Stochastic Variational Approach To Quantum-Mechanical Few-Body Problems PDF

Preface
The variational method is one widely used techniques to

of the most
solve the eigenvalue problems of quantum-mechanical Hamiltonians.
Its popularity is due to its simplicity and flexibility. The most crucial
point in the variational approach is the choice of a variational trial
function. One usually attempts to construct the trial function from
some adequate basis functions which contain a number of nonlinear
parameters. The direct method of diagonalizing the Hamiltonian ma-

trix on such a basis set may not be feasible, except for simple systems,
because of the large number of degrees of freedom involved in specify-
ing the system. One thus faces a problem of selecting the most suitable
basis set. It is by the stochastic variational method, that is, by a trial
and error procedure with an admittance test that we give an answer to
this problem. The stochastic variational method has been developed
through the search for precise solutions of nuclear few-body problems.
In this method we set up the basis element one after the other because
it enables us to test many parameters as fast as possible and moreover
to monitor the energy convergence.

The aim of this book is to give a unified and
reasonably simple
recipe for solutions of
few-body problems with the use
bound-state
of the stochastic variational method and to present its application to
various few-body problems which one encounters in atomic, molecular,
nuclear, subnuclear and solid state physics.
Though a unified approach quantum systems is in
to the diverse
general extremely difficult and challenging, great advances have been

made in recent years, especially for systems of a few particles and it has
become possible to obtain accurate solutions for the eigenvalue prob-
lem of various quantum-mechanical Hamiltonians. The main interest
in the few-body problems lies in, e.g., finding an accurate solution for
the system so as to understand the dynamics of its constituents, test-
VI Preface
ing the equation of motion and the conservation laws and symmetries,
or looking for unknown interactions governing the system.
Quantum mechanics plays a fundamental role in atomic and sub-

atomic physics. It is via quantum mechanics that one can understand
the binding mechanism of atoms, molecules and atomic nuclei, that
is, the structure of the building blocks of matter. The interaction be-
tween the particles depends on the system: For example, the long-
range Coulomb interaction dominates in atoms and molecules but the

very different mass ratio of the electrons and the atomic nucleus plays
a key role as well. In contrast, the protons and neutrons in nuclei have
almost equal masses and the interaction between them is short-ranged.

The variational foundation for the time-independent Schr5dinger
equation provides a solid and arbitrarily improvable framework for
the solution of diverse bound-state problems. As mentioned above,
the most crucial point in the variational approach is the choice of the
trial function. There are two widely applied strategies for this choice:
One is to use the most appropriate functional form to describe the
short-range as well as the long-range correlations among the particles.
Such calculations, however, are fairly complex for systems of more
than three particles, and the integration involved is performed by the
Monte Carlo method. Another way is to approximate the solution as
a combination of a number of simple basis states which facilitate the
analytical calculation of matrix elements. We follow the latter course

in this book and show that the stochastic variational method selects
the most important basis set without any bias, keeps the dimension of
the basis low and, most importantly, provides a very accurate solution.
The book is conceptually divided into two parts. The first seven
chapters present the basic concepts of the variational method and the
formulation using Gaussian basis functions. The latter four chapters
of the book cover applications of the formulation to various quantum-
mechanical few-body bound-state problems. In Chap. 2 a general for-
mulation is developed to express the physical operators which are
needed to specify the Hamiltonian in terms of an arbitrary set of inde-
pendent relative coordinates. The linear transformation of the relative
coordinates induced by the permutation of identical particles is also
established in this chapter. In Chap. 3 we review the basic princi-
ples of the variational method with particular emphasis on the case
where the variational trial function is given as a linear combination of
nonorthogonal basis functions. We introduce in Chap. 4 a key algo-
rithm used in this book, the stochastic variational method, and show
Preface Vii
that its trial and error search procedure makes it possible to select the
most important basis functions without any bias in the function space
spanned by the basis functions. Some other methods to solve few-body
problems arebriefly introduced in Chap. 5. Chapter 6 defines the type
of variational trial functions used extensively in the book, the corre-
lated Gaussians and the correlated Gaussian-type geminals. They are
chosen because they enable us to evaluate matrix elements analytically
and because they provide. us with precise solutions for most problems
of real interest. A simple but powerful angular function is introduced
to describe orbital motion with nonzero orbital angular momentum.
To facilitate the systematic and unified evaluation of matrix elements,
it is shown that the above Gaussian basis functions are all obtained
from a generating function. In Chap. 7 we show that the
generat-
ing function plays a deriving the matrix elements of the
vital role in
Gaussian basis functions for an N-body system of essentially any in-
teraction. Explicit formulas axe given in this chapter for the simplest
possible Gaussian basis functions, because they are already found to
be very useful. The matrix elements for a, general case are detailed
in the appendix. We show also in this chapter that the. method can
be extended to evaluate the matrix elements of nonlocal potentials
and the seniirelativistic kinetic energy as well. Chapters 8-11 present
application of the stochastic variational method to various systems:
small atoms and molecules (Chap. 8), baryon spectroscopy (Chap. 9),
excitonic complexes and quantum dots in solid state physics (Chap.
10), and nuclear few-body problems (Chap. 11).
We hope that this book will be found useful by students who want
to understand and make use of the variational approach to quantum-
mechanical few-body problems, while it may also be of interest to
researchers who axe familiar with the subjects. It will be our pleasure
if this book serves to bridge the gap between graduate lectures and the
literature in scientific journals, as well as to give impetus to further
development in the deeper understanding of quantum-mechanical few-
body systems. We assume that readers have taken courses on quan-
tum mechanics and mathematical physics at an undergraduate level.
No special knowledge is assumed of, e.g., atomic physics or nuclear
physics. To help readers to understand the text, we have attempted

to make the book self-contained, put as much emphasis as possible on
clarity, and given several Complements of an explanatory nature. The

Complements are intended to further develop or to reinforce the argu-
ments and ideas presented in the text. We have collected the derivation
VIII Preface
of formulas that may possibly be difficult for readers as exercises with

solutions at the end of the chapters.
Depending on their interest, readers may adopt several reading
strategies. A thorough-going reader is advised to read all of the text
including the Complements. A reader who wants to understand only
the basic formulation can omit the Complements. Anyone who is fa-
miliar with the variational method may skip Chaps. 2 and 3. Experts,
or those readers who are only interested in the performance of the
stochastic variational approach or the physical consequences of the

results, may skip Chaps. 2-7.
The Gaussian basis has long been used in many areas of physics.
The correlated Gaussians were first introduced in quantum chemistry
by S.F. Boys and K. Singer. The application of the Gaussian basis
is one of the key elements of the success of the ab initio calculations
in quantum chemistry. The stochastic variational method is actually
very similar to the so-called "random tempering", that has been used
to find the optimal parameters of the basis in quantum chemistry. In
the random tempering pseudo-random parameters axe generated, and
the best basis functions are selected by sorting out the states which
improve the energy. There exists another method which is also similar
to the stochastic variational method, called the stochastic diagonaliza-
tion. This method, originally developed in solid state physics, attempts
to find the lowest eigenvalue of huge eigenvalue problems by randomly
testing the contributions of different states. The random selection of

the parameters of a Gaussian basis, named the stochastic variational
method, was first used by V.I. Kukulin and V.M. Krasnopol'sky in
nuclear physics.
We are grateful to Prof, V.I. Kukulin for his interest and encourage-
ment during this work. We are indebted to our collaborators, K. Arai,
Y. FujiNvara, L.Ya. Glozman, R.G. Lovas, J. Afitroy, C. Nakamoto, H.
Nemura, Y. 0hbayasi, Z. Papp, W. Plessas, G.G. Ryzhikh, N. Tanaka,
J. Usukura, and R.F. Wagenbrunn for useful discussions and for many
of the calculations which are included in this book. We wish to ex-
press our hearty thanks to Prof. K.T. Hecht, Prof. D. Baye, Prof.
M. Kamimura and Prof. R.G. Lovas for their careful reading of the
manuscript and for much advice and many comments. Despite all these
efforts mistakes may still remain. Needless to say, these are our own
responsibility. Suggestions and criticisms from our readers would be
welcomed. We are grateful for the use of the computer VPP500 of
the Institute of Physical and Chemical Research (RIKEN). One of the
Preface Ix
authors (K.V.) is grateful for the hospitality of the Physics Depart-

ment of Niigata University and the Linac Laboratory of RIKEN. We
would like to thank the funding agencies, Grants-in-Aid for Scientific
Research of the Iffinistry of Education, Science and Culture (Japan),
OTKA Grants (Hu gary), Japan Society for the Promotion of
the
Science (JSPS), the Science and Technology Agency (STA) of Japan,
the U. S. Department of Energy, Nucleax Physics Division, and the
Hungarian Academy of Sciences (HAS), for their support which has
been vital for the completion of the work. Finally, without the support
and patience of our families, this book would not have been possible.
Niigata Y. Suzuki
Argonne K. Varga
September 1998
Table of Contents
I.. Introduction ........................................

1
2. Quantum-mechanical few-body problems ...........

7
2.1 Hamiltonian ...................................... 7
2.2 Relative coordinates ..............................
9
2.3 Symmetrization ..................................
15
2.4 Permutation of the Jacobi coordinates ..............
16
Complements .........................................
18
C2.1 An N-particle Hamiltonian
in the heavy-particle center coordinate set ...........
18
C2.2 Canonical Jacobi coordinates ......................
19
3. Introduction to variational methods ................

21
3.1 Variational principles .............................
21
3.2 The variance of local energy .......................
30
3.3 The virial theorem ................................
33
4. Stochastic variational method .......................

39
4.1 Basis optimization ................................
39
4.2 A practical example ...............................
43
4.2.1 Geometric progression ......................
44
4.2.2 Random tempering .........................
47
4.2.3 Random basis ..............................
47
4.2.4 Sorting ....................................
50
4.2.5 Trial and error search .......................
50
4.2.6 Refining ...................................
53
4.2.7 Comparison of different optimizing strategies . .
54
4.3 Optimization for excited states ................. ....
56
I
Complements .........................................
61
XII Table of Contents
C4.1 Minimization of energy versus variance ..............

61
5. Other methods to solve few-body problems .........

65
5.1 Quantum Monte Carlo method:
The imaginary-time evolution of a system ...........
65
5.2 Hyperspherical harmonics expansion method .........
67
5.3 Faddeev method ..................................
70
5.4 The generator coordinate method ...................
72
6. Variational trial functions ..........................

75
6.1 Correlated Gaussians
and correlated Gaussian-type gerninals ..............
75
6.2 Orbital functions with arbitrary angular momentum . .
82
6.3 Generating function ..............................
87
6.4 The spin function ................................
94
Complements .........................................
96
C6.1 Nodeless harmonicoscillator functions as a basis .....
96
C6.2 Solidspherical harmonics ..........................
105
C6.3 Angular momentum recoupling .....................
106
C6.4 Separation of the center-of-mass motion
from correlated Gaussians .........................
112
C6.5 Three electrons with S =
1/2 ......................
115
C6.6 Four electrons in an arbitrary spin arrangement ......
115
C6.7 Six electrons with S = 0 ...........................
116
Exercises ............................................
118
7. Matrix elements for spherical Gaussians, ............

123
7.1 Matrix elements of the generating function ..........
123
7.2 Correlated Gaussians ..............................
125
7.3 Correlated Gaussians in two-dimensional systems . . . . .
129
7.4 Correlated Gaussian-type gendnals .................
131
7.5 Nonlocal potentials ...............................
134
7.6 Semirelativistic kinetic energy ......................
137
Complements .........................................
143
C7.1 Sherman-Morrison formula ........................
143
Exercises ............................................
145
Table of Contents XIII
8. Small atoms and molecules .........................

149
8.1 Coulombic systems ....... ........................
149
8.2 Coulombic three-body systems .....................
150
8.3 Four or more particles ............................
154
8.4 Small molecules ................................... 165
Complements .........................................
167
C8.1 The cusp condition for the Coulomb potential ........
167
C8.2 The chemical bond: The H,+ ion ....................
169
C8.3 Stability of
hydrogen-like molecules .................
171
C8.4 Application of global vectors to muonic, molecules ....
174
9. Baryon spectroscoPy ................................

177
9.1 The trial function in the constituent quark model .....
178
9.2 One-gluon exchange model ........................
178
9.3 Meson-exchange model ............................
181
10. Few-body problems in solid state physics ...........

187
10.1 Excitonic complexes ..............................
188
10.2 Quantum dots ...................................
191
10.3 Quantum dots in magnetic field ....................
196
10.4 Quantum dots in the generator coordinate method ....
202
Complements .........................................
204
C10.1 Two-dimensional electron motion in a magnetic field. 204
11. Nuclear few-body systems ..........................

213
11. 1 Introductory remark on nucleon-nucleon potentials ...
213
11.2 Few-nucleon systems with central forces .............
216
11.3 Realistic potentials ...............................
223
Complements .........................................
230
C11.1 Correlations in few-nucleon systems ...............
230
C11.2 Convergence of partial-wave expansions ............
233
C11.3 Quark Pauli effect in s-shell A hypernuclei ..........
239
C11.4 The "C nucleus as a system of three alpha-particles .
242
Appendix ...............................................
247
Matrix elements for general Gaussians ..................

247
A.1 Correlated Gaussians .............................
247
A. 1. 1 Overlap of the basis functions ................
247
A.1.2 Kinetic energy .............................
249
XIV Table of Contents
A.1.3 Two-body interactions ......................

250
A.1.4 Density multipole operators ..................
256
A.2 Correlated Gaussians with different coordinate sets ...
257
A-3 Correlated Gaussian-type geminals .................
262
AA Spin matrix elements .............................
263
A.5 Three-body problem with central, tensor
and spin-orbit forces ..............................
265
Complements .........................................
280
CA. 1 Matrix elements of central potentials ...............
280
CA.2 Matrix elements of density multipoles ..............
283
CA.3 Overlap matrix elements of the correlated Gaussians
for
a three-particle system .........................
285
Exercises ............................................
288
References ..............................................
299
Index ...................................................
307
1. Introduction
There are a countless number of examples of quantum-mechanical few-

body systems: constituent quarks in subnuclear physics, few-nucleon
or few-cluster systems in nuclear physics, small atoms and molecules
in atomic physics or few-electron quantum dots in solid state physics,
etc. The intricate feature of the few-body systems is that they develop
individual characters depending on the number of constituent parti-
cles. The mesons and baryons, the alpha-particle and the 'Li nucleus,
or the He atom and the Be atom have very different physical proper-
ties. The most important causes of these differences are the correlated
motion and the Pauli principle. This individuality requires specific
methods for the solution of the few-body Schr6dinger equation. Ap-
proximate solutions which assume restricted model spaces, mean field,
etc. fail to describe the behavior of the few-body systems.
The goal of this book is to show how to find the energy and the
wave function of any few-particle system in a simple, unified approach.
The system will normally be in the minimum energy quantum state.

As forewarned, however, to find this state, the ground state, is a com-
plicated matter. The present stage of the development of computer
technology, however, makes a very simple approach possible: Searching
for the ground state by "gambling". Without any a priori information
on the true ground state, completely random states are generated.
Provided that the random states axe general enough, after a series of
trials one finds the ground state in a good approximation. The reader
may find this a little suspicious but there are indeed a number of fine
tricks in the trial and error procedure which makes the whole idea
really practicable.
Before bombarding the reader with
sophisticated details, let us
demonstrate the random search with example. Let us try to de-
an
termine the energy of a Coulombic three-body system: a positron and

two electrons, the positronium negative ion, Ps-. The example is sim-
Y. Suzuki and K. Varga: LNPm 54, pp. 1 - 6, 1998

© Springer-Verlag Berlin Heidelberg 1998
2 1. Introduction
ple enough so t1i at a graphic illustration of the wave function can be

given, but its solution is far from being trivial. The ground state of
this particular system can be calculated by different methods accu-
rately. In the rest of this chapter we will give just the outline of the
gambling method without paying much attention to the details.
To look for the wave function of the ground state of the system, we
generate random functions. The functions are random in the sense that
we assume some functional form which depends on some discrete or
continuous parameters or "quantum numbers" and these paxameters

a.re randomly chosen. For example, we assume a simple Gaussian form
Tf (a, 0, x, y) = e`2 - 3Y2
7 (1.1)
where x =
1r2 -7'31 denotes the distance between the two electrons and
Y =
I (r2 +r3) /2 r 1 1 denotes the distance between the center-of-mass
-
of the electrons and the positron.

A certain number of parameter sets a and randomly chosen
axe
and then the energy expectation values of the generated states are
calculated and compared. The one among them giving the lowest value
is selected to be a successful paxameter set. Figures 1.1(a)-(d) show
examples of four random states and their energy expectation values

which are obtained with such successful paxameters. Not surprisingly,
the energies of the "configurations" appreciably depend on the shape
of the functions, that is, on the random paxameters. The lowest energy,
-0.11 in atomic units (a.u.) in Fig. 1.1(d), is higher than that of
the exact ground state (-0.262). Here in the atomic units M, (the
electron mass), e (the electron charge magnitude), and h axe chosen
to be basic quantities of units, so the unit of length is Bohr radius
ao =
h?/(Mee2) 5.29x1O-11 m, theunit of energyis me4/h? 27.2
-
=
eV, and the unit of time is h3/(me4) 2.42 x 10-17 S. (Actually the
=
minimum energy calculated analytically with one term of Eq. (1.1) is
-0.177, but it may not be known for a general case.) That means that
the trial function is not general enough to describe the ground state.
To improve the trial function, let us take a linear combination of
two of the above functions of Eq. (1.1), where one of the two is fixed
as the one already selected and another is newly selected after a num-
ber of random trials. Figures 1.2(a)-(d) show examples for random

wave functions which are calculated by the two terms. The energy
improved because the model space increased but we still miss a sub-
stantial amount of binding energy. Increasing the model space further
by adding more and more functions to the linear combination one by
one, we reach the exact energy and the wave function (Fig. 1.3) with
1. Introduction
E--0.09 E--3.15
E=0.18 E--0.11
Fig. 1-1.Examples of the energy expectation value and the (normalized)

wave functionxyTf (a, 3, x, y) of Ps- for one random basis state. Figure
1(a) is in the upper left, 1(b) in the upper right, 1(c) in the lower left,
and 1(d) in the lower right. In each section the x axis denotes the distance
between the two electrons and the y axis denotes the distance between the
center-of-mass of the two electrons and the positron. Atomic units are used.
1. Introduction
E-0.14
E=-0.18
0.1
0.0
B-0.12
E=-0.17
0.1
0.0
Fig. 1.2. Examples of the energy expectation value and the wave function,
xyjcj-T1(aj, Oj y) +c2Tf(a2, 02 Xi Y) ji of Ps- for combinations of two ran-
-, x, 1
dom. basis states. Figure 2(a) is in the upper left, 2(b) in the upper right,
2(c) in the lower left, and 2(d) in the lower right. See also the caption of
Fig. 1. 1.
1. Introduction 5
a combination of about 150 functions of the form (1.1). The conver-
gence of the energy versus the number of the functions in the linear
combination is shown in Fig. 1.4. The energy gain is large in the first
few steps and then it slowly approaches the exact energy.
E=-0.262
Fig. 1.3. The energy and the wave function of Ps- obtained by the trial
and error selection of 150 basis states
The alert reader may question the importance of the steps that
are taken to increase the model space and can ask why not start
with, say, a linear combination of 150 functions and then, by random

trials of all the parameters, one may find the solution. The reason for
increasing the number of functions in the combinations is to control
the "convergence" of the energy. By comparing the energy gain in the
successive steps one can guess how far the exact ground-state energy
is. There is no guarantee that the ad hoe 150 functions would be
6 1. Introduction
-0.250
-0.252
M
41
-0.254
-0-256
-0.258
Z
-0.260
-0.262
20 40 60 80 100
Dimension of the basis
Fig. 1.4. The energy convergence of Ps- as a function of dimension of basis

states that are increased one by one with the trial and error selection. The
solid, dashed and dotted curves correspond to three different random paths.
sufficient to reach the solution in cases where the exact energy is not
known.
A
skeptic may say that one should, instead of the above gambling
method, try a deterministic parameter search such as that furnished
by the Newton or conjugate gradient method. While this may be true
for small systems, the random trials axe more successful in most of
i I
the by avoiding being trapped

cases in the
omnipresent local a.
Moreover, one may have discrete paxameters or quantum numbers

where the deterministic seaxch may not be suitable.
One may also wonder if the solution we reach is really (close to)
the ground state energy. These and other questions will be answered
in the succeeding chapters.
2.Quantum-mechanical few-body
problems
As first step toward the goal of giving a unified and reasonably

a
simple recipe for solutions of few-body bound-state problems the basic

notations and concepts axe introduced here.
The motion of the few-body system isgoverned by the Hamil-
ton operator (Hamiltonian) and described by the eigenfunction of the
Hamiltonian. This wave depends on the positions and other
function
degrees of freedom of the particles. See, for example, [1, 21 for text-
books on quantum mechanics. One can define the positions of the
particles in several different ways by using single-particle or relative
coordinates. The single-particle coordinates are useful if the particles
move ah-nost independently of each other. The relative coordinates
are advantageous to emphasize correlations between the particles and
to separate the center-of-mass motion. One can relate the different

coordinate systems to each other by linear transformations. The defi-
nitions and the properties of these transformations are elaborated in
this chapter.
One often deals withindistinguishable particles. To comply with
the Pauli principle the wave function has to be properly symmetrized.
In the spatial paxt of the wave function the symmetrization induces
permutations of the coordinates. It is shown that these permutations
can be imposed by linear transformations in the relative coordinate
space.
2.1 Hamiltonian
Let us consider isolated system of N particles with masses ml.,

an
-, mAT
and let 7,1,, ,,rN denote the position vectors of the particles.
*
All of the particles may be identical or different, or some group(s)

of them may be identical. The Hamiltonian of the system, with the
center-of-mass kinetic energy Tcm being subtracted, reads as

8 2. Quantum-mechanical few-body problems
N 2 N
H=E 2mi -T,. + Vij. (2.1)
As the interaction unchanged by the displacement of the

Vij remains
origin, the above Hamiltonian is translationaUy invariant and depends

on the internal degrees of freedom only.
One may encounter few-body, problems where the system is sub-

jected to some external field or the particles move in a single-particle
potential. The Hamiltonian then reads as
IV 2 N N
H =
E2 + 1: Ui + 1: Vij. (2.2)
i=1 i=1 i>i=l
Even with the presence of the one-body potential Ui it may happen

that one wants to remove the contribution from the center-of-mass
motion. This will be considered later in evaluating the matrix elements
of various operators. An extension of the nonrelativistic kinetic energy
to the relativistic kinematics will be discussed in Sect. 7.6.
Three-body
potentials can be treated in principle but axe suppressed for the sake
of simplicity.
The two-body potentials are assumed to be local (but nonlocal
potentials will also be considered in Sect. 7.5) and can in general be
expressed as
Vij =
1: VP(ri -
rj) 0?.
P
=
1: f VP (r) J(ri -
rj -
r) 0?. dr. (2.3)

P
Here p is the short-hand notation to specify the component of the

potential which is characterized by 0?. (e.g., central, spin-orbit, etc.
and VP(r) is the corresponding form factor. To specify the particle
motion, several degrees of freedom are in general needed besides the
spatial one. For example, the nucleon has spin and isospin degrees
offreedom, and the quark has spin, flavor and color degrees of fre&-
dom. The specific potentials related to nuclear, subnuclear and other
systems will be discussed in later sections.
2.2 Relative coordinates 9
2.2 RelatiVe coordinates
The separation of the center-of-mass motion is

important to describe
intrinsic excitations of the system. The center-of-mass motion is taken
care of most conveniently by introducing relative and the center-of-
mass coordinates;i (XIi X2 xN). The symbol- stands for a trans-
pose of a matrix. In particular, x is often used to stand for an N x 1
one-column matrix whose-ith element is xi, whereas ,'c a I x IV one-
row matrix. Here XN is chosen to be the center-of-mass coordinate of

the system and the rest of the coordinates fXI, - -
X N-I I is a set of
-,
independent relative coordinates. They axe in general related to the

single-paxtide coordinates i -=-
(rj, -' TN) by a linear transformation:
xi Uijrj, ri =
E (U-I)ijxj (i N). (2.4)
j=1 j=1
We show two examples of relative coordinates x. See Fig. 2. 1. One

is the Jacobi coordinate set, which is defined by the matrix
C
2
4
3
0 0
(a) (b)
Fig. 2.1. Examples of relative coordinates for the four-particle systein. (a)
the Jacobi coordinate set, xi =rl -r2, X2:-- (MIrI +M2P2)/(MI +M2)-
'r3 7 X3 = (Tnlrl + M2r2 + M3 r3)/(Ml + M2 + M3) -P47 and (b) the heavy-
particle center coordinate set, xi = rl r47 X2 = r2 r47 X3 = r3 r4-

- - -
1 -1 0 ... 0
MI M2
0
M12 M12
Ui M2
(2.5)
M1
M12 ... N-I M12 ... N-I
M1 M2 MN
... ...
Tal2---IV Tnl2---.LV M12 ... N
Its inverse matrix is given by

M2 M3 MIV
...
M12 M123 M12 --- M
Tni M3 MN
...
M12 M123 M12 ... N
(Uj) M 12
(2.6)
0
M123
0 0 -"M12 ... N-1
M12 ...N
where M12 i Tnl+Tn2+

...
+mi and, especially, 7nl2---N is the total
-
mass of the system. Another is the "heavy-particle center" coordinate

set defined by
1 0 0 -1
0 1 0 ... -1
UC - (2.7)
0 0 ... ... -1
Mj_ M2 MN
... ...
M12 --- N M12 ... M M12 ---N
and its inverse matrix is given by

M, M2 MN-I
M12 ... N M12 ... M M12 ... N
MI M2 M1V-I
M12 ... N M12 ... N M12 N
(UC )-I M2
M12 ... N
MI M2 MN-I
M12 ... N M12 ... N M12 ... N
(2.8)
The choice of the heavy-particle center coordinates may be natural
when the mass of the Nth particle is heavier than the masses of the
other particles. See Complement 2.1. Note, however, that it is always
possible to transform from one coordinate set to another. The Jacobian

corresponding to the coordinate transformation (2.4) is unity for both
U matrices.
Corresponding to the transformation (2.4), the momentum pi is
expressed in terms of the momentum -xj -ih ya conjugate to the =
axj
coordinate xj:
IV N
Pi =
1: Ujilrp Wi =
1:(U-l)jipj (i =
1, ..., N). (2.9)
j=1 j=1
N
Note that -7rAT =
Ej=1 p, is the total momentum. The center-of-mass
kinetic energy is given by T,,n ir2,v/(2MI2 N). The kinetic energy

=
...
operator with the center-of-mass kinetic energy subtracted is then

expressed as
i=1
2mi
i=1 j=1
(27ni 6ij 2m12 ...
N) Pi Pj
M-1 N-I
Aij-xi -
-7rj, (2.10)
2
i=1 j=1
where
Aij 1: UikUjk Mk (i,j =

11 ....
N -
1). (2.11)
k=1
To evaluate the potential energy matrix elements for the wave func-
tions that contain no dependence on the center-of-mass coordinate, it
is convenient to express the single-particle coordinate relative to the
center-of-mass, ri xN, the interparticle-distance vector, ri -,rj, and
-
the momentum difference, pi pp in terms of x and

-
-x, respectively.
This is easily done by using Eqs. (2.4) and (2.9):
1V-1
ri -
XN =
1: (U-l)ikXk -W X, (2.12)
k=1
N-I
'ri -
r3 1:
k=1
((U-l)ik (U-I)jk)
-
Xk = W(j) X, (2.13)
N-1
2
(Pi Pj 2
E (Uki -
Ukj)7rk (ij )Ir- (2.14)

k=1
Note that the contribution from the term proportional to the total
momentum, (Mi -
Mj)/(2Ml2 N)7rN, ...

is omitted in Eq. (2.14). This
is legitimate when we are in the center-of-mass system (,7rlV =
0) and
interested in the intrinsic motion of the system. The Hamiltonian of
type (2.1) is now expressible in terms of independent relative coordi-
nates alone. In what follows x and 7r are meant to represent only the
relative coordinates unless otherwise stated, so that they are (N-1) x 1
one-column matrices. In Eqs. (2.12)-(2.14), (N -
1) x 1 one-column
matrices (i.e. (N -
I)-dimensional vectors) 0), Oj), and (W) are
introduced to simplify the notation. E.g., Wx is a Caitesian vector

which is formed by the usual matrix multiplication of iv(i) and x, i.e.,
-W X =
EN-I
k=1
W(i)
k Xk-
Note that Oj) and (W) satisfy the following
equalities
M-1
W(ij)
k (k
3
-((ij)
W( ij = (-ipw(ij)
k=1
W(ij)(1(ij)
k
W
(ij) F(ij)
i>i=l i>i=l kI
N
Jk, R, I N -
1), (2.15)
2
where the relations UNTj =

?ni/?nl2---N, (U-1-)iN =
1, EN Uki = 0
for k =7 - N are used.
For an (N -
1) x (N -
1) symmetric matrix A, let the quadratic

form, bAx, represent scalar products of the Cartesian vectors:
N-I N-1 N-1
..bAx =
E xi-(Ax)i Aijxi-xj. (2.16)
i=1 i=1 j=1
It is easy to show, with this convention, that

1V
i=1
ai(ri -
xv)
2
i=1
ajw(')J)) x,
1V N
1: (ij) -(ij)
I: a
ij (ri _,rj)2 =.i ajjw X. (2.17)
i>i=l i>i=l
As a special case of the above relations, we have the following equalities

for the moment of inertia around the center-of-mass:
N N N-1
MzMj 2
Mi(ri -
XN )2 N(ri -
Tj) (2.18)
M12 ...
where the reduced mass ILi corresponding to the Jacobi coordinate xi

is defined by
?ni+17nl2 ...
z
N 1). (2.19)
M12 ...
i+1
We also obtain
2
N N
(ri -
T'j )2 = N E(ri XN)2 _
(,ri -
xiv) (2.20)
i=1
The term E , (ri -
XN) vanishes when the masses of all the particles

are the same.
The total orbital angular momentum operator L in the center-of-

mass system (-xv =
0) can be expressed in terms of the operators
relevant to the relative coordinates:
N N-1
hL (ri X Pj) (XN X 7N) (Xi X 7ri), (2.21)
i>i=l
((,ri -
Tj) X
2
(Pi -
Pj))
N
or i -
'r X Rij) -X
i>i=l
Mj
+ ((ri -
rj) XMi 7rN

2M12 ... N
W('j)X X
(('j)-7r) =NU. 2
(2.22)
In Eq. (2.22) use is made of Eq. (2.15) in the last step. Thus the equal-
ity of Eq. (2.22) holds in the case one considers the orbital angular
momentum in the system of 7rjV 0 or treats a system of paxticles
=
with equal masses.

In atomic physics the trial function of Hylleraas type or correlated
exponential type is often used with success. This function contains
the exponential form expressed in terms of the interparticle-distance
coordinates. T-nstead of the exponential function let us consider its
Gaussian analogue
IV
T =
exp E aij (ri
-
rj )2 (2.23)
2
i>i=l
Here N(N -
aij determine the falloff of the interpar-

1)/2 parameters
ticle functions and may be considered variational parameters. Since
not all of the N(N 1)/2 interparticle-distance vectors, ri rj, are
-
-
independent, e.g., TI T3 (Irl T2) + (T2 -r3), it is convenient to

-
= -
rewrite T in terms of the independent coordinates x. This is needed to

calculate, e.g., the norm of T. Equation (2.17) enables one to rewrite
Eq. (2.23) as
-1
T =
exp ( -
2 iAx), (2.24)
where the (N -
1) x (N -
1) symmetric matrix A and the parameters
aij are related by

N N
-
(ij) (ij)
AkI ajjWk WI =
I: aij (W (ij) W(ij)
) kl'
i>i=l i>i=l
N-I N-1
aij
-
1: 1: UkjAkIU1j =- -(CTAU)ij (i < (2.25)

k=1 1=1
By substituting Akj of the first relation to the second relation one

can check the validity of the second relation. The above two forms,
(2.23) and (2.24), are thus equivalent. We introduce both because in

some applications the first form or in some others the second form is
more advantageous. See Complements 6.4 and 7. 1. The norm (Tf I Tf)
becomes finite if and only if A is positive-definite and is then given by

I
((270N-1 /detA) 2 -with the use of Eq. (6.32) or Exercise 6.2.
A necessary and sufficient condition of the positive-definiteness of
a symmetric matrix A is that the matrix A is expressible as A =
2.3 Symmetrization 15
dDG, where G is an (N -
1) x (N -
1) orthogonal matrix containing

(N 1) (N 2) /2 parameterg and D
- -
is a diagonal matrix, Dij diJij, =
including (N 1) positive parameters di.

-
closing this section, we remark that the matrix Uj of Eq.

Before
(2.5) orthogonal in general. In some cases it is convenient to use a
is not
coordinate system that is obtained from the single-particle coordinates
by an orthogonal matrix. We show such an example in Complement
2.2.
2.3 Symmetrization
The function of the system must have a proper symmetry for the
wave
interchange of identical particles. This is achieved by operating on the

basis function with a suitable operator P, which is the symmetrizer
for bosons or the antisymmetrizer for fermions. The operator P can in
general be given by a combination of the permutations P with suitable
1 2 'V
phases. The permutation P =
(PI P2 PN ) of paxticle indices
transforms the single-particle coordinates as ri -+

rp,:
Pr =
Tpr, (2.26)
where the matrix Tp is
(Tp) ij =
Jj pi (i, j =
1, ...' N). (2.27)
E.g., the matrices Tp are given as follows in the case of three particles:
1 0 0 0 1 0
T123 0 1 0 1
T123 1 0 0
123 213
0 0 1 0 0 1
0 0 1 1 0 0
T123' 0 1 0 T123 0 0 1
(321 132
1 0 0 0 1 0
0 1 0 0 0 1
T123
( 23 1
0 0 1 T123
(3121
1 0 0 (2.28)
1 0 0 0 1 0
The operation of P on the wave function is thus not a problem when

it is written in terms of the single-particle coordinates.
When the wave function isexpressed in terms of the relative co-

ordinates, we have to note that the permutation P induces a lineaX
transformation of the relative coordinates as follows:
Px =
Tpx, (2.29)
where by using Eqs. (2.4), (2.26) and (2.27) the matrix Tp is now
given by
N
(TP)ij 1: Uik(U-1)Pki (i,j =

'I ...
I
N -
1). (2.30)
k=1
Though we use the same notation of Tp, note that the size of the
matrix is now (N 1) x (N 1)
- -
because the center-of-mass coordinate,
which is unchanged under the permutation, is eliminated. We Will show
in Chap. 6 that the properties of Eqs. (2.26)-(2.30) can be used to
advantage for Gaussian functions. See Eqs. (6.28) and (6.29).
2.4 Permutation of the Jacobi coordinates
One cangenerate different sets of Jacobi coordinates by interchanging

the paxticle labels by permutations. Figure 2.2 shows an example of
three different sets, x('), X(2) ,
and X(3), of the Jacobi coordinates for
a three-body system. These sets are completely equivalent from the
point of view of the dynamical description of the system, but they can
be advantageously used in the calculation of the matrix elements.
In the Jacobi coordinate set, defined by Eq. (2.5), the first Jacobi
coordinate is equal to I'l -r2. By permutations one can create other
Jacobi coordinate sets X(k) where the first Jacobi coordinate takes
the form of ri rj. The index k refers to the kth permutation and
-
serves to distinguish the permuted Jacobi coordinate set. As the two-

body potential depends on this coordinate, the set created by the
permutation will prove to be useful. See Appendix A.5 for an example
of using the permuted Jacobi coordinate set.
Another peculiarity of the Jacobi coordinates is that if we are in
the center-of-mass system (-xiv 0) then by using Eqs. (2.5) and (2.9)
=
p1V =
--xiv-11 that is, one of the single-paxticle momenta is equal to
one of the relative momenta. By permutations we can express any
of the single-paxticle momenta in terms of a certain momentum of
the relative momenta. This simple expression can be exploited in sev-
eral ways, because instead of evaluating matrix elements of operators
2.4 Permutation of the Jacobi coordinates 17
a- is
*+-- Ak a&- Ah
Mw- IRW MW
2 3 3 71 1 2
2 3
Fig. 2.2. Different sets of the Jacobi coordinates for the three-particle sys-
tem. From left to right: x(l), X(2), and x(3), respectively.
containing single-particle momentum we can do so with operators of

relative momentum and vice versa. See Sect. 7.6 for more details and
an application to the calculation of matrix elements for the semirela-

tivistic kinetic energy.
The permutation of the Jacobi coordinates is defined by the
permutation of the column vectors of Uj and the corresponding
exchange of the masses. For example, for the cyclic permutation
1 2
(1, 2, ..., N) =
(2 3 N)
1
the transformation matrix between
the singlt-,particle coordinates r and the Jacobi coordinates x(f) takes

the form
0 1 -1 0
M2 M3
0 ...
0
M23 M23
U(k) (2.31)
M2 M3 MAF
-1 ...
M23 ... M M23---.LV M23---N
MI M2 M3 MN
...
M12---N M12 ... N M12 ... N M12 ---

N
The relative coordinates in the different Jacobi coordinate sets can be

expressed by each other through the transformation:
X(k) = T(k) X T(k) = UMU-1.

i (2-32)
Thesymmetrization in Sect. 2.3 and the permutation of the Jacobi
coordinates are formally very similar. The main difference is that while
the symmetrization is assumed for identical (and therefore equal mass)

particles, in the latter case we allow for particles with different masses.
18 Complements
Complements
2.1 An N-particle Hamiltonian in the heavy-particle center coordinate
set
Suppose that the mass of one of the particles, say the Nth paxticle,
is the heaviest among all the paxticles. In this case the choice of the
heavy-paxticle center coordinate may be natural paxticularly when the
position of the Nth paxticle is close to the center-of-mass of the system,
like in an atom, where the Nth particle is a nucleus. The Hamiltonian
(2.1) can be reduced to the form which contains no explicit dependence
on the variables of the Nth particle.
By using Eqs. (2.7), (2.10) and (2.11) the Idnetic energy operator
is expressed in the heavy-paxticle center coordinates as
N 2
N-I N-I
, 1
P7
E 2mi -
T.- =
EE
i=1
( 2mi k 2mjV) + -
I-ii .7rj
i=1 j=1
N-1 N-1
70z 1
1: 2jLjjV
+ -
TaN
E Iri .7rj, (2-33)
i=1 j>i=l
'where jLjN mimN/(Tni + Talv) is the reduced mass for the ith and
=
Nth particles. The potential energy VjV (i 1, ..., N 1) is a function = -
of ri -rN =
xi and can be considered the one-body potential Uj acting
on the ith particle. On the other hand the Potential energy Vij acting
between the lighter particles function of ri -,rj
is a xi = -
xj. The
HamiltoTdan can thus be reduced to the type of Eq. (2.2)
N-I N-I N-1

2
^7r'
H=E 2AjN
+EUi+ 1: (Vij+ MN
7ri-7rj). (2-34)
i=1 =1 j>i=i
The problem is thus reduced to that of an (N-1)-paxticle system with

an external field Uj and an additional separable "two-body potential"
7ri -7rj. In the limit of m1V

-
oo, jLjjv goes to mi and the 7ri .7rj term
-->
disappears. The effect of the finite nucleax mass in the helium atom
was discussed in [3].
The above formulation can also be applied to a system of identi-
calparticles when some, or most, of them form an inert core. As an
example let us take up the nuclear shell model, where the nucleons
are divided into two groups, the passive or inactive nucleons and a
relatively small number of active nucleons. The passive nucleons form
C2.2 Canonical Jacobi coordinates 19
an inert by filling the lowest possible single-particle orbits and

core
exert a potential field Ui on each of the active nucleons. The active

nucleons may occupy several single-particle orbits belonging to several
major shells outside the inert core. In this approximation it is desir-
able to describe the motion of the active nucleons without inclusion
of any excitations arising from the center-of-mass motion. It is clear
that such excitations can be excluded by taking the full Hamiltonian
to have the form (2.34) with the core as the heavy particle.
2.2 Canonical Jacobi coordinates

As was noted in Sect. 2.2, the Jacobi coordinates x have the property
that the transformation matrix Ui of Eq. (2.5) is not orthogonal, that
is, neither UjUi- nor Uj-Uj is equal to the unit matrix. Instead, the
following relations are fulfilled:
UjA-'Uj = L-1, UiLUi =

A, (2.35)
with
Lij Jij Aij Jij, (2.36)

Mi Ai
where pi (i 1, N=
1) is the reduced mass belonging to the ith
-
Jacobi coordinate and is defined in Eq. (2.19), while 1LV is equal to the
total mass 'MI-2 ... N. By using the transformation between
the single-
paxticle and the Jacobi coordinate systems, x Uir and p UT-x, of = =
Eqs. (2.4) and (2.9), the above relations lead to the following identities
N N N N
pi2 Ir2i
E Tnir? z
=
E Nx -
Mi
E Ai
-(2.37)
Alternatively, one can introduce a "canonical" set of Jacobi coor-
dinates:
IV
Vijrj with V = A 2
UiL 2. (2.38)
j=1
The square root matrices of the diagonal matrices L and A axe sim-
ply given by the square root of the elements. This system of Jacobi
coordinates belongs to an orthogonal transformation:
Vf" = f7v 17 (2.39)

and therefore
20 Complements
1V 1V IV
EVji j and E 2 Er2 i

=
i* (2.40)
j=I i=1 i=1
Let the momentum canonically conjugate to i be denoted qi.

Simfla,rly to the transformation of coordinates, we obtain the trans-
formation of momenta as follows:
N N
71i =
E ViiPi, A =
E Viini (2.41)
j=I j=I
The total Idnetic energy does not take a diagonal form with respect
to q and can be expressed as follows:
I?
E Mi E E (VLfr) ij?7i -77j wLf7rl. =
(2.42)
i=1 i=1 j=I
The two systems of the Jacobi coordinates Ujr and the canon-
x =
ical Jacobi coordinates = Vr can easily be related by

X = ujf ' =
V(Uj)-Ix. (2.43)
For a three-paxticle system the transformation matrix V reads as
M12
L2
-VEM12 M1
0
12
V
rT; 1'2?T 1
M
M123
3 M2M3
12MI23
M12
M123
(2.44)
VMM1123 1
VMM1232: 2
V-Ml'23
M3
and for particles with equal mass
I 1
-
0
V/2- vf2-
I I 2
V -
(2.45)
v/6- T6 V"6_
I '1 1
v/3- 73T 73
The disadvantage of the canonical Jacobi coordinates is that if the
masses mi axe not equal then the center-of-mass motion is not sepa-
rated easily ( N is not the center-of-mass coordinate).
3. Introduction to variational methods
The variational method is of the most

popular approaches to tackle
one
quantum-mechanical few-body problems. Though it gives only an ap-

proximate solution except for some special cases (the Ritz variational
method, for example, gives only an upper bound of the energy), one
can get a virtually exact solution with an appropriately chosen func-
tion space. The function space is defined by basis states and the wave
function of the system is expanded in that, basis. In this chapter we
briefly introduce the theorems requisite for obtaining a vaxiational
solution.
3.1 Variational principles

Let us physical system whose Hamiltonian H is self-adjoint
consider a
(Hermitian), bounded from below and time-independent. We are in-

terested in finding the discrete eigenvalues of H and its corresponding
(normalized) eigenstates:
H!P,, =
En(fi, n =
1, 2,... (3.1)
The energies En axe real and are ordered such that El :! E2< ....
We
assume that the ground state is non-degenerate.
Although H is known, this does not mean that E,, and (fin axe
known. In general, it is difficult to solve the eigenvalue equation (3. 1).
When we do not know how to diagonalize H. exactly, the vaxiational
method becomes useful for any type of Hamiltonian.
Theorem 3.1 (Ritz Theorem). For an arbitrary function Tf of
the state space the expectation value (Mean value) of H in the state Tf
is such that
RIHITI)
E = > El, (3.2)
R_ I TI)
-

22 3. Introduction to variational methods
where the equality holds if and only if Tf is an eigenstate of H with the
eigenvalue El.
Proof. The proof elementary and can be found in

of this theorem is
textbooks of quantum mechanics. See for example [1, 2]. The ftmetion
T1 may be expanded in terms of the energy eigenstates
00
ai(fii. (3-3)
In this expansion the eigenstates with continuous

eigenvalues are in-
cluded and the sum must be extended appropriately to include an
integration over them. Then we can show that for an integer n

0
(Tf jHn Itp) Faci=2 A Eln)Jai 12

n
Y
-
7
En
I
-
-
(3-4)
(TrI Tf ) 00, jai 12
Ei=
Clearly the right-hand side of Eq. (3.4) is non-negative for n I =
because Ei E,
-
> 0 for i > 2. It vanishes if and if
only ai 0 for =
i > 2, that is, T, is an edgenstate of H with eigenvalue El. This proves

the Ritz theorem. Only the case of n I was used to prove the Ritz
=
theorem but other cases will be needed later to derive Temple's bound.
The Ritz theorem is the basis for approximate de-

a method of
termination of the ground-state energy El, namely exploited as it is
the minimization principle of the mean value of H. Suppose that we
choose a family of functions Tf (a) which are characterized by a finite
number of parameters denoted a. In the case of Eq. (1.1), there axe
two parameters a and P. We calculate the mean value E(a) of the
Hamiltonian H in this trial function, and minimize E(a) with respect
to a. The minimal value obtained in this way is an approximation
to tile ground-state energy El of the system. Clearly, whether this
variational method produces a satisfactory result or not substantially
depends on the choice of the trial functions Tf (a).
The Ritz theorem can be generalized to excited states as well:
Theorem 3.2 (Generalizecl Ritz Theorem). The expectation

value of the Hamiltonian H is stationary in the neighborhood of its
discrete eigenvalues.
Proof. Let us calculate an increment JE of the mean value us' g

E(TfIT) =
(TfjHjTf) when Tf is changed +JT-1, where JTf is as-
to Tf
sumed to be infinitely small. Neglecting terms of higher order than a
linear term in JT1, we obtain
3.1 Variational principIcs 23
(TfITf)JE =
J((TIIHITI)) -
EJ((TIITI))
=
(JTf IH -
EITI) + (TfIH -
EIJTf). (3.5)
The mean value E is stationary if JE = 0 for any infinitesimal JT,
that is, if
(,NTtIH -
EITI) + (TIIH -
EIST) = 0. (3.6)
If JTf is chosen to be -(H -
E)Tf, where - is an infinitesimal real
number, the above equation implies that the norm of the function
(H E)Tf is zero, and thus the function (H E)Tf itself must be

-
-
a null function, namely HTf = ET. Therefore the mean value E is
stationary if and only if the state Tf from which E is calculated is

an eigenstate of H, and the stationary value E is the corresponding
eigenvalue of the Hamiltonian.
This generalized Ritz theorem allows an approximate determina-

tion of the eigenvalues of the Hamiltonian. If the E(a) has several
extrema, they are the approximate values of some of the energies E,,.
In most cases for practical applications the. trial function is given
as a linear combination of a finite number of independent functions
!P(a):
K
Tf ciTf (ai). (3.7)
The linear independence of the functions will be discussed soon later.

We do not always assume that Tf (a I) T' (a K) axe mutually orthog-
, ...
i
onal because the use of nonorthogonal. functions is in fact quite useful.

They can, however, always be made orthogonal if necessary, e.g. by a
Gratn-Schmidt orthogonalization procedure. The variational method
then reduces to the eigenvalue problem of the Hamiltonian inside
the state space VK spanned by the set JTf((YI),...,Tf(aK)Ji that is,
the space containing all linear combinations of T (a,), Tf(aK). The ...
mean value E is given by
Ctlic
E = (3.8)
CtBC'
where -
c is a IC-dimensional column vector whose ith element is ci
and ct is the Hermitian conjugate of c. The K x K Hamiltonian and
overlap matrices W and B are defined by

Rij =
(Tf(ai)JHJT'(cvj)), Bij =
(TI ((Yi) I Tf (aj)). (3.9)
The linear parameter ci can be determined by the generalized Ritz
theorem. The condition that E is stationary with respect to an ar-
bitrary, infinitesimal change of ci leads to the generalized eigenvalue
problem
K
Wc =
E&, i.e., E(Rij -
EBij)cj = 0 (i =
I,-, K). (3.10)
j=J
The restriction of the eigenvalue problem of H to the subspace VK

can thus simplify the solution.
We discuss the linear independence of the functions Tf(al),...,
Tf (a K). The linear independence of the functions means that no vector
c except for c = 0 can make a linear combination , EK
i=1 ciTf(ai),
a null
function. In other words, the combination becomes a null function if
and only if c = 0. This is equivalent to saying that the
equation Bc 0 =
for c has unique solution of c

a = 0. (To understand this, we show that
K
E ci!F(ai) 0 is equivalent
= to Bc = 0. Whe'n EK ci!P(ai) =
0,
we obviously have Ef I ci (TI(aj) JTf (ai)) 0 for j I,-, K, which = =
is nothing but (&)i 0 for j I,-, K. Conversely, when (Bc)j

= = 0 =
for j I,-, K, by multiplying cj* (the complex conjugate of cj) and

=
summing over j, we obtain ctL3c =(.EK CiT,(Cv,) I EK

I 1 ciTf (ai)) =
0,
K
which leads us to Fj-1 ciTf (ai)
0.) =a unique As the existence of
solution c = 0 for the
equation Bc 0 is possible only when detB --A 0,
=
the linear independence of the functions is assured by the condi-

tion detB :A 0. Because the overlap matrix B is at least positive-
semidefinite, that is (Tf JTf) ctBc > 0 for any vector c, all eigenvalues
=
p of B become real, positive when the functions Tf (a,), Tf (a K) ar e ....
linearly independent. The positiveness of M is understood as follows:

For the eigenvector c :A 0 corresponding to the eigenvalue
JL, we have
the relation pctc (Tf Ifl. As
= the basis functions Tf (ai) are linearly
independent, Tf is not identically zero because otherwise we can make
!rf =EK i= I
ciTf (ai) vanish identically with c =A 0, which contradicts the
assumption of the linear independence of the basis functions. Thus
(TfITf) is positive and of course ctc is positive, so the eigenvalue y has
to be positive. We can also state that if there exists a vanisl-iing eigen-
value of B then the basis set JTf(a1),..-,Tf(aK) is linearly dependent.
The solution of the eigenvalue problem, pc(A), gives us an BcG l =
orthonormal set:
3.1 Variational principles 25
01.1 1: C(ii.') Tf (C'j),

2
(3-11)
i=l
where c(l') is assumed to be normalized to c(") tc(/") = 1. In many
practical problems it may happen that B has one or several very small
eigenvalues. Then the eigenstate corresponding to the small eigenvalue
has very large expansion coefficients cil") If this occurs then a
- Flj,.
small error in the matrix elements of R or B can lead to a larger error
in the solution of Eq. (110).
When the ill condition mentioned above does not occur, the gener-
alized eigenvalue problem (3.10) can be solved safely. The eigenvalues
q (i =
1, ...' K) are arranged inincreasing order el :! 62 : ...
The low-
est eigenvalue El may be a good approximation to the ground-state
energy El if the state space V_T<- is chosen to include the physically
most important configurations. Two functions T and TI' belonging to
the eigenvaluesq and ej, respectively, have the overlap
(Tf I Tfl) = CtBC' 1 (3.12)

where c -md c' satisfy the following equation
Rc = EiBc, and 'Hc =,EjBc'. (3.13)

To normalize T', c has to be normalized to ctBc I and likewise =
&t&' 1. Using the Hermiticity of R and B and the reality of the

=
eigenvalues Ej in Eq. (3.13) leads one to (Ej cj)ctBc' 0. Thus - =
the two eigenstates belonging to different eigenvalues Ei and ej are

orthogonal. In the case of degeneracy, they can be made orthogonal
by an appropriate procedure.
The relationship between the eigenvaluesEi of the truncated prob-
lem and the eigenvalues E,, of the full Hamiltonian is elucidated by
the Nfini-Max theorem.
Theorem 3.3 (Mini-Max Theorem). Let H be a Hermitian
operator with discrete eigenvalues El :5 E2 < ....

Let El ! e2 <_ -
<
EK be theeigenvalues of H restricted to the subspace Vrc of a linearly

independent set of K functions Tf (a,), ...
I Tf(aK). Then
El <'Eli E2:51E2i ...

i
EK : 16K- (3.14)
Proof. Let WK be the subspace spanned by the orthonormal eigen-
states!Pl,(k I!PK of the operator H. We will first show that there
...
is at least one normalized function in Vrf with the property

26 3. Introduction to vaxiational methods
(TfJHJTf) > EK. (3.15)

To show this we consider the projected function PTf for any normalized
function Tf in VK, where P is the projection operator that projects
onto WK; p =,EK I!Pi) (!Pi 1.
i= I
There are two possibilities:
(i) there exists a function Tfo in Vr<- such that PTfo = 0.

(ii) PTf :A 0 for all functions T, in Vl<-.
In case (i) Tro is a linear combination Of 43K+1, T)lf+2,..., and hence
the normalized Tfo has the mean value (Trollll%) : : EK+, ! EK. In
case (ii) any two different normalized fimetions Tf, and Tf2 in Vl<- are
projected to different functions PTfl and PTf2 in _PVK because otherwise
P(TV, Tf2) 0, which contradicts the assumption. Namely, any two
-
=
different normalized functions in V.[<- axe projected by P to different

functions in Wl,(. As both VK and )IVI<- have the same dhnension, it
follows that there must exist a normalized function T, in V.Tf with the
property M wPrf (a 4_ 0). Then !Tf can be expressed as a!PK + bo,
=
where 0 is a normalized flinction given as a linear combination of

PK+1,!PK+2, ... JaJ2 + JbJ2
; 1. This function has the
= mean value,,.-
(TfJHJTf) =
Ja 12EK + lb 12(01H10) Erc + lb 12((01H10) _
EK) Ell;C.
Because 16K is the largest eigenvalue of H restricted to V_Tf, we have
IEK 2! (T'IHIT') > EK. (3.16)

Next we define the (K- I)-dimensional state space Vr.C-1 to be the set
of all those functions in Vif which are orthogonal to the eigenvector
belonging to 45K and by induction can show 'EK-1 ! EK-1 and so on.
It is easy to derive the following theorem ft-om. the Nfini-Max the-

orem.
Theorem 3.4. Let E, ! , E2 < EK be a number K

of the lowest eigenvalues of a Hermitian operator H. Let Tf(al),
Tl(a2)i..., Tf (aK) be a set Of linearly independent functions. Then
1: Ei :! Tr(B`R), (3-17)
i=1
where the matrices B and R are defined by Eq. (3.9).
Proof. Let el : 162 :5 --- :: - IEK be the eigenvalues of H restricted

to the subspace Vl<- of the K functions Tf(Cfl), ...
Tf (Ci K) and let
i
3.1 Vi-triational principles 27
be the orthonormal. eigenstates corresponding to the
eigenvalueS 6 11 E21 ...

i 16K. Then Oi is given by
K
E UijT1(aj) (3.18)
j=1
with U satisfying the orthonormality condition UtBU = 1. From The-

orein 3.3 we obtain
K K, K
J:Ej :5 EEi=E(OilHloi)=Tr(UtHU)=Tr((BU)-IHU)
Tr(U-'L3-1WU) =
Tr(B-1R). (3-19)
right-hand side of Eq. (3-17) is determined only
It is noted that the
by the subspace spanned by the functions and not by the particular
choice of the basis because any non-singular linear transformation of
the basis does not change the trace (3.17). Namely, when a new basis
FC
set Xl,...,XK is given by a transformation Xi Ei= J Tij Tf ((--ij) (i =
K) with a non-singular matrix T (deff :A 0), we can show that

Tr(B-'?i), where L3i'j
21
-
(XilXi) and R'ijtj =

(XjjHjXj).
This leads to the density-functional density can formalism where the
be used as a calculating the energies of excited states
basic vaxiable for
as well as the ground state. See [4, 5] for this interesting idea.
The next theorem is fundamental to answer a question of how

the eigenvalues obtained in a restricted subspace change as the basis
dimension is increased.
Theorem 3.5. Let el ! - 162 :! 'EK be Hie eigenvalues of a Her-

mitian operator H restricted to the subspace VK of linear combinations
of independent functions Tf (a,), If (a2), ...
i Tf(01K). Let El < e2 :5 ...
<
EI be the eigenvalues of H restricted to the subspace VTC+l of linear
K+1
combinations of independent functions Tf (01)i llf (a2), ...
7
Tf (aK), and
!P(OK+I). Then
Ell < El :5 E12 :-5 C:2 :5 ... :5 '61K :-5 61K,(+i* (3.20)
Proof. Let 011 02 OK be the orthonormal eigenstates correspond-
1 ...
i
ing to -the eigenvalues 61,162 16K- Any function !P in VK+j can be
expressed as
K+1
Tf CA, (3.21)
28 1 Introdnetion to variational methods
where the normalized function 01-<-+, is constructed from Tf(aK+l) to

inake, it. orthogonal to any function in Vfc:
K 2
OIC+I (Tf (a Tf (a 1,C+ 1)) 1 (0i I Tf (Ct 1,;C+l 12
x
(3.22)
The eigenvalue problem using the basis set takes

the simple form
61 0 -
0 h, ) C, ) C1 )
0 62 -
0 h2 C2 C2
=E
0 0 -
EK hK CK CK
h *1 h;2 ... h I*c hrc+1 CTC+I CJf+
K
(3.23)
where hj =
%JHJOrc+I) and hj* is the complex conjugate of 11j. The
characteristic function D(E) to determine the eigenvalues reads as
K K K
D(E) (hK+1 -
E) II(Ei -
E) -
Ihi 12 11 (Ej _
E)
K K
Ih-I2
II(Ei -
E) hIc+i -
E -
E
0. (3.24)
i=1 j=I
EJ
Here it is assumed that all hi (i

I,-, K) axe nonzero. The (K + 1)
=
eigenvalues are obtained by finding the roots of the equation
K
Ihi 12
E-hK+I (3.25)
E -
ei
The right-hand side of the above equation becomes negative for E < C,
and positive for E > CK. A graphic solution of Eq. (3.25) is displayed
in Fig. 3. 1. It is clear that the cross points of curves, namely the new
eigenvalues E , satisfy the relation (3.20).
When there are n vanishing hi's, the corresponding ei's become the
solutions of Eq. (3.23), i.e., E'i Ei. The remaining (K + I
==
n) new -
3.1 Variational principles 29
2
1N
'=1E-Cj
.................
.......I.. ............V.. ......... .... ..... ... ........... ..................
611' F-21 I 'EK
E-hK+1
Fig. 3.1. Graphic solution of Eq. (3.25) to obtain the eigenvalues of Eq.
(3.23)
eigenvalues are obtained through the eigenvalue. equation, which has

the same structure as Eq. (3.23). The relation (3.20) trivially holds
also in this case.
The significance of this theorem is

that, by including a further term
in the basis, the K lowest eigenvalues; cannot become worse. Therefore,
this theorem implies the Mini-Max theorem because, in the limiting
case that the subspace approaches the full Hilbert space as the basis
dimension increases, the K lowest-eigenvaluesEj converge to the exact

eigenvalues; of the Hamiltonian.
Suppose that one wants to see how the lowest eigenvalue changes
when one vaxies only the (K+l)th function-!V(arc+j) while keeping the
rest of all. the K functions unchanged. In this case one does not need to
solve the generalized eigenvalue problem of type (3. 10) but only needs
to find the smallest root of Eq. (3.25). -Of course the latter is much
simpler and faster than diagonalizing the matrix. This advantage is

used to sample more random trials in selecting a suitable basis function
in the stochastic variational method, as explained in Sects. 4.2.5 and
4.2.6.
3.2 The variance of local energy
It would be nice if there judge the accuracy of

were a method to
the solution obtained variationally. The expectation value E is the
upper bound of the ground-state energy El. If we can calculate a lower
boun,d, the. difference between tile, two bounds gives an estimate for
how dose E may be to El. This makes sense if and only if the lower
bound can be calculated as closely as possible to the ground-state
energy. In order to discuss the lower bound, we define the vaxiance of
the energy expectation value, o-', by the norm of the residue function
(H -
E)TfI(Tf JTf) '21
2 -
(T1J(H-E)2JTf) (TfJH2JTf) E2
0' -
(3.26)
(TIITf) R- I Tf
According to Weinstein [6, 71 the following theorem can be proved.
Theorem 3.6. There ii at least one exact eigenvalue in the inter-
val [E -
o-, E + o-J.
Proof. By using Eq. (3.3), the variance can be expressed as
2 EMI (Ei E)2 JaiJ2 -
0'
(3.27)
_
00, JaiJ2
Fli=
Suppose that Ek is the eigenvalue which is closest to E. Then we have
00 CO
2
J:(E i _E jai 12 > (Ek -
E)2 jai 12. (3.28)
Thus we have o-2 > (Ek -
E)', which proves the theorem.
The Weinstein criterion guarantees that there is an eigenstate

whose energy is in the interval [E G-, E + o-] but does not indicate -
which one. In case E is sufficiently dose to the ground-state

energy,
the theorem gives the lower bound as
E- < El.
G-
(3.29)
Another lower bound called Temple's bound [8] is also expressed
in terms of the variance of the energy as
0'
E < El.
-
e-E
- (3.30)
3.2 The variance of local energy 31
Here - is an energy such that E < e < E2, where E2

arbitrary is the
conserved quantum
energy of the first excited state which has the same
numbers as the ground state. The best bound is clearly obtained by
setting e equal to E2. Temple's bound is easily derived as follows:
0.2
E, (E- --E )
1 (TI I H2 I TI)
= -
e -
E [ -
(,- + Ej)(E -
Ej) +
(Tf I Tf)
2]
Elf
i=2 (Ej-,-)(Ej-Ej)jajj2
EZ
I
00
> 0, (3.31)
E Ei= 1 jai12
where use is made of Eq. (3.4) for E -
El (with n =
1) and for
(TfIH2lTf)l (Tflyf-) _EJ2 (with n 2). =
The two bounds, (3.29) and (3.30), indicate that the trial function
2
that gives the smallest value of (7 is best among various trial functions
giving the same expectation value. Temple's bound often gives better
lower bounds than the Weinstein bound. See [91 for the extension of
variational bounds.
The variance of the expectation value can also be defined through
the local energy
HTf
Ej., (R) I
(3-32)
!P7
,
where R stands for the "configuration?' of the particles. More precisely,

if the coordinates of the ith particle by ri and pi, where pi
are denoted
stands for the coordinates other than the position coordinate ri, then
R stands for Ir I i P1 ilr2) P2 i 1. The variance defined by Eq. (3.26) is
...
equal to the variance of the local energy

(TfjjEjo,(R) -
E121!Tf)
0'
2
= P(R)IEI.,:(R) -
Ej2dR (3-33)
with
I Tf (R) 12
P(R) -- (3-34)
R I Tf)
where E, the energy of H in the state
mean Tf, can also be rewritten
as an average of the local energy:
(TIjEjo,(R)jT1)
E =
010f)
-
I P(R)Ejoc (R)dk (3.35)

Here P(R), the

probability density of finding the system at the config-
uration point R, is non-negative and satisfies the. relation f P(R)dR
1.
H the trial function Tf is the exact ground state of the Hamiltonian,
the local energy becomes R-independent and equals the ground-state
energy El. A similax statement holds for excited states as well. If Tf is

the exact eigenstate with energy-&, the. local energy would be equal
to E,,. This opens up another possibility of the vaxiational method
instead of the minimization of the mean energy: The niinimization of
the variance of the local energy [101. The adjustable parameters of the
trial function may be varied to minimize the variance. The calculation
of the variance is in general much more difficult than that of the mean
energy. An advantage of minimizing the variance of the local energy is,

however, that the quantity to be minimized has a known lower bound,
namely Since in any eigenstate the vaxiance of the local energy
zero.
is zero, this minimization principle applies to obtain excited states as

well as the ground state.
The stationarity condition for o-' is obtained by calculating an
increment So-' in the done for the

same. way as was mean energy in
Theorem 3.2. By multiplying both sides of Eq. (3.26) by (Tflfl and
calculating the increment to a linear change in JTf, we obtain the
following equation
(Tf Jqf-)jO.2 =,6((Tf JH2 Ifl) -
J(E2 (Tf IT,)) _
0.2j((T1JTf))
=
(,5qf- I H2 _ E2 _ 0.2 1 Tf) + (Tf I H2 - E2 _ 0.2 1 jqf-) -
2E(Tf JTf) JE
2
=
(JTf JH2 -
2EH + E _ 0.2 1 TI)
+ (Tf JH2 -
2EH + E2 _
0.21,5fl. (3.36)
Here use is made of
Eq. (3.5) in the last step. The stationarity condi-
tion is equivalent requiring (H2 2EH + E2)(y 0.2TI. When the
to -
=
trial function is given as a sum of a finite number of functions Tf

(a)
as in Eq. (3.7), this condition reads as the following
equation which
can be used to determine the linear parameters ci
(Q -
2EW + E2L3)C = 0.2BC7 Qij =
(!rf(a,) 1112 1 Tf (Cj)). (3.37)
Since E defined by Eq. (3.8) depends on c, the matrix in the round
bracket of the above equation also depends on c. Therefore the pa-
rameter c must be determined self-consistently: One assumes an initial
value for c, calculates the energy-R, and solves Eq. (3.37) to deter-
mine c and 0-2. As the value c obtained in this way may in general
3.3 The virkd theorem 33
not be equal to the initial c value, one now uses the solution c as the
initial value and repeats this cycle until both values of c become the
same. The minimization of o-' becomes cumbersome even though the
calculation of Qij is feasible.

The variational Monte Carlo (VMC) calculations employ Eq.
(3.35) to determine the mean energy. Tlie VMC method [111, or the
method of the Amalgamation of Two-body correlations into Multiple
Scattering (ATMS) [121 similar to it, usually chooses a trial function
that attempts to incorporate both the sfi7ort distance behavior and
the asymptotic behavior of the correct wave function, leading to com-
plicated functional forms. Thus the analytic evaluation of the matrix
elements is in general hopeless. Instead, the VMC requires only the
evaluation of the wave function and its Laplacian to determine the
local energy. If the interaction between the particles contains the spin-
orbit force, the first derivatives of tl-ie wave function are required as
well. These derivatives are obtained numerically by appropriate differ-
ence formulas. The mean energy is then estimated by a Monte Carlo
integration. The sampling of a set of configuration points R is made

by, e.g. the Metropolis algoritl-nn [131. The accuracy of the VMC or
ATMS calculation is limited by the choice of the trial function as well
as by the statistical error inherent in the Monte Carlo integration.
To reduce the statistical error one has to sample a great number of

points, which is computer time consuming. To optimize the variational
parameters of the trial wave function one repeat the mean en-
has to
ergy calculation using a different set of parameters. This is a hard job
considering that the VMC or ATMS has an inherent statistical error
particularly in the case one has anumber of parameters.

The variational method is a very flexible approximation method
which can be adapted to diverse -problems. The variational method
isparticularly physical intuitions lead us to an idea of
valuable when
the qualitative form of the solution. It easily gives good values for the
energy. The variational solution may, however, present unpredictable

erroneous values for other physical observables. It is unfortunately
very difficult to evaluate their error.
3.3 The virial theoren,
In this section summarize two theorems, the Hellmann-Feyntnan

we
theorem and the virial theorem, which are valid for the exact solution.
They can be used to check the quality of the wave function.
Theorem 3.7 (Hellmann-Feynman Theorem). Let H(A) be a
Hermitian operator which depends on a real parameter A, and POO a

normalized eigenstate of H(A) of eigenvalue E(A). Then the theorem
states that
d
TA E(A)
=
( P(A)J DA (3-38)
Proof. Differentiating, with respect to A, the relation
E(A) (3.39)
we have
d a
E(A) =
( P(A)J H(APKA))
dA (9A
(9 19
+ (fi(A) IHN ON) + R(A) W(A) 1 A
(3.40)
OX
The Iast two terms on the right-hand side. of Eq. (3.40) vanish because
(9 a
PNIH(A)O(A)) + OPNIHNI CA))
E(A) IN
J-P(A)) + (!P(A) I OA
jA- P(A))]
d
=
E(A) d,X ONON) =
01 (3.41)
where use is made of the conditions that H(A)!P(A) E(A)!P(A) and
When the evaluation of the matrix element of H(A) does not
pose any Oficulty, the

Hellmann-Feynman theorem can be used to
check the consistency of the vaxiational solution. If H(X) Ho +,XHI, =
then the matrix elements needed are just the terms of the energy
expectation value. It is straightforward to generalize the 11611mann-
Feynman theorem to the case that the Hamiltonian contains a number
of parameters AI A2 7
We note that the 11ellmann-Feymnan theorem can be stated in an

integral form
(!P(A2)JJff(A2)
E(A2) -
E(Al) =
(!P(A2)J!P(A1))
3.3 The viri.-,U theorem 35
(3-42)
"1
OA
The firstequality is trivial for the exact solution and has nothing to
do with the I-Iellmann-Feym-nan theorem. The equality holds, however,
only approximately for the variational solution and may be used to
check the quality of the solution.
The following theorem plays a key role in the derivation of the
virial theorem.
Theorem 3.8. Let (P be an eigenstate of a Hermitian operator H.

For any operator A we obtain
(T)I [H, A] I!P) = 0. (3.43)
The proof is easily obtained by noting H!P = ET, with eigenvalue

E. To test the accuracy of the vaxiational solution, one often uses a
special operator of the form

.
N N
(9
A= ri-pi ri -
(3.44)
h r.
'ri
i=1
because the commutator [H, A] is easily calculated. We note that the
operator A is a generator of the dilation operator, e aA, which has the
property, in a single-variable case,

d
exp (ax dx ) f (x) =
f (eax). (3.45)
This is easily by expanding f (x) in power series and using

shown
(X_A_)kxn
dX
nkXn, valid
= for any integer value of n.
Assume that the Hamiltonian for an N-particle system takes the
form
N
Pi
W(ri, r2, rN), (3.46)
H = T+ W
E=I2mi -
T,, . + .--,
Whether or not the center-of-mass kinetic energy is subtracted from

the Hamiltonian is irrelevant to the discussion on the virial theorem.
The potential W is assumed to contain no momentum operators. We
can then show that
C7 P-
IT, ri-
Ori
2(-2mi 'n
M12 Ar
--N
-Pi'IrN),
..
0 OW
1W, ri.
c9ri
1 '=
-ri-
ari
(3.47)
where M12 ... N= EN mi is the total mass of the system and 7rN
EN
..,i= I Pi
is the total momentum. Using Eqs. (3-44), (3.46) and (3.47)
we obtain
N
aw
[H, Al = 2T -
WA with WA (3.48)
ri
i=1
Substituting this into Eq. (3.43) leads to the following theorem.
Theorem 3.9 (The virial Theorem). Let!P be an eigenstate of

the Hamiltonian (3.46). For the operator A of Eq. (3-44)we obtain
2(!PjTjfl -
((fijWAj(fi) = 0. (3.49)
Therefore a quantity z7 defined by
(!PIWAI'P) '
1
77 =
(3-50)
2(!PjTj!P)
vanishes for the exact solution and can be used to check the quality
of the solution.
One has to know WA to make use of the virial theorem. It becomes
particularly simple if the potential is a homogeneous function of degree
s, namely
W(Ar I Ar2 i ArN) =

W(AX1 AY1 AZI i 7 .... ATNi AM AZN)
=
A*W(rjr2, ...
I rN)7 (3.51)
because then we obtain by the use of Euler's theorem or by differenti-
ating, with respect to X, both sides of Eq. (3-51) and setting A equal
to X = I
WA = SW- (3.52)
In this special case Eq. (3.49), together with (!PjTj!P) + ((fijWjfl =
Ej
yields the well-known relation
2
((!PITI! P) E, PIWI(P) = E (s :A -2). (3.53)
s +2 s+2
Because the homogeneity condition is fulfilled for -the Coulomb poten-
tial (s =
-1), the quantity t7 is conveniently used to check the quality
3.3 The virial theorem 37
of the solution for a system of particles interacting via Coulomb po-

tentials.
For a general non-homogeneous potential function of the form
N IV
W(rwr2, ... 'rIV) Ui(ri) + E Vij(ri -
rj), (3.54)
i=t i>i=l
Eq. (3.52) must be replaced by

IV N
Oui(r) OV,ij (r)
WA r-
Or ) r=ri
+ (r Or )r=ri-rj
(3-55)
For a spherically symmetric potential the operator r-ar- reduces simply
to r-4-.
dr
If the calculation of matrix elements of the. operator WA is not
difficult, the virial theorem can be used as in the Coulombic case.
When the potential W depends on a parameter A, we obtain
(9 d
(!PIWAI(P) =
((filWA +A WI-P) -
A
dA
E(A). (3.56)
Here the Hellmann-Feynman theorem is used to express the expec-

tation value of A'& W in terms of the energy eigenvalue E(A). In a
molecular system X may be a set of parameters which stand for the
positions of nuclei and if W consists of only the Coulomb potentials
WA + A--
bX
-W simply reduces to -W. See Complement 8.2 for an ap-
plication of this relation.
4. Stochastic variational method
The most direct approach to the variational solution of quantum-

mechanical bound-state problems is to diagonalize the Hamiltonian in
a state space spanned by some appropriate functions TV (a-,) Tf (a2) I 7 ...I
Tf (CeK). The applicability of this approach is, however, very limited,

because the diagonalization may not be feasible if the dimension K of
the state space is very large. This is typical in many-particle problems
such as the Hubbard model or the nuclear shell model. This method
can be called a approach", because one sets up a basis in a
"direct
well-defined way, e.g., by using a complete set of states that contains
no parameters, and then obtains the energy by a diagonalization.
Another possibility is basis optimization, which is actually de-

signed to avoid the problem of the huge basis dimension in the direct
approach. In this case one specifically selects the basis states that are
really essential to get the energy and the wave function of the sys-
tem to certain accuracy. It is obvious that this selection may be
a
state-dependent: Some functions might be adequate to describe the

ground state, but some others would be
appropriate to approx-
more
imate an excited state, particularly when the ground state and the
excited state have different spatial extensions. The stochastic varia-
tional method uses this second route by selecting the most appropriate
basis functions in a trial and error procedure.
4.1 Basis optimization
It goes without saying that the quality of the variational approxi-

mation crucially depends on the choice of the basis functions. Our
primary aim here is- to solveIV-particle problems. For this we will

choose basis functions that meet the following requirements:
1. They can be easily generalized for an N-body system.

40 4. Stochastic variational method
2. Their matrix elements analytically calculable.

are
3. They are easily adaptable to the permutational symmetry of the

system.
4. They are flexible enough to approximate even rapidly changing
functions.
The important condition (3) is non-trivial because the permutational

symmetry looks very complicated when expressed in terms of the rel-
ative coordinates to be used.
A possible choice for the basis functions fillfilling the above condi-
tions is the correlated Gaussian of Eqs. (2.23)-(2.25):
-1 N-1
exp
1
c
Ax) exp NE E Aij xi -
xj
i=1 j=1
IV
expf -2 E aij (ri

-
rj (4.1)
i>i=l
The matrix elements equivalently, aij are nonlinear paxameters

Aij or,
of the basis. These functions are spherically symmetric. To take into
account non-spherical states, the basis function has to be multiplied
by an appropriate orbital angulax function. In addition to the spatial

degree of freedom, the particles may have other degrees of freedom,
such as spin and flavor, and thus the function has also to be multiplied
by suitable trial functions in these additional spaces. These functions
may bring other parameters or sets of quantum numbers. These sets of
quantum numbers (e.g., total and intermediate spins, orbital angular
momenta, etc.) can be considered as discrete parameters, and will
often be referred to as channels in the following. (More details on the
trial function will be given in Chap. 6.)
The actual form of the basis function is not very important at this
stage. The above discussion serves to draw the reader's attention to
the fact that the basis function depends on many linear, nonlinear
and discrete parameters. The parameters define the shape of the basis
function and determine how well the variational function space con-
tains the true eigenfunction. To find the best possible solution, one
has to optimize the paxameters. To have a crude guess of how much
work the optimization amounts to, let us consider an N-particle sys-
tem with the simplest basis function (4.1). This correlated Gaussian
has N(N -
1)/2 parameters, and, by assuming that we need a lin-

ea,r combination of K functions, -we face an optimization problem of
4.1 Basis optimization 41
K(N(N 1) /2) -
paxameters. Tb is number increases quadratically with

the number of particles. By taking N = 4 and K = 200 as a typical
case, end up with 1200 parameters.
we
The main problem of the minimization of a function is the om-
nipresence of local minima. A local minimum is the point where the

function reaches a minimum in a finite interval of variables and the
number of such minima tends to increase exponentially with the size
of the problem. There are plenty of different methods for the
opti-
mization of a function, and these optimizations can be divided into
two categories: the deterministic and the stochastic optimizations.

A deterministic optimization moves downwards on the slope of the
function according to a certain well-defined strategy. There exist many
elaborated algorithms (conjugate gradient, Powell (direction set), etc.
[141) and they are deterministic in the sense that, starting from a
given point, they always reach the same (local or global) minimum.
The drawback of these techniques is that they are time consuming
and tend to converge to whichever local minimum they first encounter.
The solution in these cases may not be the global minimum but a local
minimum. These methods are sensitive to the starting point and are
unable to search further for a better solution after a local minimum is
reached.
Stochastic optimizations address the problem of finding the global
minimum in the presence of a large number of undesired local minima
by making random steps [15, 161. One can, for
example, start deter-
ministic searches from several random starting points and then pick
up the minimum of these. Numerous strategies have been developed
in the last few years, e.g., simulated annealing [171, genetic algorithm
[18] etc. The simplest (and actually not very economical) stochastic
optimization is a random search where one picks up random points
and tries to find the minimum. This may not sound very sophisti-
cated, but the optimization with a random trial and error procedure
seems to be the most efficient one. In some cases it would be nearly
impossible to apply other strategies than simple random trials.

To reduce the load of optimization, an alternative is to shift the
burden of minimization from a large number of parameters to a smaller
number of more sensitive, "tempering", parameters. Several such pos-
sibilities have been explored, e.g., geometric progression [19, 20], ran-
dom tempering [21, 22, 231, and Chebyshev grid [241. The common
property of these methods is that they use a "grW in the parame-
ter space. The grid is given by some ad hoe rule and each point of
the grids defines a basis function. The grid can be defined by some
simple functions which may depend on some additional- parameters
to be optimized. The number of parameters contained in the func-
tions which define the grid is chosen to be much smaller than that of
the original function to be optimized. For example, in the case of a
three-particle system, each of the basis functions has three parame-
ters: All, A12 A21 and A22 or a12, a13 and C923- Some COnVenient
=
choices for basis parameters are shown in Table 4.1.
Table 4. 1. Choice of parameters. In the random tempering p is an index to

denote a prime number that is used to generate pseudo-random numbers.
See Eq. (4.8) for the notation < i, i >.
Geometric progression
-2
All =
(aiqk-i)
I (k mi)
k -2
A22 =
(a2q2-1) (k M2)
A12 = 0
Random tempering
a12 =
exp(d, < k,p > +d2 < kp + I > (k I,- mi.) ,
OL13 =
exp(d3 < k,p + 2 > +d4 < k,p + 3 > (k 1: ... M2)
a23 =
exp(d5 < k,p +4 > +d6 < k,p + 5 > (k 1 ... M3)
Chebyshev gTid
2k-1
All = altan ( 7r
2 2 mi
(k 17 mi)
2k-I
a2tan (
7r
A22 =
2 2M2
(k 17 M2)
A12 = 0
4.2 A practical example 43
distribute a certain number of basis functions to approximate the bulk

part of the wave function and then one might increase the density of
the basis elements in regions where finer resolution is needed.
4.2 A practical example
To illustrate the point of the previous section, let us consider an exam-

ple for basis optimization. The example will at the same time provide
insight into some other aspects of the methods as well. We consider
the system of a positron and two electrons, Ps-, used in Chap. 1. The
energy of this system has been very accurately calculated by various

approaches and it has been found to be -0.262005 in atomic units
(a.u.). (Energy and length given in a.u. throughout this chapter

are
unless otherwise mentioned.) The quality of the different optimiza-

tion strategies used in this section can be judged by comparing their
results with this value. Let the positron be labelled I and the elec-
trons labelled 2 and 3. With the electron spins coupled to S 0, =
the antisymmetry requires that the orbital part of the trial function
ought to be symmetric with respect to the interchange of the electron
coordinates. The wave function is thus expanded as
K
1
Tf =ECk(1+P23) exp (-2-; ,Akx), (4.2)
k=1
where the exchange operator P23 is introduced to assure the sym-

metry requirement. The coordinates are x, rl r2 and X2
= -
-
(rI + r2)/2 r3, -

and Ak is a 2 x 2 symmetric matrix with three
nonlinear parameters All, A12 and A22. The wave function Tf has
3K parameters to be optimized. The linear parameter Ck is deter-
mined by solving the generalized eigenvalue problem (3.10). To reach
the accuracy required in atomic physics, the minimum basis size of
a three-body problem is at least K 100, as will be shown later.
=
That means that even in this simple example we would have 300 pa-
rameters. This is already almost beyond the capability of most of the
computer codes for optimization and this "full" optimization is out

of the question for larger systems. The fact that the optimization of
a large number of parameters is not feasible is just a small part of
the problem. To have an efficient optimization one needs fast function

evaluation as many times as required. There are two steps which are
necessary to calculate the energy expectation value: The calculation of

the overlap and Hamiltonian matrix elements and the diagonalization.
In a fall optimization, one has to recalculate all matrix elements (the
required time increases as K2) and to rediagonalize the Hamiltonian
(a K3-process) at each calculation of the energy. This is a considerably

heavy computational load
small system.
even for a
Instead, one can try a partial optimization. One can, for example,
fix the parameters of all basis states but one. In this partial optimiza-
tion only one row (column) of the Hamiltonian and the overlap -nn atrix
has to be recalculated (the required time is of order K). Let -us assume
that the Hamiltonian has been diagonalized over the fixed basis states.
Then, in the successive step, only one row (column) is changed. As has
been shown in Theorem 3.5, after the N x N diagonalization there is
no need for an extra diagonalization to solve the eigenvalue problem
on the (N + I)-dimensional basis. Consequently, the computational
time required by the partial optimization is only a small ftaction of

that of the full optimization, and, moreover, only a small number of
paxameters (three in the present example) has to be optimized.
4.2.1 Geometric progression
First let us try to use a grid defined by a geometric progression. One

can define three different sets of Jacobi coordinates, depending on
which two particles are connected first: (23)1, (31)2, and (12)3. See
Sects. 2.4 and 7.6 and Fig. 2.2. Let us denote the Jacobi coordinates
in these systems by x('), X(2) and X(3) The trial function is assumed
IT
to be given in the form
3 K
qf
P=1 k=1 1
(P)
Ckl exp(--l (-P)A(P)x(P))0(11)00(x(P)),
2
(4.3)
with the angular function
0(11)00 (X) =
[Y1 (XI) X Y1 (X2)100
I
(-1)1-M
-
-: 2= ,+1Y1Ta(X1)Y1-m(X2)- (4.4)
Here p denotes the arrangement channel and Y1. (,r) is a solid spherical
harmonic
YIM M = 7"YM (P), (4.5)

where Yl,,, (i ) is a spherical harmonic. The solid spherical harmonic is

a polynomial of degree 1 in the Cartesian coordinate (see Complement
6.2). A()
The matrix
k
is always chosen to be diagonal. Therefore, the
spherical part of Eq. (4.3) is considered a special case of Eq. (4.1). (See
also Complement 8.4.) The trial function of type (4.3) is used in the
so-called Coupled Rearrangement Channel Variational Method [20,
25, 26]. The basic idea of this approach is that, by taking into account
the different Jacobi sets, one can introduce various correlations. This
method has been used with great success especially for Coulombic
three-body problems [201.
Because of the symmetry requirement imposed on the orbital part
of the trial function, we may assume that
(2) (3) A (2) A (3)

C
k1
=
cki k
-
"k (4-6)
The 2 x 2diagonal matrix A(P)
k (k 1, K) has two nonlinear
=
...' pa-
rameters and they are taken as a geometric progression
(Ak(')),, (P)
k-1 (i =
1, 2). (4.7)
ai(p) (qi )
In principle one can use different geometric progression parameters,
ai(P) and qi(p), corresponding to the different arrangement channels and

the Jacobi that
is, they would depend on p and i in order
coordinates,
to get better convergence. They might even depend on the angular
momentum 1. For simplicity, in this example we use the same param-
eters, a and q, for all. arrangement channels and all sets of Jacobi
coordinates. The main reason for using the geometric progression as
parametrization is clear: The number of parameters of the basis is
reduced to just 2 (a and q), so the optimization is simple. Another
reason for the choice of this specific parametrization is that the over-
lap integral of the basis functions can be easily controlled by a choice

of q and thus the danger of the linear dependence of the basis func-
tions can be avoided. (See Sect. 3 for the danger arising from the linear
dependence of the basis functions.)

The disadvantage of the use of these diagonal matrices is that,
without having some polynomials like the scalar product of Eq. (4.4),
the energy does not converge to its accurate value. As is shown in
Table 4.2, one reproduces only the first two figures (E -0.2618530) =
of the
ground-state energy without the polynomial part, i.e., by taking
only I
0 in the expansion.
=
Table 4.2. The energy of Ps- in different arrangement channels. "Exact"

energy is -0.262005.
Channel Partial wave E (a.u.)

(23)1 1 = 0 -0.2068096
(23)1 1 =0,2 -0-2392726
(31)2+(12)3 1 = 0 -0.2609626
(31)2+(12)3 1 =
0,1 -0-2619622
(31)2+(12)3 1 =
0, 1,2 -0-2619717
(23)1+(31)2+(12)3 1 = 0 -0-2618530
(23)1+(31)2+(12)3 1 =
0, 1 -0.2619804
(23)1+(31)2+(12)3 1 =
0, 1, 2 -0.2619816
At this moment a comment is in order. As the spherical harmonies

form complete
a set for
angular functions, may think that the one
partial-wave expansion using only one of the Jacobi coordinate sets

would be sufficient to represent the wave function. The convergence
as a function of I is, however, very slow if only one set of the Jacobi
coordinates is used. slow convergence in the partial-wave ex-

(The
pansion will also be studied in detail in Complement 11.2.) E.g., the
energy calculated with the (23)1 channel using the partial waves of
up to I = 2 is just -0.2392726, which is muchhigher than the energy
of the three-channel calculation (E =
-0.2618530) using only I 0. =
Since the lowest threshold of Ps- is the Ps+e- channel at the en-
ergy of -0.25, this result indicates that this single-channel calculation

cannot bind the system. Moreover, the expansion with I has for prac-
tical purposes to be truncated to low values because the calculation
of the matrix elements of high I waves generally becomes more time
consuming. Table 4.2 shows that, by combining the expansion in the
Jacobi coordinate sets, the energy converges faster and only the first
few terms of the partial-wave expansion are needed. By using 10 grid
points for A(')
k
we have found the optimal values of the parameters to
be a = 0.06 and q 2.6. The variational energy obtained with this

=
600 (= 2 x 10 x 10 x 3)-dimensional calculation is -0.2619816. The

computational load with one particular basis set is the calculation of
180300 (= 600 x 601/2) matrix elements and the solution of a gener-

alized eigenvalue problem of dimension 600. For the optimization this
has to be repeated a few dozen times. By increasing the basis size one
can get closer and closer to the exact energy.

4.2.2 Random tempering
Another popular tempering method is random tempering [21, 22, 231-

This approach involves the generation of (pseudo-) random numbers
with the basis parameters defined by the following prescription:
ak exp 1: dj < k, j > (k =

1, ..., K), (4.8)
j=1
where < k, j > is a

pseudo-random numer, the fractional part of
(k(k + 1)/2)V/P---(j) P(j) being a prime number in the sequence
with
2, 3, 5, 7,..., that is, P(I) 2, P(2)=
3,... By using this tempering
=
formula, one optimizes a number I of parameters, dj, instead of the

original 3K parameters. A possible application of the formula for a
three-particle case is given in Table 4.1.
The origiTi of this formula is the following. One may assume that
the ground-state wave function is an integral transform of some known
function with some weight function. The known function in our case
is the correlated Gaussian and its nonlinear parameters are the inte-
gration variables. The simplest way to carry out the integral trans-
formation is to use the Monte Carlo method. The
quadrature points
required in this
integration can be generated by the above formula. As
this formula provides "good lattice points" for the integration, these
nonlinear parameters can be thought to be adequate to represent the
wave function in a variational approach.
While the random tempering works in many examples in a superb

way, one has to note that it has a serious problem: It often leads to
(almost) lineaxly dependent bases. In our actual example, the best
energy of Ps- with random tempering is -0.261872, and further hn-
provement of this value was difficult due to the linear dependence.
The parameters of the random tempering used are, in the notation
of Table 4.1, ?nj Tn2 7, Tn3 =
5, di 4, d2
=
-8, d3
=
4, d4 = =
-7, d5 =
-1, d6 -11, p =
1, and the basis dimension is K = 245.
The parameters di are nearly optimized.
4.2.3 Random basis
The above calculations have suggested that not all of the

grid points
are equally important but the different
grids give nearly the same
can
energy. The explanation is simple: The basis functions are nonorthog-

onal to each other, none of them is indispensable; they are dense, that
is,any of thein can be omitted because some others

will compensate
for the loss. This property of the basis functions and the success of
the random tempering suggest the idea of a completely random distri-
bution of the parameters. To illustrate this possibility, let us generate
K 100 sets of basis parameters in the expansion (4.2). The eleraents
=
of Ak are randomly chosen from a "physical" interval:

1 1
0 < < 10, 0 < < 10, 0 < < 20. (4.9)
V/a12 fa13 -\/FY23
The advantage of using the parametrization by aij instead of Ak
is that 1/.\,ra-jj corresponds to the "distance' between the particles.
(Note that, for a e-l"2, the expectation value rel-
Gaussian function 2
evant to the distance is, e.g., (r) V 4/(7a), or A,/-(r-27) VF3/(2a).)

= =
The paxameters inside this interval describe the (e+e-e-) system,

where the average distance between the positron and the electrons is
expected to be less than 10 a.u. and the distance between the electrons
is expected to be less than 20 a.u. This limitation serves practical pur-
pose, and it is based on the physical intuition that the paxticles
in a
bound system cannot move very fax away from each other.
The difference between the random tempering and the random
distribution of the parameters is that in the former there axe several
parameters which generate the parameters of the basis functions and
these generating parameters areoptimized, whereas in the latter the
paxameters of the basis functions are randomly chosen.
energies of different random bases are shown in Fig. 4.1. By

The
increasing the size of the basis, the energy goes very close to the exact
value. See Table 4.3. The energy is better than that obtained with the
basis set of geometric progression type, but the comparison is diffi-
cult. In the case geometric progression, the matrix of nonlinear
of the
paxameters is diagonal and the partial-wave expansion is introduced
to enable the trial function to span the full model space. This paxtial-
wave expansion increases the basis size significantly. When we use the
full A matrix and randomly distribute aij, this partial-wave expansion

is not necessary (see Complement 11.2). One may try to choose aij in
a geometric progression, but our experience is that both the geometric
progression and the random tempering of aij tend to lead to linear

dependence.
To avoid the linear dependence in the basis (which does not occur
frequently with the fully random basis), the overlap of the random
basis fimctions should be smaller than a prescribed limit. Any ran-
dom points which do not satisfy this condition are to be omitted and
-0.250
-0.252-
I
-0.254-
Ci
-0.256-
-0.258
-0.260
-0.269
20 40 60 80 100
Fig. 4.1. Energies of Ps- for different basis sets. The parameters of the
basis states are completely randomly chosen and no preselection is made.
Table 4.3. The energy of Ps- (in a.u.) for K basis states that are selected
randomly. "Exact" energy is -0.262005. The energies Ei are obtained by
starting from different random points. The energy of "Best 100" is the one
calculated by sorting the best 100 basis states from 400 basis states, where
50 random trials are probed at each step of the basis selection. Neal is the
number of matrix elements evaluated during the optimization.
K = 100 K = 200 K = 400 Best 100
El -0.2617619 -0.2619798 -0.2620032 -0.2619995

E2 -0.2617281 -0.2619793 -0.2620026 -0.2619978
E3 -0.2617918 -0.2619669 -0.2620016 -0.2619956
E4 -0.2618826 -0.2619675 -0.2620008 -0.2619953
E5 -0.2616285 -0.2619824 -0.2620024 -0.2619987
Nevai 5050 20100 80200 80200
replaced by a new trial. Note that this cure cannot be applied to the
procedures of the random tempering and of the geometric progression.
In these procedures there is no such prescription. One can just omit
some grid points depending on the actual parameters.
The energies listed in Table 4.3 are surprisingly good, but the size
of the basis seems larger than absolutely necessary. For the solution
of a three-body problem a basis dimension of K 400 is excessive.
=
We would like to decrease the basis size by selecting a minimal, indis-

pensable set of basis functions. This will be described in the following
subsections.
4.2.4 Sorting
One may wonder: Are all the basis states equally important? What
happens if we omit a few states from these bases? To answer this ques-
tion, we reordered the basis states by the following random selection
process:
1. A number n of basis states were picked up randomly and the one
giving the lowest energy was selected to be the first element of a
restricted basis.
2. Then n basis states, again randomly chosen from the remaining
pool, were tested in a 2-dimensional calculation with the first
state. Again the state that produced the lowest energy was se-
lected.
3. This reordering is continued until the last basis state.
4.2.5 THal and error search
The randomly chosen nonlinear parameters give good results provided

the basis dimension is big enough. For an N-particle problem, how-
ever, the basis dimension might become prohibitively large to use

this
simple procedure. One can obviously improve the convergence by se-

lecting the most important states and by not admitting all random
trials in the basis set. This increase of basis dimension by searching
the best among many random trials is a key of the stochastic varia-
tional method (SVM). The original procedure of the SVM, proposed in
[271, has recently been developed further [281 and successfully applied
to multicluster descriptions of light exotic nuclei [29, 30]. Learning
from these applications to nuclear few-body problems, we have gener-
alized and refined the method further to encompass diverse quantum-
mechanical few-body systems emerging in nuclear and atomic physics
[31, 32, 33].
The sorting method has shown that from large number of random
a
basis states one can select a smaller set without a significant loss of
accuracy. Moreover, the accuracy can be improved by increasing the
basis size. This experience motivates one to apply the following trial
and error procedure, combined with an admittance test, in order to set
up the most important basis set: Let Ak be the parameter set defining
the kth basis function, and let us assume that the sets A,,- Ak-1 I
have already been selected, and the (k l)-dimensional eigenvalue

-
problem has already been solved. The next step is the following:
Competitive selection
s 1. A number n of different sets of (Ak', ...' Ank) are generated ran-
domly.
s2. By solving the n eigenvalue problems of k-dimension, the corre-
sponding energies (Ek,..., Ek) are determined.
s3. The parameter set Ak' that produces the lowest energy from
among the set (El,...,Ek)

k
is selected to be the kth parameter
set.
s4. Increase k to k + 1.
The essential motivating this strategy is the need to sample

reason
different parameter sets as fast as possible. The advantage of this pro-

cedure is that it is not necessary to recompute the whole Hamiltonian
matrix nor is it necessary to perform a new diagonalization at each
time when a new parameter Ak is generated. See Theorem 3.5.
The convergence of the energy with the basis selected in this way is
shown in Fig. 4.2. The convergence is much faster than in the previous
case in Fig. 4.1, and it can be made even faster by increasing the
number n of trials. The energies found by using K = 10- and K = 100-
dimensional basis sets are shown in Tables 4.4 and 4.5, respectively.
-0.250
-0.252
-
-0.254
Ca
-0.256
-0.258
-0.260
-0.262
2-0 40 60 80 100
Fig. 4.2. Energies of Ps- for different basis sets that are selected by the
trial and error procedure
Table 4.4. Energy of Ps- (in a.u.) by different optimization strategies.

"Exact" energy is -0.262005. The basis size is set to K 10. The energies
=
Ei are obtained by staxting from five different random points. The number
n denotes the random candidates probed in the trial and error procedure
of SVM. Ne,,,,l is the number of matrix elements evaluated during the opti-
mization. Tn the case of the full optimization by the Powell algorithm 3900
diagona,lizations were also required. The time in units of seconds is on a
Digital Alpha 2100 (250MHz) workstation. The refining cycle is repeated
10 times.
Method Powell SVM(n=l) SVM(n=10) Refining(n=10)

El -0.261251 -0.244364 -0.251083 -0.261180
E2 -0.261398 -0.234532 -0.250729 -0.261165
E3 -0.261283 -0.236207 -0.252659 -0.261040
E4 -0.261259 -0.243323 -0.252956 -0.261175
E5 -0.261193 -0.226697 -0.251260 -0.261039
N,-:.v.i 214500 55 550 88055
Time 60 0.8 1.0 11
Table 4.5. Energy of Ps- (in a.u.) by different optimization strategies.

The basis size is set to K = 100. In the case of the fiffi optimization by the
Powell algorithm 7200 diagonalizations; were required. The refining cycle is
repeated only once. See also the caption of Table 4.4.
Method Powell SVM(n =

1) SVM(n =
10) Refining (n =
10)
El -0.26200016 -0.26176191 -0.26199427 -0.26200231
E2 -0.26199947 -0.26172812 -0.26199382 -0.26200351
E3 -0.26200164 -0.26179182 -0.26199733 -0.26200312
E4 -0.26200135 -0.26188261 -0.26199778 -0.26200301
E5 -0.26200193 -0.26162811 -0.26199836 -0.26200271
Neval 35466150 11615 86150 805050
Time 7200 27 43 195
These tables also contain a compaxison with the performance of a
deterministic procedure. One can conclude that one can easily reach
energy convergence with this simple selection procedure. Moreover,
the energy converges to the exact value.
4.2.6 Refining
It is obvious that the basis size cannot be increased forever. More-

over, when the Kth basis state is the previous states are
selected,
kept fixed. This means that we try to find the optimal state with re-
spect to previously selected basis states, but actually some of the basis
states selected earlier might not be so important anymore because the
succeeding states took over their role. So one may include a refining
procedure where the previous states are probed again as described by
the following steps:
Refinement cycle
ri. (A!,..., Ain)
random paxameter sets are generated.
r2. The parameters of the ith basis state are replaced by the new
candidates and the energies ElK ...Ek axe calculated.

r3. If the best of the new energies is better than the original one, then
replace the old parameters with the new ones, otherwise keep the
original ones.
r4. Cycle this procedure through the basis states from i = I to K.
Note that, again, there is no need for diagonalization because of The-

orem 3.5.
Table 4.6. Rnprovement of the energy of Ps- in the refining steps. The
basis size is K = 100 and 5 random states are probed. Atomic units are
used.
Refinement cycle E (a.u.)

(Starting value) -0.261992767
Ist -0.262002742
2nd -0.262003677
3rd -0.262004063
4th -0.262004178
5th -0.262004255
6th -0.262004314
7th -0.262004339
8th -0.262004372
10th -0.262004398
"Exact" -0.262005
4.2.7 Comparison of different optimizing strategies

for a longer time, we might get somewhat better energies, but the
available computer time is limited in any case. Moreover, the other
methods reach at least the same energy in a fraction of this time as
we have seen. Tables 4.4 and 4.5 show that the full optimization seems
to converge to different points depending on the starting point, which
indicates that the procedure leads to different local minima. This is
confirmed by explicit inspections of the parameters.
In the third column of Tables 4.4 and 4.5 the energies obtained by
using different sets of completely random parameters are given. The

next column shows the results of a trial and error selection and the last
one gives the energy after refinement cycles. We have carried out
10
refining cycles (each basis state has been cyclically probed 10 times) in
the 10-dimensional, and one refinement cycle in the 100-dhnensional
case.
In summary, one may conclude that the stochastic selection of ba-

sis parameters leads to energy convergence independently of different
random paths and, moreover, to very accurate results in a fraction
of time and by much computational load than the full optimiza-
less
tion. And while the latter is certainly superior in principle (one may
need fewer basis states to reach a given accuracy), the former suits
more practical applications. Moreover, in more complicated problems
where, in addition to the nonlinear parameters, one has to find the

most adequate channels (quantum numbers, etc.) to describe the sys-
tem, the random selection seems to be the only viable method. Since
the energy convergence attainable is of course not precise in a mathe-
matical sense, one should check the quality of convergence of the wave
function as well.
In the competitive selection the relatively best candidate found
in the steps is always admitted to be a basis state even though its
contribution to the energy may sometimes happen to be very small.
This indicates that the newly selected basis has very strong overlap
with (linear combinations of) the previous basis states. To avoid this
an alternative approach called a utility selection [281 can be used to
select the basis states. The steps go as follows:
Utility selection
A. Generate randomly a Ak.

parameter set
s2. Determine the energyEl(k) by solving the k-dimensional eigen-

value problem.
s3. Admit the parameter set Ak to be the kth element if the energy
gained by including it is larger than a preset value Je, namely
,El (k) < Ej (k -
1) -
JE.
Otherwise return to the step (sl) for the next attempt. If no pa-
rameter set was found to pass the utility test out of n consecutive
attempts, reduce JE to, say half of its original value, and return
to (sl).
s4. Increase k to k + 1.
When competitive selection is used, convergence is inferred by the en-

ergy curve's flattening out, while when utility test is used, convergence
is signalled by an insistent failure to -find further elements that pass
the test.
Instead of the random trial strategy described in (sl)-(s2)-(s3)-(s4)
and (rl)-(r2)-(r3)-(r4), one may think of more efficient and sophis-
ticated approaches, like simulated annealing or genetic algorithms.
The latter approaches may give faster convergence, although the ran-
dom selection strategy may have a slight advantage: It is easily imple-
mentable in a parallel way because the random trials are absolutely
independent of each other.
One should note that the basis elements selected by the above ad-
mittance tests are in general never optimal, not even locally. A better
candidate which will gain more energy could be found in the neigh-
borhood of the basis element admitted. The incorporation of a fine
tuning, that is, an additional search for even better parameters I the
vicinity of the element that passed the admittance test would clearly
accelerate the energy convergence. When the evaluation of the matrix
elements does not require heavy computational loads, this fine tuning
is recommended to reach faster convergence. Or even a determini tic
selection such as Powell's method [34, 351 might be used to find out
the locally best basis element.
4.3 Optimization for excited states
The Mini-Max theorem (see Sect. 3.1) shows that by diagonalizing

the Hamiltonian in a K-dhnensional basis one gets an upper bound
not only for the ground state but also for the excited states as well. It
may happen that the basis set found for the ground state is fairly good
to predict the energies of the excited states with the same conserved
4.3 Optimization for excited states 57
quantum numbers (angular momentum, parity etc.) ground

as the
state.By optimizing only the ground state, however, the energies of the
excited states will not necessarily converge (see Fig. 4.3). The energies
certainly decrease because we increase the basis size, but except for
the energy of the second excited state, the upper bounds are not very
accurate. (See Chap. 8 for the detail of the basis function used in the
present section.)
0.4
171
0.3
Cd
CY)
0.2
W
0.1
0.0
0 200 400 600
dimension of the basis
Fig. Convergence of the energies of the first five 'S states of the Helium
4.3.
atom when only the ground-state wave function is optimized. The energy
difference Ej Ei'act is shown in the figure. See Table 4.7 for the exact
-
energies.
As the ith eigenvalue Ej obtained by diagonalization is the upper

bound of the energy of the ith excited state, one may try to optimize
this upper bound to get an accurate estimate for the energy of the
excited state. In practice we repeat the same procedure as before, but
now the basis selection is governed by the requirement that the ith
eigenvalue, not the ground-state energy El, should be improved. To

get the ith eigenvalue we need at least an i-dimensional basis to start
with, but practical numerical considerations suggest that it is better

to start with a basis in which all the lower eigenvalues (k i 1) = -
axe already "stabld. This means that we need a first guess for the
lower eigenvalues, otherwise it may happen that when improving the

first excited state, for example, we pick up such components that lower
the ground state.
As the energies of the excited states of the Helium atom are known
to a high accuracy, we will test our strategy in this case. First we
optimize the ground state of the He atom on a K 100-dimensional

=
basis. The energy of the ground state is very accurate and even the
energy of the first excited state is acceptable. Starting from this K =
100-dimensional basis, we increase the basis size one by one, picking

up basis states which improve the energy of the first excited state. This
procedure quickly improves the energy of the first excited state, while
the energy of the ground state also improves a little bit (the important
thing to note is that it does not get worse) because the basis size is
increased. After reaching the basis size of K 200 we switch to the
=
second excited state and so on. The convergence of the result is shown
in Fig. 4.4. One th at the energy quickly converges to the exact

can see
value after its turn of optimization started.

The aboveprocedure gives us a basis where all the excited states
are accurate up to a certain digit. We can of course create bases which
give an accurate energy for an individual state, while the energy of

the other states might be poor. In that case we simply carry out the
stochastic search for a given state staxting from a first guess basis (like
the K 100-dimensional basis in the previous case). Tbis optimiza-
=
tion may include refinement cycles as well, if necessary. The results

we obtained in this way are compared to the "exact" (i.e., the best
calculation in the literature) values [36, 37] in Table 4.7. One can t1aus
get as energies
accurate for the excited statesground
as for the state.
Examples of caculation of the energies of excited states win be given

for the tdA molecule in Complement 8.4, the baryon spectroscopy in
Chap. 9 and the four-nucleon system in Sect. 11.2.
Before closing this section, we remark that the SVM assumes that
the numerical accuracy of the energy calculation in each step ishigh.
If the energy is calculated by, say the Monte Carlo method andhas an
inherent statistical uncertainty, then the admittance test of the SVM
will never work. Even when the energy is calculated by a deterministic
method, such as a diagonalization, the accuracy of the numerical cal-
culations has to be good enough to guaxantee the required precision
4.3 Optimization for excited states 59
0.005
0.004
0.003
Ca
>1
CD
0.002
0.001
k
0.000
0 200 400
Fig. 4.4. Convergence of the energies of the first five 'S states of the Helium
atom. The first one hundred basis states are selected to optimize the ground
state, the next one hundred (from 101 till 200) basis states are selected to
optimize the first excited state, and so on. The energy difference Ej -
Eie"'t
is shown in the figure. See Table 4.7 for the exact energies.
Table 4.7. Energies in atomic units of the ground state and the first four
'S excited states of the Helium atom. In column A the basis is optimized
successively for all the states as described in the text (K 500). In column
=
B the basis is optimized separately for each state, leading to five different
bases (K 600) tailored for the respective states. The "exact" values are
=
taken from [36, 371.

State A B "Exact"
Ei- -2.9037243758 -2.9037243769 -2-90372437698

E2 -2.1459737740 -2.1459740452 -2.14597404605
E3 -2-0612718887 -2.0612719880 -2.06127198974
E4 -2-0335865085 -2.0335866779 -2.03358671702
E5 -2.0211312479 -2.0211768312 -2.02117685157
of the energy. An analytic evaluation of the matrix elements is thus

an essential prerequisite in the SVM.
Recently an importance sampling algorithm called the stochastic

diagonalization method [381 has been presented to compute the small-
est eigenvalue and the corresponding eigenvector of extremely laxge
matrices. It is interesting to note that though the algorithm used there
has been developed independently of the SVM, it includes procedures
analogous to those of SVM. The stochastic diagonalization was applied
to matrices of order up to 1035 X 1035.
All the aboveprocedures are based on the minirni a ion of the
energy expectation value. This is, however, not the only possibility
to find accurate energies and wave functions. As described in Sect.
3.2, the minimization of the variance of the local energy is another

possibility in the variance be calculated
case can analytically. This
will be studied in a simple example of the following Complement.
C4.1 Minimization of energy versus variance 61
Complements
4.1 Minimization of energy versus variance

Here we by treating a solvable
compare the two ways of minimization
problem, an S-wave motion of
a particle m in the potential of of mass
V(r) =
-Vo exp(-r/a) (Vo > 0). By expressing the wave function Tf
as X/ /Tir, the Schr6dinger equation for this problem takes the form
h? d2X
2m dr 2
-Voe'Xp I
(-?')X=EX.
a
(4.10)
To simplifT the notation, we introduce
8Tna2 VO 2E,
= h2
I
and i/
V--8Tna ji2
(4.11)
The change of variable z exp(-r/2a) reduces Eq. (4.10)

= to the
differential equation of the Bessel functions
d2X 1 dX
Z-2
+ --
z dz
+ 1,2)X
(1 -Z2 _ = 0. (4.12)
The boundary conditions for bound states, X(r) = 0 at r = 0 and oo,

lead to the solution
r
X(r) = cJ, ( exp (_
r
2a
with C-2
=f'j ( exp(-- ))2
tj
0
2a
dr,
(4.13)
where the value of v is determined by the condition
J-(O =
01 (4.14)
and the smallest v value gives the ground-state energy El through Eq.
(4.11). In order to have abound state, the potential strength VO must
be such that larger than about 2.405.
is
To estimate the ground-state energy by the variational method,
let us assume a simple basis function for X, rexp(-ar/a), where a is
a variational parameter. There is no minimum of o-
2for a single basis
function. Table 4.8 compares the energy and the variance obtained by
using the minimization principle for E or o-2. The parameter values

of a are determined by the SVM. For a deeper potential of 5.0, a =
combination of two basis functions already reproduces the energy rea-

sonably well. For a very shallow potential of 2.6, however, at least =
62 Complements
Table 4.8.Comparison of variational solutions obtained by E-minimizattion.

and 2-minimization for an exponential well. Units of E and o-2 are
u
e/(8ma2) and (h2/(8ma2) )2 respectively. The mean values, (T) and

,
(T 2 1/2 ,
are in unit of a. The overlap integral of the variational solution
with the exact one is also given.
K E 0-2 (r) (T2) '21 Overlap

E-min. -3-5840 0.3236 1.4362 1.6169 0.99997
5.0 2 0-2_Min. -3.5824 0.05437 1.4301 1.6044 0.99983
Exact -3-5848 0 1.4376 1.6213 1
E-min. -0.016446 1.275x 10-4 9.4174 12.236 0.99999

2.6 5 U2_min. -0-016415 3.49Ix 10-6 9.1025 11.619 0.99940
Exact -0.016447 0 9.4331 12.285 1
C4.1 Minimization of energy versus variance 63
(T2-minimization
E -minimization
0 ...............................................
...............................................................................
.........
i wave function-I
0 10 20 30 40
r1a
Fig. 4.5. The local energy curve for the shallow potential: 2.6. The
exact wave function is also shown in an arbitrary unit.
5. Other methods to solve few-body
problems
In this chapter we briefly show the essential points of other ap-

proaches to few-body bound-state problems. There are many different
approaches and it is far beyond the scope of tbis book to discuss all of
them. Included here are either only those methods which have some
sort of connection with our approach or their results are frequently
compared with that of SVM, and therefore a short explanation might

be useful.
5.1Quantum Monte Carlo method:

The imaginary-time evo lution of a system
The evolution of quantum-mechanical system is governed by the

a
time-dependent Schr6dinger equation. When the time t is replaced by

an imaginary time, -iha (a > 0), the Schr6dinger equation transforms
into a diffusion-like equation
Xf (a)
-HTf (a), (5-1)
aa
which has the formal solution
Tf (a) = e-Ha Tf, (5.2)
provided th-ne-independent. Here T is an

that the Hamiltonian H is
initial wave function Tf (a 0). The imaginary-time evolution of the
=
wave function can be used to estimate the ground-state energy. The

basis of this idea is
Tf (a)- a,
lim _
(5.3)
a-W
(qf- (a) I qf- (a) ) -i jail
that is, the wave fimction TI(a) approaches the exact ground-state
wave function at a --+ oo. To prove this, we only need to use the

66 5. Other methods to solve few-body problems
expansion (3.3) for TI in Eq. (5-2) and

that a, :7 0, namely
assume
the initial wave function has a non-vanishi -n g overlap with the ground-
state wave function. In what follows we assume that this condition is
fulfilled.
If we consider T/"(a) a variational trial function depending on the
imaginary time a, the mean value E(a) of H in Tf (a)
gives an upper
bound to the ground-state energy El. In fact it is easy to show t'h at
(Tf (a) I H I Tf (a))

E(a)
R(a) I Tf (a))
E+E-E -2a(Ei-EI) aj 12
I i=2 ie I a
i=2 e-2a(Ei-EI)JLi!J2
1+EI ai
E'
i=2 (Ei
-
E1)e-2a(Ei-E1.) I ai
12
EI+
1+1:' e -2a(Ei-El )I
ai
12
2` > El. (5.4)
i=2 a,
Clearly E(a) goes to E, when a goes to infinity.

Another quantity called the asymptotic energy estimator is defined
by
1
In
(Tf(a) I Tf (a))
E(a, -r) =
* (5-5)
2T (Tf(a + T) I Tf (a + -F))
This has the following properties:
Jim E(a, -F) =

E(a), (5-6)
-r--+O
-2a(Ei-El) af
+ EOO
i=2
e I ai 12
E(a, -F) =
E, + -In '
(5.7)
2,r I+ J:" e -2(a+-r) (Ei -El) I.Es! 12
i=2 a,
Equation (5.6) can be derived by noting that the left-hand side is

equal to ( '1/2) -da.Lln(Tf (a) ITI(a)) and that the derivative can be easily
calculated by Eq. (5. 1). It is clear that, for T > 0, E(a, 7) ! El and
E(a, -r) converges to EI_ as a goes to infinity.
In fact, the quantum Monte Carlo (QMC) method or the Green's
function Monte Carlo (GFMC) method [40-44] uses e-Ha as a vehi-
cle toproject out the ground state from the initial wave function. The
point here is not a construction of a good state space which approxi-
mates the solution but a calculation of the imaginary-tiTne evolution of
e-H'. In general, one cannot compute e-Ha = e-(-ffo+H')' because Ho
and H, do not commute, but by dividing the time a into rn
any smaH
5.2 Hyperspherical harmonics expansion method 67
steps Aa =
a/n, we can approximate G(R, k) =- (RIe-"0'Ik) by
(RIe-H0A'e-HIA'Ik) in a factorized form. Then the full propagator
(RIe
-H,
IRI =
ff _f G(R, Rn-1) G(Rn-1, Rn-2)
- - ' -
x G(RI,k)dPfn-IdRn-2 * ..
dRI (5-8)
can be evaluated by the Monte Carlo method. In practice, one must
use several time steps Aa and extrapolate to da = 0 in order to

eliminate time step errors arising from the non-commuting nature of
the kinetic energy and potential energy operators.
When the propagator (5.8) to calculate
we use E(a) in Eq. (5.4),
both the numerator and denominator of Eq. (5.4) a're subject to the
statistical errors of Monte Caflo
integration. Then the nice feature
of having an ground-state energy may be lost in
upper bound to the
the QMC method. One of the most serious problems in the QMC is
the so-called minus-sign problem [43, 451. The propagator G(R, k)
is not always positive. This makes it difficult to use importance sam-
pling techniques based on the classical stochastic process such as a

Maxkov process or a Molecular Dynamics technique in computing the
propagator by the Monte Carlo method.
5.2 Hyperspherical harmonics expansion method
We briefly introduce an approach which is based on the expansion of

the trial function in terms of hyperspherical harmonic (HH) functions.
Just as spherical harmonics are useful to expand the angular func-
the
tion of a single-particle motion, the basic idea of the HH method is
to generalize this simplicity to a system of particles by introducing a
global length p called the hyperradius and a set of angles Q.

Let us suppose that all of the particles oscillate harmonically with
angular frequency w
N 2 N
Pi
HHO + rniw2Ir2 (5.9)
2mi 2
Using the Jacobi coordinate set, one can separate the Hamiltonian
into the intrinsic and center-of-mass parts
N-I N-1
7r2 1 7r2N 2
HHO +
_2
/-Ziw2X2i + N+ 2 'rn12---NWXNi
2yi 2rn12 ...
(5.10)
where Ai is defined by Eq. (2.19).

Among the 3(N 1) degrees of freedom for the intrinsic motion,
-
the hyperradius p is defined by

N-I
2
P [lix? (5.11)
The hyperradius contains only intrinsic coordinates, but it can also

be expressed as E i_=, Mi(Ti XN )21A, which is proportional to the
-
moment of inertia of the system (see Eq. (2.18)). The choice of IL can
be axbitrary, The hyperradius is apparently symmetric with respect to
the interchange of particle labels. The potential energy of the intrinsic
part is simply expressed as -Ilzw
2
2P2. Just as the single-particle kinetic
energy operator is separated into radial and angular parts, the intrinsic
kinetic energy operator is reduced as follows
N-1 2
7r2 h2 02 3N -
4 0 L
I: 21.Lii
i=1
-
2jL [5;;72 +
P -ap p2
(5.12)
where L is called the grand angular momentum, which is expressible

in terms of 3N 4 angles S2.-
Although there is no unique way to choose the angles, one possi-
bility is to choose 2(N 1) angles &j and 0j, the polar coordinate of
-
xi (i 1,
= N 1) and the other N 2 angles -yk (0 < -yk E,
...,
-
2
k -
2,3,..., N 1) called the hyperangles as follows

-
-VFILIV-1 XN-I =
V/jL P COS7N-1,
V Ak Xk =
- ,FA P Sin7N-1 ... Sin'Yk+1COS7k (k =
2, ...,
N -
2),
,//-Ll x, =
Vrj-Lpsin-yjv-j ... siny3sint2. (5.13)
Then L is related to the usual angular momentum 1k = "Xk
h
X -irk as
follows
2
L2 = LN-D
2
a2
Lk -
-
a7k
2 3(k -
2)cot-yk + 4cot2-yk la-lk

Lk-I
+ + (k 2, N 2),
COS 'Yk sin -YA;
5.2 Hyperspherical harmonics expansion method 69
2
L1 =121. (5-14)
The eigenfunctions of L 2, called the HH functions,
L 2y/C (S?) =
K(K + 3N -
5) Yjc (Q) (5.15)

are chaxacterizedby the quantum number )C which comprises K (K
0, 1, ...) 5 eigenvalues of the angular operators.
and 3N -
Expressing the intrinsic wave function as a sum, over k, of products

of radial functions RK(p) and the HH functions Y)c(Q),
3N-4
-V Ep- 2
RIC (P) Yk (5-16)
/C
the N-body Schr8dinger equation, after eliminating the center-of-mass

motion, is reduced to the matrix equation for the radial functions
d2 'C(C + 1)
(_W [ dp2
21L P2 E) (p)
+ 1: (YK 11: Vii I yic) RIC, (P) =

0, (5-17)
with
L=K+ 3(N
2
-
2). (5.18)
The volume element is given by

N-i
fj dxi P3N 4dpdQj

MIL2, AN-I
N-1
x fj df2kSin3k-4,YkCoS2,YkdYk, (5-19)
k=2
where df?k Sin?ykd'OkdOk is the usual surface element. Because C is

=
positive for N > 3-particle systems, it turns out that the "centrifugal
baxrier" is always present in N > 3-particle systems even for zero total
orbital angular momentum states.
For the HH expansion method to be practical, of course, the evalu-
ation of the potential energy matrix elements must be feasible. More-
over one has to have aguide for the truncation of the HH basis, even
though the
expansion converges with a few K values, because the de-
generacy of the HH functions is high. It is also necessaxy to construct
the HH functions of proper symmetry for a system of identical par-
ticles. This is a difficult problem, and it limits the
application of the
HH expansion method to systems of rather small numbers of particles
(N < 5).
A sophisticated version of the HH expansion method [461 employs
N
a correlation factor F =
113'>i=l hi ri -
rj) [471, replacing TV in Eq.

(5.16) with FT. In [36, 481 the correlation factor is chosen to satisf ,
a special boundary condition like the cusp condition for the Coulom-
bic system. (See Complement 8.1 for the cusp condition.) The use of
the correlation factor accelerates the convergence and leads to precise
solutions.
The HH expansion method was introduced in 1935 by Zernike and
Brinkman and reintroduced 25 years later by Delves and Smith. See
for example [49, 50] for details and [511 for recent developments.
5.3 Faddeev method
T =
'01 + V)2 + ?P31 (5.20)
where for the bound state each component is related to Tf by
*j =
GoViTf, or (E -110)0i =
ViTf, (5.21)
where the free three-body propagator is given by
I
Go -
-
- (5.22)
E -
Ho'
Here Ho is the three-body kinetic energy with the center-of-mass -ki-
netic energy being subtracted. The potential Vi is expressed in terms,
5.3 Faddeev method 71
of "odd man out"

notation, that is, V1 V23 V2 V31 V3 V12 It
:::--
i
: --
i -
is easy to see that Eqs. (5.20)-(5.22) is equivalent to the Schrbdi-nger

equation (E H) Tl'= 0 with H
-
HO + VI + V2 + V3 =
We show two forms of the Faddeev equations which are widely

used for numerical solutions. One is the differential form which results
by rewriting the second form in Eq. (5.21):
(E -
Ho -
VI),01 = V1 (02 + 03)
(E -
Ho -
V2)02 =
V2 (03 + 01)
(E -
Ho -
V3)7P3 =
V3 (01 + 02) (5.23)
Another is the integral form of the above equation
Oi =
GiVi(Oj + Ok) (j 7 ij k =7 i, j :7 k), (5.24)
with
Gi =
(5.25)
E -
Ho -
Vi'
The integral form can be simply expressed in a matrix form
G, 0 0 0 V, Vi '01
02 0 G2 0 V2 0 V,2 2
03 0 0 G3 V3 V3 0 03
(5.26)
The bound-state solution has to satisfy the boundary condition that
all of the components Oi large distances. This condition is,
vanish at
however, met only for a particular value of energy E, which is notbing
but the energy eigenvalue of the three-body system.

The Faddeev method is thus a direct method to solve the eigen-
value problem. It does not rely on the variational principle. There are
three possible choices for the Jacobi coordinates X(1), X(2)7 X(3). See
Fig. 2.2. It is natural to express the Faddeev component 01 in terms
of x(l), and likewise 02 by X(2) and 03 by X(3) .
The most difficult part
in numerical works of the Faddeev method then comes from the eval-
uation of matrix elements between the functions expressed in terms
of the different Jacobi coordinate sets. Readers are referred to [55, 561
for some details for nuclear and Coulombic three-body bound-state
problems.
In Chap. 11 we will show that a optimization of the cor-
careful
related Gaussians with the use of the SVM gives solutions that are
in excellent agreement with the results of the Faddeev method for

nuclear three-body problems.
A real power of the Faddeev method is probably not only in bound-
state problems but in applications to scattering and continuum prob-
lems. See [541 for the recent developments on this subject.
It is possible to extend the Faddeev method to four-particle system.
The Faddeev equations are then called Faddeev-Yakubovsky equations
which consist of seven components, four for I + 3-partition of four
particles and three for 2 + 2-partition. Numerical solutions become
increasingly difficult.
5.4 The generator coordinate method
There is a theory called the generator coordinate method (GCM),

which is a natural extension of Eq. (3.7) to the case of continuous
superpositions. The GCM is a powerful method and has a variety of
applications such as the restoration of symmetries or good quantum
numbers (e.g., angular momentum and particle number) In many-body
wave functions, the microscopic description of reaction dynamics be-
tween two composite particles and collective motions of many-body
systems.
The GCM assumes that the trial function is a continuous superpo-
sition of the basis function Tf (a), which is often called the generating
function:
Tf =
f f (a) Tf (a) da. (5.27)
The function f (a) is called a weight function. A real or complex pa-

rameter a specifies the generating function. It is called the generator
coordinate to distinguish it from a physical coordinate. The symbol
a may represent more than one parameter: a =
(al , a2l ... ). The in-
tegral in Eq. (5.27) is then a multidimensional integration over all
the parameters. The choice of the generating functions Tf(a) is most
important in the GCM.

The variational principle is used to determine f (a) and leads to
the equation known as the Hill-Wheeler equation [571
f H(a, a')f(a)da' El B(a, a')f (d)da',

=
(5.28)
with
5.4 The generator coordinate method 73
(a, a) =
(Tf (a) I H I Tf (a)), B (a, a) =
(Tf (a) I Tf (a)). (5.29)
Here li(a, a) and 13(a, a') are linear integral operators and called
the Hamiltonian kernel and the overlap (or norm) kernel, respectively.
Formally the GCM looks very similar to the diagonalization of the
Hamiltonian in the basis states TV (a). This is true if the parameter a
is discretized and if only a finite number of functions Tf (al) , ... Tf(aK)
i
are used. Then the HiU-Wheeler equation reduces to the generalized
eigenvalue problem (3.10). For further details see [57-601. Examples

of using discretized generator coordinates are given in Sects. 8.4 and
10.4.
6. Variational trial functions
Due care of the correlation between the particles is important for a
precise variational solution. The variational trial functions used in this

text, correlated Gaussians and correlated Gaussian-type geminals, are
formulated by using a generating function g. Orbital functions with
arbitrary angular momenta or Cartesian polynomials around arbitrary
centers are constructed from g by using a simple, well-defined prescrip-
tion. It is also shown that the generating function can be related to
the product of single-particle Gaussian wave-packets through an in-
tegral transformation. This 'uncorrelated' form is useful to extend to
a many-body system of identical particles. The spin function for an
N-fermion system is briefly discussed.
6.1 Correlated Gaussians and correlated
Gaussian-type geminals
Most crucial for the variational approach is the choice of the trial
function. To solve an N-particle problem, it is of prime importance to
describe the correlation between the particles properly. The correlation
between the particles can be described by functions of appropriate
relative coordinates. The correlation is then conveniently represented
by a correlation factor, F rl3',,i=, fij (ri rj). There are two widely
= -
applied strategies: (1) One is to use this form of F directly by selecting

the most appropriate functional form to describe the short-range as
well as long-range correlations. Such calculations are, however, fairly
involved for systems of more than three particles and the integrations
involved require the Monte Carlo technique. (2) An alternative way
to incorporate the correlation is to approximate fij by a number,
possibly a large number, of simple terms which facilitate the analytical

calculation of the matrix elements. We follow the second course by
using an expansion over a correlated Gaussian "basis".

76 6. Variational trial functions
important that two basic requirements axe satisfied in the sec-

It is
ond course. First, in the limit of large dimensions the basis functions
should become complete, so that the results obtained by means of some
systematic increase of the number of basis functions could converge
to the exact eigenvalue. Second, for the approach to be practicable,
computational effectiveness is required for the basis functions. The

matrix elements have to be evaluated with ease; otherwise calcula-
tions witli combinations of a great number of basis functions would be
extremely difficult. In our opinion only the correlated Gaussian basis
meets these requirements. The usefulness of Gaussians was already
suggested in 1960 independently by Boys [611 and Singer [621, and

since then exploited by many authors [63, 22, 64, 65, 661. Though a
mathematically sound proof may not be available for the complete-
ness of the Gaussian functions, a heuristic argument for it is possible
as follows: As is discussed in Complement 6.1, any square-integrable,
well-behaved function with angular momentum 1m can be approxi-

mated, to any desired accuracy, by a linear combination of nodeless
harmonic-oscillator functions (Gaussians) of continuous size param-
a
eter a (Eq. (6.37)). By generalizing this to the N-particle system, the

N-particle basis function then contains a product of these Gaussians:
N
F= 113" ,,j=jexpf--!a--
2
(r- U ?,
-
rj)21 =
eXpj -.1
2
EN
j>i=l ozij(ri -,rj
)2j.
This simple correlated Gaussian is actually widely used in variational
calculations. For a discussion on the completeness of Gaussians for
specific cases, see for example [67, 351.
Of course, we have to keep in mind that the Gaussian is not eco-
nomical in describing the asymptotic behavior of the wave function at
laxge distances (see Fig. 6-2). Moreover, it does not predict a correct
value for specific quantities such as the cusp ratio. See Com-
some
plement asymptotics and the cusp ratio could be well

8.1. Both the
described with exponential functions. For example, welinow that the
Rylleraas-type functions give very accurate results for Coulombic few-
body system (see, e.g. [68, 691). The correlated exponential functions
are not, however, amenable to analytic evaluation of matrix elements
for a system of more than three paxticles. This makes it difficult to use
the exponential functions as a vahational trial function for a general
N-particle system.
We extend the above argument further to define the correlated
Gaussians. For this we consider separately two cases for two types
of Hamiltonians. First in case of Eq. (2.1) the Hamiltonian can

be expressed hn terms of a set of independent relative coordinates
6.1 Correlated Gaussians and correlated Gaussian-type geminals 77
,7
x =
(xi, ...
, xN-1). As was shown in Eqs. (2.23)-(2.25), it is then con-
venient to express F in terms of x, instead of N(IV- 1) /2 interparticle-

distance vectors, ri rj. An N-particle basis
-
function, a so-called
correlated Gaussian, then looks like
Type I: TI =
exp (_2 :TbAx) 0 (x), (6-1)
where A is an (N -
1) x (N -
1) positive-definite, symmetric matrix

of nonlinear parameters, specific to each basis element. As mentioned
in Sect. 2.2, the matrix A with these properties can in general be
written as 6DG with the use of an orthogonal matrix G and a diagonal
matrix D with all positive diagonal elements. The function O(x) is a
generalization of the solid
spherical harmonics Y of Eq. (6.56) to the
many-particle case. More details will be given below.
The second case is suited to the Hamiltonian (2.2). The motion of
each particle is governed by both the
single-particle potential and the
two-body interaction. It is thus useful to extend
Eq. (6.1) to include
an independent motion of the ith paxticle around some point R, Here
PI, is not a dynamical coordinate but just a parameter vector. In this

case we do not need to use the relative and center-of-mass coordinates
but can use the single-particle coordinates. Therefore we are led to the
following type of correlated Gaussians which are often called correlated
Gaussian-type geminals
N
expf -
2
aij (,ri _
T3)2
XeXpf -
2
Y i(ri -
Ri )210(,r _
R).
Here r -
R stands for a set of vectors firl -
RI, ---,,rN -
RNJ_ As
N
E j>i=, aj(,r, _,rj)2 can be written compactly as Ar with a matrix
A defined through Aij =

Aji =
-aij (i < j), Aii =
E''-=, aji +
N
3=i+l aij, the correlated Gaussian-type geminals can in general be
expressed as
Type 11 : Tf =
exp 1 -2 Ar -
2
(r -
R)B(r -
R) I
x O(r -
R), (6.2)
optimization of nonlinear parameters allows us to obtain high-quality

solutions with expansions containing not too many terms.
The Gaussian of type I is a special case of type IL To unify the
description for both cases, we change the notation in what follows
throughout this chapter according to the following conventions: As it
is cumbersome to use x or r depending on the type of problems, we
will use x even in the
of the presence of an external field. To have
case
N as the number of xi coordinates in both cases, we consider (N + 1)

particles for the basis of type I and N particles for the basis of type II,
respectively. Table 6.1 summarizes our conventions for the notations.
Table 6.1. Convention of the coordinates jo `:

(XII X21 ---I XN) in the cor-
related Gaussians and the correlated Gaussian-type geminals. SP and CM

are abbreviations of 'single-particle' and 'center-of-mass', respectively.
Correlated Gaussians Correlated

(Type I) Gaussian-type
geminals (Type II)
Number of particles N+I N
Meaning of x Relative coordinates SP coordinates
T ansformation matrix U: X = Ur U=1: X=r
from SP coordinates r
I IV
CM coordinate: xcm XN+1 Ei==1 mixi
7nl2---N
SP coordinates relative (U-1X)i Xi -
XCM
to CM: ri -
XCM
Relative distance vector (U-I-X)i -
(U-'x)j Xi -
Xj
ri -
rj
Before discussing the function O(x) in Eq. (6.1) we distinguish two

types of variational calculations: One is a type of calculation called a
vaTiation before projection. Here the trial function used is not neces-
sarily an eigenstate of some of the conserved quantities belonging to

the symmetries of the Hamiltonian such as paxity and angular mo-
mentum. After the variational calculation, the conserved quantities
are proje-ted out from the trial function. For example, we know that
an eigenstate of a rotationally invariant Hamiltonian should have a
good angular momentum. Yet one often uses a variational trial func-
tion that does not have the proper rotational symmetry but, after the
80 6. Vaxiational trial functions
variational calculation, the symmetry is restored by angular momen-

tum projection. The restoration of good angular momentum is ensured
by superposition of the rotated variational solutions and the weight

a
function involved in the superposition can be determined, e.g., by the

generator coordinate method of Sect. 5.4. Another type of calculation
is what is called variation after projection, where the trial function is
constructed so as not to mix different quantum numbers of the con-
served quantities. Clearly the variation after projection is superior to
the variation before projection because the variation in the former case
is carried out only in the state space which has the same symmetry as
the exact solution, while in the latter the variational solution tends to
reach a minimum in the space including the basis states with different
quantum numbers. Even a calculation of variation before projection
type, if thoroughly done, would reach a solution that has the proper
symmetry in the limit that the state space is complete. See Sect. 10.4.
Of course calculations of the variation after projection type become
more challenging because it is in general not easy to calculate matrix
elements with trial functions that are eigenstates of the conserved ob-
servables. Our objective is, however, to use trial functions with the
proper symmetry.
The function O(x) in Eq. (6.1) describes non-spherical motion. To
describe the orbital motion with the orbital angular momentum L and
its projection M, a direct generalization of solid spherical harmonics Y
(4.5) (see Complement 6.2) to a many-particle case is a vector-coupled

product of the solid spherical harmonics of the relative coordinates:
OLM* [[[Yi, (XI) )( Y12 (X2)IL12 X Y13 (X3)IL123
X ...
X Y1' *V) I LM
C. Ylimi (xi), (6.3)

r-=JM1.IM27 ... 7MNI
where c,, is a product, (IIM112M2 IL12 M1+M2) (L12 M1+M2 13M3 IL123
M1 +Tn2 +M3) - - -
(L12 N-IMI+?n2+-+TnN-1INmNJLM),
...
of the
Clebsch-Gordan coefficients needed to couple the orbital angular mo-
specified quantum numbers. Here each relative motion
inenta tG0 the
has a definiteangular momentum. See, e.g., [2, 70, 71, 721 for the de-
tails of angular momentum algebra. Angular momentum recoupling
coefficients used in t'his book will be defined in Complement 6.3.
Since theangular momentum of the relative motion is not a con-

served quantity, it may be important for a realistic description to in-
clude several sets of angular momenta (117 12 IN; L12, L123 ) This
7 .... i ... -
is the case especially in nuclear few-body problems [32, 251 because

the non-central components of the nuclear potential necessitate higher
partial waves. For example the tensor force couples S waves to D
waves. It is also noted that a faster convergence is in general obtained
by allowing the use of different sets of relative coordinates together

with suitable sets of angular momenta [30, 20, 251, because a par-
ticula,r type of correlations can be best described in the coordinate
set conforming to the type of correlation. From the fact that OLM(X)
can be expressed by different paxtial-wave decompositions in different
relative coordinate systems, one can conclude that the use of partial
waves may not be so important after all. Besides, the various possible
partial wave contributions increase the basis dimension. Moreover, the

calculation of matrix elements for this choice Of OLm (x) sooner or later
becomes too complicated. This choice is therefore
obviously inconve-
nient especially as the number of
particles increases and/or different
sets of relative coordinates are employed.
This difficulty can be avoided by adopting a different generalization
of Y for OLM(x) [33, 311:
OLM(X) V2KYLM(V)
N
V 2K+LyLob) with v Uixi = fix. (6.4)
Only the total orbital angular momentum, which is a good quantum

number in most cases (at least approximately), appears in this ex-
pression. The real vector ii =
(ul,..., uAr) defines a global vector, v, a
linear combination of the relative coordinates, and the wave function
of the system is expanded in terms of its angle b. The vector u may be
considered a variational parameter and one may try to minimize the
energy functional with respect to it. The energy minimization then

amounts to finding the most suitable angle or a linear combination
of angles. The continuity of the parameter u can be more advanta-
geous in a variational calculation than the discrete nature of the set of
the angular momenta (111 12
IN; L12, L123 -) because the change of
7 ..., i .
the energy functional can be continuously seen in the former case. The
factor of v2K+L plays an important role in improving the short-range
behavior of the basis function. A remarkable advantage of this form
82 6. Variational trial fimetions
of OLm(x) is that the calculation of matrix elements becomes much
simpler than in the former case because the coupling of N angular

momenta is completely avoided. See the appendix for details.
For the function Eq. (6.2) the construction of the
0(,r -
R) in
trial function with good orbital angular momentum may not be an
immediate concern, because the vector R is intended to determine a
specific shape of the system. The function 0(r R) is chosen in the -
spherical basis as
O(x -
R) IX, _nz.12kiy1 imi (X, _
IZZ (6-5)
or in the Cartesian basis as
N 3
0 (x -
R) fl 11 (xip -
Rip)nip, (6-6)
i=1 P=I
where the index p = 1, 2, 3 stands for x, y, z components of the vectors

xi and R, and nip is a non-negative integer for the p-direction of the
ith particle. simple transformation for the polyno-
Because there is a
mial parts between the spherical basis and the Cartesian basis, the
above two representations are actually equivalent. See Complement
6.2. In the following we assume that 0 is given by Eqs. (6.3) or (6.4)
for the correlated Gaussian of type I and by Eq. (6.5) or Eq. (6-6) for
the correlated Gaussian-type geminal of type II.
The Gaussians of type I have definite parity of either
or (_l)L, depending on the choice Of OLM(X), while the Gaussians of
type II are not eigenstates of the parity operator. To project out good
parity from the trial function, one has to take a combination of two
Gaussians with the centers at R and -R.
6.2 Orbital functions with arbitrary angular

momentum
The construction of a trial function with good angular momentum

is very important for obtaining solutions of a rotationally invari-
ant Hamiltonian. In the previous section we introduced the forms of
OLM(X) to describe the orbital fimction with good angular momentlan.

Since Eq. (6.3) appears to be better established but Eq. (6.4) takes
6.2 Orbital functions with arbitrary angular momentum 83
a particularly simple form and makes the calculation of matrix ele-

ments very simple, it is useful to understand the relationship between
the two angular functions.
Theorem 6. 1. Any functions of type

N
v2KyLM (V) with v uixi
can be expressed in terms of a linear combination of terms
2ki
XI
2k2
X2 x2k' [[[Yl (XI) X Y12 (X2)] L12 X Y13 (X3)] L123
X * * *
X YIN (XN)l LM7
where ki and li are all non-negative and satisfy 2k, + 11 + 2k2 +12 +
- -
-+2kv+IN =2K+L.
Proof. We prove the theorem by induction. For a two-variable vector

V =
U1XI + U2X2, the assertion is true because we know the following
equality [73, 747 75] (see Exercise 6.1 for a simple proof):
KL
v2KYLM(V) D
kj.Ijk212
U2k3-+llU2k2+12
1 2
2kl.+11+2k2+12=2K+L
XX 2klX2k2
1 2 [Y11 (X1) X Y12 (X2
LM7
(6.7)
where DKL is given by

kjIjk212
DKL
(2K + L)! Bkj.,,Bk2l2
kjIjk212
-
-
Q1112; L). (6.8)
(2k, + I,)! (2k2 +12)1 BKL
Here Bkj is defined by
47r(2k + 1)!
BkI -
(6.9)
2kk!(2k + 21 + 1)!!'
and C is the coefficient needed to couple two spherical haxmonics with
the same argument [70]:
1YI 00 X YI'001 LM =
C(Il; L) YLm (i), (6.10)
and reads as
1 +::1:
Qll'; L) F2 (21+1)(211+1)
L
4v(2L
47(2
4v(2 + 1)
(101'OILO). (6.11)
The coefficient C vanishes unless I + I' + L is even.
For a general case of the vector v =

ulxl +-,LulvxN , we put v
* -
VN-1 +UNXN With VN-1 =

U1X1 + * * *
+UN-lXN-1 and use the above
formula to show that V2KyLM(V) can be expressed in terms of the
2K12 N-1
of the terms of
...
vector-coupled products v
1 YL12 ... N-I (VN-1)
2kN
and x
N Y1N (XIV) with 2K12 m-1 + L12 N-I 2K + L 2kN 1N-
... ...
= - -
By further decomposing VN-1 to VN-2 + UN-IXN-I, one can apply

2KI2 N-1
(VN-1)
...
the same argument to vN-1 YL12 ... N-1

th is process
and use
repeatedly. Then it is clear that we can show that the statement is

true.
Theorem 6.2. A vector-coupled function of two solid spherical

harmonies
[Y11 (Xl) X Y12 (X2)] LM

with (-1)11+12 =
(_I)L
can be expressed in terms of a linear combination of terms
X2ki.x2k2V2qyLM(V),
1 2 (6.12)
where 2k, + 2k2 + 2q =
11 + 12 -
L and the vector
of each term has v
the- form Of V =
UIXI + U2X2 with appropriate coefficients ul and U2 -
Proof. For a given L value, 11 + 12 -

L is even and non-negative, and
thus 11 + 12 may be set equal to 2k + L with a non-negative integer k.
We use induction with respect to k. First we show that the statement
is true for k 0 (11 + 12
=
L). As Eq. (6.7) shows, for the K = 0
=
case there are only terms with k, (see also Eq. (6. 108)), and
=
k2 = 0
YLM(V) With V =
UIXI + U2X2 consists of (L + 1) terms of [YI(xi) x
YL-I (X2)] LM (I =
0, ..., L), each multiplied by UI1 UL-1.
2 By taIdng (L +
1) different vectors, vi =
aix, + X2 (i L + 1) with ai =7 0, we
obtain
YLm(vi) =
E(aj)'
1=0
-21+
4-Ax(2L
1
71
1.
1+ 1) (2L
+ 1)!
21 + I)! -
X [YI(XI) X YL-1(X2)]LM- (6.13)

One can view the above equation as a system of simultaneous linear
equations for [Yl(Xl) X YL-I(X2)]LM (multiplied by the square root
6.2 Orbital functions with arbitrary angular momentum 85
factor). The determinant of (L + 1) x (L + 1) coefficient matrix (ail) is

nothing but Vandermonde's determinant and it becomes nonzero for
different ai's. Therefore the solution of the linear equation exists, and
hence it is possible to express [YI (XI) X YL-1 (X2)] Lm as a combination
of terms of YLm(vi). Thus our assertion has been proved.
Next we assume that the statement holds for k < K 1, and will -
show that it also holds for k K. To prove this, we note that a general
=
[YII (XI) X Y12 (X2) I LM With 11 + 12 2k + L 2K + L which satisfies

= =
the triangular relation Ill -121 :! L < 11 +12 takes the form [YK+l (xi) x
YK+L-1 (X2)] LM (1 0, L). What we have to show is that the term

=
...,
[YK+1(Xl) YK+L-I(X2)]LM is expressible as a combination of terms

X
of the form of Eq. (6.12). Now, looking at Eq. (6.7) in its full generality,
the expansion of V2KyLM(V) contains all of these (L + 1) functions,
K+I K+L-1
each multiplied by u u and some further terms. These have
2kt+li 2+12X 2kjX2k2
the form const. x u 1 U2 1 2 lyll (XI) X Y12 (X2)]LM7 where
at least one of k, and k2 is nonzero and 11 + 12 is equal to 2 (K -
k, -
k2) + L with k = K -
ki -
k2 < K -
Thus, by the assumption that

1.
statement of the theorem holds for k < K 1, all factors [YI, (xi) x
-
Y12(X2)ILM in these terms are expressible in the form of Eq. (6.12)

with appropriate v vectors. Equation (6.7) can now be rewritten as
L
K+1 K+L-1
V2KyLM(V) E u
1
U2 CKLI [YK+l (X1) X YK+L-1 (X2)ILM
1=0
+ ...' (6.14)
where CKLI is a suitable constant factor and the symbol indicates
the terms that axe already expressed in the form of Eq. (6.12) as
stated above. By taking (L + 1) different vectors vi in V2KyLM(V)
in exactly the same manner as before, Eq. (6.14) can be viewed as
a system of lineax equations for [YK+I(XI) X YK+L-1(X2))LM, which
is also solvable. The solution yields [YK+I(XI) X YK+L-I(X2)lLm as
combinations of terms of Eq. (6.12). This completes the proof. As
is clear from the proof, note that the vector v in Eq. (6.12) is not
uniquely determined.
It is easy to see that repeated application of Theorem 6.2 leads to

the following result.
Theorern 6.3. A vector-coup led product of solid spherical haT7non-

ics
[[[Yll-(Xl) X Yl,2(X2)IL12 X Y13 (X3)IL123 X X YIN (XAr)lrm

With (_l)L12 (-1)11+12, (_j)L123

= =
(_I)L12+13, ...,
and (_ 1) L
(_I)Ll2 N-I+IN can be expressed in
...
terms of a linear combination of

terms
(Xi Xj) kjj V2qyLM(V)

-
I
with 2 j=1
kii Ei> j=1 kij + 2q X:N
+2 i=
li =
1
-
L, where the vector v
of each term is given by EN 1 UiXi With

v = approp,riate. co j cient8
Uj's.
Since the factors xj2 and (xi xj)

-
are scalar, they play no active
roles in desci ibing the rotational motion. Theorems 6.1 and 6.3 estab-
lish the equivalence between the angular functions of Eqs. (6.3) and
(6.4) under the condition that the intermediate angular momenta a-Te
restricted as stated in Theorem 6.3 and that the parity oL the basis
L
function is given by (_ 1) through the angular momentum L. The ba-
sis function whose parity is given by (_l)L is called tohave a natural
parity-
The construction of a general angular function with unna ural par-
ity must be based on the vector-coupled form of Eq. (6.3). Unfortu-
nately there is no simple function analogous to Eq. (6.4) for the un-
natural parity case. One way to construct the angulax function with
unnatural parity is for L > 1
OLM (X) .
V2K [YL-1(V) X W]LM
with
v = iix and w Sij (xi x xj), (6.16)
where S is an skew-symmetric matrix which satisfies Sij

N x N
-Sji. For the special case of LI 0-, YL-I(v) must be replaced=
with YI(v). A slightly simpler angular function would be possible by

introducing another vector v' as follows:
OLM(X) = V2K [YL(V) X VILM) Vr = I?X- (6-17)

We note in Eq. (6.7) that in the case of K 0 both ki and k-2 are =
limited to zero and only the stretched coupling, namely 11 + 12 L, =
is allowed. See Exercise 6.1. With an increasing K value the possible

values of partial waves 11 and 12 increase including the case of non-
stretched coupling. This applies to the case of many variables as well.
6.3 Generating ftmction 87
To increase K is thus one way to include higher partial waves in the
calculation. The matrix A of the correlated Gaussian is often assumed

to be diagonal in order to reduce the number of nonlinear parameters.
In this case one has to increase K when the contribution of high
partial waves expected
is important. However, in the case where
to be
A is not diagonal, additional and important partial-wave contributions
come from the cross terms of the exponential part of the correlated
Gaussian even with K 0. E.g., the term exp(-Aijxi xj), when

= -
expanded into power series, contains many terms of the form (X,.X,)n,
which can describe high partial waves associated with the coordinates
xi and xj. This is easily understood by noting the following
relation
for arbitrary vectors a and b:
2k
(a .,r)n Bkja r2k Yj,,,(a)*Yj,,,(r)
2k+l=n M=-1
E Bkja 2kr 2k(_1)1,, F21-+l[Yl(a)xYl(r)loo, (6.18)

2k+l=n
which results from Eq. (6.45) and the addition theorem (6.54) for
spherical harmonics. This implies that even the basis function with
K 0 is expected to be useful if a general matrix A is used in the
=
variational calculation. The calculation for the dtl-L molecule given in

Complement 8.4 will clarify the point discussed in this paragraph.
6.3 Generating function
The calculation of the matrix elements becomes simpler if one uses a

function for the correlated Gaussian. In fact, the following
generating
function g, which contains an auxiliary "vector" A (8 1, ..., SV), is =
found to generate the correlated Gaussians of both type I and type 11

conveniently:
g(s; A, x) =
exp (-2 ;Mx 9x). + (6.19)
To relate the generating function g to the correlated Gaussian we
use the following formula
B kja2k+l r 2ky1M (,r) =

fYjm (ol) (a .
r )2k+lda
( 92k+l
=
f Y,.( a)
,
OA2k+l
e
Aa-r
A=O
dal (6.20)
which is easily proved by using Eq. (6.18). Then the vector-coupled

product (6.3) can be generated from the factor egx of the function g
as follows: By choosing si aiti with a unit vector ti we obtain
=
exp -I bAx)
2
YI,., (xi)
dii- Y ,,., (Fi) g(a It; A, x)

B01i Oai
(6.21)
By a symbol alt we mean a "vector" such that each component is
given by aiti, where & (a, aiv) with ai =
I - - -
7
a real number and
(ti, ..., tN) is a I x N one-row matrix of Cartesian vectors ti.

The key point in the above
equation is Eq. (6.20) which relates
the solid spherical harinonics to e ". Since Eq. (6.20) yields a more
general term r2kyl., (,r) than just the solid spherical harmonics, it may
also be useful to use a simpler relation, which generates just the solid
spherical haxmonics. The relation (6.58) serves for this purpose. By
choosing tj (1 T,j 2, i(I +,ri 2), -27-j) (j 1, N), we obtain
= =
...,
IV
exp
2 TbAx) Y1,m, (xi)
N
I 01i a Ii-Mi
H=1 Climi Oaili O-Fili-mi

x g(alt; A ,X) 1 t =(,-, 2,i(l+,
j j j
U=11 ---
2),-2-rj)
IN)
(6.22)
where
47r(I M) 1 -
Cim =
(-2)111 -
(6.23)
(21 + 1) (1 + m)!
Note that the vector ti satisfies ti-tj -2(-Fi-'F,j)2, particularly ti 2 =
0. This formula requires only differentiation in contrast to Eq. (6.21),

6.3 Generating function 89
where both differentiation and integration are needed. To couple the

solid spherical harmonics to the desired function in Eq. (6.3), one has
to multiply Eq. (6.21) or (6.22) by q, and sum over x.
To construct the correlated Gaussian-type geminals from g we note
exp 2RBR -
9R) g(s + BR; A + B, x)
1 1 ---
=
exp f -
-2
iAx -
2
(x -
R)B(x -
R) + (x -
R)j- (6.24)
The factor 0('-R) serves to generate the function O(x -
R) of Eq.
(6.5) or Eq. (6.6). It is easy to show that
exp f -
2
Jc- Ax
-2 (x
-
R)B(x -
R) I
IV
X
IIIX,_Ri12ki y1irni (X, -
Ri)
N
02ki+li
fj BkjIj f dtiYjj.j (ti) Oaj 2kj+1j
i=1
-1 kBR
x exp ( -
2
-
-J-tR)
x g(alt + BR; A + B, x) 1 (6.25)
aj.=O,...,CXJV=O
it, 1=1-, ItAr 1=1
...,
or
fn (A, B, R, x)
I I
expf -2 j Ax 2(x - -
R)B(x -
R)j xip
-
Rip )nip
i=1 P=j
N 3
anip
fj fj atipnip
i=1
exp( 2kBR iR)- -
P=I
x g(t + BR; A + B,
x)) 1 (6.26)
where n stands for the set of In,,, n12,n13,...,nNj,nN2,nN31- Tn the

Cartesian representation the parameter a plays no active role but
the x,y, and z components of each vector ti, (tillti2lti.), serve to
construct the function O(x -
R).
To construct the vector-coupled product OLM(X) of Eq. (6.3), one
has to sum over mi's with appropriate Clebsch-Gordan coefficients in
Eqs. (6.21) or (6.22). Apparently this is a very tedious task particularly
when the number of particles is large.
The choice Of OLM(X) of Eq. (6.4) leads us to the following very
simple equation which relates the correlated Gaussian to g. By choos-
ing t1 t2 tN Ae with a unit vector e, we obtain for
=
V = iiX
fKLM (u, A, x) =_ exp

(_2I FcAx v 2KYLM(v)
' ( d2K+L
=
BKL f YL M (' )
WA-2_K+
L
g(Aeu; A, x) dL
(6.27)
We can see from
Eqs. (6.21), (6.22), and (6.25)-(6.27) that the
correlated Gaussians are explicitly constructed from the generating
function g. Depending on the choice of the vector s, g leads to dif-

ferent forms of the correlated Gaussians, when followed by suitable
operations acting on s. These are surn-ma J ed in Table 6.2. The con-
struction of fKLM (u, A, x) is simplest among others and it has a wide
range of applications as will be shown in later chapters.

The correlated Gaussian of Eq. (6.27) contains only the relative
coordinates and the center-of-mass motion is dropped from the out-
set. Thus there is no problem arising from the coupling between the
intrinsic motion and the center-of-mass motion. If one uses the single-
particle coordinatesr instead of x in Eq. (6.27), the coupling between
them occurs in general and has to be taken caxe of appropriately in
order to calculate the energy of the intrinsic motion. A suitable choice
of A and u will, however, lead to the result that the center-of-mass mo-
tion can be
separated from the intrinsic motion. This will be discussed
in more detail in
Complement 6.4.
In Chap. 2 we discussed the linear transformation of the coordi-
nates x induced by tJae permutation P. It is important to Imow the
6.3 Generating function 91
Table 6.2. Relationship of the two types of the correlated Gaussians to

the generating function g of Eq. (6.19). The symbol alt indicates a set of
vectors f aiti amtN I. Bkj and Cl,,, are defined in Eqs. (6.9) and (6.23),
respectively.
Correlated Gaussians
exp -I.;v-
2 Ax) rIN i=1
Y1 irni (Xi)
(I rIN i= 1.
-
1
P01-i f
diiYiini(Z) jg(alt; A, x))
exp -!.,'cAx) IIN
2 i= I
Ylimi(Xi)
I ali ali-Mi
rIN i=1
climi aaili ajli-mi
x g(alt; A, x) I t =(I-, 2,i(l+,
j j j 2),-2-rj) i=O"ri=O
(i=l,.. N)
.,
exp ld Ax) IV12Ky

2
L IV J(V) (V =
EN i=1
Uixi)
d2K+L
f dgYLm( ,) ( ).X=O,e=je-j=3-
1
j,-X2K+L g(Aeu; A, x)
=
F3 K-L ,
Correlated Gaussian-type geminals
exp -UMx
2
-
.1
2 (x -
R)B (x -
R) I fT7 =,
Ixi _RZ12ki Y1 imi(Xi_ 14)
a2ki+li
(In,iv ,
-
I
Bkjjj
f dii-Yi iM
x exp 2
RBR -
JiR) g(a It + BR; A + B, x)

-j=O,jtjj=j
expf .1.7cAx I(x--RR)B(x R) I IIN JJ'=j(Xip

3
-
2
-
2
- -
,
i=- p
-
Rip )ni,
a7"P
=ffrff, lip-=, t-,77ri- jexp(-1:YZBR-!R)
i=
3
P
p 2
g(t+BR; A+B, x))

x
ti=o
92 6. Vahational trial functions
effect of P on the correlated Gaussians. Since the correlated Gaus-

sian is generated from the generating function, it suffices to exam-
ine the transformation property of g due to P. By using the relation
Px =
Tpx (see Eqs. (2.26) and (2.29)), we obtain a very simple result
Pg(s; A, x) =
g(Tp_.9; Tp-ATp, x). (6.28)
One only needs to change the matrix A and the vector s appropri-
ately. An important fact is that the generating function preserves its
functional form under P. This is also true for a more general linear
transformation T of the coordinates x, e.g., a transformation from one
set of coordinates to another. Combining this fact with Eq. (6.27),
we obtain a very useful property of the correlated Gaussian fKLM-

Namely for the transformation of Tx =
Tx, we have
TfKLM(u, A, x) 7--
fKLM(TU TAT, x). (6.29)
Thanks to this nice property one only needs to redefine the parameters
A and u of the basis function to construct the transformed wave func-
tion. This property plays an important role in evaluating the matrix
elements.
The generating function (6.19) plays a key role in generating the
correlated Gaussians and, moreover, facilitates the evaluation of the
matrix elements of physical operators. It is therefore desirable that
one can use the formulation based the
generating function In a
on
many-body system as well. In extending the results to laxger systems

of identical paxticles we need to cope with the symmetry adaptation
of the wave function. Tn such a case it would sometimes be useful if
the generating function were expressed in an "uncorrelated7 form of
the coordinates x, because the symmetry adaptation can then be shn-
plified by using the technique of Slater determinants or permanents.
In fact we can show that the generating function can be related
to the product of the Gaussian wave-packets centered around -4 =
(R,,..., RN) through an integral transformation. Using the definition

of Eq. (6.51) of Complement 6.1, we can express the product of the
Gaussian wave-packets as
N
det-V 4
1 1
_'i
ORj (Xi) ) exp
2
:iFx + k_Vx -
2 kFR)
(6-30)
with an N x N diagonal matrix
6.3 Generating ftmetion 93
-YJ 0 ...
0
0 72
(6.31)
0 'YN
A direct calculation using
exp
1:Mx + x) dx exp -9A-18 (6.32)
2 detA 2
proves the following equation which relates g of Eq. (6.19) to the
product of the Gaussians of Eq. (6.30):
g(s; A, x)
(detr)3 4
(47r-) N (det(rT expf 'g(r, Arisl

-
2
-
X
g(_V(.V-A)-1s;A(F-A)-'FR) 'Y'
ORi (xi) dR.
(6.33)
-1
Note that A(r -
A)-' r =
r(r -
A) r F is
symmetric matrix.
-
The function g in the right-hand side of the above equation depends

on the integration variable R and serves as just a weight which is
needed to convert the product of the Gaussian wave-packets to the

generating function. Equation (6.32) is proved in Exercise 6.2.
We have seen that the correlated Gaussians are all generated from
the same generating function as summa ized in Table 6.2. Further-
more, the latter can be obtained by the integral transformation of Eq.
(6.33) involving the product of single-particle Gaussian wave-packets.
From this we see that the calculations can be reduced to the matrix
elements of N-particle wave functions involving the product of the
Gaussian wave-packets. The width parameters -yi of the wave packets
can be chosen arbitrarily. To choose uniform width parameter for all of
them is most convenient if xi indicates the single-particle coordinate

of the identical Particles, because then the permutational symmetry
of the wave function is simply reduced to that of the "generator coor-
dinates" R. Even when x denotes the set of relative coordinates, this
94 6. Variational trial fimcdons
nice property can be used

by including the center-of-mass coordinate
as shown in[311. The
integral representation of the generating func-
tion has been successfully employed in accurate solutions of few-body
problems [311 as well as in microscopic descriptions of nuclear systems

in multicluster models [29, 30].
6.4 The spin fanction
X.L M
2
=
1
cosO -1 .1 (0),
22
.1;
2 -IM)
2
+ sin# .1 .1
22 (1),
.1;
2
.1
2 M). (6.34)
Alternatively, we may set up the spin function with a continuous

parameter such as that of Eq. (6.34) by an elementary method instead
of using the successive coupling. Its merit is that the construction of
the spin function is simple, that the evaluation of spin matrix elements
is easy, and moreover that one can use continuous parameters such as
V in Eq. (6.34), in the variational calculation.

The construction is done as follows. The Young diagram for the
spin function with spin S for the N-fermion system is [n+ n-I =
[(N12) +S (N12) -
S]. The maximum weight function with M =

S,
XSS, must have n+ spin-up functions and n- spin-down functions.
The number of terms distributing n+ spin-up functions among the N
particles is NS =
(,+). Therefore xss is expressed
n+
as a sum over these
terras:
6.4 The spin fimcdon 95
NS
XSS (A) =
E Ai ii) -
(6.35)
i=1
The coefficients characterizing the spin function,

ANs)
are not independent satisfy the condition,
of. each other but must
S+Xss(A) 0, where S+ is the spin-raising operator. Acting with
=
S+ on XSS (A) yields n+ + 1 spin-up functions and n- 1 spin-down -
N
functions in each term and thus leads to (n++,) independent terms.
Since the coefficient of each term must vanish, one has
I N (2S + 1)N!
V (n+ 1)
(nl+) +
1 =
(IN
2
+ S + 1)! (IN
2
-
S)!
(6-36)
independent parameters to specify the vector A completely. Here the

last minus one of Eq. (6.36) comes from the normalization of the
spin function. XSM is easily obtained from Xss(/\) with the use of
the spin-lowering operator and thus denoted as XSM(A). The over-
lap of two spin functions is independent of M and sh-nply given by
(Xsm(/\)IXSM(/\')) (XSS(A)IXSS(/X'))
= iV. The independent pa-
=
rameters needed to specify /X can be varied continuously in the varia-

tional calculation.
The isospin function is also constructed in exactly the same man-
ner. The action of the permutation P on the spin and isospin parts
simply produces a reordering of the terms in Eq. (6.35), leading to a
linear transformation of the vector /X to another vector denoted A(P).

96 Complements
Complements
6. 1 Nodeless harmonic-oscillator functions as a basis

In variational calculations for bound states it is
very important to
have a set of basis functions which can approximate square-integrable
functions to any desired accuracy. For a single-variable function one
may use the well-known complete, orthonormal eigenfunctions of, e.g.,
the harmonic-oscillator (HO) Hamiltonian. Or
one may try to use
nonorthogonal functions which may not be complete mathematically

but are able to cope with many problems flexibly and, from a practical
point of view, accurately. The purpose of this Complement is to ex-
amine the possibility of nodeless HO functions as a set of such "basis"

functions. We will see the performance of the nodeless HO functions
by studying how well
they approximate a given function f (r).
The nodeless HO functions with angular momentum Im have a
continuous parameter a:
14
-Palm (T) =
Nal
( V3)T
exp
2
I/r
2)y7n(,VF
1 I/r (6.37)
with a solid spherical harmonic (see the following Complement). where

the normalization constant Nj is given by
2
Nal
(2 1+2 a" 2
=
(6-38)
(21 + 1)!!
Here v is introduced to scale the
length. The Gaussians employed in
Complement 8.1 to solve the ground state of the hydrogen atom are
special cases with I 0. The overlap of two nodeless HO functions is
=
simply given by
2 2
Na I IVa'I aa1
(-Vaiml-Va-'im) =
_
.
(6.39)
+.,1)2 (a+a)2
2
2
We attempt to approximate a normalized function fl,,, (r) with

angular momentum ITn in terms of combinations of
Am (T) cirailm (T) (6.40)
The error of the approximation is estimated by calculating the vaxl-

ance
C6.1 Nodeless harmonic-oscillator functions as a basis 97
2
0- 2(f) =
fI fl (r)
,, cillil,,-, Or)Idr. (P.41)
One may choose a set of parameters jal, a2,..., aKI by some appro-
priate way or may set them up
by the trial and error procedure of the
SVM. Once they are selected, minimizing 0-2 leads to a linear equation
to determine the ci's
E(Failml-Vajim)cj =' (Faiiralfim) (i =

1, ..., K), (6.42)
j=1
K
andthena 2 is given by I -Y:ij=iCi*Cj(-Vai1mIrajIm)-
As a test function fim(r) we first take up the HO wave functions
*,,Im (r) angular frequency w hy/m, where n is the number

with the =
of radial nodes. We employ a phase convention such that the radial HO

wave functions are all positive for r greater than the outermost nodal
point. The overlap of the functions Fal (r) and is needed to

..
calculate o-2. It can be obtained by using a generating function of the

HO functions
1
A(k,,r)
4
exp
Ik 2
+ Nf2-vk-r -
VT
21
2 2
0,,jm(r)*P,,jm(k), (6.43)
n1m
where Pnjm(k),polynomial of the complex Bargmann space variable

a
k, is the Bargmann transform [76, 771 of the HO wave functions in

a spherical basis and its explicit form is given in terms of the solid
spherical harmonics of (4.5) [781
P,,lm(k) :(2 1)
Bn1
n + 1)!
(2n
n
-
(k k) ny1m (k),
-
(6-44)
where Bnj, defined in Eq. (6.9), is the coefficient needed to express xm

in terms of the Legendre polynomials Pl(x):
21+1
xm =
E 4r
BnIP1( (6.45)
2n+l=m
Expanding the overlap (F ,lm (r) I A(k, r)) in terms of the polynomials
(6.44), we obtain
98 Complements
(2n + 21 + 1)!! 1 2Va 2
(1+ )n (
-
a
(-PaimlOnlm) -
2n)+!!
(2n)!! (21 + 1)11.1 a 1 + a
(6.46)
Obviously, the approximation in Eq. (6.40) becomes less trivial as
the number of radial nodes of f increases. For Onlm (r) wit"h n 1 =
or 2, the combination of merely a few (::5: 3) terms leads to very good

approximations, whereas for n 5 it is hard to make u 2 less than
=
10-10 with a double precision calculation. This is so because all ai's

tend to take a value of unity, reflecting the fact that such a HO func-
tion can be expressed by combinations of higher order derivatives of
Falm(r) with respect to a at a 1, and the optimization procedure
=
tends to construct such derivatives out of -Tal,,, of almost equal pa-

rameters. One thus has to deal with an almost singular overlap matrix
((Faiiml-Vajim)) to obtain the solution of ci's.

In stead of approximating each of the 0,,,. (r) with different sets of
ai's, weminimized the average value, (U2 )N =

(j:IV
n=O 0r2(0nIM))/(N-.L k
1), with a single set of ai's. The values of the ai's were assumed 11-o
follow a geometric progression and its first term and the ratio of suc-
cessive terms were determined by Powell's method [141. A combination

of ten nodeless HO functions yielded (a 2)8 = 0.6 x 10-10 for I = 0.
The next test example is the shifted Gaussian defined by
I
( 3)4 (_IV(,r2+S2))jI(I/'Sr)y
V
(r) =
Fj -
exp 7
IM (6.47)
7F 2
where the function ii(x) is the modified spherical Bessel function of

the first kind
X2k+1
ij(X) =
F '7-x'I,+.! (X)
2
1: (2k)!! (2k + 21 + 1)!!'
k=O
(6.48)
and where the normalization constant Fj is
4e 2
F,j with 'us 2

(6.49)
2
To get the normalization constant, we have used the formula

00 2
I
fo x e-ax21, (bx) 1, (ex) dx =
2a
exp(b +c2),,'(bc)'
4a 2a
(6.50)
which is valid for Rea > 0, Rev > -

1. The function (6.47) is named
shifted Gaussian because it is closely related to the single-particle
Gaussian wave-packet centered around s
3
1
,ps' (r) -
exp, -
11 (6.51)
T 2
through the relation
4V'7-r
Os 1M (6.52)
Fj
1ra
Here use is made of the equation
eVr_5 = evrscos'o =
E(21 + 1)ij(vrs)Pj(cosO) (6.53)
1
with the angleO between r and s, and the well-known addition theo-
rem for the spherical harmonics
I
4w-
Pi (CosO) =
21+1
1: yjm(i )Yjm( )*
ra=-l
=
4-x-(-I)' lyl(p) X YIM100. (6.54)
/2_71-+I
The radial paxt of the shifted Gaussian is hence peaked at r 8 -
/_2, Iv.
It would therefore be challenging to approximate the shifted
Gaussian with large s in terms of the combinations of Note that
1 2
I/
V
0,v(r) expressible
is in terms of A of Eq. (6.43) as e-211 A (NF2'91 T).
The overlap between the functions F,,la(r) and 7p,jn(r) is
(r,im 1,0SW
v/-2-
N,, 1 Fs
1(1 + a)'+' 2
P(_ 1+a
ex (6.55)
Assuming again that the ai's follow a geometric progression, 0-2

was minimized by Powell's method for a given value of ;. For the case
2
of I =0 it is possible to make o- < 10`0 with ten nodeless HO
functions for the shifted Gaussians of up to :! 10. All ai then tend
to take values close to each other for large,;, and this requires a very
precise solution of the equation (6.42). Increasing the number

linear
2
of nodeless HO functions gives us even smaller o- values and widens
the range of maximum (; value in which the shifted Gaussian can be
approximated well.
The above examples strongly suggest that the nodeless HO func-
tions can approximate square-integrable functions to any desired ac-
curacy, though the number of nodeless HO functions needed depends

100 Complements
on the shape of the test function. A remarkable example given in [66]

shows that the linear function r itself, defined in finite range, can be
a
approximated to high accuracy in terms of a combination of Gaussians

e-ar2 )
We stress two remarkable points of the nodeless HO functions. One
is that a combination of a few nodeless HO functions can approximate
the shifted Gaussian even with a large value of ;. Forexample, it

2
is easy to make a less than 10-1-0 with a combination of only 15
nodeless HO functions. In an expansion in terms of HO functions,
one would need many more terms to obtain the same accuracy. This
indicates the flexibility of a nonorthogonal Gaussian "basis" compared
to the orthogonal basis such as the HO functions. Another point is that
there are many, possibly an infinite number, of sets f a,,..., aKI which
approximate a given function equally well, even though the number
K is fixed. This is what we have experienced in the above examples.
We display graphic illustrations of Gaussian expansions for dif-
ferent functions with I 0, m 0. The most appropriate nonlinear
= =
paxameters ai are determined by optimization, while the linear ones

ci are given by the solution of the least square equation (6.42). To
point out the dependence on the number of Gaussians, we use differ-
ent numbers of terms in the expansion (K 5, 10 and 20). =
In the first example we approximate the HO function 05oo(r) by

Gaussians. This function is smooth but oscillates (it has five nodes)
and asymptotically falls off like a Gaussian function. As shown in Fig.
6.1, K 10 and K
= 20 Gaussians give a perfect fit to the
= exact
function, so these curves are practically indistinguishable.
In the second case we to fit
an exponential function
attempt
f (r) e-', the wave function of the ground state of the hydrogen
=
atom. The asymptotics of this function is quite different from that

of a Gaussian. To approximate the asymptotic part of this function
one needs many terms of Gaussians as is illustrated in Fig. 6.2. By
increasing the number of Gaussians one has better and better agree-
ment in the asymptotics. After a certain distance, the Gaussians fall
off much more rapidly than the exponential function. In many practi-
cal applications, especially for bound states, however, one can always
use enough Gaussians to reach the required accuracy. We note that
the Gaussian fit gives a poor value (zero) for the derivative at the
origin (the exact value is -1) as will be discussed in Complement 8.1.
In the next example we try to approximate the absolute value
fimction f (r) 12.5-rl. The Gaussian expansion, again, does a pretty
=
C6.1 Nodeless harmonic-osefflator functions as a basis 101
10
-5
-10
0 2 4 6 8
r
Fig. 6.1. The harmonic-oscillator function (n 5, 1 0, m 0) and its

approximations by Gaussian expansions. The solid curve is the (UM3.ormal-
ized) harmonic-oscillator function f (r) V 500, and the dotted, dashed and
-
long-dashed curves are the K 5-, 10- and 20-term Gaussian expansions.
=
The 10- and 20-term. expansions are practically indistinguishable from the
exact curve.
100
20
IT
-C 10-40
--60
10
40
10
0 10 20 30 40 50
r
Fig. 6.2. The exponential function and its approximations by'Gaussian

expansions. The solid curve is the exponential function f (r) exp(-r), =
and the dotted, dashed and long-dashed curves are the K 5-, 10- and =
20-term Gaussian expansions.

102 Complements
good job (Fig. 6.3), and by the inclusion of further Gaussians one
would expect an even better fit.
1 V I I
0 .
0 1 2 3 4 5
r
Fig. 6.3. The absolute value function and its approximations by Gaussian
expansions. The solid line is the absolute value function f (r) 12.5 -,rl, =
and the dotted, dashed and long-dashed curves are the K 5-, 10- and =
20-term Gaussian expansions.
The approximation for the step function, f (r) 1 if 7- < 2.5, =
f (r) 2 if r > 2.5, is less impressive (Fig. 6.4), but it goes without
=
saying that it is not trivial to fit that function. One can improve the
fit by increasing the number of Gaussians, but the convergence might
be pretty slow.
The last example shows that a Gaussian expansion may be in-
2
12
adequate in some cases. Let us try to approximate f (,r) = r e-T
with Gaussians. This function is practically zero near the origin, and
it simulates the behavior of a wave function of a system -with very
strong repulsivecore. The Lenard-Jones or other hard-core potentials
may produce such a function. Figure 6.5 shows that the Gaussians
generally give a good fit to this function. By scrutinizing the inner

part of the approximation (see Fig. 6.6) one can see, however, that in
reality the Gaussians do not produce exact zero near the origin, but
C, -
0
0 1 2 3 4 5
Fig. 6.4. The step function and its approximations by Gaussian expansions.
The solid line is the step function f (r) = I if r < 2.5, f (r) 2 if r > 2.5,
and the dotted, dashed and long-dashed curves are the K 5-, 10- and
20-term. Gaussian expansions.
the approximate function oscillates around the exact one even for 20
Gaussians. This oscillation may be very unpleasent: To tame the hard
core potential, we need a wave function which is effectively zero near
the origin. The oscillating function leads to numerical problems, which

makes it rather difficult to obtain the solution of problems where the
interaction has a very strong hard-core part.
It is
important to know the
completeness of the basis functions in
2
L (square integrable functions), H1 and H2 (first and second Sobolev)
spaces because the bound state in quantum mechanics is traditionally

formulated in L 2, and the mathematical solution of the Schr8dinger
equation is formulated in H' and H2 spaces. (The space HI is a set of
functions whose derivatives of up to the mth order are all square inte-
grable. Note that the kinetic energy operator requires the second order
derivative of a basis function.)The convergence properties of the Ritz
variational method are demonstrated in H1 and H2 spaces, for exam-
ple in [791, and the completeness in L 2 is not sufficient to guarantee
the convergence of the Ritz method. The proof of the completeness
in the space of L 2 and in the spaces of H' and H2 is presented in
[801. This work proves that any function can be approximated to any
104 Complements
200
150
100
50
-50
-100 -j
0 1 2 3 4 5
r
Fig. 6.5. The fimction f (r) =,r 12e_,r2 and its

approximations by Gaussian
expansions. The solid curve is the ftmction and the dotted, dashed and long-
dashed curves are the K =
5-, 10- and 20-term Gaussian expansions. The
10 and 20 term expansions are practically indistinguishable from the exact
curve.
0.002 .
. .
. .
i .
. I I
0.001
0.000
-0.001
-0.002
0.0 0.1 0.2 0.3 0.4 0.5 0.6
r
Fig. 6.6. The in Fig. 6.5, but the inner paxt is

same as magnified. The
curves with the K 5- and 10-term Gaussian expan i n
= axe out of the
scope and not drawn here.

C6.2 Solid spherical harmonics 105
prescribed accuracy with a linear combination of Gaussians, provided

that the number of Gaussians in the expansion is sufficiently large and
the parameters of the Gaussians are appropriately chosen.
6.2 Solid spherical harmonics

The solid spherical harmonic Yl,(r) r1YI,(i6) of Eq. (4.5) is a =
solution, regulax at r 0, of the Laplace equation, V2f (,r)

= 0. ==
The irregulax solution is given by r-1-'YI (f). The solid spherical ..
harmonic is a homogeneous polynomial of degree I in the Caxtesian

coordinate:
21+1
+
rlyl.(,P) 47r
(1 + Tn)! (I -
m)!
p! q! p -
q)!
pq
X
(_X+i )p (X _iy)q
2
Y
2
ZI-p-q 7 (6.56)
where p and q are positive integers which satisfy the conditions

p + q < 1, p -
q = m. The complex conjugation leads to Yl,,(,P)*
(-I)MY1-09-
It is easy to show that with a special complex vector t =
(I -
7-21i(l + 7-2) -2-r) we obtain
(t-T)i =
(X + iY -
-F
2
(X _
iY) _
2TZ)I
I
C1 -F'-'YI,,, (r), (6.57)
M)!
where C1. is given Eq. (6-23).

For a given 1 the power of -F must
in
satisfy the condition 0 < 1 21, which guarantees the range of m

-
m <
as it should. Equation (6.57) leads to the equality
( al
eat-r =
CIMYIM(r)- (6-58)
aal a=O
This simple relation was used in Sect. 6.3 to generate the angular part
of variational trial ftmctions.
2
I a a
By expressing the Laplacian as V2 =
T2- Yr- (r 2 -a-, we see that
the solutions of the equation
[V2_(n-1)(n+1+1) I f(r)=0 r2
(6-59)
106 Complements
can be expressed in polar coordinates as r'Y,,,,(r) or

When n = 1 or n = -1 1, the equation reduces to
- the Laplace
equation.
The inverse relation of Eq. (6.56) is expressed as follows:
XPY qZn-p-q =,rn BPq Yl,,, (i6), (6.60)
npq
where 1 takes the values n, n -
2, ...,
0 or 1. A hint to calculate BIM
is given in Exercise 6.3.
jjIM1j2TtI2j3Ta3) j
(6.61)
IT =
U1 +j2) +j3 J12 +j3
=
il + U2 +j3) ii + J231 (6.62)
it is possible to choose a basis in which j2,
1 j2,
2 j2,
3
j212, j2, J, are diagg-
onal
ljlj2 (J12) h; jM) 7

(6.63)
2
1 j2,
j2, 2 j2,
3 123, J2i Jz
or a basis in which axe diagonal
jil j2h (J23); JM)

i
- (6.64)
The transformation from the one basis to the other is unitary and
it is expressed in terms of a U-coefficient as
ljlj2 (J12), h; JIVI) =

E U(jlj2 Jh; J12 J23) jil, j2h (J23); JM)
J23
C6.3 Angular momentum recoupling 107
(6-65)
The U-coefficient is often called a unitary Racah coefficient. It is given
by the overlap of the two basis states
U(jlj2 Jh; J12 J23) :::::

(il j2h (J23); JM ljlj2 V12) h; JM) (6-66)
i i 1
and is in fact independent of the magnetic quantum number M.
By using Clebsch-Gordan coefficients, the unitary Racah coefficient

can be expressed as
U(jlj2 Jh; J12 J23) (jITn1j2?n2jjI2MI2

MIM2M3MI2M23
X (JI.2Ml2j3M31JM)(j2M2j3M3lJ23M23)UIMIJ23M231jm)-
(6-67)
Note that in the above equation M is fixed to a certain value in the
range of -J < M < J, so that actually only two m values are in-
dependent in the summa i n. As the Racah coefficient is real and
unitary, the inverse transformation of Eq. (6.65) is simply given by
(U-1 Ut = =
CT)
jil j2h (J23); JM)
7 U(j1j2 Jh; J12 43) ljlj2 (42) h; JM) 1
J12
(6-68)
If the basis of Eq. (6.63) is to be transformed to a basis in which
the angular momenta j, and j3 axe first coupled to J13 and then
it is coupled with i 2 to the total angular momentum J, then the
transformation coefficient is given by
ljlj2 (42) h; JM)

1
=
(-l)jl+j2-J12 ji2il (J12) i h; JM)
=
(-l)jl+j2-J12 1: U(j2jI Jh; J12 J13) jj2 i j1h (J13); JM)
J13
=
E(-J)jl+J-J12-JI3 U (j2jl Jh; J12 J13) ljlj3 (J1-3) 7 j2; JM)
J13
(6-69)
The unitary Racah coefficients are transformation coefficients be-
tween complete sets of states, so that they obey orthonormality and
108 Complements
completeness relations. The orthonormality condition (jl,j2j3(J231);

JMjjI j2h (J23); JM) Jj ,3 j,,,,, reads, with the use of Eq. (6.68), as
7
E U(jlj2 Jh; J12 J23) U(jlj2 Jh; J12 J231) :_

JJ23 J23'- (6-70)
J12.
The completeness relation may be expressed as
E (jlj3 (J13) 7 j2; JM ljlj2 (J12) h; JM) i
J12
X (jlj2 V12) h; JIVII-ljlij2j3(j23); JM)

7
-
Ulh V13) j2; JMIj1 j2h (J23); JM)
i i (6.71)
This equation can be transcribed in terms of the Racah coefficients to
the following relation
E (-I)jl +J-J12 -JI3 U(j2jI jj3; J12J13) U(jlj2Jj3; J12J23)

J12
(-1)32+j3-J23 U(jlj3Jj2; J13J23)- (6.72)

There are symmetry relations with respect to the interchange
among the six angular momentum labels. These relations can be best
seen by introducing the so-called 6j symbol [70, 72, 751 written by
curly brackets
U(jlj2Jj3;j]-2J23)=(-I)jl +j2 +j+j3 Af(2JI2+1)(2J23+1)

il j2 J12
X
1 h J J23
(6.73)
because the symmetry relation of the 6j symbol is rather simple. E.g.,

one can show that
U(jlj2Jj3; J12J23) =
U(j2jlj3J; J12J23) =
U(Jj3jlj2; J12J23)
U(j3Jj2jl; J12J23) =
U(jlJj2j3; J2342)
(-l)j2+J-JI2-J23 rL2 (2j2

+ 1)(2J,
J1 223+1)
+ 1) (2J + 1)
X U(Jl2jlj3 J23; j2 J)
C6.3 Angular momentuin recoupfing 109
J,2+ 1)(2J23 + 1)
(_l)jl+j3-J12-J23 2
(2JI2,+
+ 1) (2j3 + 1)
(2j, +
X U(j2 J12 J23 J; j1h) - (6.74)

If one of the labels, ji, j2, j3, or J, is zero, the unitary Racah co-
efficient has the value +1. If J12 or J23 is zero, then we have
U(jlj2jlj2; J120) =
U(jljlj2j2; OJ12)
=
(-l)ll+j2-J12 p2
(2j,
2j,+
2J12 +1
+ 1) (2j2 + 1)
(6.75)
The above discussion is extended to the case involving four com-
muting angular momentum operators, jlj2,j3,j4. A basis with a

good total angular momentum J and its z component J, is con-
structed as, for example,
jj1j2(JI2)J3j4(J34);JM) or ljlj3(J13)ij2j4(J24);JM)- (6-76)

The transformation ftom. the basis in which J12 and J34 are good
quantum numbers to the basis in which J13 and J24 are good quantum
numbers is again real and unitary:
ljlj2 (J12) j3j4 (J34); JM)

i
31 32 J12
h j4 J34 ljlj3 (43) j2j4 (J24); JM)
1 -
J13 J24 J13 J24 J
(6-77)
The transformation coefficient is recoupling coefficient between four
a
angular momenta called a 9j symbol in unitary form. It is independent

of M and given by (jIj3(JI3)J2j4(j24); JMjjIj2(j12)J3j4(J34); JM)i
that is
il j2 J12
h j4 J3 4
J13 J24 J M I M 2 M3 M4 IVI 2 M3 4 MI 3 M2 4
(jlMlj2M2ljl2MI2)(j3M3j4M4lJ34M34)(Jl2Ml2J34M341JM)
(il7nlj3Tn3lJl3Ml3)(j2M2j4M4lJ24M24)(Jl3Ml3J24M241JM)-
(6-78)
110 Complements
Note again that M is fixed to a certain value in the range of -J <

M < J, so that only three m values axe independent iTi the above
summation. The inverse transformation is given by
ljlj3 (J13) j2j4(J24); JM)

i
A j2 J12
h j4 J34 ljlj2 V12) j3j4 (44); JM)
i
J12J34 J13 J24 J
(6-79)
The orthonormality and completeness relations for the unitary 9j
coefficients can be derived as before:
il j2 J12 il j2 J121
h j4 J34 h j4 J341 JJ12JI2"JJ34JS4"*
J13 J24 J13 J24 J J13 J2 4 J
A J2 J12 il j2 J1-2
(-l)j3+j4-J34 h j4 J3 4 j4 h J3 4
J12J34 J13 J24 J J14 J2 3 J
L J L j
A j3 J13
j3-j4-J23+J24
j4 j2 J2 4 (6.80)
J14 J23 J
There are many symmetry properties of the unitaxy 9j coefficient.

These will have their simplest form in terms of the familiaz 9j symbol
[70, 72, 751 written in curly bracket
il j2 J12
j3 j4 A4 -\/(2JI2+1)(2J34+1)(2Jl3+1)(2J24+1)
J13 J24 J
il j2 J12
X
1 h
J13
j4
J24
J3 4
J
(6-81)
Using the symmetry properties of the 9j symbol, we have for example
il j2 J12 J3 J13
h j4 A4 j2 j4 J24
J13 J24 J J12 J34 J
C6.3 Angular momentuin recoupling Ill
j2 J12 31
(2J12
j + 1) (2J34 + 1) (243 + 1)
j4 J3 4 j3
(231 + 1) (2h + 1) (2J + 1)
(2jj -
-
J24 J J13
J13 J24 i
=V,
(243 + 1) (2J24 + 1) (2J34 + 1)
3
il j2 J12
(2j3
(233-
(2.
3.3 + 1) (2j4 + 1) (2J + 1)
J34
j3 j4
33 34 J34
(-l)jl+j2+j3+j4+JI2+J34+JI3+J24+J il j2 J12
J13 J24 J
(6.82)
One may consider other basis states than those given in Eq. (6.76).
For example, the basis in which J12 and J34 are diagonal can be trans-
formed to a basis in which the angular momentum is successively cou-
pled as follows:
ljlj2 (42) j3j4 (J34); JM)

7
E U(J12j3 Jj4; J123 J34) jj1j2 (J12) 7 j3 (J123) j4; JM)

7 -
J1,23
(6-83)
The 9j coefficient can be
expressed in terms of products of unitary
Racah coefficients. The use of
Eqs. (6.69) and (6.68) enables one to
rewrite the state in Eq. (6.83), which is obtained by the successive
coupling, as follows:
jj1j2 (J12) h (J123); j4; JM)

i
(-J)jl+J123-JI2-JI3 U(j2jlJl23j3; J12J13)

J13
X jj1j3 (43), j2 (J123), j4; JM
(-j)j1+J123-J12-J13 U(j2jlJl23j3; J12J13)

J13 J24
X U(J13j2 Jj4; J123 J24) jj1j3 (43), j2j4 (J24); JM)
(6.84)
112 Complements
Substituting this result into Eq. (6.83) and comparing with Eq. (6.77)
leads to a useful identity
il j2 J12
h j4 J34
J13 J24 J
E(_j)jI+JI23-JI2-JI3 U(Jl2j3Jj4; J123J34)

J123
X U(j2jljl23j3; J12JI3) U(Jl3j2Jj4; J123J24)- (6.85)

As application of the above equation, we note that if one of
an
nine angular momenta is zero the 9j coefficient is expressible i-n terms

of Racah coefficients. E.g., we have
il- 0 31
h j4 A4 U(jlj3Jj4; J13J34)i
J13 j4 J
0
:
h
J13
j4
J24
J
J
6 (2j,
2j,
243 + 1
2
+ 1) (2j3 + 1)
U(J13jI Jj4; h J24) i
0 j2 32
h j4 J34 (-l)j4+J-J24-J34 U(j2j4Jj3; J24J34)- (6-86)
h J24 J
6.4 Separation of the center-of-mass motion from correlated Gaussians
The correlated Gaussian (6.27) in the global vector representation

is, among the several possibilities of the correlated Gaussians, most
easily constructed from the generating function. Thanks to this sim-
plicity it has wide range of applications. The relative coordinate
a
; =
(xi, ___j xiv-1) is used to represent the correlated Gaussian, so
that the center-of-mass motion is dropped from the beginning. The
aim of this Complement is to showthat, even when the correlated
Gaussian is expressed in terms of the single-particle coordinates r as
fKLM (u, A, r) =
exp,
(_2'fAr) V2Ky M(V) L
C6.4 Separation of the center-of-mass motion 113
V wri, (6-87)
it ispossible to separate the intrinsic motion, expressed in terms of

the relative coordinates, from the center-of-mass motion by requiring
some special conditions on the parameters A and u. Here A is an N x N
symmetric matrix and u is an N x I one-column matrix. Therefore, in

the correlated Gaussian basis the use of the relative coordinates is not
always absolutely necessary but the single-particle coordinates can be
used as well. We shall show the conditions in the following.
By using Eq. (2.4), the correlated Gaussian of Eq. (6.87) can be
expressed by the relative coordinates x and the center-of-mass coor-
dinate xN, which are again denoted as x:
fKLM (Ui Ai 7) =
fKLM (U A! X) i 7 (6.88)
with
A' = U---IAU-' and U, = U--IU. (6.89)

The matrix elements A!Nj =
A!jN (i =
1, ...'
N -
1) connect the relative

and center-of-mass coordinates and give rise to a center-of-mass de-
pendence of the function. The element A!Nlv is a parameter describing
the center-of-mass motion which has Gaussian form. Thus the sepa-
ration between the intrinsic motion and the center-of-mass motion is
made possible by requiring that
XNi = 0 (i =
I,-, N -
1) and XNjV =
c,
where c is an arbitrary positive constant. As (U-I)iN=l (i =

11.... N),
the above conditions are rewritten as follows:
N N
I:I:Ajk(U-I)ki=O and
j=1 k=1
N N
E 1: Aik = C- (6-90)
j=1 k=1
The center-of-mass contamination in the angular function V2K

YLm(v) can be removed by requiring that u1V =
0, that is,
IV
Euj = 0. (6.91)
i=1
114 Complements
The conditions (6.90) and (6.91) ensure that the correlated Gaus-
sian (6.87) is free from any contaminations due to the center-of-
mass motion. The intrinsic motion is described by the correlated
Gaussian of type (6.27) and the center-of-mass motion is given by
exp(-cx'N /2). The number of parameters contained in the function
(6.87) N(N + 1)/2 + N, whereas since there axe (N + 1) condi-
is
tions we (N 1)(N + 2)/2 free parameters. This number is of
have -
course equal to the one which the correlated Gaussian (6.27) has:
(N 1)!V/2 + (N 1) (N 1) (N + 2)/2.
-
-
= -
It is instructive to see how Eq. (6.90) leads to the separation of the

center-of-mass motion. For this purpose we note that the exponential
part of Eq. (6.87) can be rewritten as
I 1
exp (- 2 FAr) exp
-2
a j(,r, _,rj)2 2
(6.92)
where aij and,3i are related to the matrix A by
N
aij =
-Aij (i 7 j), gi =
1: Aki- (6-93)
k=1
The coupling between the intrinsic motion and the center-of-mass mo-
tion arises from the second term in the exponent. As ri is expressed

as
IV-1
7'i E (U-l)ijXj + XN7 (6.94)

j=1
we have
N N-I ( N
Ep,r? i
X2N +2 Pi(U-l)ij xj *
xN +
i=1 j=1 i=1
(6-95)
Here the symbol ... denotes the terms that arequadratic in the relative
coordinates and have dependence no on xN. Substituting 3i of Eq.
(6.93) and using the condition (6.90), we obtain that EiN=1,3i = c and
ENj= 1)3i ((,T-l)ij k= I EN

EN j= 1_ Aki(U-l)ij
= 0. =
C6.5 Three electrons with S =
1/2 115
6.5 Three electrons with S =

1/2
Let us construct the spin function with S =
1/2. The basic terms for
M =
1/2 are
11) =
I ITI), 12) =
1111), 13) =
I ITT), (6.96)
where the particle indices are assumed to be in increasing
order, e.g.
in the state 11) the first and second electrons are in a spin-up state
while the third electron is in a spin-down state. The requirement that
S+X(A) must vanish leads to the condition Of /XI + -X2 + A3 0. By tak- =
ing account of the normalization the spin function can be parametrized

by a single variable &(-,xl2 < 6 < 7r/2) as follows:
1
X.L i
22
(A)
2
sinO
(V/jcos,0 VilsinO) I TIT)
+ -
( F, coso 4 sin i ) I IT).

+ I (6.97)
The,O value is chosen such that,& = 0 corresponds to the spin function

with the intermediate Spin S12 = 0 whereas V =
7r/2 to the one with
S12 = I-
By acting with S- on Eq. (6.97), we obtain
2
X.1
2
_.L
2
(A) = _
:3sin,& 1111) (61 41sin,&) -

c Os'& -
+ (Vj cos,0 4 sin,&) 1111).

+ (6.98)
These spin functions can be used for the spin part of three-quaxk
baryon wave functions. See Chap. 9.
6.6 Four electrons in arbitrary spin arrangement

an
Arithmetic similar to that of Complement 6.5 gives us the following

result for the general spin function of the four-electron system. For
S = 0:
I
Xoo (A) in,& +
(2Icos,& -
inO)
(2 COO +
2 sinO) 2
cosO +
ril-isind)
116 Complements
+
(2
I
COO -
FWIi.6)I lilt) F3 sinO I JITT),

+ (6-99)
where the paxameter 6 satisfies -T/2 < V < -Ft/2 and is chosen such
that,O 0 corresponds to the spin function with S12
=
07 S123 1/2; = =
whereas V =
corresponds to the one with S12
-r,/2 1i S123 = =
1/2. In the
positronium molecule consisting of two electrons and two
positrons, the most important spin function is such that the spins of
the identical paxticles are coupled to zero, which corresponds toO 0.
See Sect. 8.3. For S = 1:
X11 (A) ezzsindsin o

2
+
3
sin,& cos o
F1!is
-
in,&
sin o)
+( ilc osv -
4's in,& cos p -
r!2is in ? sin
p)
1
-(61 cos'O +
4S inO cos W +
r!is in 6 sin
W) 7
(6-100)
where the parameters V and W satisfy --F,/2 < 0 < -F,/2 and -Ti-/2 <
W < 7r/2. The three independent spin functions with definite S12 and
S123 values correspond to the V and (p values as follows: for S12 0 =
and S123 1/2, =O 0; for S12 I and S123 1/2, 6=

T-/2 and =
W 0;
= and for S12 I and S123 3/2, 6 z/2 and =
o Ti-/2. =
6.7 Six electrons with S = 0

The spin functions may be constructed from those of a smaller group
of electrons through angular momentum couplings. As an example, let
us construct the
spin function of the six-electron system with S 0, =
Xoo (A), from two groups of three electrons. Clearly X00 (A) is given as
a combination of two terms with S123(= S456) =
1/2 or 3/2:
xoo (A) =
122 00)
cosO .1 .1; + sin V
1!!; 00).
22 (6.101)
As wasalready mentioned, the spin function with S123 =

1/2 has
two independent states with SI-2 0 or 1, and similaxly
= the fimc-
tion with S456 1/2 has=two independent states with S45 = 0 or 1.
C6.7 Six electrons with S = 0 117
Therefore the first term of the right-hand side of Eq. (6.101) can be
parametrized as follows:
I I
00) = COS
611 (S12 =
0)"17
2 (S45 =
0) '1;
2
00
sin 61 COS 62 1 (S12 =

0) 2 (S45 =
1) '1;
2
00
sinVI sin V2 COS V31 (S12 =

1) 2 (S45 =
0) *';
2
00
sin 01 sin #2 sin

631 (S12 =
1) "2 (S45
I
=
1) '1;
2 00) (6-102)
On the other hand, the spin function with S123 3/2 or S456 3/2 = =
has just one independent state and needs no angle parameters. Each
term of the above equation and the second term of the right-hand side
of Eq. (6.101) can be easily written down in the form (6.35), e.g.,
(S12 =
0) 211 (S45 =
0) '1;
2
00
2v,f2-
-
(I +
-1 ITIJIT) + + (6.103)
The four parameters 6, #1-, 02,,03 determine the relative weight

of the five independent spin states for the six-electron system with
S=0-
It is of course possible to start with other groupings such as two
groups of four electrons and two electrons or three groups of two elec-
trons.
118 Exercises
Exercises
6.1. Prove Eq. (6.7).

Solution. Let us consider the quantity
IT =
-
f YLm (&) (a V)2K+Ldb With U =

2 1 + -C2- (6-104)
Of course I is equal to BKLa 2K+L V2KyLM(V) from Eq. (6.18). To

calculate I in another way, we use the relation
(2K+L)!
(a.V)2K+L =
E p!q!
(a-xj)P(a-X2)q. (6.105)
p+q=2K+L
[YI, (a) x Y1, (xl)loo [Yi, (a) X Y12 (X2)] 00
11 11 0
12 12 0
A A 0
x [[YI, (a) x Y1,,(a)],\ x1YI1 (XI-) X Y12 (X2)]AIOO
2A+I
(211+1)(212 +,)a11+12C(l, 12; A)
X IYX(a) X 1Y1J-(XI) X Y12(X2)1,\100- (6.106)
The coupled form [Y,,_ (a) X Y12(a)],x, is reduced to a 11+12C(II 12; A)
Y by using Eqs. (6.10) and (6.11).
When using Eq. (6.106) in Eq. (6.104) and integrating over b., we
see that A is restricted to L, because otherwise the integral vanishes
due to the orthogonality of the spherical harmonics. We thus obtain
f YLM01)[YA(a) X 1YII(XI) X Y12(X2)],X]00d&

Exercises 119
JAL [YI XI) X Y12 (X2)1 LM - (6-107)
Combining these results leads to Eq. (6-7).

Consider Eq. (6.105) for the special case of K 0. Then p + q is =
equal to L. Since 11 and 12 have to satisfy the condition L < 11 + 12 "::

p + q, 11 + 12 must also be equal to
L. Equation (6.7) then simplifies
to the well-known formula
YLM(XI + X2) =
JX1 + X2 I'F'YLM( XI +X 2)
L
=
EDOL 010 L-I 1XI (Xl) X YL-1 (X2)] LM
1=0
=
E
1=0
1+41')
47r(2L + 1)!
4-)
7)
1)! (2L 21 + 1)!
(2 + 1)
(21 -
[Yl(Xl) X Yr,-I(X2)]LM-
(6-108)
6.2. Prove Eq. (6-32).

Solution. A positive-definite, symmetric matrix A can be diagonalized
by a suitable orthogonal matrix T, namely tAT becomes a diagonal
matrix D with its diagonal element Dii di being positive. With the
=
use of a transformation x --> y (x Ty), the integral.T of Eq. (6.32)

=
is evaluated as follows:
I exp
2
Dy + Ts Y) (detT) 3
dy (6-109)
3
where (deff) is the Jacoblan (functional determinant) det(ax/Oy).
Though detT is 1 in general, we may assume it to be +1 in order to
preserve the volume element, dx dy. =
Since D is diagonal, the integration can now be performed sepa-
rately in each component of y, which reads as
3
2
1
j exp
2
djyj2 + (ts)j -yj dy (27r)di
e 2di (6.110)
Therefore we obtain
3
N
2
(27r)
d- exp( 2di (6-111)
120 Exercises
N
It is easy to note that n,'=, di is equal to detD =
det(TAT) =
detA,
and E Jtsfi/dj
j= Z
can be expressed as E ,(D-1)jj(!fs)j ( fs)j
j=
- =
-Z- -
TsD-lts =
9T(TA 'T)ts
-
- 9A-1s. Using these results

Eq. in
(6.111) leads to the proof.

It is useful to generalize Eq. (6.32) for the case where all the vectors
xi and si are d-dimensional. Following the above derivation we get
I
1.,r-,Ax +
exp
2 x) dx detA ) exp
2 M-1s), (6.112)
where the scalar (inner) product (a b) for d-dimensional vectors d

-
=
(a,, a2, --., ad) and (bl, b21 bd) is to be understood as (a b)

....
- =
i:d
Some useful formulas related to thisintegral axe collected below.
By differentiating both sides of the above equation with respect to the
mth component of the vector si, (si)m, we obtain
(xi)m exp (-2I: Mx 9x) + dx
((A-'s)j)m ( detA ) exp (2 9A-1s) (6.113)
Further differentiation with respect to (sj),, leads us to
I
(xi),a (xj),, exp ( -
2
TcAx +
9x) dx
f(A-l)ijJm,, ((A-1S)iW(A-18Wnj
+
d
2
X
detA exp(19A-1s).
2
(6.114)
Setting m = n and summing over m leads to
I
f (xi xj) (
-
exp -
2
7cAx +
.
9x) dx
d(A-')jj ((A-1s)i (A-1s)j)
+ -
X
detA
exp
2 s). (6.115)
Exercises 121
Tn the case of d =
3, let us define bl,,,,, a tensor product, [a x =
(Im ln I rIL) a,,,b,, with a,,, /-4-x/3aYj"' (&,) (M 1, 0, -1)

= =
etc. for three-dimensional vectors a and b that is, a, -L: (a,; +iay), ,
= -
\/2
ao =
a, a-, --!-(a,,
v/2-
--
iay). Here n 0 or I or 2 and -n < [t < r,.
-
=
The scalar product is a special case of the tensor product, that is,
I
(a b)- =
-vf3-[a x bloo = whereas r, = I and
r, =
give a vector (outer) product and a second rank tensor. See also
2
Eq. (11.2). The use of Eq. (6.114) leads to the following result
1
[xi x xjl,,,,exp -
2
:IAx +
9x) dx
v/3- &,OJ ,O (A-1)ij + [(A-1s)i (A-1s)j]
x
X
detA
2exp 2 gA`s). (6-116)
6.3. Derive B,,,Pq in Eq. (6.60).

npq
Solution. The explicit formula for Bj'. is a bit lengthy, and here we
give a hint for its derivation. First we recall that
x=
V
r V:r(Yj--j(P)-Yjj(P)),
3
y=i
V
F -r(Yj-j(P)+Yjj(P)),
V
Z=
VL37'y 7F
10 (6.117)
Thus the power XpyqZn-p-q can be expressed as
P q
(7 )n (p) E (q) (-1?

r 2 p+q
XPY qzn-p-q = rn
3
2n- 2 iq E A I/
t1=0 V=O
"+" p+q-g-v n
X (yi, (i6)) (YI -1 (,P)) (y10 (,,)) -p-q. (6-118)
The power of the spherical harmonics can be expressed as
k (k,m)
(Y1jr (j )) =ED L
a YL km (f) (6.119)
L
(k,m) and it is
where D L can be derived by using Eq. -(6.10) successively
calculated ftom the equation
122 Exercises
(k,m)
DL = V/4-
7r E
LIL2 ... Lk-I
xll(C(Li-ll;Li)(Li-l(i-1),ralmlLiim)),
i=1
(6.120)
where LO = 0 and Lk =L. In the case ofTn 1, Li is uniquely

=
determined as Li =
i, but for rn 0 there are several possibilities for
=
the value of Li. Thus the product of the spherical harmonies I Eq.
(6.118) can be reduced to the following:
A+- p+q-A-v n-p-q

(Y11 (i6)) (Y, -1 (f)) (Y10(f))
(p+q-,p-t/,-I)
D(A+"")Dp+q-IL-il
11+V E D L(n-p-q,O)
L
X Y[t+v li+v(P) Yp+q-A-v -p-q+IL+v (f) YLO (i ) - (6.121)

The product of the three spherical harmonics is coupled to a single
spherical harmonic by using Eq. (6.10):
YIL+z, IL+v (i6) Yp+q-tt-71 -p-q+IL+zl (i ) YLO (i6)
E C(IL+v p+q-1-t-7,1; L) C(L'L; 1)

L11
X(A+v1-t+vp+q-tz-v -p-q+A+YIL'-p-q+2A+2z;)
x (L' -p-q+21j,+2v L 0 11 -p-q+2A+2v) Y,
(6.122)
The above formulas provide us with the ingredients needed to derive
B npq
I.
7. Matrix elements for spherical Gaussians
In this chapter we show that the generating function introduced in the

previous chapter can be used to advantage to derive the Hamiltonian
matrix elements of an N-particle system for both the correlated Gaus-
sians and the correlated Gaussian-type geminals. Matrix elements of
various physical operators between the generating functions will be
tabulated in a compact form. Examples of matrix elements for spher-
ical or primitive Gaussians are given in this chapter and more general
cases will be treated in the appendix. We can thus calculate
matrix
elements of essentially any interaction for a many-body system. We

also show that the matrix elements involving the correlated Gaussians
become particularly simple in two-dimensional problems. An exten-
sion to the matrix element for nonlocal potentials is also presented.
We discuss the kinetic energy operator for the relativistic kinematics
and calculate its matrix element for the correlated Gaussians as an
example.
7.1 Matrix elements of the generating function
As the correlated Gaussian basis function can be constructed from

the function g (6.19), it is convenient to derive the matrix
generating
element between the basis functions from that between the functions
g.This is true even though we use more than one set of coordinates
because the transformation of the coordinates simply leads to a change
of the parameters A and s, as shown in Eq. (6.28).
It is convenient to use g not only because thereby all types of
correlated Gaussians can be constructed in a simplified manner but
also because the matrix elements between functions g can be evaluated
with ease. Table 7.1 lists formulas for the matrix elements of some
basic operators 0

124 7. Matrix elements for spherical Gaussians
Table 7.1. Matrix elements, M (g(s; X, x) 101g(s; A, x)), o-Foperators

=
0 between generating functions g of Eq. (6.19). Here we take all vectors d-

N
dimensional. x is a short-hand notation for E,=1 wi xi. A vector product
(a x b) and a tensor product [a x b]2,, are defined for three-dimensional
vectors a and b. B A + X, v s + s' and y
= XB-1s AB--s. P is
= = -
a permutation operator and the matrix Tp is defined by Px Tpx. =
d
2
(2,X)
MO detB eXP(2!' B-1V)
X fvB-1-vMo
aiQx IdTr(B-'Q) +i B-1QB-1vjMO
(Q: a symmetric matrix)
( X X X) (iv-B-1-v x B-lv) Mo
[fV-X X CX121L [fvB-lv x (B-1Vj2ttMO
-ih&Mo
(7rj = -ih
axj
-iA7r h2
IdTr(AB-XA) -
VAy JMO
(A: a symmetric matrix)
L. -i E,, (B-')ij(s'i x sj),Mo

(rzL =
Ei(Xi X 7ri))
2
L2 (d -
I)WB-1-s -
( Ej(B-1)jjZz Sj) I.M0 X
(fVX X r) -ih (fvB-1-v &) Mo x
.4
J(iv-x -
r) Mi. =- (27riv-B-1w)-2
x exp f 1
2iv-B-lw (r -
bB -lV)2
I MO
AB-lw
J(i7vx -
r) (iv-x x r) -ihf (r X &) + ivB--Iw (r x

ibB-lv) IM,
P (g(s ; A, x) Ig(Fp-s; TpATp, x))
7.2 Correlated Gaussians 125
A4 =
(g(s; W, x) 10 Ig(s; A, x) , (7.1)
where A and X arepositive-definite, symmetric matrices. The
N x N
integration over a number N of vectors, x fxj,..., XNJ, can be

=
carried out by using Eqs. (6.112)-(6.116).

leaving this section, we note that the formulation with the
Before
generating function, though quite useful and powerful, is still severely

restricted in its application to a many-body system by the number of
paxticles because of the following properties of the method:
1. The number of nonlinear parameters contained in the matrix A
of one basis function is N(N + 1) /2. By increasing the number

of particles the optimization of these parameters becomes pro-
hibitively time consuming.

2. To calculate the matrix elements between the basis functions, one
needs to calculate the determinant and the inverse of the N x N

matrix B A+ A. The number of operations needed to calculate
=
the determinant and the inverse increases as N3. The symmetriza-

tionpostulate would require the
repetition of this calculation N!
times for the overlap and the kinetic energy and (N(N 1)12)N! -
times for the potential energy matrix elements, respectively.

3. As the system gets more complex, the number of basis functions
needed to represent the wave function increases considerably and
more matrix elements have to be calculated.
As was shown in Chap. 4, a number of random trials are probed in the

SVM to select a basis element. To calculate the energy for each trial,
the determinant and the inverse of the corresponding matrix B have
to be calculated many computational load mentioned in
times, so the
the second item would become very heavy even for systems containing
a relatively small number of particles. Fortunately, when one repeats
the calculation of detB and B-1 by changing only one element or a few
elements of B, as is often the case in the SVM selection procedure,
one can use the Sherman-Morrison formula [141 to reduce the load
involved in the calculations. This will be explained in Complement

7.1.
7.2 Correlated Gaussians
We assume that the Hamiltonian of an N-particle system takes the

form of Eq. (2.1). The set of coordinates :Z =
(xl,..., xN-I) stands
for the relative coordinates chosen to describe the basis function. As

a concrete example of correlated Gaussians we assume in this section
that the basis function is given by Eq. (6.27):
fKLM (u, A, x) exp

(-2I bAx IEX12K YLM(ijX).
(fKILIMI (U 7 X7 X-)IOIfKLM(U; A7 X))
I
*( d2K+L+2K'+L"
BKLBK"L'ff de--d f
YL m (e^-) YL, lw,, (, f)
dA2K+LdAj2KI'+L-'
X (g(A'eu; X, x) 10 Ig(Aeu; A, x)) X=O,,N,=O' (7.2)
e=jej=1
e"=jeIj=I
As the matrix element (g(Aeu'; X, x) 10 Ig(Aeu; A, x)) between the

generating functions is given as a function of A, A', e, and e' for various
operators by using Table 7.1, one only needs to perform the operations
prescribed in the above equation. As will be shown in the appendix, the
operations can be done very systematically for many cases of practical
interest for arbitrary L.
The correlated Gaussiajas with L 0 what we call spherical Gaus-
=
sians take the particularly simple form in the special

case, K 0. As =
they give a fairly good solution for a variety of few-body problems,

we list the explicit for-m of the matrix elements for this case in the
following. Because the parameter u of Eq. (6.27) is redundant in this

approximation, we denote v'r4-7rfooo (u, A, x) simply as
FA(x) =
exp, (-2 Jc'Ax). (7-3)
7.2 Correlated Gaussians 127
Since this function is equal to g(s =

0; A, x), a special case of the
in fact
generating function (6.19), the needed matrix elements are
given in Table 7.1.

The overlap of the spherical Gaussians is given by
(2T)N- I 2
(FA' IFA ) =
( det (A + A) ) . (7.4)
The matrix element of the kinetic energy is obtained through Eqs.

(2.10) and (2.11) by
2
P
(FAI -
Tm IFA)
2mi
3OTr (A(A + A) -'A!A) (FA, I FA). (7-5)

2
The matrix element of the two-body potential is given by
(FAIIV(ri-rj)IFA)=(FAIIFA)(
Ci
27j)3j V(r)
2 .1
e-2 C,,r2 dr (7.6)
where
I -
= w(ij) (A + W) -lw('j) (7.7)

Cii
is given through the column w('j) defined by Eq. (2.13). The

vector
matrix element of the total potential energy is easily obtained by sum-
ming Eq. (7.6) over i, j. As a by-product of Eq. (7.6), we show that
the matrix element of Iri rj I' can be easily obtained because then
-
the integral in Eq. (7.6) can be reduced to
r
a
e
- LCT2
2 dr 47r
f
0
r
a+2
e- 2 LCr2dr
a+3
2-x
(2) C 0
x 2 e -xdx
----
"+3
2
1
27r
(2 )C
(a+ 2
(7-8)
where F is the Gamma function. Thus we obtain

2 2
(FA'117'i -
rjl'IFA) =
(FA, IFA) (7.9)
T X Cij 2
where cjj is given by Eq. (7.7).

For a spherically symmetric potential the integral
W
I(k, c) =
fo V(r)r
k
e-
-1
!
CT2
dr (7.10)
can be calculated analytically for a certain class of potentials. If V(r)

is given in the form
2-br
V(r) = re-" (7.11)
then a closedexpression for I(k, c) becomes possible for an integer
n(n > -k) by using the following formula
00
10
ee-ar2-brdr
n+I
T
-
1 n ni
7F(- I)n 2 ,fa- Ek=O (n-k')! k! fk gn-k for a > 0
n! a=O
for
bn+l b>0
(7-12)
with
V b
fo go =
exp (4a) erfc ( 2V/a-
and for k > I
[k121
k! b k-2i
A E V (k
i=O
-
2i)! )
2
A =(-I) kH'k-I (7.13)
Here Hn(x) is the Hermite polynomial and erfc(x) 1 -
erf(x) fis
the complementary error function, where erf(x) is the error function:
erf (x) =
(2/V9T f e-t2dt. See Exercise 7.1 for the derivation of Eq.
(7.12). In case where an analytic evaluation of the integral is impossi-
ble, one has to rely on a numerical integration.
Examples of application of FA(x) to various quantum-mechanical
few-body problems are presented in [811. See also Complements 11.1
and 11.3.
7.3 Correlated Gaussians in two-dimensional systems 129
7.3 Correlated Gaussians in two-dimensional

systems
In this section we review the trial wave function and its matrix ele-
ments in the case of two-dimensional (2D) problems. A real application
of the results in this section will be made to the study on few-body
problems in quantum dots in Chap. 10.
In 2D problems the motion of the particles is constrained to the
xy-plane. The vector r has therefore two components r =

(x, y), and
allexpressions for relative coordinates and other vectors introduced
in the previous sections remain valid by setting z 0. The scalar =
products and integrations over spatial variables will include only the
xy components.
The main difference between the 2D and the 3D case is the or-
bital angular momentum. In 3D the basis function was taken as the

2
eigenfunction of the operators L ,
L ,:
rlyi-00- (7.14)
2 2
In 2D L = L so that one of these operators becomes redundant
Z ,
reflecting the fact that the system may be invariant under rotations
around the z-axis only. The corresponding orbital angular functions
take the form
r'e"mw =
(X W, (M >_ 0), (7-15)
that is, the states are characterized by the quantum number "m". This
function, apart from normalization, can be derived from its 3D coun-
terpaA by setting I m and z 0 (or 6
=
r/2 in polar coordinate
= =
system).
The matrix elements derived for the 3D case cannot be directly
used because they already include an integration over the z-component.
One can, however, derive the matrix elements in a very similax way by
using generating functions and exploiting the simplifications present
in the 2D The first step toward this is to construct a generating
case.
function. Similarly to the 3D case, we assume a trial function, by using

the global vector representation, in the form:
I
fm(u7 A7 x) =
(V1 + iV2)' exp (_2 -Je' A x , (7.16)
where x is the vector of relative coordinates in 2D and v, and V2 are
the x and y components of the vector

N-1
V Uixi. (7.17)
Note that the complex conjugate of fm (u, A, x) is an eigenfunction of

L, with eigenvalue -m-
To derive matrix elements, one can use a particularly simple gen-
erating function for f,,, (u, A, x):
dr"
f (u, A, x)
..
--
g(s, A, x) (7.18)
dtra
t=O
with
g(s, A, x) =
exp
1-2 - CAX + *1 + iV2) (7.19)
where the 2D vector sj (j =

I,-, N -
1) is defined by a complex
*
vector (tuj, ituj). Note that they satisfy si sj 0, si sj'3 - = - =
2t-eui Uill
The matrix elements of the generating function remain valid as listed
in Table 7.1 by substituting d 2 and assuming 2D vectors. By using
=
the matrix elements of the generating function all the necessary matrix
elements can be easily calculated by carrying out the differentiation
prescribed by Eq. (7.18). The matrix elements for some basic operators
are collected in Table 7.2.
Table 7.2. Matrix elements A4 ...... 0 If,,) for two-dimensional case.
B is equal to A+ X. A is defined in Eq. (2. 11). Q, R, p, p, -y, -y, c are the

same as defined in Eqs. (A.2), (A.9), and (A.18), but the factor 3 in R must
be changed to 2. The integral I is defined in Eq. (7.10).
0 Mram'
M,,,J with (27r)N-1

.. ., Mm detB 1(2p)m
IV P2
Ei=l 2mi -
T,,.,n -lh2(R
2
+ mQ)MmJmm,
P
=
IiA7r
L;, MMmJnm'
V(J ri -
ri 1) (27r)'V-'-c(7nI)2I5mM, J:M
detB n=O
(2#)m-'(-y-y')'
(m-n)!(nl)2 1(2n+1, c)
7.4 Correlated Gaussian-type geminals 131
7.4 Correlated Gaussian-type geminals
We assume that the Hamiltonian of the N-particle system now takes

the form of Eq. (2.2). We use the single-particle coordinates j =
(Irl, ... rN) in this section. The transformation matrix U defined in
Eq. (2.4) is now redundant or can be set equal to the unit matrix.
The basis function is chosen to be the correlated Gaussian-type gem-

inal of Eq. (6.26). The function is characterized by three parameters.
The matrix A describes the correlation among the particles, the di-
agonal matrix B determines the spread of the single-particle wave-
packets, and k (RI, RN) give the centers of the packets. This
=
...,
type of basis function is employed in molecular physics and nuclear

cluster models. The permutation P of particle indices transforms the
coordinates as defined by Eqs. (2.26) and (2.27). This transformation
changes the basis function to
Pf,,(A, B, R, r) 'n
(TpATp, TpBTp, TpR, r), (7.20)
where use is made of the relation (Tp)-' Tp-i

= = Tp- and Tp-n
stands for fnq,,, nq,-2 , nq13 , .... nqN, , nq,2, nq,3) for
1 2
p-I PI
1
P2
2
...
...
PIV
N ) (q, q2
...
...
N)
qN
Thus the permutation acting on the basis function just leads to the
simple transformation of the parameters, A, B7 R, and n and still

keeps the form of the basis function invariant. It is not nice, however,
that the value of n changes under the permutation.
The integral needed to evaluate the matrix element with the cor-
related Gaussian-type geminal is 3N dimensional and of multi-center
type. Such integrals are extremely hard to calculate and to handle.
See, e.g., [82, 831 for some developments made for those multi-center
integrals which appear in molecular physics. To avoid the complica-
tion, a few simplifications are introduced in practice. For example, only
one pair of particles is allowed to be correlated in each basis function.
This is equivalent to assuming that the only non-vanishing elements

of the matrix A are, Aii = Ajj =
aij and Aij =
Aji =
-aij, if the
correlation between the ith and jth particles is taken into account by
expf-aij(ri _rj)2 /21. Another simplification is to use only primitive
Gaussians, that is, the basis function with n 0. Primitive Gaussians =
yield solutions of good quality for few-electron systems. In this section

we calculate the matrix element in the primitive Gaussian basis
I I
fo (A, B, R, r) FAr R)B(T R)
=
exp
-2
-
2(r - -
=
exp
2 -hBR) g(BR; A + B, r), (7.21)
allowing the most general type of correlations. The formulation pre-

sented here will be found to be very simple and lead to compact results
for various matrix elements. The matrix elements with higher angular
momenta, when needed, can be obtained from the matrix elements
in thegenerating function by straightforward but tedious differentia-
tion. Analgorithm for a systematic evaluation of the differentiation is
presented in Appendix A.3.
The overlap matrix element is readily available from Table 7.1. It
reads as
MG =
(fo (A!, B', k, r) Ifo (A, B, R, r))
3
2
detC
exp 2'DC-1v 2RBR 2kB
- -
(7.22)
where
C=A+B+X+B', v = BR+ B'k. (7.23)

Note that formulas derived in this section are valid for a general sym-
metric matrix B or B' provided that A + B and A! + B' are positive-
definite.
The matrix element of the Idnetic energy is also available from
Table 7.1:
N
I?
P'
(fo (A! B'; k
7 I fo (A, B, R,
2mi
h2
=
2 JKI ((A + B)C-'(X + B)A) -
VAyjMG (7.24)
with
y =
(X + B)C-'BR -
(A + B)C-'B'k, (7.25)
where A is a diagonal matrix with Aii =
I/mi. The matrix element of
the center-of-mass Idnetic energy is calculated very easily as -well. As
Eq. (2-10) shows, it is calculated by the above formula with a shnple
7.4 Correlated Gaussian-type geminals 133
modification: The matrix A is now to be taken as Aij :--

1/?n12 ... N for
all i and j.
To calculate the matrix element of both one- and two-body poten-
tials we make use of the following relation
(fo (W, B/I k, r) I V (I iv- r -

S 1) 1 fo (A, B, R, r))
f V(x) (fo (A!, B1, k, r) IJ(,Cvr -

S -
x) I fo (A, B, R, r)) dx,
(7.26)
where iv- :--
(WIi ...
i WN) is an auxiliaxy parameter and S is a 3-
dimensional vector. Substituting the matrix element of the J-function
in Table 7.1 into Eq. (7.26) leads to
(fo(W, Bf, k,,r)IV(Ifvr -
Sl)lfo(A, B, R, r))
C
=MG 27r)
C
2
V(X) expf _
2
(X + S _
VC-IV)2ldX
17
0,0
FTCVSC 1CS
e_2
2f (I
0
V(x) x e-!i
-I CX2
(ecx -
e- csx)dx,
(7.27)
where
c-1 =,W-1w and s =

jW-1v -
S1 (7.28)
were introduced.
We show that both two-body matrix elements are
one-body and
calculated in a unified way with an appropriate choice of w. The matrix
element of the one-body potential, V(Iri SI), can be obtained by -
choosing wj Jij (j
=
N). If V(x) takes the form of Eq. (7.11),
=
then the above integral is obtained analytically for n (n > -1) As -
the most important application of this formulation, we give below the

matrix element for the Coulomb potential:
1
(fo (X, BI k' r) 1
I 1 fo (A, B, R,
Iri -
S1
MG
S
erf
( F?)' c
(7.29)
where c-' is now defined by (C-1)jj and s is given by I (C-lv)i -
S1 -
The two-body matrix element can also be obtained in exactly

the same manner by the use of
Eqs. (7.27) (7.28). and To derive
the matrix element of V(Iri -
rj SI) one
-
only needs to choose w
as Wk :--
Jik Jjk (k
- =
1, E.g., the matrix element of the
Coulomb potential, 1/ Iri -
rj 1, can be calculated from Eq. (7.29) With

the substitution of c-1 (C-1)jj + (C-l)jj (C-1)jj (C-1)jj
= - -
and
s =
J(C-'v)j -
(C-lv)jl. When V(x) takes the Gaussian form fac-
tor e-P' 2, the matrix elements for both the central and the spin-orbit
potentials can be expressed in particularly simple forms:
(fo(X,Bf,k,r)je-P(' '_-r)2 Ifo(A,B,R,r))

-9
2
c
c+2p ) expf -
c
CP
+ 2p
(iV-C-1V)2j. (7.30)
(fo (A!, Bf, k, r) I e-P(fVr) 2(for x p) I fo (A, B, R, r))

Ii
C 2 CP
=
-ih( W-lv X &)-MG ( 2p) eXpj (CVC-1V)2j.
c + c + 2p
(7-31)
It is clear that the above formulation applies to the calculation of even
N-body matrix elements as well.
7.5 Nonlocal potentials
In this section show

simple example of the extension of the calcu-
we a
lational technique to nonlocal potentials. The application of nonlocal

potentials will be presented for positronic atoms in Sect. 8.3 and for
the nucleus 12C iTj Complement 11.4. The potential is assumed to be
of separable form
V E ki (Ti -
rj)) (W. (ri -
Tj) (7.32)
a
with a Gaussian form factor
-21 ar
2.
O a Or) = e (7-33)
This form is quite useful in many practical cases, but generalization
to other cases is straightforward by using the same trick as for the
7.5 Nonlocal potentials 135
potential functions, that is by substituting the function o,, by a delta

function J(ri -
rj -
r).
The first step of the calculation of the matrix elements of the above
potential is to substitute the relative coordinates x by a new set of
Jacobi coordinates Y fY1; --- YN-11. The Jacobi coordinates y are
=
obtained from the original set of coordinates x in such a way that yj =
,ri -rj (see Sect. 2.4). The matrix of the corresponding transformation
y=T (k)X, (7.34)

whose Jacobian is det(Ox/Oy) =
1, is defined in Sect. 2.4. The kth
T (k) is defined in such a way that the first
permutation in deriving
Jacobi coordinate is equal to ri -
rj.
The next step is to show the matrix elements for the generating
function (6.19). The matrix element of one term of the potential reads
as
(g(s'; X, x) I o,, (ri -
rj)) (W,, (ri -
rj) Ig (s; A, x))

(g(ts; tXT, y) IW,,,(yj) (W,,(yjjg(ts; tAT, y)), (7.35)
where the integration has to be carried out first over yj in both sides
and then over the remaining coordinates, yj's. We use the same nota-
tion y to denote these remaining (N 2) variables, as well. To carry
-
out the integration implied in Eq. (7.35), we introduce the following

notations:
-
B: an (N -
2) x (N -
2) matrix obtained by suppressing the first

row and the first column of the matrix tAT.
-
b: an (N -
2)-dimensional vector I (TAT) 12, (TAT) I N-1

-
bi: the first diagonal element of the matrix (TAT), i.e., (TAT),,.
-
t: a set of (N -
2) vectors j(t8)2, ---I (tS)N-11-

-
tj: the first vector (Ts) 1.

The notations corresponding to the bra side are introduced in exactly
the same way and distinguished with a prime symbol from the quan-
tities of the ket side. By separating the yl-dependence explicitly from
the generating function g as
g(Ts; TAT, y)
g(t; B, y) exp (-2 bjy2 1

_
by. yj + tI. yj (7-36)
and by using the formula (6.32) one obtains

(w,, (yJ lg(i s; i AT, y))

i
2
(2T
d
exp
(2d t2 g (t -
d
bti; D, y), (7.37)
where
1
d=a+bl, D = B - bb. (7.38)
d
As this function is again a generating function, the integration over the

remaining variables y is just the overlap of the generating functions,
and from Table 7.1 one obtains the final result as
(9(8'; A!, X) kPa (Ti -
Ti)) ((Pa (Ti -
rj) Ig(s; A, x))

3
-
(27r)N 2
ddI det (D -+D

1) )
1 I
X exp
(2d Iti2 + 2d' t/ 2+ iTo (D + D) -1vO
2
1 (7.39)
with
I
vo =t- Ibt, + t' -
b't'l. (7.40)
d &
The above result becomes paxticularly simple for the correlated

Gaussian with L = 0 used in Sect. 7.2 and reads as
(FAd Wa (ri -
TA) (Wa (?'i -
Ii) IFA)
32
(27r)N
ddIdet(D + DI) ) (7.41)
To calculate the matrix element for nonlocal potentials of an arbitraxy

form factor, we note from Eq. (7.36) that
(J(y, -
r) lg(i s; i AT, y))
g(t -
br; B, y) exp (-2 blr2 +tl-r (7-42)
which leads to the general result
(g(s'; A!, x) 1,5(,ri -
rj -
r)) (J(ri -
rj -
r) Ig(s; A, x))
7.6 Semirelativistic kinetic energy 137
(2r)N-2 2
det (B + B)
Ib' rt2 + t/ rf+Iij (B + B') -1v

x exp
(_2Ibir2+tl-r -
2
1 1
.
(7.43)
with
v =t- br +t'- b'r". (7-44)

By introducing the short-hand notations
-1
ti -
b(B + B) (t + iV)
.Z W =
I
ti -
(B + B)-'(t + t)
b, -
6(B + B')-b -6(B + B")-lb'

C =
1 (7.45)
41(B + B)-lb V, - 61 (B + B') -Ib'
Eq. (7.43) can be expressed in terms of g in compact form:
(g(s'; X, x) I J(ri -
rj -
r')) (J(ri -
rj -
r) Ig(s; A, x))
3
2 N 2
(2 -1
exp -
(t + t) (B + B) (t + t)
det(B + i ') 2
X g(W; C' Z). (7.46)
7.6 Semirelativistic kinetic energy
This section shows the calculation of the matrix element of the semirel-
ativistic Idnetic energy. This will be needed in the application to sub-
nuclear systems among others. See Chap. 9.
We assumed the nonrelativistic Idnematics in Chap. 2. The single-
particle Idnetic energy T is then related to the momentum p by
T =
p 2/(2m). In the relativistic Idnematics it is replaced by T =
V4c__2p2+,rn2c4 mc2. The speed of light c is set to unity and the

-
term, -mc2, is dropped unless otherwise stated as it just shifts the

energy expectation value. The kinetic energy we consider takes the

form
IV
T.r =
I" rIzP12 +Ira? z
(7.47)
The evaluation of the matrix element of the operator (7.47) is never

trivial, because the separation of the center-of-mass motion requires
special care. The contribution of the center-of-mass motion is removed
for the nonrelativistic case in Sect. 2.1 by subtracting the center-of-
mass kinetic energy from the Hamiltonian. This prescription led us
to the expression (2.10) for the intrinsic kinetic energy, which, as it
should, does not contain any dependence on the total momentum -r, IV.
However, this method no longer applies to the present case and must
be replaced by a more general procedure.
We assert that the sought procedure is to evaluate the matrix
element in the center-of-mass system, that is, in the system of
IV
,7r,v :=
E pi 0. (7.48)
i=1
Note that, when applied to the nonrelativistic kinetic energy, this,

new procedure reduces to the previous prescription of subtracting
the center-of-mass kinetic energy from the total kinetic energy as ex-
pected. This is easily seen as follows: By substituting Eq. (2.9) into
N
Tnr =
i:N p2/ (2mi),
j= I i (1/2) ENJ
we obtain Tn,
j= E3=,Aij-7ri--rj
=
with Aij being defined by Eq. (2.11). (The center-of-mass kinetic en-
ergy was subtracted from the beginning in Eq. (2.10), so that the
suffices i and j go up to N 1 and Aij was defined for i, j :! , N 1. In
- -
the case where the center-of-mass kinetic energy is included, it is not

difficult to see that i and j take I to IV and Eq. (2.11) is still valid for
(i, j =
11 ....requirement (7.48) then restricts the sum over
N).) The
i and j up to N practice instead of up to N. Thus the matrix
-
I in
element of the operator T,,, evaluated in the center-of-mass system is
equal to that of the intrinsic kinetic energy of Eq. (2.10).
Now we will show a method of calculating the -matrix element of T,
in the center-of-mass system in two steps. The first step is to express
the operator in terms of the operators defined in the relative coordi-
nates. As Eq. (2.9) shows, the single-particle momentum operator pi
can be expressed in terms of the N momenta 7rj (j =

N). If we
use the Jacobi coordinate set x defined by the tran r-ma i -n matrix
7.6 Semirelativistic kinetic energy 139
(2.5), the momentum PN becomes just --7rN-1 thanks to the condition

(7.48). This peculiarity was already noted in Sect. 2.4. Note that none
of the other momenta takes such a simple form; they contain more
than one -7r's. Let us denote this particular Jacobi coordinate set x, the
corresponding matrix Qj, and the momentum 7rIv_1 defined in this co-
(N)
Ov) I and -7r N_
ordinate set as X(N), J respectively. Obviously we can
define (N -
1) other sets of Jacobi coordinates x(k) (k =

1, ...'
N -
1),
each of which corresponds to the relative coordinate set obtained by
1 2 N
acting with the cyclic permutation (1, 21 ..., IV) =
(2 3
...
...
1
k-times on the original pattern of the Jacobi set x(M. See also Sect.
2.4 and Fig. 2.2. Corresponding to each set X(k) a linear transforma-
tion analogous to Eq. (2.4) defines TT(".
i
It is easy to see that U(1)
i
can
N)
be obtained by rearranging the N column vectors of U(i according
to the permutation (1, 2, ...' N) and at the same time by the
cyclic
permutation of the masses mi,...' miv.. Likewise, U(2) can be obtained
froM U(1)
J
U(3)
1 7
from U(2)
i ,
and so on. As the transformation between
the single-particle and relative coordinates is
always given in the form
of Eq. (2.4) for any Jacobi set X(k), the corresponding transformation
of the single-particle and relative momenta is also given as in the form
of Eq. (2.9):
Pi E (U( i )) 7r,(k) i =
1, ...' N). (7.49)
3
ji
i=1
Using the special form of the matrix U(k) and the condition (7.48),
we obtain the following useful relation
(k)
Pk =
-7r.-i (k =
1, ..., N). (7.50)
That is, the kth single-particle momentum is equal to the negative of
the N 1-th relative momentum defined in the kth Jacobi coordinate
-
set. Therefore, with the help of Eq. (7.50) the semirelativistic kinetic
energy (7.47) can be effectively replaced in the center-of-mass system
by
N
T.r _(7rN-02
(')
)2 +M2
J2 Mi'
MI + (7-51)
Note that this operator has no cross terms such as -7ri--7rj. The price to
be paid is that one has to transform the coordinate systems conforming

W terms when
to 7rN-,. There is no difficulty in evaluating the cross one
can work with a definite set of the relative coordinates. That is the
reason why followed that route for the nonrelativistic kinetic energy
we
in Sect. 2.2. In the case of the sen-lirelativistic kinetic energy the cross
term has to be avoided as will be shown below.
The second step is to show how to evaluate the -matrix element of
the operator (7.51). One has to calculate the matrix elements term by
term. For brevity we show a method only for the simplest correlated
Gaussian FA(x) of Eq. (7.3). Let us assume that x denotes the last
set of the Jacobi coordinates X(N) =
jX1'...' x1v-11. To calculate the
matrix element of (7r(i) J2 +,rn?,

N- Z
we have to transform the coordi-
nate system conforming to the ith Jacobi set. This can be achieved by
using the relation
X = V(i)X(i) 1
(7.52)
where V(') is an (N -
1) x (N -
1) matrix obtained by omitting the
last row and coWmn of U(N) (U('))-l. By using this transformation

we have
(FA, (x) I (7r(') N-I )2 + Ta2i IF ,(X))
(detV(') )3 (FA,(i)(x('))IV(,7r(')
N- 1)2+Ta,21FA (,)(X(i))), (7-53)
where (detV('))' is a Jacobian corresponding to the change of inte-
gration variables and
A(') = 00AV('), A!(') = 00WO). (7.54)

The kinetic energy operator and the relative coordinates of the corre-
lated Gaussians are now expressed in the same coordinate set.

To cope with the square root, we go from the coordinate space to
the momentum space:
(FAf(,)(X(i))l /(7r(i) N-I )2+Ta3jIFAW (X(i)))
ff (FA,(i)jk')(k'j (-7r'(') 1)2+mj2jk)(kjFA(i))dkdk,

IV- (7.55)
where k Iki kiv-11 and (xlk)

. ....
(27t-)-2R-'V-1-)6i 0 is normalized
=
as (k" I k) J(k -k). The Fourier transform of the correlated Gaussian

7.6 Sen-Arelativistic kinetic energy 141
and the matrix element of the square root operator can be written in
a very simple form:
( kIFA) =
(xlk)* exp (-2 FcAx) dx
=
(detA) 2
FA- (k),,
(7.56)
(k'I -,)2
fixj(v' + mj2 I k) =
Vh 2 k,2v- j
I + ;j J(k -
k'). (7.57)
By substituting these expressions into Eq. (7.55), one can reduce the
integration over k as follows:
(FA, (x) V/ (w _1)2 + M2 FA (x))
=
(detV(') )3 (detA(')detX('))-32
x (FA,(j) -i (k) I Vr(7r,(i 1)2)2 + M3

mj2i IFAM (k))
=
(detV('))3(detA(')detA!('))--:32i f vh2q2 + Ta2i
x (FA,(j) -i (k) IJ(klv-l -
q) IFA(j) -, (k))dq. (7.58)

To calculate the matrix element of J(kN-I-q), werewriteit as J(iv-k-
q) with an auxiliary I x (N -
1) one-row matrix Cv =
(0, 0, 1) and
make use of the formula in Table 7. L Then we obtain
(FA, (x) I V (7 (Vi

r
1 )2 + M i2 I FA (x) )
(detV(') )3 (detA(') detA! (FA, (j) -1 (k) I FA(j) -1 (k))
x
27r if
2
e-2Iciq2
vh2q2 + Mi2 dq
3
IV-
(2-x)
(FA, I FA) f (ci, mi) (det(A + A)
f (Ci, 7ni), (7.59)
with
Ci
(f,7(i)A(A+A!)-'A'V(')) N-1 N-1,
(7.60)
where the function f is defined by

X je-2-Ixq2
2
Ax M) (2v) h2 -+M2 dq
X a2
j,)O I 2
47r
( )
27-1 0
e-!! xq \//-h2q2 + M2 q2 dq, (7-61)
and use is made of the fact that the overlap has the following property
with respect to the change of the Jacobi coordinate set:
(FAIIFA) -= (FAI(x)IFA(x))
=
(detV(') )3 (FA,(i) (x(')) IFA(i) (x(')))
i
(detV(i))3 (detA(')detA'())-i (FA,(j)-i- (k)IFA(i)-i (k)).
2
(7.62)
The overlap (FA, I FA) is explicitly given by Eq. (7.4).
143
Complements
7. 1 Sherman-MorTison fonnula
To show that the Sherman-Morrison formula [14] can be used to ad-
vantage in selecting a basis element from among many random trials,

let us use the correlated Gaussians of Eq. (4. 1) or Eq. (7.3) as the basis
functions. The matrix elements Aij or, equivalently, aij are nonlinear
parameters of the basis, and they are related to each other through Eq.
(2.25). As is shown in Sect. 7.2, the calculation of the matrix elements,
(FA, I FA) and (FA, I HI FA), requires the evaluation of the determinant
and inverse of the matrix B = A + X.
Let us assume thatoptimize the symmetric matrix
we attempt to
A of nonlinear paxameters by changing only one element, say Ak1 =
Alk, or aij (j > i), but by keeping all other parameters unchanged.
Instead of changing all of the nonlinear parameters randomly at once,
we change one particular element randomly and then proceed to other
elements step by step. The latter type of optimization of A is certainly

a very restricted way, but in this case the computer time required for
the evaluation of the matrix elements is tremendously reduced, as will
be seen below. And it is actually very useful in selecting a successful
candidate from among a number of random trials.
We see from Eq. (2.25) that changing only one element aij to
aij + A produces a change in A as follows:
A --+ A+ Aw(0w(ii), (7.63)

where the (N -
I)-dimensional column vector Oi) is defined in Eq.

(2.13). Note that OPOP is an (N -
1) x (N -
1) matrix, whereas
Oi) 0A is just a number. On the other hand, changing directly only

the k, I (and 1, k) element AN Alk of A as AN
=
AN + X (and ,
Alk --+ Alk + A) can be achieved by introducing (N I)-dimensiOnal -
unit vectors e(') as follows:
A
A A+ e(k) + e(l) eo(k) (7-64)
1 + Jkl
where e(') is defined by (e('))i =
Ji, (i 1). By letting
N -
ui and vi stand for (N -
I)-dhnensional vectors, both of the changes

given by Eqs. (7-63) and (7-64) are in general expressible as
P
A ---> A+ Aiuigi, (7.65)

144 Complements
where p is either I or 2.
As B is equal to A + X, the change of A as given by Eq. (7.65)
leads to the following modification of B,
P
B ---+ B+ Aiuigi. (7.66)
Therefore the calculation of the matrix element for the above change
of A results in the calculation of the inverse and determinant of the
special form of matrix, B + a, Aiuigi. When the modification is
given by just one term of the form AuO, the Sherm an-Morrison formula
can be used to obtain
(B + Aub)-' B-1 -
-AB-1u,6B-1,
1 + AbB-lu
(7.67)
and
det (B + Aub) (I + A,5B-1u)detB. (7-68)

See Exercise 7.2 for the derivation of the Sherman-Morrison formula.
The advantage of these formulas is appaxent: By knowing B-' and
detB one can easily calculate the right-hand side of the equations, and
the A dependence is given in a very simple form. To change A, therefore
there is no need for the evaluation of inverses and determinants of the
modified matrix B (which would require (N _
1)3 operations), but we
get the desired results by simple multiplication and division.

a
When the modification of B is given by Eq. (7.66) in fiL11 gener-

ality, then the inverse and determinant can be calculated by using
the Woodbury formula [141, which is the block-matrix version of the
Sherman-Morrison formula. If one wants to change a few of the aij's
or one-column (and one-row) elements of A at the same time, the
summa ion in Eq. (7.66) has to be further extended appropriately.

Exercises 145
Exercises
7. 1. Derive Eq. (7.12).

Solution.The case of a = 0 is simple. In the case of a > 0, we may get
the integral as follows:
dn
1 00rnCar2-brdr
0
=
(-I)n
dbn 1000e-ar2-brdr
dn 1 V b
(_I)n
dbn 2 I-exp ( ) erfc( \,Fa
a 4a 2 -
(7-69)
By putting x =
bl(2,Va-), the above equation becomes
00
fo rne-ar2-brdr
n
1 n+1 nt
( - fa- )
=
V,-(_I)n A (X) gn-k (X) , (7-70)
2
k=O
(n -
k)! k!
where
(k)
fk (X) =
eXp (_X2) (eXp (X2))
A (X) =
eXP(X2) (erfC(X))(k). (7.71)
It is easy to show that fk (x) defined above is equal to the one given
in Eq. (7.13). Remembering that the Hermite polynomial is given by
(n) (1)
(-I)n eX2 (exp(_X2)) _X21 V/,-X-,it is
-
.(X) = and (erfc(x)) = -2e

not difficult to check that gk(x) can be expressed as in Eq. (7.13).
7.2. Derive Eqs. (7.67) and (7.68).

Solution. Let X represent u,&. Then the inverse (B + AX)-1 may be
calculated as follows:
-1
(B + AX) =
(I + I\B-'-X) -'B-1
CO
=
j:(-A)n(B-IX)nB-1. (7-72)
n=O
Because of the special form of X, XB-'X reduces to cX, where the

constant factor c. is given by OB-'u. Therefore repeated use of this
relation leads to the follwoing result:
146 Exercises
(B -I.X)n B-1 =
B-'(XB-'XB-'X ......
B-'X)B-1
= d'-'B-'XB-1 (n > 1). (7.73)
By substituting this result into Eq. (7.72), we obtain
A
(B + AX)-' = B-1 -
B-'XB-17 (7.74)
I+Ac
which is nothing but Eq. (7-67).

To calculate the determ In ant of the matrix (B + AX) with its (i, j)
element given by Bij + Auivj, let us suppose that the determinant is a
function of A: P(A) det(B+AX).
= The function P(A) is a polynomial
of at most degree (N 1) and can be expanded
-
as follows:
P(A) =
ao + aj.X + - - -
+ aN-O:I F-1, (7-75)
where the coefficient ak is calculated by k!p(k) (0). The rule of dif-
ferentiating determinants leads us to the conclusion that ak 0 =
for k > 2 because of the special form of the matrix X. We have
ao =
P(O) = detB. The coefficient a, is obtained as follows:
B1, B12 B, N-1
N-1
a, Ui V1 V2 VN-1
Biv-, I BN-1 2 BN-1N-1
N-1 N-1 IV-1 N-1
E Y uivj,6ij E T ujvjdetB(B-')jj
i=1 j=1 i=1 j=1
(,DB-1u) detB. (7.76)

Here Aij is the (i, j) cofactor of the matrix B and use is made of the
relation Aij detB (B-1)jj. Thus P(A) is eq ial to (I+ADB-'u) detB,
=
which is what we want to derive.
7.3. Calculate the one-body density matrix for the generating function
g:
Pi W, r)
(g(sf; X, x) IJ(,ri -
xN -,r'f)) (,5(,ri -
xN -
r) lg(,s; A, x)).
(7.77)
Exercises 147
Solution. As Eq. (2.12) shows, ri -
xN can be expressed as EN-I

1=1
W
w, xj. The argument made for the nonlocal potential in Sect. 7.5
can, therefore, be applied to this case
in exactly the same way. The
only necessary change is to replace w('j) with 0). The density matrix
When the wave function
pi (r, r) takes the same form as Eq. (7.46).
has the proper symmetry for a system of identical particles, the one-
body density matrix does not depend on the suffix i.
7.4. Reproduce the nonrelativistic kinetic energy formula (7-5) by

using the formulation presented in Sect. 7.6.
Solution. By using Eq. (7.50), the matrix element of the nonrelativistic
kinetic energy operator in the center-of-mass system becomes
IV N (i)
2
(7rN-l)
(FA'(-X)JE IFA(X)) =
(FA, (x) 11: IF",(X))
2Tnj 2m,
3h2
(FAI IFA) (7.78)
2mici
where f (x, m) 3h2 / (2mx) for the nonrelativistic kinetic energy is

=
obtained by replacing Vlfh-2q2 + M2 in Eq. (7.61) with h2q 2/ (2m).

To perform the summation over i in Eq. (7.78), we need to know
the special matrix element M9 N-1:
k
N
-I
VM
kN-i
-
(U(N) )kI (TT(i)

i
IN-I
(7-79)
Let us recall that 0)

i
can be obtained by rearranging the N column
vectors of Uj(N) EE Uj and at the same by rearranging the masses

time
TnI , M2 , mv according to the operation with the

cyclic permutation
....
(1, 2, ..., N) i times. Rom this construction it is easy to see that U(')
i
(N) M
can be obtained from Uj' in exactly the same manner as U is
N)
constructed from Uj(' but by rearranging the row vectors instead of
the column vectors. Then we obtain (see Eq. (2.6))

Mi
M12---N
for 1: k i
UW (7-80)
I N-1 'i
-(I -
) for I = i.
M12---N
148 Exercises
Using this result in Eq. (7.79) and noting the relation E'V
I= I
-
( j(kv)
U
0 for k N enables us to obtain the desired matrix element as
VW
kIV-1
-
(U'(V))ki (UJ)ki (7.81)
The siimma i n over i in Eq. (7.78) can easily be done to obtain
IV
3h2
-
3eT (A(A + X) -'A!A) (7.82)
2mici 2
with the matrix A being by Eq. (2.11). This agrees with Eq.
defined
(7.5), which was obtained by explicitly subtracting the center-of-mass

kinetic energy from the total kinetic energy.
This exercise serves as an indirect evidence for the validity of the
formulation given in Sect. 7.6.
8. Small atoms and molecules
This chapter examples for the application of the method to

contains
atomic and molecular systems. The interaction between the charged
particles is the Coulomb force. The long-range character of this force
makes the solution of the few-body problems difficult, especially in the
case of scattering. We restrict our attention to bound states, where
many different methods have been elaborated in the past decades.

These calculations provide an excellent possibility to test the efficiency
of our method. Relativistic effects in atoms are withi-n the reach of ex-
perimental precision of today, which calls for very accurate theoretical
calculations.
8.1 Coulombic systems

The systems of charged particles can be characterized and classified
by the masses (MI jn2 , mN) and the charges (qj,q2,...,qN) of
....
the constituents. One distinguish systems formed by either equal

can
(unit) or unequal charges, and depending on the masses of the particles

one can classify the systems as adiabatic and nonadiabatic ones. ITI the
unit charge systems of more than two particles the constituents form
a molecule and the binding energy depends only on the mass ratio(s)
of the particles. Atoms are good examples for systems with unequal
charges. The distinction between the adiabatic and nonadiabatic cases
is dictated by the possibility of a simplified treatment in the former
case. In the adiabatic case the masses of a group of particles are con-
siderably heavier than those of the rest. The classical examples are the
H2 molecule and the H+ 2
molecular ion (see Complement 8.2). In these
cases the electrons move faster, while the protonic frame may rotate
and vibrate by moving considerably slower. This physical picture is

expressed in mathematical form as the Born-Oppenheimer approxi-
mation, where the electronic motion is first calculated by assiimin

150 8. Small atoms and molecules
form
k
CkA JeXP 2 i AkX) jVk 12K+L YL M (f,-I") X Slus
Vk UkiXi (8.1)
The operator A is introduced to impose proper symmetries on iden-

tical particles (antisymmetry for fermions and symmetry for bosons)
of the system. In most calculations the index K is assumed to be zero
or at most zero or one. In the case of the PS2 molecule of Sect. 8.3,
K is set equal to zero. The details of the calculations can be found in
[33, 84, 851 some of the system presented in this chapter. The vari-
for
ational parameters included in each basis function are the elements of
the -matrix Ak and the coefficients Uki, which define the global vector
vk. Since the Hamiltonian used in this chapter commutes with the
spin operator, the variational trial function can be chosen to 'have a

definite spin value S. The value of S influences the symmetry of the
orbital part of identical particles. Therefore, possible S values 'have
to be tested in general to obtain the ground state. For treating the
adiabatic system of small molecules in Sect. 8.4, we use combinations
of the generating function g itself as the variational trial function.
8.2 Coulombic three-body systems

8.2 Coulombic three-body systems 151
13-16 figures, that the solution of the Coulombic three-body prob-

lem is one of the most useful benchmark tests to compare different
methods. The accuracy pursued in Coulombic cases is not of purely
academic interest, but highly motivated by the high precision of the
experiments. For example, the fine-structure splitting of the 18 22p2pj
state of the Li atom has been measured with an accuracy of 20 ppm
(parts per million). To explain the
experimental numbers one has to
include relativistic and quantum electrodynamics (QED) corrections,
including terms of the second- and third-order in the fine-structure
constant, which require very high accuracy. In addition to the accu-

rate reproduction of the energies, another motivation is that in the
variational calculations, even though the energy is good, other phys-
ical observables might be less accurately determined. The increased
accuracy of the energy, as we will demonstrate, eventually will lead to
very accurate values of other observables as well.
The stability domain ofgeneral (m+M+m-)-type
a
A B C three-body
system with unit charges has been thoroughly explored [861, and the
"
requirement for stability can be phrased as an empirical rule: like
charges have to be borne by equal or nearly equal masses" [861. In ac-
cordance with this prediction, systems such as H- (pe-e-), H2+ (ppe-),

Ps-(e+e-e-), HD+(pde-), HT+(pte-), or tdg- are all bound, while
(ppe-) or (pe+e-) are most probably not. Here p, p, d, and t are pro-
ton, antiproton, deuteron (2 H) and triton (3H) respectively. In the lat-
,
ter case, the particles with opposite charges form an atom, which does
not bind the third particle. Some systems, e.g., the muonic molecules
such as (ttft-) or (tdft-), remain bound even for L =-;,k 0 orbital angular
momenta.
A second group of the Coulombic three-body problems is formed
by systems where not all the particles carry unit charges. The rep-
resentatives of this category are the helium atom (ae-e-) and the
helium-like ions, where a stands for the 'He nucleus. These systems
often form bound states with L =,4 0 as well. The antiprotonic helium
atom (ape-) has been observed [871 and studied in very bigh orbital
angular momentum states (L = 30 -
40) [88, 891.

We have challenged our method to calculate the energies of some
of these systems. In the calculations to be presented the number of

basis functions superposed is mostly limited to be modest because our
primary purpose is to demonstrate the overall performance of the cor-
related Gaussians and not to compete with well-established methods

sharpened for these systems. We will increase the basis dimension to
reach high accuracy only in cases where such calculations are con-
sidered to be important. The results of the calculations axe listed in

Table 8.1. Table 8.2 presents the parameters used in the calculations.
The results are compared with those of other (mainly Hylleraas-type
or correlated hyperspherical haxmonics basis) calculations. Our results
axe reasonable agreement with other calculations. In most cases a
in
basis size of K 200 was used in the SVM. The precision of the
=
results can be
improved by increasing the basis size as can be seen
on the selected examples of Table 8.1. In these cases we reach almost
the same precision as the other methods. The calculation extends to

nonzero orbital angulax momentum states as well, including an e,,c -
a,mple for the L 31 state of the antiprotonic helium atom or the

=
slightly bound 3pe states of the H- ion. These nonzero orbital angu-
lar momentum states as well as L = 0 states have been investigated
by using the
global representation. The recovery of the results
vector
of other calculations (which are based on several different represen-
tations of the orbital part of the wave function) convinces us of the
usefulness of the global vector representation. See Complement 8.4 for
a compaxative calculation of the tdl-t molecule with the global vector
representation.
To illustrate the convergence of the energy and the expectation
values of average separation distances, the results at different basis
sizes are tabulated in Table 8.3 for Ps-. Actually the limit of further
increasing the accuracy is the conventional precision of the computer
itself. One can notice that at the basis size of K = 100 the energy
is accurate up to six decimal digits, but only the first four figures of
the separation distances can be precisely determined. By increasing
the basis size, the virial coefficient falls below 10-'0, showing by the
high accuracy of the calculation that all the digits of the reference
calculation are recovered.
Quite a few very accurate methods have been developed to solve
the three-body Coulomb problem. It is very difficult to go beyond
their precision. This is especially true for the methods which have
been elaborated for a given system only, incorporating as much phys-
ical intuition as possible into the trial function or into the solution. In
contrast with these methods, we use the same trial function, which is
of Gaussian nature and therefore it is not tailored to Coulomb prob-
lems at all. Still, as the examples prove, one can get a sufficiently
good solution in a unified and automatic way without a priori built-
inknowledge about the systems to cope with. The real power of the
8.2 Coulombic three-body systems 153
Table 8.1. Energies of different Coulombic three-body systems in atomic

units. K is the basis dhnension. See Table 8.2 for the constants which are
used in the calculations denoted superscripts a, b and c.
System State K SVM Other method K Ref.
Ps- Ise 600 -0.262005070226 -0.2620050702328 1488 [691

COI-i- Ise 600 -0.527710163 -0.527751016523 850 [90]
H- 3pe 200 -0.1252865 -0.1252865 90 [911
UP Ise 200 -112.97300a -112.9730179a 500 [23]
a
ttA 1PO 200 -110.26210a -110.2621165 500 [231

b
UIL 1D' 200 -105.98292 -105.982930b 2250 [921
ttA IF' 200 -101.43131' -101.43' 200 [931
MIL IS' 200 -111.36444a -111.364511474a 1400 [231
b b
MIL IP, 200 -108.17923 -108.179385 2662 [20]
td[t 1D' 200 -103.40849a -103-408481a 1566 [20]
'He Ise 600 -2.9037243769 -2.903724376984 700 [941
He Ise 200 -2.9037242 -2.903724372437 100 [951
He 3se 200 -2.1752291 -2.175293782367 700 [681
He IP0 200 -2.1238423 -2.123843086498 700 [68]
He 3po 200 -2.1331635 -2.133164190779 700 [681
He 1-D' 200 -2.0556201 -2.055620732852 700 [68]
'He 3D' 700 -2.055338993068 -2.055338993337 700 [68]
He IF' 200 -2.03125504 -2.031255144382 700 [681
He 3F' 200 -2.03125506 -2.031255168403 700 [681
He IGe 200 -2.02000069058 -2.020000710898 700 [681
He 3Ge 200 -2.02000069062 -2.020000710925 700 [681
VHe+ L=31 300 -3.50760 -3.50763486 1728 [891
'Li+ Ise 300 -7.279913 -7.279913 [96]
Table 8.2. The constants used in the calculations. The masses are in units
of the electron mass. m,,=7294.2618241, mp=1836.1515. R. is the Rydberg
energy in eV.
Set a Set b Set c
Mt 5496.918 5496.92158 5496.918

Md 3670.481 3670.483014 3670.481
MI, 206.7686 206-768262 206.769

2R,, 27.2113961 27.2113961 27.2116
Table 8.3. Energy and different separation distances for the (e+e-e-)
Coulomble three-body system as a function of the basis dimension K. The
virial ratio 71 is defined by q 11 + (V)/(2(T))I. See Eq. (3.50). Atornic
=
units are used.
SVM SVM SVM Hylleraas

(K=100) (K=200) (K=600) [691
-E 0.26200465 0.2620050648 0.262005070226 0.2620050702328
5.489 5.48962 5.489633252 5.489633252
8.548 8.54856 8.548580655 8.548580655
1
(,r2+_) 2
6.958 6.95832 6.95837 6.95837
1
(7-2 9.652 9.65284 9.65291 9.65291
77 0.46 x 10-4 0.34 x 10-6 0.54 x 10-10 0.23 x 10-10
approach will be more highlighted in the following sections, where the

number of particles is more than three and the method still works al-
most as easily as before, while the other methods need tedious efforts.
8.3 Four or more particles
The expedition to larger Coulombic systems starts with the positro-

nium molecule (PS2). This exotic molecule consists of two electrons
and two positrons. The possibility that the PS2 molecule or in general
the electron-positron system consisting of p positrons and q electrons
form a bound system was originally suggested by Wheeler [97], and
this question has been extensively studied since then. The existence of
the positronium. negative ion Ps- (p 1, q 2) has experimentally
= =
been observed [981. The binding energy Of PS2 was first calculated by
Hylleraas and Ore [991. To date, it has not been observed yet due to the
dffficult experimental circum tances, and this fact has intensified the
theoretical interest in solving- this Coulombic four-body problem [65,
100, 69, 101-1041. Actually the positron-electron annibil i n limits
the lifetime of Ps2 to few nanoseconds.
I-n obtaining the solution for the
PS2 molecule, it is useful to note
that the Hamiltonian for PS2 is invariant with respect to the charge
permutation, that is, the exchange of positive and negative charges.
The trial function should therefore either remain unchanged or change
itssign under the charge permutation operation. The ground state of
PS2 turns out to be even under the charge permutation.
8.3 Four or more particles 155
The convergence of the energy Of PS2 against the increase of the

basis dimension is shown in Table 8.4. The fact that the best vari-
ational calculation [103] in the correlated Gaussian basis is already
surpassed at the basis size of 400 illustrates the power of the random
trials.
Table 8.4. The total energies (in a.u.) of the ground state and the bound
excited-state of the PS2 molecule in atomic units. K is the basis dimension.
Method PS2 (L =
0) PS2 (L =
1)
SVM (K =
100) -0.516000069 -0-334376975
SVM (K =
200) -0.516003119 -0.334405047
SVM (K =
400) -0.516003666 -0.334407561
SVM (K =
800) -0.516003778 -0.334408177
SVM (K =
1200) -0.5160037869 -0.334408234
SVM (K =
1600) -0-516003789058 -0-3344082658
CG [1031 -0-5160024
QMC [1041 -0.51601+-0.00001
We have examined the possible existence of bound excited-states of

the PS2 molecule [84, 85] by taldng all possible combinations of states
with L =
0, 11 2,3 orbital angular momenta and S 0, 1, 2 spins. By =
a bound excited state we mean such a state that cannot decay to any
dissociation channels. The results of the calculation were negative in

all but one case. In the case of L I (with negative parity) and
=
S =
0, the calculation predicts the existence of a second bound-state
of the PS2 molecule. This unique bound-state has been found to be
odd under the charge permutation operation. The convergence of the
excited-state energy is shown in Table 8.4. Figure 8.1 summa izes the
energy spectra of the bound states made up of two positrons and two
electrons together with the relevant thresholds.
One may ask the question of why the second bound-state cannot
decay to two Ps atoms in spite of the fact that it is located above the
threshold of Ps (1S) +Ps(IS). Since the total spin of the state is coupled
to zero, it can dissociate into two Ps (ground state) atoms provided
that they have equal spins and the relative orbital angular momentum
between them is L 1. (Recall that the Hamiltonian preserves spin,
=
orbital angular momentum and parity.) However, this is apparently

impossible because two Ps atoms of equal spins are bosons and their
relative motion can only have even L. Consequently, the PS2 molecule
0
Ps(2P)+e++e-
-0.1
-0.2 -
Ps-+e+ Ps(lS)+e++e-'
Cd -0.3 -
IP0
>1
0)
Ps(IS)+Ps(2P)
-0.4 -
-0.5
-Ps(IS)+PS(IS) Se
-0.6 PS2
-0.7
Fig. 8.1. The energy spectrum of electron and positron systems. Energies
are given in atomic units.
with L = I and negative paxity cannot decay into the ground states of
two Ps atoms, that is, the lowest threshold of Ps(IS)+Ps(IS). Since
the energy of this L 1 state is calculated to be E
=
-0-3344 a.u. (see
=
Table 8.4), which is lower than the next threshold(-0.3125 au.) of

Ps(IS) + Ps(2P), this state is stable against autodissociation into this
channel. The binding energy of this state is 0.5961 e-V from this second
threshold, about 40% more tightly bound than that of the ground state
Of PS2 whose binding energy is 0.4355 eV from the lowest threshold.
The expectation values of vaxious quantities for the PS2 molecule
are listed in Table 8.5.
In the
PS2 molecule we deal with antiparticles, so the electron-
positron pair can annihilate. The second bound-state may decay either
by annihilation or by an electric dipole transition to the ground state
[851. The most dominant annihilation is accompanied by the emission
of two photons with energy of about 0.5 MeV each. To have an esti-
mate for the decay width due to the annihilation we have substituted
Table 8.5. Properties of the ground and excited states of the PS2 molecule.
The positrons; are labelled I and 3 and the electrons are 2 and 4. Because
of charge permutation symmetry, some equalities hold, e.g. (r12) (r14) =
(r32) =
(r,34). Atomic units are used.
PS2 (L =
0) PS2 (L =
1)
(r12) 4.4871530 7.56881891

(r13) 6.0332070 8.8575844
(r212 29.112633 80.173836

2
r13 46.374735 96.085514
3
( r12) 253.04611 1041.3251
3
(r 13 ) 443.85244 1226.7955
(412) 2807.2718 15612.112

4
(r )
13
5202.0371 17939-574
(r 21)
12
0.36839693 0.24082648
(r.3' 0.22079007 0.147244820
-2)
r12 0.30310361 0.16081514
-2)
r13 0.073444303 0.032230158
(1'12'TI3) 23.187368 48.042757
('r12 'T14) 5.9252651 32.131079
(5(rl2)) 0.0221151 0.0112091
(5(rl3)) 0.0006259 0.00014591
(V2)
1
-0.258001894 -0.16720401
(V1 V2) * 0.1307732538 0.091656853
(VI'V3) -0.0035446132 -0-016109693
11 + (V) /(2(T)) 0.3 x 10-9 0.36 x 10-6
the probability density of an electron at the position of a positron,

(J(r12)) ,
into the formula [1021
rannihi = 47
(MC622 ) 2hc(TIjJ(rj-T2)jTf)
62 4
=
4ir-(-
hc) hcaOI(5(rI2))i (8.2)
where (,S(r12)), given equal to J0 (Tf 15(rl r2) ITf) with ao

in a.u., is -
being Roughly speaking, the Metime is inversely pro-

the Bohr radius.
portional to the probability. The Metime due to the annihilation is
estimated to be 0.44 ns. This is twice that of the ground state (0.22
ns).
dipole transition from the excited state emits one pho-
The electric
ton with energy of 4.94 eV. The decay width Idjp&,, for this transition
is calculated through the reduced transition probability B(EI) for tile

electric dipole operator D.:
16v
Tdipole ':--
9
(E) 3B(EI;
he
I- -->- 0+), (8-3)
where
I
B(EI; 1- --+ 0+) =

)7 I (Oind I DI,
,
_
M) 12 1 (8.4)
with
4
Djz qi 1ri -
X4 I Ylp (ri -
X4) -
(8-5)
IIIere X4 is the center-of-mass of the P-92 molecule and E is the exci-

tation energy (4.94 eV) of the second bound-state. We calculated the
B(El) value and obtained B(EI) = 0.87e2a 2.
0
The lifetime due to the
electric dipole transition has been found to be 2.1 ns. The branching
of the electricdipole transition is thus about 17 % of the total decay
rate. Therefore, both branches contribute to the decay of the excited
state of the PS2 molecule. Its lifetime is finally estimated to be about

0.37 The excitation energy of 4.94 eV found for PS2 is different by
ns.
0.16 eV from the corresponding excitation energy (5.10 eV) of a Ps

atom. This difference to be
large enough to detect its existence,
seems
e.g. in the photon absorption spectrum of the positronium gas.

Before discussing the spatial distribution of the PS2 molecule, let
us recall that the average distance (r+-) between the positron and
the electron is 3 a.u. in the ground state of the Ps atom, while it is

10 a.u. in its first excited state. The root-mean-square radius (rms),
(ri -
X'j )2), of the ground state of the PS2 molecule is cal-

culated to be 3.61 a.u. The rms radius of the second bound-state is
5.66 a.u., 1.5 times larger than that of the ground state. This is not
surprising if one assumes that the second bound-state is essentially

a system of a Ps atom in its ground state and a Ps atom in its first
(spatially extended) excited state. To check the validity of this as-
sumption, we have restricted the model space to include only this

type of configurations. This can be achieved by a special choice of
the uki parameters of Eq. (8-1). The energy converged to -0.323 a.u.,
that is, the Ps(1S) +Ps(2P) system with zero relative orbital angular
momentum forms a bound state with energy close to that of L = 1
state of the PS2 molecule, therefore this configuration is lik-ely to be
the dominant configuration in this molecule. There is a second config-

uration, the Ps- + e+ or Ps+ + e- with L = I relative orbital motion,
which intuitively may look important because two oppositely charged
particles attract each other, but it is barely bound (E -0.315 a.u). =
The average distances in Table 8.5 show that in the L 1 state =
the two atoms are well separated. In fact we can estimate the root-
mean-square distance d between the two atoms by
2
I'l +7'2 7'3 +7'4
d2
2 2 )
4 (2(r12 )
2
+ (,r213) -
2(rl2*rl4) (8-6)
The symmetry properties of the PS2 wave function are used to obtain
the second equality. Using the values of Table 8.5 yields d 6.93 a.u. =
for the L = I excited state and d = 4.82 a.u. for the L = 0ground
state. One cannot give a direct geometrical picture of the ground or
excited state because the variance Arij =
(r,2 (,rj)2 is laxge.
The correlation function defined by
C(r) =
(TfIJ(ri -
rj -
r) ITf) (8.7)
gives more detailed information on a system than just various average
distances. This quantity can be calculated for the correlated Gaussians
by using Eq. (A.30) or (A.136). For the ground-state wave function
Tf with L =
0, C(r) becomes a function of only r, which is called the
monopole density. For the excited-state wave function with L =
1,
C(r) monopole and quadrupole densities.
consists of the two terms of
Figure 8.2 displays the electron-electron and the electron-positron
correlation functions r 2C(r) for the ground-state of the PS2 molecule.
The peak position of the electron-electron correlation function is
shifted to a larger distance than that of the electron-positron correla-
tion function. The latter has much broader distribution and reaches
farther in distances compared to the corresponding function of a Ps
atom.
Figure 8.3 displays the electron-electron and electron-positron cor-
relation functions for the second bowid-state of the PS2 molecule. As

mentioned above, the correlation function for the L I state consists =
of the monopole and quadrupole densities and their shapes depend

on the magnetic quantum number M of the wave function. Of course
the M-dependence of the shapes is not independent of each other

0.020
0.015
Cd
0.010
0.005
0.000
0 2 4 6 8 10 12 14
r (a.u.)
Fig. 8.2. The correlation functions r 2C(r) for the ground state of the PS2
molecule. The solid curvedenotes the electron-electron correlation and the
dashed curve the electron-positron correlation. For the sake of comparison,
the electron-positron correlation function for a Ps atom is drawn by the
dotted curve.
but is related by the Clebsch-Gordan coefficient. See Eq. (A.136).

The quadrupole density is contributed from only the P wave of the
electron-positron relative motion, while the monopole density is con-
tributed by both S and P waves. Figure 8.3 plots the correlation func-
tions for both (a) M = 0 and (b) M = 1. As the correlation function
is axiaUy symmetric around the z axis and has a reflection symmetry
with respect to the xy plane, the correlation function sliced on the
xz plane is drawn as a function of x (x > 0), z (z > 0). The electron-
electron correlation function has a peak at the point corresponding

to the average distance of 7.57 a.u. The electron-positron correlation
function has two peaks reflecting the fact that the basic structure of
the second bound-state is a weakly coupled system of a Ps atom in
the L 0 state and another Ps atom in the L
= 1 spatially extended =
state. The peak located at a larger distance from the origin is due to
the P-wave component of the PS2 molecule.
The hydrogen and positronium molecules can be considered as
members of the same family as both are quantum-mechanical fermio-nic
four-body systems of two positively and two negatively charged identi-
cal particles. But they are at the opposite ends of the (M+M+m-m-)-
x (a.u.) x (a.u.) z (a.u.)

z (a.u.)
z (a.u.)
x (a.u.) 0
x (a.u.)
z (a.u.)
Fig. 8.3. The correlation fimctions rC(r) in atomic units, raultiplied by

one thousand, for the second bound-state of the PS2 molecule. The z com-
0 for (a) and M I for

ponent of the orbital angular momentum is M = =
(b). Drawn on the xz plane are the corresponding contour maps.

type Coulombic systems called biexcitons or excitonic molecules. The

biexciton (or biexciton molecule), a bound complex of two electrons
and two holes, is observed in a variety of semiconductors [105, 106].
See also Sect. 10.1. The biexciton molecule is characterized by the
mass ratio o- =
m/M. Apart from this connection, however, their
properties are radically different, e.g., H2 is an adiabatic but PS2 is
a highly nonadiabatic system. Moreover, while in the case of the H2
molecule many bound excited-states have been observed experimen-

tally and later studied theoretically, in the case of the PS2 molecule
only the ground state and the unique excited state discussed above
have so far been predicted theoretically. See also Complement 8.3 for
the stability of the biexciton molecule.
Figure 8.4 displays the dependence of the binding energy of the
biexciton molecule on the mass ratio a m/M. The changes of the
=
binding energies in the ground state (L 0) and the excited statues

=
with L 1 and negative parity is si-milar. Both the ground and excited
=
states become less bound by changing the mass ratio from H2 to PS21
though the binding of the excited state decreases to a somewhat lesser

extent. The energy of the transition from the excited state to the
ground state is also shown in this figure. This transition may take
place in an external field, for example.
By increasing the mass M of the positively charged paxticles tbo-
wa,rd infinity, one arrives at the energy of the C I H,, 2p-x state of the
H2 molecule. This state is formed by an excited H-atom and a ground-
state HI-atom. Consequently, a statement similar to the case of the PS2
molecule is valid for the biexcitonic molecule: The second bound-state
of the biexciton molecule is dominantly formed by an interacting pair
of a ground-state exciton and an L = 1 excited-state exciton.
The rule that the Pauli
principle forbids odd partial waves between
identical bosons also applies to the biexciton with L I and negative =
parity. The second bound-state of the biexciton molecule cannot decay

to two ground-state excitons. A somewhat similar situation exists in
the 3p, state of the H- ion as well, where its second bound state can-
not decay due to the parity conservation. By changing the mass ratio
in that (M+m-m-) system, however, this kind of state disappears

for o- I and the Ps- ion is known to have only one bound state.
=
Tables 8.6-8.8 show our results for various other Coulombic sys-
tem .
The investigation of the stability of positronic atoms has been nio-

tivated by the use of positrons as a tool for spectroscopy (positron
0.15
0.10
CU
0.05
0.00
0.0 0.2 0.4 0.6 0.8 to

M/M
Fig. 8.4. The binding energy of the biexciton molecule as a function of the
mass ratio o- =
m/M. The dotted curve is the binding energy of the ground
state, and the solid curve is that of the first excited state with L = I and
negative parity The dashed is the energy difference, multiplied
curve by one
third, between the first excited state and the ground state.
Table 8.6. Energies of different Coulombic four-body systems in atomic

units. K is the basis dimension.
PS2 Ise 800 -0-516003778 -0.516002 400 [102]

PS2 IP0 800 -0.334408112
Li Ise 600 -7.478058 -7.47806032 1589 [107]
Li 1PO 1000 -7.410151 -7.410156521 1715 [1071
Li 'D' 1000 -7-335520 -7.335523540 1673 [1071
'HPS Ise 1200 -0.7891964 -0.7891794 [1031
Table 8.7. Energies of different Coulombic five-body systems in atomic

units. K is the basis dimension. The Li- energy with K = oo is the ex-
trapolated one [1091, where E = -7.500577 is given by multiconfiguration
Hartree-Fock calculations with K = 2997.
Be Ise 500 -14.6673 -14.667355 1200 [1081

Li- Ise 600 -7-50012 -7-50076 00 [1091
(27r+, 3,-1-) Ise 200 -0-5493
Li + e+ Ise 1000 -7.53218
Table 8.8. Energies of different Coulombic six-body systems in atomic
units. K is the basis dimension.
System State K SVM
(37,-+,37r-) IS' 300 -0.820

Li + Ps Ise 600 -7.73855
Be+e+ Ise 1000 -14.692
annihilation spectroscopy) in condensed matter physics. An intrigu-

ing question is whether
or a chemically stable system containing a
not
positron or a positronium could be formed in the various targets. This

question can be answered
only by a sophisticated calculation or exper-
iment because the mechanism responsible for binding the positron to
the neutral atom is the polarization potential present in the atom+e+
system. The boundness of the hydrogen positride (positronium hy-
dride) HPs was predicted theoretically by Ore [99] in 1951 and it has
recently been created and observed in collisions between positrons and
methane [1101. The properties of HPs is discussed in [851. The use of
the SVM proved for the first time that the positronic lithium (Li+e-r
[1111 and the positronic beryllium (Be+e+) [1121 are stable. We see
from the tables that the positron separation energy of the positronic
lithbun is 0.054 a.u. Below the Li+e+ channel the Li++Ps channel is
open and the binding energy of the

positronic lithium is only 0.0022
a.u. against the dissociation into the Li++Ps channel. A calculation
has to be accurate at least to 10-3 a.u. to answer the stability of

the positronic litbium. Likewise, the positron separation energy of the
Be+e+ system is only 0.025 a.u. Due to the tiny binding energies of
typically about 0.01 a.u. one has to. be able to reach high accuracy
8.4 Small molecules 165
in these 6-particle systems. A naive picture of these systems is that

the positron orbiting around the neutral atom slightly polarizes the
negative electron cloud, and the positron is bound by the resulting
attraction.
To extend the method (as a fall N-body solution) for the investiga-
tion of the stability of much larger systems (e.g. Sodium plus positron)
is out ofquestion. One can, however, try to use a "frozen core approx-
imation?'. In this approximation the positively charged core is consid-
ered to be passive (its polarization is neglected) and the problem is
solved in model space where the single-paxticle orbitals are orthog-
a
onal to the core orbitals. One has to solve the modified Schr6dinger
equation of the form
(H + AP)Tf = ETI- with P Oi) (Oi (8-8)

iGoccupied
which produces wave functions that are orthogonal to the core orbitals
provided the positive constant A is large enough. The projection op-
erator P is an example for the nonlocal potentials discussed in Sect.
7.5. See also Complement 11.4. One can validate this approximation
by comparing it to the "exact" fuU N-body calculation for Li+e+.
This approximation turns out to be very accurate, reproducing the
first six digits of the result of the full calculation [1121. Assuming that
the accuracy holds for larger systems, one seems to find the stability
of positronic sodium (Na+e+) [1121.
8.4 Small molecules
As it was mentioned at the beginning of this chapter, the molecular

case demands a special treatment. The variational trial function is
assumed to take combinations of the form (6.19)
1
g(s; A, x) =
exp (_2 Mx +
9x), (8.9)
where Aij and s =

f8li S21 -7 SN-11 are parameters of the basis
functions. The "generator coordinates" s are chosen conforming to
x. Note that x is a set of relative coordinates. Our aim here is

to calculate the energy of the system which can be directly com-
pared to experiment without recourse to the adiabatic treatment like

the Born-Oppenheimer approximation. Each basis function includes
!V(IV 1) /2 + 3 (IV 1) parameters to be optimized. These parameters
- -
describe various correlations. The matrix A describes the electronic

correlations and motions, while the generator coordinate s makes the
wave function flexible and allows us to represent several "peaks" of
the density distribution when, for example in the hydrogenic limit,
the holes well separated and the electrons
are axe on "atomic orbits"
around them. By choosing s 0 the function
=
(8.9) 'has its maxhnum
at the origin and this limit is suitable around a- m/M 1, when

= =
the paxticles with nearly equal masses are moving equally fast. At the
hydrogenic limit, when the motion of the heavy particles are very slow
compared to the light ones, the density distribution has several peaks
axound the attractive centers, and to represent these configurations
we need to shift the maximum of the trial functions out of the origin
by choosings appropriately.
The usefulness of the generator coordinates in the basis function
(8.6) can be understood by the following example. Let us try to cal-
culate the energy of the IH+ 2 by using
this basis with and without
(that is by setting them to zero) the generator coordinates. The latter
form corresponds to the correlated Gaussians for L 0. The energy
=
of that system is -0.6026 a.u.. Without the generator coordinates a

basis of K = 300 Gaussians give -0.5999 a.u. for this molecule. The
inclusion of the generator coordinate immediately h-nproves the con-
vergence and one can get -0.6024 a.u. by using K 10 basis states =
only!
Table 8.9 shows examples of calculations for the molecules and the
ions consisting of protons and electrons..
Table 8.9. Ground state energies of small molecules in atomic units. K is

the basis dimension.
System K SVM Other method K Ref.
"OH+
2
50 -0.602634429 -0.602634214 160 [1131
0OH2 100 -1-17445 -1-174475714 1200 [1131
"OH+
3
100 -1.34351 -1.343835624 600 [1141
167
Complements
8.1 The cusp condition for the Coulomb potential
It is desirable that the trial function satisfies the proper asymptotic
behavior or the special boundary condition as demanded by a given
Hamiltonian. We discuss special boundary condition, the cusp con-
a
dition [115] known for the Coulomb potential, by using the hydrogen
atom as an example. The local energy for the hydrogen atom is given
by
h2 1 192 Tf 2 (9Tf h2 1 2
e2
-5r- )
Ej"'c -
+ + 1 IIf (8-10)
2m Tf -r-2 r .M r2 Tf r
where hl is the angular momentum operator. As was discussed in

Sect. 3.2, the local energy for the exact wave function turns out to be
a constant. The Coulomb potential in the local energy gives a singular
behavior at r 0. For the local energy to be a constant, this singular

=
behavior must be compensated for by the kinetic energy term. For

an S wave, where the wave function Tf has no angular dependence,
12Tf 0 and the constancy of El.,r requires that the second term in
=
the bracket in Eq. (8.10) cancel the singular behavior of the Coulomb
potential:
2
I
Tf 9r
aTf) r=o
-me
h2 ao
where ao is the Bohr radius. It is easy to see that an exponen-

exp(-r/a) for the radial part of the ground-state wave
tial form of
function leads to a constant local density if and only if a is cho-
sen to be ao. The constant of the local energy is then equal to
-0/(2ma2) =
_Me4/(2h2) = -El as expected, where E, is the hy-
drogen atom ionization energy without the proton recoil effect, that
is, the well-known energy of 13.6 eV.
For a general case angular dependence, it has to take
when Tf has
2
care I/r
of both the I/r singularities. Then Tf may be expressed
and
as a product of radial and angular parts: Tf r'R(r)Y(S?), where s =
is a positive constant and R(r) is assumed to be nonzero at r 0. =
Substituting Tf into Eq. (8.10), we obtain

h? 1 a2 R 2(s + 1) 1 M 8(8+1)
Eloc
2m ( R -5r-2 +
r R 9r
+
r2
2
h2 1 2
+ __1 Y -
_. (8-12)
2Tar2 Y r
168 Complements
As R does not vanish at r =

0, 1IR gives rise to no singularity at the
origin. The condition that the local energy is constant leads to the
following result:
I 1
( OR)
9r
R =o (s + 1)ao'
12y =
S(S + 1)y (8.13)
We know that the second equation is satisfied if and only if s is a
positive integer 1, and then the first equation determines the correct
behavior of R near the origin as R(r) oc exp(-r/(l + I)ao). Equations
(8.11) and known to be the cusp conditions.
(8.13) are
Let us attempt to solve the S-wave hydrogen atom variationally

with Gaussian basis functions, exp[-(a/2)(r/a0)2J' where a is a vari-
ational parameter. We use Gaussians as an example of such basis
functions that lead to a rather accurate solution but are poor in
satisfying the cusp condition. When a single basis function is used,
the optimal value of a is 16/(9z), giving the minimum energy of
-8/(37t-)EI -0.849EI. A combination of a few terms approximates
=
the ground-state energy quite well. The parameter values of a are

determined by the SVM. Table 8.10 shows sample results of such cal-
culations obtained with the code given in [81]. The calculation with
five Gaussians already reproduces the energy up to three digits. The
wave function obtained with ten Gaussians can reproduce both the
energy and the mean values of r and I/r fairly accurately. The over-
lap of the wave function with the exact wave function is very close to
unity. We may conclude that the Gaussian basis can predict physical
quantities to high accuracy. Of course, the solution does not satisfy the
proper asymptotic behavior at large distances and, moreover, always
gives zero for the cusp value of Eq. (8.11). The local energy displayed
in Fig. 8.5 for the variational wave function indicates that with in-
creasing K it tends to show smaller and smaller deviations from the

exact wave function except for the singular points mentioned above.
The local energy at large r deviates from the correct value because
the Gaussian basis has the wrong asymptotic behavior.
It is possible to generalize the above arguments for the cusp value
in a system of particles interacting via Coulomb potentials. Evaluating
the cusp value for a pair of particles with charges qj and qj, we obtain
(If 16(ri -
2'j) alriarj I Of) I-tijqiqj

(8-14)
(T/IJ(ri -
rj)ITI') h2
where tzij is the reduced mass of the two particles. The left-hand
side of Eq. (8.14) is expressed with the matrix elements involving
C8.2 The chemical bond: The H+
2
ion 169
J(r) =
J(r)/(47rr2) The cusp values for a pair of paxticles
.
are often
used to test the quality of the variational solution at the particles'
coalescence.
Table 8.10. Variational solution for the hydrogen atom with a number K
of Gaussian basis functions. The last row shows the exact values. Ei is the
hydrogen atom ionization energy and ao is the Bohr radius.
E
K
E,
((_L_)-2)
ao ao ao
((_E_)2)
ao
Overlap
1 -0.8488264 1.131774 0.8488284 1.499996 2.650706 0.9568351

3 -0.9939585 1.903352 0.9939409 1.491519 2.922759 0.9987560
5 -0.9996191 1.986219 0.9995692 1.499147 2.991284 0.9999446
10 -0.9999998 1.999700 0.9999958 1.500004 3.000014 0.9999999
-1 2 1 1.5 3 1
K= 1
0-, K=3
K=10
IN.
-4- 1 %
0 5 10 15 20
rlao
Fig. 8.5. The local energy curve plotted for the variational solution of the
hydrogen atom. K denotes the number of Gaussian basis functions.
8.2 The chemical bond. The H+

2
ion
Quantum mechanics enables us to understand the chemical bond,

which is responsible for the formation of molecules from isolated
atoms. The chemical bonding phenomenon involves the delocalization
of electrons in an atom to gain attraction from the other nuclei when
170 Complements
the atoms come close to each other. We take up the simplest possi-
ble molecule, the H2+ ion, to understand what an important role the
Hellmann-Feynman and virial theorems play for clarifying the origin
of the chemical bond. See [1] for detail.
ffilly quantuin-mechanical description of a molecule is a com-
The
plex problem. This problem is usually simplified by using the Born-
Oppenheimer approximation, where the electronic motion is separated
from the nuclear motion, considering the fact that the electron mass is
much smaller than that of the nuclei. One starts with determining the
motion of the electrons for a fixed configuration R of the nuclei and
obtains the ground state, of energy U(R), of the electronic system.
Then one assumes that,when R varies, the electronic system always
remains in the ground state corresponding to R, that is the electrons
follow adiabatically the motion of the nuclei. The chemical bond is
then determined by studying the nuclear motion in a potential energy
V(R) which comprises the Coulomb repulsion between the nuclei and
U(R)-
Let R be the distance vector between the two protons of the Ht
2
ion and v be the position vector of the electron with respect to the
center-of-mass of the protons. The electron motion is determined by

the Hamiltonian
p2 e2 e2
H,e( R) (8-15)
2[t IT -
Al
2 IT + RI
2
Note that R is just a parameter stage. The reduced mass jL

at this
is equal to 2Mm/(2M + m), where M is the mass of the proton and
m the mass of the electron. The electronic energy U(R) is the lowest
eigenvalue of the Hamiltonian (8.15). It is clear that U(R) becomes
a only. The Schr8dinger equation for the Hamiltonian
function of R
(8.15) completely separated in elliptical coordinates with respect
is
to the foci, R/2 and -R/2. We do not need its exact solution in
the following discussion, but note that it is well approximated by the
variational calculation using a trial function of the form
Tf -01" (Z' R) +'OL' (Z' R),

2
_
2
(8.16)
where 01, (Z, s) is the ls hydrogenic orbital of radius ao/Z centered

at s. The charge Z is a variational parameter and its optimal value
is deternUned to miniraize the energy for each R. The optimal value
of Z decreases froin Z 2 for R 0 to Z = 1 for R ---* co. At
= =
R --+ cc the system will switch over to a configuration of the hydrogen

C8.3 Stability of hydrogen-like molecules 171
atom and the proton. Between these two extremes, Z is a decreasing

function of R. The energy U(R) is -4E, (2M/(2M + m)) for R = 0
and approaches -E, (MI (M + m)) for R oo.
By using the virial theorem (3-49) and Eq. (3.56) with A

R, Ry7 R, we can show that the expectation values, (T) and (W), of
the kinetic energy and potential energy of H,..(R) satisfy the relation
d
2(T) + (W) +R U(R) = 0. (8.17)
dR
'9
Here we have used the fact that WA+R. aR W = -W and Rr-aR -U(R)
R d U(R). equation, together with (T) + (W)
This U(R), enables us =
dR
to express (T) and (W in terms of the potential between the protons,
V(R) =
U(R) + (e 2IR), as follows:
d d
(T) =
-U(R) -
R U(R) =
-V(R) -
R V(R),
dR dR
d d e2
(W) =
2U(R) +R U(R) =
2V(R) +R V(R) -
-(8.18)
dR dR R
For the chemical bond to occur in the H+2 system, the potential V(R)
must have a minimum V(R
deeper than --+ oc) --Er at some point
Ro, that is, V(Ro) =
V(Ro:) < -E-r have to be met. Since
0 and
(T) ,
Er and (W) -2E, at R
- oc (see Eq. (3.53)), we can con-
-- -
clude from Eq. (8.18) that, at equilibrium (R RO), the electronic =
kinetic energy is increased and the electronic potential energy is de-

creased. The lowering of the electronic potential energy is large enough
to cancel the repulsion between the protons, and that is responsible
for the chemical bond. According to an exact calculation, Ro 2.00ao
-
and V(Ro) -
V(R -+ oo) = -2.79 eV.

challenging to perform a nonadiabatic calculation in which no
It is
separation of the electron and nuclear motion is made. The validity of
the Born-Oppenheimer approximation can be tested in such a calcu-
lation. Furthermore, the development of the nonadiabatic treatment
for a smaU molecule is of importance in its own right because the
adiabaticity may be questionable when the electron is replaced with
heavier particles like the muon or the pion. The excellent results ob-
tained in Sects. 8.2-8.4 indicate that realistic nonadiabatic, calculations
are in fact possible in the correlated Gaussian basis.
172 Complements
8.3 Stability of hydrogen-like molecules
The existence of bound states of systems composed of particles with

unit charge attracts considerable attention. We discuss this problem
here by applying some of the principles discussed in Chap. 3.
The system of the hydrogen-like atom, (M+m-), is always bound
and its binding energy is equal to jL/2 in units of e = I and h =
1, where p =
Mm/(M+m) is the reduced -mass. What about the
stability of ahydrogen-like molecule (M+M+m-m-)? This system is
characterized by the mass ratio o- m/M. Two well-known examples
=
include the hydrogen molecule (a < 1) and the positronium molecule

PS2 (o- 1). Another example is the biexciton molecule [105, 1061. See
=
Sect. 8.3. The value of o- of the biexcitons can vary between the two
Emits, 0 < c- < I. A molecule is bound provided that the threshold
of any dissociation channel is higher than the lowest energy of the
system. The lowest dissociation channel is (M+m-) + (M+m-) for
this system, and its threshold energy is Eth =
-ft. The stability of
the hydrogen-like molecule has been studied numerically in Sect. 8.3
(see Fig. 8.4). However, theoretical argument [1011 makes it possible

a
to prove that the system is bound for arbitrary values of o-, that is,
the ground-state energy E of the system is lower than Eth. The proof
relies on the scaling property of the Coulombic Hamiltonian and the
stability Of PS2. The basic point of the proof given in [101] will be
shown below.
The Hamiltonian of the system (M+M+m-m-) is written as
-ff =
-ffS + HA, (8-19)
1
_as =
(P21 +P22+P32 +P42)
4[t
1 1 I 1 I I
+ e
2
( 7'12
+
r34
-
r13
-
r14
-
r23
-
T'24
), (8.20)
I 1
HA
4-M
-
4m ) ( 2+Pi P22_P2_
3 P44:)
2
-
(8.21)
HS is nothing but the Hamiltonian of a system (X+X+X-X-) with

the mass mX of
particle X being equal to 2p. By applying
a a scaling
transformation ri Ari (,X m,/(21L)), we obtain
-+ =
I-Is =
A H(PS2), (8.22)
M,
C8.3 Stability of hydrogen-like molecules 173
where m,. is the electron mass and H(PS2) is the Hamiltonian for
the positronium molecule. This equation indicates that the ground-
state energy of HS, ES, can be obtained from that Of PS2, E(PS2),
by Es =
(2A/Tne)E(PS2). Note that this relation can also be obtained
from the Helh-nann-Feynman theorem and the virial theorem. In fact
the use of Eqs. (3.38) and (3.53) enables one to obtain (mx 2/-t) =
d a 1
ES =
( fisjTax-Hsjfis) =
-- !PSjTSjPS)
dTax Tax
I
-
ES, (8.23)
Tax
which leads to the solution that E,,/mx independent of mX. Here

is
(Ps is the ground-state wave function of HS and TS is the kinetic
energy of HS.
According to the Ritz theorem the ground-state energy El of H
satisfies the following relation
El :! ((fiS IHS + HA PS) - (8.24)

The term HA is odd under the interchange P of the masses M and
m, i.e., PHA -HAP. Then by using that (PS is invariant under
=
P we ((PsIHA10s)
have (0SjpHApj(pS)=
_((pSjHAp2j0S) = =
-(0sjHAj(Ps) 0. The right-hand side of Eq. (8.24) is thus equal

to ((fisjHsj0s) ES, and we obtain
2A M.
El -
Eth :! ES -
(-A) =
M,
(E(PS2) -
(-
2 )) -
(8.25)
The energy of -m,/2 is the threshold for P,92 to decay into two
Ps atoms. If the stability Of PS2 is established, that is, E(PS2) +
(m,/2) < 0, then we can conclude immediately from Eq. (8.25) that
the hydrogen-like molecule is always stable. In fact Hylleraas and Ore
[991 calculated the energy Of PS2 by using a simple trial function and
showed its stability.
It is of interest to examine the stability of a hydrogen-antihydrogen
like molecular system, (M+m+M-m-) [1041. The lowest threshold of
this system is now Eth -(M/4) (m/4) corresponding to the dis-
-
sociation (M+M-) + (m+m-) and gets deeper and deeper as M in-
creases, which is in contrast to the case of the hydrogen-like molecule.

It is un I ikely that the hydrogen-antihydrogen system gets lower energy
174 Complements
than the threshold. It is thus expected that there is a critical mass ra
tio,M/m beyond which the system becomes unstable. According to

the calculations hn [104, 1161, the stability limit is M/m < 2.1.
8.4 Application of global vectors to muonic molecules

The aim of this Complement is to show the utility of the global vector
v of Eq. (6.4) for describing the angular part of the variational trial
function. As a test example we take up a Coulombic, three-body sys-

tem, the tdIL molecule, which has attracted much attention in relation
to muon catalyzed fusion [1171. The excited P state especially plays a
key role in the fusion since it lies close to the threshold for the decay
to the ttt atom and the deuteron.
The basis function we use is fKLM of Eq. (6.27). The Hamiltonian
matrix elements for this function can be calculated with the method of
Chap. 7 and the appendix. Without loss of generality the variational
parameters ul and U2 which define v can be normalized to satisfy

U +U22 1. Each basis function for
= a given set of KLM values thus
contains at most four nonlinear parameters, three of which come from

the matrix A. The matrix A has to be positive-definite to assure the
finite norm Of fKLM, and can in general be expressed as A 6DG, =
cosO sin,&
where G is a 2x2 orthogonal matrix ( -sinO cos ?
specified by just
one parameter V and D is a diagonal matrix including two diagonal
elements of positive d, and d2 values.
The accuracy of the variational solution depends on how the pa-
rameters ul, V, di, and d2 are given [331. The most naive choice would
be to take G as a unit matrix (6 =
0) ,
which is equivalent to us-
ing only a single set of relative coordinates x, and then to try to
reach convergence by including successively higher partial waves im-
plied by appropriate choice of K and ul values. Many examples

an
show [20, 29, 30, 311, however, that this type of single-channel calcu-
lation does not work well, especially in the case where the adiabatic
approximation is questionable as in the present example.
The matrix G may be chosen to let y Gx correspond to other =
particular coordinate sets such as the so-called rearrangement cha n

nel [201. (For Gx to correspond to the rearrangement channel, the
length scales of x and y in general have to be modified appropri-
ately.) Three possible rearrangement channels expressed in the Jacobi
coordinate set are drawn in particles d, t, and tL are
Fig. 2.2. If the
labelled particle 1, 2, and 3, the three patterns in Fig. 2.2 correspond
to (tlL)d, (/-td)t, (dt)IL arrangements, respectively. If x is understood
C8.4 Application of global vectors to muonic molecules 175
to stand for the coordinate set of the (tl-t)d arrangement, the co-
ordinate setcorresponding to the other

by arrangement is obtained
choosing such an appropriate,& value that is uniquely determined by
the change of the coordinates. With this choice of ?Y, we can write
i Ax ,'N DGx
=
Dy dj y', + d2y22
= =
*
The following three simple types of bases were chosen:
(i) K = 0 and A =
dDG, where G is restricted to the special ma-
tricesconnecting only the three sets of rearrangement channels as
explained above.
(ii) K = 0 or 1 and the choice of A is the same as in case (i).
(iii) K = 0 and A = dDG, where G is now an arbitrary orthogonal
matrix.
As mentioned in Sect. 6.2, the angular part with K

was 0 de- =
scribes only the stretched configuration, 11 + 12 L, and therefore -
the basis of type (i) allows rather limited angular correlations be-
tween the particles. In fact, the possible (Ili 12) values are given by
(1, L -
1), (1 =
0, ..., L). Basis (ii) is an extension of basis (i) to in-
clude the non-stretched coupling. With K =
1, possible (11,12) =
(I + 1, L -
I +
1), (1 0,..., L) are also allowed. Though K is set to
=
zero in basis (iii), the angular correlation can be taken into account
through the cross term of the exponential part of the basis function.
In bases (i) and (ii) the factors exp(- FbAx/2) are always "Correlation-
free" in a particular channel, that is, it contains no cross term in the
exponential part. In these bases it is through the inclusion of all rear-
rangement channels that one takes care of the correlation. In a basis of
type (iii), on the contrary, one needs to consider just any one arrange-
ment (that is, coordinate set) explicitly; the correlations are allowed
for through the general form of A.
Table 8.11 shows the results of calculations. As expected, the re-
sults with case (i) are poor compared to other bases. The correlation
contained in basis (i) is too restricted to obtain a realistic description
of the system. The angular correlation, which is taken into account
in basis (ii), certainly improves the energy over case (i). The basis
functions (iii) give even better energies. In fact they reproduce the
first six figures of the most precise variational calculations [20, 23, 921
for L = 0 --
2 states. Even with K
0, the use of the full A matrix
=
enables incorporate
one to the
important correlations between the
particles. A Gaussian basis of type (ii) is employed in the Coupled-
Rearrangement-Channel Gaussian (CRCG) basis variational calcula-
tion of [201, where the angular part is, however, represented by the
176 Complements
successive coupling of type (6.3). The fact that the D state energy
with the type (iii) calculation of dimension 200 becomes slightly lower
than that of [20] with 1566 basis functions confirms the importance of
a optimization of the nonlinear parameters. Further increase
caxeful
of the dimension can improve the accuracy rather easily in the angu-
lax function of Eq. (6.4) as seen in the case of S and P states. The
basis function used in [23, 921 is correlated exponential (CE) which
takes the form of exp(-air, -
T21 -
JT2 -
T31 -
7IT3 -'r1j) Multi-
plied by some polynomials for the orbital motion with

nonzero orbital
angular momentum. The parameters a, , -y are determined by the

random tempering explained in Sect. 4.2.2. These calculations using
the exponential function give very precise results.
Table 8.11. The total energies of the tdIL molecule. Parameter set a of
Table 8.2 is used except for the case indicated by superscript b, where set
b is used. Atomic units are used.
L Method Dimension E
S SVM (i) 200 -111.29346

SVM (ii) 200 -111.36398
SVM (iii) 200 -111.36444
600 -111.3645077
CRCG[20] 1442 -111.364507
CE[231 1400 -111.364511474
P SVM (i) 200 -108.13122

SVM (ii) 200 -108-17914
SVM (iii) 200 -108-17940
CE[231 1800 -108.1795424
b
SW (ifl) 800 -108-179361
CRCG[20]b 2662 -108.179385
CE[921b 1900 -108.1793881
P* SVM (iii)b 800 -99-660367
CRCG[201b 2662 -99.660548
CE[921b 1900 -99.6605507
D SVM (i) 200 -103.37067

SVM (ii) 200 -103.40824
SVM (iii) 200 -103.40849
CRCG[201 1566 -103.408481
9. Baryon spectroscopy
This chapter is devoted to applications of the SVM to the constituent

quark model of baryons. There are several different models of baxyons
and this is a very hot topic nowadays. The discussion of the different
models and their relative merits are far beyond the scope of this book,
and our only aim here is to show the applicability of the SVM to
solving tbree-body dynamics in constituent quark potential models.
The constituent quark model assumes that the baryons comprise
three quarks with dynamical quark masses and the quarks interact via
phenomelogical potentials. These potentials contain a confining term
and other terms which determine the fine structure of the spectra.
As there are no free quarks observed in nature, the confining inter-
action hinders the "ionization7 of the the baryons in the constituent

quark potential models. This interaction is usually taken as a power
law potential (V(r) rP). The second part of the interaction is tra-
-
ditionally chosen as the one-gluon exchange potential (OGEP) which

is a color analogue of the Breit interaction of quantum electrodynam-
ics (QED). There is an alternative possibility, the meson exchange
interaction, which is motivated by the fact that, despite its quite suc-
cessful reproduction of the ground-state spectra of the mesons and
baryons, the OGEP may not be adequate for describing the excited
states as mesonic effects or quark-antiquark excitations become im-
portant. This chapter includes examples for both choices to illustrate

how the SVM can be used for these systems with different interac-
tions and with different (nonrelavistic and semirelativistic) forms of
the kinetic energy.
These simple quaxk models have obvious limitations (e.g., large rel-
ativistic corrections can be anticipated for light quarks), but their suc-
cess indicates that relativistic and field theoretical effects are somehow
incorporated in the paxameters. In so far as sophisticated approaches

178 9. Baryon spectroscopy
based quantum chromodynamics (QCD) do not provide results for

on
these systems, the application of constituent quark models is justified.
9.1 The trial function in the constituent quark

model
The carries space-, spin-, flavor-, and color degrees of free-

quark (q)
doTn. The flavor is an extension of the isosphn to include up(u),
down(d), strange (s), charm(c), bottom(b), and top(t) quarks.
Thebaxyons are colorless: The three quarks in a baryon form a
color-singlet state, that is, they are totally antisymmetric with respect
to the interchange of the color coordinates. For example, the eight
baryons of spin 1/2, N (p, n), A, Z (Z+, V), Z-), and EF (EE,",
belonging to a family of octet baryons, consist of three quarks of u,
d and s flavor varieties. The product of the spin-, flavor-, and space-
part of their wave functions has to be symmetric to comply with the
generalized Pauli principle.
The spin paxt is represented by the spin functions X(S,,)Sm, (see
Complement 6.5):
xmi a 2
TIT), X(O).1.1
22
-v/2 (I TIT) -
I ITT)),
r6 (
11
X(1) 'TT 21111) -
I III) I ITT) (9-1)

-V
The flavor wave functions are shown in Table 9.1, where the
baryons with the same chaxge are arranged in each column.

The trial wave function for baxyons is a combination of the terms
-1
IR =
I (
S exp -
2 6Ax) X(S12)SM (T12)TMT (9.2)
where X and are the spin and flavor functions, and S is a sym-
metrizer.
9.2 One-gluon exchange model
The one-gluon exchange potential (OGEP) model has proved to be

quite successful in describing the spectra of the ground states and
various properties of the baryons [118, 119, 1201. In this model the
9.2 One-gluon exchange model 179
Table 9. 1. Flavor wave ftmctions of baryons arranged columnwise according

to charge
Baryon +2 +1 0 -1
N uud ddu
A UUU uud ddu ddd
A 71,= (ud du)
2
'
-
UUS
71,
72 (u
d + du) s dds
SSU ssd
S? SSS
Ac 7L (ud
2
,
-
du) c
du) c ddc
El- 72= (ud
UUC +
Ab 72 (ud du)b
-
1
Eb uub
vr2-
(ud + du) b ddb
quarks exchange massless but color-charged gluons and the qq poten-

tial is taken as the color analogue of the Breit interaction [3, 1211.
Consequently, the simplest version of the OGEP includes a Coulomb-

like I/r piece and a color-magnetic ("hyperfine") piece:
7r
-
(AI7. AC
S j)(ffi-0j)j(rj-rj)' (9-3)
i<j
6mimj
where the Gell-Mann matrix A9 (a) (a

z
=
1, 8) is the color SU(3)
generator for the ith quark and (A'9 Ajc)
-
stands for a scalar prod-
uct, 1:8 =1 AF (a) AC (a). (In a more elaborated framework one can
introduce spin-orbit, tensor and other higher order terms.) The quark
masses mi in the denominator introduce avery weak flavor depen-
dence, but the interaction is essentially flavor-blind. In a nonpertur-
bative treatment the contact interaction leads to a singularity. Because
of this and because of the finite spatial extent of the constituent quarks
the delta function should be smeared out, and it is to be substituted
by a Gaussian or a Yukawa form factor.
Among the several different parametrizations of the OGEP we have
chosen the AM potential [1221 as an example:
Vij H 3(A9
8
-
A(7)
2rr.' exp
Tij
ij
x --+Ar-A+ 1. 3
a - -
O'j , (9.4)
r 3mpaj 7F 2
Pij
where the
length r is given in units of inverse energy (I fin 1/197.327 =
MeV-'), and the smearing parameter pij is given iTi the forioa
-B
( 2mimj
Pij
_
"
A
mj + Taj ) (9-5)
The quark masses used in the calculation are Tn,, Tnd 315 MeV, = =
m , 577 MeV, m,
= 1836 MeV and ?nb= 5227 MeV. The parameters=
were determined by a fit on a large sample of meson states in every
ffavor sector [1221: They are n 0.5069, rs' 1.8609, A

= 0.1653 = =
GeV2 A 0.8321 GeV, B

7
= 0.2204 and A 1.6553 GeVB-'.
= =
The matrix element of the color part can be easily obtained because
the baryons are in a color singlet state. By noting the relationship
between the color-exchange operator P53 and (AF Ajc), -
C
PC== (A9-Af)+-,
' (9.6)
zy J
2 3
the matrix element of (A9.A'7)

S 3
between the color-singlet states reduces
simply to -8/3.
The SVM results [81, 1231 are compared with the calculations of
[1221 in Table 9.2. The agreement between the two results is good.
The experimental data [1241 are included for guidance. One has to
compaxe the energies relative to the nucleon ground state and one has
to bear in mind that this potential gives good results for the mesons
and no free parameter has been introduced to fit the baryons.
Table 9.2. The masses of baryons (in MeV) as predicted by the AU poten-
tial [1221. The numbers in parentheses axe the masses relative to the nucleon
mass.
Baryon Ref. [1221 SVM Experiment [1241

N 998 995 939
A 1306(311) 1232(293)
A 1154(156) 1149(154) 1116.(177)
E 1231(233) 1228(233) 1192(253)
1343(345) 1339(344-) 1318(379)
0 1674(676) 1674(679) 1672(733)
A, 2296(1298) 2290(1295) 2285(1346)
Z. 2466(1468) 2467(1472) 2455(1516)
Ab 5642(4644) 5635(4640) 564150 (470250)
Zb 5849(4851) 5849(4854)
-Fb 5808(4810) 5803(4808)
9.3 Meson-exchange model 181
9.3 Meson-exchange model
The OGEP model is successful in describing ground-state energies

and properties of baryons, but for the excited states a number of
delicate problems remain unsolved. No model has been able to explain,
for example, the correct level
orderings in light- and strange-baryon
spectra in a simple three-quark description of baryons. The source of
the problem can be easily understood by the following argument: The
ground state of the nucleon N(939) has positive parity followed by the
positive paxity 1/2+ Roper resonance N(1440) and the negative-parity
1/2- and 3/2- states N(1535) and N(1520), respectively. The parity
of the states follows as (+, +, -, -). In the case of the A, the order of
states follows a (+, -1 +1 -) pattern. The order of parity of the spectra
in a simple harmonic-oscillator quark model is (+, -, +, In the
case of the linear confinement the level spacing changes, but the order
of states remains the same. The wave function of N and A differs

only in flavor, but the OGEP is flavor independent (though it slightly
depends on flavor through the mass difference of light and strange
quarks in Eq. (9.4)). Therefore it predicts the same level order for
both N and A in contradiction to the experiments.
To resolve this problem an explicitly flavor-dependent meson-
exchange interaction has been introduced [1251. The essential differ-
ence between the two approaches is that while the OGEP of type
Ei<j (AF AjC)

%3
(ai o-j) V(rij) acts on the color and spin space, the
- -
meson-exchange potential Ei<j AFA- 37(aj-o-j)V(rjj) acts on the flavor

'1
and the spin degrees of freedom.

The meson-exchange potential combined with a semirelativistic
form of the kinetic energy is the second example of the application of
the SVM for baryon spectroscopy. In this example no gluon-exchange
mechanism is explicitly introduced to describe the qq interaction but
only a meson-exchange potential is employed to generate the baryon
spectra.
The
three-quark Hamiltonian is a sum of the semirelativistic Id-
netic energy, the meson-exchange potential V and the confinement
potential VO + Cr:
3 3
H=E /--2+ ?
P71i +M?
Mi2 + (Vii + V" + Clri -
ri 1) (9.7)
where the meson-exchange potential V reads as

3 7
Vij (r) = V7r (r) 1: AF (a),XjF (a) + VI (r) AF

i (a) AjF (a)
ij 3
a=1 a=4
2
+ VI7 (r) AF
ij i (8) AjF (8) + 3 Vj'j7' (r) Oj -
aj
The form factor of the meson-exchange potential is paxametrized as
g2 e-A-yr 6 -Ay
1 f
V
ij 47-, 12mimj I
2
A 'I
r
A2-
'Y
r I 1 (9-9)
where the coupling constant gy is taken to be g8 for the pseudoscalar

octet mesons (7r, K, n) or go for the pseudoscalar singlet meson (n,),
respectively and the cut-off mass Ay is assumed to be given by
Ay = Ao + riLy (9.10)
for each -/ =
q, 77. For the constituent quark masses we take the
-Ft, K,
typical vdues, Mu, Md 340 MeV and m,
= 500 MeV. Parameters
= =
of the Hamiltonian are listed in Table 9.3.
Table 9.3. Parameters of the semirelativistic constituent quark model with

meson-exchange potential
Fixed parameters
Quark masses [MeVI Meson masses [MeVI

928
Mui Md M, Ar AK An A'a 41r
340 500 139 494 547 958 0.67
Free parameters
(golg8)2 Ao [fin-] K Vf0 [Me,V C fln-21

1.34 2.87 0.81 -416 2.33
The -matrix elements of the semirelativistic kinetic energy in Eq.

(9.7) can be calculated with the method presented In Sect. 7.6. The
matrix elements of AF(a)AjF(a) in the flavor space can be easily calm-
71
culated by-using Table 9.1 and the explicit form of the GeH-Ma=
matrices. It is useful to know the action of the ffavor-changing oper-

ator AF(a)A-3 (a)
Z
on the flavor wave functions. As is summarized in
Table 9.4, for example the ffavor function uidj, when acted upon by
3
E I
AF (a) M'(a), transforms to 2djuj
Z
uidj. -
a= 3
Table 9.4. Effect of flavor-changing operators on quark pairs
Operator UU dd SS ud us ds
3
1
Ar (a) X3 (a) UU dd 0 2du-ud 0 0
4
AF (a) AF
Z j (a)
0 0 0 0 2su 2 sd
/\F(8),XjF(8) -1 UU -1 dd 4UU lud -!us -ids

3 3 3 3 3 3
_
In Fig. predictions of this model axe shown for all

9.1 the SVM
light- and strange-baryons with mass up to M < 1850 MeV; the nu-
cleon mass is normalized to its mass of 939 MeV, which determines
the value of the confinement potential parameter V0. All masses cor-
responding to three- and four-star resonances in the most recent com-
pilation of the Particle Data Group [124] are included.
The result is good, reproducing the spectra of all low-lying light
and strange baryons [123, 126]. In particular, the level ordering of the
lowest positive- and negative-parity states in the nucleon spectrum is
reproduced correctly, with the 1/2+ Roper resonance N(1440) falling
well below the negative parity 1/2- and 3/2- states N(1535) and
N(1520), respectively.
Likewise, in the A and Z spectra the positive-parity 1/2+ excited
baryons, A(1600) and E(1660), fall below the negative parity 1/2--
3/2- states, A(1670)-A(1690), and the 1/2- state E(1750), respec-
tively. In the A spectrum, at the same time, the negative-parity 1/2--

3/2- states A(1405)-A(1520) remain the lowest excitations above the
A ground state.
At this remark is in order about the necessity of employing
place, a
a semirelativistic kinetic energy operator in the three-quark Hamilto-
nian (9.7). Certainly, this is only an intermediate step toward a fully

covariant treatment but it already allows us to include the kinemat-
ical relativistic effects. In any nonrelativistic approach these effects

get compensated by the potential parameters, but there one faces a
disturbing consequence of v1c > 1, where v is the mean velocity of the
constituent quark and c is the velocity of light.
1800-
1700- El E----1
= L-J
1600-
1500-
M
1400-
[MeV]
1300-
1200-
1100-
1000- N A
900- 1
1+ 1- 3+ 3 5+ 5 1- 3+ 3-
2 2 Y Y Y Y i -2
Fig. 9.1. Comparison of low-lying baryon spectra predicted by the meson-

exchange potential model with experiment. The solid lines denote the cal-
culated spectra, and the shaded boxes show the experimental masses with
theirerror bars.
(Continued on the next page.)
One may think that the idea of the meson-exchange interaction

acting between quarks is unfamiliar though it has produced the good
results. To judge its merits the present calculations have to be ex-
tended into several directions (prediction physical observables, e.g.

of
form factors, decay widths, etc.).
The complexity of the NN interaction is basically due to the com-
positeness of the nucleon. As N, A or Z belong to the same octet
family, it is natural to attempt a unified description of the NN, AN
or EN interaction from the underlying qq potentials. This goal is at
present too difficult to achieve directly on the basis of QCD. In so

far as QCD as yet presents no results, the effective theory formu-
lated in a microscopic quark-cluster model [1271 seems to be justified.
(Continued from Fig 9. 1.)
17.71
E]
F-7 -
r__,_T
1800 -
t H p
1700 -
r--.7. G= - p"
r_.j
0
1600 -
1500-
M
1400-
[MeVj
1300-
1200-
1100-
1000- A
900 1 1 1 1 1 1
11
_2 2 _2 _2 _2 _2 _2 _2
1800
1700
1600
1500
M
1400
[MeVJ
1300
1200
1100
1000
900
1+ 1- 3+ 3- 3+
The intermediate-range attraction of the baryon-baryon interactions

cannot be accounted for by the OGEP because it produces no con-
tribution between the colorless baryons. It is not clear yet whether
the meson-exchange potential introduced above leads to a realistic de-
scription of the NN interaction or not. It is probable that the OGEP
combined with the meson-exchange potential is a useful, effective ve-
hicle to obtain a reasonable description of both baxyon spectra and
baryon-baxyon interactions at the same time. See, for example, [128]
for the attempt along this line.
10. Few-body problems in solid state
physics
This chapter is a collection of examples of the application of the SVM

to few-body problems in the field of solid state physics. The examples
include excitonic complexes, biexcitons, and quantum dots with and
without magnetic field.
The biexciton (or excitonic molecules), the bound state of two holes
and two electrons, provides an interesting few-body problem, since it
may be thought as a positronium. or a hydrogen molecule with variable

electron and hole masses realized in semiconductors. The biexcitons
are also often confined and can be modelled as two-dimensional (2D)
systems.
In recent yeaxs there has been much excitement in the possi-
ble applications of ultrasmall. systems with a length scale of 10-100
A,("nanostructures"). The technological motivation (in electronics and
optoelectronics) is that the smaller components are faster. It is possible
to manufacture very small nanostructures, often called "dots", which
contain only a very few electrons (N < 10). The quantum-mechanical
effects in these small systems axe very important and their properties
are strongly N-dependent. The energy spectra of few-particle quantum
dots call for theoretical interpretation. In a nanostructure the electrons

are confined in a box of sizes L, L., and L,. If L, -
Ly >> L, the
quantum dot is quasi 2D. The confinement is usually modelled by a
harmonic-oscillator potential. This system is somewhat similar to an
atom in nature, where the "confining" potential is the Coulomb at-
traction of the electrons to anucleus in the atom, and therefore the
quantum dots are often referred to as "artificial atoms".
Throughout this chapter we use an effective mass me* for an elec-
tron. When the dielectric constant of a material is denoted by r., the
length and the energy will be measured in "atomic units" with Bohr
radius a* =
h2n/ (M*e2
e- ) and haxtree (two times the Rydberg energy)
2R* =
e2/(Ka*), respectively.

188 10. Few-body problems in solid state physics
10.1 Excitonic complexes
The excitonic complexes have considerable importance in the develop-

ment of semiconductor physics and spectroscopy. The bound state of a
positively charged hole with an effective mass m and an electron with

an effective mass m,* is called an exciton in semiconductor
physics. The
mass ratio o- m*/m*h
= can change between c- 0 (hydrogenic limit =
e
and o- I (positronium limit) depending on the material and other

=
factors. Like in the case of hydrogen, where not only the hydrogen
atom but the hydrogen molecule (H2), or the H2+, H-,
H,+ ions are
bound, the system of N, electrons and Nh holes can also be bound.
The latter system is called an excitonic complex. These excitonic com-
plexes, including the charged excitons, have been subject of intensive
experimental [105, 1291 and theoretical [99, 130-1351 investigation.
The properties and structure of these systems strongly change with
the mass ratio, and, by approaching the two limiting
cases, one arrives
at two completely different worlds. The interest in these systems has
been intensified when the advance in semiconductor technology has
made possible the fabrication of artificial nanostructures with diam-
eters comparable to atomic distances. This restricted geometry has a
prominent effect on the dynamics of the excitonic species.

In the following we present the ground-state
energy and some
other properties of 2D excitonic complexes. The general Hamiltonian
can be written as
IV. N Ne
H
pi, pi, e2
= + TC. +
i=1
2m, 2Mh
i=IV'+I 3>z=l
N N, N
e2 e2
E r"Iri
E E
j>i=N' +l
-
rj I i=1 j=N,+1
where x is the dielectric constant of the material, and the position

vectors 2D vectors in the xy plane. In atomic units introduced
axe
in this chapter, the Hamiltonian becomes dimensionless and the en-
ergy depends only on o-. (Note that

specify what
we do not need to
value of K is used because the
dependence is hidden in the atomic
r,
units.) This is easily understood by introducing dimensionless coor-

dinates and momenta by ri* ri/a* and pi* pil(hla*). Then the
= =
Hamiltonian (10-1) is reduced to

10.1 Excitonic complexes 189
N, / N 2
H 1 2 *2 0-
2R* 2P' + E 2
P
2(o-N, +N -
N,)
P'
i=N +I
N N'- N
+ + E Jr -
'r I
E E J'r j* r I'
Z 3 7, 3 3
i>i=l j>i=N,+l i=1 j=N,+l
At this point, a comment is due. What we consider here is a well-

defined quantum-mechanical problem to be solved. This is, however,
only a model of the
experimental situation. In the realistic case there
are many other factors to be
considered, for example the effect of other
electronic bands, the possibility that the interaction is different from
the pure Coulomb force due to the confinement in the z direction, etc.
The discussion of the importance of these effects is beyond the scope
of this book.
As theproblem here is practically the same as that discussed in
the chapter of small atoms and molecules (Chap. 8), the same trial
function is used (see Eq. (8.1)), except that it is tailored to the 2D
case, that is, the position vectors have only x and y components.
The results for o- = 0 and o- = 1 are shown in Table 10.1 for the 3D
case and in Table 10.2 for the 2D case. The most striking difference
between the 3D and 2D results is the large increase in the binding
energy in the 2D case. E.g., the binding energy of R2 (with o- 0) in =
2D is about 8 times of that in 3D. The increase of binding has been

found experimentally as well.
Table 10. 1. Energies and binding energies (in a.u.) of 3D exciton complexes
of electrons and holes for two cases of the mass ratio o-. L and S are the total
orbital angular momentum and spin of the exciton complex, respectively.
The asterisk refers to the states which are found to be unbound. The binding
energy is understood with respect to the nearest threshold.
System (L, S) E(o-=O) E(o-i) B(o-=O) B(o-=I)

eh (0,0) -0.500 -0.250 0.500 0.250
eeh (0,1/2) -0.528 -0.262 0.028 0.012
ehh (0,1/2) -0.602 -0.262 0.102 0.012
eehh (0,0) -1.174 -OM6 0.174 0.016
eeehh (0,1/2) *
eehhh (0,1/2) -1.344 0.169

Table 10.2. Energies and binding energies (in a.u.) of 2D excito-n complexes
of electrons and holes for two cases of the mass ratio o-. M is the z component
of the total orbital angular momentum and S is the total spin. See also the
caption of Table 10.1.
System (M, S) E (o- =

0) E (o- =
1) B (o- =
0) B (o- =
1)
eh (0,0) -2.00 -1.00 2.00 1.00
eeh (0,1/2) -2.24 -1.12 0.24 0.12
ehh (0,1/2) -2.82 -1.12 0.82 0.12
eehh (0,0) -5.33 -2.19 1.33 0.19
eeehh (0,1/2) *
eehhh (0,1/2) -6-82 1.50

10.2 Quantum dots 191
distances have to fullfil the equality 2(r+-) =

V(r--)2 + (r++)2. Ta-
ble 10.3 shows that this is not satisfied. Moreover, the fact that the
uncertainties Arij -
(,ri2j (rij 2 are large means that no such in-
terpretation is possible. There is only one case where the uncertainty is

in
very small, that is the distance between the (heavy) positive charges
the hydrogenic limit. In that case one can assume, in accordance with
the adiabatic approximation, that the positive charges are fixed at the
distance of 0.37 a.u. The equilibrium distance in the H2 molecule in
3D is 1.4 a.u.
Table 10.3. Average distances in 2D biexcitons (eehh) as a function of the

mass ratio o-. E.g., (r--) is the mean distance between the two negative
charges. Atomic units are used.
0- = 0 o- = 0.4 a = 0.7 0- = I
0.67 1.26 1.55 1.80

(r++) 0.37 1.14 1.49 1.80
(,r+-) 0.47 0.93 1.16 1.38
2
(r _) 0.59 2.09 3.19 4.33
(r 2+) 0.14 1.69 2.95 4.33

2
(r _) 0.31 1.29 2.05 2.88
10.2 Quantum dots
Rapid advances in semiconductor technology have led to the fab-

rication of nanostructures called quantum dots or artificial atoms
[136, 137, 1381. In quanturn dots a few electrons, typically 2 to 200
electrons, are bound at semiconductor interfaces. These few-electron

systems arise when homogeneous two-dimensional (2D) electron gases
of heterojunctions or metal-oxide-semiconductor (MOS) structures are
laterally confined to diameters comparable to the effective Bohr ra-
dius of the host semiconductor. The interest in quantum dots arises

not only from the prospect of new technological applications but also
from the desire to understand the physics of a few interacting electrons
in anexternal field. Needless to say, we concentrate on the few-electron
case where the effect of electron-electron interaction and their corre-
lation seems to be very important.
The axtificial atoms we consider here are modelled by a system

of N electrons confined in 2D by harmonic-oscillator potential. Be-
a
fore starting with artificial atoms in 2D we present a calculation for

the energies of 2D "natural" atoms where the "co-ofi-ohn ' potential is
Coulombic. The results for few-electron atoms and ions are given in
Table 10.4. The binding energies are again, due to the 2D geometry,
much laxger than in the 3D
case. Otherwise the properties of these
systems axe rather similax to those of their 3D counterparts. For ex-

ample, akin to 3D, the H, Li or Be atom can bind an extra electron,
while the He atom cannot.
Table 10.4. Energies of atoms and ions confined in 2D. M is the z compo-
nent of the total orbital angular momentum and S is the total spin. Atomic
units are used.
System (M, S) Energy

H (0,0) -2.00000000
H- (0,1/2) -2.24
He (0,0) -11-89
Li (0,1/2) -29-87
Be (0,0) -56.77
Theex6mple of quantum dots considered here is a system of 2D

electrons in a quadratic confiniD potential. The Hamiltonian reads as
N 9 N N
Pi I e2
H
2m*
+ _M
2
*WO2,r? + (10.2)
(V__ + ivy)
M
exp( IiUr),
-
2
(10.3)
where v., and v. are the x and y components of the 2D vector v defined
by
V Uiri. (10.4)
The parameters ui and A of the basis function are determined by the

SVM-
Due to the external field (the single-particle potential), the Hamil-
tonian (10.2) of the system is not translationally invariant. Corre-
spondingly, the basis function (10.3) is written in terms of single-

particle coordinates instead of relative ones. This basis function de-
pends on the center-of-mass coordinate, and the energy obtained will
be the total energy of the system, unlike the previous cases where
the energy of the center-of-mass motion was always subtracted. The
case of the quadratic confinement is very special, because in that case
the energy of the center-of-mass can be separated. In general, for an

arbitrary single-particle potential, however, this cannot be done.
The case of N 1 electron is trivial and its solution is a simple
=
harmonic oscillator. For N 2 an analytical solution exists for cer-

=
tain frequencies [139]. For example, for hwo I a.u., the ground state
=
energy is 3 a.u., which is precisely reproduced by the numerical cal-

culation. Not surprisingly, as is shown in Table 10.5, if the quadratic
confinement is strong, the spectrum is close to that of a harmonic
oscillator. In the realistic cases (hwo < I a.u.), however, the eigen-
values strongly deviate from those of the harmonic oscillator. This
example illustrates the importance of these systems: by changing the
confinement one can "tune" the properties of these artificial atoms.
Table 10.5. Energies of N-electron quantum dots with the z component of

the total orbital angular momentum M and the total spin S. The numbers
in parentheses are the eigenvalues without Coulomb interaction. Atomic
units are used.
N (M, S) hwo = I hwo = 100
2 (0,0) 3.000 (2.000) 212.198 (200.00)

3 (0,1/2) 7.220 (5.000) 525.12 (500.00)
4 (0,0) 10.600 (6-000) 654.45 (600.00)
Figure 10.1 shows the energies of the ground and excited states
(in units of the energy of a harinonic-oscillator quantum) of two elec-

trons as a function of the inverse square root of the oscillator constant,
-1/2 If there would be no Coulomb interaction between the elec-
(hwo) .
trons, the energy divided by hwo would be given by lines parallel to

the horizontal axis., The deviation from the straight lines is entirely
due to the Coulomb interaction. Another interesting feature is that
level crossings occur between the excited states but the ground state
always remains M 0.
12
10
0 2 4 6 8 10
(hcoo )-1/2
Fig. 10.1. Energy levels of two electrons confined by a harmonic-oscillator
potential with an oscillator constant wo. Atomic units are -used.
3.5
3.0
2.5
2.0
;-4
t__I
(14 1.5
1.0
0.5
0.0
1 2 3 4
r (a.u.)
Fig. 10.2. Electron densities of two-electron quantum dots as a function
of the radial distance from the origin. The solid, the dotted and the dashed
curves correspond to hwo 0.2, hwo 2, and hwo 5, respectively. Atomic
units are used.
1.2
1.0
0.8
A 0.6
0.4
0.2
0.0
0 2 4 6 8 10
r (a.u.)
Fig. 10.3. Electron densities of two- (dashed curve), four- (dotted curve),
and six-electron (solid curve) quantum dots. The oscillator frequency of the
quadratic confining potential is hwo = 0.2. Atomic units are used.
peak density inreases with the number of electrons, that is, the equi-
librium configuration is realized on rings of expanding diameter. In
the case of N=4 electrons, for example, the paxticles may move along
a ring, while they are situated at the vertices of a square to TniniTni e
the Coulomb repulsion.
10.3 Quantum dots in magnetic field
A dramatic feature axises when the quantum dots are subjected to

magnetic fields: The ground states are stabilized into magic-number
states. When the Tn agnitude of the magnetic field is vaxied, the ground
state jumps from one magic number state to another. Magic numbers
appear owing to the combined effect of the electron correlation and
the Pauli principle [140].
The Hamiltonian of the system is given by
N N
I I
H =
E 2m* (pi + e-A,)2 +
- -
C
-M
2
,, 2
Wd 'ri
2
N
e2
+ E 1' Iri -,rj I*
(10.5)
i>i=l
Let us assume that a homogeneous magnetic field B is applied down-

ward in the z direction. The above Hamiltonian can then be rewritten
in the form
P' 2
1
H 2-
+ -m w r?- -
hw, Ii,
2m* 2 2
N
e2
rdri -
rj I
where u), eB/(m*c) is the cyclotron frequency and w =

V/W--22.
rl4+wo
We ignored the
magnetic interaction energy due to the electron spin.
See Complement 10.1 for details. It is shown there that the part of
the Hamiltonian corresponding to the single-particle motion can easily
be diagonalized by the use of associated Laguerre polynomials. De-
pending on the strength of the magnetic field, both the single-particle
10.3 Quantum dots in magnetic field 197
motion and the correlation due to the Coulomb repulsion play deci-
sive roles in determining the structure of the quantum dots. The trial
function is the same as in the previous section.
For the two-electron case our calculation is in perfect agreement
with the analytical solution
given in [1411. The spectrum of two elec-
trons in the spin-singlet state (S =
0) is shown in Fig. 10.4. See also
Complement 10.1. The harmonic-oscillator frequency wo is taken as
hwo 0.01 a.u. The most intriguing phenomenon seen is the level
=
crossing. Due to these level crossings, the ground state changes with
the strength of the magnetic field. This is shown in a magnified scale
on the right-hand side of Fig. 10.4. If there is no Coulomb interaction
between the electrons, no level crossing occurs.
10.0
18
9.5
16
9.0
14
+Z +It 8.5
12
8.0
10
7.5
7.0
0 1 2 3 4 0.0 0.5 1.0 1.5 2.0 2.5
(J)C/(00 (,)C/ ("0
Fig. 10.4. Energy levels of a two-electron quantum dot in magnetic field

as a function of the cyclotron frequency w,. The figure on the right-hand
side magnifies the level crossing. The oscillator frequency of the quadratic
confining potential is hwo 0.01. The total spin is S
= 0. Atomic units are =
used.
Figures 10.5 and 10.6 plot the energy levels of a three-electron

quantum dot (with the total spin being S 3/2) against the orbital
=
25
20
15
10
0 2 4 6 8 10 12 14
COC/ coo
Fig. 1.0.5. Energy levels of a three-electron quantum dot belonging to the
z component of the total orbital angular momentum M = 3 (solid c=e),
IVI = 6 (dotted curve), and M = 9 (dashed curve) orbital motion. The
frequency of the quadratic confining potential is hwc) = 1. The total spin is
S =
3/2. Atomic units are used,
15
14--1
13
12
44 10-
0 1 2 3 4 5 6 7 8 9 10
M
Fig. Energy levels as a function of the z component of the total

10.6.
orbitalangular momentum, M, of a three-electron quantum dot in a low
magnetic field (hw, 0.2, hwo
= 1). The total spin is S
=
3/2. Atomic=
units are used.

of these orbits larger (see Eq. (10.16)). The energy of the dots
axe
is determined the
interplay of the single-paxticle energies and the
by
interaction energy. By increasing the strength of the magnetic field,
similarly to the case of low magnetic field, the ground state does -not
take all possible values of M. By changing the magnetic field the states
with M 3,6,9,12,... "magic numbers' become ground states.
=
40-
35-
30-
25 -
201
10
0 1 2 3 4 5 6 7 8 9
M
Fig. 1.0.7. The same as in Fig. 10.6 but for high magnetic field (hw,- =
6,
hwo =
1). Atomic units are used.
change of the ground-state angular momentum quantum num-

The
bers as a function of the magnetic-field strength is illustrated In Fig
10.5 for a three-electron quantum dot with spin S 3/2. This figure =
shows the energies of the states with M 3 (solid curve), M = 6 =
(dotted) and M 9 (dashed)

= orbital angular momentum. The cal-
culation shows that by changing the strength of the magnetic field
the orbital angulax momentum of the ground state of this system is
M 3 for w,lw < 5, it is M
= 6 for 5 < w,lw < 11, and then it
=
switches over to M =
9, reproducing the "magic number" sequence
(M =
3,6,9,..). States with other orbital
angular momentum never
become the ground state and are not included in the figure.
The previous examples dealt with spin polaxized electrons (S =
3/2). Actually, spin polarized (S 3/2) and spin unpolarized (S

= =
1/2) states compete to be the ground state as the magnetic field

changes. This very interesting phenomenon is illustrated in Fig. 10.8.
As the magnetic field increases, the ground state changes as (M, S) =
(1, 1/2) -+ (2,1/2) --+ (3,3/2) and so on. At low magnetic field the
lowest energy state is spin unpolarized (S 1/2) and as the mag-
=
netic field increases the spin becomes polaxized. The orbital angular
momentum (M) changes continuously (in steps of unity).
C5 2
0.0 0.5 1.0 1.5 2.0 2.5
(j)C/ (1)0
Fig.1-0.8. Spin (S) and z component of the total orbital angular momentum
(M) quantum numbers of the ground state of a three-electron quantum dot
as a function of magnetic field strength. hwo = 1. Solid and dotted lines
denote M and S, respectively. Atomic units are used.
10.4 Quantum dots in the generator coordinate

method
In Sects. 10.2 and 10.3 the correlated Gaussian basis of Eq. (10.3) is
employed to obtain the solution of the quantum dots. However, th is is
not the only Gaussian basis function for the quantum dots. In fact, the
quantum dots offer a very niceapplication of another basis function
(8.9) that was used for the study of small molecules. That basis func-
tion is characterized by the matrix A which describes the correlated
motion of the particles and also by the generator coordinates s which
allow us to represent several peaks of the density distribution of the
system.
In this section we consider the quantum dots in 3D and apply
no magnetic field. The
ground state of the quantum dot
can belong
to nonzero orbital
angular momentum (for example, in the N 3- =
electron case). The application of the basis function (8.9) gives a very
simple way to obtain accurate energies of the dots even with nonzero
orbital angular momentum, without the hassle of partial-wave expan-
sion or angular momentum algebra. As this basis function has no def-
inite orbital angular momentum, in principle one has to project out
the component with good angular momentum quantum numbers. In
practice, however, the converged wave function already belongs to the
correct ground-state quantum numbers and the energy gain due to
an orbital angular momentum projection is negligible. As the Hamil-

2
tonian and the L operator commute, the optimization of the wave
2
function filters out the eigenstate of L as well. As was mentioned
in Sect. 6.1, this type of calculation is called a variation before pro-
jection. In the limit that the variational trial function is chosen to be

flexible enough, the solution of this type of calculation would reach the
2
eigenstate of the conserved quantity like L to a good approximation.
In Table 10.6 the results of the SVM with the basis function of
Eq. (8.9) is compared to that of a large scale shell-model calculation.
The shell model works extremely well in this case because the con-
fining potential is a harmonic-oscillator well. By choosing the single-

particle wave functions to be the eigenfunctions of this single-paxticle
potential, the diagonalization of the Hamiltonian with the relatively
weak Coulomb interaction between the electrons is simple and gives
an accurate solution. The results of the two methods are in perfect
agreement. This agreement shows that the optimization of the basis

of Eq. (8.9) finds the correct ground state automatically even for the
10.4 Quantum dots in the generator coordinate method 203
case with nonzero orbital angular momentum provided that the ba-
sis size is sufficiently large. An orbital angular momentum projection
may lead to a faster convergence, but as it can only be carried out by
a three-dimensional numerical integration, it would prohibitively slow
down the calculation.
Table 10.6. Energies of few-electron quantum dots in 3D calculated by

SVM and by a large scale shell-model (SM) basis'. L is the total orbital
angular momentum, S the total spin, and -x is the parity. K is the basis
dimension. hwo 0.5. Atomic units are used.
=
N (L, S, -ri) E (SVM) K E (SM) SM space (in hwo)

3 (1,1/2,-) 4.01324 100 4.01324 39 hwo
4(0,0,+) 6.35025 300 6.3506 14 hwo
5 (1,1/2,-) 9.00331 500 9.0032 6 hwo
a
P. Navratil, private communication.
204 Complements
Complements
10.1 Two-dimensional electron motion in a magnetic field

In the previous section we presented a have vaxiational solution to the
few-body problems of quantum dots. In this approach a correlated
Gaussian basis has been used and the problem has been formulated In
a relative coordinate framework. The examples from atomic physics
(Chap. 8) convincingly prove that this is a very powerful -way to cope

with the correlation between the electrons in the field of an attractive
center. This is, only possible approach. One can,
however, not the
alternatively, use a basis set made up from products of single-particle

states to diagonalize the Hamiltonian. The properties of these single-
particle states axe discussed in this Complement.

The Hamiltonian of interest is given by
H
-2m* (P eAi)2
+
c
+ 9*ABB -
2
aj I + vori)
+ (10.7)
X 1ri -,rj I
i 'i
where m* is the effective electron mass, AB eh/(2m,c) the Bohr =
magneton of the electron, g* the effective g-factor of the electron,

and r. is the dielectric constant of the bulk material. The magnetic
interaction energy due to the electron spin is included in the Hamill-
tonian. The magnetic field is directed downward in the z direction,
i.e. B =
(0, 0, -B) with B
> 0. For high magnetic field of, e.g.
B - 10 T, the value of ILBB becomes 0.5788 meV. The vectors ri
and pi axe 2D vectors, which have x and y components. The one-
body potential V(r) is to confine the quantum dots and is chosen

to be quadratic, m* W02T 2/2. A typical value of the confinement en-
ergy is hwo 5 meV. By taking a symmetric gauge vector potential,

-
A = -r x B/2 B(y/2, -x/2, 0), we obtain
e
2 _2B 2 e. hB
(P +
c
A
) _Ji2' j +
4C2
2
c
(10.8)
where A + _ey2- and I-, is the z component of the angular mo-
mentum.
To express the Hamiltonian in terms of dimensionless vaxiables, -we
introduce
C10.1 Two-dimensional electron motion in a magnetic field 205
e,B hI
W, =
M
W =
WC2 + W02 and p =
. W
U)
(10.9)
Here wc is the cyclotron frequency and p is the magnetic length. (Note

that the magnetic length is expressed by p using Bohr
=
- -2R*-1hwa*
radius a* and hartree 2R*.) Then the Hamiltonian is expressed by the
dimensionless variable ri:
1 1
H = hw
2
A, + lr,2
2 2
hw, E lj , -
2
g*AB B aj ,
i
e2 2R*a*
+XE 1ri -
rjI
with X =
rIP
-
P
(10.10)
j>i
Here the length is in unit of the magnetic length. The cyclotron energy
is hwc =
(2ra,/m*)ABB. The ratio of m,/rn* is of order 15 for the 2D
electron system in GaAs-Ga,,All-,,As heterojunctions.
The strength of the Coulomb potential is of the order of X, while
the level spacing of the single-particle energy is determined by hw.
Therefore the ratio Xlhw -
VI'2--R*-Ihw is the quantity which charac-

terizes the electronic motion. If the ratio is much smaller than unity,
the independent motion of the electrons becomes a good approxima-
tion. As it increases, the correlated motion of the electrons overwhelms
the independent motion.
Recall that the second and third terms of the above Hamiltonian
are consistent with the well-known fact that a spin 1/2 particle with
mass M and charge qe in the magnetic ffied B has the magnetic
interaction energy -IPB, where the magnetic moment tL of the particle
is in general expressed by tL g1l + gs in units of ehl(2Mc). Here
=
g, and g, are the respective gyromagnetic ratios for the orbital and
spin angular momenta. The gi value is equal to q, while the Dirac
equation would give g, 2q for a point charged paxticle like the
=
electron. The precision measurement gives I g, I 2.0023 ILB for the =
electron and the deviation from 2 can be very accurately computed

by the higher order corrections of quantum electrodynamics. On the
other hand, the experimental values for the nucleons are far from the
Dirac value expected for point paxticles: g, 5.586 for the proton =
and g, = -3.826 for the neutron in units of the nuclear magneton

AN =
eh/(2mpc). This fact is ascribed to the internal structure of the
nucleon.
206 Complements
The eigenvalue of the single-particle Hamiltonian, Ho hw (-,A/2 =
+r2 /2) -(hw,12)1., -(g*p.BB12)o-,,, is obtained easily. Rewriting the

spatial coordinate as a complex variable z (x iy)/-\/-2, we defme -
I
at =
72= (
Z
*
-
0),
OZ
a=
(z +
OZ*
I '0 0
--.,f2- (z- OZ* ), (z*

bt= b +
OZ
with the commutation relations [a, at] =

[b, bt] = 1 and [a, b]
[a, bt] =
[at, b] =
[at, bt] = 0. We can then express the single-particle
Hamiltonian as
Ho =
hw(ata + btb + 1) -
Ihw,(ata
2
-
btb) -
Ig*ttBBc-z-
2
(10.12)
The operator at creates a right circular quantum of counterclockwise

rotation about z axis, and bt creates a left one. Let a non-negative
integer N denote eigenvalue of the total oscillator quanta ata+btb
an
and m denote an eigenvalue of the angular momentum 1,, at a b b. = -
Then the eigenvalues of A and btb are given by (N + m)/2 and

(N m)/2, respectively. Since they should be positive integers or zero,
-
the possible values of m for a given N are m N, N-2, N-4, -N+ =

...,
2, -N. The single-particle energy becomes

I
E,,,,,, =
hw(N+1) -
hwm -
g*ABBS,
2
I
=
hw(2n + Iml + 1) -
hwm -
g*ABB$, (10.13)
2
with Sz being either 1/2 or -1/2, and where the number of total
quanta N is set equal to N = 2n + Iml with a non-negative integer
n. For a given N, possible values of n are 0, 1 .... [N12]. The energies
of hw(N + 1), with N + I-fold
degeneracy, corresponding to the 2D
haxmonic-oscillator are independent of m conforming to N. The single-
particle energy is shown in Fig. 10.9.

In the case of no magnetic field or very weak magnetic field
(w,lwo < 1) the single-particle level shows a shell structure with

degeneracy of N + 1. The single-particle level is then in the order of
(n, m) =
(0, 0); (0, 1), (0, -1); (0, 2), (1, 0), (0, -2);... The level cross-
ing of the low-lying single-particle levels occurs at W,/WO =

V/-I/-2 Z-'
0.71 and the level with (n, m) (0, 2) becomes lower than the
= one
with (0, -1).

CIO. 1 Two-dimensional electron motion in a magnetic field 207
If wo =
0, that is, there is no confining potential present or the mag-
netic field is very high (wc.Iwo > 1), w is (nearly) equal to (112)w, and
the energy becomes E,,,,, hw,(n+(Iml -m)/2+1/2) -9*ABBS,, de-
=
pending only on n + (Iml -
m) /2. Clearly the energy becomes indepen-

dent of m for positive m. This gives rise to infinitely degenerate levels
called the Landau levels with the Landau level index n + (IM I -
m) /2.
The fact that the single-electron motion is quantized to the Landau
levels with an infinite degeneracy number in the case of no confining

potential reflects the fact that the energy is independent of the center
of the cyclotron motion. In the presence of the confining potential, the
single-particle energy increases with m as seen from Fig. 10.9.
By using the polar coordinates r and 6 the eigenfimetions of the
single-particle Hamiltonian are expressed as

2
1-1 r2
rp2 ni
n!
TP 2(n+lml)!
(n-t (r)
P
L (I ' 1)
n
(p2
-
ey-P
r
2p2
(10.14)
where Ln(c) (x) is an associated Laguerre polynomial defined by
d
L,(,o I (X) =ex-'
n! dXn
(e-XX n+a
n
F(n+a+l)
E k! (n (-X),. (10.15)
'k=O
-
k)! -V(k + a + 1)
The parity operation changes V to V + 7, so the parity of the single-
particle eigenfunction is given by (-l)'.
For the lowest Landau states (n 0, m > 0) the larger angu-
=
lar momentum state corresponds to a wave function which is more

extended from the origin:
2
(0o,jr loom (Iral + 1)P2. (10.16)
The Han-d1tonian (10.10) contains the Coulomb interaction be-
tween the electrons, which leads to interesting effects on quantum
dots. The interplay between the magnetic field and the Coulomb in-
teraction is an essential ingredient for studying the dots. The relative
importance of various pieces of the Hamiltonian depends on m*, K, g*,
and B.
For two electrons the Hamiltonian can be separated in the relative
and center-of-mass coordinates of the two electrons. By introducing
the coordinates
7
C)
Is
6
0 1 2 3 4 5 6 7 8 9 10
O)C /0)0
Fig. 10.9. The single-particle energy E,,,,, of a quantum dot as a function

of magnetic field strength. The Zeeman spin splitting of the energy is not
included.
C10.1 Two-dimensional electron motion in a magnetic field 209
I
R= ('rl +r2), Ir = I-(Irl -r2), (10.17)
-\f2- A/2
the Hamiltonian is reduced to
I
H =
hwf -
,AR + 1R21
-
'&V,, IR ,
.
C
-
g*[tBB(Sj,, + S2 ,)
2 2 2
X 1
+ hW "A'r + 21,r2 hWcl ,T Z + -
(10.18)
2 2 vF2 r
The center-of-mass motion is determined by exactly the same Hamilto-
nian as the single-particle Hamiltonian and its energy and eigenfunc-
tion are given by Eqs. (10.13) and (10.14). The eigenfunction b(T)
of the relative motion Hamiltonian may be obtained in polar coor-
dinates. By separating the angular part as b(r) u(r)e " (m = =
0, 1, 2, ...), the radial function u(r) and the eigenvalue E, are ob-
tained from the following equation
d2u 1 du V2-X 1 M2 W, 2Er

+.- -
r2 + + 7n -
U =
0, (10.19)
r dr hW r
where u(r) has to satisfy the condition that its norm, fo' drru(r)2, is
finite.
Paxticular analytical solutions are available provided Xlhw belongs
to a certain enumarably infinite set of values [1411. For a general case
we can obtain the eigenvalue by numerical integrations. The eigenvalue
may be labelled by a quantum number n 0, 1, 2,... for each m. We =
note the symmetry property of the eigenfunctions. The interchange

of the position coordinates ri and V2 leads to r -r, which is ---).
equivalent to the change of 6 # + 7r. Therefore the eigenfunction

--*
labelled by n and m receives a phase change e"' (-I)m, so that it =
is symmetric for even m and antisymmetric, for odd m. Figure 10.10
displays the energy E, of the relative motion as a function of the

cyclotron frequency. The most striking feature is the level crossing
in the lowest level (ground state). With increasing magnetic field the
ground state changes from (n, m) (0, 0) to (0, 1), (0, 2), (0, 3),
= so ..
-,
that the angular momentum m of the ground state increases one by

one. (This figure is basically the same as Fig. 10.4 which is obtained by
the variational calculation, but included here to show the significant

role of the Coulomb interaction.)
Finally we note that theimportance of the correlated motion of
electrons in a strong magnetic field has already been recognized in
the quantum Hall effect [1421. The broadening of the Landau levels
210 Complements
12-
11
10-
I f
I I
9-
8
It
*11
IA
41/ *,
-
3-
1 -
0 1
0 1 2 3 4 5 6 7 8 9 10
0)
C
/ (00
Fig. 10.10. The energy E, of the relative motion of a two-electron quantum
dot as a function of magnetic field strength. The Zeeman spin splitting of
the energy is not included. The solid curves are for states with even m and
the dashed curves for states with odd m. hwo = 1. Atomic units are used.
CIOA Two-dimensional electron motion in a magnetic field 211
is caused by scatterers such as impurities and lattice defects. The

correlations due to the Coulomb repulsion between many electrons
lead to the Anderson localization and the fractional quantum Hall
effect. Laughlin [1431 successfully used a correlated basis of Jastrow
type to describe the electron motion.
11. Nuclear few-body systems
The stochastic variational method was originally developed to solve

few-body problems in nuclear physics. In this chapter we collect a
few possible applications of the SVM for nuclear systems. The nuclear
force, due to the effect of the underlying quark structure and relativis-
tic motion, is a very complicated interaction. h addition to the spin-
isospin dependent central part, the two-nucleon interaction contains
spin-orbit and tensor forces, and L2- and (LS)2_dependent terms. The
two-body interaction designed to fit the nucleon-nucleon phase shift
and the properties of the deuteron fails to reproduce the binding ener-
gies of the three- and four-nucleon nuclei, and a three-body interaction
has to be introduced to get the correct binding energy. As the nucleons
are not structureless point-like entities, the spatial part of the nuclear
force contains a strong repulsion at short distances. These circum-

stances altogether make the solution of the nuclear few-body problem
a formidable task.
application is restricted to central forces. These schematic
The first
forces are used in simple model calculations and serve for benchmark
tests. Examples of more realistic problems are shown J-n the last sec-
tion.
11.1 Introductory remark on nucleon-nucleon

potentials
The nucleon (N) has spin and isospin degrees of freedom in addition
to the spatial one. The operator 0?.
V
of Eq. (2.3) is a product of these
factors, one acting on each coordinate of the three types:
0?.
73
=
(0,,ij (space) 0,.ij (spin)
-
Oij (isospin),

214 11. Nuclear few-body systems
where, e.g., 0,,,,j (space) (IL =

n, r. -
1, -K) is a spherical tensor

operator of rank spatial part, and
K acting in the a scalar product
(T,,,.U,,) product (-I)"A/2__K+

stands for the tensor I[T,,, x U,,Ioo, which
is defined with Clebsch-Gordan coefficients by
[Tr-1 X Ur-2]r-393 =
E < rv1jL1K2A2jK3A3 >TK,jL1 Ur-2/L2' (11.2)
IIIA2
See, e.g., [2, 70, 71, 72] for angular momentum algebra.
The NIV interaction is a typical example of the strong interaction
and is still not known in fiiU detail. There are, however, several versions
of the nuclear potential whose parameters have been determined fairly
accurately so as to reproduce two-nucleon bound state and scattering
data. The nuclear force reaches at most up to 2 fla (I fin = 10-13 cm).
An NN potential which is consistent with the scattering data is called
"realistic". The most important ingredients of the realistic potential
include central, ,tensor, and spin-orbit components.
The central part of the potential is characterized by four compo-
nents: 0-'12 ==
Ii TTT27 0'1*(7'2, and(71'72) (o-j-a2). The central potential
is thus expressed as
V12 =
V(,r) + V' (,r) -rl'72 + VCO_ (r) 01 0'2 *
+ VcTO (r) (Irl 72) (Ol U2)

* *
These components expressed by the space- (Pj'2), spin-

can also be
1+'rr72
(P12 2
and isospin-exchange (P172-
2
) operators. =
The generalized Pauli principle requires the relation
Pi2P12'P12_ = _1* (11.4)

Therefore the central force can also be written as
VI 2 =
VW H + VM (r) PJ'2 + VB H Pl' 2 + VH (r) PI'2 PI 2, (11.5)
V12 =
(V-t (r) + V` (r) 7, '72) S12) (11.6)
11.1 Introductory remark on nucleon-nucleon potentials 215
with the tensor operator
S12 3 (o-1 *
f6) (0`2 f6)

* -
U1 Cr2 *
"
/E4T (Y2 X 0212)
V 5
The deuteron consisting of a proton (p) and a neutron (n) becomes

bound through the coupling of the 3SI and 3D, waves due to the tensor
force. Here 2s+'Lj denotes the two-nucleon channel with the spin S
and the relative orbital angular momentum L being coupled to the
total angular momentum J.
However, no such coupling occurs in the
case of two protons or two neutrons because the Pauli principle (11.4)
forbids two like nucleons with L 0 or 2 to be in a spin-singlet state.

=
The interaction between like nucleons is just too weak to produce a

bound state.
The spin-orbit force has non-vanishing matrix elements only in
spin-triplet states, too. It is parametrized in the form
V12 =
(Vb (r) + Vb-r (r) -ri '72) (L S) 12 -
with
(L-S)12 r X
2h
(PI -
P2)) 2
(Ol + Cr2)
h
(r X P)' (Ol +
2 6'2)) 2
(Ol + 472))
where p =
-ih-L,
ar
hl =
(r x p), and ( x ) indicates a vector (outer)
product. The form factors of the various components are usually ex-
pressed in terms of Y(x) e-xlx and its derivatives.

=
As is evident from the above discussion, one of the characteristic

features of the realistic NN potential is its strong state-dependence,
namely it differs in the four states of singlet-even, triplet-even, singlet-
odd, and triplet-odd character, where even and odd refer to the parity
of the orbital motion.
The tensor force makes it necessary to introduce at least D-waves
in the calculations. Another important feature of the most commonly
used family of NN potentials is the strong repulsion at short distances,
say within 0.5 fi-n. This requires due care in so far as short-range corre-
lations must be properly taken into account in theoretical models. All
of these aspects of the NN interaction really make a variational cal-
culation for a nuclear system very challenging. In these circumstances
it is ratherfrequent to use some "effective" potentials which axe not

necessarily designed to reproduce the scattering data but determined
to fit some bulk properties of nuclei such as the binding energy and
the size. Although relativistic effects should be fairly large in nuclei
((V/C)2 0.08, which is estimated from the kinetic energy TF 40

- -
MeV belonging to the Fermi level), the ambiguity of the nuclear force
makes it rather difficult to extract these clearly from experimental
data.
The complexity of the NN interaction is due to the quark-gluon
substructure of the nucleon as was discussed in Chapt. 9 and to the
compositeness of mesons exchanged between the nucleons.
11.2 Few-nucleon systems with central forces
We have tested the performance of the SVM by performing model cal-

culations with different NN central potentials, e.g., the spin-averaged
Malffiet-Tjon (MT-V) potential [1441, the Volkov no. I (Vi) "super-
soft" core potential [1451, the Afnan-Tang S3 (ATS3) potential [1461,
the Afinnesota potential [1471 and the Brink-Boeker (BI) potential
[1481. Parameters of these potentials are given in Table 11.1. Some of
these model problems have already been solved to high accuracy by
various methods and therefore we can directly compare the solutions.
The results presented in this section do not include the contribution
from the Coulomb potential.
Each calculation has been repeated several times staxting from
different random points to check the convergence. The energy as a
function of the number of basis states is shown in Fig. 11.1 for the
case of 6Li with the V1 potential. The energies on different random
paths approach each other after a few initial steps, and converge to
the final solution. The energy difference between two random paths
as well as the tangent of the curves give us some information on the
accuracy of the method for a given size of the basis. The root-mean-
square (rms) radius is calculated in each step and found to be rapidly
convergent to its final value. By increasing the basis size the results
can be arbitraxily improved when needed.
The number of basis states required to reach energy convergence

increases with the number of particles, but it depends on the form of
the interaction as well. This latter property is illustrated in Fig. 11.2
for the case of the alpha-paxticle. The soft-core Volkov (Vi) potential
11.2 Few-nucleon systems with central forces 217
Table 11J. List of the parameters of the nucleon-nucleon central po-

tentials. The potential consists of a few terms; each is expressed as
Vf (IL, r)(W + MP' + BP' + HP'P'). See Eq. (11.5). For the Gaussian
potentials (VI, ATS3, Minnesota, Bj), the form factor f ([L, r) is eXp(_/tr2)'
the potential strength V is in units of MeV, and the potential range p is
in units Of fM -2, while for the Yukawa potential (MT-V) f (A, r) is defined
by exp(-[tr)/r, V is given in units of MeVfin, and IL is in units of fin-,
respectively. The Majorana, mixtures m and m' are set equal to zero in the
calculation. For realistic values of m and n see the original papers. The
parameter u of the Minnesota potential is set equal to unity.
Potential V A W M B H
MT-V 1458.05 3.11 1 0 0 0

[1441 -578.09 1.55 1 0 0 0
-2
VI 144.86 0.82 1 -
M M 0 0
-2
[1451 -83-34 1.60 1-M M 0 0
ATS3 1000.0 3.0 1 0 0 0

[1461 -326.7 1.05 0.5 0 0.5 0
-166.0 0.80 0.5 0 -0.5 0
-43.0 0.60 0.5 0 0.5 0
-23.0 0.40 0.5 0 -0.5 0
Minnesota 200.0 1.487 u/2 (2 -
u)/2 0 0
[1471 -178.0 0.639 u/4 (2 -
u)/4 u/4 (2 u)/4

-
-91.85 0.465 u/4 (2 -
u)/4 -u/4 (-2 + u) /4

-2
Bi 389.5 0.7 I-M M 0 0
[1481 -140.6 1.4-2 1 -

n MI 0 0
shows rapid convergence, while the repulsive-core ATS3 interaction
requires more basis states to get an accurate solution. The relatively

fast convergence for the MT-V potential of a strong repulsion can be
explained by the simplicity of the spin-independence of this interac-
tion.
Table 11.2 shows the SVM results, together with results of others,
for the application of the MT-Vpotential to N=2-7-nucleon systems.
We chose h2/m =41.47 MeV fi-n2. In the table the ground-state ener-
gies E and point matter rms radii (r 2)1/2 are given. The basis dimen-
sion K of the SVM listed in the table is such that, beyond it, the energy
and the radius do not change in the digits shown. For three-body sys-
tems, the solution of the Faddeev equation is known to be the method
of choice (see Sect. 5.3), but the SVM can easily yield energy of the
-65.5
-65.6
-65.7
-65.8
5 -65.9
0)
-: -66.0
W
-66.1
-66.2
-66.3
-66.4
-66.E
0 50 100 150 200 250 300 350 40(
Fig. 11.1. Convergence of the 6Li energy on different random paths. The
Volkov (VI) potential [145] is used.
-29.0
-29.5-
Minnesota
.................................................................
-30.0-
ATS3
2 -30.5- Vo-lkov
W
-31.0
MTV
-31.5
-32.0
0 20 40 60 80 100 120 140 16(
Fig. 11.2. Convergence of the alpha-particle energy for different central

potentials
Table 11.2. Energies (in MeV) and root-mean-square (rms) radii (in fin) of
N-nucleon systems interacting via the Malfliet-Tjon (MT-V) potential [14 .
The value of K denotes the basis dimension beyond which the energies and
the radii of the SVM calculation do not change in the digits shown.
N (L, S) J' Method E (T 2)1/2 K
2 (0,1)1+ Numerical -0.4107 3.743

SVM -0.4107 3.743 5
3 (0,1/2)1/2+ Faddeev[150, 1511 -8.25273

ATMS[121 -8.26 ZLO-01 1.682
CHH[1521 -8.240
GFMC[411 -8.26 0.01 1.682
VMC[1531a -8.27 0.03 1.68
SVM -8.2527 1.682 80
4 (0,0)0+1 FY[1491 -31.36

ATMS[121 -31.36 1.40
CRCG[261 -31.357 1000
GFMC[411 -31.3 0.2 1.36
VMC[1531a -31.3 0.05 1.39
SVM -31.360 1.4087 150
(0,0)0+
2 CRCG[261 -8.50 2000
SVM -8.49 150
5 (1,1/2)3/2- VMC[153]a -42.98 0.16 1.51

SVM -43.48 1.51 500
6 (0,0)0+ VMC[1531a -66-34 0.29 1.50

SVM -66.68 1.52 800
7 (1,1/2)3/2- SVM -83.4 1.68 1300
'Calculated with Coulomb potential, the Coulomb contribution then sub-

tracted perturbatively. The potential strength used in the VMC [153] cal-
culation is Vi=1458.25 and V2 -578.17 MeV, which is
= slightly different
from that used in the calculation. See Table 11.1.
same accuracy. As the MT-V potential is a preferred benchmark test

of the few-body calculations, there are numerous solutions available.
Table 11.2 includes a few of the most accurate results. The nice agree-
ment for the four-nucleon case corroborates the fact that the SVM is
as accurate as the direct solution of the Faddeev-Yakubovsky (FY)

equations [1491, the method of the Amalgamation of Two-body corre-
lations intoMultiple Scattering (ATMS) [121 or the variational Monte
Carlo (VMC) [153] and the Green's function Monte Carlo (GFMC)
[411 methods. The basis used in the Coupled-Rearrangement-Channel.
Gaussian-basis (CRCG) variational method [26] is similar to that of
the SVM, but the Gaussian parameters follow geometric progressions.
The fact that the basis size needed in the SVM for both the ground
and first excited states of the four-nucleon system is much smaller
proves the efficiency of the selection procedure. See Chap. 4. The re-
sults of the VMC calculation for the five- and six-nucleon systems are
also in good agreement with the results of the SVM- The MT-V
po-
tential has no exchange term; therefore, unlike nature, it renders the
five-nucleon system bound, and the nucleus tends to collapse as the
binding energy increases with the number of particles.
Neither the MT-V nor the V, potential contains the full set of ex-
change components of the central potential, so the calculation becomes

rather simple. The Minnesota potential, however,
contains space-,
spin-, and isospin-exchange operators, and is considered to be more
realistic. In fact it has been successfully used in microscopic cluster-
model calculations [28, 29, 301. An extensive calculation with this po-
tential was carried out for N=2-6 nucleon systems in [31], where all
possible spin and isospin configurations were allowed for and all an-
gula.r functions that give non-negligible contributions were included.
The agreement with experiment was surprisingly good. The energies
and the'radii of the triton and the alpha-particle converge to realistic
values. The Minnesota potential, correctly, does not bind the N=5
system, but it binds 6He and slightly overbinds ILL The radius of
6He is found to be much larger than that of 4He, consistent with the
neutron-halo structure of 6He [29, 301. The SVM calculation of [311 has
made itpossible to test for the first time the Minnesota force without
assuming any cluster structure or restricting the model space by any
other bias.
The last example is the application to Efimov states [1541 to show
how well the correlated Gaussian basis describes the few-body asymp-
totics. A three-body system with short-range forces may have several
bound states. If the two-body subsystem has just one bound state
whose energy is (close to) zero, the three particles can interact at long
distances and an infinite number of bound states may appear in the
three-body system. These "Efimov states" are extremely interesting
from both experimental and theoretical points of view because of their
distinctive properties. These states are very loosely bound and their
wave functions extend far beyond those of normal states. By increasing
the strength of the two-body interactions these states disappear.

A system of three identical bosons of nucleonic mass is consid-
ered for simplicity. The potential between the particles is taken as the
Poeschl-Teller interaction because this potential is analytically solv-
able for the two- body case, and therefore the two-body binding energy
can be easily set to zero;
625.972 1251-943
V(r) = in MeV. (11.10)
sinh(l.586 r)2 cosh(I.586 r)2
Here r is in units of fi-n. A spatially very extended basis is needed to
represent the wealdy bound states. In this
predefined case we use a
basis. The predefined basis is created

diagonal ele- by using only the
ments of the matrix A. These diagonal elements are taken as geometric
progression:
1
A,-, -
('
(i =
1, ..., n),
1))2
-
(aoq6
1
A22 -
(j n). (11.11)
( ao q(j-'))2
10
This construction defines an n 2-dimensional basis corresponding to
(i, A =
(I, I),-, (n, n). The parameters are chosen as n 20, ao = =
0.1, q0 = 2.4. One can find the first three Efimov states with this
basis size. The ratios of the energies of the bound states follow the
Ei+IlEi =
expJ-2vJ rule [1541. By increasing the basis size one
can reveal more bound excited states. Note that the value of aoqo'-'
roughly corresponds to the spatial extension of the basis. The present
basis in this example goes out up to aoqon-I = 0.1 x 2.419 = 1674990
(fin).This extension is necessary to get the third bound state. For
a comparison, to calculate the ground-state energy of -4.81 MeV it
is enough to choose n 7 and the basis covers only the [0, 101 (fi-n)
=
interval. The rms radii of the ground, first and second excited states
are about 1.5, 25 and 6000 fin, respectively. As shown in Fig. 11.3,
these facts illustrate the tremendous spatial extension of the Efimov

Ground state
.0
.0
0
10.0
Efimov state
0.1
0.0
M
Fig. IXM, Wave fimetion amplitudes of the ground state and the first
Efimov'stAte. Note the different scale (in fin) of the lengths of the Jacobi
coordinates x and y.
11.3 Realistic potentials 223
states. These results indicate that the extended correlated Gaussian

basis gives a practical means to describe the asymptotics characteris-
tics of the Efimov states. See Complement 11.4 for the
possibility of
Efimov states in the 12C nucleus.
11.3 Realistic potentials
The trial function is given as a combination of the correlated Gaus-

sians:
'O(LS)JMTMT(-'C 14) 1
=
Ajexp(- 1.+,Ax) [OL (X)
2
X XSJ jM TITMTli (11.12)
where x =
jX1i ... XN-11 is a set of relative (Jacobi) coordinates,
the operator A is an antisymmetrizer,, XsMs is the spin function, and
nTmTis the isospin function. The matrix A is an (N -
1) x (N -
1)
symmetric, positive-definite matrix of nonlinear variational parame-
ters. The function angular part of the wave
OLML (x) represents the
function. It is taken as a vector coupled product of partial waves
OLML (X) :`
i3)]L123 I LML This is re-
[[[Yli 41) X Y12 (ZAL12 X Yls (X3 ... .
ferred to as the method of partial-wave expansion. The spin and

isospin functions are also successively coupled, for example XSM, =
X(E12SI23 )SMs...
=
[[[XI/2 X XI/21SI2 X X1/21S1,23 ...J,sms For the triton,
we used the partial-wave channels with (11, 12)L, where 1i < 2 and
L =
0, 1, 2. For the alpha-particle, allpartial waves ((11, 12) L 12 13) L 1
satisf3 g the condition 11 + 12 + 13 :! 4, 1i :! _ 2, 111 121 < -
L12 ! 11 + 0, 1, 2 have been tried. As the triton has

12 and L -
J =
1/2 and T 1/2, there are three spin and two isospin chan-
=
nels; (S12)S (0)1/2,(1)1/2,(1)3/2 and T12 0, 1. For the alpha- =
particle (J 0, T 0) there are six spin and two isospin chan-

=
nels unless very small isospin mixing is taken into account; (S12 S123) 7
=
(0, 1/2)0,(0,1/2)1,(1,1/2)0,(1,1/2)1,(1,3/2)1,(1,3/2)2 and (T12,
T123) (0, 1/2), (1, 1/2).
=
The spin, isospin and orbital angular dependence of re- momentum

alistic nucleon-nucleon interaction is extremely important. To extend
the application of the SVM calculations to nuclear systems interacting
with realistic potentials, in addition to the previously applied random
selection of the nonlinear parameters of the correlated Gaussian, a ran-
dom selection of the spin and orbital components of the wave function
224 11. Nuclear few-body system
is introduced. The main motivation for the random selection is that

the numbers of channels and nonlinear parameters are prohibitively
large, therefore the calculation of all of the matrix elements and di-
agonalization including potentially important basis states is out
all
of the question. In the case of the alpha-particle, for example, there
are 12 spin-isospin channels and 32 partial-wave channels even in our
truncated-partial wave expansion. Just to show up an alternative, it
is worthwhile assessing what dimensions a deterrni-ni tic optimization
would require. The simplest choice of the nonlineax parameters of the
Gaussian basis is to use a diagonal matrix A. One of the best deter-
ministic choices for the diagonal elements is a geometric progression
[251. To reach good accuracy, at least five values for each of AJ-1 A22 7
3
and A33 have to be used, requiring 5 125 functions in a given
=
channel. The spin-isospin and space paxt, therefore, would result in a

basis size of about one hundred thousand (2 x 125 x 12 x 32) in such
a "direct" calculation for the alpha-particle.

The literature is very rich in papers devoted to few-nucleon prob-
lems [155, 156, 157, 12, 25, 158-1611. For N 3 nuclei it is possible
=
to include enough channels and an exact Faddeev calculation can be

performed, but other methods give essentially the same results. For
N = 4 nuclei some of these methods become too complicated, and the
results of the existing calculations agree only within a few hundred
keV.
To test the SVM we used two, from the point of view of
spatial
form, rather different interactions. The first, the Argonne potential
(AV6 and AV8) [1621, is a well-behaved smooth function. The sec-
ond, the Reid potential (M) [1631 is more singulax and includes a
linear combination of exp(_/tr2)/,rk-type terms. potential The latter

can certainly cause some problems. The AV6 potential includes only
central and tensor components and serves as a good test case for the
inclusion of non-central forces. The AV6 and AV8 potentials axe used
without Coulomb potential, while in the case of the RV8 potential the
Coulomb interaction is added.
To keep the number of nonlinear parameters low, we restricted the
matrix A to be diagonal in one of the possible Jacobi coordinate sets.
In the of the three-nucleon system we have only one possibility
case
of the Jacobi coordinate, (NIV) + N. For the alpha-particle we use

both possible coordinate sets, (3N) + N ("K") and (2N) + (2N)-
type ("IF) Jacobi coordinates. Note that the basis function is fuHy
antisymmetrized. This choice is also dictated by physical intuition as
it is natural to consider 3H+P 3He+n and d+d-type partitions in the

,
alpha-paxticle. Test calculations show that tbis choice already gives

satisfactory results.
As is shown in Table 11.3, the convergence is relatively fast both
for the triton and the alpha-particle with the AV6 and AV8 potentials,
but in the case of the RV8 potential, due to its singular nature and
stronger repulsive core, the convergence is slower. About 50 basis
states for the triton and about 200 basis states for the alpha-particle
give fairly good binding energies. The energy of the alpha-paxticle

with the basis size of 200 is already within 0.5 MeV of its final value.
It is very satisfactory that the SVM has succeeded in reproducing a
reasonably good energy with just 400 basis states.
Table 11.3. The convergence of energies (in MeV) for the triton and the
alpha-particle for different potentials, AV6 [1621, AV8 [1621, and RV8 [163].
K is the basis dhnension.
311 'He
K AV6 AV8 RV8 K AV6 AV8 RV8
25 -6.63 -7.36 -6-53 100 -24-37 -25.15 -23-35

50 -7.04 -7.69 -7.41 200 -25.05 -25.50 -24.15
75 -7.11 -7.74 -7.54 300 -25.28 -25.60 -24.35
100 -7.15 -7-79 -7.59 400 -25.40 -25.62 -24.49
The results of the SVM and other calculations compared in

are
Tables 11.4 and 11-5. We show and compare the results not only
for the energy but for the rms radius and average kinetic and po-
tential energies as well. The results remain the same by repeating
the calculations several times staxting from different random values.
The comparison for the triton with RV8 demonstrates that the SVM
resultsare in very good agreement with the Faddeev results. Since
Faddeev calculations are considered exact for three-body systems (see

Sect. 5.3), this agreement confirms that the basis selection of the SVM
has been performed efficiently even for the singular potential of RV8.
The correlated hyperspherical harmonics (CHH) expansion method
(see 5.2) also agrees with the SVM for the triton. The results of
Sect.
the SVM and GFMC are very close to each other. Except for the case
of the alpha-particle with AV6, the energies of the SVM axe within
the statistical errors of the GFMC, though the expectation values of
Table 11.4. Energies (in MeV) and radii (in fin) of the triton calculated by
different methods and with different interactions. (T), (V), and (V6) are the
expectation values of the kinetic energy, the total potential energy, and the
potential energyof the central and tensor components, respectively. (, 2) 1/2
denotes the root-mean-square radius. The value of PL shows the percentage
of the component with the total orbital angular momentum L.
SVM GFMC Faddeev VMC CHH
AV6
(T) 44.8
(V6) -51.9 -52.0(3.00) -43-7(1-0)
(r 2 1/2 1.76 1.75(0.10) 1.95(0.03)
PS 91.2
PP < 0.1
PD 8.7
E -7.15 -7.22(0.12)a -7.15' -6.33(0.05)'
AV8
(T) 46.3
(V6) -52.9
(VLS) -1.2
(T 2)1/2 1.75
PS 91.1
PP < 0.1
PD 8.9
E -7.79 _7.79d -7.79'
RV8
(T 52.2 54.0(0.20) 52.2

(V) -59.8 -62.0(0.20) -59.8
(T 2)1/2 1.75 1.68(0.07) 1.76

PS 90.3
PP < 0.1
PD 9.7
E -7-59 -7.54(0. 10)b -7. 59b -7.44(0.03)' -7-60'
aRef. [1641 bRef. [1651.

.
'Ref. [1571. dRef. [1591. 'Ref. [1611.
Table 11.5. Energies (in MeV) and radii (in fin) of the alpha-particle by
different methods and with different interactions. The isospin mixing due
to the Coulomb potential is not taken into account. See also the caption of
Table 11.4.
SVM GFMC FY vMC CHH
AV6
(T) 100.1
(W) -125.4 -122.0(3.0) -122.0(1.0)
(r 2)1/2 1.49 1.50(0.04) 1.50(0.01)
PS 84.3
PP 0.5
PD 15.1
E -25.40 -25.50(0.20)a -22.75(0.01)a
AV8
(T) 98.8
M -124.4 -124.20(l.0)
(W) -121.5
(VLS) -2.9
(r 2)1/2 1.50 1.51(0.01)

PS 85.5
PP 0.3
PD 14.2
,d
E -25.62 -25.75(0.05) -25.31- -25.60'
RV8
(T) 111.7 109.2(0.20)

(V6) -139.1 -137.5(0.20)
(VLS) 2.1 2.45(0.23)
(KOUI) 0.75 0.71(0.02)
(T 2)1/2 1.51 1.53(0.02)
PS 84.1
PP 0.4
PD 15.5 15.5(0.20)
E -24.49 -24.55(0.13)b -23.79f -24.01e -23-9c
aRef. [1641. bRef. [165]. cRef. [1571 dRef. [1591

. .
eRef. [1611. fRef. [1551.
the kinetic and the potential energies are somewhat different. Both of
VMC and GFMC have a statistical error involved in the Monte Carlo
integration. Though the error bar of VMC is influenced by the choice
of the trial function as well, the error of GFMC is considered purely
statistical in the limit of infinitesimal time step (see Sect. 5.1).
Variational calculations were considered to be inappropriate for
realistic potentials, as they might not be able to

reproduce large the
cancellation between the kinetic and the potential energies [1581. Our
calculation shows that, with the careful optimization of the nonlinear
variational parameters, this is not the case. We can obtain accurate
energies even in the case of the RV8 potential.
The experimental energies of the triton and the alplia-particle are
-8.48 and -28.295 MeV, respectively. We see that the calculation
using the realistic potentials underbinds; about 0.9 MeV for the triton
and about 3.8 MeV for the alpha-particle. This indicates that the
binding energies of few-nucleon systems cannot be accounted for in
two-nucleon potential models but call for other mechanism to get more
attraction. One possibility for the mechanism is a three-body potential
which can be produced by two-pion exchanges among three nucleons.
Once we have a basis for a given potential, one naturally expects
that the same basis may give fair ground-state energy for other po-
tentials of similar nature as well. We have checked if it is really the
case. The basis optimized for the RV8 potential gives excellent results
(within 100 keV) for AV6 and AV8 as well. Due to the singular nature
and stronger repulsive core of RV8, however, the basis optimized for
AV8 gives about 500 keV less binding for the alpha-particle with RV8
than the basis optimized for RV8 itself. This result is still not so bad
and can be easily improved by refining the nonlinear parameters as
explained in Sect. 4.2. Therefore, one does not have to look for an
entirely new basis set for another interaction, but can use the same
basis for a given system, with some "fine tuning" if necessary.
It is obviously important to calculate accurate binding energies of
light nuclei with realistic forces. For example, the two-neutron sep-
aration energy of 6He is about I MeV, so that a few hundred keV
less binding is thought to change its neutron "halo" structure signifi-
cantly. One of the advantages of the SVM is that it is relatively easy
to extend it to N 5-, 6-, 7-nucleon systems [31]. The low dimension
=
of the bases required to solve the N 3- and 4-nucleon problems

=
confirms that the SVM is suitable for treating larger nuclei with re-
alistic forces. The formalism and the computer code itself is general
[311, so the applicability is mostly limited by the memory and speed of

the available computer. There is no dffficulty in including three-body
forces if necessary.
utility of the global vector representation of Eq. (6.4)
We show the
in describing the angular part of the system interacting via realistic
potentials. We calculated the ground-state energies of the triton and

the alpha-particle by using the AV6 interaction [162]. The results by
the global vector representation and by the partial-wave expansion
are compared to each other in Table 11.6. The model space allows
for natural parity states only. The partial-wave expansion is restricted

for partial waves with orbital angular momentum 0, 1 and 2. These
partial waves give a good description of the ground states of these
nuclei [321. The table shows that one can reach the same results with
the global vector representation as with the partial-wave expansion
even in the more complicated case when the tensor force couples the
different orbital angular momenta. The case of the alpha-particle is a
nice illustration of the main advantage of the

global vector representa-
tion over partial-wave expansion: Even in the expansion truncated
the
to the natural parity states, one needs 24 partial-wave combinations
due to the large number of combinations of the individual partial
waves, while in the global vector representation only 4 sets of (K, L)
((K, L) (0, 0), (1, 0), (0, 2), (1, 2)) were used to reproduce the same
=
result [1661. By increasing the number of particles the partial-wave ex-

pansion would contain a prohibitively large number of combinations.
The number of sets (K, L), on the contrary, does not depend on the
number of particles. Only the number of parameters ui increases (lin-
early) by adding particles to the system.
Table 3-1.6. Total energies of the ground states of the triton and the alpha-
particle [1621. Only natural parity states have been
with the AV6 interaction
taken into account for the orbital motion. Two types of angular functions
are used; the global vector representation (GVR) and the paxtial-wave ex-
pansion (PVVE). K is the basis dimension.

3
11 411
Method K E (MeV) Method K E (MeV)

GVR 100 -6.80 GVR 300 -23.29
PWE 100 -6.80 PWE 300 -23.27
230 Complements
Complements
11.1 Correlations in few-nucleon systems
To elucidate the significance of the correlation we calculate the bind-
ing energies of the triton and the alpha-particle by using the corre-
lated Gaussians and the correlated Gaussian-type geminals. The spin-
isospin parts of the triton and the alpha-particle wave functions are
considered totally antisymmetric and, as the nucleon is a fermion,
their spatial parts must be totally symmetric. The spatial motion in
the ground state of the N 3- or 4-nucleon system may thus be
=
treated as motion of bosons. We obtain variational solutions for the

triton and the alpha-partile interacting via a spin-isospi-n independent
two-nucleon potential.
The orbital wave function is assumed to be given by combinations,
of the correlated Gaussians with L 0, see Sect. 7.2: =
Tf
k=1
CkS eXP (_2 ,7c A k X
Here x is the Jacobi coordinate set and S is a symmetrizer. Each of the

positive-definite symmetric matrices Ak contains nonlinear variational
parameters, three for the triton and six for the alpha-axticle. The
paxameters can be determined by the SVM. The calculation was done
with the code of [811. The number K of the basis functions has been
chosen so as to reach convergence.
As another type of trial function, one may try the following corre-
lated wave function

1V 3
Tl'= F P =
Ffj (-v)'Iexp(
IF
-
1Vri2).
2
The function P, a product of the nodeless harmonic-oscillator func-

tions (Gaussians), describes the independent single-particle motion,
whereas the function F takes into account the correlations. The
Jastrow-type correlation factor [471 is taken, in a single Gaussian form,
as
IV
F= fj (I+aexpj-b(rj _
ri)21
i>i=l
KN
E ak exp (__2 i;BkT')

k=1
CILI Correlations in few-nudeon systems 231
with KIv =
1, where ak is given by an integer power of a,
2 2
and Bk is an N x N symmetric matrix of type, 2b Ei uiiii, where

ui is an N-dimensional vector which has only two nonzero elements.
Therefore the trial function TJ- becomes a combination of the corre-
lated Gaussian-type geminals (6.26) with n = 0 and R =

0, and the
needed formulas for matrix elements involving such special geminals
were in that section. Three parameters v, a, and b can be varied
given
to minimize the energy. Note that Tf is totally symmetric and that
its dependence the center-of-mass coordinate x1v can be factored
on
out from the intrinsic motion. This makes it possible to calculate the
intrinsic energy.
We also consider a special type of the correlation factor which
contains only the pair correlation

K N
F2 = I + 1: ak
k=1
exp bk (ri _,rj)2)
Here the pair correlation can be better described with combinations
of K Gaussian functions than with Eq. (11.15). It is expected that the
Jastrow-type correlation becomes more useful in cases where various
correlations of more than two particles play a major role. The nonlin-
ear parameters bk and v and linear parameters ak are determined to
minimize the energy.
We use three potentials, the MT-V [1441, V, [1451 and B, [1481*
potentials listed in Table 11.1. The space exchange operator PI of the
potential can be set equal to unity because of the bosonic symmetry
of thespatial wave function. The energy is thus independent of the
Majorana mixture parameters.
Table 11.7 summarizes the results. The result for N = 3 and 4
[311 calculated with the correlated Gaussian basis is known to be in

good agreement with other accurate calculations available in the lit-
erature. It is interesting to note that, except for the alpha-article with
the MT-V potential, the pair correlation with K 2 terms-already =
gives lower energy than the Jastrow-type correlation described with a

single Gaussian function. (The alpha-paxticle has more compact struc-
ture than the triton. Because of this property, the strong repulsion of
the MT-IT potential has to be fully taken caxe of in order to obtain
the ground-state energy particularly for the alpha-particle. The cor-
relation factor of Jastrow type appears to be more suited than just
the pair correlation of Eq. (11.16) in this regard.) The small difference
232 Complements
between K = 2 and K = 3 calculations indicates that the pair correla-

tion can well be described with two Gaussians. A better description of
pair correlation appears to be important. The trial function with
the
K 0 is nothing but a product of the single-particle wave functions
=
and contaiins no correlation. This function misses lot of energy. This

a
becomes more evident in singular

more potentials such a.-, the MT-
V potential. Note that there issignificant energy difference between
a
the CG asid CGG results with the MT-V potential. This suggests that
at least a triple particle correlation, e.g.,
IV
E exp( -
b(Tj -
Tj)' -
b(Tj -
Tk )2 -
b(Tk _
T,)2), (11.17)
k>j>i=l
has to be taken into account in the correlation factor for this poten-
tial to obtain accurate energy. This problem was investigated in [1671,
which takes into account the two-, three- and even more-particle cor-
relations and shows the importance of triplet correlations in producing
accurate energy.
Table 11.7. Energies of the triton and thealpha-particla in the correlated

Gaussians (CG) of Eq. (11-13) and the correlated Gaussian-type geminals
(CGG) of Eq. (11.14), which axe chosen to represent correlations in one of
two forms. The correlation factor F of CGG is defined by Eq. (11.15) for
Jastrow and by Eq. (11.16) for the number K of pair correlations. See Table
11.1 for the nucleon-nucleon potential. The energies am in units of MeV.
CG CGG
Jastrow K = 0 K=I K=2 K=3
MT-V [1441
triton -8.25 -6.00 unbound -5.78 -7.11 -7.27
alpha -31.36 -29.04 -6.41 -27-03 -28.47 -28.93
V1 [1451
triton -8.46 -6.99 -6.66 -6-99 -8-11 -8-11
alpha -30.42 -29.14 -27.92 -29.13 -30-06 -30-13
B1 [1481
triton -11.64 -9.95 -7-00 -9.87 -11.20 -11.20
,alpha -38.34 -36.73 -28.19 -36.02 -37.60 -37.60
When the interaction has spin-isospin dependence, the correlation

factor is also expected to be state-dependent. The evaluation of matrix
CII.2 Convergence of partial-wave expansions 233
elements with such correlation factors would then be more involved.

Refer to [1681 for a recent development on the application of the cor-
related wave functions to heavier nuclei.
11.2 Convergence of partial-wave expansions

Here we examine the convergence of the partial-wave expansion (PVVE)
in obtaining bound-state solution for
few-body systems. We consider
examples of three-body systems consisting of p, n, and n (triton) or
of p, n, and a hyperon A. The A is a neutral baryon with mass
mAO 1116 MeV. It has spin 1/2, isospin 0 and strangeness -1. The
=
NA interaction is not strong enough to form a bound state between

a nucleon and a A. When one or few A's are embedded in a nucleus,
they form a bound system called a A hypernucleus. The simplest hy-

pernucleus is the (p, n, A) system 3H,
A
which is called a hypertriton
and its ground state has P 1/2+. =
The binding energy of 3A H is 2.4 MeV from the three-particle

threshold. As neither subsystem of pA nor nA forms a bound state, A
couples weakly with p and n. In fact the A separation energy of 3A 1-1
is only about 0.2 MeV. Most of the binding energy of 3A 11 thus comes
from that of the pn subsystem, namely the ground-state energy is just
0.2 MeV below the threshold of deuteron (d) +A. It would be natural
to attempt to describe the ground-state wave function Tf in terms of
the coordinates xI =
rp -
r, and X2 =
(I'P + Tn)/2 -
rA.
simplify the analysis we assume that the pn subsystem has the
To
same spin and isospin state as d, that is spin 1 and isospin 0. This
assumption, known to be a rather good approximation, leads us to

the condition that Tf must be an even function of xI. In addition
we assume that the total orbital angular momentum of the system is
zero, and that the three 1/2 spins are coupled to J 1/2. Under these =
assumptions Tf can in general be expressed as follows:
Tf =
f (XI I X2, XI *
X2) [xi (pn) x X.L

2
(A)].LM, 2
where the space part f is a function Of X1, X2 and X1'X2. For Tf to be

an even function of xI, f has to be an even function Of Xl'X2-
The Hamiltonian of the system is given by
H=T+V
h2 a2 h2 192
2
---
+ Vpn + VpA + VnAI

21L, ax, 21L2 C9X2 2
234 Complements
where IL1 =
MN12, and A2 2MNMj1(2MN + IVIA). Only the central
=
potential is included in the Hamiltonian. The pn interaction VP" in

the triplet even state is taken from the Minnesota potential [147] with
u = 1 (see Table 11.1). The NA central potential is assumed to have
the form of Minnesota type
I
VNA= VR+I(I+Po')Vt+-(I-P
2 2
-)V
x
(lu+ 1(2-u)P')
2 2
(11.20)
with
2
VR = VORe-ISRT2, Vt =
-Vfote-"" 1
V, -Vfoe-S' 2. (11.21)
The u value is set equal to 1.5. We assume that the pA interaction

and the nA interaction the same, though there is some evidence
are
that they are hi fact slightly different. The potential parameters are
determined so as 3, 4 binding-energy data: The strength

to fit A =
parameters in units of MeV axe VOR 200.0, Vot 109.8, and Vo,
= = =
121.3, and the ranges in units of h3j-2 are r,-,R 1.638, nt 0.7864, and = =
ms =
0.7513, respectively. This NA potential will be used in the next
Complement to calculate the binding energies of s-shell hypernuclei.
The equation of motion for f (XIi X27 XI *X2) is obtained by substi-
tuting Eq. (11.18) into the Schr6dinger equation (H E)Tf 0 and -
=
integrating over the spin coordinates. Owing to the symmetry prop-

erty of the wave function and the assumption of VpA VnA, we only =
need to evaluate the spin matrix element of, say VnA. The matrix el-
ement of PI exchanging the spin coordinates of the nA pair can be
calculated by rewriting the spin function of Eq. (11.18), expressed in
(pn)A coupling order, to that of the p(nA) order by using the Racah
coefficient in unitary form (see Complement 6.3):
[Xj(pn)xX.j(A)j.iM= 2 2
1: U(-!-!.!1;IS)[X.L(p)xXs(nA)].jM
2222 2 2
S=0'1
Vf3_ 1
[X.L (p) x Xo (nA) I.LM + [Xi(p)xXj(nA)jjM. (11.22)
2 2 2 2 2 2
Then the desired matrix element is
&1(pn) xx.L(A)ji].LMjP'j[xj(pn)
2 2 2
x X.L(A)jfljm)
2 2 2
C11.2 Convergence of partial-wave expansions 235
1 .1 .1
1) S+I- U(I:2222
1.1.1; IS) U( 2222 .1; is) .1.
2
(11.23)
S
The equation of motion for f reads as
h2 a2 h2 a2 1
21L1- axi 2 2A2 19X2 2

-
+ Vp. (XI) + VpA (I X2 +

2
XII,P
I
+V,A(IX2 -
2
x, 1, P) -
E f (XI i X2 i XI'X2) =
0, (11.24)
with the boundary condition that f has to vanish for large xl_ and X2-
Here VpA(r, P') is given by
VpA (r, P') VR + 1Vt + 4

4 3V,) (lu 2
+ 1(2 u)P')
2
-
(11.25)
One has to note that P' exchanges rp and rA, which induces the
transformation of the coordinates: x, --+ x1/2 X21 X2 --+ -3x,/4 - -
X2/2. The potential V,,A(r, P') is defined in exactly the same manner,
where pr now transforms x, and X2 to xl_/2 + X2 and 3xj/4 -
X2/2,
respectively.
We obtain the ground-state energy and the solution of Eq. (11.24).
The variational trial function may be chosen as a combination of the
correlated Gaussians (CG) of Eq. (7.3):
f(XliX2iXI*X2) =
ECkfFAk(X) +FPAkP(X)II (11.26)
k=1
1 0
where the second term with P
(0 -1 ) assures that the trial
All A12
function is an even function of xj, Le.., for A PAP
A12 A22
All -A12
becomes PAP
-A12 A22
It is interesting to analyse the paxtial-wave contents of the solution.
For this purpose we calculate the quantity
2
C1 =(f(XliX27Xl*X2)lpllf(Xl7X27Xl*X2)) (11.27)
with
PI =
I[yl(XI) X YI(X2)j00)([YI(Xl) X Yl(X2)1001- (11.28)
236 Complements
The result is given in Table 11.8. The admixtures of D, G andhigher

waves axe small and the PWE is very effective in the present case. The
root-mean-square (rms) radius of the hypertriton is calculated to be

4.87 fm. The distance between the proton and the neutron is 3.61
rms
fm, which is slightly smaller th an the corresponding distance (3-90 fin)

obtained for the free deuteron, and the rms distance between the A
and the center-of-mass of the proton and the neutron is as large as
9.98 fin. Therefore the hypertriton is barely bound by polarizing the
deuteron slightly in order to increase the binding between the deuteron
and the A.
Table 11.8. Partial-wave decomposition of the hypertriton wave function
1 0 2 4
C2 0.99913 0.79 X 10-3 0.6x 10-4
The trial function (11.26) gets higher paxtial-wave contributions

from the cross term X1'X2. The
parameters A12 determine the appro-
priate weights of various partial waves. It is of course possible to solve
Eq. (11.24) by expanding f in partial waves:
f(X1iX2iXI'X2) "'
E A (XI X2) [YI (ill)
i
X YI (F2)] 00 .29)
I=even
Then Eq. (11.24) reduces to a coupled equation for fI(X1, X2)'s. Refer
to Appendix A.5 for a method of evaluating the matrix element with
this type of basis functions in the case that A (XI X2) is expanded as 7
2
-bX2
a combination of Gaussians (XlX2)le "
1 2.Table 11.9 shows the
energy convergence in the partial waves included in the calculation.

fin conformity with Table 11.8 the inclusion of S wave alone is a good
approximation to the hypertriton and bigher paxtial waves give modest

contributions to the binding energy.
Noncentral forces necessitate the inclusion of higher paxtial waves
and the PWE will be slowly convergent in general. Even in the case of
central potentials the energy convergence in the PWE in one particu-
la,r Jacobi coordinate set depends on a system and an interaction. To
examine the contribution of higher partial waves in the case of central
forces we consider another example of the triton (nnp), which is more
tightly bound than the hypertriton. The energy has been calculated
C11.2 Convergence of partial-wave expansions 237
Table IL9. Convergence of the hypertriton energy in the partial-wave

expansion (PWE). The CG indicates the result obtained with the correlated
Gaussian of Eq. (11.26). The energy is from the three-particle threshold. The
deuteron energy is -2.202 MeV.
PWE CG
1.. 0 2
E (MeV) -2.305 -2.358 -2-378
by using the wave function of type (11.29), where x, is the relative

distance vector between the two neutrons and X2 is the relative coor-
dinate between the proton and the center-of-mass of the neutrons. The
two neutrons are spin-singlet state. Two potentials
assumed to be in a
are used: One is the soft-core Volkov (VI) potential [1451 and the other
is the repulsive-core ATS3 potential [1461. Table 11.10 compares the

PWE results with other methods, the hyperspherical harmonics (HH)
expansion method (see Sect. 5-2) and the CG calculation. The PWE
gives fairly fast convergence for the V, potential and the inclusion of
S and D waves already produces the energy that is close to the ac-
curate value by the HH method. However, for the ATS3 potential the
convergence is rather slow and the partial waves of up to I 6 are not =
sufficient to get the energy which is close to the value calculated with
the CG basis (11.26). (The two neutrons are assumed to be in a spin-
singlet state, but, as the ATS3 potential is spin dependent, to allow
them to be in a spin-triplet state as well leads to lower energy than
the result of Table 11.10. The CG calculation including this possibility
gives -8.765 MeV (the rms radius is 1.67 fin) in K 100 dimension, =
which is in perfect agreement with the energy [170] obtained by the

Faddeev method (see Sect. 5.3).
Here we explain why the ATS3 potential leads to the slower con-
vergence than the V, potential. The matrix element for the central
potential between the two neutrons vanishes for the functions with
different I values, so the potential is not responsible for the mixing-
in of different partial waves. The potential between the neutron and
the proton, however, induces the I-mixing. To see this, we note that
the latter potential has the form V(Iax, + bX21) (for the triton case
a =
:E(1/2), b 1, but it is extended to a general case.), which is a
=
function Of X17 X2 and the angle,& between x, and X2. The potential
can be expanded in terms of the multipole operators as follows
238 Complements
Table 11.10. Convergence of the triton energy in the partial-wave expan-

sion (PWE). The two neutrons are assumed to be in a spin-singlet state.
See Table 11.1 for the nucleon-nucleon potential.
V, [1451
PVVE HH[1691
1niax 0 2 4 6
E (MeV) -8.005 -8.390 -8.447 -8.460 -8.4647
ATS3 [146]
PVVE CG
Imax 0 2 4 6
E(MeV) -2.948 -6.215 -7.210 -7.484 -7-616
00
V(Iax, +bX21) =
EV1(XI,X2)P1(COS'6) (11.30)
1=0
with
7r
=21+1
VI (X1 i X2)
2 fo V(Iax, + bX2 I)PI (COS 6)Sin'0d#- (11-31)
For example, for the Gaussian potential of V(r) = Voe-t2T2 we have
(see Eq. (6.53) or Eq. (A.81))
VI (X1 X2)
i
2
-g (a 2 xj+b
2
2)
= Vo (21 + 1) c(l, ab) il (2 M2 jabjX1X2) X2, (11.32)
1
where E(I, ab) = I for ab > 0, E(I, ab) =
(- 1) for ab < 0. Since the
Legendre polynomial PI(cosO) is proportional to the tensor product

[YI(i-1) x Yj(i2-)joo (see Eq. (6.54)) and has non-vanishing matrix el-
ements between the functions of different partial waves, we find that
the various multipole components contained in the potential bring
about the mixing of the higher partial waves. The relative strength of
the multipoles thus determines the extent to which each partial wave
is contained in the solution. For a long-ranged potential (small p)
the magnitude Of V1 (X1 X2) is determined by the factor IL21/ (21
7 1) 11, -
so that only low partial waves give main contributions to the solu-
tion. However, this is not the case for a short-ranged potential. The
shorter the is, the larger the contribution of thehigher
potential range
inultipole components. E.g., in the extreme case of V(I ax, + bX2 1) -
C11.3 Quark Pauli effect in s-shell A hypernuclei 239
J(ax, + bX2), by putting Vo (fk2/7)3/2 and letting 1L oc in Eq.

=
,
(11.32) We get VI(X1, X2) ((21+ 1)/(47))J(Ialxl jbjX2)1(jabjXIX2),

- -
which indicates that the multipole strength is equally strong for arbi-
trary 1. The result of Table 11.10 canbe understood in this way from
the fact that the ATS3 potential contains the strong short-ranged
repulsive-core part.
Clearly the matrix element for the central potential between the
neutron and the proton vanishes for the functions of diffrent I values if
they are expressed in terms of the different Jacobi coordinate set where
the first coordinate is chosen to the relative coordinate between the
neutron and the proton. This consideration leads us to the following
ansatz for the solution
f (X(I), X(1)) [Y,(X(,)) YJ(X(1)) 100

1 2 1
X
2
+ fl(2) (X (2), (2)) [yJ(X(2))

1
X
2 1 1
X y1I
(X(2))j
2
00
+ fl(3) (X (3), (3)) [yJ(X(3))

1
X
2 1 1
X Yj(X 2(3)) 100 (11-33)
where x('), x(') (i

1 2
=
1, 2, 3) stands for the ith Jacobi set (see Sect. 2.4).
This type of functions is used in the Faddeev method and the CRCG
[26] method. It is known that this expansion converges much faster
than the one using one particular Jacobi set, and the inclusion of low
partial waves is expected to be sufficient to get accurate solutions (see
also Sect. 4.2.1 and [30]). In contrast to this approach, the CG basis
of Eq. (11.26) uses particular Jacobi coordinate but takes
only one
account of the contribution of the high partial waves by the cross term
1412XVX2 in the exponent of the trial function. This ensures accurate
solutions in the correlated Gaussian approach. It is noted that the
calculation of matrix elements with high I values is computer time
consuming (see Appendix A-5), whereas in the CG basis no such PWE
is employed, which makes it possible to calculate matrix elements very
quickly.
11-3 Quark Pauli effect in s-shell A hypernuclei

3
Among a few known A hypernuclei AH, 4j-
4
and 5A Ije are Called
A 1, AIIe,
-
s-shell hypernuclei because in the simplest version of the shell model

all the constituents can reside in an s orbit. By applying the L 0 =
240 Complements
correlated Gaussians of Eq. (7.3), we calculate the binding energies

of s-shell Ahypernuclei in order to reveal the significant role of the
quark substructure of baryons in resolving the long-standing problem
of 'He.
Since the
pioneering work of [1711, the S-shell A hypernuclei have
offered several intriguing problems, among others, the anomalously
small binding of 5AHe. According to a recent survey of hypernu-
clear physics [1721, "The anomalously small binding of AHe5
remains
an enigma. Simple model calculations based upon AN potentials,
parametrized to account for the low-energy AN scattering data and
the binding energy of the A 3, 4 A-hypernuclei, overbindS A5He by
=
2-3 MeV."
The trial function is given as a combination of the correlated Gaus-
sians:
TfJMTMT CkAjFAk (X)XkJM?7kTMTli (11.34)

k
where the operator A antisymmetrizes the nucleon coordinates, XkjjU

is the spin function
(Since L 0, the spin S is equal to the total
=
angular momentum J), and 77kTMT is the isospin function. The spin
and isospin functions are obtained by successive couplings, for exam-
Ple, XkJM= X(S12SI23 )JM [[[Xl/2 X XI/21SI2 X X1/2JSI23-IJTM with
...
:::--:
k representing a set of the intermediate spins S12, S1237 The op- ....
timal set was chosen for each matrix Ak. The matrix Ak containing
nonlinear parameters was selected by the SVM.
The NEnnesota potential [1471 with u I was used for the NN=
interaction. The MA potential is the same as was used in the previ-

ous Complement. Table 11.11 lists the results of the calculation. The
calculation reproduces the data well except for the case Of AHe, for
which the theory overestimates the binding energy by about 1.9 MeV,
consistent with what is mentioned above. Another thing to be noticed
4 4
is that A
He is strongly
more bound than H. It would be difficult to
A
understand this without introducing a charge-dependent component
as the IVA interaction, because then 4AHe would be less tightly bound
because of the Coulomb repulsion.

We have so far treated a baryon as a structureless particle and
distinguished A from N. What happens if we take their quaxk sub-
structure into consideration? Recently it has been shown in [173] that
the anomalous binding problem of 5A He can be, at least partly, resolved
by considering the quark substructure of the baryons.
C11.3 Quark Pauli effect in s-shell A hypernuclei 241
Table 11.1-1. Binding energies (in units of MeV) of s-shell hypernuclei. The
A separation energy BA of the hypernucleus AAX is defined by the difference
of the binding energies, B(-A'1X)-B(-'1-1X).
3 5
AH A411 4AHe A
He
B(AX)
A
2.38 10.62 9.91 34.93
B(`1-1X) 2.20 8.38 7.71 29.95

BA(Cal.) 0.18 2.23 2.20 4.98
BA(Exp.) 0.130.05 2.040.04 2.390.03 3.120.02
Let us assume for the sake of simplicity that baryon is

a a coinpos-
iteparticle of three quarks, p (u, u, d), n =
(u, d, s), =
(u, d, d), A =
and each quark moves in the Os orbit in the harmonic-oscUlator well.

with size parameter b. The spatial part of the three-quark baxyon is
described by the function
1
OB -`
(7rb2)
-1
4
exp (_ (P2 P2 P2))
2b2 1 + 2 + 3
b2 Z3 3
( '3
-
exp
(_ _b 2) O(int) 2rI B (11.35)
where pi is the quark's position vector, rl =

(P1 + P2 + P30 is the
baryon's position vector, and 0(int)

B
is the spatial part of the intrinsic
wave function of the baryon depending on p, P27 (PI + P2)/2 P3* - -
The value of b can be estimated to be about 0.86 fin by requiring that

the function (11.35) reproduce the charge radius of the proton [1741.
When the spatial part of the two-baryon wave function is described
with the configuration of expf -3/(2b2)(r21 + r2)1,2
we interpret it as
indicating that all six quarks move in the Os orbit, exp(-p 2/(2b 2)),
of the common harmonic-oscillator well. with size parameter b. By
increasing the number of baryons, we thus expect that the many-
baxyon wave function may receive a special constraint arising from the
quark Pauli principle that any single-particle orbit can accommodate
at most six quarks (three colors and up-down spins) for each flavor.
It is easy to see that we have no apparent quark Pauli-forbidden
states up to A 4 hypernuclei. However, this is not the case for
=
5
5He:
A
Four nucleons of A He, when they are on top of one another,
have already six u-quarks and six d-quarks in the Os orbit, so nei-
ther u nor d quark of A can take the same Os orbit. This leads us
to the conclusion that the five baryons cannot take a configuration of
242 Complements
exp(-3/(2b 2) 1:5i=1_ r i2) .

To be more specific, the (normalized) Pauli-
forbidden state for 5AHe, with its center-of-mass motion being sepa-
rated, is given by
5
3
TfPF (X) = JV '(7b2)-3eXP(_
4 -
3 _b2 Dr'
i=1
X5)2 (11.36)
where X5 is the center-of-mass coordinate of the five baryons. The

normalization constant is given by JV (4M2 + M2)/(4MN + MA)2'=
N A
where MN is the mass of N. Other Pauli-forbidden states might pos-
sibly exist, but this is the simplest and most evident one.
The solution TI obtained above for 5He was found to contain the
A
Pauli-forbidden component of Eq. (11.36) by only 0.44:'YD, but its conse-
quence is very If this Pauli-forbidden component is simply
significant:
subtracted from the wave function Tf, the BA value would have been
changed from 4.98 MeV to 2.74 MeV, no anomalously small binding
of A5He! The reduction is due to the fact that the energy ex-
mainly
pectation value in TfF(x) is very large, i.e., 513.6 MeV. However, this
may be a premature conclusion from the viewpoint of the variational
principle because the reduction in the binding energy is based on a
calculation of the so-called variation before projection type (see Sect.
6.1), that
is, the variation has been done before the elimination of the
Pauli-forbidden component is made from the trial function.
To calculate the
binding energy more precisely by taldng into ac-
count the quark Pauli
effect, we have repeated the calculation by re-
placing the correlated Gaussian in Eq. (11.34) with the one that 'has
no Pauli-forbidden component:
FAk (X) =
FAA: (X) -
TfPF (X) (TfPF (X) IFAI, (X)) -

(11.37)
12
11.4 The C nucleus
system of three alpha-particles
as a
As an example of the
application of nonlocal potentials, we take up
a simple model for the 12 C nucleus, the 3a model. In this model 12C
C11.4 The 12C nucleus as a system of three alpha-particles 243
>4
3
Pq Exp.
0.0 0.2 0.4 0.6 0.8 1.0 1.2
b [fml
Fig. 11.4. The A separation energy BA of 5H

A as a function of the size
parameter b of three-quark baryons
containing six protons and six neutrons is approximatedas a bosonic

system of three alpha-particles. The physical reason behind this pic-
ture is in the fact that the alpha-particle is a strongly bound, very sta-
ble system compared to light neighboring nuclei. One needs an energy
of about 20 MeV to excite the alpha-particle. Because of this unique
feature, some of the
light nuclei tend to form
subsystem a consist-
ing of two protons and two neutrons, which is called an alpha-cluster.

The alpha-cluster would never be identical to the alpha-particle but it
may happen that a model descriptioii assun-Ang such subsystems can

explain many properties of the light nuclei [60, 175, 59].
To calculate the binding energy of 12C in the 3a model, one needs
to know the potential between two alpha-paxticles. It is hard to de-
rive a potential which acts between composite particles through the
underlying interactions of the particles composing the composite par-
ticles. Thus we use a phenomenological potential which successfully
reproduces the alpha-alpha scattering phase shifts. The one employed
here is an I-independent local potential consisting of both the nuclear
paxt of Gaussian form and the Coulomb part [1761:
2 4e2
_P, erf (,3r),
V(r) =
-Vo e + (11-38)
r
244 Complements
where VO = 122.6225 MeV, p 0.22:ftn-2, and

= 0.75 fin-'. Though
this potential fits the scattering data very well
30 to about Em =
MeV and reproduces the 0+ resonance at 92 keV quite well, we have

to note that it predicts "redundant bound states", two (-72.625 and
-25.617 MeV) in the I = 0 wave and one (-21.999 MeV) in the
I = 2 wave. Since the two alpha-paxticles are known to form no bound

state, these bound states axe considered spurious and must be excluded
ftom the configuration space. The existence of spurious states for a
system of two nuclei is due to the Pauli principle and can be easily
understood in the simplest version of the nuclear shell model. This is
the basis of the orthogonality condition model [1771 which succeeded
to give a foundation to the deep local potential of type (11.38) from
the microscopic theory of scattering between composite particles.
Let us denote the spurious states as (r), (n 0, 1 for I =
0, and n 0 for I=
2). We require that the wave function TI, for the
=
3a system be free from the spurious components in the alpha-alpha

pairs, that is,
( Pnim(ri -
rAff) = 0- (11.39)
Here ri is the position vector of ith alpha-particle. Unless this con-
dition is imposed, the ground-state energy with the potential (11.38)
would be strongly overbound because the alpha-clusters, as bosons,
would occupy the lowest possible states. An alternative, convenient
approach to eliminate reasonably well the spurious components is to
use the nonlocal potential of projection operator type
1M
Fini ==
I (ri -
rj)) (W,,,M (ri -
rj) 1, (11.40)
and include it in the Hamilto-nian as a kind of "pseudo potential" [178]
N N
I = T -
Tn, + E Vij(lri-rjl)+A E EI'i'! (11.41)

j>i=l j>i=l n1m
Here A is a positive constant chosen to be very large. The idea is

that if the function contains the spurious components, then the
wave
energy would be comparatively high. Therefore, the variational lowest

solution would approach a state that has a negligible overlap with the
spurious components.
The spurious bound state can be well approximated in the form
2
lar
O,am (r) Cae- 2 Y1. (r) (11.42)
a
12
CIIA The C nucleus as a system of three alpha-particles 245
Then the nonlocal potential summed over m takes the form
f
_Vnlra
ij
M
21+1 r2_ 1 af rf2

-47r
(rr') R (cosv) E ca*ca, e-la 2
(11.43)
aa,
where V is the angle between r and r. The trial function for the
3a system was chosen as a linear combination of the correlated Gaus-
sians, FA(x), of Sect. 7.2. The matrix element of the nonlocal potential
(11.43) in this basis can be calculated with the use of Eq. (7.46).
Thehigh sensitivity of the energy on A calls for numerical cal-
culations with a high accuracy in order for the pseudo potential to
play the role of the Pauli projector. It was found [1791 that the pro-
jection effect is so strong that the partial waves 1 0 and 2 do not =
contain bound states, and a contribution of more than 95 % to the

ground-state energy is due to the partial wave I 4. Table 11.12 lists =
the energies of the ground state and the excited states obtained in a
variational calculation [1801 using the basis function of Eq. (11.33).

The radial part h (X1 X2) is expanded as combinations of Gaussians
2
x11 x12 exp(-ax 1 _OX2)
2
The calculation does not reproduce the experi-
mental energies very well. In [1801 the energies of the 2+1 and 41+ states
are also given. They are -3.77 and -2.25 MeV, respectively, which
are compared to the experimental energies of -2.84 and 6.80 MeV.
Here the discrepancy between theory and experiment is more serious:
The energy of the 2+,
state is lower than that of the 0+1 state and the
energy of the 4+1 state is much lower than the experimental energy.
Note that there is no Pauli-forbidden state in the 1 > 4 waves, which
makes the contribution of the high partial waves too much important.
This calculation suggests that the 3a model with the local aa po-
tential has only limited success and is not very realistic to reproduce
the experimental energy spectra. This doe not, however, exclude the
Table 11.12. Energies of the 0+ states of 12C calculated in the 3a model

[180]. The energy is measured from the threshold of the 3a breakup.
0+
1
0+
2
0+
3
Cal.(MeV) -3.38 -1.43 3.70

Exp. (MeV) -7.27 0.38 3.03
246 Complements
a + a 8Be, and 'Be(a,-y)12c.

(
Appendix
Matrix elements for general Gaussians
The matrix element for the correlated Gaussian with arbitrary

angular momentum is obtained in a closed form for most
physically
important potentials including central, tensor and spin-orbit compo-
nents. A simple and straightforward method is presented to calculate
the matrix element for the general correlated Gaussian-type geminals.
A.1 Correlated Gaussians
We start with Eq. (7.2) to evaluate the matrix element for a general
case. The matrix element between the generating functions g is given

in Table 7.1 and one has to perform the operations prescribed in Eq.
(7.2).
A.1.1 Overlap of the basis functions
The overlap matrix element can be obtained through
(fK" LM (u, A", X) 1 fKLM (u, A, x))
I _ d2K+L+2K+L
BKLBK-'L ff d9de'YLm(i;_)YLm(e')
dA2K+LdAI2K'+L
3
X
detB
exp [q A2 + q"A/2 + PAA/e. er] \,=0
, (A. 1)
where

248 Appendix
B=A+A', q= fiB-lu, q B-lu , p =; B-lu.

2 2
(A.2)
To perform the operation prescribed in Eq. (A.1), we use the ex-
pansion
exp [qA2 + q, A/2 + pAA/ e. el]

CO 00 CO
E 1: E H(n, q).ff(i , q) H(m, P) Vn+m y2n+m fn

n=O n'=O m=O
(A-3)
where H(n, x) is introduced to simplify the notation and defined by
Xn
H(n, x) 10n!
for
otherwise.
non-negative integer n
(A.4)
Differentiating with respect to A and Y, followed by A 0,

gives non-vanishing contribution only when 2n + m 2K + L and =
2n + m 2K' + L, while the integration over the angles of e and e'

=
becomes nonzero if m is equal to L + 2k with non-negative integer k

(see Eq. (6.18)):
if di d iYLM(e-)yLM(, 1 j)*(e.eI)2k+L =
BI.-L (A.5)
Rewriting n K k and n
= -
= K' -
k, we obtain the result for the

overlap matrix element
(fKIILM(UI, A', X) IfKLM(u, A, x))

3
(2K+L)!(2K'+L)i (27)N-1- 2
BKLBKIL detB
min(K,K')
X
E H(K-k,q)H(K-k,q)H(L+2k,p)BkL- (A.6)
k=O
The B,,I value is given in Eq. (6.9). Note that the values of K and K'
canusually be chosen to be small. in practical cases, and then the sum
over k is limited to just a few terms. In particular, for the special case
of K = K' = 0 the above result simplifies to

A. 1 Correlated Gaussians 249
(ALM(Uli A i X) IfOLM(Ui A X))

3
(2L + 1)!! (2v) N_I 2
4v ( detB ) P
L
(A.7)
A.1.2 Kinetic energy
The kinetic energy with the center-of-mass kinetic energy subtracted

is given in Eqs. (2.10) and (2.11). It is simply written as FrAn-/2. The
matrix element of the kinetic energy is then easily obtained by using
Table 7.1 as
I
(fK"LM(UIjXjX)j ?rA7rjfKLM(UjAjX))
3
h2 (2r) N-I 2
2BKLBKIL detB if de-- do YLm (&) YLM (4 )
( d2K+L+2K+L
X
dA2K+LdA/2K'+L [R + pA2 + p/A/2 + QAA'e-.e-f
X exp [qA2 + q' A/2 + PAAte.e/ (A.8)
where
R =
M(AB-'A!A), P = -i!B-1A!AA!B-'u,
P'= -i B-'AAAB-lu , Q = 2 B-'AAXB-lu. (A.9)
The matrix element of the kinetic energy can be obtained by using a
manipulation similar to the overlap case:

I
(fKfLM(Uf7A!jX)j 2 rA7rjfKLM(u, A, x))
3
h2(2K+L)!(2K+L)! (2r)N-1 2
2 BKLBK"L detB
x
E f RH(K -
k, q)H(K' -
k, q)H(L + 2k, p)
k
+PH(K -
k -
1, q)H(K' -
k, q')H(L + 2k, p)
250 Appendix
+P'H(K -
k, q)H(K' -
k -
1, q)H(L -IF 2k, p)
+QH(K-k,q)H(K'-k,q')H(L+2k-l,p)lBkL- (A.10)
The case with K K' = 0 reduces to a very simple result
(ALM(U17A x) I --iAw
2
I fOLM (u, A, x))
,
h? (2L + 1)!! (2,x)'V-1 2
2
(R+LQp- )-
4-x detB ) P
L
(A.11)
A.1.3 Two-body interactions
Next we derive the matrix element for the interaction of Eqs. (2.3) and
(11.1). To evaluate the matrix element of the operator expressed as
a tensor product, it is convenient to make use of the famous Wigner-
Eckart theorem [70, 71, 721, which states that a matrix element of a
spherical tensor operator 0.-, between states with angular momenta
JM and XIW can be expressed as a product of a Clebsch-Gordan
coefficient and a reduced matrix element which is independent of the
z-components of the angular momenta:
(JMrILJYM-)
(J'M'JO-AIJM) =
V 110. 11 J). (A.12)
,V2J' + I
Here the reduced matrix element is expressed by the so-called double-

barred matrix element. The factor I/V -2-Y+I is factored out because
then the reduced matrix element is symmetric to the bra and ket
interchange
IEV 110. 11 X) =
(-I)J+r.-J,(y 110. 11 J), (A-13)
provided that the matrix element is real and the Hermitian conjugate
t
of OKIL is (0'1') =
E(- 1) "L
Or,, -i-L with a phase factor E(E2 =
1).
By applying the Wigner-Eckart theorem we can express the matrix
element for the orbital and spin angular momentum coupled wave
function as follows:
(Tf(LI SI) JM IV(I'ri -rj 1) (Or ij (space) 0,,ij (spin)) ITf(LS) JM)
-
U(LnJS; L'S)
V,f(2L' + 1) (2S + 1)
A.1 Correlated Gaussians 251
x (L' 11 V(Iri -
rj J)0,,jj (space) 11 L) (S' 11 0,,jj (spin) 11 S).

(A.14)
Here the form factor of thepotential V(r) is assumed to be a function
of r only. Eq. (A.14) and its extension to a general
The derivation of
coupled tensor operator is given in Exercise A.I.
The reduced matrix element of the spatial part is obtained through
(LMn1ijL'M')
(L' II V(I ri -
rj 1) 0,,i, (space) I I L)
v/2--L'+I
=
(L'IWIV(Iri -rj (space) ILM)
=
jV(r)(LM'jJ(rj -rj -
r)O,,,,,j (space) ILM)dr. (A.15)
In the case where one uses the orbital and spin angular momentum
uncoupled wave function, the matrix element (A.14) can be obtained
by
R(L"S")JM IV(Iri -
ri 1) (0,,,j (space) 0,j (spin)) lTf(LS)JM)

-
+ I U(LKJS'; LIS)
(_1)r. 2S' 2S+1 (LMLts1-tjLIML')(SMSrvjS,MS)
x
(TfL,MLS,ms, IV(Iri -rj J)0,st,,j (space)Oc,jj (spin jTfLmLsms)
(A.16)
Since the spin matrix element is easily calculated as will be shown
in Appendix AA, we will focus on the spatial matrix element and
evaluate it for the most important components of the nuclear potential,
that is, central, tensor and spin-orbit components.
(i) central and tensor interactions

be I for the central and
The operator Or-A,j (space) can potential
Y2,,(r-i---r-j)
for the tensor operator. See Eq. (11.7). Thus both the
central and tensor components can be treated by assuming the form
of Y,,,,, (,r-i ---rj) for 0,,,,,3. (space). Expressing ri -
rj as w (ij) x with the
(N -
1) x 1 column matrix Oj) defined iii Eq. (2.13) and using Table
7.1, we obtain
(fK'L'M'(UI74 x)IS(ri -
rj -
r)Y,,j,(rj -
rj) IfKLM(U7 -47 X))

252 Appendix
'::::- Yrg( 6)(fKLM(UIi W 7x)IJ(W(ij)X-r)jfKLM(u,A,x))

3
2 1 (27r)N-2C 2
)
1
e-fc7'Y-"(i )BKLBKL' detB
d 2K+L+2K'+L'
X
if de^- d, iYjLM(e,-) YJL, M,
dA2K+LdAI2K'+L'
XeXp[q,X2+ A/2+ Aye.e/+7Ae.,r+,Y/,XIe/.,r]
(A-17)
where
C w(ij)B-lw (ij) c w(ij) B-1u, 7 cw(ij)B-1U',

I
q=q-
2c
^/2 q'-
2c
712 =P- 1'7YI.
C
(A.18)
All of these quantitiesdepend on i and j but we omit the labels i and

j to simplify notations. The integration over the angles of e and e'
can be performed by expanding exp (pAVe- e' + -/Ae- r + -l'.Ve,- r) in - -
power series and using the relation
(e.e )nj(e.T)n2(eI.,r)n3 Lel=l

n2 +n3
LL'r.
njn2n3
RL Lfn [[YjL( ) X Y X Y.
00
(A.19)
with
Rnjn2n3
2 LIts =
(_I)nl+n2+n3 Bnl-LI7 Bn2-12 12 Bn,3 -13
2 '1 2 2 3
111213
E(2L+:1:1 ) (2
V 212+1
r. + 1)
C(1112; L)C(1113; L)C(1213; K)
U(IIIILK; L'12). (A.20)

See Exercise A.2 for the derivation of this relation. In the defining
equation for R the sum over 11 is limited to ni, ni 2, I or 0, and 12 -
...'
and 13 have similar ranges, respectively. Possible values of L, L", and r,

A.1 Correlated Gaussians 253
for a given set of values of ni, n2, and n3 axe limited by the conditions
that L takes the values nj +n2, nj +n2 2, ..., I or 0, and L' and r, take
-
similar values given by ni, n3 and n2, n3, respectively. In addition, the
sum over L, L, r. is restricted to even values of L + L+ K. Conversely,
for given values of L, L', and K, all of the nj + n2 L, nj + n3 L, - -
and n2 + n3 -
K have to be non-negative and even, otherwise Rnin2n3
LL'r.
njn2n3 pnjn3n2
would vanish. R has the symmetry:
Clearly RLLII,, "L'Ln
The reduced matrix element becomes
(fK'L'(UIi -4 i X) 11 V(17'i -
ri Dyt# i -rj) 11 fKL (u, A, x))

'R
(2K + L)! (2K' + L') 1 (27) N-2

Lf
BKLBKILI ( detB C.)2

x
E I(n2+n3+2,c)RnL,-n2n3
LIn H(ni, #)H(n2, -y)H(n3, 71)
n,n2n3
2K + L -
nj -
n2 2K+L'-ni-n3
XH
2 2
(A.21)
where the integral I is defined in
Eq. (7. 10)
The matrix element of thepotential for different pairs of particles
can be calculated with the above formula by changing only 01) in
Eq. (A. 18).
We note some useful applications of Eq. (A.21). For example, one
can calculate the matrix element of I ri-rj I' simply by putting V(r)
r' and K 0. The two-paxticle correlation function
=
(fK-'Ll (u, A', x) J (I ri -
rj a) Y,, (ri -
rj) fKL (u, A, x))
(A.22)
can also beeasily calculated by taking J(r a) for V(r). -
We note again that the formula (A.21) simplifies for K K' 0, = =
that is, the triple sum reduces to a single sum over, say ni (nl_ =
0, min( L, L', (L + L' r,)/2)) and n2 and n3 have to satisfy

-
nj + n2 L and nj + n3
=
L, respectively. In addition, in this
=
njn2n3
special case R is simply calculated by a term with 11 nj, 12 =
n2, 13 n3 alone in Eq. (A.20). This simplicity will be used to obtain

=
the matrix element of a density multipole operator. See Complement

A.2 for the details.
254 Appendix
(ii)spin-orbit interaction
The operator 0,,,,,, (space) for the spin-orbit potential is the orbital
1- See
angular momentum 1,1i, =
((ri -
rj) x
2h (pi -
pj)) ,. Eq. (11. 9).

Expressing 1,_,,j Fij)x W)7r) x
1,
in terms of x and -r, with the
use of Eqs. (2.13) and (2.14) and substituting the corresponding ma-
i
trix element in Table 7.1 into Eq. (7.2), we obt
(fK"L"M"(Ufi Al7X)jV(jT'i -
I'jj)1jzjjjfKLM(u, A, x))
3
(27)N-2 2
drV(r)e-2
.1 Cr2
BKLBK'Ll detB C)
I
d2K+L+2K'+L'
dA2K+LdA/2K'+LI
x exp q + qf A/2 + PAy e. ef + yXe.,r + -Y/Ale/.
-
X
ifn* X r)t, -
q'Af (e, X r),, (A.23)
with
77 = (MXB-lu, q' = (MAB-luf. (A.24)

Note that to derive Eq. (A.23) use is made of the relation
V(r) e-c('-a)2 (r x a)dr = 0 (A.25)
for an arbitrary vector a provided V(r) is a function of r. See Eq.

(A-163).
Before performing the operation prescribed in Eq. (A.23), we note
that the matrix element vanishes in the case of L 0 L. Because fKLM
has parity (_I)L and because the spin-orbit potential does not change
A. 1 Correlated Gaussians 255
parity, IL -
L'I has to be even. On the other hand, the tensorial char-

acter of the spin-orbit potential imposes the condition IL LI :5 1. -
Both the conditions are met only when L is equal to L'. This spe-
cial result is entirely due to the unique feature Of fKLM and does not
always hold for general wave functions.
Combining Eq. (A.19) and the relation
i
fqX(e X r) 0 -
77W (e' X r)
4vf2--x
3 rjqA[Yj(ii) x Yj(i )Jj, -,qA[Yj(( ) x YI(,P)JI,
(A.26)
yields the reduced matrix element as
(fK,ILI (Ul A! X) I I V(I'ri

I I -rj 1) Iij I I fKL (u, A, x))
'3
4 V2--x (2K + L)!(2K' + L)!. (27r)

3
JLLI (-1)
L
BKLBKIL ( N-2C.)2
detB
x q E l(n2+n3+3,c)H(nl,p)H(n2,,y)H(n3,-y')
n,n2n3
2K+L-nl-n2-1 L
(2K'+ nj
- -
xH
2
H
2 n3,,,)
n., n2 ns
x C(AI; L)U(LAII; 1L)RAL1
A
+77' I(n2+n3+3,c)H(nj,#)H(n2, -y) H(n3 7f) ,
njn2n3
2K+L-nl -n2 2K+L-nj -n3 -

1
H
( 2 q)H( 2
njn2n3
x C(AI; L)U(LA11; 1L)RLAI (A.27)
The above derivation is straightforward and easy to follow. As a

simple check of the above formula, the matrix element of the orbital
angular momentum is calculated in Exercise A.3. It is again possible,
however, to get a simpler formula by performing the'P integration first,
as was done in Complement A. 1. This task is reserved for Exercise A.4.
256 Appendix
A.1.4 Density multipole operators
The matrix element of density multipole operator plays a substan-

a
tial role in investigating the properties of a system, e.g. the density,

the deformation, the electromagnetic transition rates and the electron
scattering form factors. The basic element of the Multipole operator
takes the form
0,,,L, (space) :---

f (Iri -
xm I)YI, (A.28)
Note that the argument of the density multipole operator is not ri
but is correctly taken as ri xN, which is the single-particle coor-
-
dinate measured from the center-of-mass coordinate. Comparison of

Eqs. (2.12) and (2.13) immediately suggests that the matrix element
of the density multipole operator ought to be calculated in exactly the
same manner as that of the two-body potential. In fact the reduced
matrix element
(fK"L-'(U/7 X7 X) 11 f (ITi -
XNI)yn(Tii -
N)
XN fKL (u, A, x))
(A.29)
can be calculated by the same formula as Eq. (A.21) with the trivial
replacements of Oj) --+ 0) inEq. (A.18) and of V(r) --+ f (r) in Eq.
(7.10).
Just as the matrix element for the density operator can be calcu-
lated from that of J(ri -
xN -
r), the matrix element for the cor-
relation function can also be derived ftom. that of J(ri -
rj -
r).
As we have already noted in the above derivations, both of ri xN -
and ri -
rj are expressible in terms of the relative coordinates x as

Cvx with iv- being an appropriate 1 x (N 1) row matrix. There- -
fore all of these matrix elements are reduced to the one of type
(fK'L'Mf(U'jA!jx)jJ(iv-x -
r)jfKLM(u,A,x)). This matrix element

was in fact obtained in the derivation following Eq. (A.17), so that
we give the fi-nal result:
(fKfL'M" (U , X7 X) IJ(17VX -
r) IfKLM(u, A, x))
32 L'
(2K + L)! (2K' + L)! (2T, -1)
BKLBK'L' )N-2C)
detB -v/-2Ll
-e
+ I
2
x
E(LMr.M -MjLM)Y,,m, -m(i )*
M
A.2 Correlated Gaussians with different coordinate sets 257
X
E rn2+n3 Rnin2n3
LLIn H(nl, P) H(n2, -y) H(n8, -y')
nin2n3
2K+L -nI -n2 2K' + L' -
n, -
n3
X H H
2 2
(A.30)
where c, 7, 7/, given in exactly the same manner as in
p are
Eq. (A.18) with w(W being replaced by w. Here, r, takes the values
L+L', L+L'-2,..., IL-L'I, and it is of course restricted by IAF-MI :!
is.The above matrix element becomes very simple for the special case
of K K'
= 0 as was noticed in the previous subsection. Since the
=
matrix element has useful applications, we show in Complement A.2

that Eq. (A.30) can be simplified to just a double sum.
A.2 Correlated Gaussians with different coordinate

sets
As the correlated Gaussians treated in the previous section have a

very simple transformation property implied by Eq. (6.29), we as-
sumed that they are all expressed by a particular set of the relative
coordinates x. However, the correlated Gaussians with the angular
function OLM(X) given by Eq. (6.3) do not have such a nice property.
Moreover, the use of different sets of the relative coordinates for these
correlated Gaussians leads in general to a faster convergence because
they allow us the possibility of describing naturally different types
of correlations and asymptotics. The lineax transformation of the co-
ordinate sets, however, leads to a formidable task even for a system
of only four particles, because the function OLM(X) shows no simple
transformation rule. See Appendix A.5 where the matrix elements for
a three-body system are explicitly evaluated by transforming the cor-
related Gaussians from one Jacobi coordinate set to another. The aim
of this section is to outline a method of calculating the matrix ele-
ments for anN-body system using the correlated Gaussians which
are expressed by different sets of the coordinates.
By making use of Eq. (6.22) to generate the correlated Gaussian,
we have to evaluate the following matrix element for the operator 0:
N-1
exp,
2 x-fXx) 11 Yij, j, (x )
j=1
258 Appendix
N-1
x
1
0 exp -1:Z.Ax)
2
Yli.i (xi)
N-1 N 1
(-')Mj 01j, 01j,i +Mi'
-
01i ali-Mi 3
Climi,9ai 97i C11 -mlj aal - ar'1j'+mi'

j=1 j j j
x (g(a'Jt';A!,x')J0Jg(aJt;A,x)) (A.31)
ri=O,-r =O
3
Here use identity (Yl,,,(xi))*

is made of the (-1)'YI-,,,(xi) to =
take care of the complex conjugation in the bra side. Note that the
vector ti is defined by 2, i(l + i 2), -2-Fi) and likewise t!j by
(1 T,
j
/2' i(I + T/2),
j -2,Tj The vector of g in the bra side should
tj'
not be complex conjugated. Assume that x' is related to x by a linear
transformation T as x' = Tx. With the use of Eq. (6.28) the ma-
trix element (A.31) between two g's which are expressed in terms of
different sets of relative coordinates can be reduced to
(g(a'lt'; X, x')J0Jg(aJt;A, x))

(g(Ta'lt';i XTx)101g(alt;A,x)), (A-32)
and can be obtained from Table 7.1 for most operators.
For the unit operator 0 =
(A.32) takes,
1, the matrix element
-IbB'V
except for a trivial constant factor, the form e2 where v alt+ ,
=
Ta'It' and the (N 1) (N 1)

x -
matrix B' is given by (A+ TXT)
-
To simplify the notation we rename alt, -= aN-1tN-1

si, -.-,
SN-1, a.el =
aNtIV =
SN, ---, afN-,eN-l =
a2N-2t2N-2 82N-2-
Then 6B'v can expressed by a (2N 2)
be -
x (2N -
2) symmetric
matrix B by 9Bs, where B is expressed in terms of B' and T as
follows
B' Bi
B =
( TB' TB'i
(A.33)
Thus the operation prescribed in Eq. (A.31) can be reduced to the

form
N-
Oi",` exp,
(2IBs + ibs
)) .1=0'---'12M-2=0,
(A.34)
1=0,---,'2N-2=0
with
(A.35)
i9ail 0-Til-m
Here the factor iv-s is included because it is needed in the case of the
potential energy matrix elements. See Table 7.1. It is worthwhile not-
ing that the present formulation leads to a unified prescription of Eq.
(A.34), which is independent of the choice of the relative coordinate
sets but requires only a very simple calculation of the matrix B. It
would be extremely tedious if one were to try to rewrite the angu-
N 1
lar function of fL'=I Yj . (x'.) in terms of those angular functions
3 3 3
conforming to the coordinates x.

Using Leibniz's formula we can express Eq. (A.34) as
2 2
(1i -
Mi)!
T
=1 Aj! (Ij -
Ai)! (Ai -
Mj)! (1i -
mi -
Ai + Ai)!
'XiAi
2N-2
X Oj" exp
(2 Os)
-ri=O
2N-2
X ai ie, (A.36)
aj=O
-ri=O
The last factor is easily evaluated by using Eq. (6.58):

2N-2
11 ali-'Xi'mi-Ai
i
e-vs
Ui=O
-ri=O
2N-2
CIi-'XiMi-/-tiY1i-)LiMi-Ai(Wi)- (A-37)
Possible values of Ai and [ii are determined by the conditions that

O<Ai<li, 0 <Aj-/tj:! 1j-mj and -(Ii-Ai) <mj-Aj:! Ij-Aj.
The last condition assures that the coefficients do not
vanish. Because of a special form of si-sj mentioned below, another
condition that Ai -
pi < 2Aj has also to be met.

To evaluate the remaining factor, we note that ABs is quadratic
both in the ai's and -ri's because of si sj -2aiaj (ri -r.j) 2. (S ince,
- = -
for each aj, -ri appears at most in quadratic power, the number of
260 Appendix
differentiation with respect to -ri cannot exceed two times the number
of differentiation with respect to aj, that is, /Xi pi :! 2Aj.) Therefore,
-
.!Bs
when e 2 is expanded in power series in both ai's and -Fi's, and when
these are set to zero after the differentiation, only the term (.IgBs)Q
2
2N-2
contributes provided that Ej=j Ai 2Q and 1:2N-2 (Ai ILi)
i=1 2Q.
Thus we have
2M-2
19Bs (-2)Q
a,\"/L'exp
i (2 Cti=O
Q!
-ri=O
2N-2 Q
X
EBjjajaj(Tj-T.j)2 (A-38)
i=1 i<j aj=O
-ri=o
To have non-vanishing contributions, each of Q terms must be dif-

ferentiated with respect to aj, aj, and either twice with respect to
7i or -Fj, or once each with respect to both 7-i and -rj. This operation
yields 2Bij, where the miniis sign comes from the differentiation with
respect to -ri and -Fj. This sip is denotedEij. Therefore we obtain
2N-2
a,"' '14 exp (2sBs) ai=o

-ri=O
(-4)Q
Q!
Q
x Ejj, Bi,j, (A-39)
k=1
'k<jk
where the sum extends all the

possible combinations distributing
over
the given multiple differentiation into pairs of differentiations. In fact

the operation prescribed in Eq. (A.38) can be done easily by using
the software Mathematica [1841. The case of a three-particle system is
given in Co mplement A .3.
The matrix element for the kinetic energy can be obtained simi-
larly. In fact the matrix element (A.32) for the kinetic energy takes
the form (9Cs)e12'BS, but this can be dealt with easily. Using the
relation
d
1Ms
(Ws) exp (2 ) 2
dp
exp
( 19 (B +pC) s)
2
P=O
(A.40)
(112N-2 aj " "Ws

j=1
e2
IgBs
) ai=O,ti=O
can be obtained by simply replac-
ing one of the Bi,,j,,'s in Eq. (A.39) with Ci,j,.
The
potential energy matrix element for the spin-orbit interac-
tion requires a slight modification. Noting that the general form
of the matrix element (A.32) for this case can be expressed as
(Sk)yeXP (19B.9
2
+ ivis) (see Table 7.1), we make use of the relation
(Sk)jzeXP (2IMs + iv-s ( -1)"

ly(Wk)-M
exp
(219Bs + fvs
(OWk)jj (21ABs ) exp + iv-s I (A.41)
where the spherical components a,, of a vector a are defined by

a, =
- 2=(a., + iay), ao =
a,,, a-, =
!,, (a--
v 2-
-
iay), and the spher-

ical components of a differential operator V,, are given by
aa
( aa)a
1
=
-72=
1
( a
-9a.
+i
Da,
a a
-67a--,
(a
-5a I o
=
Da-,
a
7Eo I
la
"&a )-1
=
.1
V-2 ( aax
'9
-
aa,
-
aal
'9
. By using Eqs. (A.41) and (A.34)-(A.37),
2N-2
the matrix element for the spin-orbit interaction, ( fjj=j 19i1Xi'lLi (Sk)g
-!9Bs+fvs
e2
)ai=O,-ri=07 can be obtained by replacing YIk-Ak -k-Mk (Wk)
in Eq. (A.37) with ( aTwul,
a
YIk-AkMk-Ak(Wk), which is easily calcu-
lated by the gradient formula [70, 71]

,r -
Vaf (a) Yjm(et)
I+ I
-Ti -1 (f(a) +
a
f (a) [T x Yj-j-(et)]j,,.,
+ I
:21EI++11: f(a) -
1f (a)
a
[r X Y (A.42)
In particular we obtain
Yjm(a)
aa
A
-(21 + 1) V I--11 (1/-t1mjI-Im+tt)Yj-jm+, ,(a). (A.43)

262 Appendix
A.3 Correlated Gaussian-type geminals
The matrix element for the general correlated Gaussian-type geminals

of Eq. (6.26) can be calculated from the equation
(W, BI k r) 10 1 f,,, (A, B, R,

I 7
N 3 N 3
anip qn,',
ff H H H yt-i iP
i=1 p=1 j=1 q=1 p t,. n ,,
3q
exp( -
IiZBR -
iR -
liiBk -
Pk)
2 2
(g(t+B'k;X+B',r)101g(t+BR;A+B,r))
tj=O' -=O
(A.44)
It is crucial to know the dependence of the matrix element in the
generating functions on i =
(tj,...,tN) and P =
%,...'t') to do
multiple differentiations. As can be seen from Table 7.1, they appear

as a polynomial of at most degree 2 in exponential functions for the
most important operators 0 = I and J(fvr -
S). The matrix element

for the Idnetic: energy has an additional factor which is again quadratic.
The basic elements of the t and t' dependence include ti-tj, -ej--ej, ti.-ej,
and S-ti, S-tj. As the differentiation can be made separately in each of
the Cartesian coordinates tip and and the basic elements have
% ,
as
no cross terms in different directions of the Cartesian coordinates such

as tiltj2, it is possible to do the needed differentiation in each of the
x, y, and z components.
The above considerations reduce the operation prescribed in Eq.
(A.44) to the mathematical problem given in Exercise A.5. The for-
mula developed in Exercise A.5 can be applied in each of the x, y and z
components to complete the operations needed in Eq. (A.44). It is not
difficult to extend the formula given in Exercise A.5 to get the matrix
element for the Idnetic energy which has an additional polynomial of
degree 2. In fact, Eq. (A.169) can be generalized to a relation which
is valid for a general function:
( cNf (X1 ,X2,..., XK)) xl=o,---,XK=O

AA Spin matrix elements 263
(W#-N f (AA AA AK 'O))'&=O

N! f (AlZi A2Z, ...
i AKZ)
dz. (A.45)
27ri ZN+I
IZ1=1
A.4 Spin matrix elements
The interaction may depend not only on position coordinates but also
on intrinsic degrees of freedom such as spin and flavor.. By represent-
ing the single-particle spin function by Im =

1/2), a basic matrix
element is given by
V_3 (.12 m I IL 1 .1
(MI I o't, I m) =
2 m!) - (A.46)
The matrix elements for spin operators, (S'AF 10,,,,, (spin) I SM), axe
usually by using the Wigner-Eckaxt theorem (A-12). Here a

evaluated
reduced matrix element which is needed to evaluate the matrix element
is the one for Pauli matrix and it is given by
2
o-
2
(A.47)
For the spin function discussed in Sect. 6.4 no technique involv-
ing angular momentum algebra is needed to obtain the spin matrix
element. An elementary, direct method enables one to calculate the
matrix element. We list the spin matrix element that appears in eval-
uating the nucleon-nucleon potential energy.
(Tr l M12 1 0"1 *

a2 I M1 M2 ) =
3(-1)M'1"_M1jMI+M27M/j+Mf2
I
X ("2 MI I MI1-?nl I "2 Tdi) ("2 M2 1 MI2 --M2 2
Mf2 (A.48)
(MIMf
1 2 1 [01
X a212m I MI M2 3 JM,MII+M2I-MI-M2
X ("2 MI I Tnf1-MI Tn 1) (-1 M2 1 Tnf2-M2 2
MI2
2 2
X (1 R I_Ml "rnf2--M2 12 m). (A.49)
(M /M-I
1 2 I [r
X (Ul + a2)] 1M IM1 M2 V3 E E r
qj.
qj=-I q2=-l
264 Appendix
IJ,,,,,n, ('
2 2 ml_ 1 q2I "
2
M)
1 + Jmjm,('I2 M2 1 q2I .12 m)
1
2
(I q, I q2 I I m). (A-50)
(IMIMI
1 2 i(T*0'2)(al)m+(r*0"1)(0'2)-raiMlM2)
3
f (-l)M2-M2 JM?M'j-MIrM2-M2
+ (-I)M, 1 -MI
JM7M2-M27MI-M'I'
*1
x (I2 mi I m---7n,
1 2 Tn/1) (I2 M2 I m'--Tn2
2 2 7nf2) (A-51)
(al M12 I (r'0'1) (r 0*2) *

MI Tn2) = 3 (-J)MI-MI+M2-M2
I
X rMl-M"1 rM2-M'2 (I
2 TnI
1 7n/1 -'rnl
2
Mf1 )(1
2
Tn2 1 TnI--Tn2j
2 2
MI)-
2
(A.52)
(al M2 [T X O'll I X [r X O'21112m IMI M2
3 (.12 mi 1 m--7n, 2 7nfl) ("2 M2 I Tnf2--Tn2 2
7nf2
X
7'ql 7q2(l qj+m'j-mj 1 q2+rn"2 -M2 12 m)
ql.=-l q2=-l
x (I q, I m'j-m, ql+TWI--rni) (I q2 I m'2--M2 q2+m/2-M2)

(A.53)
Here r is a 3-dimensional vector and rn stands for its spherical com-
ponent
Matrix elements for the isospin part can be obtained in exactly the
same manner.
A.5 Three-body problem with central, tensor and spin-orbit forces 265
A.5 Three-body problem with central, tensor and
spin-orbit forces
The matrix element for N-body system can be derived from the
an
general formulas in the

previous sections. In order to illus-
presented
trate that there exist alternative possibilities, here we use a sligthly
different way of calculating the matrix elements for three-body prob-
lems with central, tensor and spin-orbit forces. In this approach the
radial and angulax integrations are done separately and this involves
the extensive use of angular momentum algebra. Needless to say, this
approach cannot be easily pushed far. One may try to attempt to de-
rive similar formulas for the four-body case, but to go beyond that
seems to be too tedious.
(i) a choice of basis functions

The trial function is taken in the form
TfJM,TTMT(123) C,,,,A p ac(1, 23) (A.54)

ac
with expansion coefficients Q, and functions A o,,(I, 23) that are

constructed to be fully antisymmetric under the exchange of particles:
Aw,,, (1, 23) Wc (1, 23) + W,,, (2, 3 1) + W,,c (3, 12)
oac(l, 32) -
(pac(2, 13) -
(pac(3, 21), (A-55)
(the construction of a function is the same,

totally symmetric wave
but then the last three terms should have positive sign in the above
equation). The basis function W,,(k,pq) represents a specific wave
function in a Jacobi channel k (k 1, 2,3) (see Fig. 2.2). The

= nota-
tion (k, pq) stands for
system where the particles p and q axe first
a
connected and then the center-of-mass of the (pq) pair is connected

with the particle k. Thus the coordinate used in the kth Jacobi chan-
nel, X (k) ==
JXk, Ykb is given by
MpT'p + Tnqrq
Xk =
Tp
-
rq, Yk _
rk. (A-56)
Mp +rnq
(k)
(Note that we use f Xk
Yk I in this section, instead of
, Ix 1 I
X(k)I,
2
to
denote the first and second Jacobi coordinates.)
The basis function o,,,,(k,pq) is given by a product of space, spin
and isospin parts:
266 Appendix
W,,r (k, pq)

'
FL'(kpq) x X'(k,pq)]jMj?7TtM,(k,pq),
S (A-57)
where the spatial part is chosen to be a correlated Gaussian of the
form
.FL'ML (k, pq) =

exp
(-2 ;(-k) Ax(k) 2v+,\ 2n+l
Xk Yk YL' ML
") (iik, Vk)
(A.58)
witil
'3
A=
( 6 -Y
62 0) (A.59)
and
Y L('ML Oik -
k) =
[YX(ik) X YI (Vk)ILML (A.60)
The index a stands for a set of discrete labels a =
(Y, n, A, 1) and
for the matrix A which contains the nonlinear paxameters 0, -y7 J.
Here Y, n 0, 1, 2,..., while A is the orbital angular momentum of
=
the relative motion between the and q, and I is the orbital

particles p
angulax momentum between the center-of-mass of the (pq) pair and
the particle k.
The index c denotes the labels (L, s, S, t), where L is the total
angular momentum, S the total spin, and s and t are the spin
orbital
and the isospin of the (pq) pair, respectively. (The total isospin T
is here assumed to be aquantity, so that no T-mbdng is
conserved
considered in the trial function.) The spin and isospin parts axe thus
given by
S
X S M, (k, pq) =
[X, (pq) x Xi
2
(k)] sm, (A.61)
,qTtMT(k,pq) =
[?7t(pq) xqi(k)1TMT, 2
(A-62)
where
Xsm (pq) [Xi (p) 2

x X.1 (q)
2
,qtm (pq) [77.1 (p)

2
x q.1 (q)jtm.
2
(A-63)
The function o,,, (k, qp) is obtained from p,,, (k, pq) by changing
the sign of xk and the order of couplings in the spin and isospin
functions. The sign change of xk results in a change of J to -J and
multiplication by the phase factor and the spin and isospin
functions change as follows:
XsrrL(qp) (-1)1-'X ,,,(pq),

iltm(qp) (-1)1-lqtm(pq). (A.64)
Hence the function p,,,(k, qp) is obtained from V,,,(k,pq) by changing
the sign of 6 and by multiplying by the. phase factor (-I),\+'+t.
The kinetic energy operator with the center-of-mass kinetic. energy
subtracted can. be written in terms of the relative coordinates as
3 2
0 0
Trel -Tein
k) 'k
(k)
A
Yk =Tk, (A.65)
2rni 21j,2
where
(k) Mprn, (k) (Mp + Mq)Mk

III 2 (A-66)
Mp +Mq mp + ni., + Mk
This expression is valid in any of the kth Jacobi coordinate set.
(ii) transformation of the basis function

We, have to calculate the matrix elements between the basis functions
expressed in terms of different coordinate sets (1,23), (2,31), To - ..
achieve this one has to transform the basis function from one Jacobi
coordinate set to another. This can be accomplished by using the
transformation
Xk
Yk ) T (kq)
( X
Yq
q
(A.67)
where the matrices connecting the different Jacobi sets are given by
U12
Tn2+rn3
T (21) =
MS(7nl+M2+M3) MI
(M2+M3)(tnl+M3) Mj+M3
M3
7n2+Tn3
T (31)
M2(Ml+M2+M3) M1
(1712+M3)(M1+M2) Tn1+M2
M3 __
I
tn I +rn:3
T (32)
mj(mj+Yn2+Tn.3) M2
(tnj+mj)(Tnj+tn2) "Ll +TrL2
268 Appendix
-
MI
T(12) M3+Mj
TrIS(Ml+M2+M3) M2 7
(7n3+7nI)(M2+7n3) M2+M3
MI
MI+M2
T(13)
1112(IILI +IIL2'1-11L.1) ?IL'j
(7rII+M2)(?tL2+7J'L.3) "12 +7,11.3
_M2
MI+M2
T (23) (A.68)
MI(MI+M2+M3) M.4
(Tnl+Tn2)(M3+Ml) M3+?nI
The space function TLc'M,, (k, pq), expressed in the kth Jacobi co-
ordinate set, can be transformed to the one expressed in terms of the
qth Jacobi coordinate set as follows:
TL'M (k, pq)=1: L3(kq)

11 L (ad),TLagr (q, kp), ,
(A.69)
ry
where
, MI, (q, kp)
=
exp (_2 ) X
2V+X
q
Yq2fz+Ty(Al)
LM , Q,
(X q, Yq (A.70)
and where a stands for a set (T/, ii, , 1), which has to satisfy the relation
2P + + 2h + F = 2v + A + 2n + 1. (A.71)
The matrix A is uniquely determined from A by the transformation
(A.67) as
j_(kq) AT(kq) (A.72)
The 6(kq)
L (ad) in Eq. (A.69) is the transformation coefficient of
the polynomial part:
Yk2n+Iy(AI)
2v+.k
LML, (Xki Yk
X
k
=
EB(kq)(ad)X2F/+XY2Ft+Ty(XO (XV Yq
L q LML,
-
(A.73)
a
To get this expression one has to make. use of Eq. (6.7) and to recouple,
the product of the two angular functions (see Complement 6.3) by
[y(1112) (:j, ) X y(1314-) (:j, )JLM

12 134
Ef 1121121314-134L y(113124)
LM (:j (A.74)
113124
113124
where the coefficient E is given by a unitary 9j symbol and the coef-

ficient C of Eq. (6.11) as
11 12 112
El'- 121121314134L 13 14 134 Q11 13; 113) C(12 14; 124) (A-75) -
113124
113 124 L
Note that the coefficient E may be defined by the reduced matrix

element
- -
121121314134L
N 2L T 1 Ell
113124
'/2L + I (y(113124) (k C y(1112) (60, ) X y(131A)

I )JLM)
LM I
712 ,34
= (-j)112+134-L(y(113124)(XA' ) Y(1112)(:j
112L
' ) 11 y(1314)(
134
C
=
(-1)13+lA (y(1112)
1 (:j, y(113124)
L
y(1314) ( C'
12 34-
(A.76)
which is easily verified by using Eq. (A.147) and the relation
"
(YI Yk Y1, 1) v/2k + I QW; k)
V21 + I C(kl'; 1). (A-77)
The transformation coefficient B(kq)

L (Cla) is then expressed as
(kq) D"X D
nI
&I- X2,XI1121L
BL (aa) V1XIV2-X2 nj1jn212 X[
vl"*-1"2-"2
njIjn2I2
21/2+1\2 2n1+11
X
T(kq))21/1-+Ai- 12
kq)
(T(kq))
21
2n2+12
(,p(kq)
X
- 22 ) I (A.78)
where the coefficient D is given in Eq. (6.8) and where the summation
is constrained by the conditions
2vl_ + A, + 21/2 + A2 = 27/ + A, 2n, + 11 + 2n2 +12 = 2n + 11

270 Appendix
2vi + A, + 2n, + 11 = 2 i+ X, 2v2 + A2 + 2n2 +12 = 2Tz + 17
(A.79)
and by the
triangular inequality for the angular momentum among
(/XI7'X2iA)7 (11712 07 (Al 117 X) and (A2 12 1), respectively.
7 . 1 7
The transformation to the pth Jacobi coordinate set can be per-

formed as well in a similar manner.
The spin part of the wave function transforms as
X8S Ms (k, pq) (-l)'21+'-SU(.! ISI; gs)X SIVIS (q, k-P)

22 2
E(-I)'!+-SU(.! 2
22
-IS-!;
2 sft S Ms (p, qk). (A.80)
The transformation of the isospin part is given in exactly the same
way-
The radial and angular integrations are separated in this approach.
However, no such separation is made hi the factor e-6xk-yk which
appears in the exponent of the function.FLIM,: (k, pq). Therefore in the
integration one has to expand this term into partial waves (see Eq.
(6.53)):
,-wx-Y = 4r v"2-1-+I e(l, w) il (I w I xy) 01)

100 (-- 7 (A.81)
where ii(x) is the modified spherical Bessel. function of the first kind
(6.48) and E(I, w) = I for w > 0, 6(1, w) (-I)' for w < 0. In the
=
radial paxt one encounters an integral of the form

cc
Ivy 2
I(n, 1, v, I w fo Y
2n+1+2
e-2 ij(jwjy)dy
r-(2n)!! w I'
-2 vn+l+a2 eXp
(W2 ) L('+'!) ( W2)
2v
n
2
2v
(A.82)
where n is a non-negative integer and Ln(')(x) is the associated La-
guerre polynomial (10.15). We define the following integral
I(v, n, 1, u, v, IwI; V)
0 CO
fo JO
1
2v+1+2 2n+1+2 UX2 _
IvY2
X Y e- V(x)il(lwlxy)dxdy
A.5 Three-body problein with central, tensor and spin-orbit forces 271
00
IUx2
1 0
x
2u+1+2
e
--
F
V(x)l(n, 1, v, lwlx)dx. (A.83)
The one-dimensional integration in the above equation, depending on

the functional form of V(x), can be analytically or numerically evalu-
ated. In the special case when V(x) = 1 we obtain
I(v, n, 1, u, v, IwI; V=1) (2n)!!F (n + I + 2 3

n+l+ 2
8 V
X
n
F(k + U + + 1)2
(w2)k
2v
, (A.84)
kl.(n k)!F(k + I +
-
k+v+l+A2
k=O 2
2 w2)
where F is the Gamma function given in Eq. (7.8).
(iii)overlap matrix elements

Now we axe ready to calculate the matrix elements. The overlap be-
tween the functions in different Jacobi sets k and q can be calculated
separately in the space, spin and isospin parts:
(k I q) =
( pc(k, pq) I Wycl (q, kp))
"I"
0
=
JLLI JSSI (.FL'ML (k, pq) I )7LMI (q, kp))
X
SMS (q,kp))(,qtTm,(k,pq)lqt
(X'SMS (k,pq)IX" m,(q,kp))-
(A.85)
The spin part is easily obtained by using Eq. (A.80) as
" +'-S
.1 S
(X'SMS (k, pq) I X"
SMS (q, kp))
=
(- 1) 2 U (.1
22
.1;
2 s's) -(A.86)
Likewise, the isospin part becomes
(,qtTM, (k, pq) lqt'TMT (q, kp)) =

( _j)-!+t-TU(I22ITI;
2
2 tt).
(A.87)
The integration of the space part can performed by transform-
be
ing the function in the bra to the one depending on the qth Jacobi
coordinate. By substituting Eqs. (A.69) and (A-70) we obtain (the
suffix q to label the coordinate set is suppressed in the integral)
(-'FL'ML (k, pq) I

"
(q, kp)) =
E L3(kq) (aa)
L
272 Appendix
X
if x
2P+X+2Y'+)L" 2fz+[+2n'+I'- 2IUX2_lVy2_WX.,y
y e
X
(Y'X) oi, m) *Y(""/
LML LML
dx dy (A.88)
with
U= +o V =!+-t w =S+j'. (A-89)

The angular integration can be done easily by the use of Eqs. (A.81)
and (A.74) as
e- wX-Y
(YL( ML ( C' m L ML d
4w V2_x +I1,E(r w) Ety.OX'1"LL i.(Iwlxy).

_X-+ (A.90)
One can easily verify that 1+1'-r, has to be even and non-negative,
and then one can use the integral formula (A.84) to obtain
"'
(.FL'ML (k, pq) I -1
LML (q, kp))
"
4y"13(kq)
Ir
, L
X + 16 (M' W) EX"jj'-0,X'1
Vf2-1
K
LL
A+AI-K
X I
(P+vl+ 2 2
'K,U, V, IWI; V=I)
(A.91)
(iv) matrix elements of kinetic energy operator

The next step is the calculation of the matrix elements of the kinetic
energy operator
(k ITrel I q) =
(W,, , (k, pq) ITq I W& d (q, kp)). (A.92)
In the above expression we used the fact that the kinetic energy op-
erator can be expressed in different relative coordinate sets (see Eq.
(A.65)), and thus we defined Tr,,l in conformity with the coordinate

set in the ket. To calculate the matrix elements of the kinetic energy
operator one first notes that
(X 2v+.X
Y
2n+l e-.21 ,6.2_:I_YY2_,E._y YLM
30X2 + 02X4 + j2X2Y2 -
20(2v + A)X2 + 2Y(2z/ + 2/X + 1))

2_1
21/+)L-2 2n+1 --Ox ilyy 2_,SX.y Y. (,Xj)
X X Y e 2
LM 4 M
-2( x
2v+X-1 2n+l+l
Y e
_.L,3X2_
2
1,
YY2_,5x.y
F3 7F
X (2v + 2A + I 13X2)
x
EU(A-I1Ll;AK)C(11;K)Y('X-I LM
-V2A +
+I
(2v -
Ox 2)
x
LM (A.93)
A proof of this relation is given in Exercise A.6. A similar expression

holds for the second relative coordinate y. By the help of the above
equation the effect of the kinetic energy operator applied on the basis
function is given by a linear combination of terms akin to the basis
function. The matrix element of the kinetic energy can therefore be
given by repeatedly using the expression obtained for the overlap of
the basis function.
(v) matTix elements of central potentials

The central two-body interaction can be written as
V =
V(I'r2 -
7'3 1) + V(Irl -
T3 1) + V(jrI -r2 1)
=
V(Xl-) + V(X2) + V(X3) (A.94)
that is, the potential is a sum of terms depending on the first relative
coordinate in different Jacobi sets. In calculating the matrix element
for the case
(k I V(xp) I q) = ( O,, (k, pq) I V(xp) I Wa,,, (q, kp))

-
JLLIJSSI(-'r-7LaML (k, pq) I V(xp) I.F&L (q, kp))

X (X'SMS (k,pq)IX"
SMS (q,kp))(TItTm,(k,pq)lqt m,(q,kp)),
274 Appendix
(A.95)
the spin and isospin paxts can be calculated as before but the spatial
functions in the bra and the ket have to be transformed into the pth
Jacobi coordinate set in which the potential is defined.
The space paxt of the matrix element is calculated by using the
transformations k -+ p and q --+ p:
"'
(FL'M,,(k,pq)IV(xp)l J,w, (q,kp))
B(kp)
L (aa) L3(qp)
L (ala/)
x (-'FL'M.,,(p,qk)IV(xp)IFL' (p, qk)). (A.96)

Here the matrices corresponding to the transformations axe defined
by
T(kp) AT(kp)
A' = T(qP) A!T(qP) (A.97)
respectively. By introducing
U + V +7 W j+ &I (A.98)
one obtains
(.FL5'ML (p, q k) I V (xp) 1.FL IL (p, q k))
4 jr /-2-r.+1,E(n, w) Et nOXI'P'LL
+
X
I(P+Vf+ 2
ii+?P+
2
K) U, V, IWI; V) -
(A.99)
(vi) matrix elements of tensor potentials

The tensor interaction is defined through the tensor operator (11.7)
/-4r
2
Crk] 2) (A.100)
SP V -5 (Y2 (ip ) [O'q
'
X -
To calculate the matrix element of the tensor interaction one has to

follow the same steps as for the central interaction, except that now
the spin part of the wave function has to be taken into account, and the
spin part should also be transformed into the samecoordinate system
as the potential. The use of the Wigner-Eckart theorem (A. 14) enables
one to obtain
(kjV(xp)Spjq) -= (W,,(k,pq) IV(xp)Spl W,,,e(q, kp))
',E47 VF(-2L+
247r
7r
5
r U(L2JS; LS)
1) (2S' 1) +
x (,)7L' (k, pq) II V(xp) Y2 (x-)

P II FL',f (q, kp))
x (X'S (k, pq) II [Cq X O'kj 2 11 X'S/ (q, kp))
x (nTtMT(k, pq) Iq tm, (q, kp)). (A.101)

Here the spatial part of the matrix element is calculated as before by
transforming the basis functions to the pth Jacobi coordinate set
(Jc'L'(k,pq) 11 V(xp)Y2( p-) Ij.FL,(q,kp))

13(kp)
L (ad) 13()
L' (a& )
x (YL' (p, qk) I I V(xp) Y2 (x-)

P II FL',f (p, qk)), (A.102)
where the last term is obtained by using Eqs. (A.81) and (A.82) as
(JC L"(p, qk) ip

V( Xp) Y2 (Xp) i (p, qk))
4ir V2- r, + 1 c (r., w)
X (YL' 1) YO 0 Y2 (- O) YL
X
2
ii+?P+
2
r" U, V, IWI; V)
(A.103)
Here u, v, w are defined in Eq. (A.98). Note that f+ r, has to be
even and non-negative.
276 Appendix
The reduced matrix element of the angular paxt is calculated by

using the angular momentum recoupling technique (see Complement
6.3) and Eq. (A.76). One obtains
( Y.,L YO (I K
Y2 ( 0) YL(XI F) ( c
I
U(2n2r,; WO) Q2K; W)

Kf
X (YL 01 Y2 Ll
/2r,'+ 1
EV C(2r.; K)EX LXII,-'L2. (A.104)
2K+I
The
spin matrix element is also easily calculated by transforming
the spin functions to the pth Jacobi system and by using Eqs. (A.147)
and (A. 149) as
(X'S (k, pq) 11 [Crq X Uk]2 11 X"

S" (q, kp))
=
E(-l).2k+g-SU(-nS-1; S) E(-1)-!+""-S'U(.!
22 2
2
22 2
X (X S (p, qk) [G'q X Ok] 2 11 XS P, (p, qk)), (A.105)

where
(X S (p, qk) 11 [Uq X Uk12 11 XS P, (p, qk))

(2S + 1
(_I)s,,-St+s-g V U(Ii
2 S2; S'9)
29+1
X (("221) 9 11 [O'q X UkI 2 11 ("22'0 i (A.106)

with
(('1'1)9
22 11 [O'q X Uk12 11 22
8
1 2 2
4+ 1
1
1
1
2
9
(,V6) 2 =2v/_5JgjJ pj. (A.107)
2 2
(vii) matrix elements of spin-orbit potentials

The spin-orbit force is given through the operator (11.9)
I
LV Sp - with Lp =
-ixp x Vp, Sp =
2
(0'q+ffk)- (A.108)
As in the case of the tensor interaction, with the use of the Wigner-
Eckart theorem (A.14) one obtains
(kjV(xp)Lp-Spjq) _= ( o,,,(k,pq) IV(xp)Lp-Spl W,,,,(q, kp))

U(L'lJS; LS) r
(FE'(k,pq) 11 V(xp)Lp 11 FL, (q, kp))

-1)(2S'
V-(2L + + 1)
x (X'S (k, pq) I I Sp I I X'S, (q, kp)) (,qtTMT (k, pq) lqt'TMT (q, kp)).
(A.109)
The spatial part of the matrix element is calculated as before by trans-
forming the basis functions to the pth Jacobi coordinate set
(.FL' (k, p q) V (xp) Lp YL, (q, kp))
L3L(kp) (aa) L3(qp)

LI (af&i)
x (.FL5 (p, qk) 11 V(xp)Lp 11 Jc'L",(p, qk)). (A.110)

When the orbital angular momentum operator acts on the basis
function (provided both the basis function and the operator axe ex-
pressed in the same coordinate system) one obtains
A0
e--21- '3X2 _!IY _IYI _,5X_y YL(M
L
(x 21/+X 2n+1
Y
2_1
ij(x X Y)X" +A Y2n+l e_21,3x 'YY 2_dX.y
YLI'Xj) (ic
M
1
1)3X2 YY2 A0
+X
2v+,\ 2n+1
Y e- 2
_
'f
-6"YLYL(M (A.111)
Note that the outer product i(x X y),, is equal to the tensor product
vf2-[x x yll_,, =
(4-\,F2-x13)xyYj(,',1) ). Therefore the spatial part of
the matrix element, (.FL6(p, qk) 11 V(xp)L, 11 J:L` ,(p, qk)), can be ob-
tained in exactly the same way as in the central potential. It is given
by a sum of two terms: One is
4vf2-7
47r J vf2-_K+I c (K, w)
3
X (YL` r) ( C' m I I YO(Kn) ( C' m YI(I-I-) ( C' m I I YL(" F") (:t' o

278 Appendix
X1
(P+1/1+ 2
n+W+
2
K) U, V, IWI; V)
(A.112)
and the other is
4-,T v/2 r. + 1 E (r,, w) (YL(

, YO(MM) (i L II YL(fk'[')
X Fj+P+ fi+??+ KI U1 V, IWI; V

2 2
(A.113)
Here u, v, w axe defined in
Eq. (A.98). Note that in the first term
T+ P + 1 -
r, has to be
and non-negative, and likewise in the
even
second term I + 11 r. has to be even and non-negative.

-
The angulax part in the first term can be obtained by using Eqs.
(A.74) and (A.76) as
(YL YO( MIS) YI(l 1) YL
pq (A.114)
pq
The angular part in the second term becomes
(YL( ( b, YO(K n) L YL('," i,
L
L YL(, "") (A.115)
where
(YL(X"") L YLI
2-L+
1
(_,)L-L" V 2XI +I
U L 1; L'XI) (I Xi, i L YX, ( b)
(A.116)
with
(Yl(;k) 11 L 11 Yl(.-b)) =
0(1 + 1)(21 + 1). (A.117)
The spin matrix element can be calculated as in the tensor inter-
action. The result is
(x'S (k, pq) I I Sp I I X'S'/ (q, kp))

Sg) E(-I) 12+S'-S'U(.1
1
-21+-SU('.!S.1; S' 1;
22 2 22
x (X S (p, qk) 11 Sv 11 X S-', (p, qk)), (A.118)
where
(X'S (p, qk) I I Sp II X'S'f (p, qk))

S+I
2
-S,+S-
=29+
S+IU (Iksi; Sg)
1 2
X (("I22'1) 9O'q + Uk I I (I22'I) k) (A.119)
with
((*"1)
22 9O'q
+ O'k 22 R)
E:
8+1
'9
122 -j;
1)
2+
U(-! -191; -1)
2
06- ((- 1) +
206-JgJi j. (A.120)
280 Complements
Complements
A. I Mabix elements o central potentials
The matrix element of the central and tensor potentials can be ob-
tained through a step which is slightly different from the procedure
presented in Appendix A. 1.3. Then the formula becomes simpler than
Eq. (A.21). Here we outline the step which leads to this simpler result.
Instead of using Eq. (A.19), we first perform the integration over
the angle i6 in Eq. (A. 17), which leads to
Tj I r)
(fK'LIMI (7Z, X, X) [ YxIL ( ri -
rj ) I fKLM (u, A, x))

r2
3
_.LCT2 47r (2r)N-2 2
e
BKLBKILI detB c) if d& d '
d2K+L+2K+L"
X YLMWYLIMfO )* dA2K+LdA/2K'+Ll
x exp [ A2 + q/A/2 + AA/e. et] in (rV)Y"'j#Y) (A.121)
Wi0
V =
-tAe + -Y'Ye'. (A.122)
If either -y or -y' vanishes, v is independent of the scalar product e-e',
and the integration and the differentiation prescribed in Eq. (A.121)
becomes very easy. When neither -y nor -y' is zero, the term in the
2
exponent can be expressed in terms of v as
qA2 + qiA/2 + pAA'e - e'
2-y-y'
V
2
+ (q_ 2-y A2 + (qf 2-/
A/2. (A.123)
A
One first has to expand exp(2-y,,1V 2) i,,,(rv) in terms of powers of v
2n+n
(see Eq. (6.48) for the expansion). Secondly, its general term v
multiplied by the spherical haxmonic must be expanded in
the coupled tensor form of [YL ( -_) x YL, (,E )] The second step can
be done with the use of Eq. (6.7). In this derivation one can avoid
recoupling the angular momenta, and hence would get the simplest
possible formula for the matrix element.
CA. 1 Matrix elements of central potentials 281
In the following we work out the details for the central potential
with x = 0. However, we shall not follow the above route exactly but
make a little detour in order to make transparent the relation to the
overlap matrix element. To this end we rewrite the left-hand side of
Eq. (A.123) as follows:
A2 + I A/2 + pAA/ e. e/ = qX2 + q'A12 + PAA/ e. e V2. (A.124)

2c
V2
The term e 2c jo (VV) is expanded in powers of v with the use of the
Hermite polynomials:
V V 1
exp 2c)
-
i o (r v) =_ exp 2c)
-
2rv
(e'r
V
-
e -rv)
00
V22 r) (V2c2)n.
C
c,
H2n+l (A.125)
,,F2c n=0
(2n + 1)!
Next we expand exp (qA2 + q'A/2 + pAAI e. ef) (,,2,)n

2c
as follows:
ni
exp q A2 + q' A/2 + p,\A'e -
e'
V2)n
) (2c _
(2
PPIIMMI
H(p -ra, qA2)H(pI -
n , q'A/2 )H(I -
n + m + m', pAA'e-e)
H(M' 2A2 )H(a, 7/2,X/2 )H(n -

m -
a, 27-yAYe-e)
1: FPnPj (c, q, q, P, t,,Yf),X2p+'A/2p'+l (e-e')' (A.126)

PPfj
with
n! I
Fpnp,,(c, q, q, p, 7,7') =
1: H(p -
m, q)H(p -
M Iq
MW
n+m-nz! in-m+m'
7 7
x H(I -n+m+m',p) 2m+m'm! (A.127)
ml 1. (n -
m -
ne)!'
Here p, p, I,m, and m' are all non-negative integers and the ranges
of their values are determined the conditions p + p' + I > n and
by
m < p, ne < p', n I < m -
+ m! < n. The operation in Eq. (A. 121)

is now easily done, leading to the result
(fK' L' MI (Uf, X) X) I V(I Vi -
'rj 1) 1 fKLM (U, A, X))

282 Complements
2
K+K+L
(2K + L)! (2K' + L)! (2
JLLIJMM'
BKLBK-'L detB
E J(n, c)
n=O
min(K,K")
X
E Fk7-k KI-k L+2k (c, q, q A'Yi ^//) BkL i (A.128)
k=O
where J(n, c) is the integral defined by

I
J(n, c) =
-
-V/7-r-(2n + 1)!
"0
x
f V(jx)e-X2HI (X)H2n+l(X)dX-
0 C
(A.129)
In the case of V(r) = I the formula (A. 128) reduces to that of the
overlap matrix element (A.6), because then the integral J(n, c) be-
comes nonzero if and only if n = 0 and FF2k Kf -k L+2k (c, q, q, p, -Y, -y')
reduces to k, q)H(K' k, q')H(L + 2k, p). Though this rela-
H(K - -
tionship is not apparent in Eq. (A.21), both equations should give the
same result for the central potential.
For the special case with K K' 0, Eq. (A.128) reduces to a

= =
much simpler form which essentially includes a double sum:
(fOLM (U I
A! 7 X) I V(ITi -rj 1) 1 fOLM (u, A, x)
=
(fOLM(W,X,X)jfOLM(u,A,x))
L 1
D 77
n
X
E (L-n)! ( cp ) J(n, c), (A.130)
n=O
where the overlap (fOLM(UIiWiX)jfOLM(u,Ax)) is given by Eq.

(A.7). The above formula hasa particularly nice feature for power
k
law potentials V(r) r because then the c-dependence of J(n, c),
=
,
now denoted Jk (n, c), is factored out as follows:
k
00
2
-2 ai2n+l
(2)
2%
Jk (n C) =
C Vir(2n + 1)! fo X
k+I
( dX2n+1
-e
-X
dx
k
n
2
1 (-l)m22n-2m+l
(2) C 7,=, E M=0
mI(2n-2m+I)!
-V n-m + k+3).
2
(A.131)
CA.2 Matrix elements of density multipoles 283
Since the particle j are contained only in c and -y-y'lcp,

indices i and
this factorization makes the summing up of Eq. (A.130) over i and j
much faster. Especially for the Coulomb force (k -1), we get =
2c (-l)n
J-1- (n, c) -
(A-132)
-x (2n + 1) n!
For other types of potentials the route given below Eq. (A.123) leads
us to the following simpler result including only a single sum:
(f0LM(UIiWiX)jV(jri -
rjj)jf0LM(UiA7X))
L
)3 1:
I
(2L + 1)!! (27r)N-1 2
L! 77
n L-n
4v detB
n=O
n!(L -
n)! c
00
2n+2
X
V'r(2n + 1)!! 10 V(eX) X2n+2 C
e-'2dx. (A.133)
A particular simplicity occurs for a Gaussian potential as one can then
perform the n sum easily. For V(r) e-'2, we obtain =
(ALM (U X X) I eXP I_ X(r,

i i _,rj)21 IfOLM(U7 Ai X))
3
3 f L
(2L + 1)!! (21r)N-1, -I 7,y
4v detB )2 + 2r,) (P
C
+
C i+
(A.134)
A.2 MatTix elements of density multipoles

Here we Eq. (A.30) takes a simple form for the special case
show that
of K K'
= 0. As we have seen in the chapters of applications, the
=
correlated Gaussian with K 0 is, with the use of its fall general A
=
matrix and u vector, a very useful basis function. The simplified form
of the matrix elements was in fact used in our calculations. TrI the case
of K =
0, L nj n2 has to be non-negative
- -
even and in order to have
a non-vanishing contribution from the term H((2K+L-ni -n2) /2, q).

Likewise Lf nj -
n3 has to be non-negative
- and even. On the other
hand, the recoupling coefficient Rnin2n3

LLIr.
of Eq. (A.20) becomes non-
vanishing provided that ni + n2 L, n, + n3 L' and n2 + n3

-
K -
-
are all non-negative and even. Therefore we are led to. the result that
the summation in Eq. (A.30) is restricted to the case of nj + n2 L =
and nj + n3 L. By renaming nj as n, we obtain n2

= L n and = -
284 Complements
n3 = L' -
n, where n runs from 0 to min(L, L'). Possible values of is
are limited
by the condition that n2 + n3 -
n = L + L' -
2n -
is be
non-negative and even and also K > max(IL -
LJ, IM -
M'j) due to
the triangular relation of the Clebsch-Gordan coefficient (L M K AV-
MIL'IW).
Furthermore, we note that under these conditions RnLjn2n3
L-lis
=
nL-n L-n
RLL'r. is contributed by a single term with 11 =
ni =
n, 12 =
n2 = L -
n, 13 =
n3 = L' -
n in Eq. (A.20). This is because

11 :!ni, 12
13 :5 n3 and 11 + 12 > L, 11 + 13 ! L. Using the
< n2,
well-known formulas for the Clebsch-Gordan coefficients (10 P 0 1 L 0)

and the Racah coefficient [751
U(n L-n L r,; L' L-n)
(2L 2n+ 1)! (2L' 2n)! (L +L' is) 1 (L +L' + n + 1) 1

- -
-
(2L + 1)! (2L')! (L + L' 2n is) 1 (L +L' 2n + r, + 1)!

- - -
(A-135)
we obtain the following result:
(fOLIMI'(Ul X, X)IJ(17VX -r)jfOLM(U, A, X)

min(L,Ll)
(2T) M-2 2
( detB
3
cl e-'Ycr
1 2
E pnL-nI/ IL-nrL+L'-2n
n=O
L+L'-2n
x
1: ZLL'nn (L M r, M -MJL'M)
r.=max(jL-L'J, IM-M,'I)
x Y. M' --M ('0 *' (A-136)

where r. + L + L' has to be even and the coefficient ZLL'nr. is given by
ZLLfnr. -
(-1) 2 (L-L'+x)
F 2L 47r
(2 +1)
x
JL-K)!(L
T L, + L' + -+I)!
-
(2n)!! (L + V -
2n -
n)!! (L + L' -
2n + x + 1)!!
x
F L-(LL --+L"5;
L
-1i ((L ---LL++-'KS (A-137)
CA.3 Overlap matrix elements for a three-particle system 285
With an appropriate choice of the (N -
I)-dimensional vector w
one can calculate various matrix elements. To calculate the matrix

element of J(ri -
xN -
r), one only needs to set w equal to w(') which

is defined in Eq. (2.12). The values of c, p, -y, -/' are obtained from Eq.
(A.18) with w(ij) being replaced by 0). In exactly the same manner,
one can also calculate the matrix element of J(ri -rj -r) by employing
the vector w('j), defined in Eq. (2.13), for w. By using the well-known
formula
J(Ifvxl -
r)
J(iv-x -,r) Y"11 WX Y"tt (A.138)
r2
r. it
Eq. (A.136) immediately enables one to obtain the matrix element of

the density multipole operator J(ji7vxj r)Y,,,,(w` _x). -
The matrix element for the central potential, V(j ri -rj 1), is obtain-
able from Eq. (A. 136) as well. By integrating the equation multiplied
by V(r) over r, we see that only the term with n 0 contributes to =
the matrix element of the central potential. It is easy to see that the
integration leads us to the same result as Eq. (A.133).
A.3 Overlap matrix elements of the correlated Gaussians for a three-

particle system
We tabulate in Table A.1 results of the calculation for the quantity
/2N-2
61,
i9aili (9-rili-mi
exp
( -2IS-Bs) (A.139)
for thethree-particle system (N 3) assun-Ang li < 2. The vector si is

=
defined by aj(1 i 2, i(l + 2), -2-ri). The calculation was done with
Mathematica [1841. The quantity apparently has a symmetry prop-
erty with respect to the interchange of li, mi +-+ 1j, mj. E.g., if one
calculates this quantity for a given set of angular momenta, then the
quantity corresponding to a set (111 MI 13 M3 12 M2 14 M4) can eas-
1 , , 1 1 1
ily be obtained from it by simply interchanging the suffices 2 and 3

of the B matrix. Note that the table can be used for the four-paxticle
system as well if one of the angular momenta is restricted to zero.
286 Complements
Table A.I. Tabulation of Eq. (A.139) for the three-particle system. The
vector si is defined by ai (1 -
*Tj 2, i(l +7,i 2), -2-ri).

The matrix B is a 4x4
symmetric matrix and its elements are denoted a B12, b B137 C = = =
B14, d B23, e-
=
B241 f
= B34. The tenth to twelfth colunu3.s give the
=
types and the coefficients of terms needed to construct the solution, where
each coefficient must be multiplied by a factor given in the ninth column of
the corresponding entry. For example, in the case of 11 MI 12 M2 = = = =
2, 13 =
2, M3 -2, 14
=
2, M4 =
-2, the quantity of Eq. (A. 139) is given
=
by 12288 x (3b 2e2 + 3c2d2 + 12bcde).
11 MI 12 M2 13 M3 14 M4
-4 f
--
0 0 0 0 1 1 1 -1
0 0 0 0 1 0 1 0
192 f2
0 0 0 0 2 2 2 -2 1
0 0 0 0 2 1 2 -1 -1
0 0 0 0 2 0 2 0 1
32 ef
0 0 1 1 1 1 2 -2 6
0 0 1 1 1 0 2 -1 -3
0 0 1 1 1 -1 2 0 1
0 0 1 0 1 0 2 0 2
0 0 1 0 1 -1 2 1 -1
0 0 1 -1 1 -1 2 2 1
16 af be cd
I 1 1 0 1 0 1 -1
1 0 1 0 1 0 1 0 1 1 1
(Continued on the next page.)

CA.3 Overlap matrix elements for a three-particle system, 287
(Continued from Table A. I.)
1536 def
0 0 2 2 2 0 2 -2 -2
0 0 2 2 2 -1 2 -1 3
0 0 2 1 2 1 2 -2 2
0 0 2 1 2 0 2 -1 -1
0 0 2 0 2 0 2 0 2
256 af2 bef cdf

I 1 1 2 0 2 -2 -6 -6
1 1 1 2 -1 2 -1 9 9
1 1 0 2 1 2 -2 6
1 1 1 0 2 0 2 -1 3 -6
1 1 1 -1 2 2 2 -2 -3 -6
1 1 1 -1 2 1 2 -1 3 3
1 1 1 -1 2 0 2 0 -3 -1 -1
1 0 1 0 2 2 2 -2 3
1 0 1 0 2 1 2 -1 -3 -3 -3
1 0 1 0 2 0 2 0 3 4 4
1 0 1 -1 2 2 2 -1 3
1 0 1 -1 2 1 2 0 1 -2
-1 1 -1 2 2 2 0 -1 -1
-1 1 -1 2 1 2 1 1 1
12288 a2f2 b2e2 C2d2

abef acdf bede
2 2 2 2 2 -2 2 -2 3 3
12
2 2 2 1 2 -1 2 -2 -3
-6
2 2 2 0 2 0 2 -2 3
2 2 2
2 2 2 0 2 -1 2 -1
-3 -3 6
2 1 2 1 2 0 2 -2
-2 -2 4
2 1 2 1 2 -1 2 -1 3 3
3 3 6
2 1 2 0 2 0 2 -1 -3
4 -5 -5
2 0 2 0 2 0 2 0 3 3 3
6 6 6
288 Exercises
Exercises
A.I. Derive Eq. (A.14).

Solution. Noting the definition of the scalar product in Eqs. (11.1) and
(11.2), expanding the angular momentum coupled states of bra and
ket functions, e.g.,
(LS) JM) =
E (Lm, SM2 I JM) Lrni) I SM2) (A.140)
TaIM2
and using the Wigner-Eckart theorem (A. 12), we have
((L'S') JM 1 (0,, (space) 0,,, (spin)) (LS) JM)

-
1) 1'(Lm, SM2 I JM) (LW1 SW2 I JM)

AMIM2M'MI
1 2
X (L' 11 0,,(space) L)
-vF2-L'+1
(SM2K --PISIMf2)
X (S' 11 0,,(spin) S). (A.141)
V12--S'+1
The use of a symmetry property of Clebsch-Gordan coefficients
+
(SM21S --AISIMI)
2
=
(-1)IS
V (KILSIMf2 I SM2) (A.142)
2S+1
enables one to rewrite the above matrix element as
((L'S') JIVI 1 (0,, (space) Q, (spin)) I (LS) JM)

-
(V 11 0,,(space) 11 L)(S' 11 0,,(spin) 11 S)

(-W
-1)(2S
V/(-2L' + + 1)
X
E (Lm1r,/LJL'm'I)(L'm'IS'm'2JJM)
JL'ra1M2?-afIM2r
x (Lm1SM2JJM)(K.ASITnI2JSTn2)- (A.143)
The sum over A, Tn1 , M2 , MfJ , TnI2 of the product of the four Clebsch-
Gordan coefficients is just the recoupling coefficient U(Lr.JS';L"S)

(see Eq. (6.67)). This complets the derivation.
Exercises 289
A generalization to the matrix element of a coupled tensor operator

canbe made similarly. Let us calculate the matrix element
((L',5 )XM'j [01(space) x Ox(spin)j,,, J(LS)JM
(LIMI1 SIM/2 I JM) (Lm, SM2 I JM) (1mAp I rw)

MIMMIM2MA
1 2
(Lmj1mjL'm
11 (L' 01(space) 11 L)
,f2--L'+
(SMALI S/ M2I)
(S' OX(spin) 11 S). (A-144)
V -2-Sl+1
The sum over ?n j, m'2 , Tn1 , M2 i M7 A of the product of the five Clebsch-
Gordan coefficients
(LmISM2 I JM) (1mXMjnv) (LmIlmILal)

MIM2MtlMflM'2
X (STn2l\l-tlS'M2)(VM'IS'M'21XM') (A.145)
can be taken by means of the unitary 9j symbol as
L S J
(JMr,vIYM') I A K (A-146)
L' S' X
_ _
To derive this result multiply Eq. (A.145) by (JMr'VjYAF) and sum

over M and v with M' being fixed, then what one has is a sum of
the product of six Clebsch-Gordan coefficients involving nine angular

momenta, and it is nothing but the 9j symbol in unitary form given in
Eq. (A.146) (see Eq. (6.78)). Then with the help of the orthogonality
relation of the Clebsch-Gordan coefficients Eq. (A.145) itself has to be
equal to (A.146). Substituting this result into Eq. (A.144) and using
the definition of the reduced matrix element one obtains
((L'S')X 11 [01(space) x 0,\(spin)l,, 11 (LS)J)

L S J
2JI + 1
1 A
(2LI + 1) (2SI + 1) P S,
-
x (L' 01 (space) L) (S' I I 0,\ (spin) S). (A.147)

290 Exercises
Inspecial cases one can set A = 0 or 1 = 0 in the above formula

and by using
(S' 11111 S) =
JSS, A/2S +I (A.148)
derive useful formulas as given below:
((L'S')J' 11 01(space) 11 (LS)J)
jssl(_I)L-J+J-L' V1-2LI-j,-+1U(SLXl;JL)
+I
x (L' 11 01(space) 11 L . (A.149)

((L'S')X 11 OX(spin) (LS)J)
JLLI
V TS-,+-, U(LSJA; JS) (S' Ox (spin) S).
(A.150)
A.2. Derive Eq. (A.19).

Solution. Using Eq. (6.18), we have
(e. e ) ni (e. r) n2 (ef.,r) n3
rn2 +n3 Y, E Y, BkllBk2l2Bk3l3

2ki+1,,=nj. 2k2+12=n2 2k3+13=n3
(-1)11+12+13-v/'(211 I)T(2172 1) (2K3 +T) [Y11 ( -) X Y711 ( 0100

[y12 (' -) Xy12 ('0100 [Y13 0 ) X Y13 ('06' (A.151)
As was done in Exercise 6.1 (see Eq. (6-106)), we recouple the product
IY12 0 X Y12 (f9]OO [y13 0 ) X y13 0 )100

12 12 0
13 13 0 Q12 13; "0 11Y12 (0-) X Y13 0 )Ir- X Yr-(f6)100
X 0
-
2K+1
Q12 13; K)
212 + 1) (213 + 1)
Exercises 291
X 1[Y12 (P-) X Y13 0 )I r-

X YK M 1 00' (A.152)
The last step is to combine the two scalar products as follows:
1YII (10 X Y11 (0 )100 RY12 ( -) X Y13 ( I)Jr- X YK00100

IRY11 (P-) X Y11 0 )10 X [Y12 X Y13 0 )]Klr- X Yr-00100
EELI111101213r-r- [[YL(&) X Y Lf( I& X YIs (j )100, (A.153)

LL'
where the coefficient E is given by Eq. (A.75). When one of the angu-
lar momenta is zero, the 9j symbol is reduced to the Racah coefficient
(see Eq.(6.86)). Using this simplification in the coefficient E and sub-
stituting the above results completes the derivation.
A.3. Calculate the matrix element of the orbital angular momentum
IAjj for the correlated Gaussians.
Solution. The matrix element of lAij can be obtained from that be-
tween the generating functions of Table 7.1 or by performin the inte-
gration over r in Eq. (A.23) with V(r) being set to unity. The result
is
(fK'L'M'(Ul i
A 10) 1 IjLjj I fKLM (u, A, x))
3
(2v) M-1 2
1
BKLBKILI ( detB ) c
*Y, +,q,-Y)
d2K+L+2K+L'
ff dMi YLm ( _) YL, m, (o )
dA2K+LdX12K-+L'
AAexp [q A2 + q1 A/2 + pAYe-e'] i(e x e),,
(A.154)
The integration over _ and o can be done by using the formula
ff de^_d4! YLm( -_)YL,m,(, )*(i(e e),,(e-e')') x

e=I,eI=1
_2 1 110111,
d,T V '2__Ll+1 (LMIjLjL'M) E Bklv _21+IEW

4
2k+l=n
292 Exercises
(A.155)
Clearly I has to be L I and also L' 1. In order that a possible value
of 1 exists, either L L' or IL L'I
= 2 has to be satisfied. The latter
-
=
case is obviously impossible because the condition IL L'I < I has to -
be met. Thus we reach the same conclusion that L and L' must be
equal, otherwise the integral would vanish. Using
QL-I 1; L) -
7r(2L
____
3L
47r(2L + 1)'
-
QL+1 1; L) 3(L
3(L 1)
7(L2L
+
7F(2L 1)'
47(2L +
U(L-llLl;Ll)= 2+L U(L+l IM; L1) V:2LL+ 2
(A.156)
together with the relation
2k + 2L + 1 2k
BkL-1 =
BkLi Bk-I L+I Bk-Li (A.157)
2k+L 2k+L
we can show that
r 11
JJ dM4 YLm( -)YL,Iv,(i )*(i(e- x e'),(e-e)'
JLLI VFL (L + 1) (LMly I LM) Bn+I-L (A.158)

n+I 2
In order for the integral not to vanish, n +1 -

L must be non-negative
and even.
Employing this result in Eq. (A. 154) and compaxing its result with
Eq. (A.6) leads to a solution
(fKILIMI (u!, X, X) 11pi, IfKLM(u, A, x))
05LL'1\1L(L + 1) (LMI[tILM)
1
X -
('Y?71 + 7177) (fKI LM (W, X, X) I fKLM (u, A, x)). (A.159)

PC
As a check of the above formula, one may use this to calculate

the matrix element of the total orbital angular momentum LIL =
2
7 I:i<j IlLij (see Eq. (2.22)). The sum-ma i n, over i < j, of .1C (-y?7' +
^11,q) can be easily done with the use of Eq. (2.15), yielding just z2 Lp.
Therefore, we expect that the desired matrix element becomes
Exercises 293
(fK"LfM' (ul A!, x) ILIIfKLM(UA, X)

,
--`
JLLI - fL-(L + 1)
x (LM11-ilLM)(fK'LM(WiA iX)IfKLM(UIAIX))- (A-160)
On the other hand, we can reach the same result through the Wigner-
Eckart theorem:
(LM IAILM)
(a'L'M'IL,IaLM) :::
JLLI' (aL 11 L 11 aL)
,F2L + I
(LM 1/ilLM)
JLLf VIL(L + 1) (aL 11111 aL)
,V -2-L+1
JLLI (LM lpILM) VFL-(L

+ 1)(aLMIaLM), (A.161)
where the labels a and a' stand for the other labels of the wave func-
tions. Here we used the fact that L,, does not change the magnitude of
the angular momentum of the wave function because it is a generator
of the rotation group.
A.4. Calculate the matrix element of the spin-orbit potential following

the method of Complement A.I.
Solution. The integration over P is first done in Eq. (A.23), leading to
(fK'LIM" (U A X) I V(I'ri
7 i -ri 1) IlLij I fKLM (Ui A X))
3
00
47r (27r)N-2C2
n)Idr IBKLBK"L' (
Cr2
(,yq +'y r2 V(r) e-
detB
d2K+L+2K'+L'
ff de^- di YLm( -)Ypm, (i )- d/\2K+Ld/X/2K+L'
AA/exp [qA2 + q/A/2 + #AA'e. e] i(e x e%,rii(rv) V

X=0'XI=0
e=I,ef=I
(A.162)
Here v is the length of the vector v defined by Eq. (A.122) and use is
made of the formula
f 0"(b x r),,& =
47rrii(ar)(b
a
x a),,, (A.163)
294 Exercises
which can be proved by using Eqs. (6.53) and (6.54) and noting (b x
r), , = -Nf2i [b x rl V.L3'ir [b x Yj- (i6)] I,

v2
The term e-2c 'i, (rv) is expanded
-
as was done in Complement
V
A.I:
CO
2 2
V r 1 2n+2 n
exp( 2c) -
-ii(rv)
v
3
(2c) - ' r n=0 (2n + 3)1

(V2c)
(6cr) ( 6r)
C C
c
X H2n+3 + 2(2n + 3)H2n+l - (A.164)
By using the expansion (A.126) and integrating over - and 4 with

the use of Eq. (A.158), we obtain
(fKf L' (u', A, x) I I V(I'ri -rj 1) lij II fKL(u, A, x))

32
(2
JLLI 'FL(L+,)(2L+,)(2K+L)!(2K'+L)!
V (
BKLBK-L detB
K+K+L-1 00
n+I
10 V(jX) e-X2
1
X
C
(7771 + 71,q) I: -
F-x(2n + 3)!
V C
n=O
X H, (X) H2n+3 (X) I + 2 (2n + 3) H2n+l (x) I dx

min(K,K)
1
X
E Fk-kKI-kL+2k-I (c, q, qr L+2k
BkL.
k=O
(A.165)
In the case when V(r) = I only the term with n = 0 contributes
to the matrix element. The function Fr2k KI-k L+2k-1 (CI q, q, p, -y, -y)
L+2k
then becomes H(K -
k, q) H(W -
k, q) H(L + 2k, p) Therefore

.
P
we can confirm that Eq. (A.165) reduces to Eq. (A-159).
A.5. For a given function f (X1, X2 XK) of the form
f (X1, X2, ..., XK) =1 TcAx + 6,

2
where:i =
(X1-i ---7 XK)7 (bl,..., bK), and A is a K x K symmetric
matrix, calculate
Exercises 295
,)ni
OX,ni
ef(Xl,X27---,XK
11==02 ...
7XK::--O*
(A-166)
Solution. A solution can be obtained with the use of Leibniz's formula

as was done in A.2. Here we show another way to solve this problem.
Let an operator L denote
a
L A, + A2 + *
-+AK (A-167)
XI (9X2 19XK
Where (All ...

7 AK) is an auxiliary parameter. Then we have
e' 'Cef (xI 7 ... 7XK) = ef (x'+6, ' 7 ... 7XK+IMK)
ef(XI 7 ... 7XK) exp1UXV2

(2 + W + Ax
11) - (A-168)
Let N be equal to EK i==l ni. Expanding

the above equation in power
series of 0, collecting the terms of the Nth power in 0, and putting
Xi= 0 (i =. 1, K) leads to ...,
(f-Nef(xl 7 ... )XK)

) Xj=O .... 7XK=O
[N
21
N!
=
E k! (N
k=O
-
2k)!
(1U ) k( b) N-2k.
-
(A.169)
Theproblem of finding a solution to Eq. (A. 166) is thus reduced to
a simpler problem of expanding the right-hand side of Eq. (A.169) in
the Ai's. To this end ( b)N-2k and (WA)k

2
are expanded as follows:
(N -
2k)!
( b) N-2k (Albl)", (AA) 12 (XKbK) 1K,
11! 12! IK.
k k!
2 AA) (I)k2
Mij
Mll! M12! 7nKK!
(All A,2)mll
x (2Al2AlA2 )M12 ...

x (AKKAK 2)MKK. (A-170)
Collecting those terms which have powers of A,nj A2 n2 AK nK we
obtain a solution
296 Exercises
K
ani
-e
9X,ni
f(XI,X27 ....
XK))x 1=07 ... IXK=O
i2L k K
k=O
(')
2
E 11 ni!H(mij, Aii)H(li, bi)
Mij i=1
XH H(mij, 2Aij), (A.171)

j>i=l
where mij's and li's are all non-negative integer and must satisfy the
following relations
K K
mij =
k, mij + E mji + li =
ni (i 1, K). (A. 172)
j>i=l j=1
A.6. Derive Eq. (A.93).

Solution. First we note that
( 2
'+Ae-2IpX2_,5X.y YLM
(Al)
x
x 41 M)
(,Axe- .j'8X2_jx.Y) X2v+Ay(,Xl)
2
LM
e-12 jax 2-,6X.y A.'X 21/+Ay(,\I)

+ LM PC M)
+2 (Vxe-2.jpX2_jx`.Y) (VXX -
2v+,X AI)
YL(M P9M (A.173)
The differentiation in the second term of the right-hand side of the
above equation can be performed by using Eq. (6.59) as
AI)
(AxX 2v+,\
YL(M
AO
2v(2z/ + 2A + I)x 2v+,X-2 YL(M PC M I (A.174)
The third term can be reduced, by means of the gradient formula
(A.42), to
(Vxe- .IpX2_jX.y 2
VXX
2v+A AI)
YL(M (ic'
Exercises 297
1
X2v+X-l e oX2 -,6x.,y
F+' OX 1
(2v + 2A + 1) X YX-1 ('' )4 X Y' Ml LM
-A-+I
2A+1
2v[[(- x-Jy)XYX+I( 3)11\Xyl( )ILM
(A.175)
By using the angular momentum algebra the first term in the round
bracket of the aboveequation is reduced to
[[(-OX -
JY) X Y)'-' POD' X Y' M] LM
,rL3 OxC(1A-I;A)YL(A')(:t, ) M
r.)
+ Sy U(A-1 I L 1; A r,) C(11; r)Y(X-l
LM
A
2A+lt XYL X1)
M ( O,
Jy E U(A -
11 L 1; A K) C(11; r,) YLM
K
(A.176)
where use is made of Eq. (A.156) for C(l A -
1; A) C(A -
11; A).
Similarly the second term becomes
11(- X-*X YA+'(" )IXX Y1MILM
XY(,X1)
L M PC M I
yE U(A+I ILI; Ais)C(11; r,)YL(A+1r')(1b, M (A.177)
The contribution of the first term is easily obtained by using

298 Exercises
e- .l#X2_,5X.y
2
1 2
-,5x.y
+ 02X2 + j2y2 + 2,3dx -,y) e-'2,8X 1 (A.178)
where the last term involving the scalar product x -
y can be reduced
to
X. Y Y(Ai)
,
LM
=
-v -3[x x y]oo0'1) LM
-V 3 E U(11 ; 0"0 Ry X [X X Y X
1WILM
Y
r.=Al
X Ily X YA-1 4A A X Y' MI LM
_X
2,X + I
Ry X Y +'P X YI(MILM* (A.179)
The two terms of the right-hand side of the above equation appear in
Eqs. (A.176) and (A.177), respectively.

Combining these results we obtain the desired equality (A.93)
l,3X2_,6x.y AI)
Ax (X 2 Ll +,X
e-2 YL(M
30X2 +,32X4 + j2X2Y2 -
20(271 + A)X2
LpX2_jX.y
+ 2zj(2i/ + 2A + 1)) X
2v+,X-2
e- 2 YL(M
-2Sx 2v+A-l y e-LPx

2
2
_
1 2
_,SX.y
V/ 3
7F
X
I TA):+ I
(2y + 2A + 1_ OX2)
x EU(A-IlLl;Aiz)C(11;K)YL(A-l)(. c, ) M
Is
'A+I
-
(2y _
OX2)
2A +I
x
I:U(A+IILI;Ar,)C(11;r,,)YL M (A.180)
K

Stochastic Variational Approach To Quantum-Mechanical Few-Body Problems PDF

Загружено:

Сведения о документе

Оригинальное название

Авторское право

Доступные форматы

Поделиться этим документом

Поделиться или встроить документ

Параметры публикации

Этот документ был вам полезен?

Это неприемлемый материал?

Авторское право:

Доступные форматы

Stochastic Variational Approach To Quantum-Mechanical Few-Body Problems PDF

Загружено:

Авторское право:

Доступные форматы

Preface

The variational method is one widely used techniques to

parameters. The direct method of diagonalizing the Hamiltonian ma-

to monitor the energy convergence.

general extremely difficult and challenging, great advances have been

Quantum mechanics plays a fundamental role in atomic and sub-

range Coulomb interaction dominates in atoms and molecules but the

almost equal masses and the interaction between them is short-ranged.

analytical calculation of matrix elements. We follow the latter course

physics. To help readers to understand the text, we have attempted

clarity, and given several Complements of an explanatory nature. The

of formulas that may possibly be difficult for readers as exercises with

stochastic variational approach or the physical consequences of the

testing the contributions of different states. The random selection of

authors (K.V.) is grateful for the hospitality of the Physics Depart-

I.. Introduction ........................................

2. Quantum-mechanical few-body problems ...........

3. Introduction to variational methods ................

4. Stochastic variational method .......................

C4.1 Minimization of energy versus variance ..............

5. Other methods to solve few-body problems .........

6. Variational trial functions ..........................

7. Matrix elements for spherical Gaussians, ............

8. Small atoms and molecules .........................

9. Baryon spectroscoPy ................................

10. Few-body problems in solid state physics ...........

11. Nuclear few-body systems ..........................

Matrix elements for general Gaussians ..................

A.1.3 Two-body interactions ......................

There are a countless number of examples of quantum-mechanical few-

The system will normally be in the minimum energy quantum state.

termine the energy of a Coulombic three-body system: a positron and

Y. Suzuki and K. Varga: LNPm 54, pp. 1 - 6, 1998

ple enough so t1i at a graphic illustration of the wave function can be

continuous parameters or "quantum numbers" and these paxameters

of the electrons and the positron.

examples of four random states and their energy expectation values

minimum energy calculated analytically with one term of Eq. (1.1) is

ber of random trials. Figures 1.2(a)-(d) show examples for random

Fig. 1-1.Examples of the energy expectation value and the (normalized)

a combination of about 150 functions of the form (1.1). The conver-

with, say, a linear combination of 150 functions and then, by random

Fig. 1.4. The energy convergence of Ps- as a function of dimension of basis

the by avoiding being trapped

Moreover, one may have discrete paxameters or quantum numbers

As first step toward the goal of giving a unified and reasonably

simple recipe for solutions of few-body bound-state problems the basic

are advantageous to emphasize correlations between the particles and

to separate the center-of-mass motion. One can relate the different

Let us consider isolated system of N particles with masses ml.,

All of the particles may be identical or different, or some group(s)

Y. Suzuki and K. Varga: LNPm 54, pp. 7 - 20, 1998

H=E 2mi -T,. + Vij. (2.1)

As the interaction unchanged by the displacement of the

origin, the above Hamiltonian is translationaUy invariant and depends

One may encounter few-body, problems where the system is sub-

Even with the presence of the one-body potential Ui it may happen

r) 0?. dr. (2.3)

Here p is the short-hand notation to specify the component of the

2.2 RelatiVe coordinates

The separation of the center-of-mass motion is

row matrix. Here XN is chosen to be the center-of-mass coordinate of

independent relative coordinates. They axe in general related to the