Lecture 17: The adiabatic approximation:

Berrys phase & Aharonov-Bohm effect (11/17/2005)

In classical physics, the axis of the motion of a pendulum rotates a bit if we walk around America
with the pendulum. In 1984, M.V. Berry noticed a similar effect a geometric phase in quantum
mechanics that no one had observed for 59 years.
Additional reading if you wish: Griffiths 10.2, Sakurai sup. 1
Earlier, Aharonov and Bohm explained that particles are able to detect a magnetic field even in
regions away from their trajectories. We will look at insights of Berry and of Aharonov and Bohm
and their relationships.

Classical analogue: pendulum

Start at the North pole with a pendulum in your hand. Walk towards the equator, along a great
circle. Then turn left, walk around the equator by the angle . Then turn left again and return to
the North pole. What will happen with the pendulums swing? It will clearly rotate by . Note
that this angle is equal to the solid angle subtended by the path:
1
A = (4R )
2
2

= R2 = R2

This statement is clearly true for any path. Draw a picture.

A Foucault pendulum is an example. It is located at latitude . Earths rotation effectively
causes the pendulum to move with respect to a non-rotating ghost within Earth. But the effect
is identical. After one day, the pendulum subtends the solid angle
=

cos

d(cos(0 )) = 2(1 cos )

which is also the angle by which the axis of the pendulum rotates after one day. In the rotating
frame, we typically attribute the very same angle to the Coriolis force, but you can see that its
origin may also be argued to be geometric.

Quantum mechanics: closed path in the space of Hamiltonians

In quantum mechanics, we replace the pendulum by any quantum mechanical system we like, and
the closed path is replaced by a path in the space of Hamiltonians. We will follow this path slowly
i.e. adiabatically. And the phase were interested in is nothing else than the complex quantummechanical phase of the physical system we study. You can see that this is a special case of Bailes
phase we discussed last time but because the path is closed, the resulting change of the phase is
called Berrys phase.
In other words, the latitude and longitude are replaced by external parameters on which the
Hamiltonian depends.
Finally, we should try to calculate Berrys phase. Recall that the standard phase associated with
time-independent Hamiltonians is
(t) = n eEn t/ih
1

What about time-dependent Hamiltonians? The most straightforward answer is to replace En t in

Rt
the exponent which is not well-defined because En changes by the integral tif En (t0 )dt0 . Note
that it gives the same answer if the energy is constant. A natural guess Ansatz is therefore
Rt

(t) = n (t)e

dt0 En (t0 )/i

h i(t)

where we had to use n (t) the eigenstate of the Hamiltonian H(t) whose character depends on
time; the usual dynamical phase n (t), expressed using an integral; and Berrys phase ei(t) to
account for the phase errors we may have overlooked. What should we do with this Ansatz? Much
like with other Ansatze, we should plug it into an equation. In this case it is the time-dependent
Schrodinger equation
(t)
= H(t)(t)
i
h
t
The left-hand side contains a time-derivative of a product. It is clear that all resulting terms will be
proportional to n (t) exp(i(t)), so let us divide the resulting equation by this factor (both phases)
immediately:
i
h

n (t)
dn (t)
+ En (t)n (t) h

n (t) = H(t)n (t) En (t)n (t)

t
dt

Differential equation for the phase

You see that the terms containing En (t) cancel and we have, after dividing by i
h,
dn (t)
n (t)
= i
n (t)
t
dt
We can obtain a formula for the time-derivative of itself if we multiply the equation by hn |:
dn
n
= ihn |
i
dt
t
Dont forget our goal: to compute the total phase increment resulting from a closed path in
the space of Hamiltonians. Imagine that the Hamiltonian depends on several parameters such as
~ it is a
the size of various fields etc. and we formally unify all these parameters into a vector R;
vector in a d-dimensional space where d can be any integer and the space has nothing to do with
the space where we live. The closed path in the space of possible Hamiltonians is determined by
~
a vector-valued function R(t)
of time. What is the change of the eigenstate with time? Recall the
rules for derivatives of composite functions:
~
n (t)
~ R dR n R1 + n R2 + . . .
=
t
dt
R1 t
R2 t
Now we can simply integrate the equation for dn /dt to get
n = i

~
~ R n (R)id
~ R
~
hn (R)|

~ came from dt dR/dt.

~
~ were one-dimensional, you can see that n
where dR
If the space of vectors R
will be zero because the contributions from the two parts of the path (back and forth) will cancel;
they only differ by the sign.
As we mentioned last time, the total probability the squared norm of the wavefunction is
conserved. This means, in our case, that n must be real. Indeed, the whole factor in the integrand
2

~
~ R n (R)i
~ is pure imaginary, making the whole integral real (point by point). Why is it pure
hn (R)|
imaginary? Note that
~ n (R)i
~ =1
hn (R)|

~ R hn |n i
0=

Using the Leibniz rule,

~ R n |n i + hn |
~ R n i = hn |
~ R n i + hn |
~ R n i
0 = h

~ R n i, and because the real part

The last sum is nothing else than twice the real part of hn |
vanishes, the object itself must be pure imaginary. This proof was mathematically identical to the
~ operator acted on a different
proof that the momentum operator is Hermitean even though the
space.

Analogy with the magnetic flux

We should notice that our formula for the phase
n = i

~
~ R n (R)id
~ R
~
hn (R)|

is reminiscent of the magnetic flux through a surface :

=

~
~ d Area
B
=

~
~ A)
~ d Area
(
=

~
~ dl
A

The last step used Stokes theorem. Applying Stokes theorem in the opposite direction to our
~ lives in a three-dimensional space for a moment, we see that
situation and assuming that R
n = i

~
~ R n (R)id
~ R
~ =
hn (R)|

~
~n dS
V

~n is our counterpart of the magnetic field B:

~
where V
~n Im
~ R hn |
~ R n i
V
We can bring the last expression to a more symmetric form if we realize that
~ (f g)
~ = f
~ g
~ + f (
~ g)
~ = f
~ g
~

and therefore

~n = Im h
~ R n | |
~ R n i
V

Note that the cross-product with itself would vanish if the gradient were real, but because it is
complex, it is a cross product with someone else (its complex conjugate) and it does not vanish. Let
us also insert a complete set of states:
~n =
V

m6=n

~ R n |m i hm |
~ R n i
Im h

where we omitted the term m = n because it vanishes, being the imaginary part of a manifestly real
number.

Instead of calculating gradients of the eigenvectors m , it may be more convenient to find an

expression that involves the gradient of the Hamiltonian itself (and its matrix elements). To achieve
this new form of the formula, we first prove that for m 6= n,
~ R H|n i
hm |
En Em

~ R n i =
hm |
Why is it true? It is because
n i = En n
H|

~ R H)|
n i + H|

~ R n i = (
~ R En )|n i + En |
~ R n i
(

and because the last equation can be multiplied by hm |:

~ R H)|
n i + hm |H|

~ R n i = hm |(
~ R En )|n i + hm |En |
~ R n i
hm |(
The first term on the right-hand side vanishes because it is proportional to a vanishing inner product.
~ R n i and the new form of the equation follows
The second terms on both sides give (En Em )hm |
directly. Using this new form, we may also rewrite V~n as
~n =
V

m6=n

Im

~ R H|m i hm |
~ R H|n i
hn |
(En Em )2

Example: particle in rotating magnetic field

~ magnetic field whose direction is slowly
Consider an atom with an arbitrary spin in a constant B
~ B
~ is indeed three-dimensional. The Hamiltonian and the corresponding
moving. In this case, R
energy eigenvalues are
B)
~ = gS
~ B/
~ h
H(

Em = gmB

The gradient of the Hamiltonian in the parameter space (parameterized by the magnetic field) is
~ BH
= gS/
~ h

We always measure m with respect to the axis given by the instantaneous value of the magnetic
~m , we will need the matrix elements of the spin S.
~ Only m 6= m0 contribute
field. To calculate V
~m . Recall that the matrix element hs, m0 |S|s,
~ mi only has nonzero matrix elements for m = m0
to V
0
(which we do not need) and for m = m 1, and these nonzero matrix elements are those from the
raising and lowering operators:
m0 = m + 1 :
m0 = m 1 :

~ mi =
hs, m0 |S|s,

~ mi =
hs, m0 |S|s,

h
2

h
2

~m , we have
Substituting into our formula for V

s(s + 1) m(m + 1)(~ex i~ey )

s(s + 1) m(m 1)(~ex + i~ey )

2 2
~m = g /2 ([s(s + 1) m(m + 1)] + [s(s + 1) m(m 1)]) ~ez = m~ez
V
g 2 2 B 2
B2

Note that the parameters g and canceled and the result has a simple geometric form. The Berrys
phase is simply the flux
I
1 ~
B dB = m
m = m
C B2
4

and it is simply given by (m) times the solid angle surrounded by the closed path. The result is
completely analogous to the classical result for the pendulum.
The problem was mathematically equivalent to the flux of the magnetic field of a point-like
magnetic monopole except that the contours (and other geometric objects such as the solid angle)
~ rather than the ordinary threewere taken in the three-dimensional space of possible values of B
dimensional space. Finally, you should notice that
~ Instead, it was determined
the total Berrys phase was independent of the magnitude of B.
purely by the solid angle, a geometric quantity. The cancellation of all the dimensionful
parameters and the independence on the speed of the changes is what we mean when we
say that the effect is purely geometrical
in the special case m = which can occur both for fermions as well as bosons we have
exp(im ) = 1
~ = 0; at this point
the formula for the phase breaks down if we probe the singular point at B
the states are degenerate in energy (many states have the same energy).
~ 6= 0 and avoid the degenerate point, you can see that the actual phase is
even if we keep B
measured by the solid angle defined with respect to the degenerate point

Aharonov-Bohm effect
Much earlier, in 1959, a special example of the Berrys phase was figured out by David Bohm and
his Israeli student Yakir Aharonov. When Aharonov was defending his PhD thesis, the committee
agreed that the content was nonsensical but Aharonov had made such a good job in defending it
that he had to pass. Nevertheless, the effect was observed in 1960 and today virtually everyone
believes that the effect obviously exists. It has become a part of introductory lectures on quantum
mechanics like this one.
Whats the problem? The effect involves a charged particle moving outside a solenoid. In
~ r ) and B(~
~ r ), and the force acting
classical physics, the electromagnetic field is fully described by E(~
on a particle is the Lorentz force
~ + ~v B).
~
F~ = e(E
~ = 0 and B
~ = 0 along the trajectory of your particle, the force simply vanishes. There is no
If E
~ B
~ in terms of the potentials , A
~
field. You can still express E,
~
~ =
~ A ,
E
t

~ =
~ A
~
B

~ may be nonzero. As long

but we only use them for convenience, and it does not matter that , A
~
~
as E and B are zero, we say that there is no field and no physical effect caused by such a field. For
~ by the so-called gauge transformation
example, you can always change , A

,
t

~A
~ +
~
A

~ but not E,
~ B,
~ and therefore physics is unchanged.
which will change , A
~ B,
~ , A
~ fields and gauge
The situation in quantum mechanics is analogous there are also E,
transformations. But the quantum mechanical case differs in certain subtle features that actually
~ even though they do not affect E,
~ B.
~ What do we mean?
allow us to see some parts of , A
5

Recall that the quantum Hamiltonian is constructed from the classical Hamiltonian by adding
hats:
h
i
= 1 p~ q A(
~ ~r ) 2 + q(~r )
H
2m

~
Recall that (p~ q A) = m~v is proportional to the actual velocity and its expectation value is invariant
under the gauge transformations. These transformations also act on the wavefunction
(~r) (~r)eiq(~r)/h
and we will study the combined transformation later. Once again, gauge transformations change
~ but they also change ~p in such a way that the combination is not changed as we will see.
, A
~
What Aharonov and Bohm have shown is that the particle is affected by and especially A
~ and B
~ vanishes everywhere along the trajectory of the particle. Consider a very long
even though E
solenoid draw a long cylinder whose cross section is a circle of radius a. It can be produced by a
wire (with a current) densely wrapped on the cylinder. Outside the solenoid, the magnetic field is
zero. Inside the solenoid, the magnetic field is parallel to the axis of the solenoid. It is constant and
~ outside the solenoid can be calculated:
its value is B. The total flux and A
= a2 B,

~ , ) = e
A(r,
2r

~ around the solenoid gives you the flux as required by Stokes

Check that the contour integral of A
theorem.
Now we want to study a charged particle. Imagine a bead on a wire of radius b surrounding the
solenoid; b > a. Add the new wire to your picture. The Hamiltonian for the charged particle is
h
i
= 1
~
~ + q 2 A2
H
h2 2 + 2i
hq A
2m

where we could be sloppy about the ordering of the mixed term because the commutator is propor~ A
~ = 0 in our case. Because we only allow the particle to move in the direction, we
tional to
should now replace
~ e d .

b d
With this substitution, the Schrodinger equation reads
1
q
h
2 d2
h
q d
+
2 2 +i 2
2m
b d
b d
2b
"

2 #

() = E()

and it is a homogeneous (=without the absolute term) linear differential equation of the general
form
!
d2
d
2 +
+ () = 0.
d
d
Such equations have solutions of the type = exp() where are the solutions of the characteristic
quadratic equation

2 4
2
+ + = 0

=
2
In our case

h
2
= 2,
b

h
q
=i 2,
b
6

q
=
2b


2

2mE

which leads to the solutions

b
q

2mE .
=i
2
h h

Because the wavefunction must be periodic, ( + 2) = (), we see that must be i times an
integer, i.e. = in. We see that
b
q

n=
2mE
2
h h

or

q
h
2
n
E = En =
2
2mb
2
h


2

for

n = 0, 1, 2, . . .

When the flux was zero, n and n corresponded to a particle going in opposite directions with the
same speed, and these two states had the same energy. A nonzero flux breaks the degeneracy. You
see that the spectrum depends on the flux even though the particle is confined to a wire in which
~ = 0. That could not happen in classical physics. If you go to the classical limit, the spacing of
B
eigenstates becomes infinitely dense and you wont be able to determine n accurately, i.e. you wont
be able to see any effect of as h
0.

Gauge transformation
You see that the flux modified the spectrum in a rather simple way. In fact, the solutions of the full
~
~ A
~ = 0)
Schrodinger equation including the nonvanishing A-terms
in a field-free region (where
may be obtained as
Z ~r
iq/
h 0
~ r 0) d~r0
A(~
=e
,
(~r)
~
r0

~ = 0 in the same region.

where 0 is a solution of the simpler equation with A
Proof: We start with the assumed validity of the simpler Schrodinger equation
p2

+ V (~r) i
h
0 = 0
2m
t
!

The important step we want to prove is

~ = (p~ q A)e
~ iq/h 0 = eiq/h (p~) 0
(p~ q A)
~ exp(iq/
In the last step, the term q A
h) 0 cancelled the opposite term coming from the gradient of
~
the phase exp(iq/
h) in the Leibniz rule. You can use the same step again, replacing by (p~ q A),
to see that
~ 2 = (p~ q A)
~ 2 eiq/h 0 = (p~ q A)e
~ iq/h (p~) 0 = eiq/h (p~)2 0
(p~ q A)
In this derivation, all operators act on everything on the right. You see that everytime we move
~ to the simple (p~). Therefore the simple,
the phase exp(iq/
h) to the left, we transform (p~ q A)
0
~
A-free
Schrodinger equation for multiplied by the phase exp(iq/
h) is completely identical to the
complicated equation for . Of course, the 1/2m factor can be added to the equation above and
the V and E terms in the equation are unaffected by the phase redefinition. QED.
Incidentally, we could have also derived from 0 times the phase whose was time-dependent.
In that case, the Schrodinger equation for (without the potential ) would include an extra term

hq(/t) 0 times the usual phase on the time-derivative side, and this term would cancel
against the electrostatic potential energy term q for = (/t).
7

Aharonov-Bohm interference pattern

The statement we have just proved means that by changing the phase of appropriately, we can
~ = 0 in the field-free regions. However, it is not quite right because the appropriate
essentially set A
phase for (~r) depends on the contour that ends at ~r. The two contours that avoid the solenoid
in opposite directions lead to phases that differ by q/
h where is the magnetic flux. Aharonov
and Bohm realized that this phase shift can be measured by an interference (double-slit) experiment
where a solenoid is inserted in between the two slits. Draw a figure of the experiment.
The position of the interference maxima depends on the magnetic flux even though the particle
~ 6= 0. Note that the interference patterns disappear in the classical
never visits the space with B
limit, together with all these extra phases. In other words, if you send h
0, all the phases are
essentially random being proportional to a large number 1/
h.

Aharonov-Bohm effect as an example of Berrys phase

The Aharonov-Bohm phase = q/
h can also be calculated as a special example of Berrys phase
we started with today. Imagine that we have a particle in a box of medium size and we slowly move
~ around the solenoid. Let us start from the final steps and
the box whose central position is R
the result of Berrys derivation: the overall phase will be computed from our oldest integral formula
for n :
I
I
~ R n i dR
~ = q A(
~ R)
~ dR
~ = q
n = i hn |
h

which is the Aharonov-Bohm result for the phase. The last step is clear but the previous step is
not. The required identity is
~ R).
~
~ R n i = i q A(
hn |
h

~ the position of the box, to define our adiabatic parameters,

How do we prove it? We consider R,
and we define the corresponding eigenstates associated with the same eigenvalues using our previous
trick
Z ~r
~ r 0 ) d~r0
n = eiq/h n0 ,
(~r)
A(~
~
R

This implies that

~ R)e
~ iq/h 0 (~r R)
~ R n =
~ R [eiq/h 0 (~r R)]
~ + eiq/h
~ = i q A(
~ R 0 (~r R)
~

n
n
n
h

The desired identity is obtained by multiplying this ket-equation by hn |. You can see that you get
what you need from the left-hand side and the first term of the right-hand side. The second term on
~ R =
~ r because 0 only
the right hand side will vanish after being contracted with hn | because
n
~ and because the expectation value of the momentum #
~ r in a box vanishes.
depends on ~r R