Spectral Estimates for Schrödinger Operators

Alexandra Lemus Rodrı́guez

A Thesis in
The Department of Mathematics
and Statistics

Presented in partial fulfillment of the requirements for the degree of Master of

Science (Mathematics) at Concordia University
Montreal, Quebec, Canada

August 2008

Alexandra
c Lemus Rodrı́guez 2008
 Abstract
Spectral Estimates for Schrödinger Operators.
Alexandra Lemus Rodrı́guez.

In quantum mechanics, one of the most studied problems is that of solving the
Schrödinger equation to find its discrete spectrum. This problem cannot always be
solved in an exact form, and so comes the need of approximations. This thesis is
based on the theory of the Schrödinger operators and Sturm-Liouville problems. We
use the Rayleigh-Ritz variational method (mix-max theory) to find eigenvalues for
these operators.
The variational analysis we present in this thesis relies on the sine-basis, which
we obtain from the solutions of the particle-in-a-box problem. Using this basis we
approximate the eigenvalues of a variety of potentials using computational implemen-
tations. The potentials studied here include problems such as the harmonic oscillator
in d dimensions, the quartic anharmonic oscillator, the hydrogen atom, a confined
hydrogenic system, and a highly singular potential. When possible the results are
compared either with those obtained in exact form or results from the literature.

iii
 Acknowledgements

I would like to express my thanks to Professor Richard L. Hall, my supervisor, for

his support, guidance, and patience. I have learned more than quantum mechanics
and mathematics from his classes and our conversations. I say with confidence that
without him I would not be standing where I am at this point in my academic journey.
I also want to thank Professors Twareque Ali and Lennaert van Veen for being
part of my defence committee and helping me achieve this important stage in my life.
I thank my family for their faith and their support: My mother for being the wise
woman she is, and for pushing me out of the nest in her very own way. My brothers,
Gerardo and Enrique, for sharing their knowledge and showing me other paths.
Thanks to David for everything.
I give thanks to my friends (past and present), from whom I have learned what life
is about. Because I have seen myself through their eyes, I understand how much I have
grown up and how far I have traveled, but even more, thanks to them I know there
is still so much to learn and live. In particular I want to thank Claudia and Behrang
for sharing a part of their lives with me, when our roads crossed at Concordia.
I take this opportunity as well to thank and acknowledge the Department of
Mathematics at Concordia University, the Institute des Sciences Mathématiques, the
Power Corporation of Canada and the Secretarı́a de Educación Pública for their
financial support.

iv
 To my mother and father,

v
Contents

List of Figures viii

List of Tables ix

Introduction 1

1 Schrödinger Operators 3
1.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2 Operator theory . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
1.2.1 Variational characterization of the spectrum of an
operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
1.3 Schrödinger operators . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
1.3.1 The dimension d . . . . . . . . . . . . . . . . . . . . . . . . . 11
1.3.2 Examples . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12

2 Sturm-Liouville Problems 17
2.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
2.2 Sturm-Liouville theory . . . . . . . . . . . . . . . . . . . . . . . . . . 19
2.2.1 Operator form of SLP . . . . . . . . . . . . . . . . . . . . . . 20
2.2.2 Eigenvalues and Eigenfunctions of the SLP . . . . . . . . . . . 21
2.3 Schrödinger normal form . . . . . . . . . . . . . . . . . . . . . . . . . 22
2.3.1 The Liouville transformation . . . . . . . . . . . . . . . . . . . 23

vi
 2.3.2 A numerical approach to solving the SLP . . . . . . . . . . . . 25
2.4 The particle-in-a-box . . . . . . . . . . . . . . . . . . . . . . . . . . . 26

3 Variational Analysis for Schrödinger Operators 28

3.1 Introduction . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
3.2 Variational method . . . . . . . . . . . . . . . . . . . . . . . . . . . . 29
3.3 The basis for the analysis in dimension d = 1 . . . . . . . . . . . . . . 32
3.4 The basis for the analysis in dimension d ≥ 2 . . . . . . . . . . . . . . 36

4 Results 38
4.1 Implementation of the method . . . . . . . . . . . . . . . . . . . . . . 38
4.2 Dimension d = 1 . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.2.1 The harmonic oscillator . . . . . . . . . . . . . . . . . . . . . 41
4.2.2 The potential V (x) = |x| . . . . . . . . . . . . . . . . . . . . . 44
4.2.3 Polynomial potential: V (x) = x4 . . . . . . . . . . . . . . . . 46
4.2.4 Polynomial potential: V (x) = x2 + x4 . . . . . . . . . . . . . . 47
4.3 Higher dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 49
4.3.1 The harmonic oscillator . . . . . . . . . . . . . . . . . . . . . 49
4.3.2 The hydrogen atom . . . . . . . . . . . . . . . . . . . . . . . . 52
4.3.3 Confined hydrogenic atoms . . . . . . . . . . . . . . . . . . . . 53
4.3.4 Highly-singular potentials . . . . . . . . . . . . . . . . . . . . 55

Conclusions 57

A Maple Code for the Variational Analysis of the Harmonic Oscillator 62

B Shooting Method 66

vii
List of Figures

1.1 Harmonic oscillator . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14

1.2 Hydrogen atom . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16

2.1 Infinite square well potential . . . . . . . . . . . . . . . . . . . . . . . 26

3.1 Shifted particle-in-a-box . . . . . . . . . . . . . . . . . . . . . . . . . 33

4.1 Energy vs. Parameter L for the harmonic oscillator in dimension d = 1. 42

4.2 Trial wave function for the ground state ψ0 of the harmonic oscillator
in dimension d = 1 with minimizer L = 6.1. . . . . . . . . . . . . . . 43
4.3 Trial wave function for the first excited state ψ1 of the harmonic oscil-
lator in dimension d = 1 with minimizer L = 5.6. . . . . . . . . . . . 43
4.4 Trial wave function for the second excited state ψ2 of the harmonic
oscillator in dimension d = 1 with minimizer L = 5.8. . . . . . . . . . 44
4.5 The potential V (x) = |x| . . . . . . . . . . . . . . . . . . . . . . . . . 44
4.6 Quartic anharmonic potential . . . . . . . . . . . . . . . . . . . . . . 47
4.7 Energy vs. L of the quartic anharmonic oscillator in dimension d = 1. 49
4.8 Harmonic oscillator in dimension d = 2, and ` = 0. . . . . . . . . . . 51
4.9 Hydrogenic atom in d = 3 confined to a box of size b = 4. . . . . . . . 55
4.10 Hydrogenic atom in d = 3 confined to a box of size b = 72. . . . . . . 55

viii
List of Tables

4.1 Approximation of the energy levels of the harmonic oscillator in di-

mension d = 1. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 41
4.2 Approximation of the energy levels of the potential V (x) = |x|. . . . . 45
4.3 Approximation of the energy levels of the potential V (x) = x4 . . . . . 46
4.4 Approximation of the energy levels of the potential V (x) = x2 + x4 . . 48
4.5 Approximation of the energy levels of the harmonic oscillator in di-
mension d = 2. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 50
4.6 Approximation of the energy levels of the harmonic oscillator in di-
mension d = 3, 4, 5. . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
4.7 Approximation of the energy levels of the hydrogen atom in dimension
d = 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
4.8 Approximation of the energy levels of a confined hydrogenic atom in
dimension d = 3. . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54

ix
Introduction

In non-relativistic quantum mechanics, Schrödinger’s equation governs the time evo-

lution of a particle’s wave function, and is given by

∂ ~2
i~ Ψ(x, t) = − ∆x Ψ(x, t) + V (x)Ψ(x, t). (1)
∂t 2m

In this equation, ~ ≈ 6.6255 × 10−27 erg sec is known as Planck’s constant, V (x) is
a potential independent of time, and the operator ∆x is the Laplacian corresponding
to the kinetic energy.
One of the main problems in Quantum Mechanics is to determine the solutions of
this equation, in particular, finding the energy levels of a particle. There are many
different approaches to solve it, one is to find exact solutions and the other is to use
methods that will give estimates of the solutions that cannot be found in an exact
manner. The examples in physics that have an exact solution are very few compared
to the ones that do not. This motivates an approach that allows us to approximate
solutions. In this thesis we describe an approach using variational analysis.
In Chapter 1 we review the theory concerning Schrödinger operators. In particular,
its relation with functional analysis, and we present the variational characterization
of the discrete spectrum of a Schrödinger operator, related to the min-max principle.
In Chapter 2 we review the theory of Sturm-Liouville problems. This is of im-
portance because we can transform a number of problems in the Schrödinger normal
form.

1
 In Chapter 3 we describe the variational analysis of quantum mechanical problems
and introduce a special variational basis related to the solutions of the particle-in-a-
box problem.
In Chapter 4 we describe the implementation of a program to approximate the
eigenvalues and the eigenfunctions of a problem. This chapter is also devoted to the
results obtained in the application of the variational method to a variety of explicit
problems.

2
Chapter 1

Schrödinger Operators

1.1 Introduction

For the problem of finding the solutions to the Schrödinger equation given in (1), we
can rescale the equation, and given the assumption that Ψ(x, t) = ψ(x)Θ(t), derive a
time independent Schrödinger equation which is represented as follows

−∆ψ(x) + V (x)ψ(x) = Eψ(x). (1.1)

In this chapter we will relate this equation to the famous Schrödinger operator.
The results from this part are mainly studied by Griffiths [9], Gustafson and Sigal
[10], Hall et al. [13], Hannabus [14], Kryezig [16], and Reed and Simon [19].

1.2 Operator theory

The theory of Schrödinger operators relies on functional analysis. In the following

section, we describe various results that are very helpful for the purposes of this work.
Let H be a Hilbert space, in particular the space of all quantum mechanical states
of a given system. The main example used in quantum mechanics, and in this thesis

3
is the L2 -space defined as follows

 Z 
2 d d 2
L (R ) = ψ : R → C | |ψ| < ∞ .
Rd

H is endowed with the inner product given by

Z
(ψ, φ) = ψφ dx, with x ∈ Rd , (1.2)
Rd

where ψ represents the complex conjugate of ψ. The above inner product defines the
norm
1
||ψ|| = (ψ, ψ) 2 .

We recall that an operator A in a Hilbert space H is a mapping of its domain

D(A) ⊂ H into H.

Definition 1.1. A linear operator A in a Hilbert space H is an operator such that

1. the domain D(A) and the range R(A) of A are subsets of H,

2. for all ψ, φ ∈ D(A) and scalars α, β ∈ C,

A(αψ + βφ) = αAψ + βAφ.

Definition 1.2. An operator A on D(A) ⊂ H is bounded if

||Aψ||
||A|| = sup = sup ||Aψ|| < ∞.
{ψ∈D(A)} ||ψ|| {ψ∈D(A) | ||ψ||=1}

Lemma 1.1. [10] If an operator A satisfies ||Aψ|| ≤ C||ψ|| (with C independent of

ψ) for ψ in a dense domain D(A) ⊂ H, then it extends to a bounded operator, also
denoted A on all H, satisfying the same bound: ||Aψ|| ≤ C||ψ|| for ψ ∈ H.

This result is important because the domain in which an operator is defined can

4
be a very complicated set, with this lemma we can find operators that are extended
to bounded operators and not worry about the domain itself.

Definition 1.3. Given an operator A on H, an operator B is called the inverse of

A if D(B) = Ran(A), D(A) = Ran(B), and

BA = 1|Ran(A) , AB = 1|Ran(B) ,

where Ran(A) = {Aψ | ψ ∈ D(A)}. The inverse of A is denoted by A−1 .

We note that finding the inverse of an operator is equivalent to solving the equation
Aψ = f for all f ∈ Ran(A).

Definition 1.4. The operator A is said to be invertible if A has a bounded inverse.

An operator A is not invertible if it is not bounded below, that is, if there is no

c > 0 such that ||Aψ|| ≥ c||ψ|| for all ψ ∈ H.
We assume that all operators A are defined on a dense domain D(A) ⊂ H.

Definition 1.5. The adjoint of an operator A on a Hilbert space H is the operator

A∗ satisfying
(A∗ ψ, φ) = (ψ, Aφ)

for all φ ∈ D(A), for ψ ∈ D(A∗ ), where

D(A∗ ) = {ψ ∈ H | | (ψ, Aφ) | ≤ Cψ ||φ||, ∀ φ ∈ D(A)}.

Note that the constant Cψ is independent of φ.

Definition 1.6. An operator A is symmetric if

(Aψ, φ) = (ψ, Aφ)

5
for all ψ, φ ∈ D(A).

Theorem 1.1. (Hellinger-Toeplitz Theorem) [16] Let T be a symmetric linear

operator on all of a complex Hilbert space H, then T is bounded.

Definition 1.7. An operator A is self-adjoint if A = A∗ .

It follows immediately from this definition that a self-adjoint operator is also

symmetric, but the opposite is not always true, we have then the following lemma.

Lemma 1.2. [10] If A is bounded and symmetric, then it is self-adjoint.

Theorem (1.1) and the above lemma (1.2) suggest that the class of operators that
we can use is sufficiently wide.

Definition 1.8. A self-adjoint operator A is called positive, denoted A > 0 if

(ψ, Aψ) > 0

for all ψ ∈ D(A), ψ 6= 0. Similarly we may define non-negative, negative and non-
positive operators.

Theorem 1.2. [10] If A is a self-adjoint operator, then for any z ∈ C with Im(z) 6= 0,
the operator A − zI has a bounded inverse, and this inverse satisfies

||(A − zI)−1 φ|| ≤ |Im(z)|−1 ||φ||.

The above theorems are important concerning the domains of linear operators.

6
1.2.1 Variational characterization of the spectrum of an

operator

Definition 1.9. The spectrum of an operator A on H is the subset of C given by

σ(A) = {λ ∈ C | A − λ is not invertible}.

Definition 1.10. The complement of the spectrum of an operator A in C is called

the resolvent set of A, denoted by

ρ(A) = C \ σ(A)

We note that for λ ∈ ρ(A), the operator (A − λ)−1 , called the resolvent of A, is
well-defined.

Definition 1.11. A number λ ∈ C is called an eigenvalue of the operator A, if the

equation (A − λ)ψ = 0 has a non-zero solution ψ ∈ D(A).

Definition 1.12. The discrete spectrum of an operator A is

σdisc (A) = {λ ∈ C | λ is an isolated eigenvalue of A with finite multiplicity},

isolated meaning that some neighborhood of λ is disjoint from the rest of σ(A).

Definition 1.13. The essential spectrum of an operator is

σess (A) = σ(A) \ σdisc (A).

Theorem 1.3. [10] If A = A∗ then σ(A) ⊂ R.

Definition 1.14. The ratio

(ψ, Aψ)
, (1.3)
(ψ, ψ)

7
where ψ ∈ D(A) is called the Rayleigh quotient.

Let us note that if ||ψ|| = 1, the Rayleigh quotient becomes (ψ, Aψ).
Let T be a self-adjoint operator on H as previously defined. We can apply vari-
ational techniques to derive a characterization of the spectrum of T , that is, to find
its eigenvalues.

Theorem 1.4. [10] Given T as above with spectrum σ(T ) ⊂ [a, ∞), then

(ψ, T ψ)
≥a
(ψ, ψ)

for all ψ ∈ D(T ).

The above theorem states that the Rayleigh quotient is bounded below.

Theorem 1.5. [10] Let S(ψ) = (ψ, T ψ) for ψ ∈ D(T ) with ||ψ|| = 1. Then
inf σ(T ) = inf S. Moreover, λ = inf σ(T ) is an eigenvalue of T if and only if there is
a minimizer ψ ∈ D(T ) for S(ψ) such that ||ψ|| = 1.

The above result leads to the Ritz variational principle, given that for any ψ ∈
D(T ),
(ψ, T ψ) ≥ λ = inf σ(T )

and equality holds if and only if T ψ = λψ, this particular ψ is the corresponding
eigenfunction, for all other ψ ∈ H, all values of (ψ, T ψ) are upper bounds to the
operator’s eigenvalues.

Theorem 1.6. (Min-max principle) [10] The operator T has at least n eigenvalues
(counting multiplicities) less than inf σess (T ), if and only if λn < inf σess (T ) where

λn = inf max (ψ, T ψ) .

{Dn ∈D(T ) | dim(Dn )=n} {ψ∈X | ||ψ||=1}

In this case, the n-th eigenvalue is exactly λn .

8
Theorem 1.7. (The Rayleigh-Ritz theorem) [19] Let T be a semibounded
self-adjoint operator. let Dn ⊂ D(T ) be an n-dimensional subspace, and let P be the
orthogonal projection onto Dn . Let TDn = P T P . Let λ
b1 , λ
b2 , . . . , λ
bn be the eigenvalues
b1 ≤ λ
of TDn  Dn , ordered by λ b2 ≤ . . . ≤ λ
bn . Then

λi (T ) ≤ λ
bi , i = 1, . . . , n.

In particular, if T has eigenvalues (counting multiplicity) λ1 , . . . , λk at the bottom of

its spectrum with λ1 ≤ . . . ≤ λk , then

λi ≤ λ
bi , i = 1, . . . , min(k, n).

It is important to point out that the min-max principle can be used to find up-
per estimates for the eigenvalues of operators, thus having both quantitative and
qualitative consequences.
In practice, we simply analyze T in an N -dimensional linear space
DN = span{φ1 , . . . , φN }, where {φi }N
i is an orthonormal basis for DN . The eigen-

values of the N × N matrix [(φi , T φj )] then provide upper bounds to the first N
eigenvalues of T . Very often, as N is increased, these upper approximations steadily
improve. We shall use this technique in later chapters.

1.3 Schrödinger operators

Let H = L2 (Rd ) with inner product (ψ, φ) as defined in (1.2), at the beginning of this
chapter. We can write equation (1.1) as a linear operator.

Definition 1.15. The linear operator H : D(H) ⊂ H → H defined by

H = −∆ + V (1.4)

9
is called a Schrödinger operator; where −∆ is the Laplacian in d dimensions, and
V : Rd → R a given potential corresponding to a quantum mechanical problem.

It is important to know for which potentials V , the Schrödinger operator is self-

adjoint. The main purpose is also to characterize the spectrum σ(H) of the operator
H, because that gives information about the nature of the solutions of the main
Schrödinger equation.
In the following section we write a series of important results related to this
operator.

Theorem 1.8. [10] Let V (x) be a continuous function on Rd satisfying V (x) ≥ 0 and
V (x) → ∞ as |x| → ∞, then

1. H defined as in (1.4) is self-adjoint on H.

2. σ(H) consists of isolated eigenvalues {λi }∞

i=1 with λi → ∞ as n → ∞.

Theorem 1.9. [10] Let V (x) be a continuous function on Rd satisfying V (x) → 0 as

|x| → ∞, then

1. H defined as in (1.4) is self-adjoint on H.

2. σess (H) = [0, ∞), which means that H can have only negative isolated eigenval-
ues, possibly accumulating at 0.

Theorem 1.10. [10] Let Λ be a cube in Rd , and V a continuous function on Λ. Then

the Schrödinger operator H = −∆ + V , acting on the space L2 (Λ) with Dirichlet
boundary conditions has purely discrete spectrum, accumulating at ∞.

The Dirichlet boundary conditions are given by ψ|∂Λ = 0, which means that
ψ = 0 outside Λ. We shall refer to such a Sturm-Liouville problem by the term
“particle-in-a-box”.

10
 The previous results give us very valuable information in the study of the solutions
of different problems of quantum mechanics corresponding to various potentials. The
eigenvectors associated with the discrete spectrum of H are called bound states, while
the points in the discrete spectrum are called bound state energies, or energy levels.
We can use the variational characterization of the spectrum described in the pre-
vious section to analyze a series of problems.

1.3.1 The dimension d

The most common quantum mechanical problems are usually posed in dimension
d = 1, 2, 3.
If the dimension is d = 1 the two parts of the Schrödinger operator become
∂2
∆= ∂x2
, and V : R → R.
When d > 1 the structure of the Laplacian becomes more complicated. In order
to work in higher dimensions, we transform the problem from cartesian coordinates
into a more appropriate system. Let x ∈ Rd be x = (x1 , . . . , xd ), we can transform
this vector into another vector ρ = (r, θ1 , . . . , θd−1 ), where r = ||x||. Then the we
assume that the wave function is now given by

Ψ` (ρ) = ψ(r)Y` (θ1 , . . . , θd−1 ), (1.5)

with ψ(r) being the spherically symmetric factor, and Y` the spherical harmonic
factor, where ` = 0, 1, 2, . . .. The derivation of this change of variables can be found
in Sommerfeld [22].
The above analysis is particularly suitable for central potentials given as functions
of r = ||x|| rather than as functions of the individual components of x ∈ Rd . In these
cases we denote the potential function as V (r).
Given a spherically symmetric potential V (r) in a d-dimensional space, if we

11
change the system of coordinates described above, and remove the spherical harmonic
factor, as shown in Hall, et al. [13], we get the following radial Schrödinger equation

d2 ψ d − 1 dψ `(` + d − 2)
− − + ψ + V (r)ψ = Eψ. (1.6)
dr2 r dr r2

A correspondence can now be made with a problem on the half line in one dimen-
sion, we define the radial wave function

R(r) = r(d−1)/2 ψ(r), R(0) = 0.

We can then rewrite equation (1.6) as

d2 R
− + U R = ER, (1.7)
dr2

with effective potential

(2` + d − 1)(2` + d − 3)
U (r) = V (r) + . (1.8)
4r2

Given the above analysis is now clear how one can work with central potentials in
dimensions d > 1. We also note that the above equation (1.7) is in the form of a
Sturm-Liouville problem, but on the semi-infinite interval [0, ∞). We shall return to
this in chapter 2.

1.3.2 Examples

Clearly, the family of potentials which have exact analytical solutions is very small
compared to the family of those that do not. We now exhibit a series of examples that
have exact solutions. These examples are of utmost importance because they provide
us with exactly soluble test problems for our variational methods: if the method gives

12
good results for these, we can apply it to others with some degree of confidence.

Harmonic oscillator. The harmonic oscillator is a problem in quantum mechanics

that is linked to the classical spring, as it can be thought of as a particle having the
oscillating behaviour of a spring-mass system.
The one-dimensional harmonic oscillator is defined by the potential V (χ) = ωχ2 .
As mentioned before, using a scaling argument we can reduce this problem to a
family of problems related by a factor to the potential V (x) = x2 . We simply use the
transformation χ = sx, and substitute it in equation

d2
− ψ + (ωχ)2 ψ = Eχ ψ,
dχ2

obtaining the new equation

d2
 
1
− 2 ψ + ωs4 x2 ψ = Eχ ψ,
s2 dx

if we then find s such that ωs4 = 1, we will get the relation

1
ω 4 Ex ψ = Eχ ψ

Therefore we need only to solve the following differential equation defined by the
Schrödinger operator
d2
− ψ + x2 ψ = Eψ (1.9)
dx2

with solutions
x2
ψn (x) = Hn (x)e− 2 , (1.10)

where Hn is the Hermite polynomial of order n. The energy levels are given by

En = 2n + 1, n = 0, 1, 2, . . . (1.11)

13
 Figure 1.1: Harmonic oscillator

We can also find the solutions for this problem in dimension d = 3, studied in
Griffiths [9] and Greiner [8], where the symmetric potential is V (r) = r2 , and the
equation needed to solve is the following

d2 R
− + U R = ER, (1.12)
dr2

with effective potential

`(` + 1)
U (r) = r2 + . (1.13)
r2

and energy levels

En` = 4n + 2` − 1, (1.14)

where ` = 0, 1, 2, . . ., and n = 1, 2, 3, . . ..
Using algebra we can extend to the problem for dimensions d ≥ 2. This is done
comparing the effective potentials, the one for dimension d = 3 given by (1.13) and

(2`0 + d − 1)(2`0 + d − 3)
U (r) = r2 + . (1.15)
4r2

for dimension d.

14
 We set the equation

  
0 d−1 0 d−3
`(` + 1) = ` + ` +
2 2

in order to solve for `, obtaining

d 3
` = `0 + − ,
2 2

and substituting back into the values of the energy (1.14), thus obtaining

En`d = 4n + 2`0 + d − 4, (1.16)

where `0 = 0, 1, 2, . . . denotes de angular momentum of the d-dimensional problem.

Hydrogen-like atoms. The electron in the hydrogen atom is bound by the Coulumb
potential. The hydrogen atom is modeled as an infinitely heavy and fixed proton with
an electron revolving around it. The potential that defines this problem is given by
e 2
V (x) = − |r| with r = ||x|| for x ∈ Rd .
If d = 1 the problem of the hydrogen atom is as interesting as it is troublesome,
given its nature, the singularity splits the space in two pieces acting as a barrier.
The solutions to this problem are not trivial and require a thorough analysis of the
geometry of the problem as studied by [2].
Thus, the hydrogen atom is usually studied in 3 dimensions, [9], [8]. The discrete
spectrum of this problem in d = 3 is given by the energy levels

e2
En` = − , (1.17)
4(n + `)2

where ` = 0, 1, 2, . . ., and n = 1, 2, 3, . . ..
We can also scale and extend this problem to the d-dimensional case using the

15
 Figure 1.2: Hydrogen atom

arguments and algebraic analysis as in the above example. In this case we get the
following energy levels
e2
En`d = −
, (1.18)
d 3 2
4 n+`+ 2
− 2

again with ` = 0, 1, 2, . . ., and n = 1, 2, 3, . . ..

The above examples are discussed as they will serve as test problems in the following
chapters. For more insight on the solutions for this problems see literature from
Greiner [8], Griffiths [9], Gustafson and Sigal [10], and Hannabus [14].

16
Chapter 2

Sturm-Liouville Problems

2.1 Introduction

Sturm-Liouville systems arise in a large number of physical problems. They are one-
dimensional models of oscillating systems, and therefore we can also relate the notion
of energy to these particular problems.
The general form of a classical Sturm-Liouville problem is a second-order real
differential equation given by

 
d du
− p(x) + q(x)u = λw(x)u (2.1)
dx dx

defined on a finite or infinite interval a < x < b with boundary conditions.

It is obvious that the functions p, q and w are relevant in the analysis of this
differential equation. For example, if we assume that p and w are constant, without
any loss of generality we can use a scaling argument and remove them from equation.
This is, if we divide (2.1) by p and let t = γx be the new independent variable, where
1
γ = (w/p) 2 , then we obtain

d2 u
− + Q(t)u = λu, (2.2)
dt2

17
where Q(t) = q(x)/w. In fact, equation (2.2) is known as the Liouville or Schrödinger
normal form. Even more, by means of the Liouville transformation that we will de-
scribe in the section 2.3, the general Sturm-Liouville problem (2.1) can be transformed
into this normal form.
From now on we will refer to these kind of problems as Sturm-Liouville problems
or SLP.

Definition 2.1. We say that the SLP (2.1) is regular if

1. a and b are finite.

2. p, q and w are defined on the closed interval [a, b] and are continuous (maybe
except for a finite number of discontinuities), with p and w strictly positive.

3. The regular boundary conditions

a1 u(a) = a2 p(a)u0 (a)

(2.3)
0
b1 u(b) = b2 p(b)u (b)

are imposed at the end points of the interval, where a1 , a2 , b1 and b2 are real
numbers different from zero.

Our interest in regular Sturm-Liouville problems will be justified by the application

we shall use in the following chapters.
Given equation (2.1), for the rest of this chapter we will assume that the functions
p, q, w : R → R defined on the interval [a, b] are piecewise continuous, and that p and
w are strictly positive. If we take p, q and w to be complex function, then these
problems will no longer in the Sturm-Liouville form. The results from this chapter
are mainly studied by Braun [3], Coddington et al. [5], Pryce [18], and Sagan [20].

18
2.2 Sturm-Liouville theory

We can rewrite the Sturm-Liouville equation as the homogeneous equation

− (p(x)u0 ) 0 + (q(x) − λw(x)) u = 0. (2.4)

This is a second-order linear differential equation; it is not always self-evident that

the solution to this kind of equations is unique or if it even exists. We resort to a
familiar result from the theory of differential equations.

Theorem 2.1. [3]

Given the equation
d2 y dy
2
+ g(t) + h(t)y = 0 (2.5)
dt dt

with initial conditions

0
y(t0 ) = y0 , y (t0 ) = y00 . (2.6)

Let the functions g(t) and h(t) be continuous functions in the open interval α < t < β.
Then, there exists one and only one function y(t) satisfying the differential equation
(2.5) and the prescribed initial conditions (2.6) on the entire interval α < t < β. In
particular, any solution y = y(t) of (2.5) which satisfies y(t0 ) = 0 and y 0 (t0 ) = 0 at
the time t = t0 must be identically zero.

Although this theorem refers to the homogeneous case only, it can be extended to
the non-homogeneous case, [3] and [5].
From the above theorem if we let p, q and w be as we defined in the previous
section on the interval [a, b], the Sturm-Liouville equation (2.4) has unique solutions
satisfying the initial conditions

u(c) = uc , (pu0 (c)) = vc ,

19
for c ∈ [a, b].
In the case of the Sturm-Liouville problems local solutions will not suffice: we need
global solutions that satisfy boundary conditions, not only initial solutions. This is
not a trivial problem, and further analysis is required as the solutions are related to
the eigenvalues and the eigenfunctions.

2.2.1 Operator form of SLP

We can relate the Sturm-Liouville theory to the operator theory from chapter 1.
Because (2.1) is a linear equation, we can define the differential operator

   
1 d d
L= − p(x) + q(x) (2.7)
w(x) dx dx

on the interval a < x < b. The domain of this operator is contained in the set of
admissible solutions for the Sturm-Liouville problems, which are the square integrable
functions with respect to the weight function w

 Z b 
2 2
L ([a, b]) = u : [a, b] → C | |u(x)| w(x)dx < ∞ .
a

The above space has the weighted inner product defined as

Z b
(u, v) = uvwdx, (2.8)
a

v, again, representing the complex conjugate of v.

One of the properties of L is that it is a linear operator, following directly from
its differential and multiplicative form, and it satisfies

L(αu + βv) = αLu + βLv.

20
 Another fundamental property of this operator is that it is self-adjoint and sym-
metric with respect to the weight function w(x), implying that with the given inner
product (2.8) the following equality holds true

(Lu, v) = (u, Lv) .

Lemma 2.1. (Green’s Identity) [18] Let u, v ∈ D(L) be twice differentiable func-
tions defined on [a, b], then

Z b
b
(Luv − uLv) wdx = [puv 0 − pu0 v]a .
a

The above lemma is useful to solve this boundary condition problems. In partic-
ular, Green’s identity relies on the fact that L is self-adjoint, and if u and v satisfy
the boundary conditions in (2.3), we have that [puv 0 − pu0 v]ba = 0.

There exists a different and equivalent approach to the operator theory of the
Sturm-Liouville problems using the operator form

 
d d
L=− p(x) + q(x), (2.9)
dx dx

and the inner product defined in (1.2), studied by Sagan [20].

2.2.2 Eigenvalues and Eigenfunctions of the SLP

We can now pose equation (2.1) as the eigenvalue problem

Lu = λu (2.10)

where u ∈ D(L) and it satisfies the conditions (2.3).

We have the following important results.

21
Property 2.1. [18]
Let u be as above and the functions p, q and w that define the operator L in (2.7) be
piecewise continuous on [a, b], with p, w > 0, then:

1. The eigenvalues of the Sturm-Liouville problem (2.10) are real.

2. The eigenfunctions belonging to distinct eigenvalues are orthogonal with respect

to the inner product (2.8).

Theorem 2.2. [5], [18], [20]

For a regular SLP the following statements hold true:

1. The eigenvalues λk are simple, this is, there do not exist two linearly independent
eigenfunctions with the same value.

2. The eigenvalues can be ordered in an increasing sequence λ0 < λ1 < λ2 < . . .,

and with this labeling the eigenfunction uk corresponding to the eigenvalue λk
has exactly k zeros on the open interval (a, b).

3. The set of eigenfunctions {uk } form a complete orthogonal set of functions

over (a, b) with respect to the inner product (2.8). That is, any function f ∈
L2 ([a, b]) can be represented on (a, b) by its Fourier series with respect to the
eigenfunctions
∞
X
f (x) ∼ ck uk (x),
k=0

(f,uk )
where ck = (uk ,uk )
.

2.3 Schrödinger normal form

If we take p = w = 1 all the above theory is equivalent to the theory for the
Schrödinger operator in dimension d = 1. We note that the normal Liouville form

22
(2.2) is equivalent to the one-dimensional Schrödinger equation that we studied in
chapter 1.
For dimension d = 1 we can rewrite equation (1.1) as

d2
− ψ + V (x)ψ = Eψ (2.11)
dx2

where ψ, V (x) and E represent the wave function, potential, and level of energy of
the system respectively.
As mentioned before, there are a number general Sturm-Liouville problems that
can be transformed into the Liouville normal form; we demonstrate this in the fol-
lowing subsection.

2.3.1 The Liouville transformation

We can apply several transformations to equation (2.1), with the functions p, q, w

satisfying the conditions above [18].

Independent variable transformation.

Let x = x(t), this makes (2.1) transform into a new Sturm-Liouville differential
equation that is of the form

 
d p(x(t)) du
− + q(x(t))x0 u = λw(x(t))x0 u, (2.12)
dt x0 dt

where we assume that the mapping x(t) : (α, β) → (a, b) is onto and that x0 has the
same sign over the interval α < t < β.

23
Dependent variable transformation.

Let u = m(x)v, where m(x) is a given function. If we substitute u in (2.1) we get

 
d d
− p mv + qmv = λwmv,
dx dx

which is not in self-adjoint form, though we can multiply by m both sides of the above
expression to obtain

pm2 v 0

0
+ − (pm0 ) 0 m + qm2 v = λwm2 v,


(2.13)

which is now a self adjoint problem.

Liouville’s transformation.

Using both transformations above we get the transformed equation,

 
d dv
− P + Qv = λW v, (2.14)
dt dt

with

pm2
P = 0 ,
x
Q = (−(pm0 )0 + qm)mx0 ,

W = wm2 x0 .

The boundary conditions are also transformed to

A1 v(α) = A2 (P v 0 )(α),
(2.15)
0
B1 v(β) = B2 (P v )(β).

24
where

A1 = (a1 m2 − a2 pm0 m)|x=a , A2 = a2 ,

B1 = (b1 m2 − b2 pm0 m)|x=b , B2 = b2 .

Furthermore, if p, q and w are such that w/p and q/w are defined, then the general
SLP in (2.1) is finally converted to the Liouville normal form or Schrödinger form

d2 v
− + Iv = λv, (2.16)
dt2

using Liouville’s transformation given by

Z r
w
t= dx,
p
(2.17)
− 14
m = (pw) .

This is an important result: thanks to this transformation it is possible to solve

the general SLP by first transforming to the Liouville normal form (2.16).

2.3.2 A numerical approach to solving the SLP

For some symmetric problems that can be transformed to the Sturm-Liouville form,
it is possible to develop what is called a shooting method to calculate its eigenvalues
and eigenfunctions.
Given the problem (2.16) we can solve it numerically defining it as an initial value
problem satisfying the left hand boundary conditions over a specific range, depending
on the nature of the problem. The differential equation is then solved for a sequence of
trial values of the eigenvalue λ. The eigenvalue is adjusted until the required number
of zeros is obtained and the right hand boundary conditions are met. We refer to
Theorem (2.2).

25
 Though the process is quite straightforward it is not always easy to implement such
a numerical program. But it opens the door to have another option to approximate
the eigenvalues which we can later compare to the original method studied in this
work, that is the variational technique. For an example see Appendix B.

2.4 The particle-in-a-box

The infinite square well potential, also known as the one-dimensional particle-in-a-
box problem, is an important in quantum mechanics. The potential that defines the
Schrödinger operator is the following:

 0, if 0 ≤ x ≤ b;

V (x) =
 ∞, otherwise.


Figure 2.1: Infinite square well potential

The Schrödinger operator H defines the following equation which is also a regular
Sturm-Liouville problem, with ψ defined on the interval 0 ≤ x ≤ b and boundary
conditions ψ(0) = ψ(b) = 0, and with

d2
− ψ = Eψ. (2.18)
dx2

26
Solving this equation we find that for n = 1, 2, . . . we have the following energy levels
or eigenvalues
 nπ 2
En = , (2.19)
b

with corresponding wave functions or eigenfunctions

r
2  nπx 
ψn (x) = sin . (2.20)
b b

We can verify the importance of theorem (2.2) with this example. The energy
levels form an increasing sequence, and as these levels go up, each successive wave
function or state has one more node. The eigenfunctions form a complete orthonor-
mal set: from Fourier analysis we know that each ψ ∈ L2 ([0, b]) can be written
ψ = ∞ 2 2
P
n=1 cn ψn . We also note that L ([0, b]) ⊂ L (R), this basis might be useful

later on in the analysis of problems in a wider space.

27
Chapter 3

Variational Analysis for

Schrödinger Operators

3.1 Introduction

When facing a quantum mechanical problem, in order to approximate the values of

the ground state energy, we can apply the variational principle which characterizes
the discrete spectrum. There are a number of variational approaches that have been
widely used since the nineteenth century to solve physical problems, in particular,
the work of Lord Rayleigh and Walther Ritz opened up the doors to the use of these
techniques.
The simplest variational analysis consist in the following: given a particular prob-
lem with potential V (x), we define the Schrödinger operator H as in (1.1). By the
theory studied in chapter 1, the domain of this operator is D(H) ⊂ L2 (Rd ), and H
is self-adjoint. We then choose a normalized trial function ψ ∈ D(H), which may be
different from ψ0 , the ground state. Then we know that

E0 ≤ (ψ, Hψ) ,

28
where E0 is the ground-state energy. This means that by finding a suitable trial
function we can obtain an upper bound of the ground state energy [9].
Even more, if we choose ψ as a function of x and other parameters, we can
manipulate these parameters in order to obtain a better approximation, that is to
say, a lower upper bound. If we know the exact solution of a particular problem we
can compare results and study the accuracy of the variational estimate.

3.2 Variational method

Based on the Rayleigh-Ritz principle and theorem (1.6), we can develop a variational
method to approximate the energy levels of a problem.
Given a specific problem with a potential V (x), where the corresponding Schrödinger
operator H = −∆ + V is self-adjoint and bounded below, and its domain
D(H) ⊂ H = L2 (Rd ), we consider of the following steps leading to approximate
energy levels.

1. Basis. We choose an orthonormal variational basis B = {φ1 , φ2 , . . .} ⊂ H. The

functions belonging to the basis are dependent on one or more parameters. The basis
we use in this thesis is related to the solutions of the particle-in-a-box problem, though
the use of the basis and the number of free parameters depend on the dimension d of
the problem, this will be explained in the following sections.
The variational basis spans a Hilbert space B contained in H and may be finite
dimensional, say of dimension N . A general element ψ ∈ B may be written

N
X
ψ= ci φi (3.1)
i=1

where ci = (ψ, φi ) with i = 1, 2, . . . are the Fourier coefficients of the series expansion
of ψ. If the variational basis is infinite we can substitute ∞ for N .

29
2. Optimization. We suppose that ψ is the trial wave function solution to the
problem posed by H, we can then write ψ as a series of the variational basis B as in
(3.1).
The variational problem now becomes that of minimizing the energies or eigen-
values of H with respect to the free parameter or parameters of the above functions.
This is equivalent to calculating the eigenvalues of the Hamiltonian matrix H and
minimizing them over the same parameter set. The Hamiltonian matrix is defined as

 
 (φ1 , Hφ1 ) (φ1 , Hφ2 ) . . . (φ1 , HφN ) 
 
 (φ2 , Hφ1 ) (φ2 , Hφ2 ) . . . (φ2 , HφN ) 
H= , (3.2)
 
.. .. ... .
..

 . . 

 
(φN , Hφ1 ) (φN , Hφ2 ) . . . (φN , HφN )

and, as a consequence of the linearity of H, we can separate this matrix into the sum
of two parts, corresponding to the kinetic and potential energies. Thus

H = K + P, (3.3)

with
 
 (φ1 , −∆φ1 ) (φ1 , −∆φ2 ) . . . (φ1 , −∆φN ) 
 
 (φ2 , −∆φ1 ) (φ2 , −∆φ2 ) . . . (φ2 , −∆φN ) 
K= ,
 
.. .. ... .
..

 . . 

 
(φN , −∆φ1 ) (φN , −∆φ2 ) . . . (φN , −∆φN )
  (3.4)
 (φ1 , V φ1 ) (φ1 , V φ2 ) . . . (φ1 , V φN ) 
 
 (φ2 , V φ1 ) (φ2 , V φ2 ) . . . (φ2 , V φN ) 
P= .
 
.. .. ... ..

 . . . 

 
(φN , V φ1 ) (φN , V φ2 ) . . . (φN , V φN )

30
We note that the kinetic component of the matrix H is the same regardless of the
potential for any given problem, this is helpful in a numerical application of the
method.
By the variational principle, we know that the eigenvalues of H will be upper
bounds to the eigenvalues of the operator H. That is, if the eigenvalues of H are
given by ε1 ≤ ε2 ≤ . . ., and the energy levels or eigenvalues of H are given by
E1 ≤ E2 ≤ . . ., then the following hold true

E1 ≤ ε1 ,

E2 ≤ ε2 ,
..
.

Furthermore, if we are able to find the eigenvectors of H, these will be related

to the eigenfunctions or wave functions that are solutions to H in the sense that if


νk = ck1 , ck2 , . . . is the eigenvector corresponding to the k-th eigenvalue of H, with
k = 1, 2, . . ., then the k-th approximate eigenfunction ψk can be written as

N
X
ψk = cki φi . (3.5)
i=1

This means that the components of the eigenvector are the Fourier coefficients of the
series expansion, [11], [15].

3. Computational implementation. We face several obstacles in trying to imple-

ment a computational algorithm to calculate the above eigenvalues and eigenfunctions
numerically.
First, in order to implement a numerical approach to the above variational method,
we need to work in a subspace of H, implying that we can only get an approximation
of the results, not the exact solution, as we cannot work with a suitable infinite

31
basis. Then the dimension of the matrix H is truncated and denoted by the letter
N . The numerical accuracy of the method depends greatly on how large this value
is. Secondly, we depend on a computer program to solve the eigenvalue problem for
H. And last, the optimization over the parameters can be very complicated as well.
The implementation of this method will be discussed in chapter 4 through an
example.

3.3 The basis for the analysis in dimension d = 1

As stated before, the problems in dimension d = 1 can be defined by any potential

such that x ∈ R. The basis we will use for the variational analysis is the set of
solutions that are obtained from the particle-in-a-box problem. By applying a simple
transformation, we shift the box from the interval [0, b] to a new interval [−L, L],
where L > 0 is introduced as our variational parameter.
From the particle-in-a-box problem in chapter 2, let b = 1 without loss of gener-
ality. We have that the wave functions given by (2.20) can be written as

√
ψn (χ) = 2 sin (nπχ) ,

if we apply the transformation χ = x/2L + 1/2, we obtain the shifted functions

√
  
x 1
ψn (x) = 2 sin nπ + , (3.6)
2L 2

where χ ∈ [0, 1] and x ∈ [−L, L].

Even though we know that the function ψn was normalized, after applying the
transformation it loses this property, so we have to normalize it once more in order
for our base to be orthonormal.

32
 Figure 3.1: Shifted particle-in-a-box

Let
  
1 x 1
φn (x) = √ sin (n + 1)π + (3.7)
L 2L 2

be the functions that define the variational basis B = {φ1 , φ2 , . . .}. This basis has
the following properties that we require:

1. B is a complete set, in the sense that a function in the same Hilbert space


L2 ([−L, L]) ⊂ L2 Rd can be expressed as a linear combination of B.

2. The members of B are orthonormal, this is


 1, if n = m;

(φn , φm ) =
 0, otherwise.


3. The functions are even or odd with respect to the origin depending the index.
φi is even if i = 0, 2, . . ., or it is odd if i = 1, 3, . . ..

Then, with the above basis we can implement a program to calculate the eigen-
values of the matrix H as in (3.2). In this case, the calculation of the matrices H, K,
and P depends on the members of B.
The implementation of such a program, however, can be simpler if we analyze the
properties of V (x) first. We can have three different choices to build up the matrix

33
H thanks to the above basis, for any suitable potential, for even potentials, and for
polynomial potentials. The cases where the potentials are either even functions or
polynomials present interesting opportunities when it comes to the implementation
of a computer algorithm to attack the problem.
If the potential is even, in order to construct the H matrix, we can split it up in
two different matrices, the even part and the odd part, this is thanks to the property
of the functions in the basis B of being alternately even and odd: when we multiply
by the potential the functions in the basis will retain their even or odd property.
If the potential is a polynomial function, we have a very interesting result.

Property 3.1. Let Px and Px2 be the matrices corresponding to the potentials
V1 (x) = x and V2 (x) = x2 as defined in (3.4) respectively, then

Px2 = P2x . (3.8)

Proof. Given that

 
 (φ1 , xφ1 ) (φ1 , xφ2 ) . . . 
 
Px = 
 (φ2 , xφ1 ) (φ2 , xφ2 ) . . . ,

 .. .. ...

. .

and that  
2 2
 (φ1 , x φ1 ) (φ1 , x φ2 ) . . . 
 
P x2 = (φ
 2 , x 2
φ1 ) (φ2 , x 2
φ2 ) . . . ,

 .. .. ...

. .
P∞
we know that (P2x )ij = (φi , xφr ) (φr , xφj ), and that (Px2 )ij = (φi , x2 φj )
r=1

Then we need to prove that (φi , x2 φj ) = ∞

P
r=1 (φi , xφr ) (φr , xφj ).

As B is a complete basis for L2 ([−L, L]) we can write the function xφj as the following

34
series
∞
X
xφj = φr (φr , xφj ) (3.9)
r=1

for j = 1, 2, . . ..
If we multiply (3.9) by x we obtain

∞
!
X
x2 φj = x(xφj ) = x φr (φr , xφj )
r=1
∞
X
= xφr (φr , xφj ) .
r=1

Then, the inner product

∞
!
X
φi , x2 φj =


φi , xφr (φr , xφj ) ,
r=1

by the linearity of the inner product we get that

∞

X
φi , x2 φj = (φi , xφr (φr , xφj ))
r=1
∞
X
= (φi , xφr ) (φr , xφj ) .
r=1

Thus (Px2 )ij = (P2x )ij .

2.

It is straightforward to extend the above property to the power n by using an

induction argument. Thus, in an infinite basis, Pxn = Pnx .
The above result states that these matrices are equal for the infinite case, but if
we are working in a N -dimensional subspace, then the equality is not true, we can
only say that in a numerical analysis Pnx is only an approximation for Pxn .

35
3.4 The basis for the analysis in dimension d ≥ 2

We can use a similar analysis from the previous section in the case of higher di-
mensions, as we explained in chapter 1, in d ≥ 2, the introduction of a set of d − 1
coordinate angles, and the used of a suitable radial transformation, allows us to recast
the residual problem in 1 dimension, namely (1.7) is this one-dimensional problem
for which we shall use the basis functions that are described below.
For d ≥ 2, we shall consider central potentials, and for the analysis of this problem,
we need to work with the effective potential given by (1.8). In fact, the matrix H has
one more component which corresponds to the effective potential.
Let Q(`, d) = 14 (2` + d − 2) (2` + d − 3), and the matrix U defined as follows:

 
φ1 , r12 φ1 φ1 , r12 φ2



 ... 
 
1
φ2 , r12 φ2



U=
 φ2 , r2 φ1 ... 
 .. .. ..

. . .

Then
H = K + P + Q(`, d)U. (3.10)

Another challenge is added to any implementation of the variational method be-

cause we now have a potential with singular behaviour which needs to be take into
account. Not only this, as we are working with spherically symmetric potentials,
these too can be singular. The singularity of this potential becomes a problem when
we need to calculate H as the inner product (1.2) might not be defined for certain
potentials and a given set of variational parameters.

Non-singular and weakly singular potentials. Examples of these potentials

were given in section 1.3.2, the harmonic oscillator and the hydrogen atom in dimen-
sions d ≥ 2. For this kind of problem we choose the functions in our variational basis

36
B to be exactly the set of solutions from the particle-in-a-box given by

r
2  nπr 
φn (r) = sin , (3.11)
b b

with r ∈ [0, b], and n = 1, 2, . . .. That is to say, we use variational trial functions in
L2 ([0, b]) ⊂ L2 ([0, ∞)).
This basis satisfies again the properties mentioned before such as orthonormality
and completeness for L2 ([0, b]). Here the box size b is a variational parameter.

Highly singular potentials. These potentials are functions that have at least one
term of the form 1/rs with s > 2. In these cases we cannot use the above basis
because the inner product (φi , Hφj ) is not defined on [0, b], therefore we will use, in
addition to b, a parameter a > 0 so that the variational basis is in L2 ([a, b]).
As in the case of dimension d = 1 we apply a transformation that shifts the
interval [0, b] to a new interval [a, b] where a, b > 0 are the variational parameters.
r−a
The transformation is given by χ = b−a
, and we obtain the shifted functions

√
  
r−a
ψn (r) = 2 sin nπ , (3.12)
b−a

where χ ∈ [0, 1] and r ∈ [a, b].

Normalizing (3.12) we obtain the functions

r   
2 r−a
ψn (r) = sin nπ , (3.13)
b−a b−a

with n = 1, 2, . . ..

37
Chapter 4

Results

4.1 Implementation of the method

In order to use the variational method to approximate the discrete spectrum of the
Schrödinger operators studied in chapter 3, we implemented the following analysis,
using the mathematical software Maple 11.
We will explain the main idea of the implementation by using the example of the
harmonic oscillator in dimension d = 1. The implementation for other problems is
very similar to this, and we shall discuss it at the end of this section.
Given the basis B = {φn }n , with φn defined as in (3.7), we need to calculate the
matrices K and P, corresponding to the problem of the harmonic oscillator with the
potential V (x) = x2 . The components of the matrices are given analytically by the
expressions: 
i2 π 2
d2 , if i = j;
  

4L2
Kij = φi , − 2 φj =
dx  0,
 otherwise,

and 
L2 (i2 π 2 −6)
, if i = j;


3i2 π 2
Pij = (φi , V (x)φj ) =
4L2 (
4(−1)i+j ij+4ij )


π 2 (i2 −j 2 )2
, otherwise.

38
The Hamiltonian matrix is then given by the sum of the above two matrices, as in
(3.3).
For a numerical analysis the above matrices need to be finite, so the first thing to
do is to choose the dimension N , and then i, j = 1, 2, . . . , N . This means that we can
build numerical matrices of dimension N which depend on a fixed box size L, which
becomes a variational parameter. For each value of L we can obtain the eigenvalues
using the built in functions of the mathematical software, in our case Maple 11. The
problem then arises when we need to find which is the specific L that will give us an
acceptable approximation of the energy levels, in the sense that each eigenvalue we
obtain from the variational analysis is the lowest upper bound to the exact solutions.
In the case of the harmonic oscillator we can compare the results we obtain from this
numerical program with the exact solutions given by (1.11).
For the optimization process over L, we take a simple approach, justified by the
very flat region of a graph of the energy levels against the values of the variational
parameter L, near the minimum, namely, we calculate the eigenvalues of H for a
sequence of values of L. We chose an initial value of L = L0 and a step size s, the
next value of L would be equal to L0 +s, and so on, for a certain number of iterations.
The matrix would depend on the parameter Lk = L0 + ks for k = 0, 1, 2, . . . , M ,
where M is the final number of iterations. We can make an educated guess to choose
a suitable L0 depending on each problem, or choose a large value of s first, to get a
rough idea of an L that would be close to the minimum.
As we already know the exact solutions of the harmonic oscillator, we know that
its wave functions decay swiftly to 0 as x grows. Thus, we assume that the value of
an optimal L will not be too large, but this might not be the case for other problems.
The value of L might also differ from one state to another, that is, the value needed
to optimize the energy of the ground state might be different from the following state,
and so on.

39
 When we find an acceptable value of L, the one that minimizes the given eigen-
value, we can also obtain the corresponding eigenvectors of the matrix, and use this
to graph the approximate eigenfunction as a linear combination of the basis, as stated
in (3.5).
The implementation details for different problems depend on the potential and
on the dimension. For an example concerning potentials, the harmonic oscillator
potential is a simple function, and moreover, an even function: in this case the inner
product (φi , V (x)φj ) has an analytical expression that can be exactly calculated by
a program like Maple 11. If the potential is such that we need to use numerical
integration to calculate the inner product, the calculation of the matrix P becomes
slower and possibly less accurate.
On the other hand, we do have the approximation given in (3.8) by the relation
Pxn ≈ Pnx , which allows us to calculate the Hamiltonian matrix of a polynomial
potential such as V (x) = m k
P
k=0 ak x by the computation of only two matrices, K and

Px . Thus we have
m
X
H=K+ ak Pkx . (4.1)
k=0

We shall refer to this approach as the “polynomial approach” in future.

When d ≥ 2, we also have to be concerned about the angular momentum, and use
the effective potential which now contains a singular term, for this we use the bases
defined in section 3.4, depending whether the potentials are non-singular, weakly-
singular or highly-singular.

40
4.2 Dimension d = 1

4.2.1 The harmonic oscillator

We calculated the eigenvalues of the harmonic oscillator in dimension d = 1 described

in section 1.3.2. using the implementation described in the previous section. Since
the harmonic oscillator falls into the category of being a polynomial potential, we
also used the polynomial approach to analyse this problem. We then compared the
results obtained from both of these calculations to the exact solutions.
Table (4.1) shows the results obtained in the optimization process for approaches
using a matrix of dimension N = 50, and a step size s = 0.1. In this table n
represents the state, where n = 0 is the ground state. E is the exact solution for
the energy given by (1.11). EV represents the approximation to the eigenvalue using
the direct approach described in the previous section, LV represents the optimal
variational parameter, EP is the approximation using the polynomial approach, with
corresponding optimized variational parameter LP .

Table 4.1: Approximation of the energy levels of the harmonic oscillator in dimension d = 1.
n E EV LV EP LP
0 1 1.000000001 6.1 0.999999998 7.3
1 3 3.000000001 5.6 2.999999998 6.5
2 5 5.000000001 5.8 4.999999997 6.6
3 7 7.000000002 5.9 6.999999996 6.6
4 9 9.000000003 6.3 6.999999996 6.6
5 11 11.00000000 6.3 10.99999999 8.3
22 45 45.00000006 8.9 45.00000005 8.9
23 47 47.00000010 9.0 46.32075646 11.7
24 49 49.00000120 8.9 49.00000117 8.9
25 51 51.00000185 9.0 50.32353151 11.2
26 53 53.00001881 8.9 53.00001868 8.9
32 65 65.01513530 8.9 62.32028554 10.1
34 69 69.07941805 8.9 69.07905146 8.9
37 75 75.36273369 9.0 74.35832785 9.3
42 85 89.861473970 8.9 98.14418440 8.9

41
 For this case we have also plotted the energy levels E versus the parameter L,
as shown in figure (4.1): we observe that the energy levels as a function of L are
U-shaped and flat near the minima. Also, the values of L we get as minimizers for
each of the states are not that different from each other. From the graph we can
think of choosing only one value of L to obtain a “good” approximation for all the
eigenvalues considered. If N is large enough, say, bigger than 20, any value of L in
the range [6, 10] gives good approximations to the energy levels.

Figure 4.1: Energy vs. Parameter L for the harmonic oscillator in dimension d = 1.

Once the optimal L is determined, the eigenvectors of the first 3 states also ob-
tained and graphed with their corresponding eigenfunctions as defined in (3.5). Fig-
ures (4.2), (4.3), and (4.4), represent the trial wave functions obtained, each one
calculated with the values of L stated in table (4.1).
We note that the approximations obtained by both methods are good. After ex-
perimenting with different values of N we can confirm that the larger this value is the
better the approximations become, that is to say, closer to the exact solution. Also,
by this same observations we can say that up to N/2 states are good approximations

42
Figure 4.2: Trial wave function for the ground state ψ0 of the harmonic oscillator in dimension d = 1 with minimizer
L = 6.1.

Figure 4.3: Trial wave function for the first excited state ψ1 of the harmonic oscillator in dimension d = 1 with
minimizer L = 5.6.

in both approaches, the direct and the polynomial one. However, it is interesting to
note that the polynomial approach, as it involves yet another approximation, from
proposition (3.8), does not always give upper bounds. This is clear for some of the
results showed in table (4.1), for example states n = 25, 32, 37. In contrast we observe
that the variational approach always gives upper bounds even though N is not very
large. To overcome the problem of inaccuracy of the polynomial approach we need
only to increase the value of N to be greater than 100, this will give better results.

43
Figure 4.4: Trial wave function for the second excited state ψ2 of the harmonic oscillator in dimension d = 1 with
minimizer L = 5.8.

4.2.2 The potential V (x) = |x|

The function V (x) = |x| is symmetric and non-differentiable at the origin. This is a
linear potential that represents the attraction of a small particle to a large plate with
a hole in it under Newtonian gravity. The exact analytical solutions to this problem
are given in terms of the zeros of Airy functions and its first derivatives, studied in
Flügge [7] and shown in Abramowitz et al. [1].

Figure 4.5: The potential V (x) = |x|

It can be thought of as a non-harmonic oscillator for it represents an oscillating

behaviour, with the difference that the force has a sudden change of direction at
x = 0.

44
 In table (4.2) we show the results for this problem found by applying the varia-
tional method with a matrix of dimension N = 50, and a step size s = 0.25 for the
optimization process. Again, n represents the state. As there exist suitable expres-
sions for the exact solutions to this problem, we compare the results obtained in the
variational method to these. We can also use a shooting method which is known to
give good approximations for the energy levels. E and ES denote the exact energy lev-
els and those obtained by this shooting method respectively, while EV represents the
approximation to the eigenvalue using an implementation of the variational method,
and L is the minimum value of the variational parameter.

Table 4.2: Approximation of the energy levels of the potential V (x) = |x|.
n E ES EV L
0 1.01879297 1.01879305 1.01879344 6.25
1 2.33810741 2.33810752 2.33810745 8.00
2 3.24819758 3.24819770 3.24819827 8.25
3 4.08794944 4.08794959 4.08794960 9.25
4 4.82009921 4.82009938 4.82010046 9.75
5 5.52055983 5.52056000 5.52056026 10.50
6 6.16330736 6.16330750 6.16330960 10.75
7 6.78670809 6.78670828 6.78670915 11.50
8 7.37217726 7.37277509 7.37218128 12.00
9 7.94413359 7.94413380 7.94413594 12.50
10 8.48848673 8.48848690 8.48849407 12.75
11 9.02265085 9.02265108 9.02265603 13.50
12 9.53544905 9.53544923 9.53546280 13.75

Following the theory from chapter 3, and the observations made in the harmonic
oscillator the energy levels obtained by the variational methods are optimized upper
bounds to the energy levels. In this case, when we optimize over the variational pa-
rameter L, we obtained lower upper bounds, in fact, the goal was to obtain the lowest
upper bound. Here we show the best lower bound calculated using the optimization
process.
For the shooting method, as described in Appendix B, we need to solve an ordinary
differential equation numerically, which depends strongly on the software we are using

45
where we cannot control the accuracy of the results, therefore, the eigenvalues we
obtained by the shooting process are just approximations. We cannot claim how
accurate they are, but this is a matter for further numerical analysis which is not
the main point in this thesis. Though the approximations obtained by the shooting
method are very close to the exact solutions, if we face a problem that has no exact
solutions we can only use this approximations to get an idea about the eigenvalues,
not as a point of comparison to determine how accurate the results obtained from the
variational method are.

4.2.3 Polynomial potential: V (x) = x4

We considered the potential V (x) = x4 to test the accuracy of the method using
the polynomial approach, and compared it to the direct approach of the variational
method and the results obtained from the shooting method as well.
We used a matrix of dimension N = 50, and a step size of s = 0.5. Again
the subindexes S, V , and P represent the results obtained with the shooting, the
variational, and the polynomial methods respectively.

Table 4.3: Approximation of the energy levels of the potential V (x) = x4 .

n ES EV LV EP LP
0 1.060361945 1.060362090 3.5 1.060362089 7.5
1 3.799673304 3.799673024 4.0 3.799673027 6.5
2 7.455698365 7.455697938 4.0 7.455697934 4.5
3 11.64474628 11.64474551 4.0 11.64474550 5.0
4 16.26182677 16.26182602 4.0 16.26182601 4.0
5 21.23837423 21.23837292 4.0 21.23837290 7.5
6 26.52847225 26.52847118 4.5 26.52847117 4.0
7 32.09859965 32.09859772 4.0 32.09859769 6.5
8 37.92300243 37.92300103 4.0 37.92300029 7.5
9 43.98116073 43.98115811 4.0 43.98114584 7.5

From this results we note that in some cases, the eigenvalues obtained from the
variational method are smaller than the ones obtained from the shooting method.

46
This reinforces the notion that the approximations obtained by the shooting method
cannot be compared to the ones obtained by the variational one.
We do observe that for the first states, the approximations obtained from the
variational and polynomial approaches are close. The polynomial approach has the
advantage that it is fast for an optimization process compared to the variational
method, therefore if we choose a large N we can get good approximations using it.

4.2.4 Polynomial potential: V (x) = x2 + x4

The potential V (x) = x2 + x4 is a special case of the quartic anharmonic oscillator.

There interest in this problem has a long history. Simon has written an extensive
review [21]. We consider the case of dimension d = 1, though this problem can also
be studied in higher dimensions as presented, for example, by Hall [12].

Figure 4.6: Quartic anharmonic potential

Table (4.4) reflects the results after applying the method to a matrix of dimension
N = 50, and using a step size s = 1 for the optimization process. Again, n represents
the state, and because this kind of potentials do not have exact solutions, yet again
we need to make use of a shooting method to approximate the solutions given by
ES , EV , as usual, represents the approximation to the eigenvalue using the direct

47
approach, with the respect value LV of the optimized variational parameter, and
EP represents the approximation using the polynomial approach, with corresponding
optimized variational parameter LP .

Table 4.4: Approximation of the energy levels of the potential V (x) = x2 + x4 .

n ES EV LV EP LP
0 1.392351594 1.392351474 6.0 1.392351640 5.0
1 4.648813066 4.648812622 5.0 4.648812702 4.0
2 8.655050433 8.655049869 6.0 8.655049951 9.0
3 13.15680476 13.15680388 4.0 13.15680051 10.0
4 18.05755824 18.05755739 5.0 18.05751328 10.0
5 23.29744288 23.29744128 6.0 23.29743858 9.0
6 28.83533959 28.83533840 6.0 28.83525815 9.0
7 34.64085039 34.64084825 5.0 34.63779703 10.0
8 40.69038755 40.69038606 5.0 40.62769619 10.0
9 46.96501230 46.96500945 6.0 46.96212176 9.0

The results in this table confirm what we mentioned for the problem in the previ-
ous section, regarding the accuracy of the three methods used. However, calculating
these energies using the variational approach is computationally expensive, because
in order to construct the matrix H of dimension N we need to integrate numerically
for each component of P for each value of L. This makes us think that the polynomial
approach might be more useful, as it is faster, we need only to calculate the matrix
Px which has an analytical form depending on L, and multiply it by itself according
to the relation (4.1).
One of the problems is that after experimenting with L using this approach,
sometimes we get values for the higher excited states that are lower than the energy
levels of the problem, we show in figure (4.7) the behaviour of the eigenvalues in
function of L.
We can see from this that after L = 10, using the polynomial approach, the
numerical results might be not so reliable, this can be the explanation that this
approach has sometimes good results, but some other times goes even lower than
the shooting or exact results, as seen in previous examples. We also observe that

48
 Figure 4.7: Energy vs. L of the quartic anharmonic oscillator in dimension d = 1.

before this specific value of the variational parameter, the eigenvalues have a smooth
behaviour, if we choose an L before this critical value and take a large value of N ,
we then again obtain good results.

4.3 Higher dimensions

The analysis of dimensions d ≥ 2 is different from the one in dimension d = 1 in

the sense that we do not use the shifted solutions for the particle in a box. We now
consider radial wave functions, which means we need to do the analysis in the interval
[0, b] or [a, b], depending on how singular the problem is.

4.3.1 The harmonic oscillator

The energy values for the harmonic oscillator in dimensions d = 2, 3, 4, 5 and quantum
number ` = 0, 1, 2, 3 are calculated. We used a matrix of dimension N = 40 and a step
size of s = 0.25. This problem has known exact solutions (1.16), which can be used

49
for comparisons. Since the expression of each element of the matrices corresponding
to the kinetic and potential energies is given by an exact analytical function of L, the
calculations of these eigenvalues are quite fast.
We present two different tables of results. For all of them ` represents the quantum
number related to the angular momentum, n is the state (1 plus the number of nodes
in the radial function for a given `), E is the exact value of the energy levels, and EV
the variational approximation with minimal parameter b.

Table 4.5: Approximation of the energy levels of the harmonic oscillator in dimension d = 2.
` n E EV b
0 1 2 2.28622395 4.00
2 6 6.30816202 4.00
3 10 10.32728254 4.50
4 14 14.34185985 5.25
1 1 4 4.00073469 4.25
2 8 8.00185191 4.75
3 12 12.00337390 5.25
4 16 16.00524280 5.50
2 1 6 6.00000262 5.00
2 10 10.00001068 5.50
3 14 14.00002888 6.00
4 18 18.00005965 6.25

Table (4.5) shows case of dimension d = 2. It is clear that when ` = 0 we obtain

results that are not quite accurate. This is due to the behaviour of the effective
potential, which for this case is given from (1.15) by

1
U (r) = r2 − ,
4r2

and we can see in Figure (4.8). For this case the singular term makes the potential
tend to −∞ when r is close to 0. This shows that the original oscillating behaviour
might be lost and the matrix we get from the variational analysis is not as quite
accurate. This is a hint that we might have problems with singular problems, even
when they are weakly-singular. Here the difficulty is removed for ` > 0 (or d > 2)

50
since the effective potential U (r), generally given by (1.8), has a positive pole at r = 0
and the particle is confined in effectively a “soft box” between this pole and the rising
potential at large r. This point will be discussed further in the conclusion.

Figure 4.8: Harmonic oscillator in dimension d = 2, and ` = 0.

Table (4.6) shows a small sample of the results for dimensions d = 3, 4, 5 and
quantum numbers ` = 0, 1, 2, 4. Most of the results are very accurate, except for
some values of `.

51
 Table 4.6: Approximation of the energy levels of the harmonic oscillator in dimension d = 3, 4, 5.
d ` n E EV b
3 0 1 3 3.00000000 6.00
5 19 19.00000001 7.00
10 39 39.00000001 8.75
1 1 5 5.00007348 4.50
5 21 21.00167944 6.25
10 41 41.00907276 7.75
2 1 7 7.00000000 6.00
5 23 23.00000001 7.75
10 43 43.00000001 9.25
3 1 9 9.00000001 6.00
5 25 25.00000076 7.25
10 45 45.00002070 8.50
4 0 1 4 4.00073469 4.25
5 20 20.00745550 6.00
10 40 40.02454449 7.50
1 1 6 6.00000262 5.00
5 22 22.00011370 6.50
10 42 42.00094014 8.00
2 1 8 8.00000002 6.00
5 24 24.00000248 7.25
10 44 44.00004592 8.50
3 1 10 10.00000000 6.00
5 26 26.00000008 7.50
10 46 46.00000274 8.75
5 0 1 5 5.00007348 4.50
5 21 21.00167944 6.25
10 41 41.00907276 7.75
1 1 7 7.00000000 6.00
5 23 23.00000001 7.75
10 43 43.00000001 9.25
2 1 9 9.00000001 6.00
5 25 25.00000076 7.25
10 45 45.00002070 8.50
3 1 11 11.00000001 6.00
5 27 27.00000000 8.00
10 47 47.00000001 10.00

4.3.2 The hydrogen atom

The hydrogen atom is studied in section 1.3.2. We now show results obtained with
our method. Here the size of the matrix is N = 25, and the step size used in the

52
optimization process is s = 1, the step size is chosen so large because this weakly
bound system is all together large. The variational method is slow in this case, given
the singular nature of the problem and because for the matrix P (corresponding to
the potential energy) we need to integrate numerically for each term.

Table 4.7: Approximation of the energy levels of the hydrogen atom in dimension d = 3.
` n E EV b
0 1 -0.2500000000 -0.2494292776 13
2 -0.06250000000 -0.06173021569 32
3 -0.02777777778 -0.02682855454 57
4 -0.01562500000 -0.01457726183 90
1 1 -0.06250000000 -0.06231120892 33
2 -0.02777777778 -0.02747649731 60
3 -0.01562500000 -0.01526320869 94
4 -0.01000000000 -0.009656788911 143
2 1 -0.02777777778 -0.02777640178 75
2 -0.01562500000 -0.01561644406 108
3 -0.01000000000 -0.009970374676 146
4 -0.006944444444 -0.006872824074 189

The results we get are not as accurate as the ones obtained for the harmonic
oscillator. The hydrogen atom is a complicated problem; its energy levels are all
negative and they they get closer and closer to each other as n grows. The most
serious difficulty is posed by the weak binding leading to a spread-out wave function,
quite unlike a particle in a box. We can see that in the values of the variational
parameter, each following state needs a larger b, and similarly with each next quantum
number `.

4.3.3 Confined hydrogenic atoms

With our variational approach we can think that we are confining the potential we
want to study to a box of specific size, and we want to find the optimal size that
will give us the best approximations to the energy levels. In the case of the hydrogen
atom mentioned in the previous section, we find that we need bigger boxers for each

53
following state, this is because, as already mentioned, the problem is spread out.
But this opens up the possibility to study the behaviour of a system that is already
confined.
The example of an hydrogen atom confined to a spherical box has been studied by
Varshni [23] and by Ciftci, Hall, and Saad [4]. For this sort of problem the Schrödinger
equation is given by:

d2
 
` (` + 1) A
− 2 ψ(r) + − ψ(r) = Eψ(r), (4.2)
dr r2 r

with boundary conditions ψ(0) = ψ(b) = 0, and A > 0. In the latter study [4], exact
solutions are given for specific radii. We then applied the variational method to this
problem using the radii suggested by the exact solutions and obtained the following
results for A = 1.

Table 4.8: Approximation of the energy levels of a confined hydrogenic atom in dimension d = 3.
` n b E EV N
0 1 4 -0.06250000000 -0.06249993558 200
1 1 12 -0.02777777778 -0.02777772347 150
2 1 24 -0.01562500000 -0.01562500004 150
3 1 40 -0.01000000000 -0.01000000002 100
0 1 3.803847576 -0.02777777778 -0.02777730093 100
2 14.19615242 -0.02777777778 -0.02777473502 100
1 1 11.05572809 -0.01562500000 -0.01562458222 100
2 28.94427191 -0.01562500000 -0.01562321913 100
2 1 21.77124344 -0.01000000000 -0.009999999956 150
2 48.22875656 -0.01000000000 -0.009999999937 150
3 1 36 -0.006944444444 -0.006944444423 150
2 72 -0.006944444444 -0.006944444436 150

Here ` is the quantum number, n is 1 plus the number of nodes of the wave
function, E the exact solution for the energy, and EV the variational approximation
using a matrix of dimension N . In contrast to what happened with the hydrogen
atom, we find that for the confined atom the approximations are comparable. We
also show the graphs for the approximations of the confined wave functions for radii

54
b = 4 and b = 72 in figures (4.9) and (4.10) respectively.

Figure 4.9: Hydrogenic atom in d = 3 confined to a box of size b = 4.

Figure 4.10: Hydrogenic atom in d = 3 confined to a box of size b = 72.

4.3.4 Highly-singular potentials

Highly-singular potentials are problems that have been widely studied as it is difficult
to find exact solutions. There are a special kind of potentials that are called quasi
exactly solvable, this means that we can find a part of the energy spectrum exactly
when the potential parameters satisfy specific conditions. For example, Dong et al.

55
[6] studied the potential
V (r) = ar2 + br−4 + cr−6

by assuming that the radial equations for the ground state and first excited state were
given by:

√ √
R` (r) = rκ α + βr2 + γr−2 exp − ar2 + cr−2 /2,



(4.3)

with β = γ = 0, and κ = κ0 for the ground state. Although there are more
studies on this matter, where different wave functions are used to calculate the exact
solutions, and the parameters a, b, and c have to satisfy certain constraints.
We find the the exact solutions for the ground state energy in the cases where
a = b = c = 1 and a = 1, b = c = 9 mentioned in Hall et al. [13].
For a = b = c = 1 we have the potential

V (r) = r2 + r−4 + r−6

the ground state energy is given by E0 = 5. Using different implementations of the

variational approach, with both bases described in Section 3.4, the best result we
obtained was the approximation EV = 5.000008414, for a matrix of size N = 50, and
optimal parameters a = 0.01 and b = 5.
For the case where a = 1 and b = c = 9 we now have the potential

V (r) = r2 + 9r−4 + 9r−6

the ground state energy is given by E0 = 7. And our approximation is

EV = 7.000019409, where N , a, and b are the same as above.

56
Conclusions

Throughout the history of physics, scientists have studied how to find solutions to all
sort of mathematical problems posed by physical theories. In this way the explicit
implications of these theories can be revealed, and verification and falsification of the
theories become possible. In this thesis we have studied a small portion of this wide
world of physical problems, namely that related to Schrödinger operators and their
discrete spectra.
The variational method we have adopted relies strongly in the use of the sine-basis
obtained from the solutions of the particle-in-a-box problem, which we have presented
in section 2.4. One of the questions we wanted to answer was, can this basis be applied
effectively to construct a variational analysis for unconfined problems?
From the results shown in chapter 4. we conclude that this basis is appropriate
for the study of problems that have some similar characteristics to the particle-in-a-
box problem. That is why a potential such as the harmonic oscillator, whose wave
functions vanish very quickly with |x|, as though the particle were already in a box,
leads to good results. This points out that in theory, even if we are working with
potentials that are highly singular but strongly U-shaped, then the problem would
be suitable for a variational analysis in the sine-basis.
Other kinds of singularities that do not sufficiently confine the particle might
represent a problem, such as the harmonic oscillator in dimension d = 2 and with
quantum number ` = 0. An effective use of the sine-basis for this case is still an open

57
issue.
We also conclude that the sine-basis does not work too well with loosely-bound
systems such as the hydrogen atom in dimension d = 3. The explanation for this is
that such problems are not effectively confined, and require large boxes to approximate
their energy levels; and the resulting computations proved to be slow.
However, as we found for the hydrogen atom, if the quantum system studied is
already confined in a box, the approximations for the eigenvalues are highly accurate.
The study of confined systems has been of great interest in recent years, since the early
work of Michels [17]. These physical systems are important because of the necessity
of finding information about atoms or molecules trapped in microscopic cavities, or
places in high pressure environments. It appears that this sine-basis gives excellent
results when applied to estimate the spectra of this class of physical problem.

58
Bibliography

[1] M. Abramowitz and I.A. Stegun. Handbook of Mathematical Functions . Dover,

New York, 1972.

[2] M. Andrews. Singular potentials in one dimension. Am. J. Phys. 44 22 (1976).

[3] M. Braun. Differential Equations and Their Applications. Springer, New York,
1993.

[4] H. Ciftci, R.L. Hall, and N. Saad. Study of a confined Hydrogen-like atom by the
Asymptotic Iteration Method. arXiv:0807.4135v1 [math-ph] 25 July 2008, (IJQC,
In Press).

[5] E.A. Coddington and N. Levinson. Theory of Ordinary Differential Equations.

Krieger Publishing Company, Florida, 1984.

[6] S.H. Dong, X.W. Hou, and Z.Q. Ma. Schrödinger Equation with the Potential
V (r) = ar2 + br−4 + cr−6 . arXiv:quant-ph/9808037v1 21 Aug 1998.

[7] S. Flügge. Practical Quantum Mechanics . Springer-Verlag, New York Heidel-

berg Berlin, 1974.

[8] W. Greiner. Quantum Mechanics an Introduction. Springer-Verlag, Berlin, 1994.

[9] D. Griffiths. Introduction to Quantum Mechanics. Pearson Prentice Hall, 2005.

59
[10] S.J. Gustafson and I.M. Sigal. Mathematical concepts of quantum mechanics.
Springer, New York, 2006.

[11] R. Hall and N. Saad. Variational Analysis for a generalized spiked harmonic
oscillator. J. Phys. A 33 569-578 (2000).

[12] R.L. Hall. A simple eigenvalue formula for the quartic anharmonic oscillator.
Can. J. Phys. A 63, 311 (1985).

[13] R.L. Hall, Q.D. Katatbeh, and N. Saad. A basis for variational calculations in
d dimensions. J. Phys. A Math.Gen. 37 11620 (2004).

[14] K. Hannabuss. An Introduction to Quantum Theory. Oxford Science Publica-

tions, 1999.

[15] P. Kościk and A. Okopińska. The optimized Rayleigh-Ritz scheme for determining
the quantum-mechanical spectrum. J. Phys. A Math.Theor. 40 10851 (2007).

[16] E. Kreyszig. Introductory functional analysis with applications. Wiley, New York,
1989.

[17] A. Michels, J. de Boer, and A. Bijl. Physica 4 981 (1992).

[18] J.D. Pryce. Numerical Solution of Sturm-Liouville Problems. Oxford Science

Publicatons, New York, 1993.

[19] M. Reed and B. Simon. Methods of modern mathematical physics: Analysis of

operators, volume IV. Academic Press, New York, 1978.

[20] H. Sagan. Boundary and Eigenvalue Problems in Mathematical Physics. John

Wiley and Sons, Inc., New York, 1963.

[21] B. Simon. Large orders and summability of eigenvalue perturbation theory: A

mathematical overview . Int. J. Quantum Chem. 21 3 (1992).

60
[22] A. Sommerfeld. Partial differential equations in physics, volume 1. Academic
Press Inc., 1941.

[23] Y.P. Varshni. Critical cage radii for a confined hydrogen atom. J. Phys. B. At.
Mol. Opt. Phys. 31 2849 (1998).

61
Appendix A

Maple Code for the Variational

Analysis of the Harmonic
Oscillator

>restart:with(LinearAlgebra): with(plots): with(Statistics):

The basis functions are defined by :
> Psin:= (x, n, L) -> (1/sqrt(L))*sin(n*Pi*(x/(2*L) + 1/2));
The kinetic energy matrix.
> Ki:= proc (L, n)
> local i, j, p;
> p := Matrix(n,n);
> for i to n do
> p[i,i] := evalf(i^2*Pi^2/(4*L^2));
> end do;
> p;
> end proc;
The potential energy matrix.

62
> Po:= proc (L, n)
> local i, j, k, m, p;
> p := Matrix(n,n);
> for i to n do
> p[i,i]:= evalf(L^2*(i^2*Pi^2-6)/i^2/(3*Pi^2));
> for j from i+1 to n do
> p[i,j] := evalf(4*L^2/Pi^2/(i^4-2*i^2*j^2+j^4)*
(4*(-1)^(i+j)*i*j+4*i*j));
> p[j,i] := p[i,j];
> end do
> end do;
> p;
> end proc;
The Hamiltonian matrix.
> HM:= proc (L, n)
> local HM;
> HM := Ki(L,n)+Po(L,n);
> end proc;
This is the optimization process.
n : = dimension of the Hamiltonian matrix.
L : = initial value of the variational parameter
s : = stepsize so the next variational parameter used in the
process will be L+s
it : = number of iterations
dig : = number of digits in the approximation
flag : = 1 or 2
If flag : = 1 OPT will give a matrix with the first row being

63
that containing the values of the parameters L, and the next
rows will be the corresponding eigenvalues of the Hamiltonian
matrix.
If flag : = 2 OPT will give a matrix which in the first column
will obtain the ordered eigenvalues of the Hamiltonian matrix,
and in the second column we will have their corresponding values
of the optimum parameter.
> OPT:= proc (n, L, s, it,dig,flag)
> local M, P, T, i, j, l,E;
> Digits:=dig;
> l := L;
> #for i from 1 to n+1 do;
> # T[i,1]:=i-2;
> #end do;
> for j to it do
> T[1,j]:=l;
> M := HM(l,n);
> P := evalf(simplify(LinearAlgebra:-Eigenvalues(M)));
> for i to n do
> P[i] := Re(P[i])
> end do;
> P := sort(P);
> for i to n do
> T[i+1,j] := P[i];
> # print(j,l,T[i,j],i-1)
> end do;
> l := l+s;

64
> end do;
> T:=convert(T,array);
> for i from 2 to n+1 do
> E[i-1,1]:=10^10;
> for j from 1 to it do
> if T[i,j] < E[i-1,1] then
> E[i-1,1]:=T[i,j];
> E[i-1,2]:=T[1,j];
> end if;
> end do;
> end do;
> E:=convert(E,array);
> if flag=1 then
> T;
> else
> E;
> end if;
> end proc;

65
Appendix B

Shooting Method

Given a Sturm-Liouville problem in the normal form, we can develop a numerical

shooting method to calculate its eigenvalues and eigenfunctions [18].
Roughly speaking, the shooting method consists of choosing a sequence of eigen-
values and adjusting them to the conditions given by theorem (2.2), that is, for each
state n = 0, 1, 2, . . ., the function has k = 0, 1, 2, . . . nodes or zeros respectively. Then
solving for each trial eigenvalue the equation (2.16) and choosing the eigenvalue that
gives the best approximation.
As we want to focus this method to find solutions for quantum mechanical prob-
lems, we know that the solutions have to belong to L2 (Rd ), that is, they need to
vanish at infinity. For the numerical analysis we have to choose a suitable range for
which the function will start vanishing quickly, for some problems it might be bigger
than expected.
Another thing to take into account is the initial conditions of each problem; this
is of utmost importance while solving a differential equation numerically.

Algorithm. The algorithm we use is a simple refining algorithm, and it is described

as follows:
For each state n = 0, 1, 2, . . . we set up the node goal, that is, the number of zeros

66
that the function corresponding to the particular state to be ng. We then choose
upper and lower values λu and λl respectively,
1. Let λ be the average (λu + λl ) /2. This value will be our trial eigenvalue.
2. Solve the equation
d2 u
− + Q(x)u = λu, (B.1)
dx2

satisfying the initial conditions imposed by the original problem.

3. Given the numerical solution of the above equation we count its number of roots
or nodes. If this number is lesser or equal than node goal ng, then the eigenvalue λ is
too small, and we need to update the lower value λl = λ. Else, the number of roots
is greater than node goal ng, which means that the eigenvalue λ is too big, then we
update the upper value λu = λ.
4. And we return to step 1 unless the stop criteria is met.

Though the algorithm is straightforward, a computational implementation may

not be easy as it depends mainly on the numerical solution of a differential equation,
which is a very complicated topic.

67