ANNALS OF PHYSICS: !

& 101-128 (1957)

Some Nonseparable Boundary Value Problems and the

Many-Body Problem

LEONARD EYGES

Lincoln LaborUtor~,’ Massachusetts Institute of Technology, Cambridge, Massachusetts

The quantum-mechanical problem of the motion of N interacting particles,

e.g., hard spheres in a box, is a nonseparable boundary value problem in a
3N-dimensional space. The nonseparability is of a special kind in that it in-
volves a number of boundaries, for each of which the wave equation is sep-
arable, but which taken as a whole make up a nonseparable problem. It is
then suggestive for the many-body problem to consider analogous nonseparable
problems in three dimensions, and a method for treating a rather general class
of them is given. This is illustrated by a discussion of the solution of the
Helmholta equation with Dirichlet conditions for a box containing one or more
hard spheres. This method also gives the solution to the general problem of
multiple scattering by a configuration of scatterers of finite range. The diffi-
culties of using perturbation theory with hard spheres are discussed, and a
method is described for getting around these difficulties. This perturbation
method is applied to the problem of two hard spheresin a finite box and the
energy shift is found to lowest order in hard sphere radius for particles with
symmetric and antisymmetric wave functions. In the light of the above re-
sults some qualitative features of the N-body problem are discussed. It is
pointed out that the energy perturbation of the ground state of a dilute hard-
sphere Bose gas is different for periodic and for Dirichlet boundary conditions *
and that this is not a surface effect, in the sense that it does not vanish as the
volume becomes very large. Finally, a connection is shown between the van
der Waals equation and Weyl’s theorems on the asymptotic distribution of
eigenvalues for the Helmholta equation.
INTRODUCTION
The problem of the quantum-mechanical behavior of more than two strongly
interacting particles -is mainly unsolved, although its importance for nuclear
physics and statistical mechanics hardly needs be stressed. It would be particu-
larly interesting to be able to discuss the problem when the forces between the
particles are infinitely repulsive (“‘hard spheres”), even over a very short range,
since then even the conventional perturbation theory breaks down.
1 On leave, as a National Science Foundation Senior Post-Doctoral Fellow. This work
was done at the Department of Physics, Massachusetts Institute of Technology, Cambridge,
Massachusetts.
101
102 LEONARD EYGES

In this paper we present a point of view which suggests a method of attack on

this problem. We observe that the many-body problem is a nonseparable bound-
ary value problem in a multi-dimensional space and that much of the difficulty
with it comes from the nonseparability and not from the many-dimensionality of
the space. This is not a useful observation if we cannot solve nonseparable prob-
lems in two or three dimensions. But there is, in fact, a special class of non-
separable problems which can be set up and to a large extent solved. It consists
of problems in which the nonseparability arises from the fact that there are
several boundaries for each of which the wave equation is separable, but which
taken as a whole make up a nonseparable problem. For example, scattering hy a
sphere is a separable and soluble problem and scattering by a cylinder is a
separable and soluble problem. But scattering by the system of a sphere and a
cylinder is a nonseparable one. With the method described in this paper we can
make progress with some of the problems. This class of problems is far from
being the most general one. The fact that each boundary of the geometry must
itself be separable means that we cannot calculate by this method the scattering
from, say, a cube or tetrahedron, let alone several.
The many-body problem is of this class. It is essentially the same, in principle,
as the kind of two- and three-dimensional problem mentioned above. Hence, a
consideration of the methods for treating these problems is very suggestive for
the many-body problem. It remains to be seen how much the high dimensionality
of the space involved in the many-body problem, which makes no difference in
principle, makes a difference in practice.
This paper is meant to be essentially an expository one, in which we try to
exhibit the thread which ties together a large class of nonseparable problems
including the many-body one. For this reason we have not striven for the utmost
generality, but rather have taken the simplest examples that adequately illus-
trate our viewpoint. Thus, we have confined ourselves to hard sphere interac-
tions, although it is not basically more difficult to treat a more general kind of
short-range interaction. For this same reason we have not worked out our
examples in more detail than is necessary for illustration. We hope to present
more detailed calculations in a future paper.
In this paper, we begin in Section I with a general statement of what we mean
by considering the N-body problem as a non-separable boundary value problem.
In Section II we discuss two rather general kinds of nonseparable boundary value
problems in three dimensions. The first is (Section IIa) the problem of solving
the Helmholtz equation for a box containing a hard sphere, with the boundary
conditions that the wave function vanish on the walls of the box and surface of
the sphere. A basically similar problem is that of the scattering of a wave by a
configuration of scatterers, i.e., the problem of “multiple scattering” and this is
discussed in Section IIb. The point of view developed in Section II leads very
 NONSEPARABLE BOUNDARY PROBLEMS 103

naturally to a method for doing perturbation theory for problems involving hard
spheres either in three dimensions or in a higher number. This perturbation
method is the subject of Section III. In Section IV, we discuss the problem of two
particles in a finite box using the perturbation method discussed in Section III.
Finally, in Section V we make some general remarks on the many-body problem.
I. STATEMENT OF THE MANY-BODY PROBLEM

We begin with the problem of two particles of mass m confined to a box of

volume V = L3. Let the coordinates of the particles be x1 , yl , z1 and x2 , ~2 > 22
defining the vectors rl and r2 . The Schroedinger equation governing the motion
of the particles is
(Vl + IC2)$(rl, r2) = 0
where

and p =2mE
7’

We assume that the particles interact via a “hard sphere potential,” i.e., that
the wave function vanishes when the particles are closer than a distance a,
#=O when 1rl - r2 1 I a. (2)
The condition that the particles be confined to the box is
$=O whenever xl = OorL
x2 = 0 or L (3)
1
L etc.

Equations (1)) (2)) and (3) define our problem. It is a nonseparable boundary
value problem involving a six-dimensional partial differential equat.ion. As we
have remarked in the Introduction much of its difficulty arises not from the six-
dimensionality, although this of course makes actual calculations more tedious
than for a smaller number of dimensions, but from the nonseparability of the
boundary surfaces. It is as easy to find the wave function for 1O23free particles
in a box as it is to find the wave function for one. The real problem begins with
two interacting particles. The second point we wish to emphasize is that the two
boundaries defined by (2) and (3) are individually “separable” in the sense that
one can find convenient solutions of the wave equation for each of them. That
is, calling the walls of the box Boundary A, and the six-dimensional surface
1rl - r2 ( = a, Boundary B, we can easily find solutions of (1) satisfying boundary
conditions on A, using free particle wave functions and on B using relative and
center-of-mass coordinates. The essence of our method is to try to combine these
104 LEONaRD EYGES

two solutions in a way that gives the solution of the truly nonseparable problem,
when boundary conditions must be satisfied on A and B simultaneously.
Conceptually, the problem of more than two hard spheresin a box is the same.
For example, for three hard sphereswe must solve

(09” + k*)rl(rl , r2, f3) = 0

with the wave function vanishing on the three perturbing boundaries defined by
I rl - r2 ( = a, Boundary Blz
1rl - r3 1 = a, Boundary B13
1r2 - r3 1 = a, Boundary B23
and, of course, vanishing on the walls of the box, which we again call Boundary A.
Again we have a seriesof boundaries, for each of which taken by itself the wave
equation is separable, but which taken together make a nonseparable problem.
The same remark applies to the N-particle problem where we have N(N - 1)/2
boundaries corresponding to the particle interactions plus the boundary defined
by the walls of the box.
.I
II. NONSEPARABLE BOUNDARY VALUE PROBLEMS AND
MULTIPLE SCATTERING

A. SOME NONSEPARABLE BOUNDARY VALUE PROBLEMS

We consider some three-dimensional boundary value problems, each of which

suggestsor illustrates a point applicable to the many-body problem. Four typical
problems are illustrated in Fig. 1. For each of them we must find a solution of the
three-dimensional Helmholtz equation
(0,” + k2)rL(r) = 0 (4)
which solution vanishes on the boundaries indicated.
Consider the first problem described in Fig. 1, a spherical cavity A perturbed
by a central sphere B. This is a simple separable problem, but it is instructive
nevertheless. It can be considered as the two-body problem of a point particle
moving in the space between A and B, and colliding via a hard core interaction
with a jLred central particle.2 The general solution $ of the wave equation (4) in
the space between A and B is:
$ = #A + $B (5)
* For the problem of the moving particle the Schroedinger equation is just Eq. (4)) where
k* = (ZmE/k*) and r is the coordinate of the particle. The boundary conditions are that
$ = 0 whenever the moving particle is in contact with the central particle or with the walls,
i.e., they are of the same form as we use in the text.
 NONSEPARABLE BOUNDARY PROBLEMS 105

A
El?
A

0 B 6” B
0 0
00IIIB’
ll!
FIG. 1. One separable and three nonseparable boundary value problems. I. Sphere A
and concentric internal sphere-B. II. Sphere A and offset internal sphere B. III. Sphere A
and several offset internal spheres B, B’, B”. IV. Cubical box and internal sphere B.

where3
(54
(5b)
and the j, and 121are spherical Bessel and Neumann functions. Putting in the
boundary conditions $ = 0 for r = a and r = R leads to the eigenfunctions:

6.3)

where kl, 1sthe nth root of

jdkd jzW(
-=nl(lcR)*
nt(W
8 The real spherical harmonies we use are usually denoted by Yntn , where a, the second
subscript, is the order of the spherical harmonic. In quantum mechanics, on the other hand,
one usually uses complex spherical harmonies Yl,,, where I, the $first subscript, denotes the
order and is of course also the angular momentum quantum number. It is very convenient
to retain 1 to indicate the order, hence we use this instead of n and take a slight liberty
with the sequence of letters in the alphabet, writing the real spherical harmonics as Y,,,r .
106 LEONARD EYGES

It is easy to verify from this that for the radius a of the perturbing sphere small,
the perturbed wave function is expansible in powers of a. The first term in the
power series solution for the perturbed wave function is proportional to a for S
states and to ct3for P states. Thus, considering this as the two-body problem of a
fixed central particle and a moving second particle, the expansion of the per-
turbed k in powers of a goes as the first power of a for Bose particles and as u3
for Fermi particles. One anticipates a similar result for two particles in a box
even though one particle is not fixed, and this turns out to be true.
As we have said, Eqs. (5) give a general solution of the wave equation in the
volume between A and B. But for what follows it will be useful to look at this in a
slightly different way. A general expression (really an integral equation) for a
wave function 9 which vanishes on the boundaries A and B is also given by
Green’s theorem :

In this equation rs is the vector to a point on the surface, gk (r, r,) is the Green’s
function for wave number k, and the integration is over the surfaces of A and of
B. The appropriate Green’s function is
cos k 1r - rs 1
gkhrs)= ,r-r,, .

We shall need the well-known expansions

r, > r > 0 @a>

- k 2 (21 + l)n&)j2(krJPz (cos r), r > ra > 0 (8b)

z=o
where y is the angle between r and rs .
Now suppose we have an arbitrary distribution of (‘sources” on the interior
of sphere A; i.e., suppose (&j/&z) 1r=R is given as an arbitrary function of the
angles & and cpsassociated with rs ;

where the A,t’ are arbitrary coefficients. If we put (9) and @a) into (7), use the
addition theorem of spherical harmonics on the Pl (cos 7) and integrate over the
surface A i.e., over & and cps, we get an expression which is just the form of (5a),
with coefficients A,,,1 which are arbitrary, since the A,,,i’ were arbitrary to begin.
Therefore, we can say that #‘A is the field due to a general distribution of “sources”
on A, i.e., for a$/% an arbitrary function of the surface coordinates. Similarly,
 NONSEPARABLE BOUNDARY PROBLEMS 107

we can ascribe #B to an arbitrary distribution of “sources” on the surface B.

The total field #, which is the sum of #A and J/B , must then satisfy the boundary
conditions on A and on B.
With this point of view in mind it is easy to set up the solution to problem II
of Fig. 1 to which we now turn. Problem II is a truly nonseparable problem:
nonseparable because of the off-center perturbing boundary B. As such, it is
analogous to the problem of two hard spheres in a box, which is also a problem
(in a higher number of dimensions, of course) with one perturbing boundary. We
apply the point of view just discussed to get an (infinite) set of linear equations
whose solution gives the solution to this problem. First, we set up two spherical

FIG. 2. Coordinate systems for problem II of Fig. 1. An equatorial angle p, not shown,
is common to the primed and unprimed systems.

coordinate systems with z axes along the line of centers of the spheresas shown
in Fig. 2. We set up an unprimed (r, e, (p) system with origin at the center of the
large sphere A and a primed4 (r’, O’, p) system with origin at the center of the
small sphere B. We want a general solution to the wave equations in the volume
between A and B. Let J/A be the general field due to an arbitrary distribution of
“sources” on A. Then, as before

L = Clsrn Amljl(kr)Ym~(e, (01. (104

Similarly the general field due to an arbitrary distribution of “sources” on B is

GB = ~l,?n Bmm(kr’)Yh(e’, PO>. (lob)

The total field # at any point is then the sum of *A and tiB , and the boundary
condition requires that $ vanish both on the surface of A and of B.
To apply this condition we express $A in terms of primed coordinates and $B
4 We can take the angle p to be the same in both systems.
108 LEONARD EYGES

in terms of unprimed ones. To do this we use well-known formulas for the be-
havior of the functions that occur in (10a) and (lob), under translation and ro-
tation of the axes.
To begin we use the expressions (1) for transforming the Bessel and Neumann
functions.

and for r > d

$$ = D(1) go (n + 1 + ;) Cnztl” (cos 0) +$ I+$$ Wb)

where D(Z) = 21fli2 4% I’@ + 44)) I’(Z) is the gamma function and C,’ is a
Gegenbauer polynomial, as defined in (1).
What we want is not jl(kr) and nl(kr’) themselves, but these functions times
spherical harmonics so we must use these formulas in conjunction with the equa-
tions for transforming spherical harmonics (2)

Q>= dzs-0cz (2(-)“-“(I + m)!

r~zYmz@‘, - s>!(m + s)!
yn& Q) (124

and
(I + m)!
r’Lr(h Q) = d” 8 (l _ s>!(m + s> I $ ’ yrn&t Q) (1%)
.( >

The second formula is got from the first by interchanging primed and unprimed
variables and letting d go to -d. Now we define $**A as the function $B ex-
pressed in A’s coordinates and define 4 d+B similarly. Then we can write our
boundary conditions
0 = lL.4 + $B*A r=R

0 = ')B + $.4-B r' = a.

Using the formulas (11) and (12) for transformation of the functions from
primed to unprimed coordinates and vice versa, we get

0 = c Am$d~R)Ymde,Q) + g go & (->'-"%n~ (m + 2 + ;)

z.m
(133)
J)(l) (1 + m> !.iz+n(kd) nZ+n(kR) 3 ’ c 2+1/2 ccos e) ym,(e Qo)
>
(1 - s)!(m + s)! (kR)z 0d n
and
 NONSEPARABLE BOUNDARY PROBLEMS 109

0 = cz.m
Bmznz
ql) (1+ m>!jz+?tW
jz+7zW)
; 8c z+1/2 (cose’) y @' cp)
(13b)
n ?n8 , .
(2 - s) !(m + s) !(kq 0

This is not quite the end since under the triple summation in these equations
there enter terms of the form C?“2 (cos 0) Ynzs(O, P) . These must be expanded in
a series of spherical harmonics, after which one can equate to zero the coefficients
of spherical harmonics in each of Eqs. (13a) and (13b). This leads to a homo-
geneous set of linear equations for the A,1 an d B,t and hence to a determinantal
equation for Ire.
These equations are not so formidable in practice as they look here, since as
we shall see, the formally infinite set reduces to essentially a finite one. This
comes about for the same reason that one need take only a finite number of par-
tial waves in calculating the scattering from a potential of finite range, i.e., it
is based on the very convenient property of Bessel functions that for x < 1,
.idx) sz 0. This property applies in Eqs. (13) of course, since they involve the
functions jl+,(lcd) and j,+,(ka). In addition, in transforming the spherical har-
monics according to (12) we see that this involves a series in u/d after the bound-
ary conditions are put in, and this series may be terminated appropriately ac-
cording to the size of this parameter. There is a mild problem of self-consistency
in reducing the infinite set of equations to a finite one, since one must know what
kd and ku are to determine where to break off the equations determining k. In
practice this will generally be easily met by making a rough guess for Ic.
Rather than try to discuss in detail all possible relative values of the param-
eters k, a, d, R, we shall simply indicate the general principles for truncating this
infinite set of equations. First, let us determine the highest value of 1 that we
must include m the B,l , i.e., how many partial waves we must take into account
for the “scattering” from boundary B. We know that in calculating the scatter-
ing of plane waves of wavenumber k from an object of size a, we must use I M Ica
partial waves. Now, we can consider that each point of the surface A is a point
source of a wave which scatters from B. As we shall see in the next section, the
number of partial waves one must take is larger for a point source at a finite
distance than for such a source at infinity (plane wave), but at any rate it is a
definite finite number. Therefore, we can conservatively state: the order of 1
that one must take is not higher than the order needed to calculate the scattering
by B of a spherical wave emanating from the point on A which is closest to B.
The order of 1 may in fact be less than this by reason of special symmetries.
Now we can see that the number of A,1 that one must take is finite. From Eq.
(13a) we see that the coefficients B,l occur in an expansion containing jl+,(kd),
110 LEONARD EYGES

n = 0, 1,2, as factors. Now jl+,(lcd) is essentially zero for I+ n > kd; hence, this
expansion can be terminated and as a result will contain only a finite number of
Gegenbauer polynomials. From this fact it follows that the number of A,* are
limited, although the number of A nl is, of course, not necessarily equal to the
number of B,t , and the infinite determinantal equation becomes an essentially
exact finite equation.
From the above point of view it is clear that problem III, involving more than
one nonseparable boundary, is in principle no more difficult to set up than the
above. For each boundary B, B’, B”, etc., we write a solution like (lob) repre-
senting the radiation from that boundary and an expression like (lOa) for the
radiation from the sphere A. By using formulas (lla) through (12b) one can
write any of these expressions in terms of coordinates centered at any given
sphere, and then the boundary conditions can be put in, giving a set of coupled
linear equations for the amplitudes A,l, B,I , B,II, Bml” etc. These reduce to a
finite set by virtue of the same considerations as given above, and in general
everything is the samein principle but more complicated in practice.
The last problem in Fig. 1, that of a sphere inside a cubical box, is exactly the
same physically as problem II. To solve it we must write a general expression
for the field produced by an arbitrary distribution of “sources” on the interior
walls of the box, add to it an expression like (lob) for the field produced by the
sphere, and apply the boundary conditions to this total field. There is one addi-
tional complication here in that there is no simple expression which is the analog
of (lOa) for the field from the walls. A natural alternative would then be to express
the general field from the walls in terms of the unperturbed eigenfunctions of the
box. These must include eigenfunctions corresponding to Neumann conditions
(&&/an = 0) as well as those for Dirichlet conditions (# = 0), since this general
field does not necessarily vanish on the walls. Having done this, one would have
to express these eigenfunctions in the coordinates appropriate to the sphere.
Although the physics is the sameas in the preceding problems, we have not tried
to carry this out in detail and so cannot guarantee that there will be no difficulties
in practice, for example, in evaluating the integrals that will arise.
What we have done, however, is to take essentially the first step in this pro-
cedure. That is, if the sphere is small, we can assumethat the wave from the
walls is essentially the unperturbed one and can then determine the wave gener-
ated by the sphere. In short, from this point of view, it is easy to seehow to do
perturbation theory for hard sphere interactions. This is discussedmore fully in
Section III.
B. MULTIPLE SCATTERING
The method described in Section II for nonseparable boundary value problems
is, of course, also applicable to the calculation of multiple scattering by a system
 NONSEPARABLE BOUNDARY PROBLEMS 111

of scatterers. In fact, the multiple scattering problem is easier in that the wave
number Ic is prescribed and need not be determined as an eigenvalue.
The general principles are the same.5 We are given a configuration of scatterers
for each of which one could calculate the scattering exactly (e.g., square well
potentials, hard spheres, etc.). A plane wave is incident on this configuration, and
each scatterer becomes a source of a wave field. The total field at any point in
space outside the scattering potentials is that due to the incident plane wave,
plus the field generated by each scatterer. This total field must satisfy the appro-
priate boundary conditions. For example, if the scatterers are hard spheres, the
total field must vanish at the surface of each sphere. If they are potentials of
finite range, the total field and its first derivative must be continuous across the
surface of the potential. Just as in the last section one can apply these boundary
conditions by expressing the waves from one boundary in terms of the coordi-
nates appropriate to other boundaries, and this leads to a set of inhomogeneous
linear equations for the partial wave amplitudes involved.‘j In the last section
the corresponding set of linear equations was homogeneous because there was no
external applied field, such as a plane wave.
To illustrate we calculate the scattering from two hard spheres of radius a, one
at z = 0 and one at z = d on the z axis. We assume an incident plane wave from
z = - 00. It is not essentially more difficult to treat a wave coming at an arbi-
trary angle, for which case all that is involved is use of the addition theorem for
spherical harmonics. Nor is it basically more difficult to treat, say, a square well
potential; in this case one must write interior as well as exterior solutions. More
interesting than either of these elaborations, however, is to study the question,
already considered to some extent in Section IIA, of how the formally infinite
set of equations for the scattering amplitudes breaks down to a finite number.
For this purpose the problem as stated is general enough.
We call A the sphere at x = 0 and B the sphere at z = d. We set up two
coordinat#e systems with common z axes, an unprimed system (T, 0) with origin
at the center of A and a primed one (T’, 0’) with origin at the center of B. Ob-
viously there is no q-dependence. Much as in the previous section we can say
that the general wave #,, emanating from A is of the form

6 The method we have described, when applied to the multiple scattering problem, is
seen to be similar to that used by Korringa, Segall, and Morse, Refs. (S-6) for calculating
the E(k) dependence of Bloch waves in a periodic atomic lattice. This latter problem is
discussed in considerable detail in these references. Since our interest is somewhat different,
in that we are mainly concerned with this problem for the light it throws on the many-body
problem, we confine ourselves to a simple illustrative example.
6 It will be seen that a special case of this method is that due to Foldy, Ref. (6) for cal-
culating the scattering due to a configuration of S-wave scatterers.
112 LEONARD EYGES

and from B
I//B = CL, B zhz(kr’) PL (cos 0’). (14b)
For this scattering problem we use Hankel instead of Neumann functions to
satisfy the boundary condition at infinity. We expand the incident plane wave
iks =
e e ik(z’+d) in primed and unprimed coordinates
eikr = CL0 (21 + l)ilj@r)P1 (cos e>
(15)
= eQd CL (21 + l)iljr(kr’)Pl (cos 19’).
Finally, we need the formulas for expressing $A in B’s coordinates, and vice versa.
These are formulas analogous to (11) except that they involve Hankel functions
instead of Neumann functions. They are

and

!f$ = D(Z) go (-y (n + z + 1) cC+~/~

(~0s
0’)
.-jl+,b’) ht+,(“) / < ,j (16b)
(lcr’)~0”’ -
These formulas must be supplemented by Eqs. (12) for the transformation of
spherical harmonics.
The equations we need can now be written as follows. The total wave ampli-
tudes $ at any point in space is the sum of three terms, the amplitude due to the
incident plane wave, which we call 9, and those due to the fields produced by A
andbyB:

Using the boundary conditions # = 0 for r = a and r’ = a we get two inhomo-

geneous equations for the Al and BZ which are the analogs of Eqs. (13a) and
(13b). In these equations we must equate to zero coefficients of Legendre poly-
nomials. To take a simple illustrative example, suppose a << d and ka is small
enough so that we need include only Ao , B. , Al , B1 . We get for the equations
determining these. quantities

-j&a) = A&&a) + Bojo(ka)ho(kd) - 3B1-“f;) h&id)

- 3&&a) = AlhI + 3Bojl(ka)hl(kd) - 5Bl’q h&d)

 NONSEPARABLE BOUNDARY PROBLEMS 113

- j,(ku)eikd= Boho(ka) + Aojo(ka)hg(kd) + 3fll 2e hdkd)

- 3ijl(ka)eikd = &hl(ka) + 3Aojl(ka)$(kd) + 5A&9 hz(kd) .

Now we can begin to answer the question of how many terms one must take in
the partial wave expansions (14a) and (14b). To begin we observe that there are
three lengths that characterize this problem. These are the radius a of the spheres,
the distance d between them, and the reduced wave length X = (l/k). These
lengths always enter as ratios, e.g., as u/X, d/X, or u/d. Only two of these ratios
are independent, hence we can take as the two dimensionless parameters deter-
mining the properties of the scattering, the quantities ka and kd.
For single scattering ku is the quantity that determines the number of partial
waves that one must take account. For the more general case of multiple scat-
tering ku is still involved, but in addition the parameter kd enters. To see how this
works let us contrast two cases: (a) scattering by hard sphere of radius a of plane
wave, wave number lc; (b) scattering by same sphere of radiation from a point
source a distance d’ from the center of the sphere. First consider case (b). Let
the point source S be at z = -d’ and let the sphere be centered at the origin.
Here R is the vector from S to a point in space, r the vector from the origin to
the same point, and d’ the vector from S to the origin; 0 is the angle between r
and the positive x axis. We have then for the expansion of the point source in
partial waves
ikR
e
_ = ik 2 (-)l(21 + l)jt(kr)hz(kd’)J% (~0s 0).
R I-;0

Note that when kd’ gets very large we have

ikd’
hl(kd’) M (--i)‘+’ e
kd’
and the expansion above becomes
ikR ikd’ co
e
- m eF z i’(2t + l)j,(kr)Pt (cos 0)
R
i.e., just the expansion for a plane wave of amplitude, eikd’/d’, as it must in-
tuitively.
Now let us calculate the scattering from the sphere. If we take the scattered
wave to be of the form
ho = xto Dh(WPz (~0s 0)
we get, on putting in the boundary conditions, that
114 LEONARD EYGES

Dl = G->zik(2z + l)jdka)hdkd’)
07)
h&a)
For kd’ >> 1 the relative absolute values of DO , D, , Dz . . . remain unchanged as
compared to a plane wave. For kd’ small, however, we have

&(kd/) NN 1*1*3'5 *' * (21 - 1)

(kd’)l+l ’

This makes the sequence Do , D1 , Dz . . . decrease rather more slowly than for
a plane wave. To illustrate this we give in Table I the values of 1Dt 1 and cor-
responding plane wave amplitudes for two different cases. 1Do 1 is normalized to
unity for each of these since we are interested only in relative amplitudes.
Now we can use the prescription of Section IIA for determining the number
of partial waves one must take into account. As we have said before we can
imagine that each point of sphere A radiates an S wave, which is scattered by B
and vice versa. Hence, we are safe in taking the number of partial waves from B,
say, to be no more than the number for scattering by an imagined source on
that point of A which is closest to B.
There is one final point we would like to discuss. It is the problem of calculat-
ing multiple scattering given only the scattering amplitude (or phase shift) from
a single scatterer. For scatterers of finite range, the outgoing wave from one
scatterer will always be a sum of partial waves whose radial dependence outside
the range of the scatterer for the Zth partial wave is h l&r). From the quoted prop-
erties of h&) we see that the condition for a second scatterer at a distance d to
be in the wave zone of the first is that kd >> 1. If this is the case, then the scatter-
ing from two scatterers can be calculated knowing only the asymptotic form
of the field from the first. If this is not the case, one must know the local behavior
of the scattered wave. It is only for pure S scattering that the asymptotic wave
function is the local one also. Thus the solution for scattering from two particles

TABLE I
THE AMPLITUDES [ Dt 1 FOR SCATTERING BY A POINT SOURCE AND CORRESPONDING
ABSOLUTE VALVES OF PARTIAL WAVE AMPLITUDES FOR SCATTERING BY A
PLANE WAVE

ko = 0.5 ko = 1.0

hiit=Sgoy5ce Point Source

1 Plane wave kd’ = 1.2 Plane wave

0 1.000 1.000 1.000 1.000

1 0.427 0.226 0.986 0.760
2 0.0464 0.00678 0.278 0.102
3 0.00303 0.0000695 0.044 0.0045
4 0.0044 0.000081
 NONSEPARABLE BOUNDARY PROBLEMS 115

is not given in general by knowing the scattering amplitudes from each; it is

“model dependent” as Drell and Verlet (7) have clearly pointed out, and the
exact behavior of the near fields may be of the greatest importance. In particular,
the assumption of point scatterers or of scattering via “potentials that act only
in X states” can easily lead to fallacious results for multiple scattering from
closely spaced scatterers.
III. A PERTURBATION MICTHOD FOR HARD SPHERES

The method of Section IIA is general in that it leads to a set of linear equations
for the expansion coefficients of the wave function and to a determinantal equa-
tion for the eigenvalues. Nonetheless, it would be useful to have some kind of per-
turbation method for problems such as those illustrated in Fig. 1 when the
perturbing hard spheres are small, as well as for ‘the problem, essentially a three-
body one, of the motion of two small hard spheres in a box. A perturbation
method,’ although intrinsically limited, does have the advantage of giving rela-
tively quick answers for simple problems, and hence of providing orientation and
a check point for more complicated ones.
It is sometimes stated that one cannot do perturbation theory with hard
spheres since this would involve integrating an infinite potential energy in the
matrix elements of ordinary perturbation theory. This is true, of course, but it
simply means that the ordinary form of perturbation theory is unsuitable. It is
equally true that when the radii of the hard spheres becomes very small in
problems such as the one we have been considering, the effect on the energy levels
becomes very small, and it is reasonable to infer that a perturbation expansion
exists.
One procedure that suggests itself is to look on these problems as ones of per-
turbation of boundary conditions. That is, we start with a wave function which
vanishes on the walls of the box (unperturbed boundary conditions) and then we
require that for the perturbed problem the wave function vanish over the per-
turbing spheres as well (perturbed boundary conditions). Such problems are ex-
tensively treated in the literature and there is an excellent summary of the
techniques involved in (2). The basic difficulty with these methods as applied
to this problem is that they converge poorly. In them, the correct wave functions
are expanded in series of eigenfunctions of the unperturbed box. As we shall see,
the perturbed wave function is of such a nature that it does not converge well
when expanded in these eigenfunctions. This convergence difficulty is avoided
if we do not expand the wave function entirely in terms of unperturbed ones, but
instead write the wave function, as we have done, as the sum of two terms, one
term produced by the walls and the other produced by the perturbing boundary.
We turn to this method now.
’ While this manuscript was being prepared, another method of doing perturbation theory
with hard spheres appeared: cf. K. HUANG AND C. N. YANG, Phys. Rev. 106, 767 (1957).
116 LEONARD EYGES

The derivation of the next few equations is standard (2), but we reproduce it
here for convenience. We assume we have a box A perturbed by an interior bound-
ary B. The box need not be three-dimensional, nor need the boundary B. They
might for example be the “box” and six-dimensional boundary defined by Eqs.
2) and (3). We wish to solve
(V” + k2)# = 0 (18)
with boundary conditions $ = 0 on A and B. Here Eq. (18) can be the Hehnholtz
equation for any number of dimensions. We assume we know a complete set of
unperturbed eigenfunctions V)Z (and associated eigenvalues k~) satisfying the
boundary conditions (01 = 0 on the surface of the box. The cpr satisfy

(V” + kl?Ql = 0 (19)

everywhere in the box. From (18) and (19) we get

(P2Ql - QlV2#) dv
k2 - k; = s
J/Q1 dv ’
1
Using Green’s theorem,’ the fact that 1c,vanishes over both A and B, and that
Ql vanishes over A gives

This is a suitable form to begin from for computing a perturbed eigenvalue. To

compute the perturbed eigenfunctions we use

If we expand the Green’s function in eigenfunctions of the unperturbed box,

g,JR, R,) = 4s c Qp;,o~~) (22)

P P

the surface integral reduces to an integral over B alone, and we get, separating
*We use dv and ds to stand for the general “volume” element and “surface” element,
in whatever space we are concerned with, not necessarily three-dimensional. For example,
if the Laplacian in (18) were six-dimensional, then dv would be dxl ... dzg . Similarly we
let R stand for the general “position vector” in this space.
 NONSEPARABLE BOUNDARY PROBLEMS 117

out the Zth term in the sum explicitly:

This is very close in form to the ordinary perturbation theory result for find-
ing the perturbation of the eigenfunction p,(a#/an)
cpl . The integral ds is
f
the analog of the usual “matrix element” and we shall denote i by Ul, ,

Ul, = ~,GVlW ds. (24)

f B
Now, a natural procedure would be to try to approximate # by (PI in the volume
integral of Eqs. (20) and (23) and approximate a#/& by dpr/an in the surface
integral. The first approximation is valid; the second is not. The reason is that
even a very small perturbing object changes a#/& violently, although over a
limited region of space.
Since this is an essentialpoint, we shall illustrate it by reference to problem I of
Section IIA for which we know the exact solution. Let us consider the lowest
state, for which 1 = 0. Then we have from (6), letting #, stand for the ground
state :

$0 = jdkr) - ‘2) no(kr), k’Rj+

-a

The unperturbed wave function is just

$8’ = joCkor>, ko = (r/R).
Now for small r, (&,“/8n) = - (a#~“/&-) is proportional to P.On the other hand,
for the correct wave function, a*,/& goesas l/r2 for small r becauseof the singu-
larity in the Neumann function. Thus it is completely wrong to approximate
a$,/an by a&'/an. Pictorially, what happens to the wave function as the hard
sphere is “switched on” is shown in Fig. 3. For very small hard sphere radius
the wave function continues in almost unperturbed to the hard sphere boundary,
where it suddenly turns down. This sudden turning down corresponds, of course,
to a large kinetic energy, but this is so localized that the total energy perturba-
tion due to the sphere vanishes as a + 0 and a perturbation expansion is possible.
On the other hand, it illustrates clearly why we cannot approximate a&/an
by a&'/an over the surface of the sphere. This difhculty will occur whatever the
shape of the small object; it need not be spherical, nor three-dimensional. Thus,
exactly the same difficulty occurs in the problem of two hard spheres in a box
where the perturbation is the six-dimensional surface 1rl - r2 1 = a .
118 LEONARD EYGES

This difficulty is easily overcome if we look at the problem from the point of
view of Section IIA. For simplicity we shall illustrate this by considering a rather
general problem in three dimensions: an enclosure of arbitrary shape perturbed
by a sphere of radius a with center at r. . We shall use the same idea in the next
section in discussing the two-body problem.
We start with an unperturbed wave function pL(r), corresponding to a wave
generated by “sources” on the walls with no perturbing sphere in the enclosure.
If we imagine the sphere set in place at r. , this wave will scatter from it, and if
the sphere is small compared to a wavelength, it will generate an outoing wave
no(Icr’), where r’ is the radius vector in a coordinate system with origin at the

1.
R
FIG. 3. Perturbation of ground-state wave function by central hard sphere interaction:
GO0 is the ground-state wave function for a spherical box of radius R; $g is the ground-state
wave function when an inner central sphere of radius R/IQ is added.

center of the sphere. Then in the neighborhood9 of the sphere we can write
# = m(r) + ho@-‘). (25)
This is, of course, nothing more than the lowest order approximation to Eqs. (10).
The condition that this wave function vanish over the surface of the sphere gives,
forka<<l.
A = -kwl(ro).
Then over the surface of the sphere, to this lowest order approximation,
(8$/&z) = (l/a) pl(ro) and the matrix element is
U zp = 4?rw &0)&0).

we can now put this into (20) and (23), using to this order of approximation
9 See the last paragraph of this section.
 NONSEPARABLE BOUNDARY PROBLEMS 119

I &CJZdu = 1 to get the perturbed eigenvalue and eigenfunction to lowest

order.
It is interesting to ask whether one could iterate this matrix element, as one
does in ordinary perturbation theory. The answer seems to be that one can iterate
once, to get terms of order a” in the eigenvalue and eigenfunction corrections, but
no higher. To show this we begin in a negative way by pointing out that one ob-
viously cannot hope to get the terms in a3 correctly without including P waves
in the expression (25) for the wave emanating for the sphere, since P-wave
scattering goes as (ka)3 for small a. The iteration of the angle independent S wave
can not reproduce this term. On the other hand, the a” corrections arise physi-
cally as follows. The unperturbed wave scatters from the sphere generating the
wave n&r’) as given by (25). However, the expression (25) does not satisfy the
boundary condition at the surface of the box, and the wave from the small sphere
induces new sources on the inside of the large sphere in such a way that the
boundary conditions are satisfied to order a. These induced sources scatter from
the sphere again giving an a” term, and it is this that the iteration procedure can
take into account.
A final point worth mentioning is that for T small, (25) is an approximate solu-
tion, i.e., it is correct to terms of order a, For example, the lowest ordereigenfunc-
tion for problem I of Fig. 1 is to first order in a

IJ~M jo(kOr) + r$ j,‘(kor) + k0ano(kor)

so for r M a we can drop the second term. As we have said, we can interpret this
as follows: jo(kor) is the incident wave, koano(kor) is the wave scattered from the
small sphere, and (Tar/R’) jo’(kor) is the ‘Cinduced” wave from the large sphere.
Since this last term is of order a2 for T E a, we can neglect it for small r, and
in particular for the purpose of calculating a$/&.

IV. TWO PARTICLES IN A FINITE BOX

We now discussthe problem of two hard spheresin a box. As we have empha-

sized in the Introduction it is a problem involving two boundaries, namely, the
walls of the box and the surface 1rl - r2 1 = a. As such it is similar to the two-
boundary problems discussed in Section IIA, and could in principle be set up
by analogy with them, leading to an infinite set of linear equations relating the
expansion coefficients of the wave functions. The difficulty with this procedure
is the practical one of expressing eigenfunctions appropriate to one boundary in
terms of those appropriate to the other boundary. We think this difficulty is far
from insuperable and that this method offers considerable hope for a complete
solution of the problem. But for a first orientation it is perhaps wiser not to try
120 LEONARD EYGES

to be so general. In this paper, therefore, we shall use the perturbation method

,of the last section, invoking the point of view of Section II to start off the itera-
tion correctly.
We start with an unperturbed wave function cp~. We define the relative vector
.r (components 2, y, z) by r = rl - rz and center-of-mass vector R (components
X, Y, Z) by R = (rl + x$/2. Let us forget for a moment about the requirements
of symmetry on the wave function and take for (ohlo

‘Ph =
02 3
z
hx1
sin __
L
127rx2
sin -
L
.
sin ___
L
ml?ryl
sin
. mzay2
L
sin nz
L
sin n2=2
-
L (26)

where l1 . . . n2 are integers, and the single index X stands for these numbers.
If we imagine the interaction “turned on,” this unperturbed wave function
‘“scatters” from the boundary 1rl - r2 1 = a. As in the three-dimensional case
this effectively determines the wave functions in the neighborhood of the bound-
ary to lowest order. We imagine the radius a is small, and express (OXin the co-
ordinates r and R.

3
cph(R, r) = 2 sin
0L

Then to lowest order” in r

sin 12rX
Psm-sm-
L
. m17rY . m2aY
L L
. n17Z
.sin---sinP
L
. n27rZ
L
(27)

We should emphasize that pk (R, 0) is really defined as the function on the right-
hand side of Eq. (27) for X, Y, Z, between 0 and L and as being zero when out-
side this region. This will be important in taking the Fourier transform of
cpx(R, 0)) which we now do.
00
cpx(R,
0) = (2;)3’a
/.I.. -~ g(d) exp (i6.R) dd.

We wish to find a solution of the wave equation, call it h, corresponding to

the wave “scattered” from the boundary r = a, i.e., such that for small r the
function (PA + #.J approximately satisfies the boundary condition of vanishing
10 We arbitrarily choose a cubical box. It is equally easy to work with boxes of other
shapes, for example spherical.
I1 By pA(R, 0) we mean the function cpA(X, Y, 2, 0, 0, 0).
 NONSEPARABLE BOUNDARY PROBLEMS 121

at r = a. We observe that in R, r coordinates the wave equation is

( v+2 + 2v: -I- lc2 * = 0.

)
Hence, the function
exp (ia.R)no(~r)
where (a2/2) + 2K2 = k2,is an elementary solution of this wave equation, and as a
consequenceso is the function

g(d) exp (id.R) n!?!@?dd.

n&a)
Now (CX+ ti8 obviously satisfies the boundary condition, so +. is the approxima-
tion we seek for the wave function near the boundary r = a. It is essentially the
correlation function. Although this is an interesting and important aspect of the
problem, we shall not discussit further in this paper since we are mainly interested
in finding the energy.
For calculating h? and the first-order connection to the wave function we must
evaluate the matrix element UXX~, defined in Eq. (24), where X stands for the set
of quantum numbers Zr . . . n2 and X’ for the similar primed set. As before, we
cannot use (a#/&) M (+x/&z), but must use the expression (28) for the field
near the boundary to get the dominant contribution. Since at,b/anis just a$/&,
and since (a/&z) (K%(KT)) w l/ r2 f or KI’ < 1, we get, on differentiating under the

integral sign,

a*
- M -+(R,O).
an
We put this into Eq. (24) and approximate (OX* by (ph<(R,0) as is appropriate for
the surface integral. To find the surface element we observe that the volume
element is dzr * * * dz2and in the r, R coordinates this becomesdX dY dZ dx dy dz
since the Jacobian of the transformation is unity. The surface element is then ob-
viously dX dY dZa2 sin 0 de dp. Then we have for the matrix element:
L L L
UXXJ= 47ra cpx(R, 0)+&R, 0) dX dY dZ. (29)
sss
0 0 0

The typical integral that appears in the matrix element is

s0
* sinl~5.sinl~5.sinlzE.sinlz’~dE = a [y(Z1 - &, II’ - &‘) - r(& - 12 ) 11’ + 12’)

- r(Z1 + z2 , h’ - 12’) + r(h + 12 , Zl’ + Z2/)1

122 LEONARD EYGES

where
(0, a # b
y(cz, b) = T/2, a = b # 0
{
(7r, a = b = 0.
From this one gets the perturbation energy for an arbitrary state. For example,
for both particles in the ground state, we get

or

The procedure is the same when we include the effect of statistics. Then, in-
stead of taking the unperturbed wave function as (ox(rl , rz) we must take the
properly symmetrized or antisymmetrized combinations
cph , r2) f cpdr2 , rd.
As before, we imagine these waves scattering from the boundary r = a, and must
set up a wave function like (26) for the radiation from this boundary, which
wave function together with the unperturbed one satisfies the boundary condi-
tion.
We take up the symmetric case first. We call (px8 the symmetrized n, i.e.,
in r, R coordinates
2n”(R r> = AR, r> + PAR 4.
We see that to lowest order
PAYR, 0) = PAR 0).
To this order then, there is no difference between the symmetrized case and the
one we have treated. The expression (27) remains valid (it is properly symme-
tric), and the matrix element is just that given by (29). Thus, (30) represents
the energy perturbation for small a for two symmetric particles.
For two antisymmetric particles we form the antisymmetric q~ , which we call
(pAA.We must also expand this for small r. This case is morecomplicated than the
symmetric one in that (OX* vanishes to zero order and the lowest order terms are
linear in x, y, z. If we do the expansion we get

(31)
n27d
n2 cot ~
L
 NONSEPARABLE BOUNDARY PROBLEMS 123

where
m17rY . m27rY . nurZ . m27rZ
f(X,Y,Z) = L sm L sin L sin L.

loxn is of the form

where 8 and (p are the angles in the relative coordinate system, i.e., x = r cos 0,
etc. Now we can write the approximate wave function in much the same way as
before. We imagine F, , F, , F, each expanded in a Fourier integral. For example,

F,(R)
=& sss G,(d) exp (id *R) dd.

Then a solution +8 of the wave equation, such that (CX~ + & is zero at r = a,
is given by

where, as before, 20’ + (~~/2) = k’. This determines the local solution and we
use this to determine +?/I% over the surface. We get, just as for the symmetrical
case, that a#/&. is independent of K and is given by

w
- x -i qiA(R, r) Jrza (32)
an
where we have used nl(x) CT (1/x2) for small 2. We approximate y? itself by +A
over the boundary and put Eqs. (31) and (32) into the expression (25) for the
matrix element. Doing the angular integrations we get for X = X’

UAh= (g)‘? {~L~L~Lf’(x,Y,z)[(G cot y - l2 cot y )

mmY m2?rY 2
+ ml cot __ - mz cot - (33)
L L )
nlrZ
+ n1 cot __ - n2 cot __ dX dY dZ
L
The first-order correction to the energy of two particles goes as a3, as we surmised
from the example quoted previously of one particle colliding with a jixed central
particle.
For a given hard sphere radius the energy correction is much smaller for par-
ticles in antisymmetric space states than for particles in symmetric ones. The
reason is clear. Antisymmetric space states for two free particles automatically
have a node at the point r = 0. When we introduce the hard sphere interaction,
124 LEONARD EYGES

this node is simply pushed out from the origin to r = a, and this costs relatively
much less energy than to create a node in the wave function, as must be done for
particles in symmetric space states. This result is also interesting in connection
with the problem of why the shell model works, in that it provides an example
of the possibility of introducing very strong forces into a free-particle system
without perturbing the motion very much. Of course the example is academic, in
that one knows that there are strong attractive forces in the nucleus, but at least
it shows that not all strong interactions are inconsist,ent with essentially free
particle motion.

The Many-Body Problem

The point of view we have emphasized is that the many-body problem is a
multiple scattering problem in a many-dimensional space. This wave-scattering
picture is often suggestive. It supplements the alternative view point in which
one thinks of the problem as one in which particles collide and wherein one must
take into account two-body collisions, three-body collisions, etc.
To illustrate this let us look at the general qualitative picture of multiple
scattering in three dimensions, where intuition is of some help, and then extend
this to the many-body problem where intuition is less immediate. We consider
the problem of a three-dimensional box perturbed by interior boundaries, as for
example problem III in Fig. 1. If there are few boundaries and they are far apart
the wave function, as determined by the shape of the box, will not be much per-
turbed. In calculating this perturbation, we can neglect the effect of the boun-
daries on one another, or to put it another way, can neglect the multiple scat-
tering between boundaries. As more boundaries are added, however, and they
become closer and closer, this multiple scattering begins to dominate the behavior
of the wave function and finally the wave function is essentially determined by
the local boundary configurations rather than by the walls of the box.
Analogously for the many-body problem, we have a “box” and the N(N - 1)/2
boundaries Bij , defined by 1ri - rj 1 = a. When these boundaries are small
enough, or not too numerous, we need take into account only the scattering
from each of them and can neglect the multiple scattering between them. This
means, in particle language, that we take into account only binary collisions.
As the boundaries get more numerous and closer together, multiple scatter-
ing among the boundaries becomes more and more important. Now, the multi-
ple scattering between boundaries Blz and B2, is, in particle language essentially
a three-body collision between 1, 2, and 3. When these and higher order colli-
sions become important, then, just as for the three-dimensional case, the local wave
function will be determined mainly by local configurations (in the many-di-
mensional space). and this wave function will not much resemble the unperturbed
one.
 NONSEPARABLE BOUNDARY PROBLEMS 125

It is clear that under this approximation of binary collisions one could dis-
cuss the problem of more than two particles in a box, taking proper account of
the symmetry or antisymmetry of the N-particle wave function. In fact for some
special cases, we can just write down the answer for N particles by multiplying
Eq. (30), which is the result for one pair, by N(N - 1)/2, the number of pairs.
This is correct, for example, for the ground state of Bose particles in a box. For
this case we have for N particles
E = E I, + N(N - 1) .--*
27 ?rah’
N N (34)
2 4 mV
Now, this result for the coefficient of a differs from that given by Brueckner and
Sawada (8) and by Yang and Huang (7) who get for the coefficient of a
N(N - 1) 4auh2
2 mV
i.e., we have a factor 27/4 where they have a factor 4.
We would like to explain this difference. It arises because these authors use
periodic boundary conditions, whereas we use boundary conditions correspond-
ing to particles really confined to a box. The physical explanation is simple. If
we imagine two free particles in a box, with periodic wave functions, then this
corresponds to a density distribution which is constant across the box. If we im-
agine the hard sphere interaction “turned on,” the particles scatter and the
energy of the system is increased by a certain amount. Consider the same phys-
ical picture when we use the correct boundary conditions. We recall that the un-
perturbed wave functions will then have maxima at the middle of the box and be-
come zero at the end. Now if we imagine the interaction switched on, the particles
tend to collide more frequently (they are more bunched in the middle) and the
energy perturbation will be greater as (34) shows.
We would like to emphasize that this is not a “surface” effect in that it does
not disappear if one makes the volume large enough. The effect is simply to
change the coefficient of a by a volume-independent factor. No matter how large
the volume, the ground-state wave function is of the same form and the effect
persists. Secondly, we would like to emphasize that for a gas in a finite volume
the correct boundary conditions must obtain, as the gas becomes progressively
more dilute. Thus if one can use periodic boundary conditions at all with the
method of binary collisions, it can only be done for densities which are not too
high (or the binary collision expansion breaks down) and not too low (or the use
of periodic boundary conditions breaks down). The hope must be that there
is some intermediate region where both approximations are valid. Finally we
would like to remark that this result does not disagree with our usual intuitional
picture of a gas at jinite temperatures, which we imagine to fill a box uniformly.
126 LEONARD EYGES

This result holds only at very low temperature (strictly just at T = 0) and only
for a very dilute gas.
Let us consider briefly the special one-dimensional many-body problem of
hard “spheres” confined to a length L. The reason for discussing the one-dimen-
sional case is mainly pedagogical, in that it suggests a generalization to three
dimensions. It is clear that there is very little real similarity between the physics
of the one-dimensional gas and of the three-dimensional one, essentially because
particles can bypass one another in three dimensions, whereas they are con-
strained to collide, so to speak, in one dimension.
We begin with two hard spheres in one dimension. By this we mean a system
satisfying the wave equation

g + g2 + lc2#= 0
l2
with the boundary conditions
Xl I 0
*=o whenever x2 2 L (36)
1x2 - x1 5 D.
Here xl , x2 are the coordinates referring to the hard spheres, and D is the hard
sphere “radius.” The boundary conditions (36) define an allowed region in the
xl , x2 plane which, it is easy to see, is a right triangle. Thus the problem of two
hard spheres in one dimension is the same as the (nonseparable) problem of the
vibrations of a clamped triangular membrane. In other words, imposing the hard
sphere boundary conditions changes the allowed square region for free particles
to an allowed triangular one.
Of course, the same kind of result applies if we consider more than two hard
spheres. Thus for three particles we have three coordinates x1 , x2 , x3 and the
corresponding three-dimensional Helmholtz equation. The boundary conditions
are that the wave function vanish whenever x1 5 0, x3 2 L, x2 - x1 5 D,
x3 - x2 5 D. These boundary conditions define as the allowed region in x1 , x2 , x3
space an irregular polyhedron, whereas for the unperturbed problem the cor-
responding surface is a cube. For more than three particles one cannot so easily
visualize the results, but it is clear that they are essentially of the same nature.
The introduction of interactions means we must solve the Hehnholtz equation
with the boundary conditions that it vanish over an irregular, nonseparable
region.
In fact, these problems are soluble (9). But suppose for a moment that they
were not. Is there anything we can say about, them in a general way? The answer
is yes. For to all these problems we could apply Weyl’s famous theorems on the
asymptotic distribution of eigenvalues for solutions of the Helmholtz equation
 NONSEPARABLE BOUNDARY PROBLEMS 127

with Dirichlet conditions (2). For the two particle case, for example, the asymp-
totic distribution of eigenvalues for the triangle is the same as for a square having
the same area as the triangle. Similarly for three particles in a box, we can say
by Weyl’s theorem that the asymptotic distribution of eigenvalues for the al-
lowed polyhedron is the same as for a cubical box of the same volume. And so it
goes.
Let us extend this idea from one dimension to the general problem of N hard
spheres of radius a moving in a box. If for a moment we imagine the hard sphere
diameter to be zero (free particles), we must solve the SN-dimensional Helmholtz
equation satisfying the boundary condition # = 0 on the surface of a hyper-
volume defined by the equations x1 = 0 and L, x2 = 0 and L, etc. Now, just as
in the one-dimensional examples, when we imagine the diameter of the hard
spheres to grow from zero to a, we can say that we change the shape of this
hypervolume in a complicated way, which is not easy to visualize. Nonetheless,
we assume we can apply Weyl’s theorem to it. According to Weyl’s theorem, the
asymptotic level density for solutions of Helmholtz equation in a volume (or
hypervolume) depends only on the magnitude of that hypervolume. We can
relate this to thermodynamics by using the well-known relation between entropy
and the logarithm of the level density. Now, for free particles the entropy as a
function of the hypervolume Vh , Vh = VN is given, except for some irrelevant
additive constants, by

NE, vh>
= ‘G In E + In Vh . (37)
k

For the case when there are N particles of diameter a in a volume we can calcu-
late (10) for small a
4
In Vh = N In V - N$, w = - 7ra3.
3
Equation (37) leads, of course, to the equation of state of the ideal gas. If
instead we use in (37) the expression (38) for In Vh , it is easy to verify that one
gets the second virial coefficient as well, i.e., one gets essentially Van der Waals
equation for dilute gases. This is not a far-reaching result, but still it is perhaps
interesting to see it in a new light.
Finally, we would like to remark on an approximation suggested by the point
of view of the many-body problem as one of multiple scattering from boundaries
in a hyperspace. We have seen that in the neighborhood of such a boundary,
the wave function varies rapidly, i.e., it has high Fourier components. Moreover,
this behavior is essentially independent of the incident wave from the walls.
That is, the singular field in the neighborhood of a boundary is determined by the
shape of the boundary, and to calculate it we do not have to know the wave
128 LEONARD EYGES

function everywhere. It seems reasonable then that for processes such as the high
energy nuclear photoeffect, which involve high Fourier components, that the
wave function is accurately enough given by the “local field” of the boundary.
That such an approximation will give qualitatively correct results seems assured
from the success of the phenomenological treatments of high-energy processes
by Levinger (11), and others. The point of view of this paper perhaps will give
some confidence that such phenomenological treatments have a fundamental
basis in theory.
ACBNOWLEDGMENT
I would like to thank Professor ‘Herman Feshbach for discussions and for commenting
on the manuscript.

RECEIVED: March 11, 1957

REFERENCES
1. A. ERDELYI (Ed., Bateman Project Staff), “Higher Transcendental Functions,” Vol. II,
cf. p. 101. McGraw-Hill, New York, 1954.
2. P. MORSE AND H. FESHBACH, “Methods of Theoretical Physics,” McGraw-Hill, New
York,1954.
5. J. KORRINQA, Physica 13, 292(1947).
4. B. SEOALL, Phys. Rev. 106,108 (1957).
6. P. M. MORSE, Proc. Natl. Acad. Sci. U.S. 43, 296 (1956).
6. L. FOLDY, Phys. Rev. 67, 107 (1945).
7. S. D. DRELL AND L. VERLET, Phys. Rev. 99, 849 (1955).
8. K. A. BRUECKNER AND N. SAWADA, reported at Stevens Institute of Technology, Sym-
posium on the Many-Body Problem.
9. T. NAGAMIYA, Proc. Phys.-Math. Sot. (Japan) 22, 705 (1940).
10. J. C. SLATER, “Introduction to Chemical Physics,” cf. p. 193. McGraw-Hill, New York,
1939.
11. J. S. LEVINGER, Phys. Rev. 84, 43 (1951).