The Laplace Method For Energy Eigenvalue Problems
The Laplace Method For Energy Eigenvalue Problems
net/publication/362732383
CITATIONS READS
0 276
3 authors, including:
Anna Galler
Graz University of Technology
26 PUBLICATIONS 248 CITATIONS
SEE PROFILE
All content following this page was uploaded by Anna Galler on 26 July 2023.
17 August 2022
1. Introduction
If the integrand is the derivative of a complex-valued function that has the same value
at the endpoints of the contour γ, then we can immediately obtain the solution to the
differential equation. We require that R(z) satisfies the following differential equation
d
P (z)R(z) = [Q(z)R(z)] , (7)
dz
and that the function V (z) defined by
V (z) = Q(z)R(z)eξz (8)
has equal values at the endpoints of the contour (or is single valued for a closed contour).
Note that more than one contour can satisfy these conditions; indeed, for an order-m
equation, we know that exactly m linearly independent solutions are represented by m
inequivalent contours.
The function R(z) is then found by integrating Eq. (7) after dividing it by Q(z)R(z)
and recognizing the resulting logarithmic derivative. Solving, yields R(z) as
Z z
P (z 0 ) 0
1
R(z) = exp dz . (9)
Q(z) Q(z 0 )
Armed with R(z), we immediately find the solution of the original differential equation
to be
Z z
P (z 0 ) 0
Z
ξz 1
Φ(ξ) = e exp dz dz, (10)
γ Q(z) Q(z 0 )
where the contour γ needs to be chosen so that the vanishing endpoint condition is
fulfilled. In quantum mechanics, we require the solution to satisfy additional properties,
such as being always finite, or being square-integrable.
Next, we apply these methods to solve for the bound states of the simple harmonic
oscillator in one- two- and three-dimensions. We do the same for the Coulomb problem
in two- and three dimensions, as well as solving the Morse potential. We do not consider
the spherical harmonics problem, or the Pöschl-Teller and Hulthén potentials, because
all of these cases require full hypergeometric functions for the wavefunction. These can
be solved by the closely related Laplace transform method, as discussed in the work of
Tsaur and Wang [9].
We start with the application of the Laplace method to the bound-state problem. As
mentioned above, we examine the simple harmonic oscillator in one-, two- and three-
dimensions; the one-dimensional case is treated twice because there are two different
forms for the wavefunction ansatz that are commonly used. We also cover the inverse
r potential (Coulomb problem) in two- and three-dimensions. Finally, we discuss
The Laplace method for energy eigenvalue problems in quantum mechanics 5
Independent Wavefunc.
Problem Potential
Variable Form
1D SHO, µω 2
V = 12 µω 2 x2 ξ= x Φ(ξ)
Even ~
1D SHO, µω 2
V = 12 µω 2 x2 ξ= x xΦ(ξ)
Odd ~
2D µω 2
V = 12 µω 2 ρ2 ξ= ρ ρ|m| Φ(ξ)eimφ
SHO ~
3D µω 2
V = 12 µω 2 r2 ξ= r rl Φ(ξ)Ylm (θ, φ)
SHO ~
2D 2
q
V = − eρ ξ= −2µE
~2
ρ ρ|m| Φ(ξ)eimφ
Coulomb
3D 2
q
−2µE
V = − er ξ= ~2
r rl Φ(ξ)Ylm (θ, φ)
Coulomb
Morse √ √
−2µE
2 2µV0 −ax
V = V0 (e−2ax − 2e−ax ) ξ = a~
e ξ a~ Φ(ξ)
Potential
1D SHO p µω µω 2
V = 12 µω 2 x2 ξ= x e− 2~ x Φ(ξ)
Method 2 ~
Table 1. Quantum-mechanical potentials with bound states that are analyzed in this
work. For each, we give the form of the potential, the general form of the wavefunction
where Φ is the unknown part determined by using the Laplace method, and the form
for the independent variable ξ used for each problem. Notably, 0 ≤ ξ < ∞ in all
cases but the last (where −∞ ≤ ξ ≤ ∞). In the table, m is the z-component angular
momentum quantum number with eigenvalues ~m and l is the total angular momentum
quantum number with eigenvalue ~2 l(l + 1). Moreover, µ is the mass of the (effective)
particle, ω is the frequency of the oscillator, E is the energy of the corresponding
energy eigenstate, a is a real constant with units of inverse length, and V0 has units of
energy.
one-dimensional simple harmonic oscillator, we have a different form for the even and
the odd solutions, in higher dimensions, we separate out the angular and radial degrees
of freedom—in two dimensions we use ρ and φ for the polar coordinates, while in three-
dimensions we use r for the radial coordinate and θ (polar angle to the z-axis) and φ
(azimuthal angle in the x-y-plane) for the angular coordinates. The radial functions
also have a power-law behavior as they approach the origin included in the ansatz. The
second method for the simple harmonic oscillator in one dimensions removes a Gaussian
factor from the wavefunction. The independent variable (always dimensionless and
denoted by ξ) is proportional to the square of the (radial) coordinate for the simple
harmonic oscillators, is linear for the Coulomb problems and is exponential for the
Morse potential. For the second way we treat the simple harmonic oscillator in one
dimension it is linear.
Table 2. More details on the solutions for the wavefunctions using the Laplace method.
The 1D SHO can be solved in two ways with the Laplace method, and we treat them
both. In the second column, we provide the form of the Schrödinger equation obtained
by the substitutions detailed in Table 1. Notably, for all but the last case, the coefficient
of the first derivative term is either an integer or a half-odd integer, except in the Morse
potential. It is larger than 1 for all of these cases except the first row. Moreover, a0 is
the reduced Bohr radius (a0 = ~2 /µe2 ).
Note that the ansatz we use is not the standard one used in textbooks, where one
incorporates the asymptotic behavior for large argument of the wavefunction as well.
Instead, we are following the methodology of Schrödinger in his original paper, where
only the behavior at the origin is incorporated into the ansatz.
The Laplace method for energy eigenvalue problems in quantum mechanics 7
Plugging the wavefunction ansatz and the functional form of the independent
variable into the Schrödinger equation, yields a differential equation in the Laplace
form for all of the examples we are working on in this paper. The results of this exercise
are summarized in Table 2. There, you can see that all but the last row (which will
be treated separately) have Schrödinger equations that have been transformed into the
form:
ξΦ00 (ξ) + βΦ0 (ξ) + δ − λ2 ξ Φ(ξ) = 0,
(11)
where β, δ, λ ∈ R with λ = 1 or 21 > 0 in the bound-state cases we consider here (when
we later consider continuum solutions, the transformed Schrödinger equation for Φ has
the same form, but λ becomes an imaginary number, as we discuss below). Notably, the
form of Eq. (11) is the same form used to treat the three dimensional hydrogen atom, as
discussed in Refs. [1, 8]. Nevertheless, the differences in the parameters in the Laplace
form of the final differential equation, along with the different wavefunction ansatzes,
mean that the analysis can be slightly different for some of these cases. We summarize
the procedure in as general terms as possible, and show where the cases vary, as needed.
Note further that this is not the standard Kummer equation, which arises when one
includes the asymptotic behavior of the wavefunction for large argument as well. The
form we use is in many respects easier to derive and simpler to work with. Of course, we
will see the final answers are the same as one obtains with the standard methodology,
as must be so.
As an illustration of this procedure, consider the even solutions of the one-
dimensional simple-harmonic oscillator, as given in the first row. We use ξ = µωx2 /~,
so the kinetic energy operator becomes
~2 d2
2
d 1 d
− → −2~ω ξ 2 + (12)
2µ dx2 dξ 2 dξ
and the time-independent Schrödinger equation transforms into the equation in the top
row of Table 2 after we divide both sides by −2~ω and we put all terms on the left hand
side of the equation. The other rows are derived similarly.
To begin the Laplace method, we construct the P and Q polynomials as described
in the introduction. They become
P (z) = βz + δ, (13)
2 2
Q(z) = z − λ = (z − λ)(z + λ). (14)
The ratio of P/Q is then
P (z) α+ α−
= + , (15)
Q(z) z−λ z+λ
where
βλ + δ βλ − δ
α+ = and α− = . (16)
2λ 2λ
This can be immediately integrated and exponentiated to form the integrand for the
contour-integral form of the solution, which, up to a constant prefactor, is given by
Z
Φ(ξ) = dz eξz (z − λ)α+ −1 (z + λ)α− −1 , (17)
γ
The Laplace method for energy eigenvalue problems in quantum mechanics 8
where we still need to choose the contour γ. We must choose the contour γ so that the
value of the integrand of Eq. (17) multiplied by Q(z) = (z − λ)(z + λ), has equal values
at the endpoints. Closed contours will always satisfy this. If the contour is not closed,
then the function that must be equal at the endpoints is
V (z) = eξz Q(z)R(z) = eξz (z − λ)α+ (z + λ)α− (18)
and this must hold for all ξ. Note that we always have α+ > 0 and α+ + α− = β > 0,
but α− can take on positive or negative values. In general, it is difficult to find contours
where this function will be equal at the endpoints (for arbitrary values of V ) if the path
is not a closed path. However, special values, such as 0 or ∞, are easier to enforce,
because they automatically hold for all ξ; for V = ∞ one must use a careful limiting
procedure to ensure the difference of V at the two endpoints actually vanishes.
The selection of these contours, and examining the integral over inequivalent
contours is at the heart of the Laplace method. It requires a discussion of some technical
details from complex analysis. Specifically the integrand is generically multivalued when
α± are nonintegers. Consider a complex valued function of the form G(z) = z α . We can
write this as
G(z) = z α = eα log(z) . (19)
The complex logarithm is multivalued in the complex plane; consider
log(z) = log |z|eiθ = log(|z|) + iθ.
(20)
Clearly, if we fix |z| and move through θ from 0 → 2π, we will return to the same point
on the complex plane, but the value of the logarithm will have picked up an additive
shift by 2πi and thereby will not be a continuous (or even well-defined) function. So,
we instead write the logarithm function as log(z) = Log(z) + 2πim, m ∈ Z, where
Log indicates the principal value of the logarithm function, which we take to be the
complex logarithm function for 0 < θ < 2π in this example; note that this choice is a
domain that has a cut in the complex plane running from 0 to infinity along the positive
real axis. Log(z) is then single-valued, but the domain is smaller than the full complex
plane due to the cut. This part of the complex plane given by 0 < θ < 2π is a branch
of the logarithm, defined to be a domain in the complex plane on which a function is
single-valued. A branch is chosen to be maximal, in the sense that no point can be
added to it and still maintain single-valuedness.
The boundary of a branch is called the branch cut. From our definition of the
complex logarithm, it is clear that the branch cut will be the ray where θ = 0 (or more
completely, where θ = (2nπ for n ∈ Z) originating at the origin; that is, the positive
real axis plus the origin. In this case, the origin is a branch point of the logarithm
function since it is a point that is common to any branch cut one can draw for the
logarithm. The branch cut can be any curve that extends from the origin to infinity
to have a single-valued logarithm function. Note that the complex plane, C, does not
include infinity, and as such there is no “point at infinity;” instead approaching infinity,
means that the complex number continues to increase without bound (|z| → ∞).
The Laplace method for energy eigenvalue problems in quantum mechanics 9
contain any branch points. This is because such a closed contour always gives a zero
integral due to Cauchy’s theorem and the fact that the function is analytic inside the
domain defined by the branch. An open contour that goes from a finite point z to
another finite point z 0 inside the domain (that does not cross a branch cut) is also ruled
out because it will not be possible to satisfy V (z) = V (z 0 ) for all 0 ≤ ξ < ∞.
These restrictions then imply that contours must go to infinity, or end at a branch
point (technically they could also end anywhere on a branch cut, but similar to the
open contour case above, one won’t be able to satisfy the condition V (z) = V (z 0 ) for
all allowed ξ in that case). Rather than determine all possible contours next, we now
consider the condition that the solution must be finite as ξ → ∞. Since there is a
factor of eξz in the integrand, no contour can go to infinity with Re(z) > 0 without
the integral diverging as ξ → +∞. Moreover, the contour cannot have an end point at
z = λ because using the stationary phase method to asymptotically determine the limit
of the integral for large ξ, again has an exponential term that will diverge (for details
see [8]).
For all cases, we can have a branch cut running from −∞ to −λ along the negative
real axis and from λ to ∞ along the positive real axis. When β is an integer, we can
instead have the branch cut run along the real axis between −λ and λ. These two
possibilities are depicted in Fig. 1.
We describe next the three possible contours we can have for all 0 < ξ (we will
determine the behavior for ξ = 0 last). Two contours can be drawn when we have the
branch cuts extending from −λ to −∞ along the negative real axis and from λ to ∞
along the positive real axis. The first one, γ1 , is the so-called Hankel contour, which
starts at −∞ just below the real axis, loops around the branch point, and goes back to
−∞ above the real axis (see Fig. 1 a). The second possibility, γ2 , starts at the branch
point at −λ and goes to −∞ (either above or below the real axis, both end up giving
the same asymptotic behavior as ξ → 0; we draw it above the real axis here). The final
contour can be drawn when the branch cut extends from −λ to λ (which is possible
only because α+ + α− = β is an integer). This contour, γ3 , encircles the branch cut and
the branch points; it is sometimes called the dog-bone contour. Note that a similar γ2
exists when the branch cut runs from −λ to λ, but it behaves the same as the γ2 that
we analyze. A stationary phase analysis of γ3 shows that it diverges as ξ → ∞, so it is
ruled out [8].
Now, we need to determine if any of the remaining contours yield a finite result
as ξ → 0. First, in the case of the Hankel contour, the phase of the integrand will be
different below and above the real axis, since we drew a branch cut along the negative
real axis. Hence, Φ(ξ) → ∞ for ξ = 0 for the Hankel contour, because the integrand
remains finite but the contour extends an infinite length, and the difference in the phases
means those infinite contributions do not cancel. Thus, we can rule out the γ1 contour.
For the contour γ2 from −λ → ∞ (in the left half plane), if α− < 0, the integral will
diverge as z → −λ. If α− ≥ 0, we find that the contour does satisfy the condition that
V (z) has equal values (in fact V (z) = 0) at the endpoints of the contour γ2 for ξ > 0.
The Laplace method for energy eigenvalue problems in quantum mechanics 11
Figure 1. The contours we consider when α± are not integers. The branch cuts in
each case are shown by the dashed red lines.
Figure 2. Contours that lead to the correct solution for the wavefunction and provide
the quantization condition for the energy. The contour is always a closed contour in
the counter-clockwise direction encirling the point z = −λ. In some cases, shown in
panel (b), a branch point remains at z = λ and the branch cut runs from there to
z = ∞ along the real axis.
2πi d|α− | ξz α+ −1
Φ(ξ) = e (z − λ) . (24)
|α− |! dz |α− | z=−λ
From the Rodrigues formula for the associated (alternatively, ”generalized”) Laguerre
polynomials [10],
(b) 1 ξ −b dN −x N +b
LN (ξ) = eξ e x , (25)
N! dxN x=ξ
we can see that the form of Eq. (24) is similar. To establish the exact correspondence, we
define u = −(z−λ)ξ (the point at which we evaluate the derivative, z = −λ, corresponds
to u = 2λξ) and the inverse is z = (λξ − u)/ξ. Re-expressing the right-hand side as a
function of u yields
" α+ −1 #
|α− |
2πi d u
Φ(ξ) = (−1)|α− | ξ |α− | |α− | e−u eλξ (−1)α+ −1
|α− |! du ξ
u=2λξ
β−1 |α− |
2πi(−1) d
ξ −β+1 eλξ
−u α+ −1
= e u
|α− |! du|α− | u=2λξ
β−1
2πi(−1) (β−1)
= −β+1
e−λξ L|α− | (2λξ). (26)
(2λ) |α− |!
Hence, except for a constant prefactor, which will be set via normalization, we find that
(β−1)
Φ(ξ) ∝ e−λξ L|α− | (2λξ) . (27)
The Laplace method for energy eigenvalue problems in quantum mechanics 13
Because the Laguerre polynomial of order N goes like ξ N as ξ → ±∞, the asymptotic
behavior of Φ(ξ) is dominated by the exponentially decaying term. Thus, this solution
will be finite for all 0 ≤ ξ and is the wavefunction we sought to find.
The last detail we need to work out is how the integer from the quantization
condition N = |α− |, relates to the conventional principal quantum number n. In all
harmonic oscillator cases, we have n = N , while the two dimensional Coulomb problem
has n − |m| − 1 = N and the three-dimensional Coulomb problem has n − l − 1 = N .
These results are summarized, in terms of the principal quantum number, in Table 3.
Quantization Energy
Problem Form of Φ(ξ)
Condition Quantization, En
1D SHO, E 1 (− 1 )
~ω 2n + 12
N =n= 2~ω
− 4 e−ξ/2 Ln 2 (ξ)
Even
1D SHO, E 3 1
(1)
N =n= 2~ω
− 4
~ω 2n + 1 + 2 ξ 1/2 e−ξ/2 Ln 2 (ξ)
Odd
2D E 2|m|+1 (|m|)
N =n= 2~ω
− 2
~ω (2n + |m| + 1) e−ξ/2 Ln (ξ)
SHO
3D E 1 2l+3 3
(l+ 1 )
N =n= 2~ω
− 2 2
~ω 2n + l + 2 e−ξ/2 Ln 2 (ξ)
SHO
2D 1
− ~2 (2|m|)
N = n − |m| − 1 = √~
a0 −2µE
− |m| + 2 2µa20 (n− 12 )
2 e−ξ Ln−|m|−1 (ξ)
Coulomb
3D 2 (2l+1)
N =n−l−1= √~
a0 −2µE
− (l + 1) − 2µa~2 n2 e−ξ Ln−l−1 (ξ)
Coulomb 0
Morse √ √ √ 2
2µV0 − −2µE 2 2 2µV0 (2n−2δ−1)
N =n= a~ − a2µ~ n− a~
e−ξ/2 Ln (ξ)
Potential
1D SHO E 1 1
N =n= − 2
~ω n + 2
Hn (ξ)
Method 2 ~ω
Table 3. Quantization condition for the Laplace method (by tradition, the principal
quantum number n starts from 0 for all cases, except the Coulomb cases, where it starts
from |m| + 1 in two dimensions and from l + 1 in three dimensions); N is required to be
a nonnegative integer from the quantization condition arising from Laplace’s method.
By tracing back through the definitions of Φ and ξ in each case, one obtains the
standard wavefunctions for each problem, up to a normalization constant, that still
needs to be determined. The δ in the index of the associated
√ Laguerre polynomial in
the last column of the Morse potential satisfies δ = 2µV
a~
0
. All models, except for the
Morse potential, have an infinite number of bound states. The Morse potential has a
√
finite number, where we are required to have n < 2µV0 /a~.
Having now constructed the general methodology to solve bound-state problems using
the Laplace method, we show the concrete details for how the method is used for
the two-dimensional simple harmonic oscillator, which we treat in polar coordinates.
The Schrödinger equation is solved first by separating variables, that is by letting
The Laplace method for energy eigenvalue problems in quantum mechanics 14
From Eqs. (27) and (32), we find the desired solution to the differential equation is
ξ
Φ(ξ) = e− 2 Ln(|m|) (ξ) (34)
as summarized in Table 3. This then yields the following for the full wavefunction:
ρ2 (|m|) µω 2
|m| − µω
ψn,m (ρ, φ) = ρ e 2~ Ln ρ eimφ , (35)
~
up to a normalization constant, which has not yet been determined.
The solutions we derived for the one-dimensional simple harmonic oscillator may
not look familiar to many. But they actually are the standard form, expressed in terms
of a Gaussian multiplied by a Hermite polynomial, once we realize that there is an
identity relating associated Laguerre polynomials to Hermite polynomials, given by
(− 1 )
H2n (x) = (−4)n n!Ln 2 x2
(36)
1
( )
H2n+1 (x) = 2 (−4)n n!xLn 2 x2 ,
where Hn is the physicist’s form for the Hermite polynomial defined by the Rodrigues’
formula n
2 d
−x2
Hn (x) = (−1)n ex e . (37)
dxn
The Laplace method for energy eigenvalue problems in quantum mechanics 15
The remaining cases follow in a similar fashion and can be constructed by following
the summarizing formulas in Tables 1–3.
What remains is for us to discuss the second way to solve the one-dimensional
harmonic oscillator, which obtains the Hermite polynomials directly from their
Rodrigues’ formula after determining the residue at the pole. This method is
summarized in the last row of Tables 1-3 and it uses a different ansatz for the
wavefunction, which also leads to a differential equation in the Laplace form, but a
different one from all of the other cases. It is given by
where α = 12 − ~ω E
; note we must have E ≥ 0 (since the Hamiltonian is a positive
semidefinite operator in this case), so α ≤ 12 . In addition, we need to find a finite
solution for all −∞ < ξ < ∞. We now go through the steps of the Laplace method.
First, compute the polynomials
P (z) = z 2 − 2α (39)
Q(z) = −2z, (40)
and then exponentiate the antiderivative of the ratio P/Q, which now includes only
one power of z raised to a potentially noninteger exponent. This allows us to write the
solution to Eq. (38) as the contour integral
Z
z2
Φ(ξ) = dz eξz− 4 z α−1 . (41)
C
Next, we restrict the exponent of z such that α ∈ / Z. The branch points of the integrand
in Eq. (41) are at z = 0 and at infinity. To construct a single branch of the integrand,
we choose to draw the branch cut from z = 0 → z = −∞ along the negative real axis.
We next consider the possible contours as shown in Fig. 3. By the same arguments
used before, we can eliminate all closed contours that don’t contain a branch point or
end at a branch cut and all open contours between any two finite points in the complex
plane. Thus, the only candidate contours are ones with at least one endpoint at z = 0 or
that have endpoints that go to infinity. As the contour goes to infinity, it must remain
inside a cone with an angle of ±π/4 of the real axis, to be bounded. There is a cone
around the positive real axis and also one around the negative real axis.
Thus, the contours to analyze are a Hankel-like contour around the branch cut,
which we call γ4 , a contour from negative infinity to positive infinity (which does not
cross the negative real axis), called γ5 , and a contour from the origin going to infinity
with Re(z) < 0, called γ6 , or from the origin to infinity with Re(z) > 0, called γ7 ; as
long as we remain inside the cones about the real axis as we go to infinity. It is fairly
easy to see any other contour with endpoints at 0 and ∞ can be deformed into one of
these contours or can be mapped to them by taking z → −z. For example, the integral
over γ7 is converted into a contour from the branch point at 0 that runs to infinity inside
The Laplace method for energy eigenvalue problems in quantum mechanics 16
a) b)
Figure 3. Four possible contours for solving the one-dimensional simple harmonic
oscillator using the second ansatz. The contours must lie within the white “cone”
regions as they go to infinity (there is no other restriction on the contours except not
to cross the branch cut for finite values of z). In panel (a), we show the Hankel contour
γ4 , which goes around the branch cut and γ5 , a contour that runs parallel to the real
axis. In panel (b), we show two contours starting from the branch point and running
to infinity either below (γ6 ) or above (γ7 ) the branch cut.
the white cone along the positive real axis by transforming z → −z, which is equivalent
(up to a complex phase) to the integral over γ7 but with ξ → −ξ.
The first condition we have is that the function V (z) has identical values at the
endpoints for all ξ. In this case, we have
z2
V (z) = eξz− 4 z α . (42)
It is easy to see that limz→0 V (z) = ∞ if α < 0, while limz→0 V (z) = 0 if α > 0. The
2
asymptotic behavior of V is dominated by the e−z term, when |z| → ∞, so we have
limz→∞ V (z) = 0 when we lie inside the white cone about the real axes. This implies,
that we can only have z = 0 as an endpoint, when α > 0. But, in that case, any integral
that has 0 as an endpoint diverges, due to the power-law behavior of the integrand near
z = 0 having too negative of an exponent. So no contour can have z = 0 as an endpoint,
eliminating γ6 and γ7 .
The next condition to check the integral as ξ → −∞. Since the contour lies
inside the cone, it seems like it will always be bounded, and should never diverge.
2
But, in the region near z ≈ 2ξ, the integrand actually behaves like eξ and can give
large contributions. The way to evaluate such situations is to perform an asymptotic
analysis on the integrals to determine their value for large |ξ|. The standard way to do
this is called the steepest-descents approach, which notes that the contributions to the
exponential are largest near the maximum of the exponent, which is then described by a
simple quadratic near the extremum, yielding a Gaussian integral that can be evaluated
exactly. The analysis is straightforward, but is not often taught, so we describe it
carefully, starting with the Hankel contour γ4 .
The saddle-point approximation allows us to approximate integrals of the form
Z
Φ(ξ) = dz eh(z) g(z), (43)
γ
The Laplace method for energy eigenvalue problems in quantum mechanics 17
which corresponds to the integral we need to evaluate for the Laplace method solution.
Note that we have two options for how to proceed here. We can pick g(z) = z α−1 , or we
can pick g(z) = 1 via writing z α−1 = e(α−1) ln z and absorbing this term into the exponent
h(z). Both approaches yield the same final results, but the former is simpler than the
latter, because there is only one saddle point in this case, making the analysis simpler.
This means we have h(z) = ξz − 41 z 2 . Taking the derivative, to find the extrema, we have
h0 (z) = ξ − 12 z. Setting this equal to zero, tells us the extremum occurs at z = z0 = 2ξ,
which is called the saddle point; note that we have h(z0 ) = ξ 2 . In complex analysis, one
direction through the saddle point is a minimum, while the other is a maximum, yielding
a saddle-point shape for the exponent near the saddle point z0 . One must choose the
direction for the contour through the saddle point to traverse the maximum, not the
minimum. In this case, the maximum direction is along the real axis, which is simple
to see, because we have a quadratic for the exponent with a negative real coefficient.
The asymptotic analysis next deforms the contour to go through the saddle point
along the maximum direction. When ξ < 0, this saddle point lies on the branch cut, so
we will deform the Hankel contour to pass infinitesimally below it and parallel to the real
axis—once below the negative real axis and once above. This yields two contributions
for the steepest-descents integral. For the contribution from below the real axis, we
parameterize the contour as given by γ ≈ z0 + t near the saddle point, so that
1
h(z) ≈ h(z0 ) − t2 . (44)
4
Since the integrand decays quickly away from the saddle point, we extend the limits
on t to run from −∞ to ∞ and we approximate g(z) ≈ g(z0 ). This then gives us the
contribution from the saddle point below the real axis to be
Z ∞
1 2 √ 2
dteh(z0 )− 4 t g(z0 ) ≈ 2 π(2|ξ|)α−1 e−iπ(α−1) eξ . (45)
−∞
The contribution from the saddle point above the negative real axis, is similar—it has
an overall negative sign, because the contour runs from right to left instead of left to
right and the sign of the phase in the exponent is positive because we are above the
branch point. The total asymptotic estimate for the integral is then
√
Z
1 2 2
dzeξz− 4 z z α−1 ≈ −2α+1 i π|ξ|α−1 sin π(α − 1)eξ . (46)
γ4
Since we assume α is not an integer, the coefficient is nonzero and this gives a leading
2
contribution that goes like eξ . Looking at Table 1, we see that the full wavefunction
1 2 1 2
is proportional to e− 2 ξ Φ, so this solution will go as e 2 ξ as ξ → −∞, which diverges.
So, γ4 does not yield a finite wavefunction. Moreover, a similar analysis yields the same
asymptotic behavior for γ5 . Note that this analysis is similar to the Fröbenius analysis
when the series does not truncate, and we reject the solution due to the wavefunction
growing as we go to infinity.
The Laplace method for energy eigenvalue problems in quantum mechanics 18
So we do not obtain a finite solution from any of the possible contours, and we
again must change our assumption that α ∈ / Z to allow for a new contour. When
α ∈ Z, the integrand is single-valued, so there no longer is a branch point or branch
cut. Consequentially, we can now enclose the origin with a new closed contour. Our
only choice is then that α ≤ 0, yielding the quantization condition given by
E 1 1
n = −α = − ⇒ En = ~ω n + , n = 0, 1, 2... (47)
~ω 2 2
This determines the energy levels of the one dimensional oscillator. Now, we write the
(unnormalized) solution to the differential equation for this closed contour as
I
z2
Φ(ξ) = dz eξz− 4 z −n−1 , n = 0, 1, 2 · · · , (48)
with a closed contour that encircles the origin. By completing the square in the
exponential term in Eq. (48), we can re-write this integral as
I
z 2
Φ(ξ) = e ξ2
dz e−(ξ− 2 ) z −(n+1) . (49)
n ξ2 dn −u2 n ξ2 d
n
−ξ 2
Φ(ξ) = (−1) e lim n e = (−1) e e . (51)
u→ξ du dξ n
This is precisely the Rodrigues formula for the nth degree Hermite polynomial Hn (ξ).
That is, up to a constant prefactor,
r
µω
Φ(ξ) = Hn (ξ) = Hn x , (52)
~
which allows us to write the (unnormalized) wavefunction for the 1D harmonic oscillator
as: r
µω 2
− 2~ x µω
ψ(x) ∝ e Hn x . (53)
~
There are several quantum systems whose energy eigenstates have energy eigenvalues
that lie in the continuum and that we can also treat with the Laplace method; this
includes the free particle in two, and three dimensions, the continuum solutions of
the Coulomb problem in two- and three-dimensions, and the continuum solutions of
the one-dimensional Morse potential. The steps for obtaining a differential equation
The Laplace method for energy eigenvalue problems in quantum mechanics 19
Independent Wavefunc.
Problem Potential
Variable Form
2D Free q
V =0 ξ = 2µE
~2
ρ ρ|m| Φ(ξ)eimφ
Particle
3D Free q
2µE
V =0 ξ= ~2
r rl Φ(ξ)Ylm (θ, φ)
Particle
2D 2
q
V = − eρ ξ= 2µE
~2
ρ ρ|m| Φ(ξ)eimφ
Coulomb
3D 2
q
V = − er ξ= 2µE
~2
r rl Φ(ξ)Ylm (θ, φ)
Coulomb
Morse √ √
2µE
2 2µV0 −ax
V = V0 (e−2ax − 2e−ax ) ξ = a~
e ξi a~ Φ(ξ)
Potential
Table 4. Summary for how to convert the Schrödinger equation into the Laplace
equation for the five problems with continuum solutions. For each, we give the form of
the potential, the general form of the wavefunction where Φ is the part of the solution
that is found by using the Laplace method, and the independent variable ξ used for
each problem. Note that E > 0 for these continuum problems. We have m denoting
the quantum number for the z-component of angular momentum and l denoting the
quantum number for the total angular momentum.
in the Laplace form are similar to what we already showed above, and in Table 4 we
summarize the results for these different models. Note that the substitutions required
for the free-particle problems look like the final wavefunction will diverge at the origin,
but we require the function Φ(ξ) to have a high-enough order zero at the origin that the
final wavefunction remains finite everywhere. In Table 5, we show the final differential
equations obtained by this procedure, which is similar to the bound-state form in
Eq. (11), but with the sign of the λ̄2 term changed. We have written that term as
−(iλ̄)2 instead of +λ̄2 to simplify the notation that we need in solving the problem. The
Morse potential, on the other hand, keeps the form of Eq. (11), but some parameters
now become complex.
We begin with the free particle and Coulomb problems, each treated in both two
and three dimensions, because they all are treated similarly. The differential equation
in the Laplace form takes the form
for all of these cases. This means that we can write the solution as
Z
α −1 α −1
Φ(ξ) = dz eξz z − iλ̄ + z + iλ̄ − , (55)
γ
iβ λ̄ ± δ
α± = . (56)
2iλ̄
The Laplace method for energy eigenvalue problems in quantum mechanics 20
Figure 4. Rotated ’dog-bone’ shaped contour for evaluating the contour integral for
the continuum wave functions of the Coulomb problem in two and three dimensions.
multivalued. In this case, the integral must be taken along the imaginary axis from −i
to i, connecting the two points where V = 0 for all ξ. Note that this choice could also
have been used for the Coulomb potential cases (because V = 0 at the branch points for
all ξ as well), but there is pedagogical value to using the dog-bone contour because it
allows us some additional options for evaluating Φ(ξ) numerically, as we discuss below;
since the solutions corresponding to both choices of the contour are proportional to each
other (shown below), we can freely choose either one. The ambiguity in the prefactor
is always removed when the final wavefunction is normalized (but we will not discuss
normalization in this work).
It is critical to evaluate the phases properly, when determining the integrand. With
the branch cut structure we use, the function is single-valued everywhere in the complex
plane, except along the branch cut itself, but the function has different values on both
sides of the branch cut. We start by picking a reference point, which will be the origin,
just to the right of the branch cut, which we call z = 0+ , as shown in panel (a) of
Fig. 5. The multivalued function in the integrand is f (z) = (z − i)α+ −1 (z + i)α− −1
and we focus on how to determine the phase consistently for this function. We find
π
that f (0+ ) = (−i)α+ −1 (i)α− −1 , which we evaluate
with the
standard phases i = ei 2 and
π
(−i) = e−i 2 . Then we have f (0+ ) = exp i π2 (α− − α+ ) . Noting the form of α± , we
find f (0+ ) = exp(− π2 δ).
Now, to calculate f (z) anywhere in the complex plane, we first draw a path from
the reference point 0+ to z that does not cross the branch cut (see panels (b) and (c)
of Fig. 5). Then, we examine how arrows drawn from the upper branch point i to a
point on the path rotate as we move along the path from 0+ to z. This determines
the change in the phase for the factor (z − i). We repeat with an arrow drawn from
−i to a point along the path, and follow it from 0+ to z. The rotation of the arrow
here, again determines the change in the phase for the factor (z + i). Each of those
factors will have the change in phase multiplied by the corresponding exponent, and
that will determine the phase of f (z). We will show below that using this procedure
produces a single-valued function over the entire complex plane. But first, we will use
this procedure to convert the integral form of our solution to a single integral that runs
along the real axis.
We deform the contour in Fig. 4 to be infinitesimally close to the imaginary axis
along the vertical portions, and wrapping infinitesimally close to the branch points in
the circular portions, we obtain a result expressed as the sum of the contributions from
two vertical lines (downward from i → −i just to the left of the axis and upward from
−i → i to the right of the axis) and from the infinitesimal circular arcs enclosing the
two branch points. It is easy to see that the contributions from the circular arcs around
the branch points will be zero. Since Re(α± ) > 0, the integral around each arc will go
to zero, so the branch points do not contribute to the integral. The integral over the full
contour then reduces to the sum of the contributions from the two vertical lines. Note
that since they lie on either side of the branch cut, there is a phase difference between
the two integrals, and the contributions will not cancel.
The Laplace method for energy eigenvalue problems in quantum mechanics 22
a) b) c)
Figure 5. Procedure for determining the phases of the integrand in Eq. (55) along
the vertical pieces of the contour (we deformed the contour slightly for clarity in the
image). a) We pick a reference point z = 0+ and show arrows from i and −i to the
reference point. b) We draw paths from the reference point 0+ to each point z of this
piece of the deformed contour Γc (here running upward vertically to the right of the
branch cut). The arrows drawn from the upper and lower branch points to z do not
change their net direction as they move along the path from 0+ to z, thus the phase
for f (z) is the same as the phase for f (0+ ) on the right side of the branch cut. c)
For reaching the points of the contour on the left side of the branch cut, the arrow
drawn from i to z needs to rotate by 2π, while the net change in direction of the arrow
drawn from the lower branch point is zero; note that the arrow along the path from
the reference point to z is allowed to cross the branch cut, even though the path never
crosses the branch cut.
To determine this phase difference, we follow the procedure described above (see
Fig. 5 for a graphical representation). We start with the piece of the contour that runs
from −i to i along the right hand side of the imaginary axis. We can draw a path from
the reference point at 0+ to any point along this piece of the deformed contour as a
straight vertical line. We can immediately see that the arrow from i to z on the contour
does not change direction after moving along the path. Neither does the arrow drawn
from −i to z. This means the phase for f (z) is the same as the phase for f (0+ ) along the
entire piece of the contour. We write z = iy, with y real, and this part of the contour
integral becomes the following:
Z 1
π
I1 = i dy eiyξ |y − 1|α+ −1 |y + 1|α− −1 e− 2 δ . (57)
−1
The second piece of the contour runs along the left hand side of the imaginary axis from
i to −i. We draw a path from the reference point 0+ to z along the contour, by an arc
that goes around the upper branch point at i. One can immediately see that an arrow
drawn from i to the reference point 0+ , winds by 2π in the counterclockwise direction as
it goes around the path to z. So the change in the phase for the factor (z − i) is 2π. The
arrow drawn from −i to the reference point, will rotate first to the right, and then to
the left, but it ultimately accumulates no net phase, so its change in phase is 0. The fact
The Laplace method for energy eigenvalue problems in quantum mechanics 23
that the final point ended on the left-hand side of the branch point does not determine
the change in the phase along the path, it is the net motion of the arrow as we follow
along the path that does this. This point can often be misunderstood by students. It
is important to note that our rule for determining the phase of the function does not
input by hand a change of phase when crossing the branch cut. Instead, we follow the
described procedure to determine the change in the phases. Again, we let z = iy and we
note that the integral runs down the imaginary axis, so we find the contribution from
this piece of the contour is the following:
Z −1
π
I2 = i dy eiξy |y − 1|α+ −1 |y + 1|α− −1 e2iπ(α+ −1) e− 2 δ . (58)
1
The factor e−2iπ = 1 and can be ignored. The real part of α+ is either an integer or a
half-odd integer. In the former case, the added factor is 1 and can be ignored, in the
latter case it is −1. The imaginary part of α+ adds in a factor of eπδ . Hence, when we
combine the two integrals together (and recall that we have to switch the order of the
limits in the second integral) we find the total contour integral becomes
Z 1
− π2 δ π
Φ(ξ) = i e ∓e 2
δ
dy eiξy |1 − y|α+ −1 |1 + y|α− −1 , (59)
−1
where the minus sign is for when the real part of α+ is an integer and the plus sign is
for when it is a half-odd integer.
To convert this into a form that is easily expressed in terms of confluent
hypergeometric functions, we let y = −1 + 2x and substitute into the integral to find
that Z 1
− π2 δ π
−1 −iξ
Φ(ξ) = i e δ
∓ e2 2 α + +α −
e dx e2iξx |1 − x|α+ −1 xα− −1 . (60)
0
Comparing to the standard integral form of the Kummer function (for Re(b) > Re(a) >
0) (as given by Eq. 13.4.1 of Ref. [11])
Z 1
Γ(b)
M (a, b, z) = dx ezx xa−1 (1 − x)b−a−1 , (61)
Γ(a)Γ(b − a) 0
we find that
π π Γ(α+ )Γ(α− ) −iξ
Φ(ξ) = i e− 2 δ ∓ e 2 δ 2α+ +α− −1 e M (α− , α+ + α− , 2iξ). (62)
Γ(α+ + α− )
Note that for E > 0, the numerical prefactor is never zero, so we can always remove
it from further discussion in the summary of the wavefunctions. It will enter, and is
important, when we evaluate the results numerically below. One does need to complete
the normalization step for the final wavefunctions (which is usually done with delta-
function normalization), but we will not discuss that further here and instead will only
summarize unnormalized wavefunctions, with the prefactor removed. Note that the
result here corrects a sign error in the final result for the continuum wavefunction in
Ref. [8] arising from an inconsistent definition of the phase of the multivalued function.
The Laplace method for energy eigenvalue problems in quantum mechanics 24
We comment briefly here on the 3d free particle. This case results in just the
I1 term, which has the same final form as we have for the cases with the “dog-bone”
contour (just with a different prefactor). So its result falls into the same category as
the other three cases. It is just that we do not need to worry about any phase issues in
working with the integrand in this case. Because there is no branch point, we cannot
describe that solution with a closed contour, because such an integral always vanishes.
Instead, we simply integrate from one “zero” point to the other.
One other point to note, is that using the Kummer relation (Eq. 13.2.39 of Ref. [11])
we can show that the unnormalized Φ(ξ), given by e−iξ M (α− , α+ + α− , 2iξ), is real for
real ξ. In particular, we have
∗
e−iξ M (α− , α+ + α− , 2iξ) = eiξ M (α−
∗
, α+ + α− , −2iξ), (64)
where we used the facts that α+ + α− and ξ are both real. Then, if you note that
∗
α− = α+ = (α+ + α− ) − α− , and use the Kummer relation, the right hand side of the
equation becomes eiξ e−2iξ M (α− , α+ +α− , 2iξ), which is equal to the original function and
hence shows that this combination is real. Thus, if we drop the constant prefactor, the
unnormalized Φ(ξ) can always be chosen to be real-valued. Note as well that at ξ = 0,
we have Φ(ξ) = 1, because M (a, b, 0) = 1 for all cases where the Kummer function is
well defined from its power-series, which is the situation we have here. So this choice of
contour leads to the correct continuum wavefunction, as we claimed earlier. A summary
of the unnormalized wavefunctions for all potentials (which have real radial functions)
is given in Table 6. These results follow by simply plugging in the explicit values of
α± and noting that there are identities between confluent hypergeometric functions and
other functions, such as Bessel functions and spherical Bessel functions (as summarized
in section 13.6 of the NIST Digital Library of Mathematical Functions [11]). We do not
show the details for how to carry out that algebra here.
One of the benefits of using contour-integral representations for the continuum
wavefunctions is that it allows us to explore different ways to determine the
wavefunctions numerically. For example, in this case we have three equivalent numerical
representations. The first involves a real integral and is given in Eq. (62). The second
involves our original integral representation in Eq. (54), where, for concreteness, we use
a circular contour of radius R centered at the origin for evaluating the wavefunction.
The third is to develop a power series representation and then to numerically evaluate
the series. This is done by deforming the contour until it has a very large radius, and
then extracting the residue at infinity.
We describe how to determine this power series next. Key to completing the
calculation is the determination of the Laurent series near the point at infinity. This
is most easily determined by approaching infinity along the positive imaginary axis; we
add a constant shift by −i first, because we know the final result has a factor of e−iξ .
The Laplace method for energy eigenvalue problems in quantum mechanics 25
So, we let z = −i + i/y for y a positive real number near zero. The phase for the factor
z − i is iπ, because the arrow from the reference point wraps by π as we move up the
imaginary axis, while the phase from z + i is zero. This gives us an overall phase factor
of −eiπα+ . The integrand (including the change of variables factor −i/y 2 ) is then given
by
π ξ 1
ieiπα+ − 2 δ e−iξ ei y α+ +α− |1 − 2y|α+ −1 (65)
y
π
(the term e− 2 δ comes from the phase of the function at our reference point). We now
want to expand this in a Laurent series for small y. Using the generalized binomial
theorem for complex powers, yields
∞ ∞
iπα+ − π2 δ −iξ
X (iξ)m X (α+ − 1)j
ie e (−2)j y j−m−α+ −α− , (66)
m=0
m! j=0
j!
where (α − 1)j is the Pochammer symbol for the falling factorial, given by (α − 1)(α −
2) · · · (α−j). If we now perform a contour integral around the point at infinity, the result
will be given by the residue, which is determined by the coefficient of the expansion in
Eq. (66) of the term 1/y. Since α+ + α− is an integer, we can immediately determine
the residue. It is given by
∞
iπα+ − π2 δ −iξ
X (α+ − 1)j (iξ)j+1−α+ −α−
Residue = ie e (−2)j . (67)
j=α+ +α− −1
j! (j + 1 − α + − α − )!
The confluent hypergeometric functions are typically expressed in terms of the rising
factorials, instead of the falling factorials. Converting between them gives
Γ(j + α− ) Γ(α− )
(α+ − 1)j+α+ +α− −1 = (−1)j+α+ +α− −1 = (−1)j+α+ +α− −1 (α− )(j) .
Γ(1 − α+ ) Γ(1 − α+ )
(70)
The term in the denominator can be written as
Using Euler’s reflection formula Γ(z)Γ(1 − z) = π/ sin(πz) gives our final result:
2iπα+ ∞
− π2 δ e −1 α+ +α− Γ(α+ )Γ(α− ) −iξ
X (α− )(j) (2iξ)j
Residue = e 2 e . (73)
4π Γ(α+ + α− ) j=0
(α+ + α− )(j) j!
Multiplying by −2πi to determine the integral via the calculus of residues then produces
a result equal to that in Eq. (62), provided
∞
X (α− )(j) (2iξ)j
M (α− , α+ + α− , 2iξ) = , (74)
j=0
(α+ + α− )(j) j!
For the contour integral around the circular path (see Eq. (54)), we need to carefully
determine the phase of the multivalued function in the integrand again. We use
our standard approach, relating everything to our reference point. With the angle θ
measured relative to the positive imaginary axis, the parametrization of the radius R
π
integral is z = Rei(θ+ 2 ) = R(− sin θ + i cos θ); for concreteness, we will pick R = 2 in
the figures. Note that the conventional angle for describing z in the complex plane is
measured from the real axis, hence, the polar angle for z is θ + π2 . The integral then
becomes
Z 2π
π π α+ −1 π α− −1
Φ(ξ) = i dθ Rei(θ+ 2 ) eRξ(− sin θ+i cos θ) Rei(θ+ 2 ) − i Rei(θ+ 2 ) + i
0
π
× eiφ2 (α+ −1) eiφ1 (α− −1) e− 2 δ . (75)
Figure 6. Geometry for determining the relationship between the phases φ1 and φ2
and θ. In the figures, the symbol l is used for the unknown length on each triangle
and R is chosen to equal 2 for concreteness. This case corresponds to 0 ≤ θ ≤ π.
π, φ1 rotates from π/2 to π, etc. Overall, this yields the following relations
0 ≤ θ < cos−1 − R1 π
: 0 ≤ φ1 < 2
cos−1 − R1 ≤ θ < π π
: 2
≤ φ1 < π
π ≤ θ < π + cos−1 R1 3π
: π ≤ φ1 < 2
π + cos−1 R1 ≤ θ < 2π 3π
: 2
≤ φ1 < 2π. (88)
The Laplace method for energy eigenvalue problems in quantum mechanics 29
We can repeat the procedure for φ2 , which regards the arrow drawn from +i to z. We
get for φ2
0 ≤ θ < cos−1 R1 3π
: π ≤ φ2 < 2
cos−1 R1 ≤ θ < π 3π
: 2
≤ φ2 < 2π
π ≤ θ < π + cos−1 − R1 5π
: 2π ≤ φ2 < 2
π + cos−1 − R1 ≤ θ < 2π 5π
: 2
≤ φ2 < 3π. (89)
In general, when evaluating the integral, we can simply use a trapezoidal rule,
dividing the θ interval evenly. While the result is independent of the radius R, the
appearance of zξ in the exponent of the exponential function produces accuracy issues
for large R and ξ—this means, for accurate numerical work, we should use as small an
R as possible (we found R = 1.1 to be good with 100 000 steps). It also means at some
point, the direct numerical integration will fail when ξ is large enough, due to precision
issues similar to using the power series to compute e−x for large x.
As an example, we plot three results for the radial Coulomb wave function for small
ξ and three different energies in the continuum in Fig. 8. These results are normalized
according to the result from the 1 dimensional integral for comparison. The results
agree with the exact results, as expressed in terms of the confluent hypergeometric
function, but requires no knowledge of that function. It instead requires just a moderate
computing exercise. We feel that working with these numerical representations of
continuum wavefunctions, as derived from the Laplace method, may reduce the cognitive
load associated with continuum eigenstates and confluent hypergeometric functions.
All the other continuum solutions except for the Morse potential, proceed in a
similar fashion and can be evaluated with the circular contour integral. The Morse
potential is different for two reasons: (i) the contour is not given as a rotated dog-bone
contour about the imaginary axis and (ii) the function must vanish as x → −∞. We
treat its continuum solutions next.
The Morse potential differs from the other continuum cases. We see in Table 5 that
the Laplace form of the Schrödinger equation is once again of the form of Eq. (11), so
±λ ∈ R once again, but now β ∈ C. This differs from both the previous problems of the
form in Eq. (11), where β was real, as well as the previous continuum problems, where
β was an integer. As a result, the solution for Φ can still be written in the contour
integral form given in Eq. (17), but the sum of the exponents α+ + α− − 2 = β − 2 is not
an integer. This means the integrand is not single-valued as |z| → ∞, so we must draw
the branch cut as two pieces, each running from infinity to one of the branch points ±λ.
Hence, we draw the branch cut along the real axis in two pieces: one piece travels from
−λ to Re(z) → −∞ and the other piece travels from Re(z) = ∞ to λ. This branch-cut
structure does not allow us to enclose the branch points at ±λ with a contour (as we
The Laplace method for energy eigenvalue problems in quantum mechanics 30
Figure 8. Plot of the continuum Coulomb wavefunctions for l = 0 and three different
values of E: (a) E = 0.1; (b) E = 1, and (c) E = 10 (all energies are in Hartrees).
The power series is shown in red, the contour integral with R = 1.1 in green, and the
one-dimensional integral in black. The three approximations lie on top of each other
until they start to fail—the power series fails around ξ ≈ 20, while the contour integral
fails around ξ ≈ 30. The errors typically occur due to loss of digits of precision in the
expressions being evaluated.
did for previous continuum cases), nor enclose just one of the two branch points (as in
the bound-state problems). However, we can now try another possible contour which
connects the two points ±λ. For concreteness, we choose it to run through the origin,
and we already see a parallel to the other continuum cases, where we took the limiting
behavior of a contour running just next to √the imaginary axis. For the Morse potential,
we have 0 ≤ ξ < ∞. Recall that ξ = 2 2µV a~
0 −ax
e , so as x → −∞ we have ξ → ∞
and as x → ∞ we have that ξ → 0. Since the Morse potential becomes large and
positive for x → −∞, we must have ψ(x) → 0. Hence, φ(ξ) must go to zero as ξ → ∞.
This is a more stringent condition than just having the wavefunction be bounded as we
used previously. This condition eliminates the new contour from −λ → +λ, because an
analysis similar to the previous cases shows that this solution diverges as ξ → ∞.
Now, let us instead consider the contour as shown in Fig. 1b: the contour running
The Laplace method for energy eigenvalue problems in quantum mechanics 31
Immediately, we see that this integral will go to zero as ξ → ∞ because the exponential
term in the integrand guarantees a convergent integral that will vanish in the limit.
Moreover, this integral closely resembles the integral representation of the Tricomi
confluent hypergeometric function, given by (Eq. 13.4.4 in Ref. [11])
Z ∞
1
U (a, b, z) = dt e−zt (t + 1)b−a−1 ta−1 . (92)
Γ(a) 0
This means we can write the integral solution to the Morse differential equation as
One might be concerned that this function is not bounded for ξ → 0. Indeed, a simple
power-counting argument shows that the magnitude of the integrand behaves like z1
when ξ = 0. But, because the exponent is complex, it will produce oscillations, which
can allow the integral to converge and be bounded. To settle this question, we look at
the well-known asymptotics of the Tricomi function U . In the limit where the argument
ξ goes to zero, the behavior of U is governed by the value of the real part of the second
parameter, in our case β. For the Morse potential, Re(β) = 1, so the asymptotic
behavior of U as ξ → 0 is (Eq. 13.2.18 of Ref. [11])
Γ(β − 1) 1−β Γ(1 − β)
U (α− , β, ξ) = ξ + + O(ξ 2−Re(β) ). (94)
Γ(α− ) Γ(1 − α+ )
Once again, Re(β) = 1, so the ξ 1−β term will be ξ −iIm(β) , which has a modulus of
1 (ξ −iIm(β) ξ iIm(β) = 1) and a phase that varies rapidly as ξ → 0. Thus, the Tricomi
function does not diverge, but oscillates, as ξ → 0. We will see below that the behavior
generally looks like a cosine of x for x → ∞.
This contour yields a solution which will be finite everywhere, as well as going to
zero as ξ → ∞. Thus, it satisfies all of our requirements, and we can write the solution
for the unnormalized Morse potential wavefunction in one dimension as
β−1 ξ
ψ(ξ) = ξ 2 e− 2 U (α− , β, ξ), (95)
√
Recall that ξ = 2 2µV
a~
0 −ax
e in this solution. Finally, we note that we can use the
so-called Kummer relation (see Eq. 13.2.42 of Ref. [11])
Γ(1 − b) Γ(b − 1) 1−b
U (a, b, z) = M (a, b, z) + z M (a − b + 1, 2 − b, z), (96)
Γ(a − b + 1) Γ(a)
The Laplace method for energy eigenvalue problems in quantum mechanics 32
Figure 9. Plot of the continuum Morse wavefunction for three different values of E:
2 2
(a) E = 0.1; (b) E = 1, and (c) E = 10. We use ~2µa as the energy unit and a1
2 2
as the length unit. The case we consider is for V0 = ~2µa . Note how the continuum
wave rapidly decays for x < 0, where the Morse potential becomes large and positive.
Because the Morse potential decays to zero exponentially fast, the continuum solution
rapidly looks like a simple cosine wave for large positive x with a fixed amplitude. In
the region around x = 0 we see a transition between the two behaviors.
to relate the above solution in terms of the Tricomi function U to the sum of two complex
conjugate Kummer functions M , which is the form of the Morse continuum wavefunction
that appears in the literature [12, 13]. The divergence in each M as x → −∞ is exactly
cancelled by their sum leading to a finite result. We plot the continuum wavefunctions
for some typical values in Fig. 9. You can see the behavior is as anticipated. We have
a rapid decay for x < 0, there is a transition region near x = 1, and then the form is of
a constant amplitude sinusoidal oscillation as x increases in the positive direction.
We summarize all of our continuum solution results in two tables. Table 6 shows
the results expressed as functions of ξ, while Table 7 shows the results in terms of the
original variables. The normalization of these wavefunctions is subtle and depends on
the scheme that will be used, so we do not discuss the issue of normalization here.
The Laplace method for energy eigenvalue problems in quantum mechanics 33
Confluent Hypergeometric
Problem
Form of Φ(ξ)
2D Free
e−iξ M |m| + 12 , 2|m| + 1, 2iξ
Particle
3D Free
e−iξ M (l + 1, 2l + 2, 2iξ)
Particle
2D −iξ
1
e M |m| + + √i~ , 2|m| + 1, 2iξ
2 a0 2µE
Coulomb
3D −iξ
e M l+1+ √i~ , 2l + 2, 2iξ
Coulomb a0 2µE
Morse
(−1)β−1 Γ(α− )e−ξ/2 U (α− , β, ξ)
Potential
Table 6. Summary of the results of the Laplace method for continuum cases in terms
of the variable ξ. The solution for Φ(ξ), as defined in Table 5, is expressed in terms of
the confluent hypergeometric functions M (a, b, z) and U (a, b, z).
Table 7. Summary of the results of the continuum cases we solved with the Laplace
method in terms of the original independent variable. For the free particle cases, we
express the more common form of the confluent hypergeometric function.
7. Pedagogical discussion
Back in 1937, Dirac urged quantum instruction to include treatments via complex
analysis [3]. This suggestion appears to have been ignored by all (including Dirac
himself) as it never made it into later editions of his quantum mechanics textbook.
This appears to us to have been an unfortunate mistake. As one studies more advanced
quantum field theory for particle physics, or many-body physics for condensed matter,
facility with complex analysis greatly helps in making progress with the material. So,
an earlier quantum mechanics class would be an ideal setting to start developing facility
with complex analysis ideas within quantum instruction.
Furthermore, with the emphasis on the series solution (Fröbenius method) for
solving bound state problems, one does not develop the tools to tackle the continuum
The Laplace method for energy eigenvalue problems in quantum mechanics 34
solutions. Hence, in advanced quantum classes, where these are discussed, they usually
are covered in a somewhat ad hoc fashion that emphasizes analytic continuation of the
bound-state solutions and then the student being simply told what the answer is. In
our opinion, this is not teaching the material. Instead, the approach we present here,
via the Laplace method (building off of Schrödinger’s original solution for hydrogen)
provides a nice approach to cover continuum solutions and introduces a proper way
to teach complex analysis ideas within a graduate quantum mechanics setting. The
integration over the circle of fixed radius provides a nice way to determine these functions
numerically, using a relatively simple code, once one knows how to properly determine
the phases of the powers given the branch cuts used.
It is true that one can instead be told the continuum solution and substitute it into
the Schrödinger equation and, by employing identities of the confluent hypergeometric
functions, verify that it solves the differential equation. But, this is not actually done in
advanced quantum textbooks, nor is it commonly done in quantum instruction, probably
because it is deemed to be too much work to have students find and properly use the
required confluent hypergeometric equation identities (even if they are straightforward
to use). In this circumstance, the approach presented here provides a useful alternative
to other methodologies.
How do we work with these ideas with students? We think that one can present
the solution for some example problems, such as hydrogen, in class and then ask the
students to produce results for other potentials on homework exercises (except the Morse
potential, of course, because that problem requires different methods in the continuum
and is a bit more subtle in its analysis). In addition to developing a capacity for
working with these solutions, we strongly believe students should numerically calculate
these functions using one of the integral forms (we prefer the integration over a circle of
fixed radius for the continuum problems), in order to develop numerical skills within a
quantum context and to learn the challenges with accurately determining these functions
for large arguments.
8. Conclusions
We have shown how the almost forgotten Laplace method can be employed as a
powerful tool within graduate quantum instruction to teach how to determine bound
and continuum states of all problems that are solved with confluent hypergeometric
equation wavefunctions. The approach helps include complex analysis instruction
within a quantum setting, which we believe will better prepare students for more
advanced quantum field theory in a high-energy or condensed-matter setting. The
bound-state solutions provide a way to relate the orthogonal polynomials that arise to
their Rodrigues formulas, which naturally emerge via the poles in the contour integrals
needed to determine the wavefunctions. Quantization of the energies in the bound
states also occurs naturally. For the continuum solutions, we find a simple contour
integral that gives us the wavefunctions, but it requires a precise determination of the
The Laplace method for energy eigenvalue problems in quantum mechanics 35
phases of the terms that are raised to complex-valued powers—hence it requires a proper
understanding of branch cuts and how to evaluate the polar radius and phase of complex
numbers on a cut plane—a quite valuable skill for graduate students to learn.
We believe that using this approach provides students with well-needed practice in
the use of complex analysis methods. It also allows for a solution of the linear potential
problem (which we did not discuss here). That problem is treated with methods similar
to what we developed here in other textbooks, in particular, the textbook by Konishi
and Paffuti [6] has a nice treatment of this. Curiously, Landau and Liftshitz [5] use the
Laplace method to determine the properties of the confluent hypergeometric functions
in their appendix, but they do not use the Laplace method for finding solutions to the
Schrödinger equation in the main part of the text.
The materials we present here are most appropriate for graduate quantum
instruction and, if used, help differentiate the graduate class from the undergraduate
one, rather than just repeating the undergraduate one with harder problems, as is
commonly done.
References
[1] Schrödinger E 1926 Quantisierung als Eigenwertproblem (Erste Mitteilung) Ann. Phys. 384 361–
376
[2] Schlesinger L 1900 Einführung in die Theorie der Differentialgleichungen mit einer unabhängigen
Variablen (Leipzig: G J Göschensche Verlagshandlung)
[3] Dirac P A M 1937 Complex Variables in Quantum Mechanics Proc. R. Soc. London Series A:
Math. Phys. Sci. 160 48–59
[4] Dirac P A M 1947 Principles of Quantum Mechanics 3rd ed (Oxford: Clarendon Press)
[5] Landau L D and Lifshitz E M 1977 Quantum Mechanics: Non-Relativistic Theory 3rd ed (Oxford:
Pergamon Press)
[6] Konishi K and Paffuti G 2009 Quantum Mechanics: A New Introduction (Oxford: Oxford
University Press)
[7] Capri A Z 1985 Nonrelativistic Quantum Mechanics (Menlo Park, CA: Benjamin Cummings)
[8] Galler A, Canfield J and Freericks J K 2021 Schrödinger’s original quantum-mechanical solution
for hydrogen Eur. J. Phys. 42, 035405
[9] Tsaur G-Y and Wang J 2014 A universal Laplace-transform approach to solving Schrödinger
equations for all known solvable models Eur. J. Phys. 35 015006
[10] Szegö G 1939 Orthogonal Polynomials (Providence: American Mathematical Society)
[11] NIST Digital Library of Mathematical Functions http://dlmf.nist.gov/ 2022 Olver F W J Olde
Daalhuis A B Lozier D W Schneider B I Boisvert R F Clark C W Miller B R Saunders B V
Cohl H S and McClain M A eds
[12] Nicholls R W 1981 Continuum Wavefunctions for Morse Molecules Chem. Phys. Lett. 79 317–320
[13] A Matsumoto 1988 Generalized matrix elements in discrete and continuum states for the Morse
potential J. Phys. B: At. Mol. Opt. Phys. 21 2863–2870