See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/1737370

The Harmonic Oscillator in Quantum Mechanics: A Third Way

Article in American Journal of Physics · June 2008

DOI: 10.1119/1.3042207 · Source: arXiv

CITATIONS READS
49 3,405

1 author:

F. Marsiglio
University of Alberta
267 PUBLICATIONS 5,682 CITATIONS

SEE PROFILE

All content following this page was uploaded by F. Marsiglio on 12 July 2013.

The user has requested enhancement of the downloaded file.

 The harmonic oscillator in quantum mechanics: A third way
F. Marsiglio
Department of Physics, University of Alberta, Edmonton, Alberta, Canada, T6G 2J1
(Dated: December 30, 2008)
Courses on undergraduate quantum mechanics usually focus on solutions of the Schrödinger equa-
tion for several simple one-dimensional examples. When the notion of a Hilbert space is introduced
only academic examples are used, such as the matrix representation of Dirac’s raising and lowering
operators or the angular momentum operators. We introduce some of the same one-dimensional
examples as matrix diagonalization problems, with a basis that consists of the infinite set of square
well eigenfunctions. Undergraduate students are well equipped to handle such problems in familiar
contexts. We pay special attention to the one-dimensional harmonic oscillator. This paper should
arXiv:0806.3051v2 [quant-ph] 30 Dec 2008

equip students to obtain the low lying bound states of any one-dimensional short range potential.

PACS numbers:

I. INTRODUCTION encountered it using the algebraic method (series solu-

tion) and Dirac’s operator method. The former method
A standard undergraduate text in quantum tends to leave students struggling with the mathematics,
mechanics1 devotes most of its attention to analyt- and the latter method leaves many students awestruck.
ical solutions of the Schrödinger equation. Some texts2,3 The method proposed here uses matrix algebra, which is
coordinate problem solving in quantum mechanics with a more familiar mathematical method.
numerical solutions. However, the focus remains the We also illustrate wave functions, which can be read-
solution of the Schrödinger equation. ily calculated from the eigenvector solutions. We have
A typical introductory course also includes a treatment been unable to find problems of this nature in the liter-
of the more formal part of quantum mechanics, with top- ature. One reference illustrates a solution to the three-
ics such as vector spaces and the matrix representation. dimensional harmonic oscillator bounded by two impen-
This part is usually more abstract. Examples that are etrable walls, but does so following the standard route of
often used in this part are two state systems and matrix differential equations, though in cylindrical coordinates.7
representations of angular momentum. Alternatively, an The matrix methodology discussed here is at the heart
actual (say, 3 × 3) matrix is given, and students are asked of many applications in physics. For example, vibrations
to use matrix algebra to mimic a solution of a quantum in solids8 are treated using springs that obey harmonic
mechanical problem. These examples often strike stu- oscillator potentials. Bose condensation in alkali gases9
dents as artificial. The main part of such a course appears utilizes a harmonic trap, and the conventional theory
to students to be the solution of various differential equa- of superconductivity uses electron-ion interactions with
tions, and the formal part of the course often appears to harmonic oscillators.10 Calculations of a single electron
be superfluous. interacting with vibrating ions (that is, a polaron) that
Numerical solutions of the time-independent use matrix diagonalization can be found in Ref. 11.
Schrödinger equation have been discussed in vari- The convergence criterion for the harmonic oscillator
ous contexts (see, for example, Ref. 4). These solutions potential is related to energy scales, and the number of
utilize numerical techniques for solving a differential basis states required to accurately represent these solu-
equation. Even in the rare instance that an actual basis tions is dictated by the energies of the basis states. In
set is used,5 it is used to determine the time evolution for Sec. IV we briefly discuss some other familiar interac-
the time-dependent Schrödinger equation. Many quan- tions, such as the finite square well and the delta func-
tum mechanical research problems that are amenable tion potential. In some of these cases length scales also
to solution (for example, the behavior of electrons on play an important role, which leads to a discussion of the
a small lattice) are solved using matrix mechanics.6 convergence criterion connected with capturing the short
The purpose of this note is to bring attention to a length scales in the potential.
wide variety of time-independent problems that can be This paper is meant to be sufficiently self-contained
tackled through matrix mechanics at an introductory and should serve to enhance the undergraduate curricu-
level. lum in quantum mechanics, and equip undergraduates to
In the following section we set up the infinite square solve a variety of useful problems.
well problem. The solution to this problem constitutes
the (infinite) basis set with which we can tackle any sin-
gle particle potential problem that can be embedded in II. THE FORMALISM
the infinite square well. In Sec. III we illustrate the ma-
trix solutions for the harmonic oscillator potential. This A typical introductory course begins with one-
example is chosen because the students will have already dimensional problems, including the free particle, the in-
 2

finite square well, and the harmonic oscillator potential. By taking the inner product of (7) with the bra hψn |, we
The free particle has certain pathologies associated with obtain the matrix equation:
the normalization of the stationary states. For this rea- ∞
son we prefer to discuss the infinite square well, in which X
Hnm cm = Ecn , (8)
a particle is confined to the space enclosed by the walls m=1
of the well. The harmonic oscillator is also amenable to
analytical solution. Unlike the infinite square well the where
solutions (Hermite polynomials enveloped by Gaussians)
are unfamiliar to the novice student, and, as far as a Hnm = hψn |(H0 + V )|ψm i =
2 a nπx
mπx

Z
typical student is concerned, might as well be numerical.
δnm En(0) + dx sin V (x) sin (9)
For these reasons we focus on the infinite square well, a 0 a a
which is defined as
( is the Hamiltonian matrix; in the second term we have
0 if 0 < x < a, used the eigenstates of H0 as the basis set. As usual, δnm
Vinf (x) = (1) is the Kronecker delta function.
∞ otherwise.
Equation (8) along with Eq. (9) allows us to study a
The Hamiltonian is given by variety of interesting problems, albeit in the space de-
termined by the infinite square well. Note that for nu-
h̄2 d2 merical solutions of Eq. (8) we need to truncate the sum
H0 = − + Vinf (x), (2) and hence the number of eigenstates used to describe the
2m dx2
solutions. We denote the maximum basis set number by
where m is the mass of the particle. The eigenstates are N.
well known:
r
 2
 nπx 
sin if 0 < x < a, III. THE HARMONIC OSCILLATOR
ψn (x) = a a (3)
0 otherwise,

We imbed the harmonic oscillator potential, VHO =
mω 2 x2 /2, in the infinite square well. In the following
with eigenvalues,
all units of distance will be in terms of the square well
n2 π 2 h̄2 width a, and all units of energy will be in terms of the
(0)
En(0) = ≡ n2 E1 . (4) (0)
(unperturbed) ground state, E1 . We will use lower case
2ma2
letters to denote dimensionless energies. The potential
The quantum number n = 1, 2, 3, . . . takes on a positive VHO can be written in terms of the infinite square well
integer value. Students are generally unaware of pertur- length and energy scales as
bation theory at this stage, but reasons for the super- 2 
π 2 h̄ω

script (0) can be readily explained. VHO x 1 2
vHO = (0) = − , (10)
Although we have presented the well-known solutions E1 4 E (0) a 2
1
for the infinite square well problem, the emphasis in the
preceding paragraphs should be on the formulation and (0)
so that the dimensionless parameter h̄ω/E1 determines
presentation of a convenient and familiar set of basis the stiffness of the harmonic oscillator potential. We ex-
functions. We are now prepared to tackle a variety of pect that for low energy states (say, the ground state),
problems, represented by H = H0 + V (x), where H0 is the solution should be identical to that of the harmonic
as in Eq. (2), and V (x) is any potential in the domain potential alone, because the wave function will be suffi-
0 < x < a. We proceed as in most textbooks, except now ciently restricted to the central region of the harmonic
we have the specific basis set Eq. (3) in mind. oscillator potential so that it will not “feel” the walls of
We start with the Schrodinger equation in ket notation: the infinite square well. High energy states will not be
well described by the harmonic oscillator results, because
(H0 + V )|ψi = E|ψi. (5)
they will be primarily governed by the infinite square
If we use the general expansion of a wave function in well.
terms of a complete set of basis states, We first use Eq. (9) with the potential given by
Eq. (10). The result is
∞
X
|ψi = cm |ψm i, (6) (0)
hnm ≡ Hnm /E1
m=1 2 
π 2 h̄ω
  
2 6
= δnm n + 1−
we obtain 48 E (0) (πn)2
1
∞ ∞  2
X X h̄ω
cm (H0 + V )|ψm i = E cm |ψm i. (7) + (1 − δnm ) (0)
gnm , (11)
m=1 m=1 E1
 3

where
 (−1)n+m + 1  1 1  5000
gnm = − . (12)
4 (n − m)2 (n + m)2
Note that the gnm remain of order unity close to the di- 4000
agonal, but for large n the diagonal elements grow as n2 ,
so the off-diagonal elements become negligible in com-
vHo
parison. 3000

(0)
Equation (11) can be evaluated up to some cutoff for

En/E1
(0)
a given h̄ω/E1 ; these form the elements of a matrix
to be fed into an eigenvalue/eigenvector solver (see for 2000 exact
example, Numerical Recipes12 ) or in software packages
n2
such as Matlab, Mathematica, or Maple.
A typical result is shown in Fig. 1. The low lying states hω(n−1/2)/E1(0)
1000
are well described by the harmonic oscillator eigenvalues. n2+ v0
The harmonic oscillator potential becomes truncated by
the infinite square well at an energy given by the value 0
(0) 0 10 20 30 40 50 60 70
of the potential at x = a (see Eq. 10): Ecross /E1 =
(0) (0) n
(π 2 /16)(h̄ω/E1 )2 . For this example, h̄ω/E1 = 50,
(0)
so this crossover occurs at Ecross /E1 ≈ 1550, which FIG. 1: (color online) Eigenvalues obtained by the exact di-
is approximately where the numerical results in Fig. 1 agonalization of the harmonic oscillator embedded in an in-
begin to deviate from the analytical harmonic oscillator finite square well (shown as an inset). The symbols denote
result. For higher energies the numerical results begin the numerically converged result, obtained with a truncation
to resemble those of the infinite square well, albeit with N = 400. We also show the known analytical result for a har-
an added constant, v̄0 . Inspection of Eq. (9) (see also monic oscillator, EHO = h̄ω(n − 1/2) (we use an enumeration
Eq. (11)) shows that this constant is the average of the that starts with one (not zero) consistent with the enumera-
tion of the infinite square well eigenvalues). The agreement
harmonic potential.
is excellent for low-lying eigenvalues. For eigenvalues that are
What happens if the harmonic potential is consider- comparable to or exceed the crossover energy the analytical
(0)
ably less stiff? If, for example, h̄ω/E1 = 1, then none results for the infinite square well, suitably modified to take
of the eigenstates will describe harmonic oscillator states; approximate account of the harmonic oscillator potential, are
(0)
rather they all will be minor perturbations of the infinite extremely accurate (dotted curve). We used h̄ω/E1 = 50.
square well eigenstates. Nonetheless, to describe such
an oscillator is still possible, simply by choosing a much
wider infinite square well.
How well is the wave function reproduced by this trun- tion, and compare them with the expected Fourier com-
cated diagonalization technique? We use Eq. (6) and ob- ponents of the Gaussian wave function. For example, for
tain the ground state, these are
∞ r
X 2 mπx
ψ(x) = hx|ψi = cm sin , (13) ( (0)
m=1
a a i(n−1) 32 E1 (0)
exp [−n2 E1 /h̄ω] for n odd,


cn = π h̄ω

where the cm are the components of the eigenvectors in 0 for n even.

the basis of the infinite square well eigenfunctions. For (15)
numerical results we use Eq. (13) with the coefficients The exponential decay with quantum number n helps
obtained from the eigenvectors determined by the matrix to explain why so few components are required to obtain
diagonalization. For the harmonic oscillator the results an accurate representation of the Gaussian ground state
are known analytically. For example, in dimensionless wave function. When we checked the numerically deter-
units the ground state wave function is mined coefficients against those given by Eq. (15) (not
1/4 shown), we found excellent agreement.
π 2 h̄ω x 1
2
  
π h̄ω The important point is that low lying states for any po-
ψHO (x) = exp − − .
2a2 E (0) 4 E (0) a 2 tential can be obtained to any desired precision by setting
1 1
(14) up the matrix in Eq. (9). Analytical evaluations of the
In Fig. 2 we plot the numerical results superimposed with integrals are not even required.13 For the harmonic os-
the analytical ones. The agreement is excellent (beyond cillator this method requires the least amount of mathe-
what can be discerned on a graph with this scale). We matics. More importantly, it serves to illustrate in a very
can also check the components cm (see Eq. (13)) di- concrete and elementary way the matrix formalism that
rectly, because they are determined through diagonaliza- is often presented only abstractly in textbooks.
 4

12000
3
10000
exact V0

N = 20 8000
2 N = 10

En/E1(0)
N=5
6000
ψ(x)

4000 exact
1 n2
2
4n
2000 n2+v0/2

0
0 0 20 40 60 80 100
0.0 0.2 0.4 0.6 0.8 1.0 n
x/a
FIG. 3: (color online) Eigenvalues plotted as a function of
FIG. 2: (color online) The ground state of the harmonic os- quantum number n for the potential shown in the inset. Sym-
cillator using the analytical result, Eq. (14) (symbols), along bols denote numerically determined eigenvalues (for a matrix
with numerical results, obtained from matrix diagonalization, truncation N = 400, but a much smaller value yields identi-
with various truncation values. Convergence is achieved for cal results). The dashed curve illustrates the analytical result
(0)
N = 20. We used h̄ω/E1 = 50. expected for an infinite square well of width a/2. The short-
dashed curve shows the same result for a well of width a.
These analytical results agree with the numerical results for
low and high eigenvalues, respectively. Agreement for high
IV. RESULTS FOR SOME STANDARD eigenvalues improves significantly when a shift of v0 /2 (dot-
ONE-DIMENSIONAL POTENTIALS ted curve) is included to account approximately (as far as the
high energy states are concerned) for the presence of the step
There are a variety of even simpler one-dimensional potential. We used v0 = 4000; the value of N is not impor-
tant if it exceeds the maximum quantum number displayed
potentials, and this section illustrates how results from
by about an additional 10.
matrix mechanics can make contact with known analyt-
ical solutions.
Consider the potential of the form,
 represented by Eq. (8) can be substituted into an eigen-
0 if 0 < x < b0 ,
value/eigenvector solver. We now consider some special



V if b < x < b ,
0 0 1 cases.
V0 (x) = (16)

 0 if b1 < x < a,

∞ otherwise.

A. Step potential
This potential represents a generic barrier (well) for V0
greater (less) than zero. This case can be solved analyt-
ically although the algebra is tedious. The units are as One of the simplest potentials is that of a potential
before, and we let ρ0 ≡ b0 /a, ρ1 ≡ b1 /a, and ρ = ρ1 − ρ0 . barrier as depicted in the insert of Fig. 3. The expres-
Evaluation of Eq. (9) yields sion in Eq. (17) applies, with b0 = a/2 and b1 = a. Fig-
ure 3 shows the eigenvalues as a function of the quantum
(0)
hnm ≡ Hnm /E1 number n. The symbols show the numerically obtained

sin 2πnρ1 − sin 2πnρ0 
 eigenvalues, and the dashed curve shows the result ex-
2

= δnm n + v0 ρ − pected from Eq. (4), with width a/2 – hence the extra
2πn
 factor of 4. For low-lying eigenvalues the agreement is
sin (n − m)πρ1 − sin (n − m)πρ0 very good. In other words, the particle finds itself in an
+ v0 (1 − δnm )
π(n − m) essentially infinite square well of width a/2. However, for
sin (n + m)πρ1 − sin (n + m)πρ0
 sufficiently high energy states the particle should act like
− (17)
. it is in an infinite square well of width a. We have also
π(n + m)
plotted the result expected for an infinite square well of
As with the harmonic oscillator, Eq. (17) can be evalu- width a, given by the short-dashed curve. Some accom-
ated up to some cutoff, and the resulting matrix problem modation should be made for the fact that a step in the
 5

potential exists at low energies. (Imagine a state with

(dimensionless) energy more than 100 000 on the scale of
Fig. 3 — the step with height v0 = 4000 would appear
to be a small blip.) The average of this potential is just
the product of the height and the width, so the dotted
curve corresponds to the dashed curve shifted by v0 /2.
There are a number of states whose energies are close 0
to v0 , where the eigenvalues are not simply related to
those of an infinite square well. Beyond quantum num- −10 (a)
ber n = 80 the eigenvalues, with a simple accounting for
the step v0 /2, agree with those of an infinite square well.
−20
This example also illustrates how the numerical solution −4
converges to known analytical results at low and high

E/E1(0)
energies. −30

−6
−40
B. Various square wells

Consider a square well with arbitrary width less than −50 −8

0.50 0.75 1.00
a, with V0 < 0. This interaction can be solved ana-
lytically, although the algebra is tedious. For V0 < 0 −60
there will exist states that are bound within the inner 0.0 0.2 0.4 0.6 0.8 1.0
finite well. Insofar as these states don’t “feel” the walls b/a
of the infinite square well at x/a = 0, 1, these energies
and their eigenfunctions should be given by the simple
0
analytical solution found in textbooks. The ground state
is obtained by finding the lowest positive solution to the (b)
equation,1 −10
s 
πu u0 −20
tan = − 1, (18)
2 u
E/E1(0)

b q E−V0 b q V0 −30 N=1

where b ≡ b1 − b0 , u ≡ (0) , and u0 ≡ − (0) 3
a E1 a E1
5
(for V0 < 0 and V0 < E < √ 0). In the dimensionless units −40
10
used in Sec. IV A, u = ρ e − v0 , and similarly for u0 .
20
For the finite square well problem analytical results are
provided by the solution to Eq. (18),14 and numerical re- −50 b 50
exact
sults are obtained by diagonalizing the N × N matrix
with the elements given by Eq. (17). Specifically, we −60
use b0 = a/2 − b/2, b1 = a/2 + b/2, and V0 = V00 a/b.
0.00 0.05 0.10 0.15 0.20
For a fixed value of V00 , this choice of potential (see
Fig. 4) allows a study of the eigenenergies and eigen- b/a
states of square wells with width b (0 < b < a), keeping FIG. 4: (color online) Ground state energy as a function of
the product bV0 fixed. The limit b → 0 describes an finite square well width. The finite square well is imbedded
attractive delta function potential in the middle of the symmetrically within the infinite square well, as shown in the
infinite square well, which is treated analytically in the (0)
insert. We use V00 = −5E1 , and vary V0 = V00 a/b as we vary
Appendix. The other limit, b → a, lowers the baseline of b. Part (a) shows that convergence as a function of truncation
the infinite square well by an amount V0 , so that all the (see legend in part (b)) is slow for small b/a. The inset shows
(0) (0) the situation at large b/a, where the results have converged,
eigenvalues are shifted: En /E1 → n2 − V00 /E1 (for
b = a only). but not to the analytical result for a square well in an infinite
Results for the ground state energy as a function of the space, but rather, as expected, to the trivial result expected
for a slightly deeper infinite square well. In part (b) we expand
finite square well width b are shown in Fig. 4, for various
the low b/a portion of (a), and also show analytical results for
values of the cutoff N (see Fig. 4(b)). The results are the δ-function potential, shown as symbols (see Appendix).
insensitive to the cutoff for b ≈ a, while for low values of Convergence is slow because of the need to capture very small
b, there is considerable variation. The inset in Fig. 4(a) length scales.
clarifies the situation for b ≈ a. The result converges very
rapidly (just the diagonal element is exactly correct for
 6

the wave functions obtained numerically for the case of a

5 delta function potential as we vary the cutoff of the size
of the matrix retained. It is clear that for low cutoffs,
exact only variations on a large length scale can be captured.
4 δ−function potential N=5
Even for the highest cutoff used here, N = 200, the nu-
20
50
merical result still misses (though only slightly) the cusp
3 200 at the centre of the figure. The same kind of plot can
be examined for square well potentials of varying width,
ψ(x)

so that one can see directly that the matrix size required
2 for convergence increases with narrower well width b (as
already seen through the energy in Fig. 4).
1
V. SUMMARY
0
The purpose of this paper is to provide some exam-
0.0 0.2 0.4 0.6 0.8 1.0 ples accessible to undergraduate students learning quan-
tum mechanics for the first time that utilize matrix
x/a
diagonalization.16 Our goal is to provide a context within
FIG. 5: (color online) Ground state energy wave function which a simple basis set can be used. The most straight-
for a δ-function potential, as calculated by diagonalization of forward case in one dimension is the infinite square well.
matrices truncated at N = 5, 20, 50, and 200. The exact We worked through some examples that are generally
result (see Appendix) is given by symbols. There is a cusp used in an introductory course on quantum mechanics,1
at the center, which requires smaller and smaller wavelength with an emphasis on the harmonic oscillator. We showed
components to capture exactly. We used a δ-function with that by embedding a (truncated) harmonic oscillator po-
(0)
strength bV00 = −5aE1 , corresponding to the limiting case tential in an infinite square well, we can obtain extremely
studied in Fig. 4. accurate solutions.
We also illustrated how to calculate wave functions and
demonstrated that potentials with short length scales of-
ten require a large number of basis states for an accurate
b = a). Even the 3×3 matrix yields accurate ground state description.
energy values all the way down to b ≈ 0.5a, and as the full
The empowering aspect of this approach is that stu-
frame of Fig. 4(a) shows, finite N results “peel off” as the
dents can tackle bound state problems for any potential.
width decreases. Note that the analytical result given by
For example, it is simple to find the low-lying states for a
Eq. (18) differs from the numerical result near b = a (see
double well potential. The potential does not even have
inset), because in this regime the two approaches solve
to be defined analytically.13
different problems. The analytical result is for a finite
square well of width b with no background potential, and
the numerical result includes the infinite square well in
Acknowledgments
addition.14 Not surprisingly, the latter case reproduces
(0)
the analytical shift noted earlier from E0 /E1 = 1 to
(0) (0) This work was supported in part by the Natural
E0 /E1 = 1 − 5 = −4 (with V00 = −5E1 ). Sciences and Engineering Research Council of Canada
Figure 4(b) clarifies the behavior in the very narrow (NSERC), by ICORE (Alberta), and by the Canadian
well regime. Results for the delta function potential are Institute for Advanced Research (CIfAR). The author
given with symbols along the b = 0 line. The lowest curve thanks Cindy Blois and James Day for suggestions to
is given by the solution of a simple equation (see Eq. (A2) improve the presentation in the manuscript.
in the Appendix). There is a considerable variation with
the cutoff N in the low b/a regime. For example, even
with a cutoff of N = 100, meaning that unperturbed APPENDIX A: DELTA FUNCTION POTENTIAL
states with energy as high as 10000 are included, the re- WELL
sult at b = 0 is off from the N → ∞ result by about
10%. The reason is that as the finite well becomes nar-
Consider a delta function potential well with strength
rower and the wave function becomes more confined, the
α = V0 b located at position a0 , inside an infinite square
latter contains components with very small (of order b)
well between x = 0 and x = a, that is,
spatial variations. These can only be described by basis
states with a similar strong variation, and such states are (
to be found only among the very high energy states of the −bV0 δ(x − a0 ) (0 < x < a)
V (x) =
infinite square well.15 We use Eq. (13) and show in Fig. 5 ∞ otherwise,
 7

where 0 < a0 < a (the symmetric case is for a0 = a/2). The more familiar solution is obtained by allowing a →
The length scale b is introduced to make contact with ∞, in which case κ = λ/2, and a solution always exists.
the finite square well potential discussed in Sec. III. The In this problem, with a finite infinite square well width
application of matching criteria yields the solution (we a, a solution exists only if
confine our discussion to the symmetric case only):
(
sinh(κx)/ sinh(κa/2) (x < a/2)
ψ(x) = A
sinh(κ(a − x))/ sinh(κa/2) (x > a/2).
(A1) V0 4 a
> . (A4)
The bound state energy E is given by the solution of
(0)
E1 π2 b

2κ
tanh(κa/2) = , (A2)
λ
r
2m 2m
where κ ≡ − 2 E and λ ≡ − 2 V0 b. Equation (A2) A challenging exercise for students is to check how well
h̄ h̄ Eq. (A4) holds for a finite square well of non-zero width
must be solved numerically. The normalization constant
is given by b. Results for non-symmetrically placed delta functions
can also be readily obtained, but are not as concise, and
r
2 1 yield a condition for existence of a bound state which is
A= . (A3) more stringent than Eq. (A4).
a
r  λ 2 λ
1− + 2
2κ κ a

1
D. J. Griffiths, Introduction to Quantum Mechanics (Pear- Vetterling, Numerical Recipes: The Art of Scientific Com-
son/Prentice Hall, Upper Saddle River, NJ, 2005), 2nd ed. puting (Cambridge University Press, Cambridge, 1986).
13
There are many text books to choose from. I will cite this At least one of my students simply programmed Matlab
one as a typical, very good one. to evaluate Eq. (9) numerically for an arbitrary potential.
2
J. M. Feagin, Quantum Mechanics with Mathematica Though not to my personal taste, many students find such
(Springer, New York, 1994). a numerical evaluation much easier than slogging through
3
Bernd Thaller, Visual Quantum Mechanics (Springer, New analytical integrals. The numerical approach allows one to
York, 2000). solve for bound states of any one-dimensional short range
4
J. V. Kinderman, “A computing laboratory for intro- potential.
14
ductory quantum mechanics,” Am. J. Phys. 58, 568–573 The analytical solution given by Eq. (18) provides the solu-
(1990). tion for a finite square well in an infinite space. Results will
5
M. Belloni and W. Christian, “Time development in quan- therefore differ from the numerical results as the width of
tum mechanics using a reduced Hilbert space approach,” the finite square well approaches the width of the infinite
Am. J. Phys. 76, 385–392 (2008). square well. The Appendix provides an analytical solution
6
E. Dagotto, “Correlated electrons in high-temperature su- for the δ-function potential in an infinite square well, which
perconductors,” Rev. Mod. Phys. 66763 (1994). differs from the textbook solution of an infinite square well
7
J. L. Marin and S. A. Cruz, “On the harmonic oscillator in infinite space.
15
inside an infinite potential well,” Am. J. Phys. 56, 1134– These difficulties are well known. For example, bosons in a
1136 (1988). harmonic trap experience a short range boson-boson inter-
8
M .T. Dove, Introduction to Lattice Dynamics (Cambridge action. If we try to expand the many-particle wave function
University Press, Cambridge, 1993). in terms of the eigenstates of the harmonic oscillator poten-
9
M. H. Anderson, J. R. Ensher, M. R. Matthews, C. E. tial, convergence difficulties arise because of the disparity
Wieman, and E. A. Cornell, “Observation of Bose-Einstein in length scales. Our problem illustrates this difficulty at
condensation in a dilute atomic vapor,” Science 269, 198– the single particle level. I am indebted to Doerte Blume
201 (1995). and Pierre-Nicholas Roy for discussions on this point.
10 16
J .R. Schrieffer, Theory of Superconductivity (Ben- Though often referred to as “Heisenberg matrix mechan-
jamin/Cummings, Don Mills, 1983). ics” it is now clear that Born should be credited with this
11
F. Marsiglio, “The spectral function of a one-dimensional technique. See, for example, Nancy Thorndike Greenspan,
Holstein polaron,” Phys. Lett. A 180, 280–284 (1993); The End of the Certain World: The Life and Science of
ibid., “Pairing in the Holstein model in the dilute limit,” Max Born (Basic Books, Cambridge, MA, 2005).
Physica C 244, 21–34 (1995).
12
W. H. Press, B. P. Flannery, S. A. Teukolsky, and W. T.

View publication stats