyer formula for the
diffusion coefficient and the StefanMaxwell formula for the mean free path through a binary mixture, Matricon obtains

< previous page page_221 next page >


< previous page page_222 next page >
Page 222

Table 4.4. Calculated Diffusion Coefficients at 101308 Pa and 160°C


Units σ σ' M (g) M' (g) D0 D760
(108 cm) (108 cm) (1018 cm1·s1) (cm2/s)
Airmercury
3.11 5.67 29 200.6 2.88 0.17
Airmercury
3.72 4.50 29 200.6 3.30 0.19
AirOctoil
3.11 12 29 390 0.99 0.058
Hydrogenmercury
2.38 5.67 2 200.6 13.9 0.82
Hydrogenmercury
2.74 4.50 2 200.6 17.2 1.01
HydrogenOctoil
2.38 12 2 390 4.35 0.257
Heliummercury
2.2 3.5 4 200.6 19.6 1.16
HeliumOctoil
2.2 12 4 390 3.2 0.19
DeuteriumOctoil
2.38 12 4 390 3.1 0.18

D760 = D0/N760 = 5.9 × 1020D0 at 160°C

where σa = (σ + σ')/2 is the mutual collision diameter for the gas and mercury molecules, M is the molecular weight of the gas, M' is the
molecular weight of the vapor, and T is the absolute temperature of the gas within the vapor stream. The diffusion coefficient D760 (in cm2 · s1)
at 101,308 Pa (760 Torr) as used in Eq. (4.34) is obtained from

where N760 is the concentration of vapor molecules at 101,308 Pa and the specified temperature.

Values of the diffusion coefficient D760 for various gas-vapor combinations T = 160°C and 101,308 Pa as calculated by Dayton [61] from
Matricon's formula are listed in Table 4.4 together with the values of D0. The mean diameter σ' = 5.67 × 108 cm for mercury, σ = 3.11 × 108
cm for air, and σ = 2.38 × 108 cm for hydrogen are the values used by Matricon. Texts on kinetic theory give σ' = 4.50 × 108 cm for mercury
(150°C), σ = 3.72 × 108 cm for air, and σ = 2.74 × 108 cm for hydrogen. The value σ' = 12 × 108 cm for Octoil was estimated by Jacobs and
Kapff [83].

Gaede [2] measured an average value D760 = 0.18 cm2 · s1 for air through mercury vapor and an average value D760 = 0.66 cm2 · s1 for
hydrogen through mercury vapor at 140°C. It may be noted that D0 is proportional to T½ while D760 is proportional to T3/2.

Wertenstein [84] and Jaeckel [69] also arrive at Eq. (4.96) but use Enskog's [85] formula for D0,

This equation gives values of D0 about 10% lower than those from Eq. (4.97) using the same constants. Jaeckel cites σ = 2.7 × 108 cm for
hydrogen and 3.68 × 108 cm

< previous page page_222 next page >


< previous page page_223 next page >
Page 223

for air. The ratio of the diffusion coefficient for hydrogen through mercury to the coefficient for air through mercury is
about 4.8, while the ratio for hydrogen to air for diffusion through an organic vapor such as Octoil is about 4.4.

For a cylindrical nozzle with no loss from throat to exit the stream velocity, ve, of Octoil at the exit is the acoustic
velocity as given by Eq. (4.64). The molecular concentration of vapor at the exit will be the same as at the throat:

where Na is Avogadro's number and P0 is the vapor pressure in dyne · cm2. When P0 is expressed in Pa, one must use

If we now assume that the nozzle clearance is small compared to the diameter of the nozzle exit, then, as the vapor
flows from the cylindrical nozzle exit into a condenser with a cross section only slightly larger than the nozzle exit, the
vapor concentration will be only slightly decreased and the stream velocity only slightly increased, so that the product
Nv will be approximately equal to Neve = Ntvt. Then for an organic vapor, such as Octoil, Eq. (4.96) with P0 in Pa
becomes

with D0 in cgs units given by either Eq. (4.97) or Eq. (4.99) and assuming that the temperature T of the gas in these
equations equals the vapor temperature T0 and that very little condensation of the vapor jet occurs within the distance L
from the nozzle exit. As an example, at the relatively low temperature of 160°C or T0 = 433 K the vapor pressure, P0,
of Octoil is about 13.33 Pa; and using M' = 390, R0 = 8.31 × 107, Na = 6.02 × 1023, D0 = 1018 cm1 · s1, and L = 5 cm,
Eq. (4.102) gives β = 3.35 × 1029 for air through Octoil. Using D0 = 4.35 × 1018 cm1 · s1 for hydrogen through Octoil
gives β = 2.84 × 107, so that a forepressure of 1 Pa of hydrogen would limit the ultimate pressure to pu = 2.84 × 107 Pa.
Actually, there is considerable condensation of vapor over the distance L which is determined by the position of the
shock boundary created by the forepressure; and the ratio of nozzle throat area to body clearance area is about 1/2 to 1/3
for the last stage (nearest the fore-vacuum), so that the static vapor pressure at the body clearance area is about 1/2 to
1/3 of the pressure at the throat, which in turn is about 0.6 times the boiler pressure P0. Thus, even for the more
common temperature 180°C and boiler pressure P0 = 67 Pa, the β coefficient for hydrogen through Octoil can be of the
order of 107 or greater.

Theory of Pump Performance in the Forepressure Breakdown Region. A shock boundary is formed between the vapor
jet and the gas in the fore-vacuum when the forepressure rises above 1 Pa. The vapor expanding from the nozzle exit
travels with increasing supersonic velocity, and the static vapor pressure decreases as the jet expands and diverges
toward the pump walls. Any obstruction, such as a Pitot tube which has a small opening at the tip of a tapered nose,
placed in the vapor stream

< previous page page_223 next page >


< previous page page_224 next page >
Page 224

which permits the vapor to escape laterally will result in a compression shock in front of the obstruction as the flow
velocity perpendicular to the obstruction is reduced to zero. The law of conservation of momentum requires that a
pressure is exerted on the obstruction when the latter is fixed in position. According to the definition in Stodola and
Loewenstein, [68, p. 85], this pressure, known as the ''impact pressure" or "dynamic pressure," is the reading of a
manometer attached to the open channel in the Pitot tube. The magnitude of this pressure depends on the residual
momentum in the vapor stream after lateral dispersion and on the heat transferred to the Pitot tube. It has been found by
Löliger (and independently by Prandtl) that the impact pressure can be calculated by assuming that a compression shock
followed by an adiabatic compression takes place in front of the pointed end of the Pitot tube until the kinetic energy of
the flow is consumed. The shock arises from the sudden conversion of stream momentum into static pressure as the
vapor at supersonic velocity impinges on vapor moving with subsonic velocity.

The method of computing the impact pressure given in Stodola and Loewenstein involves the tracing of the "condition
loci" on the entropy diagram [68, Fig. 51]. The condition change leads from the initial condition A1 along the Fanno
and Rayleigh curves to the shock point A2, and from there adiabatically to the impact pressure Pi. The Fanno line is
given by combining the total energy per unit mass equation

where P1 is the initial static pressure and P2 is the final static pressure while E1 and E2 are the initial and final internal
energies per unit mass due to molecular rotation and vibration, with the equation of continuity

where a1 and a2 are cross sections in a tube of flow. For adiabatic expansion [86] we obtain

The Rayleigh line is obtained by combining the equation of continuity with the impulse-momentum equation

Since compression shock occurs in a very thin zone over the shock boundary, a1 and a2 will be approximately equal.
For perfect gases the algebraic solution for the intersection of the Fanno and Rayleigh lines is easily obtained, giving as
the static pressure after the compression shock the following equation:

< previous page page_224 next page >


< previous page page_225 next page >
Page 225

or for organic vapors with k → 1 we obtain

which is the total flux of stream momentum per unit area before the shock. Prandtl's solution for the stream velocity
after the shock is

For organic vapors with k → 1 we obtain

where vs1 is the value of the acoustic velocity under the initial conditions. There can be no shock unless the initial
stream velocity v1 exceeds the acoustic velocity vs1 for the initial state. From Eq. (4.110) it is obvious that if the initial
stream velocity is much larger than the acoustic velocity, the final velocity v2 after the shock will be a small fraction of
the acoustic velocity. After the shock the vapor is compressed adiabatically and the stream velocity falls from v2 to zero
while the static pressure rises to the final maximum value which represents the impact pressure, Pi.

When measuring the impact pressure at the throat of a nozzle, there is no compression shock, but the process of
adiabatic expansion is suddenly reversed to one of adiabatic compression. Then in Eq. (4.110) both v2 and v1 equal the
acoustic velocity, vs, at the throat and P2 = P1 = Pt. The measured impact pressure at the throat then equals the boiler
pressure, P0, for which the vapor stream velocity is zero. All streamlines are parallel at the throat and remain parallel to
the nozzle axis when the nozzle is cylindrical and short so that no energy is lost at the walls, but in expanding conical
nozzles of the DeLaval type the streamlines diverge and the impact pressure at the exit is less than the boiler pressure.
However, in an expanding nozzle with walls curved in a certain way, the Prandtl nozzle, all the streamlines at the exit
can be made parallel to the nozzle axis and to each other. However, for most diffusion pumps the expanding nozzles
have straight walls and the vapor acquires some momentum in directions perpendicular to the axis in passing from the
throat to the exit. In expanding from the acoustic velocity at the throat to supersonic velocity at the exit of a DeLaval
nozzle, some of the energy of random motion is converted into a flux of forward momentum parallel to the nozzle axis,
and the impact pressure at the exit is difficult to calculate. As shown by Fig. 49 in Stodola and Loewenstein [68] for the
expansion of steam through a DeLaval nozzle, the measured impact pressure at the exit is given approximately by

where At/Ae is the ratio of throat area to exit area and C is somewhat greater than 1.0.

By application of the definitions [87] of pressure in terms of momentum flux with respect to a specified surface, the
impact pressure can be related to the dynamic pressure as given by

< previous page page_225 next page >


< previous page page_226 next page >
Page 226

where n is the number of molecules in unit volume, m is the mass of each molecule,

is the average molecular velocity, and v is the stream velocity perpendicular to the surface. The impact pressure is then
obtained by omitting the last term on the right since molecules moving with velocity ua v in the negative direction with
respect to the surface normal will never strike the surface of the shock boundary. Since the vapor velocity at the throat is

one can substitute

Proceeding in this manner the author [67] obtained the following as an approximate formula for the coefficient C:

For example, for steam through the nozzle studied by Löliger [68, p. 85, Fig. 49] we have At/Ae = 0.238, k = 1.3, Pe/P0
= 0.05, Pt = 0.546P0, and C = 1.402, giving a computed impact pressure Pie = 0.33P0 close to the experimental value
0.34P0 obtained by Löliger. For oil vapor with k = 1.02, and assuming Te = Tt, the same nozzle gives ve/vt = 2.3183, Pe/
P0 = 0.0615, Pt/P0 = 0.602, C = 1.502, and Pie = 0.36P0. When Ae = At and k = 1.02, Eqs. (4.117) and (4.119) give ve/
vt = 1.1 and Eq. (4.116) gives C = 0.96 corresponding to Pie = 0.96P0.
< previous page page_226 next page >
< previous page page_227 next page >
Page 227

Fig. 4.21
Shock boundary near forepressure breakdown [15].

After the vapor leaves the nozzle the vapor diverges rapidly and the impact pressure for direct compression shock will
depend on a narrow bundle of streamlines intercepted at the shock boundary and the distance traveled from the nozzle
exit. For cylindrical or expanding conical nozzles it has been found experimentally that as a good approximation it can
be assumed that the vapor diverges in straight lines from a point source located inside the nozzle on the axis at a
distance depending on the vapor velocity at the exit. For oil vapor through cylindrical nozzles with velocity about equal
to the acoustic velocity at the exit, this distance appears to be about one-half the diameter of the nozzle exit, so that a ray
from the source point passing through the rim of the nozzle exit makes an angle of 45° with the nozzle axis and with the
wall of the pump casing. From Table 4.3 it can be seen that the static vapor pressure along this ray (ϕ = 135°) is about
0.035 times the boiler pressure, or 2.3 Pa when P0 = 67 Pa. For vapor diverging from a virtual point source the (direct)
shock boundary can be expected to form on a spherical surface perpendicular to each ray from the source. That this
actually occurs in pumps with cylindrical or conical nozzles is shown by glow discharge experiments reported by
Dayton [61] and by Kutscher [15]. The curved shock boundary shape is particularly noticeable just at the point of
forepressure breakdown as shown by Fig. 10 in Dayton [61] and by Fig. 4.21 adapted from Kutscher's article (which
includes some regions of oblique shock).

Since Eq. (4.111) indicates that the impact pressure is roughly inversely proportional to the ratio Ae/At of exit area to
throat area, as an approximation it can be assumed that the impact pressure at the wall of a pump with cylindrical nozzle
is inversely proportional to the square of the radius r from the hypothetical point source, where r is greater than the
radius to the rim of the nozzle exit. Then from Fig. 4.22, r2 = R2 + (x + xs)2, where R is the radius of the pump casing
and xs is the distance between the point source and the exit plane. The impact pressure at x will then be given by

< previous page page_227 next page >


< previous page page_228 next page >
Page 228

Fig. 4.22
Distance x of shock boundary at wall from plane of nozzle
exit with hypothetical point source of vapor flow.

where rn is the radius of the cylindrical nozzle and Pie is the impact pressure at the rim of the nozzle exit.

Assuming that at every point on the shock boundary before jet breakdown the impact pressure Pix equals the
forepressure F and that the impact pressure at the nozzle exit, Pie, is given by Eq. (4.111), then using xs = rn we obtain

where rt is the radius of the nozzle throat. From Fig. 4.22 the forepressure breakdown, F = Fb, occurs when the shock
boundary has been pushed back to x = x0 = R rn. Then for cylindrical nozzles we obtain

where B = πR2 is the body-clearance area and At is the nozzle throat area.

For most metal diffusion pumps the nozzle is constructed of concentric cylinders or cones so that the throat and exit
cross sections are annular with a width small compared to the mean radius. Then, as an approximation we can assume
that the vapor disperses in straight lines from a line source (circle of large radius) orthogonal to the plane through the
pump axis and located within the nozzle at a short distance from the exit of the order of one-half the width of the exit as
shown in Fig. 4.23.

The dashed line ti = R rB is the width of the body clearance area, B, and the angle φ gives the deflection of the
centerline of the jet caused by the presence of the streamliner (or apron) hj. For each stage, the breakdown forepressure,
Fb, at which the outer edge of the shock boundary is close to the point t, is determined by the ratio of the nozzle throat
area, Ae, to the body clearance area and the amount of the jet deflection φ. Then, as an approximation, the impact
pressure at a point on the pump casing wall located at the distance x from the plane of the nozzle exit will be given by
< previous page page_228 next page >
< previous page page_229 next page >
Page 229

Fig. 4.23
Dispersion characteristics of inverted annular nozzle.

where Pie is the impact pressure at the nozzle exit, xs is the distance of the source from the plane of the nozzle exit, rs is
the distance of the source from the pump axis, R is the radius of the pump casing, θ + φ is the angle between the vertical
line through the source and the ray from the source which passes through the rim of the nozzle exit, and n is an index
which equals 1 for annular nozzles in which the throat width w and the body clearance ti are small compared to the
casing radius R, but which approaches the value 2 as rs→ 0 corresponding to the previous case of a cylindrical nozzle
from which the vapor is assumed to issue from a point source. In general, the impact pressure at the casing wall will be
a maximum and forepressure breakdown will occur when the shock boundary reaches the point t where x = x0. At this
point, F = Fb = P1 and

so that the breakdown forepressure is


< previous page page_229 next page >
< previous page page_230 next page >
Page 230

It can be seen from the similar triangles in Fig. 4.23 that

so that, using Eq. (4.111) with C replaced by the empirical constant Kn, we obtain

which is equivalent to Eq. (4.122) when n = 2 and rs = 0. When n = 1 and R rn is small compared to rn, the ratio (rn rm)/
(R rm) will be approximately equal to the ratio of nozzle exit area, Ae, to the body clearance area, B. Then from Eqs.
(4.126) and (4.127) when n = 1 the general formula for the limiting forepressure is

Experimental data on various pumping stages has shown that the limiting forepressure is given by Eq. (4.128) with Kn
values which in general range from 0.5 to 1.1.

To calculate the β coefficient it is necessary to compute the rate of diffusion of gas from the fore-vacuum through the
vapor jet from the boundary uu' at the forepressure F to the boundary te with the high vacuum pressure p. The length of
the vertical diffusion path, Lj, between these boundaries can be calculated from the given geometry, and integration can
be performed over the upper boundary. However, it is obvious that most of the back-diffusion will occur where Lj is
smallest, or close to the casing wall through a barrier of thickness ut = x x0, where from the above equations we obtain

From Fig. 4.23 we have

Then the thickness of the diffusion barrier at the casing wall is

Substituting this for L in Eq. (4.34) together with mean values for the static vapor pressure P and the vertical component
of the streamline velocity, v, at the casing wall obtained by extrapolating Eqs. (4.117) and (4.119) in terms of the
expansion ratio Ae/B to the region x x0 would then give an approximate formula for β as a function of the forepressure
F. A more exact relationship could be obtained by extrapolating these equations from the nozzle exit to the cross section
at x to obtain the vertical component of stream velocity near the wall, vx(x), as a function of the distance x and the static
vapor pressure near the wall, P(x), as a function of x and then substituting

< previous page page_230 next page >


< previous page page_231 next page >
Page 231

these functions in place of v and P in Eq. (4.33), integrating from x x0 to x, and then using Eqs. (4.129) and (4.130).

Vapor Pressure of Fluid Condensed in the Inlet Chamber or on Baffles. The vapor pressure of the fluid condensed on
walls near the inlet of the pump (and baffle) can be reduced by proper design of fractionating pumps and baffles, as well
as the use of cold traps. For fractionating pumps with water-cooled baffles the ultimate pressure limit due to pump fluid
vapor will be slightly less than the vapor pressure in Pa at 25°C as given in the next-to-the-last column in Table 4.1 for
the given pump oil. Use of the more stable fluids, such as Convalex-10, Santovac-5, or DC-705 with the addition of a
cold trap cooled to below 40°C will usually allow ultimate pressures below 1 × 108 Pa when the forepressure is
sufficiently low. It has been reported [88,89] that using a system with an alumina or zeolite trap and gas purge
provisions to reduce contamination from the oil in mechanical forepumps and using a fractionating diffusion pump with
polyphenyl ether pump fluid, a cold cap, a cryotrap chilled with liquid nitrogen or simply by circulating methanol at 75°
C, and a high-vacuum gate valve with proper operating procedures can produce clean vacua as free of hydrocarbons and
CO as those produced with conventional (hydrocarbon lubricated) turbopumps.

< previous page page_231 next page >


< previous page page_233 next page >
Page 233

Part II
Molecular Drag and Turbomolecular Pumps

Jörgen Henning

4.5
Molecular Drag Pumps

A molecular drag pump is a vacuum pump in which gases are pumped by momentum transfer from a rapidly rotating
solid surface to the gas molecules. The rotor impulse is transmitted to the particles by superposition of the thermal
velocity of the colliding particles with the velocity component of the moving rotor surface (Fig. 4.24). The nondirected
motion of the particles is changed to a directed motion in the pumping process. When the mean free path of the particles
between the collisions with other particles is larger than the spacing between the rotating and the stationary surfaces
(molecular flow range, typically < 0.1 Pa), particles collide primarily with the rotor, resulting in an efficient pumping
process, and there is no interacting influence of the different gases.

In the laminar flow range (typically > 0.1 Pa) the action of the rotor surface is restricted by the frequent collisions
between particles. Therefore, a molecular drag pump is not capable of pumping gases against atmospheric pressure and
must be backed by an adequate roughing pump.

The first pump of this type was introduced in 1913 by Gaede [90], when he presented his "Molecular Drag Pump" using
a multistage cylindrical rotor design, with parallel slots around the circumference of the rotor, into which project
extensions from the outer casing.

Foundations of Vacuum Science and Technology, Edited by James M. Lafferty.


ISBN 0-471-17593-5 © 1988 John Wiley & Sons, Inc.

< previous page page_233 next page >


< previous page page_234 next page >
Page 234

Fig. 4.24
Principle of operation of
molecular drag pump.

A modified form of the molecular drag pump, designed by F. Holweck [91] in 1923, used a smooth cylindrical rotor
(drum) inside a housing with spiral channels, decreasing in depth toward the exhaust side. The intake was in the middle
of the housing and the gas was dragged to both ends of the housing, making use of twice the pumping speed.

In 1940 Siegbahn [92] described his disc-type molecular drag pump using a disc-rotor. A circular disc rotates inside a
container consisting of two side plates with spiral grooves. These grooves are deeper at the periphery (inlet) and
gradually decrease in depth toward the center (exhaust).

Up to the 1970s, only few researchers were interested in these "early" molecular drag pumps because of their relatively
low pumping speed and their questionable reliability; furthermore, in those days there was no real demand for these
pumps. In order to attain low ultimate pressures with these pumps the clearances between rotating and stationary parts
were made a few hundreths of a millimeter only. Therefore, any change in temperature or intruding solid particles could
result in a failure of the pump, caused by a ceased rotor. However, recently, the basic ideas of Holweck (drum), Gaede
(slotted rotor), and Siegbahn (disc) were picked up again successfully in the design of modern pumps (molecular drag
pumps and combined molecular drag and turbomolecular pumps), in order to attain extremely low pressures and/or to
make use of simple dry roughing pumps.

4.5.1
Theoretical Considerations and Performance Data

The principal design of the pumping stage of a molecular drag pump is shown in Fig. 4.25. It does not make any
difference if the pumping channel is located in the rotating or the stationary part of the pumping stage. Different
methods have been developed to describe the pump performance of molecular drag and turbomolecular pumps. The
continuum methods [93], and methods using statistical calculations [94] agree in the basic results concerning the pump
performances. The values depend on the transmission probabilities (P) of the particles moving in and against the
pumping direction. For the molecular flow, these probabilities depend on the ratio v/u of the rotor speed v and the mean
thermal velocity u of the gas particle, but not on the pressure.

For a pumping channel, in the range of free molecular flow, the compression and the net flow rate (pumping speed) can
be calculated from the law of conservation of molecules by assuming that the velocity distribution function of the gas on
both sides

< previous page page_234 next page >


< previous page page_235 next page >
Page 235

Fig. 4.25
Principle design of a molecular
pumping stage: l, channel
length; h, channel height; b,
channel width; s, gap between
rotor and housing; FA, entrance
area; FB, exit area.

of the channel is Maxwellian. For steady-state flow, the net flow rate W (Ho-coefficient) from side A to side B can be
derived from

where PAB and PBA are the probabilities that a molecule incidenting from side A or B will be transmitted through the
channel to side B or A, and FA and FB are the entrance area of side A and the exit area of side B, respectively. K = nB/
nA is the compression (n is the number of molecules) [95].

In order to calculate the performance of a molecular pumping stage, two extremes, the maximum compression Kmax (at
zero flow) and the maximum pumping speed Smax (for equal pressure on both sides of the pumping stage) must be
known.

From Eq. (4.132) the following relations can be derived:

For maximum compression, no gas flow (W = 0):


The important problem to solve remains the calculation of PAB and PBA, which, for example, was solved by Kruger
and Shapiro [94] using Monte Carlo techniques. As a result it was found, in agreement with calculations using different
methods [93], that Kmax of a molecular drag pump is exponentially dependent on the rotor speed v,

< previous page page_235 next page >


< previous page page_236 next page >
Page 236

a pump-specific factor g (rotor and stator geometry), and the square root of the molecular weight M of the particles
pumped:

For maximum pumping speed (K = 1):

Smax is proportional to the product of a specific pump factor G (rotor and stator geometry) and the rotor speed v; it does
not depend on the pressure and the kind of gas pumped, because the molecular arrival rate at the inlet is proportional to
the thermal velocity of the gas:

An operating pumping stage works in the region between these two extremes. For real pumping conditions (gas
throughput Q = Spinlet) and different pressures on both sides of the pumping stage (K = poutlet/ pinlet) the following
relation exists between compression K and real pumping speed S of a single pumping stage:

Therefore, the pumping speed of a molecular drag pump depends on the compression and the pumping speed Sv of the
backing pump, as given in Eq. (4.138). From there it can be derived that if Kmax for any gas is small, the pumping
speed for this gas is a function of the ratio S/Sv. For practical purposes a molecular drag pump with low Kmax for H2
needs a larger backing pump to pump H2 effectively (low S/Sv ratio).

Due to the inlet conductance of the pumping channel, there is a limitation of Smax and the actual pumping speed S of a
molecular drag pump becomes a function of the ratio v/u and therefore it will depend on the molecular weight of the gas
pumped.

More exact calculations take into account the losses through gaps between the rotating and the stationary parts.
However, the main relation, Eq. (4.138), remains the same.

As for any vacuum pump, the ultimate pressure pf, that a molecular drag pump can attain can be calculated from the
pressure pv at the outlet side of the pump by dividing it by the maximum compression Kmax:

< previous page page_236 next page >


< previous page page_237 next page >
Page 237

Fig. 4.26
Principle of multistage molecular
drag pump (Balzers Pfeiffer).

The increase of Kmax with the molecular weight M means that heavy molecules are highly compressed and have a low
backflow probability; this is the reason for the ''clean" vacuum without contamination by oil vapors and hydrocarbons.

4.5.2
Design Considerations

Stand-alone molecular drag pumps are designed on the basis of the Holweck principle [91]. These pumps in general
include two different sections: (1) an entrance section with a row of turbomolecular pump blades for maximum
conductance to ensure a high pumping speed and (2) subsequent molecular pumping stages (up to 5) having in principle
a drum design with a multiribbed structure to ensure efficiency and high compression (Fig. 4.26). These pumps are
available with mechanical rotor bearings or a combination of mechanical and magnetic bearings. The lubricant used for
the mechanical bearings is either grease or oil. More details of the design of molecular drag pumps are given in Section
4.6.2.

4.5.3
Typical Performance Data of Commercial Pumps

The working range for molecular drag pumps is, depending on the manufacturer, between 103 and 103 Pa. Due to the
high admissible foreline pressure, molecular drag pumps can use small, simple, dry, and inexpensive fore-vacuum
pumps (e.g., membrane pumps) with an ultimate pressure of <103 Pa.

< previous page page_237 next page >


< previous page page_238 next page >
Page 238

4.5.3.1
Compression

Depending on the manufacturer's design, molecular drag pumps have the following compression values: H2, 2 × 102 to
1 × 103; He, 1 × 103 to 2 × 104; N2, 1 × 107 to 1 × 109.

4.5.3.2
Pumping Speed

Molecular drag pumps are available in a pumping speed range from 7 to 300 liter · s1.

Type designations of molecular drag pumps often use the pumping speed for nitrogen. Therefore the pumping speed for
N2 is used as the 100% value. This value is compared to the pumping speed data for H2 and He. From the catalogues of
different manufacturers: H2, 40% to 56%; He, 53% to 67%.

4.5.3.3
Ultimate Pressure

Depending on how many molecular pumping stages are used in series, the attainable ultimate pressure is in the range of
105 to 103 Pa.

4.6
Turbomolecular Pumps

The turbomolecular pump invented by Becker [96] in 1957 (Fig. 4.27) became commercially available in 1958. Since
then it has become very popular in every field of high- and ultrahigh-vacuum technique, due to a clean, consistent, and
predictable vacuum created, the easy operation, and the advanced degree of operating reliability. The turbomolecular
pump is the only mechanical vacuum pump which, together with a roughing pump, can attain ultimate pressures in the
range below 108 Pa. It is a bladed molecular turbine that compresses gases by momentum transfer from the rapidly
rotating blades of the rotor wheels to the gas molecules. It is working on the same principle as the molecular drag pump.

When the mean free path of the particles is larger than the spacing between rotor and stator (molecular flow range in a
turbomolecular pump, typically <101 Pa), particles collide primarily with the rotor, resulting in an efficient pumping
process, and there is no interacting influence of the different gases. In the laminar flow range (typically in a
turbomolecular pump > 101 Pa) the action of the rotor is restricted
Fig. 4.27
Principle design of turbomolecular pump [96].

< previous page page_238 next page >


< previous page page_239 next page >
Page 239

by frequent collisions between the particles. Therefore, a turbomolecular pump is not capable of pumping gases against
atmospheric pressure and must be backed by an adequate roughing pump.

The Becker design avoided the obvious disadvantages of the "early" molecular drag pumps (relative low pumping
speed, questionable reliability): The turbomolecular pump is composed of a series of wheels with coaxial blades,
alternately fixed and moving. In each wheel the sides of the blades are inclined with respect to the axis, in one direction
for the blades of the moving wheels and in the other for the blades of the fixed wheel. The moving wheels have a high
rotational speed so that the peripheral speed of the blades (up to 500 m/s) is of the same order of magnitude as the speed
of the molecules of the pumped gas.

The distances between these wheels were in the range from several tenths to a few millimeters. The channels between
the inclined blades of the wheels act like elementary molecular drag pumps, similar to the Gaede molecular drag pump.
All channels (~2050) on one wheel are connected in parallel and together can yield a high pumping speed up to several
thousand liters per second.

4.6.1
Theoretical Considerations and Performance Data

A milestone for the understanding of turbomolecular pumps was the work published by Kruger and Shapiro in 1960
[94] on blading geometry of axial-flow molecular turbines in the molecular flow range. However, most later
calculations of the pumping speed characteristics made by several authors [93, 94, 97] included complicated statistical
mathematics or, from the user's standpoint, unknown geometrical factors with different formulas for light and heavy
gases. In 1983, simple pumping speed calculations were published for gases with molecular weights between 2 and 50
[98].

The main vacuum data of a turbomolecular pump, the compression, and the pumping speed can be calculated from the
data of single rotor and stator wheels. A rotating wheel with blades pumps molecules from one side to the other, and
some particles move opposite to the primary flow direction through the blades. The structure of most common
turbomolecular pumps is shown in Fig. 4.28. In order to calculate the performance of a single wheel, the same
considerations as with the molecular drag pump can be used, by replacing the wording "molecular pumping stage" by
"single wheel." Again two extremes, the maximum compression Kmax (at zero flow) and the maximum pumping speed
Smax (for equal pressure on both sides of the wheel), must be known. For a turbomolecular pump the transmission
probabilities in the molecular flow depend on the ratio v/u of the blade speed v and the mean thermal velocity u of the
gas particle, and additionally on the blade angle α, but not on the pressure. For Kmax of a turbomolecular pump Eq.
(4.134) applies and is exponentially dependent on the blade speed v, a specific pump factor g (rotor and stator
geometry), and the square root of the molecular weight M of the particles pumped.

In agreement with Eq. (4.136), Smax is proportional to the product of a specific pump factor G (rotor and stator
geometry) and the blade speed v, and it does not depend on the pressure and the kind of gas pumped:

An operating wheel works in the region between these two extremes. For real pumping conditions (gas throughput Q =
Spinlet) and different pressures on both sides of the wheel (K = poutlet/ pinlet) the relation given in Eq. (4.138) applies.
More exact calculation takes into account the gap between the rotating wheel's outer diameter

< previous page page_239 next page >


< previous page page_240 next page >
Page 240

Fig. 4.28
Structure of turbomolecular blades [98]: α, blade angle; d, blade thickness;
h, distance between blades.

and the inner diameter of the stator housing and the variation of the blade geometry with the radius. However, the main
Eq. (4.138) remains the same.

A turbomolecular pump is assembled from several wheels in series with different blade geometries. Each wheel can be
regarded as a separate pump. If the pumping speed Sv of the next wheel downstream is known and the throughput is
constant, K in Eq. (4.138) can be replaced by S/Sv; and beginning with the pumping speed Sv of the backing pump and
using the values Smax and Kmax for the first wheel at the forevacuum side, the real pumping speed of this wheel can be
calculated. The formula can be used as a recurrence formula to calculate step by step the pumping speed of the whole
pump in the molecular flow range.

Then the pumping speed of a turbomolecular pump depends on the compression and the pumping speed Sv of the
backing pump, as given in Eq. (4.138). From there it can be derived that if Kmax for any gas is small, the pumping
speed for this gas is a function of the ratio S/Sv. For practical purposes a turbomolecular pump with low Kmax for H2
needs a larger backing pump to pump H2 effectively (low S/Sv ratio).

Due to the inlet conductance of the blade area of the wheel, there is a limitation of Smax and the actual pumping speed S
of a turbopump becomes a function of the ratio v/u [98], and therefore it will depend on the molecular weight of the gas
pumped:

where F is the pumping area of the wheel, α is the blade angle, ds is the blade thickness, h is the blade distance, and k is
the trapping probability ≈1 [98]. The pumping speed
< previous page page_240 next page >
< previous page page_241 next page >
Page 241

and the compression of a turbomolecular pump decrease at pressures of above 101 Pa, caused by the interaction of the
particles with one another, as the mean free path is no longer larger than the blade distance and the blades of the rotor
wheels no longer are in the molecular flow range. This value corresponds to a foreline pressure of 110 Pa.

Again, as for any vacuum pump, the ultimate pressure pf a turbomolecular pump can attain can be calculated from the
pressure pv at the outlet side of the pump by dividing it by the maximum compression Kmax using Eq. (4.139). The
increase of Kmax with the molecular weight M means that heavy molecules are highly compressed and have a low
backflow probability; this is the reason for the "clean" vacuum without contamination by oil vapors and hydrocarbons.
The smaller compression for light gases is the reason that the residual gas atmosphere of a turbomolecular pump
consists mainly of H2. This, however, holds true only for a "clean" system with metallic flange seals. In the case of
viton or rubber seals the ultimate pressure and the residual gas look differently [99].

4.6.2
Design Considerations

4.6.2.1
Rotor and Stator Geometry

The pumping speed and the compression of a turbomolecular pump depend strongly on the rotor geometry and speed.
Starting from the 1958 original geometry of the rotor and stator wheels, new wheel geometries have been developed.
These geometries together with increased rotor speeds allowed much smaller and lighter rotors for the supercritical
speed range.

The rotor and stator stages nearest the high-vacuum inlet are designed to serve a purpose different from those near the
outlet. The flow through each stage is constant, or, stated another way, the product of pressure times pumping speed is a
constant. The blades nearest to the inlet of a turbomolecular pump are designed to have as high a pumping speed as
possible, whereas the blades nearest to the foreline port are designed for high compression. The opening angles of the
blades are decreased from the high-vacuum side of the fore-vacuum side with the aim of optimizing the compression
and the pumping speed. For economic reasons it would be impractical to make each stage different from its neighbor. A
compromise results in groups of two to four types of blades, in which each is designed for a particular speed and
compression ratio.

The methods of manufacturing the rotors and stators have an influence on the pumping speed and compression. Rotors
can be made of individually machined wheels which are heat shrunk to the rotor shaft, by machining complete groups of
wheels from a single block of material or by manufacturing the rotors using spark erosion. The individually machined
wheels offer the advantage of making them "optically opaque", maximizing them for the compression. The other
methods of rotor production yield wheels with less opaqueness and lower compression, maximizing them for pumping
speed.

The stators are either manufactured from individually machined wheels or from stampings.

Meanwhile the first commercial turbomolecular pumps were dual-flow ("horizontal") pumps [96] having a double-
ended rotor, pumping the particles from a central inlet toward both sides and reuniting the gas flow in a common
foreline; single-flow ("vertical") turbomolecular pumps, using single-ended rotors, became available in

< previous page page_241 next page >


< previous page page_242 next page >
Page 242

1969 [100]. The double-ended rotor design allows a more stable bearing design which is advantageous for easy
balancing and lower vibration levels. The single-flow design has little conductance losses between the inlet flange and
the rotor, whereas the dual-flow design suffers losses from the inlet to both sides.

Today only a few models of commercial turbomolecular pumps still use the dual-flow design.

4.6.2.2
Rotor Suspension

The development of the turbomolecular pump was quite spectacular concerning the reduction in size. This was possible
due to (a) an increase of the circumferential (tip) speed from 150 m/s in 1958 to approximately 500 m/s today and (b)
changes of the rotor geometry. These high tip speeds relate to high rotational speeds of the rotors exerting high loads to
the rotor's suspension.

Today most of the commercial turbomolecular pumps are equipped with lubricated mechanical rotor bearings, or a
combination of permanent magnet bearings at the high-vacuum side and a lubricated mechanical bearing at the fore-
vacuum side. Depending on the wheel diameter the rotational speed of the rotor goes up to 90,000 rpm. These high rpm
have become possible because of the advances in bearing and balancing technology, without sacrificing the high
reliability of the turbomolecular pump. Today, high-precision ball bearings are available, which, when specially tuned
to a certain turbomolecular pump rotor at comparable radial and axial loads, even at much higher rpm, have a superior
lifetime compared to bearings of older design.

"Ceramic" bearings (ceramic balls) are widely used today. The ceramic balls exert lower centrifugal forces and lower
stress on the races than metal balls, are harder and more temperature-stable, and, therefore, have a stable spherical shape
and minimal wear on balls and race. Their surface is smoother, leading to less friction, and the pairing of different
materials (ceramic balls/steel races) avoids micropitting. Therefore these bearings are more reliable even under
lubrication-starved conditions.

4.6.2.3
Lubrication of Mechanical Bearings

Three main requirements have to be fullfilled by the lubricant used for the mechanical rotor bearings: (1) The lubricant
has to cool the rotor bearings since these are running under fore-vacuum conditions, and a heat transfer from the inner to
the outer race is possible only by the lubricant; (2) a low vapor pressure is required; and (3) good lubrication properties
at high speed are necessary. Today most of the lubricants used have a synthetic base.

Even at low rotor weights a minimum fluid flow through the bearings for heat transfer is necessary. Many smaller
turbomolecular pumps use a wick lubrication system or, together with "ceramic" bearings, a grease lubrication system,
while most larger turbomolecular pumps have pump systems circulating the oil.

4.6.2.4
Magnetic Rotor Suspension

After some unsuccessful experiences with commercial "gas bearing" turbomolecular pumps [101], today commercial
magnetic bearing turbomolecular pumps are available in which one, two, three, or all of the necessary five degrees of
freedom of the rotor are actively controlled either by an analog or by a digital electronic system. The position of the
rotor spindle is monitored by sensors, and the electronics corrects the spindle position to the data supplied by the
sensors. There is no mechanical friction and hence no wear. The price of these magnetic pumps is still considerably
higher as compared to standard ball-bearing turbomolecular pumps, limiting their general use.

< previous page page_242 next page >


< previous page page_243 next page >
Page 243

4.6.2.5
Balancing and Vibration

The dynamic balancing of the rotor of a turbomolecular pump is of great importance to minimize vibration and noise
levels, which are related to the mechanical bearing lifetime. Due to the high rotational speed of turbomolecular pump
rotors, the centrifugal forces associated with the residual unbalance (material inhomogeneities, radial bearing play,
geometrical imperfections) attain considerable values and transmit vibrations to the body of the pump.

The balancing process modifies the mass distribution of the rotor by adding or taking off material in order to bring the
rotational axis as close as possible to the principal axis of inertia. To further reduce the influence of the residual
unbalance, the mechanical rotor bearings within a turbomolecular pump are held in elastic "antivibration" rings which
effectively dampen the residual unbalanced forces.

Dynamic multifrequency balancing of the rotor is generally done in several balancing planes. This relative complex
procedure in the last few years has become simplified by the use of computers and dedicated software. Modern
turbomolecular pumps have very low residual vibration amplitudes, below 0.02 µm. These low values are necessary for
the use of a turbomolecular pump with vibration-sensitive instruments, such as mass spectrometers and electron
microscopes.

4.6.2.6
Rotor Materials

Most of the commercial turbomolecular pumps use high-strength Al alloys as rotor material. Compared to other high-
strength materials, such as Ti and steel alloys, these Al alloys are lighter, are much easier to machine, and have
sufficient thermal stability for the operational temperatures even at the typical turbomolecular pump bakeout cycles.
The use of ceramics (Si3N4) has been reported for the use of turbomolecular pumps in very strong magnetic fields
[102]. A typical maximum stress in the root of the rotor blades of a high-speed turbomolecular pump rotor at operating
conditions is in the range between 50 and 150 N/mm2, well below the 0.2% elongation limits of adequate Al alloys
[103].

4.6.2.7
Drive Systems

The drive rotor of a turbomolecular pump is an integral part of the turbomolecular pump rotor and together with its
drive stator can be located in the fore vacuum area. Today three different motor systems are used: DC motor, AC motor,
and hysteresis motor. The somewhat more expensive DC motor has lower energy consumption and energy losses than
the other motors. The motors are driven by solid-state frequency converters. Some of these converters can operate the
turbomolecular pump at variable rotor speeds.

For special application (e.g., radiation) motor-driven frequency converters are used.

4.6.3
Applicational Considerations

4.6.3.1
Venting

After a power shutdown a turbomolecular pump should be vented; otherwise it will be contaminated by oil vapors as a
result of pressure equalization between exhaust and inlet. By venting with a dry gas to atmospheric pressure, this effect
can be suppressed and a contamination of the vacuum system avoided. Most of the turbomolecular pumps, with the
exception of the larger pumps, can be vented at the high-vacuum side at full rotor speed. The venting air bursting into
the pumps puts a force of 1 kg on every cm2 of blade surface, which for larger areas represents a high load to the rotor
suspension. If the vacuum system cannot be vented
< previous page page_243 next page >
< previous page page_244 next page >
Page 244

directly, it is advisable to vent the turbomolecular pump by means of a special vent port which opens into the
compression stages. Venting a turbomolecular pump from the contaminated foreline should be avoided, by all means.

4.6.3.2
Baking

In order to attain low ultimate pressures in the high-vacuum and ultrahigh-vacuum range, the internal surfaces of a
turbomolecular pump (rotor, stator, housing) can be baked out. Due to the temperature sensitivity of the Al alloy used
for the rotors, there is a limit to the maximum value of the baking temperature. This maximum has to be well below the
critical temperature for the rotor alloy with respect to its strength. Typical baking temperatures are in the range of
100140°C, using the bake-out systems supplied by the pump's manufacturer.

4.6.3.3
Cooling

In order to dissipate the frictional heat from the bearing areas, the motor, and the heating by gas throughputs at high
pressures, the bearing areas of turbomolecular pumps have to be cooled. Meanwhile, for small pumps, convection
cooling is sufficient, whereas larger pumps are equipped with fans for cooling. For many pumps, water cooling is
common.

4.6.3.4
Operation in Magnetic Fields

Turbomolecular pumps with their metallic rotors in the presence of magnetic fields experience induction of eddy
currents which, due to heating, can cause serious problems concerning the material strength and the tolerances between
rotating and stationary parts in the turbomolecular pump [104]. The eddy-current loss ∆P, is the amount of energy
transformed by the eddy current into heat, and can be represented by the Eq. (4.141):

The eddy-current loss ∆P is proportional to the square of the magnetic flux B, proportional to the square of the rotor's
rotational frequency f, and inversely proportional to the resistivity δ of the rotor material. In addition ∆P is dependent on
the shape of the rotor. Since in actual turbomolecular pumps the length of the rotor greatly exceeds its diameter, the
maximum heating of the rotor results from the component of the magnetic field in a radial direction.

Therefore, turbomolecular pumps with metallic rotors can be used in magnetic fields only if certain maximum values of
the magnetic flux density will not be exceeded. These maximum values are specified by the manufacturers and, for
static magnetic fields perpendicular to the axis of rotation, typically are in the range of 1030 mT. With pulsed magnetic
fields, higher maximum values of Bmax are admissable. If the magnetic field is applied for a time t1 and off for a time
t2 and since the heating of the rotor is proportional to the square of B, the following relation for Bmax(pulse) for pulsed
magnetic fields applies:

In case of higher flux densities the turbomolecular pump will have to be shielded magnetically, which, however, can
create problems with the distribution of the magnetic field itself. An experimental turbomolecular pump with a ceramic
rotor has been reported which has been tested with a magnetic flux density of 460 mT [102].

< previous page page_244 next page >


< previous page page_245 next page >
Page 245

In this context it has to be kept in mind that the electronic drive systems and the control systems (magnetic suspension)
of turbomolecular pumps create magnetic stray fields in their vicinity. These stray fields can influence other electronic
and magnetic systems. Depending on the position relative to the turbomolecular pump, typical values for these stray
field are in the range of 150 µT.

4.6.3.5
Pumping Corrosive Gases

Modern manufacturing techniques require pumping processes where the ultimate system pressure has to be in the high-
vacuum range and corrosive gases have to be handled (e.g., below 101 Pa). At these pressures the rotor material will not
be attacked by the corrosives. However, the higher gas density at the fore-vacuum side in the rotor bearing area requires
the use of special lubricants (Fomblin®, Krytox®, etc.) which are corrosion-resistant.

In the semiconductor industry, Si and Al are etched in plasma etch machines (e.g., at pressures above 101 Pa). To avoid
etching and corrosion of the turbomolecular pump's Al rotor and other parts, all internal parts of the turbomolecular
pump having contact with the corrosives (e.g., Cl2, BCl3, CCl4) are either made of corrosion-resistant materials or are
specially coated (e.g., Ni-plated). Such coatings of the rotor can reduce the attainable ultimate pressure of a
turbomolecular pump to 5 × 106 Pa. Besides using special lubricants, for further protection the turbomolecular pumps
(for plasma etching) are equipped with a purge gas system which admits inert gas (N2, Ar, typically 30 sccm) directly
into the bearing area of the turbomolecular pump, where it creates a directed flow to the exhaust side of the pump. This
flow prevents the corrosive gases from entering the bearing area.

In some processes the reactive molecules of the etch gas from AlCl3 which is pumped off with the etch gas, and at
temperatures of 69°C solidifies inside the pump. To avoid this, the pumps will be heated to keep the temperature level
above this temperature. The use of turbomolecular pumps with magnetically suspended rotor will reduce the above-
mentioned problems.

4.6.3.6
Pumping Toxic or Radioactive Gases

An additional range of problems has to be solved in the case of a turbomolecular pump pumping toxic or radioactive
gases (e.g., tritium in plasma fusion installations). Due to the safety hazards involved, the turbomolecular pump has to
be extremely tight. With a "tritium" turbomolecular pump, all seals to the atmosphere are made of metal. The integral
leak rate is below 107 Pa · liter · s1.

4.6.3.7
Turbomolecular Pumps in Combination with Other Pumps

In order to increase the pumping speed of a turbomolecular pump for hydrogen or water vapor and make use of its
constant throughput and pumping speed, turbomolecular pumps have been combined with Ti sublimation pumps [99]
or, recently, with cryopumps [105] (see Chapter 9).

4.6.4
Performance Data of Commercial Pumps

The variations in the performance of turbomolecular pump by different manufacturers depend on their actual design
(blade design, rotor staging, etc.).

< previous page page_245 next page >


< previous page page_246 next page >
Page 246

4.6.4.1
Compression

Figure 4.29 shows typical values for the maximum compression Kmax of a turbomolecular pump for different gases as a
function of the foreline pressure.

Depending on the manufacturer, turbomolecular pumps have the following compression values: H2, 1 × 102 to 1 × 105;
He, 5 × 102 to 1 × 107; N2, 5 × 106 to 1 × 1010.

4.6.4.2
Pumping Speed

Figure 4.30 shows the pumping speed of a turbomolecular pump for different gases as a function of the inlet pressure.

Turbomolecular pumps are available in a pumping speed range from 35 to 25,000 liter · s1.

Type designations of turbomolecular pumps often use the pumping speed for nitrogen; therefore the pumping speed for
N2 is used as the 100% value. This value is
Fig. 4.29
Maximum compression of turbomolecular
pump as a function of the foreline
pressure (Balzers Pfeiffer).

< previous page page_246 next page >


< previous page page_247 next page >
Page 247

Fig. 4.30
Pumping speed of turbomolecular pump as a function
of the inlet pressure (Balzers Pfeiffer).

compared to the pumping speed data for H2 and He from the catalogues of different manufacturers: H2, 34133%; He,
60133%.

4.6.4.3
Ultimate Pressure

The ultimate pressure of a commercial turbomolecular pump is generally between 108 and 107 Pa, using metal flange
seals and a two-stage rotary backing pump.

4.7
Combined Molecular Drag and Turbomolecular Pumps

In the past several years, different kinds of combined molecular drag and turbomolecular pumps have been developed to
improve the high-pressure performance of the turbomolecular pump and to attain very low ultimate pressures.
Combinations of a turbomolecular and a molecular drag pump basically work on the same principle. Momentum is
transferred from a rapidly moving rotor surface to the particles to be pumped. Therefore the theoretical considerations
are the same.

In 1975, Schittko and Schmidt [106] tried to combine a Holweck molecular drag pump with a turbomolecular pump in
order to raise the compression for light gases. In 1973, Sawada and Tanigushi [107] proposed the concept of combined
molecular drag pumps and turbomolecular pumps in which a turbomolecular pump and a drag pump stage are mounted
in series on a single shaft (''Compound Molecular Pump") for the purpose of expanding the working pressure to a higher
pressure range. Maurice [101] developed in 1974 such a pump ("Hybrid turbomolecular pump"). However, in these
attempts the gaps between rotary and stationary parts were still extremely narrow, causing poor reliability at high-speed
rotation.

With the progress in semiconductor manufacturing there has been an increasing demand for "clean" vacuum pumps
with high throughputs and capable of being used in the process range of 1103 Pa. This demand led to the development
of combined turbomolecular pumps and molecular drag pumps in the 1980s. This combination has two effects: H2
compression is increased from typically 103 to 105, allowing ultimate pressures of below 108 Pa to be obtained and,
more significantly, allowing the

< previous page page_247 next page >


< previous page page_248 next page >
Page 248
pressure in the backing line to rise above, for example, 103 Pa and higher, thereby enabling it to be backed with a dry
membrane pump. In comparison, turbomolecular pumps, without the molecular drag pump have to be backed by a two-stage
rotary pump to ensure that the compression stages do not operate under laminar flow conditions.

With an integral molecular drag pump, oil is still present in the turbomolecular pump bearings and in the drive mechanism of
the diaphragm pump (separated from the high-vacuum space), but the elimination of the oil-sealed rotary pump is a substantial
improvement.

4.7.1
Design Considerations

This pump design combines the advantages of a multistaged turbomolecular pumps and molecular drag pumps: The
turbomolecular pump section provides a high pumping speed and low ultimate pressures, and the molecular drag pump section
provides a high compression and extends the fore-vacuum tolerance up to 102 to 5 × 103 Pa. These capabilities allow the use of
simple small, dry, and inexpensive backing pumps (e.g., membrane pumps) and therefore allow a dry evacuation from
atmospheric pressure down to 109 Pa.

These pump combinations are available in several different design configurations:

TMP stages + multistage MDP (disc type, derived from Gaede [90])

TMP stages + multistage MDP (drum type, derived from Holweck [91])

TMP + multistage MDP (drum type, derived from Holweck [91])

TMP stages + multistage MDP (disc type, derived from Siegbahn [92]).

4.7.2
Typical Performance Data for Commercial Combined Molecular Drag and Turbomolecular Pumps

4.7.2.1
Compression

H2, 2 × 103 to 2 × 107; He, 7 × 103 to 2 × 109; N2, 5 × 107 to 1 × 1012

4.7.2.2
Pumping Speed

N2, 100%; H2, 24133%; He, 42133%

4.7.2.3
Ultimate Pressure

Depending on how many turbomolecular pump and molecular drag pump stages are used in series, the attainable ultimate
pressure is in the range of 5 × 109 to 1 × 107 Pa.

4.8
Backing Pumps

Because all of the above-mentioned turbomolecular pumps and molecular drag pumps and combination of these pumps need a
backing pump, it is worthwhile to look at the existing types of backing pumps, in order to be able to evaluate which

< previous page page_248 next page >


< previous page page_249 next page >
Page 249

Table 4.5. Typical Backing Pump Data


Type Ultimate pressure (Pa) Quality of Vacuum
Membrane Absolutely dry
< 103
Rotary vane Oil vapors
< 101
Piston Technically dry
<1
Multiroots Technically dry
< 10
Claw roots Technically dry
< 10

backing pump should be used together with the high-vacuum pump (see Chapter 9). In principle, the selection of an
adequate backing pump depends on its pumping speed, its pressure range, the quality of the fore-vacuum produced, its
price, and its size. Typical backing pump data are shown in Table 4.5.

Backing pumps producing an "absolutely dry" vacuum do not have any oil or grease in possible contact with the
pumping area. "Technically dry" pumps do not have any oil or grease in the pumping area; however, they have grease or
oil (e.g, in the bearing or feed-through area) that is kept away from the pumping area by dynamic seals, because there
are labyrinths, and so on.

Together with magnetically levitated combined turbomolecular pump and molecular drag pumps, even the most
demanding applicational requirements can be met.

< previous page page_249 next page >


< previous page page_251 next page >
Page 251

Part III
Regenerative Drag Pumps

Nigel T. M. Dennis

4.9
Regenerative Drag Pumps

Capacities available: 30540 m3 h1 (25150 liter·s1)

Operating pressure range: 1000 to less than 104 mbar

4.9.1
Mechanism

The regenerative pump is a high-speed single-shaft vacuum pump capable of delivering to atmospheric pressure. It is a
dynamic machine that relies on momentum transfer from fast moving blade rows that create the pumping action. The
name "regenerative" comes from the way the mechanism circulates the gas repeatedly through the same blade row, each
time "regenerating" the forward momentum. Other names are used: side channel, vortex, or peripheral flow pump.

The advantage of this pump is that the high shaft speed results in small stage size, making the pump extremely compact.
The high rotational speed also allows other mechanisms such as a molecular drag stage (see Section 4.5) to be
effectively used on the same shaft.

A single stage of the pump consists of a row of blades mounted on a rotor; these protrude into the flow channel (see Fig.
4.31). As the blades move around they generate a spiral vortex within the channel that circulates the gas through the
blades as it moves along the channel, thus creating a helical flow path. This ensures that

Foundations of Vacuum Science and Technology, Edited by James M. Lafferty.


ISBN 0-471-17593-5  1988John Wiley & Sons,Inc.

< previous page page_251 next page >


< previous page page_252 next page >
Page 252

Fig. 4.31
One stage of a regenerative peripheral flow pump
indicating direction of gas flow and where the flow
channel is interrupted by the "stripper."

high-speed gas from the blades is continually fed into the channel, diffusing into the bulk flow to generate the pressure
rise. Within a single stage the gas may circulate many times through the blades, creating a high compression ratio.
Typical performances range from a compression of 2 to about 10, with the latter occurring in the very best machines.
The blades are shaped to catch the incident gas with the minimum of turbulence, to deflect it, and to throw it forward
into the bulk gas flow. The inlet and exhaust of a pump channel are separated by a restrictive section through which
only the blades may pass. This is generally known as the stripper. Pockets of gas between the blades are also carried
through the stripper; these form a carry-over volume, limiting the efficiency and compression of the machine. The blade
size must, therefore, be small compared to the channel size. The amount of gas carried over to the inlet can be reduced
by allowing the closed pockets within the stripper to expand through a connection to an intermediate pressure point
within the stage.

Sixsmith [108] revolutionized the understanding and design of these machines and showed that practical compressors
and vacuum pumps could be made. He developed many of the features described above and demonstrated a prototype
machine with a maximum compression of 10 and a working flow of 250 m3·h1.

Blade speeds are usually below Mach 1, and Mach 0.5 to 0.8 appears to be suitable. Speeds above Mach 1 result in
shock wave problems.

The function of the blades is to impart momentum to the gas as efficiently as possible.
< previous page page_252 next page >
< previous page page_253 next page >
Page 253

Fig. 4.32
A typical pump with inlet drag stages.
Fig. 4.33
Speed and power curve of a 240-m3·h1 regenerative
pump with two drag stages.

At the present stage of development, five or more stages of regenerative compressor are used with an inlet drag stage or
stages based on the Holweck or Siegbahn principle. The regenerative stages are typically capable of reaching 1 mbar.
The drag stages extend the ultimate to pressures to as low as 1 × 104 mbar [109]. A typical pump

< previous page page_253 next page >


< previous page page_254 next page >
Page 254

cross section is shown in Fig. 4.32. The speed curve obtained from a medium-size pump of this type is indicated in Fig.
4.33.

The power requirement of a regenerative pump is high because of the poor efficiency of the mechanism which is due to
the slip required to generate a useful compression. A typical power curve is shown in Fig. 4.33.

Pumps of this type are susceptible to dust accumulation in their mechanism and, therefore, should generally be used on
clean duties. They are, however, the only single-shaft pumps to achieve this level of vacuum when working against
atmospheric pressure.

References

1. V. V. Fondrk, Chem. Eng. Prog. 49, 3 (1953).

2. W. Gaede, Ann. Phys. (Leipzig) [4] 46, 357 (1915).

3. I. Langmuir, Gen. Electr. Rev. 19, 1060 (1916); J. Franklin Inst. 182, 719 (1916).

4. W. Gaede, Z. Tech. Phys. 4, 337 (1923); German Patent No. 286,404.

5. C. R. Burch, Nature (London) 122, 729 (1928).

6. K. C. D. Hickman and C. R. Sanford, Rev. Sci. Instrum. 1, 140 (1930).

7. K. C. D. Hickman, J. Franklin Inst. 221, 215 (1936).

8. L. Malter and N. Marcuvitz, Rev. Sci. Instrum. 9, 92 (1938).

9. N. Embree, U.S. Pat. 2,150,676 (1939).

10. K. C. D. Hickman, J. Appl. Phys. 11, 303 (1940).

11. M. H. Hablanian and J. C. Maliakal, J. Vac. Sci. Technol. 10, 58 (1973).

12. B. D. Power, High Vacuum Pumping Equipment. Reinhold, New York, 1966.

13. L. Laurenson, Vacuum 37, 609 (1987).

14. R. Jaeckel, H. G. Nöller, and H. Kutscher, Vak.-Tech. 3, 1 (1954).

15. H. Kutscher, Z. Angew. Phys. 7, 218, 229, 234 (1955).

16. N. M. Bulgakova, G. A. Khramov, O. A. Nerushev and A. K. Rebrov, Vacuum 44, 749 (1993).

17. M. H. Hablanian, J. Vac. Sci. Technol. A 5, 2552 (1987).

18. B. B. Dayton, Vacuum 15, 53 (1965).

19. W. Steckelmacher, Vacuum 15, 503 (1965).

20. W. Steckelmacher and D. Turner, J. Sci. Instrum. 43, 893 (1966).

21. P. Clausing, Ann. Phys. (Leipzig) [5] 12, 961 (1932).


22. C. W. Oatley, Br. J. Appl. Phys. 8 (15), 495 (1957).

23. W. Harries, Z. Angew. Phys. 3, 296 (1951).

24. T. L. Ho, Physica (Amsterdam) 2, 386 (1932).

25. R. F. Coe, J. Sci. Instrum. 32, 207 (1955).

26. D. Latham, B. D. Power, and N. T. M. Dennis, Vacuum 1, 97 (1951); 2, 33 (1952).

26.a. P. Alexander, J. Sci. Instrum. 25, 313 (1948).

27. A. Herlet and G. Reich, Z. Angew. Phys. 9, 14 (1957).

28. S. F. Kapff and R. B. Jacobs, Rev. Sci. Instrum. 18, 581 (1947).

29. J. Blears, Nature (London) 154, 20 (1944); Proc. R. Soc. London, Ser. A 188, 62 (1947).

30. D. J. Crawley, E. D. Tolmie, and A. R. Huntress, Trans. Natl. Vac. Symp. 9, 399403 (1962).

< previous page page_254 next page >


< previous page page_255 next page >
Page 255

31. A. R. Huntress, A. L. Smith, B. D. Power, and N. T. Dennis, Trans. Natl. Vac. Symp. 4, 104111 (1958).

32. K. C. D. Hickman, Trans.