NATIONAL NUCLEAR ENERGY SERIES

Manhattan Project Technical Section

Division III-Volume 1B

THE THEORY OF ISOTOPE SEPARATION

AS APPLIED T O THE LARGE-SCALE PRODUCTION OF u~~~
THE THEORY O F ISOTOPE SEPARATION
AS A P P L I E D T O
THE LARGE-SCALE PRODUCTION O F U235

KARL COHEN
Director, Atomic Energy Division,
The H. K. Ferguson Company; formerly Director,
Theoretical Division, S A M Laboratories

Edited by

GEORGE M. MURPHY
Washington Square College, New York University;
formerly at SAM Laboratories, Columbia University

F i r s t Edition

New York - Toronto . London

McGRAW-HILL BOOK COMPANY, INC.
1951

THE THEORY OF ISOTOPE SEPAMTION
AS APPLIED TO THE LARGE-SCALE PRODUCTION OF
Copyright, 1951, by the
McGraw-Hill Book Company, Inc.
P r i n t e d i n t h e United S t a t e s of A m e r i c a
Copyright assigned, 1951, to the General Manager
of the United States Atomic Energy Commission.
All rights reserved. This book, o r parts thereof,
may not be reproduced in any form without p e r -
mission of the Atomic Energy Commission.
Lithoprinted
by
Edwards Brothers, Incorporated
Ann Arbor, Michigan
FOREWORD
The wartime project f o r development of atomic energy was a r e -
uZ3' markable feat of cooperation and accomplishment by government,
science, industry, labor, and the military s e r v i c e s aimed exclusively
at the military application of atomic energy. Our present national
atomic energy program, expanding upon the previous developments,
i s directed not only toward the assurance of national security but
also toward the realization of the immense potential benefits atomic
energy holds f o r our civilization. The record of progress and t h e r e -
sults of extensive scientific investigation~and engineering develop-
ment a r e contained in the National Nuclear Energy Series. This
knowledge, which offers the basis of world -wide benefits from nuclear
science, i s being published in the established scientific tradition, not
solely to meet the precise needs of science but a l s o in support of the
high goals of the American people s e t forth in the Atomic Energy Act.
The work reported in this s e r i e s is a tribute to all the scientists en-
gaged in both the Manhattan Project and the postwar Atomic Energy
Commission program.
Gordon Dean, Chairman
U. S. Atomic Energy Commission

ACKNOWLEDGMENT
The Manhattan Project Technical Section of the National Nuclear
Energy Series embodies results of work done in the nation's wartime
atomic energy program by numerous contractors, including Columbia
University. The arrangements for publication of the s e r i e s volumes
were effected by Columbia University, under a contract with the
United States Atomic Energy Commission. The Commission, for
itself and for the other contractors who contributed to this series,
wishes to record here its appreciation of this service of Columbia
University in support of the national nuclear energy program.
PREFACE
This volume is one of a series which has been preparedas a record
of the research work done under the Manhattan Project andthe Atomic
Energy Commission. The name Manhattan Project was assigned by
the Corps of Engineers, War Department, to the far-flung scientific
and engineering activities which had a s their objective the utilization of
atomic energy for military purposes. In the attainment of this objec-
tive, there were many developments in scientific and technical fields
which a r e of general interest. The National Nuclear Energy Series
(Manhattan Project Technical Section) is a record of these scientific
and technical contributions, a s well a s of the developments in these
fields which a r e being sponsored by the Atomic Energy Commission.
The declassified portion of the National Nuclear Energy Series,
when completed, is expected to consist of some 60 volumes. These
will be grouped into eight divisions, a s follows:
Division I - Electromagnetic Separation Project
Division I1 - Gaseous Diffusion Project
Division I11 - Special Separations Project
Division IV - Plutonium Project
Division V - Los Alamos Project
Division VI - University of Rochester Project
Division VII - Materials Procurement Project
Division VIII - Manhattan Project
Soon after the close of the war the Manhattan Project was able to
give its attention to the preparation of a complete record of the
research work accomplished under Project contracts. Writing pro-
grams were authorized at all laboratories, with the object of obtaining
complete coverage of Project results. Each major installation was
requested to designate one o r more representatives to make up a
committee, which was first called the Manhattan Project Editorial
Advisory Board, and later, after the sponsorship of the Series was
assumed by the Atomic Energy Commission, the Project Editorial
Advisory Board. This group made plans to coordinate the writing
programs at all the installations and acted a s an advisory group in
all matters affecting the Project-wide writing program. Its last
meeting was held on Feb. 9, 1948, when it recommended the publisher
for the Series.
vii
viii PREFACE PREFACE ix

The names of the Board members and of the installations which Many difficulties were encountered in preparing a unified account
they represented a r e a s follows: of Atomic Energy Project work. F o r example, the Project Editorial
Advisory Board was the f i r s t committee ever organized with r e p r e -
Atomic Energy Commission sentatives from every major installation of the Atomic Energy Project.
Public and Technical Information Alberto F. Thompson Compartmentation for security was s o rigorous during the war that
Service
it had been considered necessary to allow a certain amount of dupli-
Technical Information Division, Brewer F. Boardman cation of effort rather than to permit unrestricted circulation of
Oak Ridge Extension r e s e a r c h information between certain installations. As a result, the
Office of New York Operations Charles Slesser, J. H. Hayner, writing programs of different installations inevitably overlap markedly
W. M. Hearon * in many scientific fields. The Editorial Advisory Board has exerted
Brookhaven National Laboratory Richard W. Dodson itself to reduce duplication in s o f a r a s possible and to eliminate
Carbide & Carbon Chemicals R. B. Korsmeyer, W. L. Harwell, discrepancies in factual data included in the volumes of the NNES.
Corporation (K-25) D. E. Hull, E z r a Staple In particular, unified Project-wide volumes have been prepared
Carbide & Carbon Chemicals Russell Baldock
on Uranium Chemistry and on the Analysis of Project Materials.
Corporation (Y-12) 'f Nevertheless, the reader will find many instances of differences in
results o r conclusions on similar subject matter prepared by different
Clinton Laboratories 1 J. R. Coe
authors. This has not seemed wholly undesirable for several reasons.
General Electric Company, Hanford T. W. Hauff F i r s t of all, such divergencies a r e not unnatural and stimulate in-
General Electric Company, John P. Howe vestigation. Second, promptness of publication has seemed more
Knolls Atomic Power Laboratory important than the removal of all discrepancies. Finally, many Proj-
Kellex Corporation John F. Hogerton, Jerome Simson, ect scientists completed their contributions s o m e time ago and have
M. Benedict become engrossed in other activities s o that their time has not been
Los Alamos R. R. Davis, Ralph Carlisle Smith available for a detailed review of their work in relation to similar
National Bureau of Standards C. J. Rodden
work done a t other installations.
The completion of the various individual volumes of the Series has
Plutonium Project also been beset with difficulties. Many of the key authors and editors
Argonne National Laboratory R. S. Mulliken, H. D. Young
have had important responsibilities in planning the future of atomic
Iowa State College F. H. Spedding energy research. Under these circumstances, the completion of this
Medical Group R. E. Zirkle technical s e r i e s has been delayed longer than its editors wished. The
SAM Laboratories  G. M. Murphy volumes a r e being released in their present form in the interest of
presenting the material a s promptly a s possible to those who can
Stone & Webster Engineering B. W. Whitehurst
Corporation make use of it.
University of California R. K. Wakerling, A. Guthrie
The Editorial Advisory Board
University of Rochester D. R. Charles, M. J. Wantman

*Represented Madison Square A r e a of the Manhattan District.

The Y-12 plant a t Oak Ridge w a s operated by Tennessee Eastman Corporation until May 4 ,
1947, a t which time operations were taken over by Carbide & Carbon Chemicals Corporation.
$Clinton Laboratories w a s the f o r m e r name of the Oak Ridge National Laboratory.
§SA (Substitute Alloy Materials) was the code name for the laboratories operated by
Columbia University in New York under the direction of D r . H. C. Urey, where much of the
experimental work on isotope separation was done. On Feb. 1, 1945, the administration of
these laboratories became the responsibility of Carbide & Carbon Chemicals Corporation.
Research in progress t h e r e was t r a n s f e r r e d to the K-25 plant a t Oak Ridge in June, 1946, and
the New York laboratories were then closed.

The Manhattan Project Technical Section of the National Nuclear
Energy Series is intended to be a comprehensive account of the sci-
entific and technical achievements of the United States program for
the development of atomic energy. It is not intended to be a detailed
documentary record of the making of any inventions that happen to be
mentioned in it. Therefore, the dates used in the Series should be
regarded a s a general temporal frame of reference, rather than a s
establishing dates of conception of inventions, of their reduction to
practice, o r of occasions of first use. While a reasonable effort has
been made to assign credit fairly in the NNES volumes, this may, in
many cases, be given to a group identified by the name of its leader
rather than to an individual who was an actual inventor.
COLUMBIA UNIVERSITY PROJECT FOREWORD
Government-supported research on nuclear energy f i r s t occurred
in early 1940 when certain funds made available by the Army, Navy,
and National Bureau of Standards were used for experiments a t
Columbia University. The subsequent history of this project a s it
expanded in many other universities and industrial laboratories is
told in detail in the Smyth Report. Two of the major programs began
a t Columbia University: the uranium-graphite pile reactor and the
gaseous diffusion method of uranium 235 separation.
Lack of sufficient laboratory space a t Columbia for the develop-
ment of both programs led to the transfer in the spring of 1942 of the
uranium-graphite reactor program to the University of Chicago. The
program of separation of uranium 235 by gaseous diffusion, and vari-
ous ancillary projects, developed rapidly until not only all available
space in the Columbia University buildings was occupied but, in addi-
tion, an even greater amount of rented space.
The project for the separation of uranium isotopes by gaseous
diffusion was first carried forward through various government con-
t r a c t s under the direction of Harold C. Urey, Professor of Chemistry,
and John R. Dunning, Assistant Professor of Physics. In 1943 the
work was unified under a single contract with the Manhattan Engineer
District with Professor Urey as Director. An account of the research
on the major assignment, the separation of the uranium isotopes, is to
be found elsewhere in the National Nuclear Energy Series. The vol-
umes that this preface introduces deal with the general mathematical
theory of isotope separation, with some new experimental methods of
separating isotopes, with spectroscopic properties of uranium com-
pounds, and with the chemical and physical properties of heavy water
and other deuterium compounds. Most of this work was done at
Columbia, but parts of it were carried on a t the National Bureau of
Standards and a t The Johns Hopkins University.
George B. Pegram
Chairman, Columbia University Division
of War Research, 1941-1945

PUBLISHER'S NOTE
Although every effort has been made to ensure accuracy in ref-
erences, at the time of publication of this book some of the other
volumes of the Series had not been completed. It is therefore pos-
sible that some of the references to other volumes a r e in e r r o r . It is
hoped that the extensive c r o s s checking which has been done in the
preparation of this volume has resulted in keeping such e r r o r s to a
minimum.
AUTHOR'S PREFACE
by the theoretical division of the SAM Laboratories and its anteced-
ents, which the author was privileged to lead. The work was done
over a period of years, from 1940 to 1945, under contract to various
government agencies. However, we have in some instances drawn
on outside sources for special topics, and some of the material in the
volume is new. Since the reports have not been and probably will not
be published, we have avoided distracting the reader with references
in the body of the text which would be meaningless to him. At the end
of each chapter a reference section of relevant reports is appended
for purposes of record. Specific references to published articles a r e
footnoted a s they occur in the text.
Although the theory was developed piecemeal in response to vari-
ous demands, it is exposed a s a logical whole according to our pres-
ent understanding and experience. In Chaps. 1 to 5, on various aspects
of cascade theory, certain generalities that a r e common to many dif-
ferent separation methods a r e collected. Chapter 6 treats the cen-
trifugal method. Chapter 7 is composed of sections briefly outlining
two-phase separations (including chemical exchange and distillation)
and the thermal-diffusion method, and a section presenting some
remarks on the separation of deuterium.
The studies given a r e a selection from a much larger volume of
material. More emphasis is put on general principles and concepts
than on an exhaustive collection of particular results, no matter how
elegant. It is thought unlikely that any detailed problem will recur in
exactly the same form and that a thorough grasp of the fundamentals
will be more useful to the reader. In any event, the most important
applications of the theory cannot yet be discussed.
While the theory itself is intact, the reader may find that the moti-
vations and explanations a r e occasionally somewhat sketchy. This is
a result of security deletions. We did not feel that they were im-
portant enough to warrant extensive rewriting.
xiii
XIv AUTHOR'S PREFACE

There is no reference to direct experimental proofs of the theory,

although they a r e not lacking. This is only partly a question of space
and security. In a large measure the theory is self-evident, in the
fashion of demonstrations in thermodynamics.
Karl Cohen
1 It is a pleasure to acknowledge indebtedness to Drs. J. R. Dunning,
August 1951
ACKNOWLEDGMENTS
Development of the theory in this volume was influenced by many
discussions with people whose names appear on no theoretical report.
E. T. Booth, and H. A. Boorse, all of Columbia University, and
Dr. F. G. Slack, of Vanderbilt University, for ideas and criticisms.
A British team, composed of Drs. F. Simon, R. Peierls, K. Fuchs,
and N. Kurti, worked out independently the theory of separation a s it
applied to a particular separation. They obtained substantially the
same results, in a different chronological order, corresponding to the
different order in which the problems were attacked. Adequate liai-
son was accomplished in early 1942, at which time a very stimulating
comparison of results and viewpoints was made. This and subsequent
contacts contributed toward clarity of thought and expression.
Much of the credit for the success of these investigations i s due to
Dr. H. C. Urey: on an administrative plane, for his unwavering sup-
port of the theoretical group; and on a scientific plane, for imparting
to us his fine intuitive grasp of the essentials of separation proc-
esses. The concept of separative work, which i s basic for the com-
parison of different methods, was first used by him in 1939.
The author is personally indebted to E. V. Murphree, of the Stand-
a r d Oil Development Company, for allowing him the time to write
this volume.
The role of our colleagues in the theoretical division does not
really need to be mentioned here; their contributions will be amply
demonstrated by the reference sections. Above all, Dr. I. Kaplan,
H. Mayer, and Dr. R. D. Present have contributed substantially to the
success of this work.
The elegant derivation of the value functions and the separative
power in Chap. 1, Sec. 3, is due to Prof. P. A. M. Dirac. The value
function itself, Eq. 1.43, was first determined by Prof. R. Peierls.
The relaxation-time phenomena for square cascades were worked out
with the constant aid of Dr. I. Kaplan. The method of determining the
fundamental constants in the steady state i s due to Dr. W. I. Thomp-
son, of the Standard Oil Development Company. The mathematical
method employed in Chap. 4 on losses was shown to the author by
Dr. R. P. Feynman, then at Princeton University. The solution of the
mi ACKNOWLEDGMENTS

fluctuation problem, which avoids solving the differential equation f o r

the second o r d e r , was devised by Dr. K. Fuchs.
Without the aid of Mrs. L. W. Goodhart, of the Standard Oil Devel-
opment Company, i n editing and in all phases of preparation of the
manuscript, it could never have been finished; special thanks a r e due CONTENTS
to h e r f o r this important s h a r e of the work. Page
1
Karl Cohen Foreword .................................... v
Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . vii
Columbia University Project Foreword . . . . . . . . . . . . . . . . xi
Author's Preface . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xiii
Acknowledgments. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . xv
Introduction .................................. 1
CHAPTER 1
Ideal Cascades. ................................ 5

CHAPTER 2
Square Cascades ............................... 30

CHAPTER 3
Equilibrium Time of a Square Cascade ................. 39

CHAPTER 4

Determination of Cascade Constants .................. 62

CHAPTER 5
The Control Problem ............................ 79

CHAPTER 6
Centrifuges ................................... 103
Ss

CHAPTER 7

Other Separation Methods ......................... 126

xvii
xviii CONTENTS

Appendix A-Roots of a Transcendental Equation ......... 139

Appendix B-Equilibrium Time of Square Cascades

for N 1 -
................................... 144

Appendix C -The Holdup Function . . . . . . . . . . . . . . . . . . . 147

Appendix D -Rayleigh Distillation . . . . . . . . . . . . . . . . . . .
i
Appendix E-Properties of Concurrent Two-phase Elements. .
Index.. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . .
INTRODUCTION
150
The separation of isotopes may be accomplished in many ways. In
154 this volume consideration will be given, in more, or l e s s detail, to
centrifuges, electromagnetic methods, electrolysis, chemical ex-
163 change, thermal diffusion, and distillation. Within each of these cat-
egories there is considerable variety. For example, chemical ex-
change may take place between a liquid phase, between a liquid phase
and a vapor phase by countercurrent scrubbing, or between two liquid
phases with concurrent contacting.
All these methods, except the mass-spectrograph method in iso-
lated instances, have the common feature that one elementary process
does not produce the desired large change in isotopic abundance. The
problem of efficiently multiplying elementary processes to reach
large end results is thus of utmost practical importance, and indeed
of greater importance the smaller the change produced by the ele-
mentary process.
Fortunately, although the methods differ both physically and mathe-
matically in fundamental ways, they a r e sufficiently alike so that their
multiplication problems a r e very closely related. It will therefore be
useful to examine thoroughly the multiplication problem per se before
taking up any of the several separation schemes in detail.
The purpose of the first five chapters is to develop the theory of
cascades a s generally a s possible in order to permit the widest pos-
sible application to future separation problems. Existing applications
of the theory will b e found in the subsequent chapters on particular
separation methods. The point of view taken is that of a development
engineer, and the main concern is the significance of the results in
terms of cascade design. For example, the effect of losses is inves-
tigated in terms of the problem "How much larger do we have to build
our stages to get the same quality and quantity of yield for a given
loss?" rather than "In a given cascade how will the yield vary with
loss?" Consideration is seldom given to problems of the second
kind-problems that occur after the cascade is built-although natu-
rally the same equations govern both questions.
2 THE THEORY OF ISOTOPE SEPARATION
In Secs. 1 to 4 of Chap. 1 the concept of an ideal cascade is exam-
ined from three radically different points of view; Section 5 intro-
duces the idea of equilibrium time. Section 6 gives the general par-
tial differential equation showing the dependence of mole fraction in a
cascade on time and stage number.
The first portion of Chap. 2 is intended to show the steady-state
behavior of single countercurrent columns, or square cascades. In
Secs. 2 and 3 of the chapter, cascades of countercurrent columns a r e
considered. Section 4 covers the squaring off of the top of a cascade.
Time-dependent phenomena in square cascades a r e dealt with in
Chap. 3, along with the methods for experimental determination of
the fundamental constants of a separating element.
Chapter 4 considers the increase in size of a cascade with losses,
the efficient combination of plants using different separation proc-
esses, and the prime operating problem of the control of a cascade.
Chapter 5 deals with the mathematical theory of the effect of dis-
turbances in operation on the production of the cascade. Chapter 6
describes the different types of centrifuges and gives the partial dif-
ferential equation of the centrifuge. Other separation methods a r e
presented in Chap. 7.
Only for certain kinds of processes-"infinitesimal" processes,
countercurrent columns, and the mass spectrograph-is the theory
developed explicitly, in such a way a s to permit immediate applica-
tion. In the last chapter, which includes the deuterium problem, a r e
introduced cascade problems that a r e not treated in Chaps. 1 to 5.
But no one who has absorbed the first five chapters will be at a loss
to solve the problems. Likewise, among the infinite variety of non-
ideal cascades, only square cascades have been selected a s being of
universal interest. The reader will be able to work out for himself
such variants a s cascades in the form of a lozenge (the "cascade-of-
cascades" in British terminology).
The generality of the treatment still leaves something to be de-
sired, since the entire treatment is restricted to binary isotope mix-
tures. This is a consequence of the practical origin of the theory.
In a general way an attempt has been made to avoid lengthy mathe-
matical demonstrations, except when they a r e indispensable to the
argument. For this reason there is only a very general treatment of
the control problem, which is beset with many purely mathematical
difficulties. A detailed study of the equilibrium-time problem of an
ideal cascade has also been omitted; here again the problem has
more mathematical than physical interest. Some of the sections that
a r e mostly mathematical might perhaps be omitted on a first reading.
INTRODUCTION 3
They a r e Secs. 2 and 3 of Chap. 2; Secs. 1 and 2 of Chap. 3; Secs.
1,2, and 3 of Chap. 5; and Appendixes A and B. In the Appendixes a r e
grouped a number of extensions and amplifications that have consid-
erable practical interest. Appendixes D and E in particular contain
special results upon which some work in the later chapters is based.
The many graphs which a r e included were designed for use a s well
a s illustration, and particular pains were taken to ensure their accu-
racy. The same applies to the tables, which were in some cases
recalculated entirely for this book.
Principles of Notation and Terminology. Because of the wide
variety of topics that will be covered in this review, it will not be
possible to set up a system of notation that will be general enough to
cover every case and at the same time describe each special case
naturally. However, certain principles of notation will be followed.
A mole fraction of a particular isotope will always be denoted by
the letter N. In our case of binary mixtures, the two mole fractions
a r e of course N and (1-N). In two-phase systems, all lower-case
letters will refer to one phase, upper-case letters to the other. Thus
N and n will refer to mole fractions in different phases. With every
binary mole fraction N we will associate the corresponding molecu-
lar abundance ratio R, where
Thus corresponding to n there will be r = n/(l -n). AS material flows
through a separating element, its mole fractions a r e changed. The
mole fractions at the various orifices of a separating element will be
distinguished by primes (for example, see Fig. 1.1). Quantities of
material-generally expressed in moles per second-are designated
by L's.
The term "cascade') is used to mean any connected arrangement
of elements. The "elementsv a r e the separative units-individual
centrifuges o r fractionating columns. A "stagey7 of a cascade con-
sists of all units operating in parallel on material of the same mole
fraction. (In the special case of one element per stage, the terms
"element" and "stage" will be used interchangeably.) The feed
material of a cascade, generally of natural abundance ratio, is de-
noted by the subscript zero (No); the amount of feed is F moles/sec.
The "stage number" (counting the bottom stage a s zero) is denoted
by s. The mole fraction of material fed into the sth stage is Ng, and
the amount is Lg.The subscript S indicates the top stage of a recti-
4 THE THEORY O F ISOTOPE SEPARATION

fying cascade. The mole fraction a t the product end of a cascade is

Np ;the amount of product is P. The mole fraction of the waste i s Nw;
the amount of waste is W.
Other general terminology and notation will be covered in the sub-
sequent introductory discussion of cascades. Equations, figures, and
tables a r e numbered decimally (viz., Eq. 6.4); the f i r s t digit indicates
the chapter, the number after the decimal the order in the chapter.
Chapter 1

IDEAL CASCADES

Consider a separating element that divides an entering stream of

L moles/sec of a bicomponent mixture, with the mole fraction of one
component designated a s N, into two s t r e a m s of 0L moles and mole
fraction N', and (1 - Q)L moles and mole fraction N", respectively
(Fig. 1.1). Of course 0 < 0 < 1, where 0 is known as the "cut." Fur-
ther suppose that the action of the unit is such that

The t e r m a is called the "simple-process factor" of the separation.'

F r o m the conservation of matter

Now suppose that a cascade is built of such elements, the size of

each stage being graduated s o that the flow entering the sth stage is
L c . F o r purposes of this discussion it is irrelevant whether each
stage is built of a single element, whose size differs from stage to
stage, o r of identical small elements in parallel, whose number v a r -
ies. The elements a r e connected a s shown in Fig. 1.2, s o that the en-
riched fraction from each element is fed into the stage next above and
the other fraction is fed into the stage immediately below it. At the
top of the cascade, material is withdrawn continuously a t a rate P
moles/sec. It should be observed that there is considerable latitude

This t e r m i s usually reserved f o r t r u e equilibrium ratios of abundance ratios (cf.

Eq. 6.7 and Eq. 8 of Appendix D) which, however, have no universal relationship to the
a of Eq. 1.1. If necessary to distinguish, a of Eq. 1.1 i s referred to a s the "effective"
6 THE THEORY OF ISOTOPE SEPARATION IDEAL CASCADES 7

about the choice of the flows, Ls. For the instant they a r e completely
arbitrary.
The problem now at hand is one of choosing flows Ls and cuts 13~to
get the most efficient cascade. Clearly, the remixing of materials of
Equation 1.11 gives the flow required in the sth stage of a cascade
different mole fractions must be avoided. That is, i t is imperative
producing P moles/sec of material with mole fraction Np, and Eq. 1.6
that
gives the required cut to ensure that no mixing takes place in the
R&-1= Rs = R;+l (1.3)

From Eqs. 1.1 and 1.3, Rs = RL-I = a R s - i , whence ENRICHED OUTPUT

e L,N1

follows immediately by induction. Likewise

R; =
1
-
a Rs STRIPPED OUTPUT
(4-9)L.N''
From Eqs. 1.2 and 1.5,
Fig. 1.1-A simple separating element.

cascade. Such a cascade is called an "ideal" o r "no-mixing" cas-

Considering that section of the cascade above an imaginary line cade. In exactly what sense this is the most efficient cascade will be
drawn between the 8th and (s + 1)th stages (see Fig. 1.2), we must made clear later.
have balance between the amount of materials entering and leaving Equation 1. l l can be written a s
this whole section, i.e.,

by replacing the N's by R's and remembering that Rs = asRo and that
Substituting Eq. 1.7 in Eq. 1.8, R - Ro. In this form the Lie may be summed to get the total flow
p -
in an ideal cascade. Noting that

Equation 1.9, which i s a consequence of the conservation of matter,

holds generally for all cascades, even if mixing takes place. Intro-
ducing Eq. 1.3, the condition of no mixing, in Eq. 1.9, it readily follows that

Using the properties of the separating element itself and eliminating

N; by Eq. 1.1 and eS by Eq. 1.6,
 THE THEORY OF ISOTOPE SEPARATION IDEAL CASCADES 2

Since the size, o r number of elements, of a stage is proportional a

the flow into the stage, the total flow measures the magnitude of the
whole cascade.

1. THE INFINITESIMAL CASE

Because of the special importance of cascades when the simple-

process factor is very nearly 1, the limiting f o r m s of the equations
just found will be written for

a - 1 = e à 1

Equation 1.1 can be expressed a s

I
Nt-N =- R' R R' - R = (a -
1 + R' 1 + R - (1 + R') (1 + R)
+
FROM s 2 s+4
STAGE
which in the limit is the s a m e as

-_ _
-
_ _ __I__- --
(IMAGINARY DIVIDING LINE)
---------- In the no-mixing case Eq. 1.15 becomes

Ns+l - Ns = eNs (1 - Ns ) (1.17)

and Eqs. 1.4, 1.6, and 1.11 a r e replaced by the following:

Rs = e^R, (1.18)

-- -- -- - ENRICHED FRACTION
1
6 - - + ^ - ( 2 ~ ~- 1) (1.19)
STRIPPED FRACTION
2 4
FEED MIXTURE

The total flow in the cascade may be determined from Eq. 1.20 a s
follows:

I
I
i I
Now Eq. 1.17 may be written
Fig. 1.2-A cascade of simple elements.
10 THE THEORY OF ISOTOPE SEPARATION
IDEAL CASCADES 11
ap

Inserting in Eq. 1.21 and dropping the now superfluous subscript s,

very close to $$. Passing over to the limit, which i s clearly legiti-
mate since Ns varies slowly,

Equation 1.27 gives the r a t e of increase of N per stage a s a func-

tion of the flow per stage, here considered a s an arbitrary function.
which could have also been obtained directly from Eq. 1.13.
Now consider a function such a s the total flow in the cascade
Equation 1.19 shows that in the infinitesimal case Os varies very
.
slightly with Ns Even if -as they must be - e and 6 a r e related by
the physical nature of the separating element, e will be constant
throughout the cascade, to t e r m s of the o r d e r of e2, and the preceding
Total flow = L'L = l N p ~ ( ~ )
ds
0
ds
dN

analysis holds. But for the finite case, if an internal relation between
o r the total number of elements, each of which i s capable of proc-
a and 6 exists which differs from Eq. 1.6, a will vary from stage to
essing G moles/sec,
stage, and the equations break down. Practically, this means that the
cascade equations will not be universally applicable unless (a - 1) is
1 N~ ds
sma1l.l Number of elements = L(N) dN
0

2. RELATION BETWEEN IDEAL AND NONIDEAL CASCADES

o r more generally, the integral of any property proportional to the
Perhaps the best way of understanding the r e a l significance of ideal flow per stage
cascades is to consider their relation t o nonideal cascades. To do
this the infinitesimal case must again be treated, this time allowing
mixing.
F r o m Eqs. 1.2 and 1.16,
where Na and Nb a r e arbitrary; <A(N)does not change sign inthe range
Na < N < Nb, and

using N; instead of Ns on the right to the same accuracy a s Eq. 1.16.

From Eqs. 1.7 and 1.8,
Combining Eqs. 1.24 and 1.25,
Now, from Eq. 1.7, if P is much smaller than Ls and if the values
of Ls vary slightly from stage to stage, all the values of OS will be
the ideal flow of Eq. 1.20. The asterisk (*) i s used in this volume to
denote the Laplace transform of any function.
The design problem f o r cascades may then be expressed a s fol-
lows: Of all possible cascades that produce P moles of product with
a mole fraction Np, starting from raw material of mole fraction No,
which cascade makes I a minimum? Or, since each function L(N)
completely determines a cascade, what function L(N) makes I sta-
tionary?
Clearly, I will be a minimum when

' Cf. Appendix D: Rayleigh Distillation.

12 THE THEORY OF ISOTOPE SEPARATION IDEAL CASCADES

that is, when

3. VALUE FUNCTIONS AND SEPARATIVE POWER
L =L*
Equation 1..23, for the total flow in an ideal cascade, is the ratio of
Therefore the ideal cascade minimizes the number of separating a term
elements, and related quantities, in any section of a cascade.
When L = L*, from Eq. 1.27,

which depends only on the amount and mole fraction of material pro-
duced, and a t e r m e2/2, which is characteristic of the separating ele-
is half the maximum concentration gradient (which occurs when ment. This suggests that each t e r m separately has physical signifi-
P/L = 0). Optimum rate of production of a stage i s that rate which cance and also that Eq. 1.23 might be established in a more gen-
reduces the concentration gradient to half its value at no production. e r a l way.
It is well to bear in mind that the concept of an ideal cascade is a s - An attempt should now be made to derive a function U that repre-
sociated with the form of I. For example, to minimize expressions sents the value of a quantity of separated material, i.e., a function
such a s proportional to the number of separating elements required to pro-
duce it. Obviously U = P V(N), where P is the number of moles of ma-
Total number of stages = Js ds = LoNp L
2L - L*
dN
mFm (1.34) terial. Suppose G moles of material is passed through a separating
element of the type shown in Fig. 1.1. There results a net change
requires L = m and not L = L*. In practice it will be necessary to in value
strike a balance between reducing the number of separating elements
and reducing the number of stages. Furthermore an ideal cascade
requires each stage to be of a different size. F o r these reasons ac-
Taking (a- 1) = e -K 1 and expanding V(N1) and V(Nt1) in Taylor's
tual cascades will deviate somewhat from ideal cascades. Neverthe-
s e r i e s about N,
less, the properties of ideal cascades a r e of great practical impor-
tance because of the proposition that follows.
Developing I about L = L *,

The first term on the right is called I*, the value of I for the ideal
[ ( N ~ ' - Nd2v
) ](N)
} + ~ [ ~ G ( ~ ' ~ ~ ) ' +(N"( -
I
~ - ~ )+ G ... (1.38)

case. If IL - L* 1 =s y L * for y c 1, Eq. 1.35 gives, to higher powers The coefficients of V(N) and d v ( ~ ) / d N vanish by the conservation of
of Y, matter. From Eq. 1.2,
I - I* 5 y21* (1.36)
6(Nr -N) = - (1 - 6) (N" -N) (1.39)
Expressed in words: If the flow in a cascade at no point differs from
that in the corresponding ideal cascade by more than 20 per cent, and from Eq. 1.16, N1 - N = N(l - N). Thus Eq. 1.38 becomes
integrals of the type in Eq. 1.30 will not differ from the ideal inte-
grals by more than 4 per cent. It is this stationary property of inte-
grals for the ideal cascade which makes them useful for all cascades.
14 THE THEORY O F ISOTOPE SEPARATION IDEAL CASCADES 15

Now, specifying that this change in value be independent of N, the

- 2P V(Np) 1 - 6
GUp -
following equation must hold: Total flow = -
6U c2 6

should be pointed out that, in o r d e r f o r Eq. 1.44 to hold, there

t be no process taking place in the cascade which changes U ex-
Then Eq. 1.40 becomes the separation process; that i s , t h e r e must be no mixing.l Equa-
1.46 therefore gives the total flow in an ideal cascade, a s found
viously f o r the special case2 6 = $$.
A different choice of the integration constants is of course permis-
ble. Taking V(0.50) = dv(0.50)/dN = 0, which means taking the equi-
olar mixture instead of the natural abundance a s the fiducial point,
The t e r m 6 U i s called the "separative power" of the element.'
ves the simple formula
The differential expression of Eq. 1.41 defines V(N) except f o r two
integration constants. Following Dirac, these may be chosen by r e -
quiring that V(No) = ~ v ( N ~ ) / =~ 0,Nwhere No i s the natural abundance.
The f i r s t condition is obvious; the second makes V(N) a minimum at
N = No and s o e n s u r e s a positive value f o r V over the entire range of the value, which has the obvious advantages of depending on only
concentrations. mole fraction and of being the same for all isotopes. However,
Thus it is found that equation like Eq. 1.46 takes the form

G
Total flow = - (Up + U& - U;)
6U

V will be called the "value function." Obviously it i s the value p e r and since U' i s always used in such differences, with two relations
mole. between the p a r a m e t e r s (Eqs. 1.61 and 1.62), i t s numerical value i s
Since one element produces a change in value 6U, r e g a r d l e s s of the l e s s illuminating.
mole fraction of the material on which it operates, the number of ele- F r o m the standpoint of calculations, if there is one particular iso-
ments required to produce a change in value AU is tope-separation problem, with No known, values of V a s defined by
Eq. 1.43 a r e most useful. These a r e found in Fig. 1.3 f o r the m a -
Total number of elements = A U / ~ U , (1.44) aium problem, No = 1/140. F o r a survey of many different separation
and problems, values of V a r e required. They a r e found in Table 1.1 and
Fig. 1.4. V will be called the "elementary value function," to dis-
AU 2AU 1 - 0 tinguish it from V, which is the value function relative to natural
Total flow = G X (number of units) = G = 7 7 (1.45) abundance. U' will be called the "elementary value, " and U will be
defined a s the "value of a quantity of material."
F o r the cascade previously considered The value functions and the separative power have been derived f o r
infinitesimal p r o c e s s e s . They may also be extended to cover the case
of large values of a.

where the subscripts r e f e r to the product, waste, and feed points. 'Mixing reverses the process in Eq. 1.37 and hence decreases U.
With the convention chosen, Uo = Uw = 0 and Eq. 1.45 is merely 'Since 8 has not been restricted, in general a more intricate cascade than the one
shown in Fig. 1.2 will be required. F o r example, if 6 = '/5, the enriched fraction from

- - as -
stage s will have to be fed into stage s + 3, while the depleted fraction is fed into stage
'Note that e must be a function of 6. a s otherwise bU 6 1. s-2.
16 THE THEORY O F ISOTOPE SEPARATION IDEAL CASCADES

46 0
I I I I I I I I
Table 1.1-The Elementary Value Function

F o r N > 0.50 use V'(N) = V'(1

N
V = (2N - 1) In -
1 -N

0 0.2 0.4 0.6 0.8 1 .O

M O L E FRACTION

Fig. 1.3-Functions associated with infinitesimal cascade. No= 0007143.

Consider a symmetric process1 where (cf. Eqs. 1.5 and 1.6)

Then corresponding to Eq. 1.37

= constant

Now introduce s by the equation R = Roos and s e t

(1 + R) V(R) = g(R) = g(RoCt3) = F(s)

'The problem i s also soluble for asymmetric processes, where quite complicated
functions a r e found.
 IDEAL CASCADES
THE THEORY OF ISOTOPE SEPARATION

a + 1 2N-1
V(R) = C - -in-+
a - 1 lna
[R
Ro
a - N o ( a + l ) (N-No)
a - 1 No(l - NO)] (1.55)

Now take the constant C = 6u/G s o that

Thus

R
RO
lna a-(a+l)No
(2N - 1) In - + -
a-1
(N-No)
N ~ ( ~ - N ~ ) I (1.57)

Equations 1.56 and 1.57 a r e closely analogous to Eqs. 1.42 and 1.43.
The variation of U f fwith a exhibited in the second t e r m is caused by
the varying concentration of by-product waste material and has no
theoretical significance. The elementary value for l a r g e a is of
course still U ' .
The separative power v a r i e s a s ( a - 1)' f o r small (a - I ) , but f o r
0 0.10 0.20 0.30 0.40 0.50 large a increases only a s In. Since the number of elements in a c a s -
MOLE FRACTION cade varies inversely a s the separative power of an element, it i s
important to design and operate the elements s o that their separative
Fig. 1.4-The elementary value function. For values of N > 0.50,use v'(N) = V f ( l- N); power (Eq. 1.42 o r Eq. 1.56) is a s l a r g e a s possible. This problem
for N > 0.46, V = 2(2N - I)'.
will be considered in detail in subsequent discussions on particular
separation schemes.
With these definitions Eq. 1.50 becomes It i s also possible to calculate the separative power for some more
general types of separating elements. If it i s assumed that instead of
two s t r e a m s leaving the apparatus of Fig. 1 . I , t h e r e a r e many, each
of magnitude L ~ and , furthermore
The general solution of Eq. 1.52 is

where A and B a r e constants, which gives

a + 1 2N-1 R AN
a-1 -
V(R) = C - l n a In - + -+ B ( l -N) (1.54)
RO Ro

Setting V(Ro) = 0, and establishing the convention that the waste m a -

t e r i a l from the feed stage, of abundance ratio Ro/a, has zero value
also, the following relation holds:
20 THE THEORY O F ISOTOPE SEPARATION IDEAL CASCADES 21

The f i r s t two t e r m s onthe right vanish by the conservation of matter. ally both T - NP and N' - N a r e proportional to N(l - N); if SO
Hence
 ( T - NP)(N' - N)
6u = ~ 2 (-
1~ ) 2 (1.60)

quantity T , i.e., the amount of desired isotope moved forward, is

of which Eq. 1.42 is a special case.
ed the "desired material transport." The factor T - NP is called
For infinitesimal separating elements of the countercurrent type,
"net transport." Equation 1.60 is an exceedingly important for-
6U may be evaluated a s follows: Consider two large reservoirs (see
a for the theory of countercurrent processes.
Fig. 1.5) containing, respectively, M moles of material in which the
he separative powers for two other important special types of
ents a r e fully discussed in Appendix D (Rayleigh Distillation) and
ndix E (Properties of Concurrent Two-phase Elements).

4. CASCADES WITH STRIPPING SECTIONS

In addition to a rectifying section, it will usually be desirable to

1 SEPARATING
ELEMENT 1 e a stripping section in the cascade.l The function of the stripping
tion is to economize raw material by recovering the desired iso-
e from the reject stream of the rectifier. Figure 1.6 shows the
ions between rectifier and stripper.
r any cascade, ideal o r otherwise, provided there is no loss of
terial anywhere in the system, the following relations hold:

Fig. 1.5-Separative work of a countercurrent element.

mole fraction of desired material is N, and M' moles of slightly dif-
ferent mole fraction N' > N. The separating element puts P moles of
mixed isotopes containing T moles of (pure) desired isotope into the
upper reservoir, withdrawing the same amounts from the lower r e s -
ervoir. The upper reservoir increases in mole fraction by an amount
r e F, P , and W refer to the amount of feed, product, and waste,
pectively, and No, Np, and Nw a r e the corresponding mole frac-
s.
aking the mole fractions a s given and eliminating f i r s t F and then
etween Eqs. 1.61 and 1.62, it is found that

W = P N~ - No (1.63)
No - Nw
and the lower decreases by the amount (T - PN)/(M - P). Then

he terms "rectifier" and "stripper" are taken from fractionating-tower usage.

tower for producing alcohol from wine, for example, the section above the feed
t rectifies the alcohol, i.e., concentrates it, while the lower section strips the
aste water, i.e., recovers the last traces of alcohol from the reject.
 THE THEORY OF ISOTOPE SEPARATION IDEAL CASCADES 23

resemble closely those for an ideal rectifier. A moment's con-

deration, referring again to Fig. 1.2, shows that Eqs. 1.4, 1.9, and
0 .ll also hold for the stripping stages of a cascade, with -W and Nw
which relate the size of the waste stream and the feed stream to the replacing P and Np, s running from -1 to -B instead of from 0 to S,
rate of production. and Rw = ~ ~ / a ~Equation
+ l . 1.11 f o r the flow in the sth stage of a
rectifier becomes

a + 1 W(N, -Nw)
Ls =- (1.65)
a - 1 Ns(l-Ns)

Equation 1.13 becomes

Total flow (stripper) =

s=-1
a -1
1
(aB+s*l - 1) + -(aB+l
Ro
- a"' )
I
a+1 1-2Nw R
a-1
=-W[ In a In
Rw

1 = WASTE
W, N,.
- a -N(,(a + 1) No-Nw
(a-1) No(l-No) I
a + 1
(1.66)
( a-1)lna u'

where U$ is the value defined by Eq. 1.57. This expression vanishes

for Rw = ~ ~ / in a accordance
, with our convention of counting stage 0
a s part of the rectifier.
FEED
The total flow in a cascade composed of both rectifier and stripper,
combining Eqs. 1.13, 1.64, and 1.66, takes the simple forms

Total flow (rectifier and stripper) = a+1 [P(2Np-1)ln- RP

( a - 1) In a Ro

+ W(2Nw - 1) ln-
Fig. 1.6-Above, connections between rectifier and stripper. Below, detail of con-
nections. a+1
-
( a - 1) In a (up + uw
Since a stripping section is essentially no different from a rectify-
ing section, except that i t produces material at a different rate and -- a +1
(UP + U&- Uo) (1.67)
concentration, the equations for an ideal stripper should be expected (a - 1) In a
24 THE THEORY O F ISOTOPE SEPARATION IDEAL CASCADES

The t e r m s Up + Uw and Up + U& - Ug a r e of course AU = AU" and f r o m that of the product to that of the feed. Before the plant can
AU', respectively. Note that in Eq. 1.67 the distinction between U" oduce concentrated material a t the top i t must build up these in-
and U has disappeared, since the t e r m s that differ because of differ- ntories of partly enriched material. The time spent in this proc-
ent conventions have canceled out. is known as the "equilibrium time" o r "relaxation time" of the
F o r the infinitesimal case, in place of Eq. 1.65,
will be shown l a t e r the operation of a cascade f r o m starting-up
might be a s follows: At z e r o time the mole fraction a t the top
the cascade i s No; hence operations begin with no withdrawal. The
centrations increase gradually over the entire cascade, the top of
In place of Eq. 1.66 cascade leading. The product end reaches design mole fraction
Nw2 W(N-Nw) dN 2
st, while the r e s t of the cascade is still below design. Withdrawal
Total flow (stripper) =
f. 7 NO - N ) N(I-N)=F~W
a slow r a t e is then begun, maintaining the top mole fraction a t Np
d gradually increasing to design production a s mole fractions in the
and in place of Eq. 1.67 t of the cascade build up to their steady-state values (Fig. 1.7).
eneral definition of the equilibrium time that has meaning f o r this
2 e of operation o r any other is the number of days production that
Total flow (rectifier and s t r i p p e r ) =
â
(up+ Uw) lost between starting-up time and steady -state production.
2 A useful approximate expression for the equilibrium time of a n
= - (up +
e2
uy, - UO) eal rectifier by simple considerations will now be developed.
The total amount of material in p r o c e s s in a section of a cascade
called the "holdup" of the section.' In most cascades the amount
holdup in a stage i s proportional to the flow in the stage. This i s
rtainly t r u e if the cascade i s built up of identical small elements,
that successive stages consist of different numbers of elements in
rallel. It also holds when the stages a r e similar single units scaled
size. The holdup p e r unit flow will be denoted by h, which i s a l s o
e average p r o c e s s time p e r stage, expressed in seconds.
F o r completeness the formulas f o r the number of stages may be The "net desired material holdup" (N.D.M.H.) is the amount of
added. s i r e d isotope in a going plant over and above that which would be
contained in a like quantity of material a t normal abundance. It is,
Number of rectifying stages = S + 1 = In RP/RO the infinitesimal case,
In cr

Number of stripping stages = B = In ~ o h _w 1 N.D.M.H. = y " " ' h ~ ,(N, -No) d s

In a

Number of stages (rectifier and stripper) = lnRp/Rw - 1 (1.73)

-[ ( N P - 2 N p N o + N o ) l n ~ - 2 ( ~-No)]
p (1.74)
In a RO
F o r the infinitesimal case, In a i s merely replaced by e.
he equilibrium t i m e in any c a s e is equal to the N.D.M.H. divided by
a suitable average transport of desired isotope over the initial period
5. EQUILIBRIUM TIME of operation. Without f o r the moment specifying too closely the oper-
Obviously some cascades f o r separating isotopes a r e very l a r g e
and contain considerable material in p r o c e s s a t concentrations rang- Nomenclature again borrowed from fractionating-tower terminology.
 THE THEORY O F ISOTOPE SEPARATION IDEAL CASCADES 27

may be applied to other than ideal cascades. The usefulness of the

expression will depend on the appropriateness of the choice of the
average net transport, which will not in general be P(Np - No).
The N.D.M.H. i s one of the functions that i s a minimum for ideal
cascades, a s is the equilibrium time function E(Np) derived from it.
E(Np) i s independent of P .

6. FUNDAMENTAL EQUATIONS O F ISOTOPE SEPARATION

The partial differential equations describing the time-dependent ef -

fects in any infinitesimal cascade a r e easily derived by a slight ex-
TIME tension of the previous analysis.' Ls and 6s a r e taken a s known func-
tions that may vary with time. The mole fraction Ns i s also a function
of time: No = Ns(t).
Referring again to Fig. 1.2, Eq. 1.7 takes the more general form

where Ts(t) i s the amount of (mixed) material transported into the

cascade above the sth stage. Explicitly, if t h e r e a r e no leaks o r
losses,

H i is the holdup in the jth stage and is a l s o a given function of time.

Fig. 1.7ÑProductio a s a function of time.
ating schedule, this average transport may be taken a s the net t r a n s -
port of the rectifier in the steady state, that i s , P(Np - No). This gives
Ts (t) i s defined in t e r m s of known functions (by Eq. 1.77 o r 1.78) and
therefore is known.
Likewise Eq. 1.8, the equation of conservation of desired material,
holds in the form

RP
à [(N,, - 2NpN0 + No) In -- 2 ( ~ p- No)]
2 Ro
Equilibrium time = (Np - No) where rS(t) is the amount of desired material transported into the
section of the cascade above the sth stage. Unlike T, T is unknown.
Explicitly

It should be remarked that the formula for the equilibrium time in

the f o r m

N.D.M.H. ' F o r finite a it i s difficult toformulate the problem without specifying the separation
Equilibrium time =
average net transport method.
 THE THEORY O F ISOTOPE SEPARATION IDEAL CASCADES 29
28

Because special cases of Eqs. 1.81 and 1.82 describe the properties
In using N,'(t) on the right the difference between the mean concentra-
of apparatus in widely differing separation methods and because all
tion of the stage and Ni has been tacitly neglected. Combining Eqs.
the theory s o f a r developed for infinitesimal cascades can be deduced
1.77 and 1.79
from them by purely formal operations, Eqs. 1.81 and 1.82 a r e called
the "fundamental equations of isotope separation. "
If L is not a function of t , and H = hL, then T = P, and Eqs. 1.81 and
1.82 combine to give
Assuming Eq. 1.24 t o hold unchanged (what is involved is neglecting
s
the holdup of one stage compared to 2 Hj) and introducing Eq. 1.24
in the last equation,
Equation 1.83, with the initial condition N(s,O) = No, describes the ap-
proach to equilibrium of a cascade. F o r an ideal cascade, L is the
extremely complicated function defined by Eqs. 1.18 and 1.20. Equa-

On passing over t o the limit of continuous functions, OsLs

-
K ( t ) N(s,t), etc., this becomes
- %L(s,t),
tion 1.83 will not be solved f o r this case, since the utility of the solu-
tion i s slight. In Chap. 3 it will be solved for the practical (and con-
siderably easier) case of a square cascade, taking L a s a constant.
Should L (and also H and e ) be a functionof t, fluctuating about some
mean value, then Eqs. 1.81 and 1.82 describe the effect of fluctuations
on separation, which is covered in Chap. 5, on the control problem.

which is the required generalization of Eq. 1.27. F r o m Eqs. 1.78 and
1.80 i t is found1 that
REFERENCES
Cohen, K., Report A-60, October 1941;Report A-96, Jan. 5, 1942.
Cohen, K.,Columbia Ser. No. 4L-X28, Dec. 22, 1942 (letter to J. 8.Dunning).
Cohen, K., and I. Kaplan, Report A-98, Jan. 14, 1942.
Dirac, P. A. M., British MS, 1941.
Fuchs, K., British MS 29, Dec. 18, 1941.
Fuchs, K., and R. Peierls, British MS 12A, February 1942; British MS 47, June 26,
1942.
Kurti, N., and F. Simon, British MS, 1940.
Peierls, R., British MS 12, summer 1940; British MS 13, September 1940.

Differentiating Eq. 1.81 with respect t o s and using Eq. 1.82, 9 ~ / 9 s

may be eliminated and a second-order partial differential equation
f o r N will result. Thus Eqs. 1.81 and 1.82, with appropriate boundary
conditions, completely determine N.

' T h e integral form of Eq. 1.78, for example, i s

a
~ ( s , t =) P +f H(j,t) dj

and differentiating with respect t o s gives Eq. 1.82.

 SQUARE CASCADES 31

lies to both square cascades and countercurrent units. Indeed, it

o preserve this generality that the notation of Eq. 2.1 has been

1. STEADY STATE OF A SQUARE CASCADE (OR COUNTERCURRENT

COLUMN)

Chapter 2 In the steady s t a t e t h e f i r s t integral of Eq. 2.1 i s

SQUARE CASCADES PNp =

dN
PN + c1N(l - N) - c -
ds

Ideal cascades a r e characterized by a smoothly varying flow func- ch separates to give a second integral,
tion which i s next to impossible to realize in practice. Fortunately,
a s the discussion in Chap. 1 of the relation between ideal and nonideal
cascades has shown,it is by no means necessary to adhere strictly to
ls iNs
ds =
- + c1N2
c, dN
(c1 + P)N - PNp
the shape of the ideal cascade.
The outlines of this cascade roughly follow those of an ideal cas-
cade, and the properties of the blocked-off cascade closely resemble
those of the ideal cascade. A cascade thus obtained from an ideal cas -
cade is called a "squared-off cascade." Each section of a squared-
S =
1
-
6
tanh-I
A($)
(Ng - No) A(+)
[(Ns + No) (1 + $) - 2NsNo - 2Np+ ] (2.5)
off cascade i s itself a square cascade.
F o r a square cascade Eq. 1.83 takes the form

' a~ a2N a
^= P/C,

c 6 Ãa‘tà ‘ = c 5 à ‘ [PN + c ~ N -( N)]

~
as2 as (2.1) A($) 5 [I + 2$(1 - 2 ~+ ~ ) (2.6)
where cl, P, c5, and c6 a r e constants. Specifically
6 = c1/2c5

c, = hL = holdup p e r stage The use of <= in Eq. 2.5 agrees with Eq. 2.2; f o r countercurrent col-
c5 = L/2 mnns in which the constants c, and c5 a r e fundamental, l i s under-
c1 = e L
stood to be defined by 6 = c1/2c5. The dimensionless quantity is +
thought of as a normalized rate of production.
Equation 2.5 gives the distribution in the cascade, for given Np and
Proof will be given later that precisely the same equation governs P,a s a function of s. At the top of the cascade, Ns = No and s = S =
the behavior of a countercurrent centrifuge, a fractionating column, a number of stages (or length of column, depending on the definition
thermal-diffusion column, and most other countercurrent devices. In nd units of f ). Equation 2.5 becomes
each of these cases c, and c5 have different but analogous meanings,
and s is replaced by z , the coordinate along the length of the column.
A countercurrent unit i s therefore equivalent to a square cascade of
simple elements of the type of Fig. 1.1, and a cascade of counter-
current units i s like .a squared-off cascade. The ensuing discussion
 THE THEORY OF ISOTOPE SEPARATION SQUARE CASCADES 33

g=- 1
e A(@)tanh-l f (NP - No) A(@)
(Np - 2NpNo - No) - (Np - No)@
N~ - (1 + @)e2*^^
--
N~ 1 + i/fe2N1++) (2.14)

If values a r e assigned to Np and @, Eq. 2.7 i s an explicit equatio

which determines the necessary number of stages (or length of
countercurrent column). If S is given, Eq. 2.7 i s an implicit relatio
between Np and @.
When P = 0, then 0 = 0, A ( @ )= 1, and Eq. 2.5 becomes
-

1 NS - NO 1
s = - tanhW1 =- In RsFt,,
e Ns -2NsNo + No 2e -
or ezfS =FRACTIONATION
AT NO PRODUCTION
Rs = ~~e~~~ $ = YC,=
NORMALIZED -

Equation 2.9 is in no way peculiar to square cascades, since at -

P = 0 a l l cascades, regardless of their shape, have the s a m e distri-
bution of concentrations, because Eq. 1.27 with P = 0 i s
-

dN
-=
ds
2eN(l - N) --------------

--
dR - 26R
ds 2 4 6 10' 2 4 6 10" 2 4 6 1 2 4 6 10
N O R M A L I Z E D PRODUCTION

whence Eq. 2.9 follows.

Fig. 2.1-The effect of production on fractionation.
However, although the distribution in an ideal cascade with produc-
tion a t the designed rate i s again exponential (with half the slope of
Eq. 2.9), Referring to Eq. 2.3, since the t e r m s PN + c,N(l - N) increase with
s o also must dN/ds. The smallest value of dN/ds is therefore a t
Rs = RoeEs foot of the cascade where N i s smallest, N = No. The maximum

square cascades have the much more complicated distribution of

Eq. 2.5.
No(l - No)
In the special c a s e when both No<l and Np< 1 , . Eq. 2.7 takes the @ =
simpler f o r m s NP - No
F o r this value of @ Eq. 2.7 becomes

s = 2c(l 1+ @ ) No
l n N 0 -@(Np - No) g=- tanh'll=-
6 A(@)
34 THE THEORY O F ISOTOPE SEPARATION SQUARE CASCADES 35

Equation 2.7, for a given value of Np , may be satisfied by an infi- whose integral is (cf. Eq. 2.5)
nite range of pairs of values of S and ip, ranging from S = m, ip =
N ( 1 - No)/ (Np - No), to S = 1/26 In R ~ / % , $!J = 0. The optimum length
of cascade and rate of production a r e that pair of values (S,$!J)which
makes the average transport per stage P/S a maximum o r , what is
the same thing, makes s/c1ip a minimum. From Eq. 2.7 where
4s = pAsc1 (2.20)
A(&) = [l + Us (1 - 2Np) + @

The right-hand side i s now a function of ip alone. Since differen-
tiating Eq. 2.17 gives a transcendental equation, the minimum i s best
found by direct computation.
The length of columns in the entire sth stage which changes mate-
r i a l of mole fraction Ns to material of mole fraction Ns+l, while the
cascade a s a whole produces material of mole fraction Np a t the r a t e
P, is

2. SQUARED-OFF CASCADES (CASCADES OF

COUNTERCURRENT ELEMENTS)

In the problem of squaring-off an ideal cascade which was previ- vs Zs is a function of <*sand its minimum (e.g., at A, = &) gives the
ously considered, i t was possible to use square sections of any de- minimum length of column for stage s. From <As i t is possible to ob-
s i r e d length1 and breadth, i.e., Z and c5 arbitrary. The usual cascade tain the corresponding optimum value of us,
problem is to make a cascade from elements that a r e identical coun-
tercurrent columns o r small-capacity square cascades. Each element
has the same constants, cl and c5, and the same length, Z ?
Subsequent examples will show that cascades that a r e very close to
ideal may be expected, provided the square elements a r e sufficiently
small2 the number of columns in parallel, and also Es, the appropriate length
Instead of solving the problem just s e t up, i.e., the best Ns with Zs of each individual column,
given, i t is much more convenient to proceed indirectly. Suppose that
the steps of the cascade, i.e., the Ns, a r e given, and solve for the best
lengths Zs to fit this cascade. Then it will be seen how the solution of
the second problem solves the desired one.
The number of columns in parallel in the sth stage is denoted by
vs. The mole fraction fed into the sth stage is Ns , as before. In place
of Eq. 2.3, for each of v columns, 3. RAPID APPROXIMATE SOLUTION FOR SQUARED-OFF CASCADES
The solution, Eqs. 2.21 to 2.23, takes a simple form for moderate
values of Rs+i/Rs. It may be presumed, subject to later check, that
the argument of the arctanh is small. Since tanh-I x = x + Ysx3 + , ...
'Use is made of Z instead of S for the length of the square sub-cascades, retaining
neglecting t e r m s of third and higher order,
s a s the stage number of the grand cas.cade.
"'Sufficiently small" is here used to mean the transport per element a cascade
net transport, and length of stage a ideal cascade length [qNo(l- No)Ã P(N - No)
and z a 1/e ~n RJRJ.
36 THE THEORY OF ISOTOPE SEPARATION SQUARE CASCADES 37

which i s a minimum for
With this value of As the argument of the arctanh becomes
A column of length ZS operating with no production will, according
to Eq. 2.11, produce an enrichment of e2&. Consequently the theo-
r e m has been proved that a column o r square cascade in the body of
a cascade should be operated a t the square rootof i t s maximum f r a c -
tionation. The foregoing result is really another manifestation of the
s i m i l a r relation found for ideal cascades (Chap. 1, Sec. 2).
Although the approximate expressions (Eqs. 2.25 and 2.26) a r e
fairly inaccurate unless (Rs+l - Rs ) / R ~ is small, experience has
shown that the final relation

F o r any stage of a cascade except the top, a s Rs+l Rs, As - -

-
Ns (1 - Ns ) / 2 ( ~ p- Ns ) = finite, and the argument 0. But f o r the top-
most stage the argument is not s m a l l even for Rp very nearly equal t o
Re . F o r , when NS+, = Np, and the square-root theorem, which follows f r o m it, both hold for
-
~ s + l / ~ s2.
The recommended rapid procedure f o r finding optimum lengths and
numbers of columns when the Ns a r e given i s therefore a s follows:
and, a s Rp - -
R e , A (&.) &., and the argument - 1. Thus, except for The value of Zs is found from Eq. 2.30. Using this value in Eq. 2.23,
(ps i s found. (This gives a better value than Eq. 2.25.) Then Vs may
the last stage of a cascade, the approximations in Eqs. 2.24 and 2.25
will be valid f o r sufficiently s m a l l values of R ~ + ~ / R ~ . be found from Eq. 2.22. It should be remarked that even - though Zs
Using the value Eq. 2.25 for As, from Eq. 2.24 i s obtained from Eq. 2.30 will be slightly i n e r r o r , and likewise V g , vsZs will be
indistinguishable from the t r u e minimum because of the stationary
properties of a minimum.
The solution to the original problem of finding optimum Ns and us
when the Z a r e given follows immediately. The Ns a r e found from
Eq. 2.30, and the r e s t of the procedure i s the same a s in the pre-
F r o m Eqs. 2.22 and 2.25, ceding paragraph.
The final proof that the procedure outlined gives a satisfactory an-
swer is the comparison of the resulting cascade with an ideal (infini-
-
tesimal-step) cascade. As NS+l Ns, from Eq. 2.26,

and from Eqs. 2.26 and 2.27,

and

Solving Eq. 2.28 f o r R ~ + ~ /leads

R ~ to

where V i s the value function of Eq. 1.43.

Equation 2.32 shows that, for a square cascade o r column defined by
Eq. 2.31, the separative power p e r unit length o r per stage i s
 THE THEORY OF ISOTOPE SEPARATION
In the case of the square cascade whose elements a r e of the type
shown in Fig. 1.1, where c, = L/2 and c, = EL,Eq. 2.33 reduces to
which coincides withthe previous expressionfor the separative power
of such elements, Eq. 1.42.
4. SQUARING-OFF THE TOP O F A CASCADE
In the problem and example of squaring-off treated in the last two
sections, a scheme was adopted which would make the squaring-off
uniform over the entire cascade. If the cascade is built up of small
standard elements, this represents no difficulty. F o r example, i f the
whole cascade is assumed to consist of 10,000 centrifuges, then the
top stage i s made of 9 centrifuges, the twelfth of 28, etc.
Consider a cascade that is ideal up to a mole fraction NA and then
square up to the top. The square section is taken to have the same
flow a s the ideal section a t the point of intersection. It is necessary
to compare the properties of an ideal cascadefrom NA to Np with that
of a square cascade from NA to Np in which the flow is
The equilibrium time increases about twenty-five times a s rapidly
a s the total flow and is usually the factor that decides the number of
stages that can be squared-off. The squared-off cascade is shorter
than the part of the ideal cascade which it replaces.
REFERENCES
Cohen, K., Report A-530, Feb. 4, 1943.
Cohen, K., Packed Fractionating Columns and the Concentration of Isotopes, J. Chem.
Phys., 8: 588 (1940).
Cohen, K., Columbia Ser. No. 4L-X28, Dec. 22, 1942 (letter to J. R. Dunning).
Cohen, K., and I. Kaplan, Report A-101, Jan. 28, 1942.
Chapter 3
EQUILIBRIUM TIME OF A SQUARE CASCADE
The equilibrium-time phenomenon s e e m s a t f i r s t glance to be
merely an obstacle to be overcome in successful cascade operation.
Indeed, it causes serious production losses in large plants by causing
delays when operation i s changed, a s well a s in the initial incubation
period. Even for small experimental cascades, it prevents rapid
attainment of steady-state conditions and thus slows up the accumu-
lation of data and necessitates long continuous operation for each
experiment.
However, it is possible to use the r a t e of approach to equilibrium
a s a powerful tool f o r determining the constants c, and c5 of the fun-
damental equation (Eq. 2.1), and consequently to turn the disadvantage
into an asset. The theory will be expounded below primarily with this
end in view. Solutions will be found for those c a s e s which occur ex-
perimentally, and the results will be developed with the aid of tables
and graphs for immediate application.
The fundamental partial differential equation (Eq. 2.1)
is quasi-linear and parabolic and can be solved by an iteration meth-
od. When N <s: 1, Eq. 2.1 becomes linear, reducing to
Equation 2.1 also becomes linear even for large mole fractions
provided the change in mole fractions is small. The substitution
N = No + x gives the linear equation
 THE THEORY O F ISOTOPE SEPARATION EQUILIBRIUM TIME O F A SQUARE CASCADE 41

column (by convention the "top"). The other end i s maintained a t a

constant composition by supplying fresh feed continuously, which is
equivalent to connecting the bottom of the columns to an infinite r e s -
ervoir. There i s usually a certain amount of external holdup a t the
whose solution closely resembles that of Eq. 3.1.
product end in connecting piping, surge tanks, etc. The holdup in this
top reservoir i s denoted H.
PRODUCTION
In Type I1 operation (Fig. 3.1) no material i s withdrawn. Both ends
t feed into and out of finite reservoirs. The holdup of the bottom r e s -

RESERVOIR 4 HOLDUP 1 1 HOLDUP p- RESERVOIR

TOP
ervoir (at s = 0) i s called H'. Type I operation with P = 0 coincides
with Type I1 operation with H' = w .
A still more general type of operation (H and H ' finite; P # O), of
which both Types I and I1 a r e special cases, is of no physical signifi-
cance, since the system is gradually emptied of material and there i s
no steady state.
Before solving Eq. 3.1 it is convenient to divide through by c5,
giving

where

The definitions of 0 and c. a r e repeated f o r convenience. F o r square

+
CONSTANT- cascades of elements of the type illustrated in Fig. 1.1, A. = 2h by
COMPOSITION
FEED- I I
Eq. 2.2. If s i s dimensionless, A. i s expressed in units of time. In
HOLDUP -BOTTOM Type I1 operation, = 0 in Eq. 3.2.
J! , I

WASTE H' RESERVOIR The boundary conditions for Type I operation are1

TYPE I TYPE It
Fig. 3.1-Diagrammatic representation of the two main types of experiment on
columns.

1. (For N 1, s e e
Consider f i r s t the solution of Eq. 3.1, with N ç
Appendix B.)
There a r e two main types of experiments: In Type I operation
(Fig. 3.1) material i s withdrawn continuously from one end of the
-
'The difference in concentration between the reservoirs and the ends of the column
may be neglected if K/X? << 1, which i s almost always the case.
 THE THEORY OF ISOTOPE SEPARATION EQUILIBRIUM TIME O F A SQUARE CASCADE

The boundary conditions under Type I1 operation a r e

aN aN
Ats=O, --2eN=K -
 as at
aN F o r , integrating by parts,
Ats=S, -aasN- 2.5N = - K - a t

At t = 0, N = No

In Eqs. 3.4 and 3.5 the factors introduced were The remaining boundary conditions likewise transform. F o r Type I
operation Eq. 3.4 becomes
K =-H
c5
H'
K ' =-
c5

By f a r the most expeditious method of solving Eq. 3.2 under these F o r Type I1 operation Eq. 3.5 becomes
boundary conditions i s the use of the Laplace transform. Briefly
stated, the method i s a s follows: By the transformation

the partial differential equation (Eq. 3.2) i s transformed into an ordi- The general solution of Eq. 3.9 is
nary differential equation for G with p a s a parameter. The solution
of the ordinary differential equation gives G(s,p) explicitly. The in-
v e r s e operation to Eq. 3.7,

where Ci and C2 a r e constants and ml and m2 a r e the solutions of

then gives the required solution. In Eq. 3.8, B r , i s the f i r s t Bromwich

contour in the complex p plane, a line from -im to +i-, deformed if
The constants Cland C2a r e determined by substituting in Eq. 3.10
necessary to pass to the right of the poles of G(s,p).
f o r Type I operation o r Eq. 3.11 for Type I1 operation. Omitting these
Equation 3.2 will now be solved according to the above outline.
details, the results a r e a s follows.
Multiplying Eq. 3.2 by e-Pt dt and integrating from 0 to m gives the
Type I Operation:
equation

with
Here again b,r = 0 in Type I1 operation. In establishing Eq. 3.9 the
boundary condition a t t = 0 has been used to evaluate
44 THE THEORY OF ISOTOPE SEPARATION EQUILIBRIUM TIME OF A SQUARE CASCADE

Type I1 Operation: where the p j a r e the roots of

No 2eN0ecs sin y s - f(S,p) g(s,p) r

G(s,p) = - + (3.16) tan yS =
P PY PD(S,P) ~ ( - A1 ) - KP

where the functions f , g, and D a r e defined by Type I1 Operation:

P
1
( e - pK) sin yS - cos yS

pK1 + e
I
g(s,p) = ecs (cos y s + sin y s) (3.17)
Y
where the p j a r e the roots of D(S,p) = 0, o r
(K' - K) 6 -p ~ - A~ '
sin y~ - (K + K') cos yS
D(S,P) s (K + K' )Y
Y
tan yS =
E ( K ' - K) - ~ K -KA
and y is the same a s in Eq. 3.15, but with $ = 0,
and

The inversion of either expression for G(s,p) can be carried out in

two ways, giving two different infinite s e r i e s for the same function -
AS cos y j s
N(s,t). The difference between the two methods is that one s e r i e s [(K' - K)E - P ~ K K '- A ]
converges rapidly for small values of t, the other for large values.
7
Although either expansion can be used over the entire range of t by
carrying enough t e r m s , it i s more convenient to use both expressions,
each in i t s appropriate range. It i s possible in this way to find N(s,t)
with considerable accuracy for all values of t, using only the first two
t e r m s of the series. Equations 3.19 and 3.21 describe N a s a function of stage number
and time for square cascades (N Ç1). A discussion of the roots of
Eqs. 3.20 and 3.22 i s given in Appendix A. It i s shown there that all
1, ~ ( s , t )FOR VALUES O F t > h s 2 / l 0 the roots pj (there a r e an infinite number of them) a r e negative. Thus
the t e r m s in the sums a r e transient, and the constant t e r m s represent
The inverse of G(s,p) suitable for calculations for large t i s the N(s,m). The second root is shown to be approximately
usual Heaviside expansion. It i s obtained by evaluating Eq. 3.8 A s the
sum of the residues a t the poles of G. Carrying out the indicated
operations, the following i s obtained.
Type I Operation:

Since the coefficient of epzt i s much smaller than that of eW, the
2e sin yjs ec(l-irttS-~) second transient t e r m i s negligible when o,t < -2; that is, when
epjt

+si K +-
AS
2
A
p, sin y, S
+-[Kpj~-e(l-$)~
2y7
+ 1] [Kp-e(1-I))]
(3.19)
46 THE THEORY O F ISOTOPE SEPARATION EQUILIBRIUM TIME O F A SQUARE CASCADE 47

Consequently, f o r most purposes only the leading transient t e r m need Accordingly Eq. 3.24 takes the form
be considered after t = ASVlO.
Equations 3.19 and 3.21 a r e valid both for strippers and rectifiers.
When the cascade o r column i s acting a s a stripper, c5, P, and S
remain positive, but cl is negative. Consequently e and >/Ã a r e also
negative. The same formal solutions hold. It i s quite incorrect to Tables of the functions A and B for K/AS between 0.0 and 0.5 and
analyze a stripper a s if it were a rectifier of the s a m e length, since
for 0<2cS<1.20 a r e given in Table 3.1 and a r e plotted in Figs. 3.2
the equations a r e not symmetrical about eS = 0.
and 3.3. A wider range of these functions for the special case K = 0
Particular examples of the solutions given by Eqs. 3.19 and 3.21
is given in Table 3.2, where values of 2eS ranging from -5.0 to +4.0
may be found in the reports listed in the references for this chapter.
a r e covered.
The most important special case of the two types of experiment,
namely K ' = -, P = 0, has been considered in detail. This is the case ) VALUES O F t < X
2. ~ ( s , tFOR S ~
of a column operated under total reflux (no production), with the con-
centration at the bottom maintained a t No. The concentration a t the Of interest here a r e the inverses of Eqs. 3.14 and 3.16 in the s,t
top of the column, o r in the reservoir a t the top, varies with time plane near t = 0, that is, the expressions for G(s,p) in the s,p plane
according to the relation f o r p near + W. Dropping t e r m s of the order e-2@small compared to
unity, Eqs. 3.14 and 3.16 may be written a s follows.
Type I Operation:

neglecting the higher transients. Now

Type 11 Operation:

and yl S is the smallest root of

Therefore ylS is a function of 6s and K/\S, and pl may be exhibited in The t e r m 6 is defined by Eqs. 3.15 and 3.18. This f o r m is obtained
the form by writing

and s o on.
Likewise Equations 3.29 and 3.30 hold to within a few per cent provided
0S = 6 s > fi;that is, replacing p by 1/t,
2<
= -A(2eS, K/kS)(ezes - 1) (3.27)
AS A
1
(Kp,S - 6 S + 1) (Kp, - e )
 EQUILIBRIUM TIME O F A SQUARE CASCADE 49
48 THE THEORY O F ISOTOPE SEPARATION

Thus they hold in precisely the range of t where the Heaviside expan-
sion begins to become inconvenient.
The inversion of Eqs. 3.29 and 3.30 a s they stand, in all their gen-
erality, is difficult and not illuminating. Instead, consideration will
0.98

Table 3.1 -Functions for the Time Equation

0.96
N(S,t) = e 2 t s - ~ ( ~ 2 6 sÂ-¥ i ) e - ~ t / ~ s
NO
2eS = In of steady -state enrichment
0.94
-K- - holdup in reservoir
AS holdup in column

-
If)
'0.92
.4
\
x

HOLDUP IN RESERVOIR
K/XS =
HOLDUP IN COLUMN

Fig. 3.2-A function for the time equation A(ZeS,K/XS).

be given various special cases one by one, thereby reducing the

complexity of the equations before inversion. For the moment atten- F i r s t consider special cases of Type I1 operation.
tion should be confined to the mole fractions at the endsof the column Type 11, Case I: K = K' = 0 (no reservoirs at either end)
(s = 0 and s = S). These a r e usually the only measurable mole frac-
tions anyway, and the restriction greatly simplifies the inversi0ns.l

'The rest of the column will be considered in Sec. 3.

 THE THEORY O F ISOTOPE SEPARATION EQUILIBRIUM TIME OF A SQUARE CASCADE 51

Table 3.2-Functions f o r Approach to Equilibrium of a Square Cascade with No External Holdup

M(s^ = e 2 ~ S- Ate%s - l l e - ~ t / ~ ~ '

74 .., ,.- = HOLDUP IN RESERVOIR

HOLDUP IN COLUMN
3 Rectifier
NO

Stripper

= LN OF STEADY-STATE
ENRICHMENT FACTOR

with M (-V2, %; - e2t/X) a confluent hypergeometric function. Values

2eS of the function a r e listed in Table 3.3.

Fig. 3.3-A function for the time equation B(2cS,K/AS).

whence whence

Equations 3.32 and 3.33 a r e the desired expressions.

or tables of inverses, see G. A. Campbell and R. M. Foster, Fourier Integrals Type 11, Case 11: K ' = -, K = 0 (infinite reservoir a t bottom, no
f o r Practical Application, Bell Telephone System Monograph B-584 (1931). reservoir on top)
52 THE THEORY OF ISOTOPE SEPARATION EQUILIBRIUM TIME OF A SQUARE CASCADE 53

Table 3.3-The Confluent Hypergeometric Function M(-Y2,%; -x) The condition in Eq. 3.36 is more stringent than in Eq. 3.31 if 2eS > 1;
but if 2eS 5 1 it i s automatically satisfied when Eq. 3.31 holds.
Type 11, Case 111: K ' = W, K = finite; 2e S < 1 (infinite reservoir at
bottom, finite reservoir on top; short column)

whence

Type 11, Case IV: K , K ' both finite; 2eS < 1 (finite r e s e r v o i r s a t
both ends; short column)

whence

whence

whence

whence
Simple expressions f o r more complicated c a s e s can only be found
by further specialization. Appreciable simplification of a l l formulas
would be obtained if e 2 with respect to Ap could be neglected. The
condition for validity of this procedure to a t least 2.5 per cent i s
54 THE THEORY OF ISOTOPE SEPARATION EQUILIBRIUM TIME O F A SQUARE CASCADE

Some c a s e s of Type I operation will now be considered. In the most

general case,

whence

If 2.5(1+ i^) S < l , s o that e 2 ( l + iA)' with respect to Xp may be neglected,

the r e s u l t is precisely the s a m e expressions f o r G a s in Type 11,
Case 111. To show the essential difference between operation with
production and operation without, the following case is given:
Type I, Case 1: k = 0 (no holdup in the column itself)

whence

N(0,t) is given by Eq. 3.42.

3 . DESCRIPTIVE MATERIAL ON RELAXATION PHENOMENA

FOR COLUMNS

If the results of the preceding two sections a r e combined, the ap-

proach to equilibrium of square cascades over the entire period of
operation may be followed. Figures 3.4 and 3.5 illustrate Type 11,
Case 11, which i s a cascade of a r b i t r a r y length, with no r e s e r v o i r a t
the top and an infinite r e s e r v o i r a t the bottom, operating under total
reflux. Plotted in these graphs i s N(s7t)
No
against (&) Vi
for

various values of 2eS = In 2 N(S 4. The initial p a r t s of the curves (for

NO
t/hs2 < 0.1) a r e plotted from Eq. 3.34, and the subsequent sections
THE THEORY OF ISOTOPE SEPARATION EQUILIBRIUM TIME OF A SQUARE CASCADE
58 THE THEORY OF ISOTOPE SEPARATION
come from the two-term Heaviside expansion (Eq. 3.28). The two
different p a r t s of the curve coalesce quite satisfactorily. The curves
a r e straight lines n e a r the origin, the deviation appearing sooner on
this s c d e for long columns (large 2eS) than for short columns.
Another way of presenting the same r e s u l t s may be obtained by
noting'that Eq. 3.34 gives N(S,t)/No a s a function of e2t/A. The ex-
pression breaks down a t ~ / A s ^ o . ~ , o r e2t/A= (2&)'/40. Figure 3.6,
Fig. 3.7-Graph of a short column o r cascade (finite reservoir at top; infinite r e s e r -
voir at bottom).
N(S,t) -- No
which gives , shows how columns with differing
NO
2eS follow the s a m e curve until they break away to approach their
equilibrium values.
Figure 3.7 gives a graph of Eq. 3.37 (infinite r e s e r v o i r a t the bot-
tom, finite r e s e r v o i r a t the top) for short columns. This i s Type 11,
Case 111, o r Type I for 2eS(l + V / ) < 1. Here N(S,t) - No has been
(2eS)No
plotted against t/AS2 for differing values of the ratio
EQUILIBRIUM TIME OF A SQUARE CASCADE
K holdup in top r e s e r v o i r
--
\S holdup in column
At high ratios this gives practically a straight-line graph. (But com-
pare with Figs. 3.4 and 3.5, where the abscissa i s the square root of
t/\S2, and a straight line i s found when there i s no holdup on top.) Ob-
viously N increases l e s s and l e s s rapidly a s K increases.
Fig. 3.8-Variation of enrichment with production. Constants: 2eS = 0.8, S = 100 cm,
A = 2 sec/sq cm, K = 0.
In Fig. 3.8 the effect of changing r a t e of production on equilibrium
time i s shown. Here 2eS = 0.8, \ = 2 sec/sq cm, S = 100 cm, and
K = 0. F o r this graph the solution of Eq. 3.19 was used, keeping two
transient t e r m s . Increasing the production r a t e d e c r e a s e s both the
eventual enrichment and the relaxation time of the column. A cascade
operating a t a lower production r a t e will reach any particular value
of the enrichment sooner than one operating a t a higher production
rate, except in the very earliest stages of the operation, in which the
concentration increase i s independent of 1/1.
It is of interest to follow theoretically the development of the con-
centrations along the entire column, especially since it i s very diffi-
 THE THEORY OF ISOTOPE SEPARATION EQUILIBRIUM TIME OF A SQUARE CASCADE 61

cult to do s o experimentally. Referring, therefore, to the master

I I I I I
equations (Eqs. 3.29 and 3.30), an infinite r e s e r v o i r i s taken a t the
bottom, no r e s e r v o i r on top, and 2.5s < 1 (Type 11, Case I1 f o r short
columns), and it i s found that

The inverse of Eq. 3.44 i s

erfc
2eS(l - s/S)
4 lGq%-
- d$ exp - [ 2 e ~ ( +l s/s)12
16e2t/A

+ 2eS(l + s/S) erfc

2eS(l - s/S)
4Ve^tA 1 (3.45)

which holds a s long a s &/A < (2e~)'/40. F o r large values of e2t/A

Eq. 3.28 may be used with K = 0.
Figure 3.9 shows the behavior of such a column whose equilibrium
fraction is twofold. Thus 2eS = In 2 = 0.693, and the critical value of
e 2 t A i s 0.0125. Here N(s't) - is plotted a s a function of s / S for
No
various values of 4e2t/A. The upper curves (e2t/A 5: 0.0125) were
plotted from Eq. 3.28, the lower (e2t/A =s 0.0125) from Eq. 3.45. The
two s e t s of curves fit admirably. It should be added that, during the
early stages of the experiment, enriched material accumulates only
a t the top of the column; when this enriched slug has extended itself
down to the foot of the column, the nature of the p r o c e s s changes, and
there is a change from one form of asymptotic equation to the other.

REFERENCES

Cohen, K., Report A-530, Feb. 4, 1943.

Cohen, K., and I. Kaplan, Report A-195, June 17, 1942; Report A-395, Nov. 19, 1942;
Report A-1276, May 1, 1944; Report A-1292, May 29, 1944.
FRACTIONAL LENGTH OF COLUMN Kaplan, I., Columbia Ser. No. 4L-X136, Oct. 12, 1942 (letter t o C. A. Hutchison).

Fig. 3.9-Cascade of fixed length (2eS = In 2) (infinite reservoir at the bottom, no

reservoir at the top).
 DETERMINATION O F CASCADE CONSTANTS 63

An experiment should be considered in which the withdrawal of

product at the top i s balanced by the introduction of an equal amount
of feed at the same place (Fig. 4.1). In this way the flows up and down
a r e equa1,so that the operation of the unit resembles that of a column
in the base stages of a cascade; and the equations a r e appreciably
simp1ified.l
Chapter 4 PFg;CT FEED

DETERMINATION OF CASCADE CONSTANTS

A complete analysis of certain transient phenomena of square cas-

cades was given in the last chapter. This amount of material would
not be justified were its purpose purely descriptive. The r e a l need
for the detailed results just obtained i s given in this chapter, in which
it i s shown how to turn the equilibrium effects into a tool for deter -
mining the cascade constants.
It i s always desirable todetermine the fundamental constants of the
steady state, cl and c5, by direct experiment. In some cases, such a s
in thermal diffusion, the constants cannot be predicted because of
ignorance of the theory. In other cases, such a s centrifuges, in which
the constants a r e known functions, subsidiary parameters (viscosity,
diffusion coefficients, etc.) a r e not known to sufficient accuracy. In
all cases it i s useful to have experimental confirmation that cascade
operation follows calculation.
This restriction i s mainly responsible for the elegance of the r e -
sults-the -
generalization to N 1 will be much more cumbersome.
Take the holdup per stage o r per unit length, c , the production, P ,
and the number of stages, S, a s known; c and c a r e to be determined.
Consider f i r s t the evaluation of the constants along conventional lines,
i.e., by experiments on the steady state.
Equation 2.13, which may be written

Fig. 4.1 -An experiment for the determination of column constants.

Following out the same kind of analysis a s in Chap. 1, Sees. 2 and 6,

this case gives, analogous to Eq. 2.3,
contains the two parameters e and c , which ostensibly may be deter-
mined by any two observations of P and Np. Such a naive procedure
i s subject t o considerable experimental e r r o r , and another method,
allowing the use of many data, will be presented.
his device furnished by W. I. Thompson.
 64 THE THEORY OF ISOTOPE SEPARATION DETERMINATION O F CASCADE CONSTANTS 65

The solution of Eq. 4.1 i s

which gives directly the separative work. At the same time there
i s experimental determination of the optimum production r a t e and
fractionation.
Next, consideration should be given to the methods based on the
equilibrium phenomenon. These a r e chiefly useful in c a s e s where the
(compare with Eq. 2.14).
time to reach complete equilibrium i s of the order of days o r weeks,
Now, s e t up f o r this system the function
so that a s e r i e s of steady-state experiments would be excessively
time -consuming.
T o illustrate the use of relaxation phenomena for determining cl
and c5, consider an experiment of Type I1,Case 111 (K' = m, K = finite,
2eS < I).' The pertinent equation for short t i m e s i s (Eq. 3.37)

which i s a generalization of the work function of Eq. 1.60, defined for

1N(S t) e -" -
2cK
infinitesimal concentration changes. The function i s plotted against P,
from the experimental values of N p / ~ oand P. The maximum value
NO
+ 4
( ) (exp $ erfc ^-y- - 1)
of W i s a t
= l + % ( ~ Y + 2 e s -
6 xs2 XS
[ (yy AS^
exp - - e r f c -
AS -
K AS
(
t\^_

H, the holdup in the top r e s e r v o i r , i s known, a s well a s c6. Thus

and the maximum is
XS/K = c6S/H i s known, but e, \, and K individually a r e unknown.
Equation 3.37 holds until t/& = 0.1. Unless XS/K Ã 1 (holdup in
column much greater than holdup in r e s e r v o i r ) , the argument of the
e r r o r function
?h
It is worth noting that at P = 3 (L)
K AS2

will be small a s long as the equation is valid. Expanding the e r r o r

function in ascending powers of the argument, Eq. 3.37 behaveslike
The values of and W a r e chosen f r o m the curve. Then

c1 = 2 V P W (4.7)

and The deviation from this formula becomes l e s s and l e s s a s t -

0.
On the other hand, should the holdup in the r e s e r v o i r be negligible
(e.g., XS/K > l o ) , then unless t/AS2 i s very small the argument of the
e r f c will be large,

F o r small enrichments, eZEs= 1 + 2 4 and the maximum of W is \ S l t \ " K ( W - / i S f t

erfc - - - - exp - - -
K xS2 A S rt AS2

he sample size i s supposed to be negligibly small.

p ~ :i-sr_*- %,
9 %-
66 THE THEORY OF ISOTOPE SEPARATION
and Eq. 3.37 behaves like
over a considerable region.
Depending: on which i s the appropriate case, N's9t' - 1 i s plotted
No
against t or t%. The slope of the straight-line portion of the curve is
found, which i s either 2e/K = ci/H, o r 4e/* = 2 c 1 / i G . In either
case a relation for the desired constants is obtained.
The data for long times, which a r e more abundant, a r e worked up in
quite a different manner. The pertinent equation i s (Eq. 3.28)
Here the procedure i s to find e2" by extrapolation and then to plot
1 [ e 2 â-
‚ No]
No
against t. If the initial choice of e2" is good,this
graph will be a straight line1 whose intercept i s In [ ~ ( e -~I)].
' ~ (A is
found from Fig. 3.2.) If this i s not the case, a slightly different choice
of ezCs must be made. From the straight line finally obtained, pl i s
found a s the slope.
A consistent set of values of 2eS and p, i s thus found. From the
former, c1/c5 = 2e. From pi and the relation
where B is found from Fig. 3.3, cs is determined.
The weak point of this procedure is the choice of e2cs,the equilib-
rium fractionation, which i s not always unambiguous. However, the
value of the separative power of the cascade per stage, which i s the
really important property of the cascade for design purposes,
i s unaffected by considerable e r r o r s in 26s when determined in this
manner.
'Except, of course, for small values of t.
DETERMINATION OF CASCADE CONSTANTS 67
As an example of how simple the whole procedure can be made,
take the special case K = 0. From a graph of N(S,t)/No against t , the
equilibrium fractionation may be found, along with tv,, the time when
The nomograph (Fig. 4.2) then gives c,.c5/c2 on the right-hand line
when a straightedge i s laid against N(S,-)/Noon the left-hand line and
ti/, on the center.
1. IDEAL CASCADE WITH LOSSES
So f a r , cascades, both ideal and otherwise, in which no losses of
material occur have been considered. In actual cascades, sources of
loss a r e always present. In most chemical processes the effect of
small losses i s not serious.
It is obvious that, in a given cascade, a s the losses increase the
production diminishes, so that a cascade which has losses will have
to be bigger to produce the same quantity and quality of material.
Consider the problem of constructing a no-mixing cascade that will
accommodate losses. To solve it a different procedure1 from that of
Chap. 1 should be employed.
It i s necessary to introduce the following notation (see Fig. 4.3):
Ls = esLsN; = flow of desired isotope i n enriched fraction
Me = (1 - es)LsN? = flow of desired isotope in stripped fraction (4.12)
ZS = Os L s ( l - NL) = flow of other isotope in enriched fraction
ms = (1 - & ) L s ( l - Ni') = flow of other isotope in stripped fraction
From the conservation of matter,
Ls + Me = LsNs = flow of desired isotope into the separator
(4.13)
ls + me = Ls(l - Ns) = flow of other isotope into the separator
 THE THEORY OF ISOTOPE SEPARATION DETERMINATION OF CASCADE CONSTANTS

ENRICHED FRACTION

t LsMOLES
I s MOLES

STRIPPED
OF DESIRED ISOTOPE
OF OTHER ISOTOPE

FRACTION

Me MOLES OF DESIRED
ISOTOPE
ms MOLES OF OTHER
ISOTOPE

MOLES OF DESIRED
FEED ISOTOPE

< ( L ~ +M ~ ) MOLES OF DESIRED ISOTOPE

( l s tmsl MOLES OF OTHER ISOTOPE

Fig. 4.3-Diagramof a cascade with losses. Only the desired isotope flows areshown.
Flows of the other isotope are obtained by using lower-case rather than capital italic
letters.

The no-mixing condition leads to the analogues of Eqs. 1.3 to 1.5, the
most convenient of which i s

From Eqs. 4.14 and 4.15, eliminating rns/L,

Ls = a M s

Now the cascade i s connected in such a way (Fig. 4.3) that

Ls + Ms = Ls.l + M g + 1 - losses of desired isotope (4.17)

Fig. 4.2 -Nomograph for determining column constants. The losses per stage a r e often proportional to the flow in the stage.
Thus the losses from the sth stage of a cascade a r e set a s yLs,where
y i s a constant and Ls i s a s usual the flow entering the stage. Equa-
tion 4.17 becomes
 THE THEORY O F ISOTOPE SEPARATION DETERMINATION O F CASCADE CONSTANTS 71

o r , introducing Eq. 4.16,

&+I - (a + 1) (1 + y) Ls + G - 1 = 0 (4.19)

Equation 4.19 i s a second-order difference equation for Lg.The gen-

eral solution i s
That <t;,(l/a) = ( l / a ) u l ( a ) and u 2 ( l / a ) = ( l / a ) a i 2 ( a )i s shown b y
Eq. 4.21. Also since ul(a)u,(a)= a , the following schedule may be
written:

where u1 and u2 a r e the roots of the characteristic equation

^ - ( a + 1) ( 1 + y ) u + a = o

and can be expressed a s

with A defined a s

Using the form of Eq. 4.27 to simplify Ls and Is,

The boundary conditions a r e

Solving for A and B,

- 2 P N p ( f f + 1 ) (1 + A)a(s-s-1)/2 sinh ( S
- + 1 - s) In (1 + A) -/a (4.28)
[ a ( l + A)' - l ] N s

This i s the required flow per stage with a fractional loss per stage y.
Now from Eqs. 4.14 and 4.15, The total flow in a cascade i s obtained by summing the Ls with r e -
spect to s from 0 to S. This gives

while Eq. 4.18 holds with Z and m substituted for L and M. Therefore
Z S i s the same function of l / a that L s i s of a , and in place of Eq. 4.23
the result i s
72 THE THEORY O F ISOTOPE SEPARATION DETERMINATION O F CASCADE CONSTANTS 73

-
As before, Rs = o^Ro and R p = crS+lRo. When A 0, Eqs. 4.28 and 4.29
reduce to the familiar expressions of Eqs. 1.11 and 1.13.
Equations 4.28 and 4.29 take more easily understandable forms if
a = 1 + e. For then
The above results a r e quite sensitive to the mole fraction produced
by the cascade. Figure 4.5 shows that value of y/e2 which demands a
compensatory increase of 20 per cent in the total flow of the cascade
a s a function of Np. A cascade producing a concentration of 10 per
cent can afford more than five times the losses of one producing 90

It is found convenient to define an auxiliary symbol 6 by

From Eqs. 4.30 and 4.31, 6 = A/â‚ Using these relations gives, in
place of Eq. 4.28,

4PNp
Ls = Ns e ( l + 26) exp [$(s-s)] sinh [i( S - s) (1 + 2611 (4.32)

and for Eq. 4.29,

The relative increase in size of a stage caused by the loss of ma-

terial i s given by the ratio

Fig. 4.4-Curve showing the effect of losses on total flow (the loss per stage i s pro-
portional to the flow per stage). N p = 0.90, y = fraction of flow lost per stage, e = en-
richment per stage.

which i s a function of 6. Since 6 i s in turn a function of y/e2 = yL/Le2,

it i s seen that the increase in size of a stage i s a function of the ratio
of l o s s to separative work. The bottom stages a r e enlarged propor -
tionally more than the upper stages.
The increase in total flow for a (rectifying) cascade i s found by
comparing Eq. 4.33 with Eq. 1.23, which i s its limit for y = 0. Figure
4.4 shows the percentage increase in total flow a s a function of y/e2
for the standard case. It will be seen that a 20 per cent increase in
size of the cascade i s required when y = 0.025 e 2 .
per cent. Intuitively it should be understood that this effect is due to
the increasing penalty for losing high-concentration material on which
much work has been done. Obviously it i s worth while to take con-
siderable pains to avoid losses in the small upper stages of the cas-
cade. It may well turn out to be economical to use expensive methods
and materials in the upper stages if by so doing losses can be cut and
thus the amount of overbuilding required in the bulk of the cascade
can be diminished.
74 THE THEORY O F ISOTOPE SEPARATION DETERMINATION OF CASCADE CONSTANTS 75

The effect of losses on a stripping section i s somewhat different. The amount of material transferred between two cascades depends
The usual objective i s not that of producing a given quantity of waste, on the mole fraction of the return stream from the upper cascade a s
but rather that of economizing feed. A loss in stage s of the stripper follows: Let PC and Nc be the rate of production and mole fraction of
means that a small amount of material i s stripped down to Ns and not the product from the upper cascade, respectively. Let PB and NB be

MIXI N G

F I N A L M O L E FRACTION FOR WHICH P L A N T WAS DESIGNED

Fig. 4.5-Loss of process material resulting in 20 per cent increase in total flow (the
loss per stage i s proportional to the flow per stage). The symbol e denotes the enrich-
ment per stage, and y represents the fraction of flow lost per stage.

N w . The result i s to decrease the size of the stripper and increase

the required feed. This problem will not be pursued any further here.
The problem solved above (loss proportional to flow) is about the
only one that admits a closed analytical solution. But, since in prac-
tice the only cases of interest a r e those in which the effect of losses
is small, iteration methods a r e reasonably successful in dealing with Y

more complicated conditions (see References). LOWER CASCADE

Fig. 4.6-Diagram showing the combination of cascades.

2. COMBINATION O F PLANTS USING DIFFERENT
SEPARATION PROCESSES

In this section consideration will be given the connections between the corresponding quantities for the lower cascade. Pn i s at the same
two plants with different separation processes, giving expressions time the feed to the upper cascade. Pp and NÃ a r e the waste stream
with whose aid the factors of plant size and amount of material trans- and its concentration from the upper cascade; Pp i s the return stream
ferred between the plants may be balanced. to the lower cascade (refer to Fig. 4.6).
76 THE THEORY OF ISOTOPE SEPARATION DETERMINATION O F CASCADE CONSTANTS 77

The conservation of matter gives the relations Furthermore the return stream to cascade A i s made up of two
parts so that

Substituting f o r lA- lBfrom Eq. 4.41,

From Eqs. 4.35 and 4.36,

and rearranging,
LANA- lAnA= LBNA- lBnB- P-N = PBNB - P- N
C A C A

In these two expressions, P C and Nc a r e given (they a r e the specifi-

cations on the final product), and No i s the mole fraction at which the
change from the lower to the upper cascade is made. NA, the mole
fraction of the return stream, i s the only variable.
If the upper cascade is an ideal rectifier, Eq. 1.5 gives
Equations 4.42 and 4.43 show that cascade A is just the part up to
NA of a cascade that produces material (Pc,Nc).
Expressions for the flows and holdups in these sections may easily
be found on the assumption that they a r e ideal cascades. The total
flow in cascade section A is

For (a - 1) small, (Nn - NA) is very small, and PB and Pc a r e very

large. Thus in general it will be advisable to add a stripper to the
upper cascade, a s indicated in Fig. 4.6. The connections of the return
stream, between p a r t s A and B of the lower cascade, a r e also shown.
The upper cascade consists of an ordinary rectifier -stripper com- i f 1 + â ‚ is the simple-process factor in the lower cascade. The total
bination. Cascade B i s clearly a rectifier, receiving a feed of mate- flow in cascade section B is
r i a l at mole fraction NA and producing material of mole fraction NB.
From the conservation of matter

The total flow in cascade section C is

The conservation of matter in the reservoir (actually i t s size i s im-

material) requires that
where 1 + ec is the simple-process factor in the upper cascade.
The total flow in the stripper section A is
78 THE THEORY OF ISOTOPE SEPARATION

and in the stripper section C is

Analogous expressions hold for the finite case. Expressions for the
desired material holdup and the contribution to the equilibrium time
of the various sections above the feed point can also be obtained with- Chapter 5
out any trouble (see Appendix C).
Cascades A, A, and C a r e the same a s the corresponding sections THE CONTROL PROBLEM
of a single ideal cascade producing (Pc,Nc). But it will b e observed
that cascade B is not the same as the section between NA and NB of a
single cascade. B and C together a r e equivalent to (and slightly larger The problem of maintaining certain specified variables a t pre-
than) this section. An inefficiency in the combined cascades caused assigned values is common to any planned operation. In a chemical
by the mixing of the constituents of the return stream to cascade A reaction, for example, these quantities a r e usually the temperature,
causes a l o s s in value of the pressure, and the amount of material in process. The control
problem for isotope-separation plants differs f r o m that of other
manufacturing operations in the degree of complexity. The cascades
considered in this volume contain thousands of different interdepend-
which is made up for by the increased size of B and e. ent stages or elements. Methods of analysis which a r e successful
It will be remarked that a combined cascade, for maximum effi-
with systems containing, for example, a dozen related units will fail
ciency, i s considerably different from two cascades designed for other
here. F o r this reason the control problem for isotope-separation
duties and connected together. There is a considerable premium on
cascades must be treated a s a distinct subject.
foresight in this problem.
Furthermore it is often possible to break up a cascade into smaller
The separation into two cascades which i s illustrated here can
units. For example, a cascade of 10,000 centrifuges with nine cen-
clearly be carried on further, giving any number of sub-cascades.
trifuges in the top stage (Chap. 2, Sec. 4) might be broken up into nine
The chief feature which makes continued subdivision attractive, even
completely independent cascades in parallel, o r a cascade might be
when the sub-cascades use identical equipment, i s that the flow be-
broken down into sub-cascades along the lines discussed in Chap. 4,
tween cascades (each of which i s a rectifier-stripper) i s much l e s s
Sec. 2. The desirability of such expedients will not be clear until the
than the flow between stages of a single cascade (cf. Eqs. 4.37 and
difficulties of controlling a single large cascade a r e known. In the
4.38). This is an advantage in removing accumulated impurities
discussion that follows, therefore, consideration will be given to a
(caused by inleaks, decomposition, etc.) o r in control.
single large rectifier -stripper.
It was shown in Chap. 1, Sec. 2, that for an ideal plant deviations up
to 20 per cent from the calculated flows a r e without appreciable ef-
REFERENCES
fect on the separation. This also holds true to a considerable extent
Cohen, K., Report A-60, October 1941; Report A-530, Feb. 4, 1943; Report A-781, for nonideal plants of intelligent design. Thus to have a cascade op-
Aug. 13, 1943.
Cohen, K., Columbia Ser. No. 4R-X143, May 13, 1942.
erate according to plan, it is only necessary to s e e that the flows
Cohen, K., and W. I. Thompson, Columbia Ser. No. 4R-104, Sept. 4, 1943. throughout the plant a r e approximately correct. This question, of
Kaplan, I., Columbia Ser. No. 4M-133, Jan. 18, 1944 (memorandum to K. Cohen); maintaining the temperatures, pressures, flows, etc., near their de-
Columbia Ser. No. 4M-160, Apr. 17, 1944 (memorandum to M. Kilpatrick). sign values, of seeing that material does not accumulate in o r drain
Lefkowitz, H., and B. Schwartz, Columbia Ser. No. 4R-18, Feb. 4, 1943.
Lefkowitz, H . , B. Schwartz, and H. Mayer, Columbia Ser. No. 4R-42, Apr. 12, 1943.
out of any part of the plant, and of seeing that small distrubances do
Mayer, H., I. Kaplan, and K. Cohen, Report A-396, Dec. 2, 1942. not build up to large ones, is one part -the hydrodynamic part-of
Thompson, W. I., Experimental Operation of a Single Centrifuge, S.O.D. Report, June the control problem.
14, 1943.
 80 THE THEORY OF ISOTOPE SEPARATION THE CONTROL PROBLEM 81

There is in addition a second part which deals not with the average
conditions but with the effects of fluctuations from the average. A
slight change in the operating conditions may produce a variety of
effects: a change in the transport of material from stage to stage, a
mixing of material of different concentrations from different stages,
o r a change in the separating power of the element. In general, a
fluctuation causes mixing, and mixing means a lowering of the con-
centration of the product.
Both parts of the control problem a r e essential. Without the hydro-
dynamic part, the types, magnitudes, and frequencies of variations
The effect of perturbations in the fundamental quantities cl, c5, c6,
that a r e to be expected a r e unknown. Without a knowledge of the ef-
and P on the separation performed by the cascade is found by solving
fects of fluctuations on concentration an evaluation cannot be made of
Eqs. 5.1 and 5.2 with these four quantities preassigned functions of
the effectiveness of the hydrodynamic control measures s o far a s
time. Of primary interest a r e the small fluctuations, which if im-
plant productivity is concerned.
posed on a system already in the steady state will give small changes
By i t s nature, the hydrodynamic problem is intimately associated in concentration. Gross upsets and phenomena associated with the
with the particular separation method, equipment, and control in-
approach to equilibrium a r e excluded from consideration. The de-
struments used. Since this problem cannot be generalized without
sired solution may thus be obtained by ordinary perturbation theory,
destroying its illustrative value, an explanatory example will be
with the steady state a s the unperturbed system.
presented.
The quantities c,, c5, and c6 a r e given a s functions of stage number.
The problem of the effect of fluctuations on separation is more In cascades with more than one element per stage, they a r e to be
susceptible of general treatment. Even s o there a r e aspects which understood a s averages over the elements. The mixing losses caused
differ from one case to another. A cascade of centrifuges, for in- by deviations of parallel elements from their average a r e not in-
stance, has certain problems in common with all squared-off cas- cluded here.
cades; i t also has effects due to variations in the behavior of parallel Suppose that there is a primary disturbance of some nature from
units which a diffusion cascade lacks. But the theory developed in the an outside source. It will be written a s
succeeding sections is basic to all the problems. In Sec. 1 considera-
tion is given to equations governing the change of mole fraction with
fluctuations in the operating constants. Their solution in the general
case is developed in such a way that the results of the hydrodynamic The parameter u is not necessarily small-its function is to distin-
problem may be used in their most compact form (in the s,p plane of guish between the various o r d e r s of t e r m s resulting from the pri-
the Laplace transform). In Sec. 2 the average change in transport for mary disturbance.
periodic fluctuations is derived. Section 3 treats a particular ex- As a consequence of of there will be induced disturbances in cl, c5,
ample: the effect of variations in the rate of withdrawal of product on c6, and P. Thus c,, which before the disturbance is a function of s
a square cascade. Sinusoidal variations a r e considered, and the effect
only, becomes a function of s , t, and u. Expanding in powers of a,
of batch (instead of continuous) withdrawal of product is treated.

1. E F F E C T O F FLUCTUATIONS ON CONCENTRATION

The fundamental equations of isotope separation a s developed in where cm(s) is the undisturbed value of c,. Generally the symbol
Chap. 1, Sec. 6, using the generalized notations of Chap. 2, are1 c1(s,t;u) will be shortened to cl(s,t). Likewise

The terms c,, cc, and c. were introduced in Chap. 2, Square Cascades, by the sched-
ule in Eq. 2.2,but there i s no reason why they cannot be employed a s functions of s and
t, a s indicated.
82 THE THEORY O F ISOTOPE SEPARATION THE CONTROLPROBLEM

The notation is self-explanatory and defines the coefficients c5,, cS2,

etc. As a result of Eq. 5.5 and the f i r s t line of Eq. 5.2,

.
The functions c,(s,t) . . T(s,t) a r e considered known, a s well a s the
coefficients of their expansions. They a r e in fact the solutions of the
hydrodynamic part of the control problem.
Now, inserting Eqs. 5.4, 5.5, and 5.6 in Eqs. 5.1 and 5.2, it is de-
sirable to find N(s,t) and r(s,t). These will also be functions of a, and s o on for higher orders.
which will be written asi The boundary conditions for each order equation can likewise be
found by expansion. This will be done later in the section (Eqs. 5.23
N(s ,t) = No(s) + aNl (S ,t) + a2N2(s ,t) + . . . e t seq.). When the boundary conditions a r e known the equations may
(5.7) be solved successively. In the paragraphs immediately following an
T(s,~)= To + C T T ~ ( S+ o, ~' T) ~ ( s , ~ )+ . . - investigation will be made of the properties of the differential equa-
Of course tions, and method of solving them will be considered.
Equation 5.9 gives the steady state of the unperturbed system. The
expression for T , has the form
where N p is the undisturbed mole fraction of the product.
The convergence of Eq. 5.7 is now of concern. The natural period
of a cascade with respect to concentration disturbances is the equi- where D is the linear operator, independent of t,
librium time, which we have seen in Chap. 3, Sec. 1, is of the order
\/e2 - c.-n~50/cfo.Physically this means that Eq. 5.7 will converge
provided the average values of the disturbances on most of the stages,
over periods of time smaller than \/e2, a r e small compared to the and
undisturbed quantities.
Introducing the expansions of Eqs. 5.4 to 5.7 into Eqs. 5.1 and 5.2,
and separating the coefficients of uO, ul, 02, , the following sys- ...
tems of equations a r e obtained:
i s , after substituting No(s), which is a solution to Eq. 5.9, an explicit
function of s and t. Combining Eq. 5.12 with the second equation of
PoNpo = PoNo(s)- Cgo aNo
s
,- cioNo(l- No)
+ (5.9) Eq. 5.10 gives

a aQ (5.15)
+ TPo '60 at
a N l + -a D
s Nl=-+Nas o "
a1
s
9No
- '51 + cllNo(l - No)
(5.10) The left-hand side of Eq. 5.15 is a linear homogeneous expression
in N17with coefficients independent of t; the right-hand side is an ex-
plicit function of s and t. It is therefore easily solvable by standard
methods. In particular the Laplace transform of Eq. 5.15,

'The symbols N,,(s), N,(s,t) ..

. introduced in Eq. 5.7 a r e not in contradiction with
the previously used symbols of Nc, which have been supplanted by N(s,t) and have no
84 THE THEORY O F ISOTOPE SEPARATION THE CONTROL PROBLEM 85

is an ordinary linear second-order differential equation in s. It

be remembered that the asterisk (*) was introduced to denote
Laplace transform of any function. Thus if x(s,t) is a function
and t, where D i s the linear operator of Eq. 5.13 and

The chief advantage1 of using the Laplace transform of Eq. 5.15-

besides the fact that i t solves the equation without trouble in spite
complicated boundary conditions -is that the transforms TT , cz , a Combining Eq. 5.19 with the second part of Eq. 5.11,
c * a r e used on the right-hand side. Since the hydrodynamic par
the control problem will also be solved by using the Laplace t r a
form, these quantities a r e immediately available. In fact, s i
the inversion of T* to TI, etc., i s the last and hardest step of
hydrodynamic solution, very considerable difficulties a r e avoi The Laplace transform of Eq. 5.21 is1
by using Eq. 5.16.
The solution of Eq. 5.15 or 5.16 may be facilitated by transforming
the dependent and independent variable^.^ The expression DNl be-
comes, using the integrating factor,

'lo(' - The process can, a t least in principle, be continued indefinitely f o r

[ Ls
+ 0'
J ( s ) = exp -
'50 ds] NS, N4 . . . .
The boundary conditions for the differential equations will now be
obtained. First of all the initial condition N(s,O) = No(s) leads to

The dependent variable is then chosen a s JN,. In many cascades the

quantities clo, cco, etc., a r e most conveniently expressed in terms of
the No. For example, an ideal cascade is defined in terms of the No Assume that there a r e reservoirs a t each end of the plant, of holdup
so that H (product end) and H' (waste end). The product and waste a r e with-
drawn from the reservoirs. The mole fractions in the reservoirs a r e
c
-i-^- = 26 = constant therefore called Np(t) and Nw(t), respectively. e ere
i t is not legiti-
'50 mate to neglect the difference between Np and N(S,t) . A rapid fluctua-
tion in N(S,t), for example, leaves Np relatively unchanged because of
the buffer effect of the holdup.] The waste stream is called W(t). The
flow from the cascade into the top reservoir i s called c5(P,t);the flow
from the cascade into the bottom reservoir is c5(W,t). These quanti-
This suggests changing the independent variable from s to No, which ties also have expansions a s powers of o.
can be done with the aid of Eq. 5.9.
The expression for T, has the form
'The formula for the transform of the product of two functions i s
1
[x1(t)x;(t)]* = %/ x: (P - q)x;* (q) dq
'This feature was pointed out by H. Mayer. Br,

2Following the practice of K. Fuchs. The contour i s to the right of the poles of x;*(q) and to the left of the poles of x* (p - q).
 THE THEORY O F ISOTOPE SEPARATION THE CONTROL PROBLEM 87

W(t) = Wo + a w l (t) + 0% (1) + .. . The boundary conditions a t the top of the rectifier a r e obtained
from the relations
H = Ho + u ~ , ( t ) + u2H2(t) + . ..
H1=H; +uH'(t) +u2Hn(t) +...
Np(t) = Npo + uNpl (t) + o w 2(t) + ... (5.24)
Nw(t) = Nwo + uNwl(t) + cr2Nw2(t) + . . .
c5(P,t) = c ~ ~ ( +P uc5,(P,t)
) + u ~ c , ~ ( P +, ~.)
. a

c5(W,t) = cso(W)+ ucgl(W,t) + U ~ C ~ ~ +

(W. . ,. ~ ) On expanding in powers of o and collecting, it i s found that

The mole fractions, N(s,t), and the transports, T(s,t) and ~ ( s , t ) ,

will r e f e r to the stripper if s has negative values. Equations 5.9 to
5.22 hold for the stripper if Po and N p a r e replaced throughout by
-Wo and N W . The stage number of the lowest stage of the stripper
will be called "B" (a negative number).
The discontinuity in T and T a t s = 0 implied in the change in the These obvious relations have already been incorporated in the pre-
equations is accounted for by the introduction of feed (F,Np). ceding equations (cf. Eqs. 5.6 and 5.8). Likewise, combining the f i r s t
two equations of Eq. 5.29,

Note, however, that Np does not have to be the same a s N(O,t), o r even
N F the s a m e a s No(0). F and Np may both vary with time (the cas-
cade under consideration might be theupper cascade of a combination
of plants). Expanding Eq. 5.25 in powers of u,

The boundary conditions a t the bottom of the stripper a r e very

similar. Corresponding to Eq. 5.29,
88 THE THEORY OF ISOTOPE SEPARATION THE CONTROL PROBLEM

In z e r o order

F r o m Eq. 5.31,
The first-order t e r m s a r e

and from Eq. 5.35,

and the second-order t e r m s a r e

This elaborate s e t of conditions is very expeditiously condensed by

using the Laplace transform. With the aid of Eq. 5.23,

Substituting for 7, from Eq. 5.12 in the boundary conditions (Eqs. 5.27,
5.31, and 5.35), a complete s e t of boundary conditions may be found
for the first-order differential equation (Eq. 5.25). F i r s t a n expres-
sion for
and therefore N: and N: may be eliminated altogether by the second
p a r t s of Eqs. 5.39 and 5.40, which give

is needed. This can be obtained from Eq. 5.26 by substituting Eq. 5.9
and its analogue with -Wo for Po and N W for N p . It is found that1

with the definitions

Then from Eq. 5.27, for the boundary condition at s = 0,

I N p - No(0) i s found by substituting the solution of Eq. 5.9, giving N p in terms of

N,(O), and N,,, in terms of No(0), into Eq. 5.26

a s in Chap. 3. With Eq. 5.41, the transforms of Eqs. 5.39 and 5.40 a r e
90 THE THEORY O F ISOTOPE SEPARATION
The transform of Eq. 5.38 i s
Similar expressions can be found for the boundary conditions of the
higher-order differential equations.
It must seem to the reader that the work required has been in-
creased by carrying the intermediate quantity r all through the ex-
pansions and then eliminating it at the last step, instead of elimi-
nating i t at once from Eqs. 5.1, 5.25, 5.29, and 5.33. The procedure
was adopted to prepare for the work of the next section, where i t s
usefulness will become apparent.
With this cataloguing of the boundary conditions all the development
that can be made without special assumptions has been completed.
The treatment has been more detailed than the novelty of the content
strictly merits; i t i s hoped that i t s reference value will make up for
the difference.
2. AVERAGE TRANSPORT LOSS FOR PERIODIC FLUCTUATIONS
There a r e two aspects of the effect of perturbations which a r e i m -
portant. The f i r s t is to predict the concentration changes in detail,
s o that the success of corrective m e a s u r e s can be verified. The sec-
ond i s to find the l o s s of production over long periods of time.
If the primary disturbance uf(s,t) is nonperiodic, like a simple
pulse o r a damped oscillation, both aspects a r e of interest. The lead-
ing t e r m in the answer is N,(s,t), but often N2(s,t) is also required to
give a correct evaluation of the production loss.
If the primary disturbance i s periodic, a new average steady state,
with concentrations lower than before, will result. In this c a s e we
a r e more interested in the new steady state-the l o s s of production
and average change in concentrations-than in the full detail of the
THE CONTROL PROBLEM 91
answer. Usually1 the leading t e r m in the answer is the average
second-order change in concentration o r in transport. Of course the
average second-order quantities can be found by solving straight-
forwardly for N,(s,t), a s indicated in Sec. 1, and then averaging. But
this procedure requires considerable work to get superfluous detail;
it is possible to obtain2 T Z and Nz(s) directly without solving the
second-order differential equation. The method for doing this, which
was devised by K. Fuchs, follows.
Suppose that the solution N,(s,t) o r N*(s,p) of the equation of the
f i r s t order in u has been determined. Taking the average of Eq. 5.19
and using the relation Eq. 5.18 for D,
F r o m Eq. 5.2, a ~ / &= - a(c,.~)/at.The right-hand side must be zero;
otherwise there i s no steady state. Therefore T i s constant, except
for a discontinuity a t s = 0, and s o a r e all o r d e r s of i t s expansion,
including G. The solution of the problem r e s t s on this property of
-
r Z , which permits i t s evaluation f r o m the boundary condi4' .ions.
Equation 5.46 integrates out to give, for the rectifier,
and, for the stripper (negative values of s ) ~
'It may be supposed without loss of generality that af(s,t) has a vanishing time a v e r -
age. It will then usually be t r u e that the first-order t e r m s in the fundamental quanti-
..
ties c,, cs, co, P . have vanishing time averages:
-
Should this b e the case, then N,(s,t) = T , = 0, for, on averaging Eq. 5.10 and the bound-
a r y conditions (Eqs. 5.27, 5.31, and 5.35), i t will be found that % satisfies a n equation
almost exactly like Eq. 5.9, but with the feed concentration z e r o instead of NFo- Note,
however, that in the solution to b e given for N3 and T2 we shall not assume that c,,(s,t),
etc., and hence q , equals zero.
-
2 T h e time average of x(s,t) will be written a s X(s) instead of the cumbersome x(s,t).
The average of the product xl(s,t)x,(s,t) will, however, be written a s x,(s,t)x,(s,t).
92 THE THEORY O F ISOTOPE SEPARATION THE CONTROLPROBLEM 93

It will be recalled that (Eq. 5.17) - These linear relations will be shown, but the process of eliminating
N2(S) and N,(B) and solving for the % will not be undertaken here.
In the second equation of Eq. 5.32,

for the rectifier. J' has been used to denote the corresponding func-
tion for the stripper.

[ lS
J' (s) = exp -
2) 1 ~ l O (-
l - Wo dS
o r else N p (t) increases o r decreases indefinitely. Thus on averaging
i t is found that

It now remains only to evaluate the a s yet unknown constants T2(S)

and T(B) from the boundary conditions, and the problem is solved.
From the boundary conditions of Eq. 5.28, at s = 0, on averaging, The f i r s t equation of Eq. 5.32 gives

where A i s a known constant. Also

Eliminating Np2 between Eqs. 5.53 and 5.54 gives N2(S) in t e r m s of
-
N2(0+) = N2(0-) T , (S), a s required. Similarly, from the boundary condition (Eq. 5.36)
at s = B,
Now J(0) = Jr(0) = 1. Thus with the aid of Eq. 5.51,

(JN2Io = (J'N,).

Setting s = S in Eq. 5.47 and s = B in Eq. 5.48 and subtracting,

For many purposes Tz(S), which is the average (second-order) change

It should be understood that J in the last integral is given by Eq. 5.49
from B to 0 and by Eq. 5.17 from 0 to S.
Now the boundary conditions (Eq. 5.32) a t the top of the rectifier
give a linear relation between N2(S) and T2(S); those a t the bottom of
the stripper (Eq. 5.36) give another between N,(B) and T,(B). Using
these to eliminate &(s) and N 2 ( ~from
) Eq. 5.52, a linear relation
between ~ ~ ( and ) obtained, which, together with Eq. 5.52,
8 ) T ~ ( Bis
allows the determination of both quantities.
in production of desired material, is a sufficient answer, and it is un-
necessary to return to Eqs. 5.47 and 5.48 to find N2(s).
The solution just given requires the averaging of some second-
order quantities, and of products of first-order quantities. If these
functions a r e known explicitly in t e r m s of t, the averaging is done by
elementary integration. However, they will usually be given a s La-
place transforms. To perform the averaging without taking the in-
verse the following theorem: which holds if a steady state exists, can
be used: If lim px* = C(s) as p 0 by positive r e a l values, then X(s) =
->
C(s); otherwise ~ ( s =) 0.
'G. Doetsch, "Theorie und Anwendung der Laplace Transformation," p. 193, Dover
Publications, New York, 1943.
 THE THEORY O F ISOTOPE SEPARATION THE CONTROL PROBLEM 95

The boundary conditional the top is still Eq. 5.29 and i t s descendants,
3 . VARIATIONS IN PRODUCTION RATE FOR A SQUARE CASCADE notably Eq. 5.43.
The solution of the zero-order differential equation (Eq. 5.9) con-
In this section the theory of the preceding sections will be illus-
tains nothing new. As in Chap. 2, Sec. 1, it is found that
trated and clarified by applying it to a special case. The abstract-
ness of the demonstrations just given will be relieved by discussing
further properties of the solutions with a particular example. The
example chosen i s that of a square cascade-practical comparison
may be made with a fractionating column-on which a r e imposed a r -
bitrary variations in P , the rate of production. Suppose that the hydro- The first-order differential equation (Eq. 5.16) requires the evalu-
dynamics of the cascade a r e such that a t the same time everything else ation of QQ:/QS. From Eq. 5.14, in the present c a s e
is invariant: the holdups, the flows and separation factors, and the feed
rate and composition. To simplify the arithmetic a rectifier alone
will be considered, with the feed concentration held fixed. It is fur-
ther supposed that N i s always small. A reservoir i s included at
the top.
The t e r m s cm, ego, and ego do not vary with stage number. Let
and Eq. 5.16 becomes
cc0(P) = cHO(flow into reservoir = flow between stages). Introducing
the s a m e notation a s in Chap. 3,

where

The general solution of Eq. 5.62 i s

N: = eCts(Acosh Bs CP*
+ B sinh /3s) - Ã e2~Is
‘Ñ (5.64)
P
The new constant e' is used to condense the notation. The expansions As in Chap. 3,
of the coefficients reduce to

The constants A and B a r e evaluated from the boundary conditions

Everything else (except of course N and r ) has only zero-order terms. (Eqs. 5.43 and 5.59). The latter in the present example reduces to
In place of the boundary conditions a t s = 0 and s = B previously used,
there is the simple expression

N(0,t) = Np = constant
where
an.d hence

Performing the indicated operations and collecting,

96 THE THEORY O F ISOTOPE SEPARATION THE CONTROL PROBLEM 97

(This is a particular example of the general result proved in Sec. 2.)

N,'(S,~)= "*
P
[
ec,S I M - c t ) sinh ;3(S- s) + 13 cosh ^(S
(M- el) sinh 8s + 8 cosh (3s
- s) To get the average loss of production, the second-order t e r m s a r e
thus necessary.
Equation 5.52 of Sec. 2 reduces now t o
+ eels
~ e sinh " 0s ~
(M - ct) sinh (3s + 13 cosh (3s -
1
a 2 c ~

and the boundary condition which determines Tn i s , from Eqs. 5.53

and 5.54,
If the disturbance is taken to be sinusoidal, i.e.,

P, = cos u t and
Now from Eq. 5.17,
and then Eq. 5.68 is inverted, contributions a r e obtained from the
poles a t 2 iw, also from the roots of (M - e ' ) sinh )3S + )3 cosh BS = 0,
J = exp (- f ' 2 ~ cis) = exp (-2ets) (5.75)
which lie on the negative r e a l axis. There is no pole at p = 0. Thus
Combining Eqs. 5.73, 5.74, and 5.75 and simplifying the result with
the aid of Eq. 5.60, then

i.e., t e r m s of the same period a s the disturbance, plus transients.

This is a general result. Note also that the form of Eq. 5.68, which
is linear in P r , allows superposition of solutions, i.e., a Fourier Accordingly it i s necessary to evaluate the expressions on the
analysis of the variations in P , c , c , etc., is permissible. right-hand side of Eq. 5.76. From Eq. 5.20, Q2 = NIPl,
In what follows a type disturbance more complicated than the si-
nusoidal will be discussed. Suppose that the product is withdrawn in
batches a t regular intervals a , instead of continuously. Then the dis-
turbance function is

P I (t) = - Po + pulses including a quantity aPo a t t = 0, a, 2a. .. (5 .TI)

The transform of Pl i s easier to write This integral i s equal1 to the sum of the residues a t the poles of
..
P^(q). These a r e a t q = 2k3ri/a (k = 51, ± .). Each residue has the
value

PF(p) has no pole a t p = 0. The average of N,(s,t) i s found from Eqs.

5.68 and 5.72:
'Strictly speaking
-
Nl (s) = lim pN;(s,p)
P- 0
= 0 Q: = lirn
v-m
+ 9
k=-v
(residues)
98 THE THEORY O F ISOTOPE SEPARATION THE CONTROL PROBLEM 99

Then In the same way the effect of disturbances of the same frequency
which a r e out of phase by an angle a can be realized. Corresponding
to Eq. 5.70, Nl is

and likewise, using the relation (Eq. 5.41) for NG,,

-
and on calculating NIPl (or similar t e r m s ) there will result the con-
tributions of the separate disturbances plus a cross-product t e r m
with cos a a s a factor. Now if the two disturbances a r e unrelated
cos a i s , on the average, zero, and the resultant is purely additive.
Note, however, that fluctuations in P and, for example, c5 caused by
The meaning of Eq. 5.78 will be clear by returning to the disturb-
a common primary disturbance a r e correlated through the hydrody-
ance Pl = cos u t and following through the s a m e procedure. From Eq.
namic response of the plant and cannot be treated separately.
5.70 and .On = NIP,, this time averaging by direct integration with The r e m a r k s of the preceding two paragraphs contain the results
respect t o t, necessary f o r the composition of disturbances of different sources
and types.
c
F(s,iu)eiwt - -F(s, -iu)e-iwt] cos u t dt Returning t o the derivation of i-according to Eq. 5.76 and picking
2iu up the thread of the argument a t the expressions Eqs. 5.78 and 5.80,
consider ( l / i u ) \ ~ ( s , i u ) - F(s, -iu)] a s a function of u. T e r m F(s,p)
is defined by Eq. 5.68. The t e r m e 2 " ~does not depend on u and hence
gives nothing. Then

Equation 5.80 is the contribution to D2(s) of a sinusoidal variation F = (M- e') sinh (3(S - s) + (3 cosh (3(S - s) + sinh (3s (5.81)
of frequency w/27r. Now, if the (improper) disturbance function (Eq. (M - e') sinh /3S + (3 cosh (3s
5.71) is analyzed, representing batch withdrawal a s
where

2 e @ 5 iuky
M =
l Âiuky
and if contributions of the individual frequencies given by Eq. 5.80
a r e summed, the result is precisely Eq. 5.78. F o r C.A à e2(l + i/È)2 a condition fulfilled by all but the lowest f r e -
This is a general result, which can be understood a s follows: In a quencies because of the smallness of e ,
disturbance, two different frequencies (in Ply for example) contribute
independently to N , each giving a t e r m of its own frequency. In taking (3- Vd>x(i±i
time averages, t e r m s like e^itlu:'t always vanish. The result is a sum
of contributions from each frequency acting as if it were alone. This
whole process i s taken c a r e of automatically by the use of the aver-
sinh (3(S- s ) - cosh'(3(~ s) - %e^
- (s-s)(W

aging theorems1 F(s, 5 iu) -. e-sV^(l?i)

Then
'In comparing the solution of thedifferential equation by the functionally closely r e -
lated methods of the Laplace transform o r the Fourier expansion, i t should be r e - 1 2
-[F(s,iu)-F(~,-iu)]=--e-~^ sinsYt^X" (5.82)
marked that the former method gives a t the same time the solution for the transients. lu Ill
100 THE THEORY O F ISOTOPE SEPARATION THE CONTROLPROBLEM 101

In case B, T2 is already quite large; in fact it may be suspected that

higher-order t e r m s a r e needed. In case A, Tn becomes large if a i s
taken much larger, even though ae2/^ is still small. Thus in both
cases T becomes appreciable a t a = -I/&, although pi does not ap-
pear in the formula (Eq. 5.85). This can be understood after an ex-
The contribution to T J ' P ~ N ~from the disturbance Pi = cos u t is, by
amination of the transient t e r m s in N, (s,t). The f i r s t transient has
Eq. 5.76,
the factor

(5.84)

This factor increases rapidly for fixed pl a s apl becomes large. For
The integral was evaluated by Eqs. 5.80 and 5.83 and by the definitior-
example, taking pl = -1: for a = 0, f = 0; for a = 0.1, f = 0.048; for
of C , Eq. 5.63. The contribution from the term Pl (t)Np/Pg vanishes
a = 1, f = 0.42; for a = 10, f = 100. Thus the transients get large a s
in the same manner a s e - s G / u and can b e neglected.
The disturbance in Eq. 5.71 caused by batch withdrawal gives like-
-
a -l/pl, implying that the new steady state differs significantly
from the initial state. Holdup a t the top increases -l/pl and must
wise, comparing Eq. 5.78 with Eq. 5.80,
therefore permit larger values of a , even though this is not exhibited
by the rough formula, Eq. 5.85.

Table 5.1 -Production Loss Due to Batch Withdrawal

Case A Case B

Equilibrium enrichment 2-fold 30-fold

(no production)
2tS 0.693 3.40
provided a <s: 2rA/e2(l + fl. The holdup in the reservoir has no in- k/e2 40 days 1 day
fluence to this degree of approximation. For larger values of a, nu- First transient period:'
merical summations a r e needed.
The factors in the solution (Eq. 5.85) a r e a t e r m in the ratio of a to
the natural period, X/e2; a factor <//' that i s related to the s i z e of the 1 day 1 day
disturbance; and a factor NpO/Np that shows that the sensitivity is -3.5 % -29 %
proportional to the enrichment factor of the column. It is interesting
to note that the period X/e2 and not -1/p,, the relaxation time, enters 'Taking the holdup in the top reservoir a s zero.
in; also that the magnitude of disturbance Po enters in a s I+!J = Po/%
and not P / C ~ ~An. idea of the o r d e r s of magnitude involved is given
in Table 5.1, which compares two cases. The evaluation of the loss in production by the single term T2 will
In both cases a is about equal to the time f o r the column to reach b e correct only if the higher-order t e r m s a r e negligible, i.e., only
the enrichment N ~ ~ / NHowever,~ . in case A, ae2/A = 1/40, whereas if T i s quite small. This slightly discouraging conclusion does not
in case B ae2/X = 1, s o that the loss in case B is much more serious. deprive the calculation of i t s usefulness, which is mainly to assign
The second most important factor is the severity of the disturbance: permissible upper limits t o the disturbances.
I+!J = 2 in case A and t+b = 0.2 in case B. The weight of this factor i s ir- It i s difficult to generalize from the problems just solved to other
the other direction. By comparison the differences in the factor problems. Even the most interesting result, that the effect of high-
N ~ / Na r ~e slight. frequency disturbances vanishes a s w-%, is only true for a disturb-
102 THE THEORY OF ISOTOPE SEPARATION

ance in P alone. It does not subsist when fluctuations in the coeffi-

cients c6, c5, and c, a r e taken into consideration. The correlations
to be expected between disturbances of the various coefficients a r e
different from one separating apparatus to another. Hence this is
about a s far a s the investigation can proceed without specifying the
type of equipment in question.
The reader may have noted that, leaving out the explanations and
digressions, the solution to the problem of this section was obtained
Chapter 6
fairly easily. This is not generally true; problems on more intricate
cascade shapes and for intermediate frequencies run into heavy com-
CENTRIFUGES
putation. However, it should be remembered, in spite of the space
required to expose the theory, that small correction terms a r e in-
volved. It is therefore quite proper to reduce the numerical work
by rather drastic approximations. 1. HISTORICAL INTRODUCTION

The effect of the earth's gravitational field on the composition of

REFERENCES
Fuchs, K., British MS 71, January -November, 1943.
Fuchs, K., British Supply Council reports: MSN 4, Feb. 11, 1944; MSN 5, Mar. 4,1944;
MSN 8, Mar. 7 and Mar. 17, 1944; MSN 10, Apr. 6, 1944; MSN 12, June 6, 1944; MSN
14, June 28, 1944; MSN 15, July 11, 1944; MSN 17, July 20, 1944.
the gases in the atmosphere at various heights has long been recog-
nized. For, if pi is the partial pressure of a gas of molecular weight
Mi, according to the Boltzmann distribution law
pi(h) = pi(0) exp -
where h is the height; g is the gravitational constant; R is the gas
constant, 8.314 X lo7 ergs/OK/mole; and T is the absolute tempera-
ture. Then for two gases of different molecular weights M, and M,

The suggestion that gravitational o r centrifugal fields a r e suitable

for isotope separation was made by Lindemann and Astonl in 1919.
In fact, they proposed that samples of neon taken at ground level and
at 100,000 ft would show a sensible shift in the isotopic abundance
ratio.
In a centrifuge, vastly more powerful fields a r e available. The
energy per mole at radius r in a centrifuge rotating with angular ve-
locity u is%Mi(ur),. This expression for the energy replaces -Migh
in Eq. 6.1, giving
Mi (u r)2
pi(r) = pi@) exp 2RT
104 THE THEORY O F ISOTOPE SEPARATION CENTRIFUGES 105

Between the periphery r = r, and axis r = 0, for two different gas ing the spinning of the rotor, vapor is removed slowly through a shaft
from Eq. 6.3, along the axis. In this way a Rayleigh distillation (Appendix D) of the
liquid i s accomplished, with simple-process factor given by Eq. 6.31.
Pl(r2) _ PI(O) exp (M, - M2) ( u r 2 ) 2 An increased concentration change in the residual liquid i s thus ob-
~2b-2) ~ ~ ( 0 ) 2RT tained.
The development by Beams1 of vacuum-chamber centrifuges, which
where w r , i s the velocity of the inside wall of the rotor, the were f r e e from vibration and thermally isolated to eliminate convec-
peripheral velocity. tion currents, finally permitted successful centrifugal isotope s e p a r a -
In a binary system, if N is the mole fraction of the isotope ha tion to be obtained in 1939. Beams and Skarstrom2 a t the University
m a s s Mi, of Virginia, using the evaporative centrifuge method on CC14,reported
Pi + P2 = P a 13 p e r cent change in the C135/C137ratio. Shortly thereafter Hum-
p h r e y ~ ; using the s a m e technique on ethyl bromide, altered the
N = Pl/P ~ r ~ ' / ~ r "
abundance ratio by 11 p e r cent.
1- N = p2/p The simple-process factor of centrifugal separation depends only
on the m a s s difference. Unlike other methods (e.g., diffusion through
and p - p(r) i s the total pressure. Then Eq. 6.4 becomes
b a r r i e r s , where a v a r i e s a s AM/M) it i s no more difficult for heavy
elements than for light. In addition, it is possible to calculate a p r i o r i
(Ml - 1%) (urn)'
the separation to be expected from centrifugal apparatus without any
= (&)o exp 2RT
undetermined constants o r theoretical uncertainties.
The simple-process "flow-through centrifuge," a s the concurrent
It is usual to take M, > Ml, and accordingly flow type was called, i s a n application to gaseous flow of the continu-
ous c r e a m separator. As originally conceived, a single s t r e a m of
(Ma - MI) ( u r J 2 g a s e n t e r s one end of a rotor through a hollow shaft, and two s t r e a m s
a n = exp a r e taken off the other end, one from the periphery and the other near
2RT
the axis. This method produces a s m a l l change in concentration per
so that Eq. 6.6 takes the form machine.
The countercurrent flow type was designed to attain considerable
R(0) = a o R ( r 2 ) separation in a single centrifuge, thus reducing the number of stages
required and the amount of material circulated between stages. As
where a ni s the (equilibrium) simple-process factor of the separat originally proposed by u r e y , circulation was to be established by
The value of the simple-process factor for a peripheral velocity continuous distillation of UFs from the bottom cap of the rotor. The
300 meters/sec, T = 300°K and a m a s s difference of one unit i s , vapor was to be condensed on the top cap. The heavy liquid would
Eq. 6.6, a. = 1.0183. then be forced out to the periphery and would flow down the walls to
Shortly after the suggestion of Lindemann and Aston was made, the bottom cap, countercurrent to the vapor flow, and complete the
tempts were made, in rather crude apparatus, by Joly and Poole1 cycle. Bramley and Brewer5 and Martin and Kuhn6 in Germany pro-
by Mulliken2 to realize this separation factor, but these were uns posed to establish countercurrent flow by thermal convection.
cessful. The idea of the evaporative centrifuge was introduced at
same time by Mulliken. In this method a small amount of liquid
' ~ e v s .Modern Phys., 10: 245 (1938); Rev. Sci. Instruments, 9: 413 (1938).
introduced into the centrifuge, forming a layer a t the periphery. Du- 'Phys. Rev., 56: 266 (1939).
3 ~ h y s Rev.,
. 56: 684 (1939).
Reports on P r o g r e s s in Physics, VI: 72 (1939).
Phil. Mag., 39:,, 372 (1920). 'Science, 92: 427 (1940).
'J. Am. Chem. Soc., 44: 1033 (1922). '2. physik. Chem., A189: 219 (1940).
106 THE THEORY OF ISOTOPE SEPARATION
The aim of this chapter is to expose the theory of centrifugal sepa-
ration for various types of centrifuges. The important question of
mechanical design will be dealt with only in so far a s it is essential
to the understanding of the process design. The partial differential
equation of centrifugal separation is derived in Sec. 2. In Sec. 3 the
theoretical maximum separative power of a centrifuge is found. The
solution for the evaporative centrifuge i s given in Sec. 4, that for
the concurrent centrifuge in Sec. 5, and that for the countercurrent
centrifuge in Sec. 6. Section 6 investigates at length the properties of
the countercurrent centrifuge.
The writer was introduced to the a r t of centrifugation by C. Skar-
strom, formerly of the University of Virginia and later with the Stand-
a r d Oil Development Company. Harold C. Urey participated actively
in the early work, until his other responsibilities became too heavy,
contributing ideas and physical interpretations of mathematical re-
sults. Irving Kaplan shared in all the developments subsequent to
December 1941, notably the work on cascades of centrifuges (in-
corporated largely in Chaps. 1 to 5) and the design problem.
2. PARTIAL DIFFERENTIAL EQUATION OF CENTRIFUGE
As a preliminary to the derivation of the differential equation, con-
sider for a moment a centrifuge containing a single pure gas. The
gas is supposed to rotate uniformly at the same speed a s the rotor,
i.e., w radians/sec; to have no other motions; and to be isothermal.
Choose a cylindrical coordinate system, rotating with the gas, with
the origin at the axes of rotation. The effect of the rotation is to set
up a centrifugal force w 2 r per unit mass. Under the influence of this
force, a pressure gradient is established according to the hydrody-
namic equation
Here p i s the density of the gas mixture in grams per cubic centime-
ter. The quantities p and p a r e related by the gas law
Combining Eqs. 6.8 and 6.9 gives Eq. 6.3 of its equivalent
Mw2r2
~ ( r =) ~ ( 0exp
) (6.10)
CENTRIFUGES 107
The density distribution (Eq. 6.10), considered from a molecular
point of view, is a dynamic equilibrium between the effect of random
motion of the molecules (diffusion), which tends to erase density
gradients, and the effect of the centrifugal field, which tries to pile all
the molecules at the periphery. The diffusion counter to the density
gradient gives a current density (in grams per second per square
centimeter)
Here D is the coefficient of self-diffusion. The outward current per
square centimeter caused by the centrifugal force, which balances
this inward current, must therefore be
In mixtures, Eq. 6.11 holds for each component separately
Di i s the diffusion coefficient for molecules of type i through the whole
mixture. For isotopic mixtures the exact meaning of the diffusion
coefficient is of no significance, and henceforth the subscript will be
dropped. Equation6.12 i s the desired expression for the flux of mole-
cules caused by a centrifugal field, and it i s now possible to derive
the differential equation.
Consider then a centrifuge containing a perfect binary gas mixture,
isothermal and rotating uniformly. In order to accomplish separa-
tion, the gas i s made to flow radially and axially. In t e r m s of the
cylindrical coordinate system chosen a s before, the motion of the gas
a s a whole (drift velocity) is given by the components
(Radial) u = u(r,z)
(Azimuthal) v =0
(Axial) w = w(r,z)
Because of this motion of the gas, the equilibrium distribution (Eq.
6.3 o r 6.10) i s continually upset. The system moves material in an
effort to reestablish equilibrium; in this way a transport of the de-
sired material i s set up.

Imagine a small annular region in the rotor, bounded by the cylin-
d e r s r = roand r = r o+ 6r and by the planes z = z0 and z = z0 + 6z. In
a short time interval 6t t h e r e flows a c r o s s the cylindrical face r = rn
an amount of desired isotope (denoted by the subscript 1).
In this equation p,u is the flow caused by the drift velocity, p$cl i s
the flow impressed by the centrifugal field, -D ap1/ar i s the con-
tribution f r o m diffusion against the density gradient, and 2 r r O6z i s
the a r e a of the cylindrical face. Likewise the flow a c r o s s the plane
face z = z0 i s
The change in content of desired material in the volume element i s
therefore
Introducing Eqs. 6.13 and 6.14 into 6.15 and making use of the rela-
tion of Eq. 6.12 for u C (dropping the subscript O),
Now
and
Substituting Eq. 6.17 in Eq. 6.16, using Eq. 6.18 and noticing that Dp
=constant, it i s found that
CENTRIFUGES
which i s the desired partial differential equation of centrifugal sepa-
ration.
The three kinds of centrifuge described in Sec. 1 a r e governed by
special c a s e s of Eq. 6.19. In the evaporative centrifuge w = 0 and
r u = constant; in the concurrent and countercurrent centrifuges u = 0,
and w describes the respective types of flow. The solution of Eq.
6.19 for these c a s e s i s the subject of the subsequent sections devoted
to the particular centrifuge types.
3 . MAXIMUM SEPARATIVE POWER OF A CENTRIFUGE
In this section the maximum separative work that can be performed
by a centrifuge1 will be determined. It will be used a s a standard to
evaluate the efficiency of the various centrifuge designs.
Using the result of Sec. 6 (Eq. 1.60),
The net transport of desired material (in moles per second) a c r o s s a
surface element do is, according to the analysis given in a preceding
section,
where To i s a unit vector along a radius through do, TIcl i s the vector
velocity created by the centrifugal field directed along F0 and with
magnitude given by Eq. 6.12, and the symbol v is the vector gradient
operator.
Thus the separative power for a s m a l l slab of thickness ds, with
f a c e s do normal to VN and sides parallel to VN, i s
Following Dirac.
110 THE THEORY O F ISOTOPE SEPARATION CENTRIFUGES Ill

The symbol indicates the scalar product. dependent of the radius. The radius affects the holdup, and in counter -
Taking VN a s a variable and do d s and everything else constant. current centrifuges the flow into the rotors, but is otherwise irrele-
Eq. 6.21 i s a maximum when vant. It is usually chosen by mechanical considerations.

that is, along the direction of the centrifugal field, and one-half the
equilibrium value. Then

SURFACE do- \
\
\ A
Integrating over the whole centrifuge, of length Z and radius r2, \I ,/

Note that the expression in brackets is the exponent of an (Eq. 6.7).

The minimum number of centrifuges of length Z and peripheral
velocity w r 2 required to perform a given separation is, by Eqs. 1.44
and 6.23, AU/(6U)rnax. The value of (i5U)max depends on the peripheral
velocity of the fourth power, which puts a premium on high periph-
Fig. 6.1-Stresses in a thin rotating cylinder. (a) S t r e s s e s in a wall sector. (b)Vector
e r a l velocities. The separative power is also proportional to the diagram.
length. But for constant peripheral velocity1 the separation is in-
4. THE EVAPORATIVE CENTRIFUGE
'The peripheral velocity-not the angular velocity-is determined by the tensile In the evaporative centrifuge N is independent of z, and w = 0. Con-
strength of the rotor wall, s o that it i s the fundamental characteristic of the machine. sidering the steady state, a ~ at/ = 0 and Eq. 6.19 reduces to
The demonstration follows: Consider the equilibrium of forces acting on a sector of
the rotor wall, subtending a small angle 4. Then the triangle of f o r c e s gives (Fig. 6.1;
<t>
r + d u u 2 r = 2 T d u s i n - = T dff 4
2 *
where

where T is the s t r e s s in the wall and p i s its density. The rotor will break when T =
p d r 2 i s greater than the tensile strength of a metal.
112 THE THEORY O F ISOTOPE SEPARATION CENTRIFUGES 113

Suppose vapor i s removed from the axis a t a r a t e L moles/sec. ducti0n.l If it i s assumed that the production i s only a small fraction
Then of the charge, o r that successive batches of product a r e kept separate,
-JL2mzu = L (6.26j the separative power of the centrifuge i s (cf. Appendix D)
RT

Multiplying Eq. 6.24 by the constant 2 i i Z / R ~and integrating, 5U = -2L ( ($1

a n y+L - 1 ) 2 = ~ L 2( ~~ )L+2 ( l n a o ) 2(6.32)

y r
3N
+ 2flAr2 N ( l - N)
I = L M(0) - L N(r) (6.27~ The value of 6U is a maximum when L = 2 ~ :

Y u z
where ( 5 ~ = -(ln ~ =~Dp- (ln
) ~ a0)2 - (6.33)
4 RT 2
DP
y = 27rZ - (moles/sec) (6.28) which i s the same a s the theoretical maximum separative power of a
RT
centrifuge (Eq. 6.23).
In the development just given the gas was assumed to be isothermal.
The integration constant was evaluated by noting that the left side of This picture i s f a r too simple. When vapor i s withdrawn a t the axis,
Eq. 6.27 vanishes a t r = 0. the gas cools by expansion and a temperature gradient i s induced.
Replacing N by R a s dependent variable, The gradient i s determined by the balance between the r a t e of cooling
of the g a s and thermal conductivity. As L increases, the gradient
+ 2AAr2 R ) [
= L R(0) - R(r) ] [ + K%]
1 È6s29
increases until the gas reaches the dew point o r the adiabatic g r a -
dient. The mean effective temperature Teff i s therefore reduced a s a
function of L.
An attempt may be made to account f o r this effect by writing in
Now - R(0) s - AAr2 N(O), which can be neglected with respect to
1 + R(0) place of Eq. 6.31
unity. After thus simplifying Eq. 6.29, the solution is readily found to
be
fc^) T 2

I)&(
R = R(0) exp - A A r 2
Since T / T ~increases
~ ~
R(0) = R ( r 2 ) a o (6.34)

with L, the separation will not fall off quite so

and on introducing Eq. 6.7, which in t e r m s of AA i s
a. = exp [ + A A ~ ; ]
this expression may be written
R(0) = R(r2)ffo (&)
fast with L in practice a s Eq. 6.31 would indicate.
A unit weight of gas rotating near the periphery with angular ve-
locity w has the angular momentum r & ; at the axis it has none. When
(6.30) g a s i s moved from t h e periphery toward the axis, means must be
provided to absorb this excess angular momentum, which otherwise
tends to increase the angular velocity of the gas a t smaller radii.
This i s accomplished by radial baffles2 (looking down along the axis,
a s t a r ) which prevent the g a s from rotating a t angular velocities dif -
(6.31) ferent from that of the rotor.

Equation 6.31 gives the relation for mole fraction of product N(0) 'This result was found by Humphreys, Phys. Rev., 56: 684 (1939).
in t e r m s of the composition of the charge N(r2) and the r a t e of pro- Yearns and Skarstrom, Phys. Rev., 56: 266 (1939).
114 THE THEORY O F ISOTOPE SEPARATION CENTRIFUGES

Although the evaporative centrifuge, with L = 2y, delivers 100 per

cent of the theoretical separative power, it is not easily adapted to
continuous operation.

5. THE CONCURRENT CENTRIFUGE

Like the evaporative centrifuge, the concurrent centrifuge is a

simple-process machine since the enrichment factor per machine is
limited to a,,.
In this device axial flow i s used to obtain an axial a s well a s a r a -
dial concentration gradient. The gas enters (Fig. 6.2) in two streams
a t one end of the rotor and flows axially to the other end, where the
s t r e a m s a r e removed separately. During the passage through the
centrifuge the s t r e a m s tend to assume the (radial) equilibrium dis-
tribution (Eq. 6.6). The flow pattern adopted for the concurrent cen-
trifuge consists of two thin cylindrical s t r e a m s located a t radii r,
and r (periphery) and flowing parallel to the z axis. (The optimum
position for the inner stream i s not a t the axis, a s will be seen sub-
sequently.) There i s no radial m a s s flow.
A rotor with a double entrance (so that entering streams of two
WALL
different concentrations can be used) will be considered; previously
the literature has covered only single-entrance machines. The double-
entrance centrifuge i s more flexible than the latter. Cascades of
these elements, which a r e more general than the element .of Chaps. 1
to 5, a r e studied in detail in Appendix E. Among the more interesting
features of this study i s the result that the no-mixing case (both
entering s t r e a m s of the same composition) i s not an optimum.
The solution given here follows roughly the demonstration by Mur-
phree, which for the case considered i s superior to the more general
method that was originally used. The value of u i s zero, and the flows
a r e generally so large that back-diffusion is negligible. Then Eq. 6.19
reduces for the steady state to

In the region between r = rl and r = r2,w = 0, s o that Eq. 6.35 inte-

grates to

Fig. 6.2-Simple model of a gaseous concurrent centrifuge.

116 THE THEORY OF ISOTOPE SEPARATION CENTRIFUGES 117

where f (z) is the transport of desired isotope a c r o s s a cylindrical The separative power when N(r2,0) and N(rl,O) a r e most favorably
element of unit length between the two streams. related1 is, according to Eq. 11 of Appendix E,
But also, if L(rJ i s the flow in the stream at r = r2,

and the separative power per unit length is

and

L(r2) N(r2,z) + L(rl) N(rl,z) = L(r2) N(r2,0) + L(rl) N(rl,O)

(6.38)
= B (a constant)

Integrating Eq. 6.36 between rl and r2, considering AA N(l - N) which i s a maximum with respect to rl/r2 for any value of bz when
constant to t e r m s of the order of (AA)', rl/r2 = 0.534,

and substituting for f (2) from Eq. 6.37 and for N(r2,z) from Eq. 6.38, and a relative maximum with respect to bz when bz = 0, namely,

(F)
max
= 0 . ~ 0 7 2 4 7DP
RT
r-(~nit~)~

Integrating this ordinary differential equation for N(rl,z) and r e -
arranging the result,
If this result is compared with Eq. 6.23, it is seen that this i s 81.45
per cent of the theoretical efficiency.
For a single entrance centrifuge, it is necessary to take N(r2,0)
= N(rl,O). The separative power is then, according to Eq. 8 of Appen-
dix E,

and per unit length

where
118 THE THEORY OF ISOTOPE SEPARATION CENTRIFUGES

The maximum with respect to rl/r2 is the same a s before. The ex-
6. THE COUNTERCURRENT CENTRIFUGE
pressionl (1 - e-^)'/bz has a true maximum a t bz = 1.25643 and thus
Countercurrent separating devices in general have the property of
(y) max
=
27rDp
(0.40724)' -(In aJ2
KT
multiplying the equilibrium simple-process factor many times in one
unit. Because of the nature of the flow, a concentration gradient whose
size i s limited only by back-diffusion i s established in the direction
of the flow.
which has an efficiency of 66.34 per cent. The condition bz = 1.25643
Since a large separation i s obtained in one unit, much of the recy-
gives a definite relation between the length of the rotor and the flow,
cling between units which is otherwise necessary to multiply the sim-
namely,
ple-process factor i s avoided. Furthermore, the problem of cascade
operation i s very considerably simplified since the number of units
in s e r i e s required to effectuate a given fractionation decreases enor-
mously. The cascade becomes broad instead of long and i s easily
broken down into independent parallel sections. Countercurrent flow
If it i s remembered that the separative power i s the product of the
i s also efficient from a process standpoint because it i s possible to
concentration difference and the transport, the maximum of 5U with
maintain maximum separative power through the entire unit.' F o r all
respect to rl/r2 is easily understandable. The concentration differ-
these reasons, which have long been appreciated by chemical engi-
ence is the maximum when r,/r2 = 0, but the transport (through the
neers, countercurrent processes a r e preferred if at all possible.
cylinder of radius zero) is zero. The transport i s a maximum when
The differential expression of Eq. 6.19 becomes for the counter-
r1/r = 1, but the concentration difference is zero. The maximum of
current centrifuge
the product thus lies somewhere in between.
A fuller discussion of the significance of the dependence of the r e -
sults on 1 - e-^, the degree of equilibrium attained, is given in Ap-
pendix E. The phenomenon i s a quite general one.
Notice that b contains the factor 0(1 - 0) (L, + LJ. Decreasing The solution of Eq. 6.45 may be found by following the standard meth-
0(1 - 0) and increasing L, + Ln leaves this factor unchanged, but small ods for solution of partial differential equations. This method was
total flows a r e desirable and 0(1 - 0) should obviously be made a s utilized in June and July of 1940. However, Eq. 6.45 i s particularly
large a s possible, which requires 0 = % (cf. Eq. 6.44). simple since the gradients of the concentration in all directions a r e
small. Thus the equation for a countercurrent centrifugal column can
Further properties2 of the concurrent centrifuges may be obtained
also be solved by the method used by Furry, Jones, and Onsager2 for
from the equations given.
the thermal-diffusion column. This possibility was first pointed out,
and partially worked out, by B r a m l e ~ .A~ similar procedure was
'Note that the correspondence e-b2 - </I$ makes this take the form employed by Martin and ~ u h n . ~
The results of the two methods a r e of course identical. The second
method will be demonstrated here because it i s easier and because it

'1n contrast to concurrent devices, in which aN/ar changes with z, s o that if the
separative power i s an optimum in one place it is necessarily l e s s in every other
'For example, for any r l / r 2 the dependence on z/(L1 + L2) i s the same, and the place.
maximum for the single entrance centrifuge is at t h e s a m e place, bz = 1.256 [with the v h y s . Rev., 55: 1083 (1939).
appropriate value of In (r2/rl) in b]. But for z/(Ll + LJ preassigned (and not bz) the 'Science, 92: 427 (1940).
maximum with respect to r 1 / r varies. Z . physik. Chem., A189: 219 (1940).
120 THE THEORY OF ISOTOPE SEPARATION CENTRIFUGES 121

permits the c a s e in which the mole fraction of the product i s large to

be included without added complications. ar o
rr pw r d r , which i s just a s useful for present purposes. Substi-

The method depends essentially on the fact that aN/ar i s of the tuting Eq. 6.46 in Eq. 6.49, it i s found that
o r d e r of N(l - N)2AAr, which i s a s m a l l quantity. Thus the variation
of N with respect t o r, a s compared to the variation of r, p, o r pw,
may be neglected. F o r example

CJ^L pw r d r - ?a-!z r pw r d r + t e r m s of the o r d e r of A A ~ where

Consequently Eq. 6.45 may be integrated with respect to r. The f i r s t

integration gives directly, dropping a2N/az2 and aN/at,which a r e of
second o r d e r ,

The integration constant ( a function of z and t) w a s evaluated by r e -

-
membering that r ( a N / a r ) 0 a s r 0. -
The second integration with respect to r r e q u i r e s a little more
artifice in o r d e r to get a form in which the integration constant can
be evaluated. The boundary conditions a r e
c6 =
277
/ 'Â¥
p r d r = holdup p e r unit length (moles/cm)

Equation 6.50 h a s the f o r m of Eq. 2.1, the typical equation f o r iso-

Equation 6.48 may be transformed by integrating by p a r t s the f i r s t
t e r m on the left-hand side,
tope separation in a square cascade o r column. The properties of
such columns, individually and in cascades, f o r fixed values of c,, c ,
and c6 were studied in Chaps. 2, 3, and 4. The following discussion
covers the investigation of how these coefficients vary a s a result of
changes in operating conditions. The material in the e a r l i e r chapters
will be r e f e r r e d to constantly.
In order to examine properties of the countercurrent centrifuge,
attention will be confined for the moment to a single centrifuge. F o r
convenience it will a l s o be assumed that N<1, although t h i s i s by no
means essential to the argument. Then the steady-state solution of
Eq. 6.50 i s (Eq. 2.14)

Equation 6.49, although it does not give the integration constant that
would permit the integration of a ~ / a ritself, tells how to integrate
122 THE THEORY O F ISOTOPE SEPARATION CENTRIFUGES 123

where and the maximum possible value of N(Z)/N(O) for a given length of
centrifuge and for a given type of countercurrent flow i s

-N(z)
- - exp 2cOz
NO)

where 2e0 is the maximum of c2 + a3L2 with respect to L and i s

The quantities cl and c5 depend on the function pw, which measures

the up-and-down flow, in two ways: They depend on the flow pattern,
o r the locations and relative size of the s t r e a m s , and on the absolute
magnitude of the flow. The pattern i s built into the rotor by the posi-
tion of orifices in the end caps and position (or absence) of baffles. The corresponding value of L i s
But the magnitude of the flow i s an operating variable and can be ad-
justed f r o m outside. This independence of flow rate and internal con-
struction in a properly conceived centrifuge gives it more flexibility
than the otherwise analogous thermal-diffusion column.
Therefore a new quantity L is introduced.

The existence of a maximum in e i s physically due to the phenom-

which is the flow up the column plus the flow down the column. Then
does not depend on L, and Eq. 6.52 becomes
enon of back-diffusion. The concentration gradient is proportional to
e . When L, the total flow in the centrifuge, is large and the concentra-
tion gradient i s small, back-diffusion i s negligible (c, < a,L2) and
e v a r i e s inversely a s L. But a s soon a s L is small and e and the con-
centration gradient a r e large, back-diffusion plays an important role,
finally limiting the attainable fractionation.
It will be found convenient to measure all flows in t e r m s of Lo.
Therefore s e t

Substituting in e,

where a, = c,/L and a 3 = C J L ~ do not vary with L.

N(z)/N(o) i s now a function of L and ip. F o r any value of L,
N(z)/N(o) is greatest when the rate of, production i s zero. Then
In t e r m s of the dimensionless parameters m and ib. P is given a s
-N(Z)
- - exp (c2 ' t L 2 z,
NO) P = m^P0 (6.61)
 *
124 THE THEORY O F ISOTOPE SEPARATION
CENTRIFUGES

where o h K., memorandum to H. C. Urey, May 14, 1940.

Cohen, K., and I. Kaplan, Report A-101, Jan. 28, 1942.
Cohen, K., and C. Skarstrom, Columbia S e r . No. 4R-X138, Feb. 6, 1941.
Dirac, P. A. M., British MS, 194'1.
Kaplan, I., and K. Cohen, Report A-195, J u n e 17, 1942.
Murphree, E. V., Separation of G a s e s by Diffusional Methods, February 1942.

Equations 6.57 to 6.62 constitute a complete description of the be-

havior of the countercurrent centrifuge. When the type of flow pw i s
specified, co, Lo, and Po a r e determined. L and P a r e the conditions
imposed and a r e measured on flowmeters by the experimenter. By
Eqs. 6.59 and 6.61, m and $ a r e therefore known. And finally, sub-
stituting co, m, and ~L,J in Eq. 6.60, the fractionation of the column i s
predictable.
The maximum separationfactor occurs a t L = Lo, but this i s clearly
not the most efficient circulation rate f r o m the standpoint of separa-
tive power, since the parasitic phenomenon of back-diffusion is very
important at this flow. The relative weight of back-diffusion i s re-
duced by increasing L (thus decreasing the concentration gradient).
These qualitative considerations a r e confirmed by examining SU.
From Eq. 2.33,l

whose largest value i s reached asymptotically a s m m. -

It should be noticed that the limiting value of 6U i s approached very
rapidly: When m = 3 the separative power i s 90 per cent, and when
m = 5 it i s 96per cent of c;Z/4c3. At these valuesof m the concentra-
tion gradient i s 60 and 38 p e r cent, respectively, of the r n a x i m ~ m . ~
Thus it i s not necessary to discard the advantages of large separa-
tion in one unit, o r to increase the flows inordinately, to get adequate
separation efficiency.

REFERENCES

Bramley, A., The Centrifugal Separator, Oct. 7, 1940.

Cohen, K., Report A-54, October 1940; Report A-52, Jan. 21, 1941; Report A-50, Jaz.
30, 1941.

C o n s i s t e n t with the e a r l i e r sectionsof t h i s chapter, the symbol 6U is being used k?

the total separative power, and not the separative power p e r unit length a s in Chap. Z
Sec. 3.
'The f a c t o r is 2m/(l + m2).
 OTHER SEPARATION METHODS
127
isotopic solutions a r e perfect solutions and because the simple-
process factor is nearly one; and (2) the inclusion of the equilibrium-
time phenomenon, which is generally omitted although it is more often
significant than is commonly susnpfted-
It i s clear that the two phases need- - -not
-
be liquid and gas; two liq-
uids, o r a liquid and a solid, would be equally suitable. The phase

Chapter 7

OTHER SEPARATION METHODS

1, TWO-PHASE SEPARATIONS

The classic case of a two-phase separation is fractional distillation,

one of the most widely used unit operations of chemical industry. The
earliest use of this process in isotope separation was the fractiona-
tion of Ne20 from liquid neon by Keesom. The method was later used
to r a t e deuterium from liquid hydrogen andin the fractionation of
H,0 to give 018(Urey e t al.).
The scope of this process was much enlarged by Urey's invention
of the chemical-exchange method. Here the driving force of the sepa-
ration is the chemical equilibrium constant of a reaction such a s

Such a reaction can be carried out entirely analogously to ordinary

fractional distillation (refer t o Fig. 7.1). In the example shown, HCN
gas is scrubbed countercurrently with a water solution of NaCN in a
packed tower, just a s vapor and liquid a r e contacted in fractional dis -
tillation. The condenser of the fractionating tower is replaced by
NaOH, which converts the gas phase to the liquid phase.
Although the theory of such processes was fairly well understood
before the war, an account of it is included here by reason of its prob-
able great future importance. Also, the generality of development of
cascade theory now allows the theory to be given in a clearer and
m e concise form. The theory has some points of resemblance with
the ordinary theory of fractional distillation. The two chief differ-
ences a r e (1)the closed form of the results, which is possiblebecause
126
Fig. 7.1 -A chemical-exchange tower.
in which enrichment of desired isotope takes place will be denoted
by small letters and the other phase by capitals. Thus n, 1 , h, and d
a r e , respectively, the mole fraction,flow in moles per second, holdup
in moles per centimeter, and diffusion coefficient in one phase; and
N, L, H, and D a r e the corresponding quantities for the other phase-
At equilibrium1
'A complete discussion of the theory of the equilibrium constant may be found in
3. Bigeleisen and M. G . Mayer, J. Chem. Phys., 15: 261 (1947).
 THE THEORY O F ISOTOPE SEPARATION OTHER SEPARATION METHODS

Eliminating ni and N, between Eqs. 7.5 and 7.6,

Countercurrent contacting (Fig. 7.1) will be considered first. The

direction of flow of the enriched phase will be taken a s positive, and
length along the column will be measured in this direction. Then if T
is the rate of transfer of desired isotope to the enriched phase across 1 B b
-=-+-
the interface per unit length of column, k DCu dcu

The rate of transfer T depends on two processes : the rate of at -

tainment of the chemical equilibrium and the rate of diffusion through
fluid boundary layers. If the equilibrium process is rate determining,
then T has the form of the rate constant for a bimolecular reversible
reaction

where c and C a r e the concentrations in moles per liter of the proc-

e s s material in two phases.
Usually the slow process is the diffusion through the boundary
layers. Referring to Fig. 7.2 and denoting the mole fractions a t the
interface by ni and Ni, and the a r e a of the interface per unit length of Fig. 7.2-Relations at a gas-liquid interface.
column by a , then

a relation that is quite familiar to chemical engineers. Equation 7.7

can be simplified by the assumption that the interfacial concentrations
do not differ appreciably f r o m the concentrations in the body of the
The differential coefficients may be replaced k y (Ni - N)/B and fluid, giving
(n - ni)/b, respectively, when B and b a r e the film thicknesses. Thus
T = - k [(n - N) - (a- 1)N(1 - n)]
= - k [n(l - N) - aN(1 - n)] (7.9)

which is the s a m e a s Eq. 7.3 but with a different meaning for k. It

At the same time ni and N, a r e related by Eq. 7.1, s o that should be noted that in deriving Eq. 7.9 it has not been assumed that,
'.a - 1) i s small.
Inserting Eq. 7.9 in Eq. 7.2,
 THE THEORY OF ISOTOPE SEPARATION OTHER SEPARATION METHODS 13 1

The solutionof these equations i n t h e steady state (aN/at = &/at = 0)

an an
h -
at
+ 1 -= - k
az [(n - N) - (a - 1)N(1 - n)] f o r ( a - 1) small has been treated at length in Appendix E, a s well a s
(7.10) the associated cascading problem.

2. THERMAL DIFFUSION

These two equations may now be combined, using the assumption that
( a - 1) i s very s m a l l (for the c a s e of large values of a , s e e Sec. 3).
Noting that the operator a/at i s proportional t o (a - 1)' and that 9/9z
i s of the o r d e r (a- 1).
The history and theory of the thermal-diffusion process has been
the subject of a number of recent papers which have been widely read.'
The treatment here, therefore, will be confined to a rapid review of
the theory to show how i t fits in with other separation methods in the
general scheme.
The equation f o r thermal diffusion corresponding to Eq. 6.45 that
was derived f o r a centrifuge is easily a r r i v e d a t by replacing

which is the form of the fundamental equation of isotope separation

(Eq. 2.1) with the correspondences

by the corresponding expression for thermal diffusion2

The theory of square cascades developed in Chap. 2 can thus be ap- where a is known a s the thermal-diffusion coefficient.
plied directly. Then, i n place of Eq. 6.45,
The exact solution of Eq. 7.10 in the steady state for values of
( a - 1) that a r e not small i s also of some interest. Eliminating N by
the relation In - LN = Pnp, defining $J = P / ( a - 111, (. = k(o1 - 1 ) / 2 ~ ,
the following is obtained:
where x is the distance from one wall in a thin annular space between
two concentric cylinders. It must be remembered that because of the
severe temperature drop a c r o s s the annulus pD varies with x.
Proceeding now precisely a s i n Sec. 6 of Chap. 6, it is found that
where A($) = {[I + $ + ( a - l)$np)2 - 4a$Jnp}% and no, nz, and n~ are Eq. 7.15 resolves to
the mole fractions a t the bottom of the column, the top of the column,
and the top of the cascade, respectively. Equation 7.13 is directly
analogous to Eq. 2.5, to which it reduces f o r small values of ( a - 1).
The equations for concurrent contacting a r e easily obtained f r o m
Eq. 7.10 by changing L to -L. This gives
*L. Waldmann, 2. Physik, 114: 53 (1939); H. Furry, R . C. Jones, and L. Onsager,
3N + L -
3N = k [ ( n - N ) - ( a - l ) N ( l - n ) I Phys. Rev., 55: 1083 (1939); P . Debye, Ann. Physik, 36: 284 (1939); K. Wirtz, Ann.
H- Physik, 36: 295 (1939); A Bramley, Science, 92: 477 (1940); J. Bardeen, Phys. Rev.,
at az 57: 35 (1940), 58: 94 (1940).
(7 .14)
'see S. Chapman and T. G.Cowling, "Mathematical Theory of Non-Uniform Gases,"
h -a+n a n = - k [(n - N ) - ( a - 1)N(1 - n ) l
1- p. 244, Cambridge University Press, London, 1939.
at az
132 THE THEORY O F ISOTOPE SEPARATION OTHER SEPARATION METHODS 133

where now which requires a catalyst. It was f i r s t proposed by Urey in late 1940
to use this p r o c e s s by countercurrent washing of water and hydrogen
in a tower packed with catalyst. At the time of writing, water must be
converted into hydrogen a t the product end of a cascade using the
process of Eq. 7.17.
The steady-state equations corresponding to Eq. 1 of Appendix E
a r e , from Eq. 6.14,l

an
I - = - k [(n - N) - (a - 1)N(1 - n) ]
az
Appreciable simplification occurs when N -s: 1; since this i s the major
part of the Dn problem (initial concentration = 1/7,000) this case will
be considered. Then
In Eq. 7.17, p is the density, r i s the radius of the annulus, and a is
the spacing a c r o s s the annular gap. The relations of Eqs. 7,16 and
7.17 hold f o r any fluid, liquid a s well a s gas.
Values of cl and c,; depend on the spacing, the temperature drop
a c r o s s the gap, and the intrinsic properties of the fluid.
The further theory of the thermal-diffusion column closely r e s e m - The solutions of Eq. 7.20 a r e
bles that of the centrifuge, except that in the present instance the
m a s s flow is caused by convection and i s dependent on the tempera-
t u r e drop and the spacing. Corresponding to ^>O, the flow that gives
the maximum fractionating factor in a centrifuge, there i s a charac-
t e r i s t i c width an for which the fractionation in a thermal-diffusion
column i s a maximum. However the maximum efficiency from the
standpoint of separative work per unit of heat loss i s at a spacing
other than a. which corresponds to the maximum separative effi-
ciency in a centrifuge f o r flows l a r g e r than &,. The cascade is shown in Fig. 7.3. Then f o r the (s + 1)th stage, the
following identifications can be made:
F r o m the standpoint of flow geometry, it i s easy to show, by the
method of Chap. 6, Sec. 6, that the model described has an absolute
efficiency of about 71 per cent and that the most efficient flow pattern
corresponds to flow immediately adjacent to the walls of the annulus, Equation 7.2 1 becomes
but with no flow in the middle.

3 , CONCENTRATION OF DEUTERIUM

A possible basis f o r the production of deuterium i s the exchange

reaction

HD + HgO= + HDO (7.18) 'Since (a- 1 ) i s large, we may no longer s e t N(l -

 OTHER SEPARATION METHODS 135
THE THEORY O F ISOTOPE SEPARATION

where

ft = 1 - e x p [- k(L ~ ~ 1= fraction
) 2 of equilibrium attained (7.23)

Now Eq. 7.22 can be combined with the conservation equation

to give three equations in the four quantities Ns+l, Ns, ns, and ns.i
which can be solved in t e r m s of any one of them. For example

The difference equation (Eq. 7.25) is easily solved to give

where

and No is the concentration of HDO in the feed. The number of sec-

tions in the cascade, a s a function of the initial and final concentra-
tion, is accordingly

m
KEY

(GAS) LIQUID
Since L - 1 = P, A and B (and hence S) a r e functions of P/L, the per-
centage withdrawal1 The volume of catalyst required can be obtained
by writing

The reciprocal, L/P, is known as the "reflux ratio."

136 THE THEORY O F ISOTOPE SEPARATION

where V i s the volume of catalyst p e r stage and k' i s the transfer co-
efficient per unit of catalyst. Then Crom Eq. 7.23,

and the total volume i s VS, a function of (3 and P . The total volume of
catalyst has i t s smallest value when 0,the fractional attainment of
equilibrium, i s zero. Without proof, this limiting volume i s taken to
be the same a s that f o r the operation of a countercurrent column and
corresponds to an infinite number of stages. The minimum number of APPENDIXES
stages o c c u r s for (3 = 1 and of course requires an infinite amount of
catalyst. Appendix A -Roots of a Transcendental Equation
It can also be shown that t h e r e i s a maximum in the function
Appendix B-Equilibrium Time of Square Cascades for N -1
production Appendix C -The Holdup Function
-P
-
VS volume of catalyst Appendix D-Rayleigh Distillation

which gives the economic operating conditions when the catalyst vol- Appendix E -Properties of Concurrent Two-phase Elements
ume i s limiting. However, it is usual.1~expedient to run a t higher
values of P and get more complete recovery of heavy water from
feed.

REFERENCES

Cohen, K., Report A-118, May 1941; R e p o r t A-530, Feb. 4, 1943; Report A-531, Feb.
5, 1943.
Cohen, K., Columbia S e r . No. 4M-X148, June 25, 1941; Columbia S e r . No. 4R-X133,
August 1941.
Cohen, K., Columbia S e r . No. 4L-35, M a r . 17, 1943 (letter t o P. C . Keith).
Cohen, K., Packed Fractionating Columns, J. Chem. Phys., 8: 588 (1940).
Cohen, K., and I. Kaplan, Columbia S e r . No. 4R-6, Nov. 5, 1942.
Hutchison, C. A,, J r . , "Chemical Separation of the Uranium Isotopes," Division 111,
Volume 3, National Nuclear Energy S e r i e s .
Kaplan, I., and K. Cohen, Report A-335, Oct. 19, 1942; Report A-1276, May 1 , 1944.
Thompson, W. I., and K. Cohen, S e r . No. 4R-104, Sept. 3, 1943.
 Appendix A

ROOTS OF A TRANSCENDENTAL EQUATION

The features of the solution of the characteristic relations of Eq.

3.20 o r Eq. 3.22 for the roots pi, will be made clear by completely
discussing the simplest special case, namely K' = -, K = 0, and P = 0.
In this case it is necessary to solve for all the roots pi, of the equa-
tion
YS
tan Y S = -
â s
where

If both s i d e s of Eq. 1 a r e plotted a s functions of yS (Fig. I), the so-

lutions of the equation a r e the intersections of the curves y = tan yS
and y = y ~ / c ~If. cS is positive and l e s s than 1, there i s a root yS =
(yS), in the f i r s t branch of the tangent function, a s well a s other roots
which approximate to "

in all succeeding branches. Since p depends on (yS)', the symmetrical

negative roots may be neglected. The smallest root is i n the f i r s t
 APPENDIX A 141
140 THE THEORY O F ISOTOPE SEPARATION

Plotting both sides of Eq. 4 a s functions of flS (Fig. 2)- a root of

branch and gives pi; the other roots in the l a t e r branches give pa,
(fiS), i s observed. Since tanh (fiS), i s always l e s s than 1, s o also i s
(fi~},/es, and consequently p, i s negative.
-
h -
F o r negative eS, there i s no root of Eq. 4. All the roots yS a r e
r e a l and satisfy Eq. 1. Referring to Fig. 1, there i s a root tyS), be-
tween TS/Z and n in the f i r s t branch of the tangent, and t h e r e a r e l a t e r
.
roots which approximate to (2K - l)n/2 (K = 2.3.4.. ) i n all succeed-
ing branches. Again, p i s negative.

Q3
Fig. 2-Graph of both sides of the equation tanh BS = SS/CS a s functions of BS.

Approximate expressions may be obtained f o r the roots (yS), when

eS (of either sign) i s small o r large, o r when S i s positive and
nearly 1.
eS Large and Positive. The value of tanh (PS), is very nearly 1.
Equation 4 i s expressed a s

13S
tanh pS = 1 - 2e-2'3s + 2e-4'3s + . . . -- -
eS (6)
Fig. 1 :Graph of both sides of the equation tan yS = @ / C S a s functions of y S .
Solving by successive approximations,

If eS is positive and g r e a t e r than 1, the root in the f i r s t branch

disappears (dash-dotted line, Fig. 1). In this c a s e t h e r e is a pure
imaginary root, y , = ip,. Substituting /3 in Eq. 1,
Equation 7 to the number of t e r m s given i s good to 0.002 f o r eS > 2.
PS
tanh PS = - eS Large and Negative. The value of (yS), i s nearly TS. We set
cs
(YS), = TS + x. Then Eq. 1 becomes
and
142 THE THEORY O F ISOTOPE SEPARATION APPENDIX A 143

and solving by successive approximations, Table 1 -Roots of the Equation yS = (6s) tan y~

Equation 9 also gives ( Y S )for ~ eS very l a r g e and positive, and to the

number of t e r m s given i t i s good to 0.06 for - eS > 2.
eS Small, Positive o r Negative. The value of tan (yS), is nearly
i s nearly 7r/2. Set (yS), = (7r/2) - x. Then Eq. 1
infinite, and ( Y S ) ~
becomes
IT

expanding the cotangent. The s i z e of x i s very small and the root is

accordingly

Equation 11 to the number of t e r m s given is good to 0.001 for l e ~ lS

0.20. The expression also gives the succeeding roots ($Ik by sub-
stituting (2K - l)7r/2 for 7r/2. Explicitly,

eS Nearly 1. Since (YS), is s m a l l , Eq. 1 may be written as Equation 14, derived h e r e f o r eS < 1, can be shown to hold for ES > 1,
giving the expected imaginary values f o r (a.
It holds to 0.001 for
I s - 1 =Â 0.1.
Roots of Eq. 1 for eS = [-2.0(0.1) - 1.0; -1.0(0.5) + 1.0; 1.0(0.1)2.0;
5D] a r e listed in Table 1.
whose solution is
REFERENCES
Cohen, K., Packed Fractionating Columns and the Concentration of Isotopes, J.Chem.
Phys., 8: 588 (1940).
Cohen, K., Report A-530, Feb. 4, 1943.

Appendix B
EQUILIBRIUM TIME OF SQUARE CASCADES FOR N -1
This section outlines the procedure that will solve Eq. 2.1 in the
general case when N is not small. The differential equation i s , when
P = 0,
Consider the simple case (K'= 0 0 , K = 0, P = 0) -a cascade operated
under total reflux, with no reservoir on top, and the concentration a t
the bottom maintained constant. The boundary conditions a r e
The substitution of R instead of N as dependent variable makes the
boundary conditions linear.
At the same time, Eq. 1 takes the form
APPENDIX B 145
Now it is known f r o m the theory of partial differential equations1
that equations of the form
with boundary conditions of the type in Eq. 3 a r e solvable by iteration.
It remains t o choose the most efficient procedure.
The expression in parentheses on the right-hand side of Eq. 4 i s ,
in t e r m s of N,
aR 2 ( -~ NT)
2eR - - =
according to Eq. 1.77, where T - NT is the net transport of desired
material. Since a ~ / a sis always positive and since the transport f o r
the initial conditions chosen is also positive, the entire expression
is essentially positive. Thus an iteration process in which we begin
by overestimating this t e r m will converge to the solution from low
estimates of the equilibrium time; whereas if we begin by under-
estimating the t e r m we will overestimate the equilibrium time.
2 aR
The expression -- is the difficult term; the most efficient
l+Rds
iteration would seem to be performed onthis expression alone, getting
successive equations of the form
to solve, where Rk is the kth approximation to R and
^Maurice Gevrey, "Sur l e s equations aux deriv6es partielles du type parabolique,"
Gauthier-Villars. Paris. 1913.
146 THE THEORY O F ISOTOPE SEPARATION
We might take f o r fo either zero o r 4e%e2's/(l+ b e 2 " ) corresponding
to the initial and final values of R and aR/as (underestimates and
overestimates of f ) .
However, unless fir is particularly simple, Eq. 5 also has to be
solved by iteration. Thus practically the solution by way of Eq. 5
fails after RO. Even the choice f,, = 4 e ~ o e ~ ' ~ /+( ~l ~ e ~ ' ~ ) i cumber-
stoo
some; f i r s t - o r second-degree rational algebraic functions of s which
lead to special c a s e s of the confluent hypergeometric equation a r e
more convenient.
An iteration method that can be continued indefinitely is obtained by
solving equations of the f o r m
where
Here the homogeneous p a r t of the equation has constant coefficients,
which permits the particular solutions to be obtained operationally.
Appendix C
THE HOLDUP FUNCTION
-
The flow in any section of an ideal cascade i s equal to ~AU/E'.
Over a s m a l l section AU d ' v / d ~ ~ .Consider, instead of U = PV, the
expression H = PhK, where h i s the holdup p e r unit flow and
i t is clear that (2/e2) AH for the small section will be equal to the
holdup of desired isotope. Now a large cascade section may be con-
sidered a s composed of small sections, the product of one section
being the feed of the next. Likewise AH f o r the l a r g e section i s the
sum of the AH'S of the s m a l l sections. Hence (~/E')AHover any s e c -
tion of an ideal cascade, l a r g e o r small, represents the desired mate-
r i a l holdup in that section.
The integral of Eq. 1 with the constants chosen s o that K(0.50) =
d ~ ( 0 . 5 0 ) / d=~0 i s
This is called the "elementary holdup function." It i s the analogue of
V (not V).
To illustrate the use of H, consider the net desired material holdup
i n the cascade section B of Fig. 4.6.
Evaluating AU' around this section,
148 THE THEORY O F ISOTOPE SEPARATION APPENDIX C 149

If we define a general function

F2(N1,N2,N3,N4) = N1 ln -"2-
Ra
+ (N1 - 1) (R2 - R3) - N. Fl(Nl,N2,N3) (5)

gives the expression where the function Fl i s

R
i u l = pBV ! ( N ~ )+ ( I B - L ~ v'(NA)
) + Z B ~ B - NA) VN( A ) Fl(Nl,N2,Ns) = (2N1 - 1 ) In Ñ + (N1 - 1) (R2 - R3) + Nl (R, - Rd
R3 RyRa

Then in a s i m i l a r fashion it i s found that

F r o m Eqs. 4.39 and 4.40, this reduces to

2h
i U f = PBVfINB) - PB V f ( N ~-) PB (NB - NA) dN V' (NA) (N.D.M.H.)~ = 7PC F2(Nc,N~,No,No)
â‚

(N.D.M.HJB = 7
2h
PB F2(NB9NB,NA,No)
In exactly the s a m e manner, â‚

d
AH hPB K(NB) - hPB K(N,i) - hpefNa - NA) K(NA) 2h
(N.D.M.H.)~ = pCF2(Nc,Nc,NB,No)
ec
Substituting for V = (2N - 1 ) In R , K from Eq. 2, and 2h
(N.D.M.H.)~ = PA F~(N~,N*,N,,,N~)
â‚

2h
(N.D.M.H.)? = 7Pp F 2 ( N A 9 N A 9 N B , ~ J
cc

gives REFERENCES
Lefkowitz, H., B. Schwartz, and H. Mayer, Columbia Ser. No. 4R-42, Apr. 12, 1943.
Mayer, H.,I. Kaplan, and K. Cohen, Report A-396, Dec. 2, 1942.

and

Then the equation for the net desired material holdup i s

2 2
N.D.M.H. = -2- AH - 2 (hNo) AUf
 APPENDIX D 151

which integrates to give

Q" R" 1-N

( a o - l ) lp-=Q ln-+R ( a o - 1 ) in- 1 - N"

Here Q i s the initial quantity of the charge in the distilling flask, and
Appendix D N i s i t s mole fraction. Introducing the cut 6 = Q'/Q = (Q - Q")/Q,
where Q' i s the volume of the distillate,
RAYLEIGH DISTILLATION

The theory cascades which was presented in Chaps. 1 to 5 was

built upon elements that separate according to the law The mole fraction of the destillate, N', i s given by

R' = QR a = constant)

R~~= $R ( ft = constant) It is evident that Eq. 5 is fundamentally different from Eq. 1 and
represents a new kind of separating element.
However if a. - 1 i s small, then
Another common type of element which does not obey this simple law
will now be considered. An example of such an element was f i r s t R"
encountered by Lord Rayleigh while he was investigating batch dis- R = 1 + (ao - 1 ) In (1 -6)
- + order of (ao -
tillation.
Consider the course of a distillation. At any instant the vapor (mole R' 1-6 1 (7)
fraction n) i s in equilibrium withthe composition of the liquid remain- -R= I + ( % - 1 ) - i n - 9 1-6 + o r d e r of ( a o -
ing in the flask (mole fraction N") according to the relation
Provided 6 i s constant o r nearly so,' Eq. 7 is approximately of the
form of Eq. 1. Thus the infinitesimal c a s e of Rayleigh distillation can
be treated according to the theory of Chaps. 1 to 5, with effective sim-
Since the composition of the vapor leaving the liquid is different from ple-process factor a = 1 + e given by
that of the liquid, the composition of the residue changes a s the dis-
tillation proceeds. This is expressed by the conservation relation

~ ( N " Q)" = n d ~ ~ ' (3)

where Q" i s the quantity (in moles) of the liquid residue a t any in-
stant. The change in concentration of the residue (N") and distillate (N')
Combining Eqs. 2 and 3, relative to the initial concentration has characteristic peculiarities.
F r o m Eq. 7,

'1t was seen in Eq. 1.19 that it is so in an ideal infinitesimal cascade.

 THE THEORY OF ISOTOPE SEPARATION APPENDIX D 153

is a maximum [= - I)'] for 0 = 0, decreasing steadily because of

mixing of the successive distillate fractions (which a r e of changing
mole fraction). The value of 6V itself (column 4) increases initially
a s 6 increases, in spite of the mixing effect, just because more is
distilled: a s 0 approaches 1 the mixing is so serious that it causes
a decrease in 6U in spite of the increase in amount distilled. (If
from which Table 1 is constructed. everything is distilled, evidently no separation is accomplished.) The
maximum separative power per unit distillate may be obtained for the
Table 1 -Separation and Separative Power in Rayleigh Distillation entire charge by dividing the distillate into many small portions, in-
stead of collecting it all together. Then the separative power devel-
oped reaches G ( f f o- 1)'/2.
It is perfectly plain that any separation process that obeys Eqs. 2
and 3 is formally equivalent to Rayleigh distillation. Of interest is
the evaporative centrifuge (Chap. 6, Sec. 4), where the process is
indeed a distillation, but the concentration change (Eq. 2) is caused
by the centrifugal field instead of by a difference in vapor pressures.

REFERENCES
Cohen, K., Report A-50, Jan. 30, 1941; Report A-60, October 1941.
Cohen, K., and I. Kaplan, Report A-163, Apr. 25, 1942.
Rayleigh, J. W. S., Phil. Mag., 42: 493 (1896).

6 = cut; fraction of charge in distillate

N = mole fraction of charge
N1=mole fraction of distillate
N" = mole fraction of residue
a,,= equilibrium simple-process factor
SU = separative power
G = rate of feed to unit

By taking larger cuts it is possible to increase indefinitely the

changein concentrationof the residue (column 3 of Table 1). However,
the separativepower remains finite. If it is supposed that the element
continually separates G moles/sec of charge into 0G moles/sec of
distillate and (1 - 6)G moles/sec of residue, then by Eq. 1.42,

The value of 6U is found in column 4.

More fundamental than 6V is 6 V / 6 ~ which
, gives the separative
power per unit amount distilled (column 5). It is seen that this term
 -

APPENDIX E 155

provided

in the equilibrium state. Here k > 0 is a transfer coefficient and z

is a parameter expressing the size of the unit. F o r example, in a
Appendix E
catalytic-exchange process, z would be the volume of catalyst and k
would be proportional to the surface a r e a per unit volume of catalyst.
PROPERTIES OF CONCURRENT TWO-PHASE ELEMENTS These equations a r e consistent with the equations of conservation

A type of element which occurs inthe case of concurrent contacting L N(z) + 1 n(z) = constant = L N(0) + 1 n(0) (2)
of two separate phases, and in many other instances, e.g., the con-
current centrifuge, is illustrated in Fig. 1. Figure 2 shows a simple and solving Eqs. 1 and 2 together it is found that, after some r e -
cascade of such units. The element has entrances for two separate arranging,

N(Z)- N(0) = (1 - 9) [n(0) - ~ ( 0 +) ( f f - 1)N(1 - N)] (1 - e-bz)

n(z) - n(0) =-9 [n(O) - N(0) + (a - 1)N(1 - N)] (1- e-^) (3)

where

Fig. 1 -A concurrent two-phase element.

When z - W, the last t e r m in Eq. 3 approaches 1, s o that
streams, and unlike the element of Fig. 1.1 (Chap. I ) , there is no
particular necessity for the s t r e a m s to be of the same composition.
This extra degree of freedom gives considerable flexibility and r e -
is the percentage of equilibrium attained in the unit.
sults in more efficient cascades. There will be exhibited phenomena
The separative power of the element is
associated with the degree of equilibrium attained in each unit, and
the dependence on cut characteristic of Rayleigh distillation elements
(Appendix D) is absent.
The equations governing the transfer of material from one phase to
another a r e Expanding the value functions about V [ N ( ~ )and
] collecting,
dN
L -= -k [(N - n) - ( a - 1 ) ~ ( - 1 N)]
dz

and substituting from Eq. 3,

+ y1 [ d o ) - n(z)]
1
156 THE THEORY OF ISOTOPE SEPARATION APPENDIX E

STH OR TOP STAGE

It is seen that fiU is a function of (3 and N(0) - n(0). Taking N(0) = n(O),
which corresponds to the no-mixing case for single-entrance ele-
ments, the following is obtained:

and from Eq. 4, the separative power per unit length is

(s + 11TH STAGE

-5U
- - k(ff - l y , p
(no-mixing case)
z 2 - ln(1 - 6 )

which has a maximum at (3 = 0.715.

5U - sTH STAGE
-= 0.40724 (no-mixing case)
z 2
*
However, this choice of N(0) - n(0) is not the most efficient. Dif-
ferentiating 6V from Eq. 7 with respect to N(0) - n(0), it is found to
be a maximum when

N(O)- n(0) = -" ( a - DNO - N) (s - W H STAGE

2-6

which is

and a relative maximum with respect to 6 when 6 = 0, namely,

f ST STAGE

-6U
- - k(a-If
(optimum concentration relation)
z 4

By comparing Eq. 12 with Eq. 9 it i s seen that the optimum concen-

tration relation permits an increase in separative power of 19.6 per Fig. 2-A cascade of concurrent double-entrant units (rectifying section).
cent over the no-mixing case.
158 THE THEORY OF ISOTOPE SEPARATION APPENDIX E 159

It i s further evident from Eqs. 4, 8, and 11 that 0 = % i s the most

efficient cut, because this value r e s u l t s in a minimum value of ( L + Z)
for a given 6. Total volume = - (2 -13) In (1 - 6 )
dl - q ) 6
It is instructive to compare these two c a s e s (no-mixing and opti-
mum-concentration relation) by considering the net transport for the 1 2-/3
cascade of Fig. 2 with 9 = ?h. This is Total pumping =
- q) (3

Number of stages = 2 - 5
(1 - q)(3
On the other hand, f o r a given flow Ls the maximum possible t r a n s -
port through the sth stage is

Ls ( - 1)Ns+l(l - Ns+l)
ff

which occurs when

-- vs - 1 Ns+1
1-vs a l-Ns+l

The percentage transport ( r e f e r r e d t o the maximum) is therefore

When N(0) - n(0) a r e related by Eq. 10 and 9 = ?h,q turns out to be %;

when they a r e equal, q = (3/2. Thus in the no-mixing c a s e the net
transport is related to the degree of equilibrium.
The r e s u l t s just obtained may be generalized by introducing the
parameter q directly into the formula (Eq. 7) for <5U.It i s then found
that (for 0 = lh)

Fig. 3-Factors for total pumping,volume, and number of stages; a comparison of no-
mixing cascade and "constant per cent transport" cascade (q = 0.5).

The volume and pumping a r e both minimum with respect to q for

F r o m Eqs. 14 and 15 expressions a r e obtained for the total volume q = ?h. Figure 3 gives the three factors a s a function of fi for the two
of a cascade, the total pumping, and the total number of stages. These choices q = 0.5 (minimum volume) and q = (3/2 (no mixing). The min-
a r e , for the constant-transport case, imum volume for q = 0.5 i s a t 6 = 0 and i s 19 p e r cent l e s s than the
160 THE THEORY OF ISOTOPE SEPARATION APPENDIX E 161

minimum volume for the no-mixing case which occurs at 0 = 0.715.
Unfortunately both the pumping and the number of stages a r e infinite
It is obvious that the more general "constant per cent transport"
cascade offers greater opportunities than does the no-mixing case
which might a priori be considered the most efficient. A few interest-
ing possibilities a r e noted which illustrate the flexibility of the con-
stant-transport cascades.
The factor for minimum volume in the no-mixing case is 9.82 and
corresponds to a 71.5 per cent approach to equilibrium. In the con-
stant-transport case this volume factor corresponds to many choices
of q and 0;among others, to q = 0.5 and?! , = 0.82. In this case, how-
ever, the total pumping is 24 per cent l e s s than in the no-mixing case,
whereas the number of stages is only 3 per cent higher. It is thus
possible to save a considerable amount of pumping while losing only
slightly on the number of stages. Another constant-transport pos-
sibility with the volume 9.82 is the choice q = 0.4 and ft = 0.77. The
pumping is 15 per cent l e s s and the number of stages 4 per cent l e s s
than the no-mixing values for ft = 0.715. The no-mixing case thus rep-
resents no extremal.
A good balance between volume, pumping, and stages will be struck
by the choice q = 0.40 and = 0.80. But if one of these three factors
overweighs in importance the others, different values should be
chosen. Thus if it is especially desirable to reduce the volume to a
minimum, the choice might be q = 0.5 and (3 = 0.5. If it is desirable to
reduce to the limit both the pumping and the number of stages, the
choice should be q = 0.4 and 0 = 0.90, o r even 0 = 0.95. If the number
of stages is not important, then q = 0.5 would be chosen.

REFERENCES
Cohen, K., and I. Kaplan, Report A-302, Sept. 19, 1942.
I ----
NO MIXING Kaplan, I., and K. Cohen, Report A-301, Oct. 29, 1942.

Fig. 4-Factors for total pumping, volume, and number of stages as functions of q
and fi.

at 0 = 0. It is seen, therefore, that the significant region will lie be-

tween ft = 0.95 and ft = 0.50. Figure 4 gives the factors for volume,
pumping, and number of stages for q = 0.3, 0.4, 0.5, and 0/2 within
this range of 0.
 INDEX

Cascade, square, Type I1 operation of,

41 -44, 47-53
Aston, F. W . , 103 squared-off, of countercurrent ele-
ments, 34-35
rapid approximate solution for, 35-
38
scheme for, 38
Bardeen, J., 131
stages of, 3-5
Beams, J. W., 105, 113
number of, 3, 12
Boundary conditions, rectifier, 87
size of, 9
stripper, 87-88
stripping sections of, 21-24, 77-78
Bramley, A., 105, 119, 124, 131
Centrifuge, concurrent flow, 105, 114
Brewer, A. K., 105
countercurrent flow, 105, 119, 121-122
evaporative, 104-105, 111-114
separative power of, 109-1 11
Centrihge separation, design of, 105-106
Campbell, G. A., 50 differential equation of, 106, 109
Cascade, 3, 5 history of, 103-106
combined, 78 maximum separative power of, 109-
cut in, 5 111
design of, 11 theory of, 105
efficiency of, 6 Chapman, S., 131
equilibrium time of, 24-27, 38, 82 Cohen, K., 29, 38, 61, 78, 124-125, 136,
flow in, 13-15 143, 149, 153, 161
total, 9, 77-78 Column, countercurrent, 31
ideal, 5-10 number of stages, 31 -32
stages of, 5 fractionating, 30
ideal and nonideal, 10-12 relaxation phenomena for, 54-61
losses in, 67-74 thermal diffusion, 30
per stage, 69 Combination of separation plants, 74-78
production rate of, 12, 94-102 Concurrent contacting, 131
rectifier of, 21 Concurrent two-phase elements, 154-161
relaxation time of, 25 Control, hydrodynamic, 72-102
square, 30-44 Control problem, 79-102
equilibrium time of, 39, 144-146 Countercurrent centrifuge, 30
production loss of, 101 Countercurrent column, 31
production rate of, 94, 96 number of stages, 31 -32
disturbance function of, 96 Countercurrent contacting, 129
steady state of, 31, 62, 65 Countercurrent devices, 30-3 1
transient phenomena of, 62 -64 Cowling, T. G., 131
Type Ioperation of, 40-43, 47, 54 Cut in cascade, 5
 THE T m O R Y O F ISOTOPE SEPARATION INDEX

Transport, loss of, 90-91

maximum, 33
Debye, P., 131 Joly, J., 104 Schwartz, B., 78, 149
Deuterium, separation of, 126, 132 Jones, R. C., 119, 131 Separation, isotope (see Isotope sepa-
Deuterium separation, catalytic exchange ration)
method, 132-136 two-phase, 126-131
Diffusion, thermal, 131-132 Separation plants, combination of, 74 -78 Urey, H. C., 105-106, 125, 133
Dirac, P, A. M., 29, 109, 125 Separative power, 13-15, 19-21, 37-38
Kaplan, I., 29, 38, 61, 78, 106, 125, 136,
Distillation, Rayleigh, 150-1 53 (See also Centrifuge; Distillation)
143, 149, 161
separation in, 152 Separative work, 72
Keith, P. C., 136
separative power in, 152 Simon, F., 29
Kuhn, W., 105, 119 Value function, 13-14
Doetsch, G., 94 Skarstrom, C., 105-106, 113
Kurti, N., 29 elementary, 15, 17
Dunning, J. R., 29, 38 Square cascade (see Cascade, square)
Stages of cascade (see Cascade)

Laplace transform, 84, 89, 93

Elements, number of, 14, 19 Lefkowitz, H., 78, 149 Waldman, L., 131
Equilibrium fractionation, 67 Lindemann, F. A., 103 Thermal diffusion, 131-132 Waste material, 18
Equilibrium time, 24-27, 38, 82 Thompson, W. I., 62, 78, 136 Wirtz, K., 131
of square cascades, 39, 144-146 Transcendental equation, roots of, 139- 143 Work function, 64

Martin, H., 105, 119

Feynman, R. P., 67
Flow, total, 9, 15, 23
Fluctuations, 80 -90
nonperiodic, 90
periodic, 90
Foster, R. M., 50
Fuchs, K., 29, 84, 91, 102
Furry, H,, 119, 131
Mayer, H., 78, 84, 149, 161
Mixing, 6, 1 5
Mole fraction, 5
Mulliken, R. S., 104
Murphree, E. V., 125
,
Net transport, 21, 26

Gevrey, M., 145 Onsager, L., 119, 131

Holdup function, 25, 77-78, 147-149 Peierls, R., 29

in column, 59 Plants, separation, combination
net desired material (N.D.M,H.), of, 74-78
25-27 Poole, J. H. J., 104
in top reservoir, 59 Process factor, centrifuge, 105
Humphreys, R. F., 105, 113 infinitesimal, 9
Hutchison, C. A., 61, 136 large, 1 5
simple, 5
Production rate, variations in, 94-102

Isotope separation, 27-29

by chemical exchange, 126-127
by thermal diffusion, 131 Rayleigh, J. W. S., 150, 153
uranium, 15 Relaxation phenomena, 54-61