Double Well Quantum Mechanics

Steven G. Avery
The double well potential is a rich toy example for understanding numerous quan-
tum mechanical systems. At the time I made these notes, I was actually motivated by
thinking about the decoupling limit and the AdSCFT correspondence. Here, I outline
a method for modeling the double well dynamics that I have not seen elsewhere in the
literature. The standard reference I know (which uses instantons) is in Colemans Aspects
of Symmetry, the chapter titled The double well done doubly well.
We want to think of perturbatively relaxing the limit in which the barrier is innitely
high, and there is no tunneling. To do this we wish to add interaction terms which annih-
late a particle in the left well and create a particle in the right well (and vice-versa).
1 The Nonrelativistic Toy Problem
We want to consider non-interacting non-relativistic particles in the potential
x
1
x
2
x
3
V
U
V (x) =
_

_
0 0 < x x
1
U x
1
< x x
2
V
0
x
2
< x x
3
otherwise
. (1)
For convenience we dene
x
1
= L
1
x
2
x
1
= a x
3
x
2
= L
2
. (2)
1
Double Well Steven G. Avery
Our ultimate goal is to approximate this problem as two isolated square wells of widths
L
1
and L
2
with some small interaction. This description should be valid for energies, E,
such that
0 < V
0
< E U. (3)
For simplicity, we set L
1
= L
2
= L.
The interaction should annihlate a particle in the left well and create a particle in the
right well (and vice-versa). Thus, this problem naturally requires the use of a multi-
particle formulation of nonrelativistic quantum mechanics.
We rst review how to second quantize the schrodinger equation to nd such a
description. Then, we argue that one can drop the time-derivative kinetic term of the
action in the barrier region and integrate out those degrees of freedom. This removes the
barrier region of the action, but leaves behind boundary terms.
The boundary terms dene nonvanishing boundary conditions for the interior edges
of the two decouopled wells, and provide a coupling term. We treat the coupling term as
a perturbation. To zeroeth order, then, one can neglect it and canonically quantize the two
wells separately. Finally, one sees with what strength the coupling term couples modes
in the left mode to modes in the right well.
2 Second Quantized Schrodinger Equation
Note that the Lagrangian
L =
i
2
_

1
2m

V (x)

(4)
has the Schrodinger Equation,
i

=
1
2m

2
+V (x), (5)
and its Hermitian conjugate as its equation of motion.
We wish to canonically quantize this theory to describe multiparticle nonrelativistic
quantum mechanics. We begin by nding the canonical momenta
=
L

=
i
2

=
L

=
i
2
, (6)
which give us two primary constraints

1
=
i
2

2
=

+
i
2
. (7)
The naive Hamiltonian has no kinetic term,
H
N
=

+

L
=
1
2m

+V (x)

, (8)
double-well.tex 2 January 12, 2010, 7:15pm
Double Well Steven G. Avery
a key hint that we need to use Diracs generalization of Hamiltonian mechanics.
The two primary constraints are second-class and generate no secondary constraints.
The total Hamiltonian is given by
H = H
N
+u
1

1
+u
2

2
. (9)
Demanding that

1
=
1
, H
P.B.
= 0 (weakly), (10)
xes one of the arbitrary functions
u
2
= iV (x)

, (11)
and similarly one nds from the other consistency condition that
u
1
= iV (x). (12)
Since we are interested in quantizing the system, we nd the canonical Dirac brackets
and then we can use the naive Hamiltonian. The Dirac brackets are dened by
f, g
D.B.
= f, g
P.B.
f,
i

P.B.
_
M
1
_
ij

j
, g
P.B.
, (13)
where
M
ij
=
i
,
j

P.B.
= (
2
)
ij
= i
ij
=
_
M
1
_
ij
. (14)
One nds that the nonvanishing canonical Dirac brackets are
(x),

(y)
D.B.
= i(x y)
(x), (y)
D.B.
=
1
2
(x y)

(x),

(y)
D.B.
=
1
2
(x y).
(15)
Upon canonically quantizing, then, one can describe the system as

H =
1
2m

+V (x)

(x),

(y)] = (x y). (16)

n
+V (x)f
n
= E
n
f
n
. (17)
We can expand

in these modes as

(x) =

n
a
n
f
n
(x) [ a
m
, a

n
] =
mn
. (18)
double-well.tex 3 January 12, 2010, 7:15pm
Double Well Steven G. Avery
The Hamiltonian becomes

H =

m,n
a

m
a
n
_
dx
_

1
2m
f

n
+V (x)f
n
_
f

m
=

m,n
E
n
a

m
a
n
_
dx f
n
f

m
=

n
E
n
a

n
a
n
. (19)
3 Integrating out Barrier Degrees of Freedom
We can formally break the action into three parts
S
tot.
= S
1
+S
barrier
+S
2
=
_
dt
_
L
0
dxL +
_
dt
_
L+a
L
dxL +
_
dt
_
2L+a
L+a
dxL (20)
3.1 Integrating out the Barrier
We now want to use our assumption that the energies under consideration are much
smaller than the potential energy in the barrier region
E
typ.
U. (21)
Consider that
e
iEt
,
and therefore
L
B
=
i
2
_

1
2m

V (x)

= (E U)

1
2m

1
2m

V (x)

. (22)
In this approximation, we have replaced E U U, and thereby dropped the kinetic
term. By dropping the kinetic term, we have made the barrier degrees of freedom non-
dynamical. We can integrate them out as auxillary variables by solving the equation of
motion and then plugging it back into the action.
For this 1 +1-dimensional example, the equation for the barrier degrees of freedom is

(x) = 2mU(x) L < x < L +a, (23)

2mU(x L)
_
+Bsinh
_

2mU(x L)
_
. (24)
double-well.tex 4 January 12, 2010, 7:15pm
Double Well Steven G. Avery
We can eliminate A and B in terms of the boundary degrees of freedom by using
A = (L)
B = (L +a) csch(

2mUa) (L) coth(

2mUa).
(25)
We then plug this into the barrier Lagrangian,
L
B
=
_
L+a
L
L
B
[], (26)
to nd
L
B
=
_
U
2m
coth(

2mUa)
_

(L)(L) +

(L +a)(L +a)
_
+
_
U
2m
1
sinh(

2mUa)
_

(L)(L +a) +

(L +a)(L)
_
.
(27)
The rst term provides the boundary conditions of the solution in the two regions inde-
pendently, whereas the second term provides the coupling between the two wells.
3.2 Solving the two regions with boundary terms
We wish to work in a the limit where the second term is a small effect and can be treated
perturbatively. In which case, the unperturbed action for the left region is given by
S
1
=
_
dt
_
L
0
dx
_
i
2
(

)
1
2m

_
U
2m
coth(

2mUa)
_
dt

(L)(L).
(28)
To determine the equations of motion we should demand that the linear variation of the
action under arbitrary variations of and

vanishes. The variations and

must be
subject to appropriate boundary conditions. In this case, we should leave the right edge
unconstrained while the left edge should vanish.
Therefore, we must be careful to not drop all of the boundary terms when integrating
by parts. One nds
S =
_
dt
_
L
0
dx
_
i

+
1
2m

_
dt
_
1
2m

(L) +
_
U
2m
coth(

2mUa)(L)
_

(L),
(29)
which implies that

1
2m

(x, t) = i

(x, t)

(L, t) =

2mU coth(

2mUa)(L, t) (0, t) = 0.
(30)
The boundary term gave us a nonvanishing boundary condition. For convenience we
dene
=

2mU coth(

2mUa). (31)
double-well.tex 5 January 12, 2010, 7:15pm
Double Well Steven G. Avery
The eigenvalue problem, then, is

1
2m
g

E
(x) = Eg
E
(x) g(0) = 0 g

(L) = g(L). (32)

2mEx), (33)
which satises the DE and the rst boundary condition. The second boundary condition
quantizes the energy:

2mE cos(

2mEL) = sin(

2mEL) =tan(

2mEL) =

2mE

(34)
There is a set of solutions,
n
to this transcendental equation. The easiest way to visu-
alize them is graphically, as shown in the gure.
2 4 6 8 10
6
4
2
2
4
In region 2, we must solve

1
2m
h

E
(x) +V
0
h
E
(x) = Eh
E
(x) h
E
(2L +a) = 0 h

E
(L +a) = h
E
(L +a). (35)
We take the ansatz
h(x) = sin[
_
2m(E V
0
)(x 2L a)] (36)
which satises the DE and the rst boundary condition. The other boundary condition
gives the transcendental equation
tan[
_
2m(E V
0
)L] =
_
2m(E V
0
)

, (37)
which has a set of solutions
n
, which are related to the
n
via

n
=
n
+V
0
. (38)
Let us, for simplicity, dene

1
(x) = (x) x [0, L]

2
(x) = (x +L +a) x [0, L],
(39)
in which case we can write the general solutions as

1
(x, t) =

n
A
n
sin(

2m
n
x)e
int

2
(x, t) =

n
B
n
sin(

2m
n
(x L))e
int
(40)
double-well.tex 6 January 12, 2010, 7:15pm
Double Well Steven G. Avery
3.2.1 Properties of the Eigenfunctions
One may worry that the peculiar boundary conditions may not be consistent with Sturm
Liouville theory and therefore we may lose reality of eigenvalues and orthogonality of
eigenfunctions with distinct eigenvalues. In fact one can demonstrate that one does have
these properties for real .
We consider the eigenfunctions of region 1, but everything works in the same way for
region 2. Consider
_
L
0
dx g

m
(x)

g
n
(x) = 2mE
m

_
L
0
dx g

m
g
n
= g

m
(x)

g
n
(x)

L
0

_
L
0
dx g

m
(x)

n
(x)
= g

m
(L)

g
n
(L) g
m
(L)

n
(L) +
_
L
0
dx g
m
(x)

n
(x)
=

g
m
(L)

g
n
(L) +g
m
(L)

g
n
(L) 2mE
n
_
L
0
dx g

m
g
n
= (

)g

m
(L)g
n
(L) 2mE
n
_
L
0
dx g

m
g
n
. (41)
In the above equation, the rst term vanishes since =

. If one considers the case

m
_
L
0
dx [g
m
[
2
= 2mE
m
_
L
0
dx [g
m
[
2
= (E

m
E
m
)
_
dx [g
m
[
2
= 0, (42)
and therefore the eigenvalues E
m
must be real.
If one considers the master equation above for different n and m, and makes use of
the reality of the eigenvalues, then
(E
m
E
n
)
_
L
0
dx g

m
g
n
= 0, (43)
and therefore eigenfunctions of distinct eigenvalues are orthogonal.
We will assume from now on that the eigenfunctions have been properly normalized.
3.3 Canonically Quantizing
We canonically quantize by recognizing that
H
1
=
_
L
0
dx
_
1
2m

1
_
+
_
U
2m
coth(

2mUa)

1
(L)
1
(L)
H
2
=
_
L
0
dx
_
1
2m

2
+V
0

2
_
+
_
U
2m
coth(

2mUa)

2
(0)
2
(0)
(44)
double-well.tex 7 January 12, 2010, 7:15pm
Double Well Steven G. Avery
and following the usual procedure. We expand
1
and
2
in modes as (Schrodinger
Picture)

1
(x) =

n
a
n
g
n
(x)

2
(x) =

b
n
h
n
(x), (45)
where assuming we have properly normalized g
n
and h
n
.
When one plugs in, one needs to be careful of the boundary term:

H
1
=
1
2m

1
(L)
1
2m
_
L
0
dx

1
+
_
U
2m
coth(

2mUa)

1
(L)
1
(L)
=

2mU
2m
coth(

2mUa)

1
(L) +
_
U
2m
coth(

2mUa)

1
(L)
1
(L)

1
2m

m,n
a

m
a
n
_
L
0
dx g

m
(x)g

n
(x)
=

n
a

n
a
n
. (46)
A similar relation holds for the second region, leaving one with

H
1
=

n
a

n
a
n

H
2
=

b
n
. (47)
We now add the perturbation

H
int.
= S
int.
=
_
U
2m
1
sinh(

2mUa)
_

1
(L)

2
(0) +

2
(0)

1
(L)
_
. (48)
The rst term is of the form

1
(L)
2
(0) =

n
a

b
n
g

m
(L)h
n
(0)

1
L

m,n
a

b
n
sin(

2m
m
L) sin(

2m
n
L), (49)
and we see how the modes in one well couple to those of the other well. Note that the
last line is not equality because I have not carefully normalized the modes.
3.4 The Probability to tunnel
The interaction is

H
int
=
_
U
2m
1
sinh(

2mUa)

1
(L)

2
(0) + h.c.
=
_
U
2m
1
sinh(

2mUa)

m,n
a
m

n
g
m
(L)h

n
(0) + h.c. (50)
double-well.tex 8 January 12, 2010, 7:15pm
Double Well Steven G. Avery
Let us consider an inital and nal state of the form
[i) = [m)
1
[f) = [n)
2
; (51)
that is, the particle tunnels from eigenstates m in the left well to eigenstate n in the right
well. The matrix elements of the interaction Hamiltonian look like
i[

H
int
[i) = 0
f[

H
int
[f) = 0
f[

H
int
[i) =
_
U
2m
1
sinh(

2mUa)
g
m
(L)h

n
(0) = .
(52)
Let us assume that tunneling between these states dominates over tunneling between
other energy levels; that is, let us assume that we can treat this as a two-state system. For
this purpose, let
=
m
=
n
(53)
In the free energy eigenstate basis, the two-dimensional Hamiltonian becomes

H
_

_
=
+
2
1

2

3
+
1
=
+
2
1 +v , (54)
where
v =
_
, 0,

2
_
. (55)
The time-evolution operator, then, is given by

U(t) = e
i

Ht
e
i
+
2
t
e
ivt
= e
i
+
2
t
_
cos([v[t) +i( n) sin([v[t)
_
. (56)
Thus, the amplitude to nd the particle in the right well after time T, given that it
starts in the left well is
f[

U(t) [i) = ie
i
2
(+)t
sin([v[t)
_
0 1
_
n
_
1
0
_
= ie
i
2
(+)t
sin([v[t)
[v[
_
0 1
_
_

2
__
1
0
_
= ie
i
2
(+)t
sin([v[t)

[v[
= ie
i
2
(+)t
sin([v[t)

_

2
+
()
2
4
double-well.tex 9 January 12, 2010, 7:15pm
Double Well Steven G. Avery
= ie
i
2
(+)t
sin([v[t)
1
_
1 +
()
2
4
. (57)
The probability, then, is simply
P(t) =
sin
2
[v[t
1 +
()
2
4
. (58)
The tunneling time is given by
T
tunnel
=

2[v[
=

_
4
2
+ ( )
2
, (59)
and the probability to tunnel is
P
tunnel
=
1
1 +
()
2
4
. (60)
When the energy levels are equal there is a resonance and the probability converges to
unity. Of course, this is only an approximate result for our system, since we have ne-
glected tunneling into other energy levels.
Let us consider the case where = . Then the rate (one over the tunneling time) is
given by

tunnel
=
1
T
tunnel
=
2

[[ =
2

_
U
2m
1
sinh(

2mUa)
[g
m
(L)[[h

n
(0)[. (61)
For the above problem with V = 0, this becomes

tunnel
=
2

_
U
2m
[g
n
(L)[
2
sinh(

2mUa)
, (62)
where
[g
n
(L)[
2
=
sin
2
(

2m
n
L)
_
L
0
sin
2
(

2m
n
x)dx
. (63)
The integral, we can do:
_
L
0
sin
2
(

2m
n
x)dx =
L
2

1
2

2m
n
sin(

2m
n
L) cos(

2m
n
L)
=
L
2

1
2

2m
n
tan(

2m
n
L)
tan
2
(

2m
n
L) + 1
=
L
2
+
1
2

2m
n

2m
n
L
2m
n
L
2
+
2
=
L
2
_
1 +

2m
n
L
2
+
2
_
double-well.tex 10 January 12, 2010, 7:15pm
Double Well Steven G. Avery
=
L
2
_
1 +

2mU coth(

2mUa)
2m
n
L
2
+ 2mU coth
2
(

2mUa)
_

L
2
_
1 +
1

2mU
_

L
2
(64)
Furthermore, we can approximate
sin
2
(

2m
n
L) =
tan
2
()
tan
2
() + 1
=
2m
n
2m
n
+
2

n
+U

n
U

n
2

2
2mL
2
1
U
, (65)
which gives
[g
m
(L)[
2

n
2

2
mL
3
U
. (66)
Finally, to leading order, we have

tunnel
n

4n
2

mL
3
e

2mUa

2mU

2
(67)
where has been reintroduced.
4 Comparison with Other Methods
Now consider solving the quantum mechanics problem in the double well potential. Re-
ecting one well into the other commutes with the Hamiltonian, and so the energy eigen-
states can have denite parity. Consider the symmetric and antisymmetric states which
in each well look like the nth energy level. We call these states
[S)
n
[A)
n
. (68)
We can dene
[L)
n
=
1

2
([S)
n
+[A)
n
) [R)
n
=
1

2
([S)
n
[A)
n
). (69)
The [S) states will have slightly different energy levels from the [A) state, and thus they
evolve at slightly different rates:

U(t) [L) =
1

2
_
e
iE
S
t
[S) +e
iE
A
t
[A)
_
double-well.tex 11 January 12, 2010, 7:15pm
Double Well Steven G. Avery
=
e
iEst

2
_
[S) +e
i(E
S
E
A
)t
[A)
_
, (70)
Thus, we see that
T
tunneling
=

E
. (71)
We need some way of estimating the difference in energy between the symmetric and
the anti-symmetric state.
4.1 Estimating the Energy Levels
To begin consider the position-space energy eigenvalue equation:

1
2m

(x) +V (x)(x) = E(x). (72)

_
Asin(

2mEx) x (0, L)
Bcosh
_

2mU
_
x (L +
a
2
)
_
_
+C sinh
_

2mU
_
x (L +
a
2
)
_
_
x (L, L +a)
Dsin
_

2mE
_
x (2L +a)
_
_
x (L +a, 2L +a).
(73)
Demanding that C
1
(0, 2L + a) imposes the following conditions (after assuming
E U)
Asin(

2mEL) = Bcosh(

2mU
a
2
) C sinh(

2mU
a
2
)
A
_
E
U
cos(

2mEL) = Bsinh(

2mU
a
2
) +C cosh(

2mU
a
2
)
Dsin(

2mEL) = Bcosh(

2mU
a
2
) +C sinh(

2mU
a
2
)
D
_
E
U
cos(

2mEL) = Bsinh(

2mU
a
2
) +C cosh(

2mU
a
2
).
(74)
The symmetric solution must have C = 0, which leads to the energy eigenvalue equa-
tion
tan(
_
2mE
S
L) =
_
E
S
U
coth(

2mU
a
2
). (75)
Similarly, the antisymmetric solution must have B = 0, which leads to the energy eigen-
value equation
tan(
_
2mE
A
L) =
_
E
A
U
tanh(

2mU
a
2
). (76)
We can use the approximations
tanh x 1 2e
2x
coth x 1 + 2e
2x
x 1. (77)
Let
= 2e

2mUa
, (78)
double-well.tex 12 January 12, 2010, 7:15pm
Double Well Steven G. Avery
and then the two eigenvalue equations are
tan(

2mEL) =
_
E
U
(1 ) . (79)
We solve the problem, perturbatively, to rst order in . The zeroeth order result is
degenerate. We write
E
n
= E
0
n
+E
1
n
. (80)
Suppose
tan(
_
2mE
0
n
L) =
_
E
0
n
U
. (81)
Expanding the left-hand side to rst order in , one nds
tan(
_
2mE
n
L) = tan(
_
2m(E
0
n
+E
1
n
)L)
= tan
_
_
2mE
0
n
(1 +
E
1
n
2E
0
n
)L
_
= tan
_
_
2mE
0
n
L +E
1
n
_
m
2E
0
n
L
_
=
tan(
_
2mE
0
n
L) + tan(E
1
n
_
m
2E
0
n
L)
1 tan(
_
2mE
0
n
L) tan(
_
m
2E
0
n
E
1
n
L)
=
tan(
_
2mE
0
n
L) +E
1
n
_
m
2E
0
n
L
1 tan(
_
2mE
0
n
L)
_
m
2E
0
n
E
1
n
L
=
_
tan(
_
2mE
0
n
L) +E
1
n
_
m
2E
0
n
L
_ _
1 + tan(
_
2mE
0
n
L)
_
m
2E
0
n
E
1
n
L
_
= tan(
_
2mE
0
n
L) +E
1
n
_
m
2E
0
n
L
_
1 + tan
2
(
_
2mE
0
n
L)
_
. (82)
Expanding the right-hand side:

EU(1 ) =
_
E
0
n
+E
1
n
U
(1 )
=
_
E
0
n
U
_
1 +
E
1
n
2E
0
n
_
(1 )
=
_
E
0
n
U
_
1 +
E
1
n
2E
0
n

_
. (83)
Setting the two sides equal to each other one nds
(
(
(
(
(
(
(
(
tan(
_
2mE
0
n
L) +E
1
n
_
m
2E
0
n
L
_
1 + tan
2
(
_
2mE
0
n
L)
_
=
_
E
0
n
U
_

1 +
E
1
n
2E
0
n

_
double-well.tex 13 January 12, 2010, 7:15pm
Double Well Steven G. Avery
E
1
n
_
m
2E
0
n
L
_
1 +
E
0
n
U
_
=
_
E
0
n
U
_
E
1
n
2E
0
n
1
_
. (84)
Using the approximation that E
0
n
U, one gets
E
1
n
_
m
2E
0
n
L =
_
E
0
n
U
, (85)
and therefore the rst-order shifts in the energy-levels are given by
E
1
n
=
E
0
n
L
_
2
mU

=
2E
0
n
L
_
2
mU
e

2mUa

n
2

2
mL
3
_
2
mU
e

2mUa
(86)
This suggests that the energy difference between the symmetric and anti-symmetric states
is given by
E =
2n
2

2
mL
3
_
2
mU
e

2mUa
, (87)
and
=
2n
2

mL
3
_
2
mU
e

2mUa
. (88)
This exactly matches the estimate from the previous method.
double-well.tex 14 January 12, 2010, 7:15pm