PH YSICAL REVI EW VOLUME 82, NUM HER 5 JUNE 1, 1951

A Collective Description of Electron Interactions. I. Magnetic Interactions*

DAvtD BoHM. AND DAvID PINEst
Pabner Physical Laboratory, Princeton University, Princeton, Em Jersey
{Received December 4, 1950)

A new approach to the treatment of the interactions in a collection of electrons is developed, which we
eall the collective description. The collective description is based on the organized behavior produced by
the interactions in an electron gas of high density; this organized behavior results in oscillations of the
"
system as a whole, the so-called "plasma oscillations. The collective description, in contrast to the usual
individual particle description, describes in a natural way the long-range correlations in electron positions
brought about by their mutual interaction. In this paper we con6ne our attention to the magnetic inter-
actions between the electrons; the coulomb interactions will be discussed in a subsequent paper.
The transition from the usual single-particle description to the collective description of the electron
motion in terms of organized oscillations is obtained by a suitable canonical transformation. The complete
hamiltonian for a collection of charges interacting with the transverse electromagnetic Geld is re-expressed
as a sum of three terms. One involves the collective 6eld coordinates and expresses the degree of excitation
of organized oscillations. The others represent the kinetic energy of the electrons and the residual particle
interaction, which is not describable in terms of the organized oscillations, and corresponds to a screened
interparticle force of short range.
Both a classical and a quantum-mechanical treatment are given, and the criteria for the validity of the
collective description are discussed.

I. INTRODUCTION "
lations, represent the effects of the long-range correla-
KCAUSK of the long range of the coulomb force, tion of electron positions brought about by coulomb
the interactions in a collection of electrons involve interactions. A description in terms of these organized
many particles simultaneously. However, the usual oscillations therefore provides a natural way of treating
description of these electron interactions is based on a the long-range electron interactions, and leads to greater
free-particle approximation. In a metal, for instance, insight into the dynamical behavior of the electron gas
the motion of a given electron is assumed. to be inde- than is afforded by the free-particle approximation.
pendent of the motion of all the other electrons in 6rst Thus, it may be expected that such a collective descrip-
approximation. The effect of the other electrons on this tion will make possible a better understanding of the
electron is then represented by a smeared-out potential, interactions between the electrons in a metal.
which can be determined by using the self-consistent In a treatment of these organized oscillations, ' one
6eld methods of Hartree and Fock. This means that considers a particular fourier component of the average
the effects of the correlations in the positions of the field, proportional to

expLi(k x —
electrons brought about by the long range of the
coulomb force are almost entirely neglected; we can
(ut)].
therefore expect that in any problem in which the For small amplitudes (which are of interest to us here),
electron-electron interactions are important, as, for the linear approximation is valid, and an arbitrary 6eld
example, the calculation of the cohesive energy or the can therefore be expanded as a sum of such trigono-
electronic contribution to the specific heats at low metric terms. In response to this oscillating field, each
temperatures, the free-particle model may not be electron undergoes a small corresponding trigonemetric
adequate. In order to obtain a better mathematical change in its velocity and in its contribution to the
treatment of this problem, we have adopted a new mean charge density. Owing to the long range of the
approach based on a collective description of the coulomb force, the mean 6eld at each point can become
motion. This collective description is most appropriate quite large as a result of the cumulative effects of small
for systems of high particle density. contributions arising from each particle. The condition
It is weH known that an electron gas of high density' for sustained oscillations is that the 6eld arising from

waves.
~
"
can undergo organized oscillations resemblmg sound
These oscillations, the so-called "plasma oscil-
the response of the particles must be consistent with
the field producing this response. This requirement leads
to a dispersion relation connecting co and k. For longi-
Submitted in part by David Pines to Princeton University in
partial ful6llment of the requirements for the degree of Doctor of tudinal waves the approximate dispersion relation, good
Philosophy. for Iong wavelengths, is'
t Now at Randal Morgan Laboratory of Physics, University of
Pennsylvania, Phjladelphia, Pennsylvania.
' The electron gas must be neutra~ by an approximately res = (4wtsse'/ra)+ 3k'aT/m, 0)
equal density of positive charge. In practice the positive charge
can usually be regarded as immobile relative to the electrons, and where no is the electron density, T the temperature, and
for the most applications can also be regarded as smeared out x is Boltzmann's constant. For infinite wavelength,
uniformly throughout the system.
' L. Tonks and I. Langmuir, Phys. Rev. 33, 195 (1929). (1949); Paper A discusses the origin of mediumlike behavior;
~ D. Sohm and E. P.
.
Gross, Phys. Rev. 75, 1851 and 1864 Paper B deals with the excitation and damping of osciHations.
625
 D. BOHM AND D. PINES
this reduces to the well-known plasma frequency, In order to obtain a collective description of the
= 41ISp's /iit electrons in a metal we must use a quantum-mechanical
Cd&
treatment, since the electron gas is highly degenerate.
There are, however, certain limitations on the col- Previous treatments of the organized oscillations of an
lective description of the electron gas in terms of electron gas have been comparatively unsystematic in
organized longitudinal oscillations which arise from the that hamiltonian methods were not employed, so that
fact that these oscillations cannot be sustained for the results were not extensible to quantum theory. In
wavelengths shorter than a critical distance known as our treatment of the collective description hamiltonian
the Debye length, methods will be employed throughout.
Let us consider the hamiltonian for a collection of
Xn = (cc T/4tcn pe') & V/—
cd p (2) charges interacting with the electromagnetic 6eld, the
This length is of the order of the distance traveled particles and the Geld being described by appropriate
during the period of an oscillation by a particle moving canonical coordinates. This hamiltonian may be repre-
with the mean thermal speed, V, and thus might sented schematically as
reasonably be expected to constitute a limitation on
organized oscillation. It may be shown, furthermore,
+D +pert++inter++field
tha, t whenever there is a static field in the electron gas, where Hp„& represents the kinetic energy of the elec-
either externally imposed or arising from a lack of trons, H;„~„represents the interaction between the
charge neutrality, the electrons redistribute themselves electrons and the electromagnetic field, and Hf;, M
in such a way as to screen out the 6eld within a distance represents the energy contained in the electromagnetic
of the order of a Debye length. ' field.
Screening and organized oscillation are different but Our program is to find a canonical transformation to
related manifestations of the collective behavior of the a new set of variables which will provide a collective
electron gas, brought about by cumulative responses of description of the system leading to results classically
the particles to the average force; and the Debye length equivalent to those obtained in the noncanonical
determines the smallest distance for which this collective treatment. Thus, we shall require that the new field
behavior is significant. Thus, one electively obtains a variables oscillate independently of the new particle
separation between the long-range part of the force, variables with the characteristic frequency of organized
which is best described collectively, and the short-range oscillation. When we do this we 6nd that the hamil-
part, which is best described in terms of the coordinates tonian in the collective description can be represented
of the individual particles. The higher the density, the schematically as
shorter the. distances at which the collective description
applies, and the more useful this description becomes. pert+&sec+&pert intr (4)
For a metal, the Debye length is of the order of 10 ' where H&op & corresponds to the kinetic energy in
cm; and therefore, the collective description is applicable these new coordinates and H„, is a sum of harmonic
practically down to interparticle distances. oscillator terms with frequencies given by the dispersion
Organized transverse oscillations of an electron gas
relation for organized oscillations. Hp„~, „~ then corre-
are also possible. These oscillations are electromagnetic
sponds to a screened force between particles, which is
waves strongly modi6ed by the fields arising from the
large only for distances shorter than the appropriate
collective particle response. Such oscillations are ob-
minimum distance associated with organized oscilla-
tained in the transmission of radio waves through a
tions. Thus, we obtain explicitly in hamiltonian form
highly ionized medium, such as the heaviside layer.
the e6'ective separation between long range collective
The organized transverse oscillations can be given a
interactions, described here in terms of organized
treatment similar to that of the longitudinal oscilla-
oscillations, and the short-range interactions between
tions. ~' The dispersion relation for these waves is
individual particles. e
Cd~Cd p'+ C'k'. In our treatment, certain approximations must be
made, which are discussed in detail in Sec. II, in
The minimum wavelength for which the collective connection with the collective approximation. These
behavior is important is, in this case, approximations reQect the fact that while the effect of
the average 6eld on an individual particle is small,
these cumulative small contributions from each particle
This distance is clearly much longer than the Debye 6 In the case of longitudinal oscillations, this model for electron
length. interactions provides a physical basis for the hitherto empirical
use of a screened coulomb force to represent correlation effects.
4P. Debye and E. Huckel, Physik Z. 24, 185 (1923). This Furthermore, it predicts a screening radius which produces
phenomenon was Grst studied in connection with highly ionized agreement with the experimental results for both the electronic
electrolytes. contribution to the speci6c heat of a metal, and the width of the
~ For a treatment
along the lines of Bohm and Gross, reference tail of the soft x-ray emission curve for sodium. D. Bohm and
3, see D. Pines, PhD. thesis, Princeton University (1950). D. Pines, Phys~ Rev. 80, 903 {1950).
 MAGNETIC I NTERACTIONS 627

to the average Geld may produce a large change in Then, following the usual treatments, ' it can be shown
these 6elds relative to what they wouM be in a medium that our hamiltonian (5) becomes:
of low charge density. Thus, perturbation theory is
applied to the solution for the motion of each particle H =g(pP/2m)+g(4s e'/L') &[(p; ei, „)/m jql, „exp(ik x;)
in the average 6eld of all the others, but cannot be
applied in similar fashion for the solution of the field + P (2we'/mL')qi, „q(.eI.„s(.exp[i(k+1) x;j
equations of motion. On the other hand, because of the
high density of particles, we may assume that only the + 2+(p-l„p pl, +c'k'qg„q p„). (8)
response of a particle which is in phase with the field
producing it will be important and that other responses Here pq„ is the canonical conjugate to qq„, and we also
which depend on the position of the particle can be have
neglected. p"=(p ~»)*.
We confine our attention in this paper to the collec-
We 6nd it convenient, for reasons which will become
tive description of the interactions between electrons
clear later, to split up the term
brought about through the medium of the transverse
electromagnetic 6eld. These magnetic interactions are
weaker than the corresponding coulomb interactions
(2xc2/mL') P qq„q~„eq„e~„exp[i(k+1) x;j
by a factor of approximately s'/c' and, consequently, are —
in those terms for which 1= k and 14 k. When this —
not usually of great physical interest. However, the
is done, our hamiltonian corresponds to the schematic
canonical treatment of the transverse 6eld is more
hamiltonian of the previous section, with
straightforward mathematically than that of the longi-
tudinal 6eld; and since we would like to illustrate —
H p.„PpP/2m—, (10a)
clearly the techniques and approximations involved in
our methods, we therefore investigate 6rst the role of the
H;„t,„=g(4xe /m2L')&(p, 'sl, „)qz„exp(ik x;)
organized transverse oscillations in a description of
electron interactions. In Sec. II, we give a classical
treatment and discuss in detail the collective approxi-
P(2xe/mL') P
kbixv
q&„q&„s&„s&„
mation and the effective residual interparticle force. lW —k
In Sec. III we give the analogous quantum-mechanical )&exp[i(k+l) x~j, (10b)
treatment of the collective description. The longitudinal
oscillations wiB be treated in a subsequent paper. +~4~~ 2E[p~PP ~F+ (c k ++% )q~Pq (10c)

II. THE CLASSICAL CANONICAL TRANSFORMATION We will now show that a canonical transformation to
TO THE COLLECTIVE DESCRIPTION the new coordinates (X;, P;; Q~, IIL), which is generated
(A) Generating Function for the Transformation by S(x;, P;;qq, II~) as given below, constitutes the
desired transformation to the collective description.
The hamiltonian for a collection of charges interacting This generating function is
with a transverse electromagnetic field may be written
S(x;, P;; qg, Iig) = Qx; P;+Pqg„lip„
H= g [y;+e A(x;)/cg'/2m
+F(x;, P;; qp, Ii),), (11)
+ ([E'(x)+8'(x)g/Sx)dx. (5) where

F(x;, P;; qa, IIg) = QPg„(P~) qg„lip„—(4xc'/m'L') 4

We expand A(x) in a fourier series in a cube of volume
I, , and impose periodic boundary conditions. Thus,
XP(P; er.)
[(l P,/m)q, .+iIr, „jexpil x, (12).
A(x) = g(4 /sLc') &qg„eg„exp(ik x), (a' —(l P~)'/m'

b„(P;) depends only on the particle momenta and is of

where a~„ is a unit polarization vector, and for this the order w'/c', it will be determined in the course of
transverse Geld, p takes on values 1 and 2, representing working out the consequences of this transformation
the two possible values of polarization perpendicular to (see Eq. (25)); co represents the frequency of the
the direction of propagation. The wave number vector organized transverse 6eld oscillations. With this gener-
k is assumed to take on both plus and minus values, 7 See, for instance, G. Wentzel, The
Q24uetum Theory of Iiklds
i.e., k =2xn, /L, where' = —~ . +00, etc. Toensure (Interscience Publishers, Inc. , New York, 1949), Chapter IV.
that A(x) is real, we take It will be seen that au depends on the momentum P;. However,
this dependence is of the second order (in e/t, ), and the corre-
sponding changes in the transformation equations may be neg-
VII = ig-&~i ~ lected.
628 D. BOH M AN D D. PI NES
ating function we obtain the following equations gas in which the mean free time for such collisions is considerably
longer than the period of an organized oscillation. Actually, such
relating the old and new coordinates:
collisions tend to disrupt the organized osciOation of the system
as a whole; but if they are not too frequent, they lead only to a
aS (4zre ) &
P; a)„ '
small damping of the oscillations. The long free path needed for
pa= —
=&,~+
lzoz —
(
zz) the validity of this approximation actually occurs because of the
axe, L mL &
(1 P;)'/m'
screening of the long-range part of the force, which leaves only
the short-range interactions to be accounted for in terms of
X I (l P;/m)q&„+iII&, Ilz exp(z1 x~), (13) collisions.
(2) The organized oscillations are assumed to be of suKciently
small amplitude that each particle suBers only a small pertur-
Xo, = a5/aP;), —x,),+ (4—
~e'/mzLz) &
bation in its straight line motion due to the combined fields of all
the other particles. Thus, we will neglect quadratic Geld terms in
XZ I [z(«.)~/[~' (1 —
P') /zm]] the electron or field equations of motion and apply perturbation
lv
theory to the particle motions. This is the customary linear
approximation, appropriate for small oscillations.
X [[(l P,)/m]q(. +zII ), ] (3) We distinguish between two kinds of response of the elec-
trons to a wave. One of these is in phase with the wave, so that
+[(P,"e(,)lg/[zzz —(l P~)z/m, ']] the phase difference between the-particle response and the wave
producing it is independent of the position of the particle. This
X[[aP+(1 P;/m)z]qg, +2i(l P;/m)II („]I is the response which contributes to the organized behavior of
the system. The other response has a phase diGerence with the
X exp(il x;)++II»q»(ap»/aP;&), (14) wave producing it which depends on the position of the particle.
Because of the general random location of the particles, this
second response tends to average out to zero when we consider a
P„„=as/aq, „=n, „(1+g.„) large number of electrons, and we shall neglect the contributions

—(4 ~/ 'L')'ZI(»' .)(k P,)/ arising from this. This procedure we call the random phase
approximation. "
(4) We shall assume the smallness of (k. v)/~. In our case,
~/k~~, so that the smallness of gr v)/co follows from the smallness
m[aP —(k P~/m)z]I exp(ik x, ), (15) of o/c. For longitudinal oscillations, however, we can have small
(k v)/co only for wavelengths appreciably longer than the Debye
Q»=aS/aII z„=q»(1+&») —(4zre'/mzLz)& length, XD. This follows because Xz is essentially f /ao, where P
is a suitable mean speed. Thus, the collective description is
X+I(P"ez )/[oP —(k P./m)']I exp(zk x.). (16) applicable only for long enough wavelengths. This result agrees
with the general conclusion cited previously that both statically
and dynamically a dense ion gas exhibits organized behavior only
%'e see that there is a rather complicated interrelation- for wavelengths longer than X~ or 'Ag.
ship between the old and the new coordinates defined The collective approximation is similar to a complete
by these equations. %e will not be able to solve these perturbation theory treatment in that perturbation
equations exactly to obtain the oM coordinates in terms theory is applied to the particle motion. The diBerence
of the new coordinates or vice versa. Ke could obtain between the two approaches lies in the assumptions
approximate expressions for the old coordinates in made regarding the Gelds produced by the particles.
terms of the new, and then determine what form the For instance, in Eqs. (15), and (16) we see that the
hamiltonian will take in terms of the new variables. old and new field coordinates and momenta di8er by a
However, we feel that the nature of the collective series of terms summed over all particle coordinates. A
approximation will be revealed more clearly by a some- complete perturbation theoretical treatment would
what di8erent procedure. Ke will substitute the
require that this sum be small, but we do not make this
"mixed" expressions for p;, p», and q» [(13), (15), and assumption here. Instead we note that the small
(16)], into our old hamiltonian and thus obtain a modifications of the net Geld arising from each particle
hamiltonian in terms of both old and new coordinates. may add up to a large change of the average field.
However, this hamiltonian can be simpli6ed consider- The principal advantage of the collective approxima-
ably with the aid of the collective approximation, and tion is that it does not require a small cumulative
after achieving this simpli6cation, we then express the
response to the fields. Thus, in this approximation we
hamiltonian entirely in terms of the new variables.
shall retain terms in the hamiltonian proportional to
B. Nature of the Collective Approximation the square of the vector potential, whereas in a pertur-
bation treatment these would be discarded, since they
The approximations we find it necessary to make, in are formally of second order in the perturbing potential.
order to apply the collective description to the electron. As we shall see, in the case of high particle density,
gas, have been grouped by us under the general heading these can bring about a signi6cant modification in the
of the collective approximation. The collective approxi- behavior of the electromagnetic field.
mation involves the following requirements:
(1) The short-range electron-ion and electron-electron collisions NThis damping is discussed in detail in Bohm and Gross,
are neglected. This assumes that we are dealing with an electron reference 3.
 MAGNET IC I NTERACTIONS
C. Results of the Tranaformation; the Hamiltonian follows:
for the Collective Description Hi;. ia=E{(IIa,+8F/8q», )(II ~, +8F/8q ~,)
Let us now see what eGect the transformation has on
the various terms in our hamiltonian. Using Eq. (13) +ie'[Qi, „Q i, (8F/8II, „)
„2q»„—
we have —(8F/81ii, „)(8F/8II k„)]}
—PF;2/2m+ (4~e'/m'L') &P [(I P,)/m] +-', Q(c'k'+co„' —
H». „— co') qi„q
kis
g„. (20)

This particular combination is taken because P is a

{[(l P;)/m]qi„+iII —exp(il
i„}
X [P,"ai, ]
function of qI„-. We then obtain
x;)
ce' —[(l P;)/m]' H „.„=-,'P {11,„11,„(1+2g„„)
kp
2~e' (P," ai, )(P; aa„)(k l)/m +& (Qk»Q 2— k»)+(i—
e» +c k i0 )&»q —&»}
+mL' ~i»- [i0' —(l. P. —(k. P./m)']
&» 6»q&»q
/m)'][co' (47re' )» -P,"ap»(1+pi, »)
X {(l P,/m)qi„+iII i„}{((k P;)/m)qg„+iII k„} Lm'L')» ~' —(k P /m)'
X {—i(k P,/m)II i„+is'qi„} exp(ik x, ) —(2ae'/m'L')
{(P; ai„)(P,"ai„)[(e —(k P;/m)(k P,/m)]}
Ke consider the terms for which 1= —h, and 14 —h, XZi» —(k —(k
separately in the quadratic Geld terms in (17). When
[~e'2 P,/m)'][41' P,/m)']
1= —h, this quadratic term reduces to Xexp[ik (x, —x;)]. (21)
2s.e' (P; aa»)'k' In (21) we have neglected terms involving $i»', since
these are of order v4/c', and for the particles with
mL' ~» |m[~e' —(k P,/m)'] which we will be concerned a'/c'(&1.
The sum of the expressions (17), (19'), and (21) will
{(k P;/m)'qi»q i»+II&»II &»} (18) give us our hamiltonian (8) expressed in terms of a
mixture of the old and new coordinates. When we add
provided we assume an isotropic distribution of the P;. these expressions, we find that those terms which are
Using Eq. (13) we may write the terms arising from linear in the field coordinates are reduced to
—(4a e'/m'L') &Pgg„(P," a»„)
+1IlteF as

H;„„,= (4s e'/m'L') &Qqi»(P, ai») exp(ik "x,)

{ie2qi, »+i(k P,/m)II i„.}
X —(k exp(ik x;). (22)
4ae' (1 aa„)(P; ai„)
~02 P;/m)'
+ 'L' But this term is of order v'/c' smaller than our original
m, is&». aP (l P,/—
m)'
lg-A H; i„, Eq. (10b). We could devise a further canonical
transformation which would eliminate terms of this
X[(1 P;/m)qi, q &»+iII i,q&»] order too, but this would only introduce corrections of
order a4/c' in the interparticle force. Since we will
+-,'(a»„ai„)q»„qi„exp[i(k+ I) x;], (19) confine ourselves to the terms of lowest order (a'/c') in
the interaction between particles, we can neglect this
small term (22) entirely.
where we have split up the terms for which 1= k, — Let us now investigate the quadratic 6eld terms
and —
14 k, and applied the transversality condition resulting from the sum of (17), (19), and (21). These
P„k ai» —0. It is convenient to group the terms arising fall into two categories, which are given in (23) and
from the application of (15) and (16) to Hi;, ie as (24) below:

27( 8 (P; a&, )(P,"aq„)l k[(l P;/m)(k P;/m)qigi, „Iiq„iii„+— 2i(k P;/m)q~„II i„]
F13 iklpv— —
m'[ce' (k P,/m)'][aP — (I.P,/m)']
k~ l

2(l ag»)(P; ai„) f1

P„. i
+ j}qi»q
i»+iII igi» +qa„qi, (ai, „ai„-) exp[i-(k+1) x;], (23)
m[aP —(l
{
P,/m)']. ( m
6% D. BOH M A N D D. PI NES

(P,"ek„)'P ~k P;~
'
qkpq kg+ —II kg II kp—
mL' kk' m'[aP —(k P,/m)']
+~+ f(ekk&+co, ' —~k)qk„q 'E(ilk„il k„+w'Qk„Q k„). (24)
k„+2)k„(IIk„rI k„—~'qk„q k„)}+-,

The terms in (23) are all quadratic in the field

variables, and in addition have a phase factor
dispersion relation for transverse plasma oscillations.
Thus we see that we have reduced the field terms
"
exp[i(k+ I) x,], the argument of which never vanishes. represented in (24) to
In order to discuss these terms, we transform from the
old coordinates x; to the new coordinates X;, according kZ(ilk, iI-k, +co'Qk, Q k, ). (2g)
to Kq. (14). We will then have two types of terms;
those in which x, is replaced by X;, and those which The remaining terms are given by
arise from the difFerence between x; and X;. Now the
X;, because Hp„t, , ~ is a screened short-range inter- 2xe' (P, . ek„)(P,"ek„)[c0 —(k P,/m)(k P,/m)]
action, behave to a good degree of approximation like m'L' —(k
cikl [cd' (k Pc—
/m)'][ce' Pg/m)']
the coordinates of a free particle. We may then assume
that these are distributed at random, as in a perfect Xexp[i1c (x;—x;)]. (29)
gas. Since we have a very large number of particles,
the nonvanishing argument of the phase factor, We now wish to express this term entirely in terms of
exp[i(k+I) X;], will cause the contribution of those our new coordinates. One can show, using the linear
terms in which it is present to average out to zero. approximation and the random-phase approximation,
This is essentially the random-phase approximation. that the lowest order (in s/e) field term resulting from
The efFects of the correlations in the x; are represented substituting our expression for x;, (14) into (29), is
in the di8erence between x; and X;. However, the terms
arising from this difFerence multiply quadratic 6eld ( 4~e ) 4k~ (Pc'eke)qkP
exp(ik X,). (30)
terms; and it may be shown that the resultant products EmkLk) 'kc (yk —(k P,/m)k
either make nonlinear contributions to the equations
of motion, which we neglect in the linear approximation, This term, which is of the same sort as (22), may be
or average out to zero because they contain a phase —
neglected for the same reason that its inclusion will
factor with nonvanishing argument. ' Thus, in general, lead to fourth-order terms in m/c in H„,c; c. Thus, we "
terms which are quadratic in the field variables and obtain, to this order of approximation, only the above
contain a phase factor with nonvanishing argument term (29) with x; replaced by X,.
may be neglected in the collective approximation. Our transformed hamiltonian, expressed entirely in
When we consider the terms in (24), we see that if terms of the new coordinates, the "collective" variables,
we take thus takes the form:
——(2we'/mL')
gk„— Q 8(P,"ek„/m)'/ H&" =+(Pp/2m)+-, '+ilk„li k„

[ ' —(k P'/m)'] (25)

then thequadratic terms in G»II », reduce to
+co'Qk„Q k„(2n.e'/mkL'—)
xgk„ II»II ». We similarly see that with this choice I (P," e»)(P; ek„)[co' (Ir P,/m)—
(k P,/m)]}
of gk„, the quadratic terms in qk„q k„become X
[a)' —(k P;/m)'][ca' —(k P./m)']
2kgqk„q k„c'Ic'+~~'+(4w—e'/mL') —
)& exp[i k(X; X;) ]. (31)
This is just the hamiltonian we sought, as discussed in
{P(P,".ek„/m) [kP+(k P,/m)']} Sec. I, with
(26)
—(k

xZ,
[cd' P;/m)']' H &"p„c —QP, k/2m, (32a)
but, if we take
't' This difFers from the dispersion relation quoted by Langmuir
ca'= c'Ic + '+ (4m e'/mL')
co in the terms of order o'/c2 which he neglected. It may be shown,
using methods similar to those of Langmuir, that the above
}(P," ek„/m)'P[aP+(k P;/m)']} dispersion relation is obtained when these terms are not neglected
— see D. Pines, reference 5.
(») "The term in Eq. (14) involving 8&/8I'; leads to quadratic
[co' —(k P;/m)']' field terms multiplied by phase factors which do not vanish.
According to the general properties of the collective approximation
then this term vanishes. This choice of aP is just the this term can be neglected.
 MAGNETIC I NTERACTIONS 63i

+ose = gpff kyle-kg+ & QkyQ —key (32b) the interaction is greatly reduced for small values of k,
kp, or to put it another way, screened out for sufFiciently
e,.„;„,= ~e'/m'L')
(2— large interparticle distances. To see this in more detail,
let us evaluate Hp„t;„t in terms of the interparticle
(P; e)„)(P"eg )[co' —(k P~/m)(k P;/m)] distances. First, we carry out the sum over polarization
X directions, obtaining
[g' —(k P;/m)'j[oP —(k P;/m)'j
Xexp[ik (X,—X,) j. (32c) Hpart int
We see that the new field and particle variables no
longer interact. The new field coordinates carry out Xexp[ik (X;—X;)g. (36)
oscillations of frequency or given by the transverse field
dispersion relation, Eq. (27), and thus correctly describe For simplicity we consider only the first term in Hp
the organized transverse oscillations. The new particle and we neglect (k P,/m)' in the denominator, since it
coordinates act like those of a free particle, to the extent will always be order v'/c' smaller than c'k'. We replace
that we are able to neglect H peart &nt which represents the sum over k by an integral, multiplying by the
an effective residual particle interaction. density in k space, (L/2s)'. We thus have:

D. Effective Residual Particle Interaction

Let us now investigate H„,t,. t in some detail. It
H&" p. „t;.——[2~e'/m'(2s-)'j+P, "P,
t dk

e6'ectively corresponds to that part of the magnetic

interaction between electrons which is not describable X {[1/(o '+c'0'1 exp[ik (X;—X,) j}. (37)
in terms of the organized transverse oscillations. Since The integral over k yields just
cv'~&c'k', we see that this term will be at least of order
v'/c' smaller in magnitude than the corresponding 2s.~[exp {—[cd „/c] X;—X, }7/ X;—X, },
( ) }

coulomb interaction between the electrons, which is

so we have
given by
(2s.e'/L')P(1/P) exp[ik (x, —x.
,) j. H "„„t;„i=—(e'/2)g(P, "P,/m')
Hence, we shall consider only the lowest order (in v/c) X [exp{ —(cd, /c) X, —X;} }]/}X;—X;}. (38)
}

terms represented in (32c), neglecting those terms of This leads to a screened Biot-Savart law of interaction
order e4/c4 as compared with the coulomb term. The between the particles, with the interaction being
lowest order terms may be written: screened out at distances c/cv„. For a metal, where
3 X 10 ' cm.
e„„;„~—(2~e'/m'L') sp 10 ', this screening distance is

(P' s~.)(P" e~.) III. QUANTUM-MECHANICAL TRANSFORMATION TO

X P —
exp[ik (X, X,)]. (33) THE COLLECTIVE DESCRIPTION
*'i~~ co' —(k P;/m)' A. GeneratingFunction for the Transformation
Now to this order of approximation, we may write We now carry out a quantum-mechanical treatment
z'=c'k'+co„', so that we have of the preceding classical material. We shall see that
2s e' '
(P, eg„) (P," ep„)
the quantum-mechanical calculation yields essentially
the same results as the classical treatment in the
m'L' ~ i&l c'k'+re„' (k'. P;/m)'— preceding section.
We find it convenient to work in the following repre-
Xexp[ik (X;—X,)]. (34) sentation. We expand in terms of the creation and
annihilation operators for the transverse field, aj, „and
We now compare the above expression with the analo- (a~.)*,
gous term in the hamiltonian from which may be
derived the Biot-Savart law for the magnetic interaction A(x) = g ( sr 2k / cLid') &[a~„e px(ik x)
between particles. This may be shown to be

.
H, s„, ——(2x e'/m2L')
g
+aA, „*exp( „(39).
ik x) jei,—
These are connected with PI, „and q~„by the following
(P; ea, )(P;. ea, ) relations:
(X,—X,)j.
'i~~ c'k' —(k P,/m)'
exp[ik

Because of the presence of or„' in the denominator in

(35)
a~. = (~/2&)'[if~. +i(P ~./~) j,
a~, *= (~/2&)'LC ~. i(P "/~) j,
— g~„—(h/2') &[ay„+a p„*j, (40)
—-
Eq. (34), we see that when the ion density is large, ——i(ka)/2) &[a~„—a a„*j.
pg„—
632 D. BOH M AND D. PI NES
Using these, one can show that the classical hamiltonian rules as the (x;, p;) by
(8), which is equally valid in the quantum-mechanical
case, leads to the following hamiltonian. in this " x;= {exp(—iS/h) } X;(exp(iS/h) },
representation: p;= {exp(—iS/h) j P;{exp(iS/h) }, (44a)

H= p (p,k/2m)+ (e/m) Q (2grh/kgL') &(p; ek„) and similarly

ak„(exp(—
—iS/h) }Ak„(exp(iS/h) }. (44b)
X(ag„+a „*)exp(ik x,)+(2ire'/mL')
g„.

We may consider (44a) and (44b) as operator equations,

X P (h/kg)(ek, «„)(ak„+a k„*)(ag„+a g„') and take S to be a function of the new operators
klan,
lW
vi
— k (X;, P, , A», A»~) only. 5 will be the quantum-
mechanical analog of our classical generating function.
Xexp[i(k+1) x;)+ kkPhkg(ak„ak„*+uk„*ak„) The relationship between the old and the new hamil-
tonians may similarly be viewed as an operator equa-
h'
tion:
+kp[h(rug'+C' [ak„ak„'+uk„*a»
cd' )/-kg)—
krak H= {exp(—iS/h) }a{exp(iS/h) } (44c)
+a —kpakgg+a —kk akk ) (41)
where 3C represents the hamiltonian expressed in the
We may adopt the schematic notation of Sec. I, with new coordinates.

H. —p(p..
g /2m),
The desired canonical transformation to the "col-
lective" representation is then defined by the following
generating function
—(e/m)p(2xh/cvL')&(ak„+a
H;„g„— k„')(p, ek„)' S= —(ei/m)g(2grh/kgL') ~

Xexp(ik x;)+(4ire'/mL') p
klpcvi
lW —k
(h/2cd)(ek„«, )
X ([(P; sg.)Ag, /(cd —(1 P,/m)+hP/2m)] exp(il X, ).
—[exp( —zl X;)][Ag,*P,"eg„]/
X(ak„+a k„*)(ag,+a g„*) exp[i(k+1) xg], (42b)
Hgiegd= kphca(akggakk +akgg akk) (co —(1 P,/m)+ hP/2m) }. (45)
» ikey

—cek)/co)[a B. Consequences of the Transformation

+-gp[h(co„'+e'h' „a „*+a „*a „
In working out the consequences of this transfor-
+a „a „+a „*a „*). (42c) mation, we shall expand the exponentials, exp(&iS/h),
so that (44c) may be written:
The order of factors in (42b) is not essential because,
due to the transversality condition, y;. e» commutes H =K (i/h) [S, K) —(1/2—
h') [5, [S, K))
with exp(ik x.;). We are working in the Heisenberg +(i/6h')[5, [5, [S, se)))+ . (46)
representation and our operators satisfy the usual
commutation rules: We will classify terms according to the power of S they
—
contain i.e. , [5, 3C) is the first-order commutator of
Sky p Clv
= [ak„, ag„)=0, S and X. We shall see that the terms arising from
+kis ~lv
(43) [S, [5, [5, BC))), and higher order commutators can
[p;» x ]= (h/i) f'g,
, f'gg„, be neglected if we restrict our attention to the lowest
jib
order terms in gg/c.
Just as we did in the classical case, we now seek to The following relationships are useful in applying
find a canonical transformation to the collective de- (46):
scription, in which the held coordinates describe the
— —(i/h) [S, P;g, ]+.
p;k —P, k
organized transverse oscillations. The transformation
theory most suited to our purpose is briefly as follows. " = P,k+ (e/m) p(2s h/kgL') glk
We define new operators (X;, P, ) which possess the
same eigenvalues, and satisfy the same commutation
'~ This hamiltonian differs from the customary one, because ~ve X {[(P; sg, )Ag./(&a —(1 P;/m)+hP/2m)]
have expanded in terms of an arbitrary oscillation frequency co,
rather than taking oy=ck. X[exp(jl X;)]+[exp(—il X;)]
"Quantum-mechanical transformation theory is developed in,
[A, „*P; eg„/(c0 —(1 P;/m)+M'/2m)] j, (47)
for instance, P. A. M. Dirac, Pr~ecip/es of Quuntues 3fechanics
(Oxford University Press, London, 1935), second edition.
 MAGNETI C I NTERACTIONS

exp(ik. x;) =exp(ik X;) —(i/h)[S, exp(ik X,)]+ H;„~, which we denote by

= exp(ik. X;) —(e/mh)Q(2s h/coL') & X'; c„——

(e/m)g(2s h/ceL') &(Aa„+A a„*)

y, {P;. ac„A c,[I/(ce —(I P,/m)+ AP/2m') X(P; ea„) exp(ik X;). (53)
—1/(ce —(I P;/m)+ —hl k/m)]
(AP/2m) Thus, there are no interaction terms of this order (in
s/e) in the new hamiltonian. The vanishing of the
yexp[i(k+I) Xc]+[exp[i(k —I) X;]] linear 6eld terms in this order is one of the desired
y P; rc„A c.'[1/((u —(I P;/m)+ (AP/2m) properties of the hamiltonian appropriate for the
collective description.
—hl k/m) —1/(ce —(I P,/m)+AP/2m)] When we consider the terms arising from the second-
in P», c and -', Pa„Ace(aa„ca»*
+hk ec,[(exp(ik X;))(Ac„/(ce —(I P,/m) order commutators
+ca»aa»), we see that they will yield just minus
+AP/2m)) exp(il X,)) —(exp( — il X,)) one-half the terms given by the 6rst-order commutator
arising from 3'.~; ~„, and that all of the terms arising
X(Ac„a/(ce —(I P,/m. )+AP/2m)) from the higher order commutators in these two terms
X(exp( —ik X;))]}, (48) will bear a similar simple arithmetical relationship to
the terms in the next-lower order commutator in
cc» —
— (i/h)
Aa„— [S, A a„]+ K; „.
t, The 6rst-order commutator of 5 and 3'. ;„t, is: „
= A a„—(e/mh)Q(2s h/cdL') &[exp( —ik X;)]
—(i/h) [S, K';~c„]= —(ei/mh)Q(2s h/ Lce') &

X[(P;.s,)/( —(k P'/m)+Aha/2 )] (49) X {[S,(I'; „)](A „+A „*)(e p('k X,))

The calculation of the commutators for (47) and (49)

+(P,"sa„)[S, (Aa„+A a„*)](exp(ik. X,))
is trivial. In obtaining (48) one encounters commutators +(P; „)(A „+A „*)[S,(exp(ik Xc))]}. (54)
like
It is quite straightforward, but tedious to show that
[{1/(ce—(I P;/m)+AP/2m) }, {exp(ik X;) }]. this reduces to
Such a commutator
one notices that:
may be easily calculated, provided —(4 ~/ 'L') 2 {(P',)(P' .,)/
{I/[ce I P, /m+— AP/2m]} exp(ik X,)
[ga —((k P;/m)+hh/2m) ]}exp[ik (X,—X,)], (55a)
=exp(ik X;) {1/[ce —I P;/m+AP/2m
—hl kb;;/m] }. (50) 4me' (hh'/2')(P;. s»)'
+maL
[ce —(k P./m)]' —h'hc/4m'
That (50) is true, may be seen by multiplying both cac
sides by (ce I P /m+ AP— /2m) and commuting this with
exp(ilt X,) on the right-hand side. Using (50), we see X {Aa,Aa, *+Aa, 'Aa, +Aa, A a, +Aa, *A-a, *} (55b)
that:
[{I/(ce=—(I PI/m)+ AP/2m) —
},exp(zk. Xc) ] In reducing (54) to (55a) and (55b) we have neglected
a number of terms which are quadratic in field variables
{exp(ik X;) } {1/(ce (I P;/m)+(AP/2m)
m) 1/(ce —
and are multiplied by a phase factor with nonvanishing
hbgI k/— (I P;/m—)+AP/2m)} (51).argument, exp[i(k+I) X;]. This is the same approxi-
W'e now consider the 6rst-order terms arising from Hp t, mation we made in the classical case, when we neglected
and &Pa„ha&(cca„cca„+a»*caa„) These are g. iven by the corresponding terms (23). The justification used for
the classical approximation may be directly applied to
—(i/h)[S {E(&'/2m)+kZA~(A»A»*+A»'A») }] this case. Since, as we have noted, the second-order
commutators arising from H~, c and saba„hca(ca»caa„*
= —(e/m)P(2~A/~L') a{ (A, „+A,„*) +u»*a») yield just —as{ —(i/h)[S, Kr;„c„]},then the
combined contribution of these terms and the first-

X(Pc s») exp(ik X;)}, (52) be —

„
order commutator with 3'.~;„t, to our hamiltonian will
(i/2A)[S, K;,c„] or one-half the sum of (55a)
and (55b).
as may be verified using the commutation rules anal- In addition to the unconsidered higher order commu-
ogous (43) for the new coordinates. But these terms tators in the terms we have already discussed, there are
thus cancel one of the zero-order terms arising from two terms we have not yet investigated. One is a
 D. BOH M AN D D. PI NES
quadratic term in the 6eld variables in II; t„, It is essentially the same as our classical result (32c)
since the hk'/2m term is unimportant for most cases
(4sca/mL') P (h/2aa)(ea„si, )(aa„+a a„*) of interest.
We may neglect the terms arising from the higher
X(a~.+a ~,*) exp[i(%+I) x~]. order commutators we have not yet considered, for
the higher order commutators in that part of Hf;, i~
This term may be neglected because it is quadratic in
given in (56) are of the same character as those resulting
the 6eld variables and contains a phase factor from (55b). Thus, we see all of the higher order commu-
exp[i(k+1) x;] with nonvanishing argument. The other tators will be arithmetical fractions of those resulting
term is from H fipple, and is, to 6rst order,
from [S, [S, Xr;«„]]. But as may be seen from
-'Q(&o '+c'k' — oP)(h/co) fAa Aa *+Aa *Aa (55a) and (55b) the lowest order term arising from
[S, [S, X;«„]]
will be a factor of w%' smaller than
+Aa, A a, +Aa, *A a, * (i/h)— [S', (Aa, Aa„* H;««, and may be neglected, just as (22) was in the
classical case, since its inclusion would lead to a fourth
+Aa, *A a, +Aa, A-a, +A a, *A-a,*)]}. (56) or higher power of s/c in H, «a;~t, .
Thus, in all respects the quantum-mechanical results
On comparing the zero-order terms in (56) with the
quadratic fiel terms resulting from —
are essentially the same as those obtained with our
(i/2h) [S, Kr; &„]
classical treatment of the previous sections. We 6nd
obtained from (55), we see that the sum of these will
that we must carry out the same approximations for
vanish if we take:
this case, and the analysis of H„,t i„t will yield results
aa'= (a,'+ c'k'+ (47re'/mL') P
(k'/P; sa„)'/re'} similar to those of Sec. II.

X }1/(a0 —k P;/m)' —h'k'/4' I. (57) IV. CONCLUSION

This is our quantum-mechanical dispersion relation for In conclusion, we should like to point out that we
the organized transverse oscillations. It is almost have verified, both classically and quantum-mechani-
identical with the analogous classical dispersion relation cally, our qualitative picture of the role of organized
(27), and reduces to the latter as h 4. — transverse oscillations in electron interactions. We have
When we combine all of the terms we have considered seen that, by a suitable canonical transformation to the
thus far, and assume that the frequency of the organized collective variables, we can show that the effects of
oscillations is specified by (57), we obtain magnetic interaction are divided naturally into the two
components discussed earlier:
'Qh(o(A a„A a„*+Aa„*Aa„)
3C= Q(P, 2/2m)+ —,
{1) The long-range part, {)&c/co„). This is responsible for the
long-range organized behavior of the electrons, leading to modified
2%8 (P; ea„)(P,"ea„) transverse Geld oscillations. We may interpret these interactions
as being redescribed in terms of the coordinates of the modified
m'L' '~a~ aP —[(k P m/) +h'k/m2]2 transverse oscillations.
(2) The short-range part, (X&c/~~), given by Hp„t; t, which
Xexp[ik (X;—X;)], (58) does not contribute to the organized behavior, and represents the
residual particle-interaction after the organized behavior of the
where we have used Kqs. (55a), (55b), (56), and (57), system has been taken into account.
and taken one-half of the sum of the terms in (55a) and
(55b). This is just our desired hamiltonian for the Furthermore, in those regions in which organized
collective description. It reduces to the classical hamil- behavior is unimportant (i.e., X(c/co~), the new hamil-
tonian (31) as h — ao. H~„&;« is here given by tonian reduces to the appropriate particle hamiltonian,
in which the electrons interact according to the Biot-
27( 8 (P; sa, )(P; sa, ) Savart law, and the transverse 6elds oscillate with
Hpsrt int frequency co=ck. For, as we have seen, Hp t ' t de-
maLa '~'a~ u)a [(k P;/m)—+hk'/2m]a
scribes the Biot-Savart law for short wavelengths, and
X exp[i(k (X; X;)].. 9)
(5— in this limit our dispersion relation becomes co'= c'k'.