Wavelets in Quantum Mechanics

Shannon T. Cowell
Old Dominion University
shannon@physics.odu.edu
August 20, 1998

Mentor: Robert J. Perry, Ph.D.

The Ohio State University

Abstract
It was the intent of this project to use the expansion of basis tech-
nique to solve the one-dimensional square well problem. In an eort to
make use of the advantages of both orthogonal and non-orthogonal ex-
pansions, the wavelet basis was chosen. Though the one-dimensional
case considered here is trivial, the eciency of the approximation
sheds light on the appropriateness of the basis in pursuing more com-
plicated three-dimensional expansions.

1
1 Introduction
This paper focuses primarily on the motivation for and details of the ex-
pansion technique used to solve problems arising in Quantum Mechanics.
A discussion of the expansion technique followed by a general overview of
each expansion type is presented (Sections 2 and 3). For the purposes of
this project, an orthogonal wavelet basis was chosen. A brief prole of the
wavelet is presented (Section 4) as well as the results and conclusions drawn
from its use (Sections 5 and 6).

2 Expansion Technique
This section assumes some knowledge of Schrodinger's equation and of Quan-
tum Mechanics. Its sole intent is to provide a clear picture of the details of
the expansion technique.
The non-relativistic one-dimensional Schrodinger's equation is
p2 + V = E ; (1)
2m 0

where p2 is the momentum operator, m is the mass of the particle, V0 is

the potential energy, E is the total energy of the particle and is the wave
function of the particle in the system.
Schrodinger's equation can be written symbolically as
H =E ; (2)
where H is called the Hamiltonian of the system.
Equation 2 is simply an eigenvalue/eigenvector problem, where E is the eigen-
value and is the eigenvector. In trivial problems such as the square well
considered here, this dierential equation is not dicult to solve. However,
in many body, three-dimensional problems, the Hamiltonian is more complex
and an analytic solution may not exist. A common approach to approximate
the solution in these types of problems is an expansion technique.
Let be approximated by
X
n

= ;
i i (3)
i=1

where  is a set of functions of x or p and  is the unknown relative `strength'

i i

of  in the approximation. The stability of the approximation is controlled

i

2
by n, the number of basis functions used. As n increases, the approximation
becomes more accurate However, beyond a nite value of n the expansion
can become independent of n. Simply, once the expansion has become `com-
plete,' additional basis functions will no longer increase the accuracy of the
approximation.
Using this expansion, Schrodinger's equation can be re-written as
X
n
X n

H  =Ei i : i i (4)
i=1 i=1

Multiplying equation 4 by each  of the expansion (equation 3), and inte-

i

grating will yield n equations of the form

Z1 X
n Z1 X n

H  = E   ; j = 1 : : : n: (5)
1 1
j i i j i i
i=1 i=1

In the case of n = 2, equation 5 becomes

Z1 Z1
1H (11 + 2 2) = 1E (1 1 + 2 2);
1 1
Z 1 Z1
2H (11 + 2 2) = 2E (1 1 + 2 2): (6)
1 1
Using the notation, Z1
H =  H ;
Z11
ij i j

M = i j ;
1
ij

equation 6 can be re-written as

(H111 + H122 ) = E (M11 1 + M12 2);
(H211 + H222 ) = E (M21 1 + M22 2); (7)
or ! ! ! !
H11 H12 1 = E M11 M12 1 : (8)
H21 H22 2 M21 M22 2
Multiplying both sides by M121 , assuming that it exists, leads to
! ! ! !
M11 M12 1 H11 H12 1 = E 1 : (9)
M21 M22 H21 H22 2 2
In more general notation, equation 9 can be written as

3
 0 1 1 0 1
BB 2 C C B
BB 2 CCC
1

M H B
1
B@ C
... C = E B .. C : (10)
ij ij
A @ . A
 n  n

The matrix equation 10 is a matrix eigenvalue/eigenvector problem. The

original problem, which may not have an analytical solution, has now been
reduced to one of diagonalizing a matrix and minimizing the eigenvalues,
which can be done with standard techniques.
For this project, a variational method was used in equation 10 to approxi-
mate the eigenvalues and eigenvectors of the original Schrodinger's equation
(equation 2). Although an important aspect of the process, it is beyond the
scope of this paper.

3 Basis functions
The advantages and disadvantages of each basis expansion can be more easily
seen through an example. Consider a one dimensional double well potential,
where one well is narrow and deep and the other, a large distance away, is
shallow and wide (Figure 1).
V=0

V1

V2

Figure 1: Example: One-dimensional double square well

Without knowing the general solution, the behavior of the wave function at
points on the axis can be determined. Specically, there must be a peak in
each well and the wave function must go to zero outside of each well. Which
expansion is best suited for this problem?

3.1 Non-Orthogonal Basis

A non-orthogonal basis expansion provides a great deal of freedom. Basis

functions can be pieced together in each region to obtain the desired behav-
4
ior regardless of their relationship to one another. The wave functions can
be approximated with a Gaussian basis. In this example, Gaussian func-
tions can be centered on or around each well. Their width and placement
can be allowed to vary such that the number of basis functions needed is
minimized. In addition, because of the localization properties of Gaussian
functions, their overlap in the region between the wells will be minimal requir-
ing few additional basis functions to achieve the near zero behavior expected.
This attractive feature, however, adds to the computational complexity. The
matrix M 1 must be calculated and inverted, if possible. If, to achieve greater
ij

accuracy in the approximation, the number of basis functions is increased,

M 1 can become singular. In addition, the time needed for computation
ij

increases rapidly with n.

3.2 Orthogonal Basis

An orthogonal basis expansion eliminates the disadvantage of calculating the

matrix M 1 (M 1 = 1). However, the freedom of piecing together the ba-
ij ij

sis to suit the potential is also lost. Most orthogonal basis functions are
centered strictly around one point. Again, if a Gaussian basis is chosen, it
can be placed near or around one well, but many additional basis functions
will be needed to build the approximation around the second well. If the
basis functions are centered between the two wells, additional functions will
be necessary to approximate the near zero behavior between the wells while
building the wave function within each well. Although the inverse matrix
does not need to be calculated, computation time is not always decreased
due to the additional basis functions needed to obtain a more accurate ap-
proximation.

4 Wavelets
Wavelet theory has its roots in such areas as applied physics and engineering.
Used for several years in such elds independently, it was not formalized as
a mathematical theory until the middle of the 1980's. Since that time, many
areas such as applied mathematics, seismology, signal processing and data
compression have made great progress in developing and using wavelets. To
date, however, little work has been done in pursuing wavelets as a tool in
quantum mechanic calculations.
In simple terms, a wavelet is a small wave, or a well-localized function.
Though there exists no general formula for wavelets, they are characterized
5
by their oscillatory form. The only additional dening property of a wavelet
is that its total area must be zero.
For the purpose of this project, the wavelet chosen had to meet three addi-
tional criteria. To optimize the Heisenberg Uncertainty relation, the wavelet
needs to be well localized in both position and momentum spaces. Orthogo-
nality is required of the basis functions to simplify calculations and ensuing
computations. In addition, a simple analytic form is necessary for both po-
sition and momentum spaces to simplify computation of matrix elements.
Imposing these conditions greatly reduces the wavelets available. Wavelets
are not typically dened in analytic form, but by an iterative process. In
addition, few orthogonal wavelets have been developed. Of the wavelets hav-
ing these two features, the best l properties are found in the Meyer Wavelet
(Figure 2).

|Y0,0 (p)| Y0,0 (x)

0.4

0.5
0.3

x
0.2 -10 -5 5 10

-0.5
0.1

-1
p
-10 -5 5 10

Figure 2: Meyer `Mother' Wavelet: Left - Momentum Space, Right - Position

Space

Figure 2 shows what is termed the `Mother' wavelet. Through a standard

translation and dilation equation, the mother wavelet produces a series of
basis functions. Each basis function will possess the general form of the
mother wavelet, but it will either be located at a dierent position on the
axis, and/or `broadened.' Figure 3 and Figure 4 show the eects of the
translation/dilation equation on the Meyer wavelet.

6
 |Y-1,0 (p)| |Y0,1 (p)|
0.3
0.5
0.2
0.4
0.1
0.3
p
-10 -5 5 10
0.2 -0.1
0.1 -0.2
p -0.3
-10 -5 5 10

Figure 3: Meyer wavelet basis functions produced by translation and dilation of

1 respectively

Y1,0 (x) Y0,4 (x)

0.6
0.4 0.5
0.2
x x
-10 -5 5 10 -10 -5 5 10
-0.2
-0.4 -0.5
-0.6
-1
-0.8

Figure 4: Meyer wavelet basis functions produced by dilation of 1 and translation

of 4 respectively

4.1 Wavelets - Best of both worlds?

At rst glance, wavelets seem to provide the best of both types of expansions.
They are `movable,' meaning they can be pieced together in the same fash-
ion as the non-orthogonal basis functions. By their design, each wavelet is
mutually orthogonal, eliminating the concern of calculating the M 1 matrix. ij

They are well localized, requiring fewer additional functions to cancel out or
build up pieces of the the approximation. These advantages combined were
expected to result in easier calculations and shorter computation time.

5 Results
Using the Meyer wavelet to solve Schrodinger's equation provides its own
set of disadvantages. The most inhibitive of these are caused by the highly

7
oscillatory behavior of the wavelet and the extremes on the limits of integra-
tion caused by the translation and dilation of the momentum space wavelet.
Obtaining accurate results in the integration routines for the calculation of
matrix elements was dicult and time consuming. All calculations attempted
required long run times, and the accuracy obtained for some basis functions
never exceeded three decimal places.

6 Conclusion
Although the results of this project are not promising, several possibilities
still exist to improve future calculations using wavelets. Among these is
the possibility to develop a basis that is more aptly suited for numerical
integration techniques. The converse is also true: nding or constructing
a more accurate integration routine designed to handle the oscillations and
limits of integration variations of the wavelets more reliably. Until these
issues are resolved, further pursuit of calculations using the Meyer wavelet
will prove impractical and often yield unreliable results.