Physics 342 Lecture 1

Separation of Variables
Lecture 1

Physics 342
Quantum Mechanics I

Monday, January 25th, 2010

There are three basic mathematical tools we need, and then we can begin
working on the physical implications of Schrödinger’s equation, which will
take up the rest of the semester. So we start with a review of: Linear
Algebra, Separation of Variables (SOV), and Probability. There will be
no particular completeness to our discussions – at this stage, I want to
emphasize those aspects of each of these that will prove useful. Pitfalls and
caveats will be addressed as we encounter them in actual problems (where
they won’t show up, so won’t be addressed).
Try as one might, it is difficult to relate these three areas, so we will take
each one in turn, starting from the most familiar, and working to the least
(that’s a matter of taste, so apologies in advance for the wrong ordering – in
anticipation of our section on probability: How many different orderings of
these three subjects are there? On average, then, how many people disagree
with the current one?).
We will begin with separation of variables, a simple technique that you have
used recently – we will review the basic electrostatic interest in these solu-
tions, and do a few examples of SOV applied to PDE’s other than Laplace’s
equation (∇2 V = 0).

1.1 Separation for the Laplace Problem

Separation of variables refers to a class of techniques for probing solutions to

partial differential equations (PDEs) by turning them into ordinary differen-
tial equations (ODEs). We generally rely on some notion of uniqueness1 for
1
Uniqueness implying both a solution to the PDE and satisfaction of some boundary
conditions.

1 of 11
1.1. SEPARATION FOR THE LAPLACE PROBLEM Lecture 1

our PDE – then the logic of SOV rests on the notion that finding a solution,
even by this somewhat limited technique, suffices.
There are two basic flavors of SOV, additive and multiplicative – in either
case, the style of argument is the same, so we’ll start with the algebraically
simpler version.

1.1.1 Additive Separation

We take the Laplace problem: Find V (x, y) satisfying ∇2 V = 0 on some

region, with some specified boundary conditions (for example, in E&M, we
would have an implicit V −→ 0 at spatial infinity condition, and maybe some
conducting boundaries where we set V = 0). Working in two dimensions
(for simplicity), we start with a separation ansatz – additive, in this case:

V (x, y) = Vx (x) + Vy (y), (1.1)

then running this through the Laplacian gives:

d2 Vx d2 Vy
+ = 0. (1.2)
dx2 dy 2

Now for the standard argument: Vx00 (x) depends only on x just as Vy00 (y)
depends only on y – they cannot be equal unless each is equal to a constant.
Suppose we set Vx00 = α, then we must have Vy00 = −α. The solution to these
ODEs is:
1 1
Vx (x) = α x2 + β x + δ Vy (y) = − α y 2 + γ y + ρ. (1.3)
2 2
for arbitrary constants (β, δ, γ, ρ), to be used in imposing the boundary
condition(s). The solution, for any α, can be written as:
1
V (x, y) = α (x2 − y 2 ) + β x + γ y + κ (1.4)
2

As far as electrostatics goes, this solution does not match our usual notion of
potential – it doesn’t die at spatial infinity. That’s not always a deal-breaker,
but it is true that additive separation is of limited utility in E&M (it plays a
much larger role in, for example, Hamilton-Jacobi theory). Nevertheless,
the pattern of: ansatz, argument, solution is the same as for the more
immediately useful, and algebraically involved multiplicative separation.

2 of 11
1.1. SEPARATION FOR THE LAPLACE PROBLEM Lecture 1

1.1.2 Multiplicative Separation

The difference here is in the starting ansatz – rather than additive single
variable functions, we take the multiplicative combination:

V (x, y) = Vx (x) Vy (y), (1.5)

then Laplace’s equation becomes

d2 Vx d2 Vy
Vy + V x = 0, (1.6)
dx2 dy 2

and if we divide by V (x, y) itself, we can write this in a form conducive to

making the separation argument:

Vx00 Vy00
+ = 0. (1.7)
Vx Vy

The first term depends only on x, the second only on y, so for a solution
satisfying our functional assumptions, we must have both terms equal to a
constant, call it α2 (a convenient choice, as you will see in a moment) – for
α complex. Then
Vx00 Vy00
=α =−
2
(1.8)
Vx Vy
will certainly have ∇2 V = 0. The solutions here are familiar:

Vx = A eα x + B e−α x Vy = F ei α y + G e−i α y . (1.9)

Now we see that for a generic complex number, both Vx and Vy will be
combinations of growing and decaying exponentials, as well as oscillatory
sines and cosines. For example, if α is real, Vx is growing and decaying,
while Vy is pure sine and cosine. For α = i a with a real, the roles are
reversed. In general, we use the boundary conditions to choose a convenient
expression for α.

Example

Consider a pair of infinite, grounded conducting sheets separated a distance

d with a conductor connecting the two sheets held at V0 as shown in Fig-
ure 1.1. What is the potential in between the plates? We know that the

3 of 11
1.1. SEPARATION FOR THE LAPLACE PROBLEM Lecture 1

potential satisfies Laplace’s equation in the region between the plates (no
charge in there), and the boundary conditions are clear:
 
2πx
V (x = 0, y) = V (x = d, y) = 0 V (x, y = 0) = V0 sin . (1.10)
d

ŷ

x̂
d

! "
2πx
V0 sin
d

Figure 1.1: Two parallel plates separated by a distance d and held at V = 0,

the connecting plate is held at V0 sin(2 π x/d).

We know the solution will be of the form (1.9), but how to choose α and the
constants (A, B, F, G)? There is an additional, implicit boundary condition
– we’d like the potential to go to zero in the “open” spatial direction, y −→
∞ – this tells us that we should set α = i a for a ∈ IR to get growing and
decaying exponentials for the Vy (y) solution:

Vx = A ei a x + B e−i a x Vy = F e−a y + G ea y (1.11)

and, furthermore, we should set G = 0 to eliminate the growing exponential.

By introducing à = (A + B) and B̃ = i (A − B), we can write Vx (x) in terms
of sine and cosine. Our solution so far, then, is:

Vx = Ã cos(a x) + B̃ sin(a x) Vy = F e−a y V = Vx Vy . (1.12)

Now for the rest of the boundary conditions: take Vx (x = 0) = 0 – that tells
us that à = 0. Our solution is now much simpler, the full potential has the

4 of 11
1.1. SEPARATION FOR THE LAPLACE PROBLEM Lecture 1

form
V = Q sin(a x) e−a y (1.13)
where we have introduced yet another rewrite of combinations of constants
(Q = Ã F ). The boundary condition V (x = d, y) = 0 is interesting, this
tells us that sin(a d) = 0, not a statement about the obvious constant Q,
but rather, a constraint on the allowed wavelengths:

sin(a d) = 0 −→ a d = n π (1.14)

for integer n. Then the only a’s that satisfy the boundary condition are
a = ndπ , still an infinite family, but labelled by the integer index n. So, we
have
Vn (x, y) = Qn sin(n π x/d) e−n π y/d , (1.15)
and we put the subscript n here to remind us that any integer n provides
a solution. This type of boundary condition, one that limits the allowed
“wavelength”, is what leads to the mathematical representation of quanti-
zation when applied to the separable solutions of Schrödinger’s equation.
We have one boundary left – because Laplace’s equation is a linear PDE,
sums of solutions are still solutions, and we can make a general solution
out of our n-indexed ones. Each term satisfies Laplace’s equation and the
boundary conditions at x = 0, d and y −→ ∞, so the sum does as well:
X
∞ n π x nπy
V (x, y) = Qn sin e− d . (1.16)
d
n=1

As for the final boundary condition: V (x, y = 0) = V0 sin(2 π x/d), we can

take Qn = 0 for all n except for n = 2. We know that the solution is unique,
so this is certainly valid. But it begs the question, what about a less well-
adapted driving potential? That is a question for next time, as it is tied to
the decomposition of functions in sin and cosine “bases”, a linear algebra
issue.
The final answer:
 
2πx 2πy
V (x, y) = V0 sin e− d . (1.17)
d
This solution is just the potential at the y = 0 plate, decaying into the
region between the plates – equal contours are shown in Figure 1.2.

5 of 11
1.2. SEPARATION FOR OTHER PDES Lecture 1

0.5

0.4

0.3

0.2

0.1

0.0
0.0 0.2 0.4 0.6 0.8 1.0

! "
2πx
V0 sin
d

Figure 1.2: The potential between the plates for the setup shown in Fig-
ure 1.1.

There is not much more to say about separation of variables – the best
approach is to do as many examples as you can get your hands on – that
makes some of the choices that appear unmotivated at first (why should we
make the integration constant α2 ? Why do we pick the y solution to be
growing and decaying?) more reasonable.

1.2 Separation for Other PDEs

We will be applying the SOV technique to Schrödinger’s equation, and that

is not going to be much different than the Laplacian example from E&M.
The approach is independent of PDE, although the PDE must be linear (not
strictly speaking true, but the superposition principle we invoked in writ-
ing (1.16) relied on linearity) – just to show some of the places where SOV
comes up, we can look at the Helmholtz equation and the wave equation.

6 of 11
1.2. SEPARATION FOR OTHER PDES Lecture 1

1.2.1 Helmholtz Equation

Suppose we consider the same problem as above, but for the Helmholtz
PDE:
(∇2 + µ2 ) V = 0
 
2πx
V (x, y = 0, d) = 0 V (x, y = 0) = V0 sin V (x, y → ∞) −→ 0,
d
(1.18)
we are imagining a slight modification to electrostatics (actually amounts to
a theory of electrostatics with a massive photon, but forget about that for
now).
First we’ll find the ODE solutions to the Helmholtz PDE via V (x, y) =
Vx (x) Vy (y):
Vx00 Vy00
+ + µ2 = 0. (1.19)
Vx Vy
We see immediately that we will again get oscillatory and exponential solu-
tions, since we want the solution in the increasing y direction to decay, we’ll
V 00
call Vyy = α2 , with α real. With the constant µ2 in place, we don’t need to
set the x term equal to −α2 , we have additional freedom. We know that the
00
boundary conditions for x are easily imposed for sines, so set VVxx = −β 2 for
β real. Then the constraint we must impose to solve the Helmholtz PDE
is: −β 2 + α2 + µ2 = 0. Imposing the boundary conditions at x = 0, d gives
back β = ndπ as before, so that
n π x
Vx = A sin . (1.20)
d
Now, for the Vy equation, we are given µ, and we have β 2 = n2 π 2 /d2 , so
the growing and decaying exponentials have the form:
√ 2 2 2 2 √ 2 2 2 2
Vy = F e n π /d −µ y + G e− n π /d −µ y . (1.21)
There are interesting elements to this solution, most notably, a cutoff in n
for which Vy is actually oscillatory (n < µπd ) – ignoring that for the moment,
if we assume the above terms for Vy are indeed growing and decaying for
n = 2, we have the solution:
  √
2πx 2 2 2
V (x, y) = V0 sin e− 4 π /d −µ y , (1.22)
d
which just has a different fundamental decay length when compared to the
Laplace form for electrostatics.

7 of 11
1.2. SEPARATION FOR OTHER PDES Lecture 1

1.2.2 The Wave Equation

Separation of variables is not limited to purely spatial problems – we can

mix in time. For the wave equation in one spatial dimension (call it x),
defined in a medium with fundamental speed v, we have
1 ∂ 2 u(x, t) ∂ 2 u(x, t)
− + = 0. (1.23)
v2 ∂t2 ∂x2
Take a multiplicative ansatz: u(x, t) = ux (x) ut (t), then by the now familiar
procedure of inputting and dividing by u:
u00t u00x
− + =0 (1.24)
v 2 ut ux
and again, setting both terms equal to a constant:
u00t u00x
− = α2 = −α2 . (1.25)
v 2 ut ux
What is interesting here is that both the spatial solution ux and the temporal
ut can be oscillatory – that is not true for the Laplace equation separable
solutions we saw above. Without worrying about boundary conditions, we
have the easy solution:
ut = A cos(α v t) ux = B cos(α x) (1.26)
so we can write u(x, t) = u0 cos(α x) cos(α v t) – of course, just as obvious is
the solution û(x, t) = u0 sin(α x) sin(α v t). Since the wave equation is lin-
ear, the sum of these two solutions is also a solution, and it is an instructive
one:
u(x, t) + û(x, t) = u0 (cos(α x) cos(α v t) + u0 sin(α x) sin(α v t))
(1.27)
= u0 cos(α (x − v t)),
precisely a plane wave. Of course, in general, we would be given spatial
boundary conditions and/or an initial waveform at t = 0.

1.2.3 Poisson’s Equation

The utility of separation is not limited to source-free equations – consider

Poisson’s equation for electrostatic potential in the presence of source charge
density ρ:
ρ
∇2 V = − . (1.28)
0

8 of 11
1.2. SEPARATION FOR OTHER PDES Lecture 1

We must be given a ρ that is itself appropriately separable. The simplest

possible such distribution would be a constant, but we can imagine more
interesting charge distributions. As an example, suppose we are in spherical
coordinates, and we have a spherically symmetric charge density: ρ(r, θ, φ) =
ρ(r), depending only on our distance from the origin. Poisson’s equation
becomes, in these variables
     
1 ∂ 2 ∂V 1 ∂ ∂V 1 ∂2V ρ(r)
r + 2 sin θ + 2 V =− .
r ∂r
2 ∂r r sin θ ∂θ ∂θ r sin θ ∂φ
2 2 0
(1.29)
The separation ansatz now becomes V (r, θ, φ) = Vr (r) Vθ (θ) Vφ (φ), and we
can see what will happen – to satisfy the functional dependence assumption,
we will have constant Vθ and Vφ , leaving us with just the radial ODE:

1 d 2 0
ρ(r)
r Vr = − . (1.30)
r2 dr 0
Suppose we take ρ(r) = ρ0 a constant, then the solution to the above is:

β ρ0 r2
Vr (r) = α + − . (1.31)
r 6 0

Notice that the first two terms are just solutions to ∇2 V = 0 (an overall
constant and the potential outside a spherically symmetric distribution of
charge) – if we are inside a uniformly charged sphere, with r = 0 included
in the domain, then we must set β = 0. The constant α, of course, can be
set to zero, and we are left with the usual potential for a uniformly charged
sphere.

9 of 11
1.2. SEPARATION FOR OTHER PDES Lecture 1

Homework

Reading: Griffiths “Introduction to Electrodynamics”, pp. 127–136.

Problem 1.1

Solve Laplace’s equation ∇2 V = 0 using separation: V (x, y) = Vx (x) Vy (y)

for the following two-dimensional boundary conditions:

V (0, y) =
V (d, y) = 0, V (x, d) = 0 and V (x, 0) = V0 sin d .
2πx

(V = 0)
d

(V = 0) (V = 0)

d x

! "
2πx
V0 sin
d

Physically, we are describing the electrostatic potential inside a square that

has grounded lines on three sides, and a spatially-varying potential on the
fourth side.

Problem 1.2

The heat equation in one (spatial) dimension reads

1 ∂u(x, t) ∂ 2 u(x, t)
= , (1.32)
k ∂t ∂x2
where u(x, t) is the temperature of an object with “thermal conductivity” k
(a material property). For a complete solution, we must specify two bound-
ary conditions (since this equation is second order in x) and an “initial”
condition (the value of u(x, t = 0), for example).

10 of 11
1.2. SEPARATION FOR OTHER PDES Lecture 1

a. Using separation of variables: u(x, t) = ux (x) ut (t), and separation

constant −α2 , write down the separated heat equation and solve for ux and
ut .

b. Suppose we have a material that goes from x = 0 to x = d – we

set the temperature on both ends of the material to zero – i.e. u(0, t) =
u(d, t) =
0. Initially, at time t = 0, we have temperature: u(x, 0) =
u0 sin πdx . Using your separation solution, find the particular solution for
this set of boundary and initial conditions.

u(x, 0)
!π x"
u0 sin
d
u0

d x

11 of 11