Lecturenotes PDF

Advanced Quantum Mechanics
Notes by P.R. Kaye based on the lectures of A. Kempf

December 18, 2002
Advanced Quantum Mechanics 1
Contents
1 Notation 3
2 From Classical Mechanics to Quantum Mechanics 3
2.1 Classical Mechanics: . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 3
2.2 Whats Dierent for Quantum Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 5
2.3 Heisenberg Equations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.4 A Representation for x, p . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.5 Dynamics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 6
2.5.1 Time-Evolution Preserves the CCRs and Hermiteicity Requirement . . . . . 6
2.5.2 The Time-Evolution Operator . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 7
2.6 Measurement of Observables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 8
2.7 The General Uncertainty Principle . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 9
2.8 The Time-Energy Uncertainty Relation . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 10
2.9 Pictures of Quantum Mechanics . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.10 The Heisenberg Picture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.11 The Schrodinger Picture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 11
2.12 The Dirac Picture . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 12
3 Representation Theory 13
3.1 Dual Spaces . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.2 Bases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 14
3.3 for Matrix Representations . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 15
3.4 Eigenvectors in Quantum Mechanics: State Collapse . . . . . . . . . . . . . . . . . . . . . . . 15
3.5 The Spectral Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 16
3.6 Whats Dierent in -Dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.6.1 Vectors [) in -dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 17
3.6.2 Operators

A[) in -dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.6.3 Scalars [
A[) in -dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.6.4 Spectral Theorem in -dimensions . . . . . . . . . . . . . . . . . . . . . . . . . . 18
3.7 Sharpness of Measurement Predictions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 19
3.8 The Position and Momentum Eigenbases . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 20
3.9 Hermite and Fourier Transforms . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 21
3.10 Equivalence of Representations: the Stone/Von-Neumann Theorem . . . . . . . . . . . . . . . 23
4 Quantum Mechanics in the Schrodinger Picture, and the Position Basis 24
4.1 The Schrodinger Equation Revisited . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.2 Predicting Observables . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 24
4.3 The Green Function Method for solving Schrodinger Equations . . . . . . . . . . . . . . . . . 25
5 Feynmans Formulation of Quantum Mechanics 26
6 Density Matrices and Mixed States 28
6.1 Density Matrices and Mixed States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 28
6.2 Example: quantum system o in a heat bath . . . . . . . . . . . . . . . . . . . . . . . . . . . . 30
6.3 Decoherence and Mixed States . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 31
7 Compositions of Several Systems 33
8 Identical Particles 36
8.1 Bosons and Fermions . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 36
8.2 Special Behaviour of Bosonic and Fermionic Systems . . . . . . . . . . . . . . . . . . . . . . . 37
9 Angular Momentum 38
9.1 Generators of Symmetries . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 38
9.2 Orbital Angular Momentum . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 39
9.3 Angular Momentum and Identical Particles . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
9.4 Orbital Angular Momentum v.s. Spin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 40
9.5 Spin . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 43
9.6 Working With Spin-
1
2
Particles . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 44
9.7 Example: e
in a B-Field . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 46
10 Time-Independent Perturbation Theory 46
11 Bells Theorem 48
12 The Adiabatic Theorem 50
12.1 Dynamical Phase for a Time-Independent Hamiltonian . . . . . . . . . . . . . . . . . . . . . . 50
12.2 The Adiabatic Theorem . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 51
12.3 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 52
13 The Berry Phase 53
13.1 Geometric Phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 53
13.2 Development of the Berry Phase . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 54
13.3 When is the Berry Phase Nonzero? . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 55
13.4 The Berry Phase is Real . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
13.5 Alternative Expression for
j
(t) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
13.6 Example . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 56
14 APPENDIX: The -Function and Fourier Transforms 59
14.1 Denition of (x) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
14.2 Properties of (x) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 59
14.3 The Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
14.4 Properties of the Fourier Transform . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
14.4.1 Important Limit Property . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
14.4.2 Linearity . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 60
14.4.3 Fourier Transform of Derivatives . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
14.4.4 Multiplication Convolution . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
14.4.5 Scaling and Shifting Properties . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
14.4.6 Fourier Transform of (x) . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . . 61
1 Notation
Quantum mechanical quantities will be denoted with a hat: .
Vector quantities are denoted using bold face; e.g. x (x
1
, x
2
, x
3
).
Super/sub-scripts: x
(r)
i
denotes the i-component (i 1, 2, 3) of x for particle r.

f and g will always be assumed to be some polynomials in position and momentum variables x and p.
Time derivatives will sometimes be indicated with a dot (e.g.
dx
dt

x)
[ x, p] denotes the commutator x p p x.
The Kronecker and Dirac deltas,
m,n
, and (x), are detailed in the Appendix.
2 From Classical Mechanics to Quantum Mechanics
2.1 Classical Mechanics:
In Classical Mechanics (CM) the equations of motion are derived via Hamiltons Equation. To prove this
from more basic principles is beyond our scope here, so we just take this as a basic assumption. Suppose
f is some physically observable quantity, which is a polynomial function of position x, momentum p (and
possibly has an explicit dependence on t).
Physical Assumption 1 The equation of motion governing the time-evolution of f follows Hamiltons
equation
d
dt
f = f, H +
f
t
where , is the Poisson Bracket.
The term
f
t
on the RHS of Hamiltons equation is the explicit time-dependence of f, that is, not accounting
for the time-dependence through x and p (which are ultimately number-valued functions of t). In Hamiltons
equation, x and p are treated as abstract variables, and f is some polynomial in x and p. In particular, this
means that if f is dependent on time only through x and p, Hamiltons equation simplies to
d
dt
f = f, H ().
The Poisson bracket is a multiplication rule for polynomials in positions x and momenta p. It is dened by
two sets of axioms:
A) x
(r)
i
, p
(s)
j
=
i,j
r,s
x
(r)
i
, x
(s)
j
= 0
p
(r)
i
, p
(s)
j
= 0
B) f, g = g, f
cf, g = cf, g
f, g +h = f, g +f, h
f, gh = f, gh +gf, h
An important consequence of the axioms of , is the Jacobi Identity:
f, g, h +h, f, g +g, h, f = 0.
The choices for f in Hamiltons equation that are most important for classical mechanics are f = x
(r)
i
and
f = p
(r)
i
. For these choices of f Hamiltons equation () becomes:
d
dt
x
(r)
i
= x
(r)
i
, H (1)
d
dt
p
(r)
i
= p
(r)
i
, H (2)
As a concrete example illustrating that the equations of motion for a system follow Hamiltons equation,
consider a free particle of mass m possessing only kinetic energy. Its Hamiltonian is
H =
3
j=1
p
2
j
2m
.
Using this H in equations (1) and (2) we obtain the following equations of motion:
d
dt
x
i
=
_
_
_
x
i
,
3
j=1
p
2
j
2m
_
_
_
=
p
i
m
d
dt
x
i
=
_
_
_
p
i
,
3
j=1
p
2
j
2m
_
_
_
= 0.
These agree with what is learned in high-school physics: p
i
= m x
i
and x
i
= 0.
As another concrete example, consider a system of two point-masses m
1
, m
2
which are connected by a spring
with spring-constant k. Its Hamiltonian is
H =
_
p
(1)
_
2
2m
1
+
_
p
(2)
_
2
2m
2
+
k
2
_
x
(1)
x
(1)
_
2
.
Using this H in equations (1) and (2) give the equations of motion
d
dt
x
(r)
i
= x
(r)
i
, H =
p
(r)
i
m
r
d
dt
p
(1)
i
= p
(1)
i
, H = k
_
x
(1)
i
x
(2)
i
_
d
dt
p
(2)
i
= p
(2)
i
, H = k
_
x
(2)
i
x
(1)
i
_
In CM, x, p are real-valued functions. This means that the Poisson bracket can be evaluated using the
following representation:
f, g =
f
x
g
p

f
p
g
x
2.2 Whats Dierent for Quantum Mechanics
Quantum Mechanics (QM) also obeys Hamiltons equation, and all the axioms of the Poisson algebra.
What diers is the mathematical representations of position and momentum. The experimentally observed
phenomenon of incompatible measurements suggests that position and momentum cannot be number-valued
functions in QM. Suppose x and p denote the mathematical objects we use to represent position and momen-
tum. The condition that x and p cannot be number valued functions suggests that x p ,= p x, or equivalently
[ x, p] ,= 0 (if a representation consistent with the Poisson algebra satised [ x, p] = 0, it would be isomorphic
to a number-valued representation). If x and p are not number-valued, then the representation of , in
terms of partial derivatives that we had for classical mechanics is no longer valid. So we need a new represen-
tation of , that is both mathematically consistent with the axioms dening it, and is physically consistent
with observed data. We begin by examining the constraints given by the requirement for mathematical
consistency with the axioms of the Poisson algebra.
We evaluate u
1
u
2
, v
1
v
2
in two ways, and require that the results agree.
u
1
u
2
, v
1
v
2
= u
1
u
2
, v
1
v
2
+ u
1
, v
1
v
2
u
2
= u
1
( v
1
u
2
, v
2
+ u
2
, v
1
v
2
) + ( v
1
u
1
, v
2
+u
1
, v
1
v
2
) u
2
(1)
and also
u
1
u
2
, v
1
v
2
= v
1
u
1
u
2
, v
2
+ u
1
u
2
, v
1
v
2
= v
1
( u
1
u
2
, v
2
+ u
1
, v
2
u
2
) + ( u
1
u
2
, v
1
+u
1
, v
1
u
2
) v
2
(2)
Equating (1) and (2) we get the requirement
( u
1
v
1
v
1
u
1
) u
2
, v
2
= u
1
, v
1
( u
2
v
2
v
2
u
2
) .
This means we must have
( u
1
v
1
v
1
u
1
) = K u
1
, v
1
and
( u
2
v
2
v
2
u
2
) = K u
2
, v
2
where K is some entity that is independent of the u

1
, u
2
, v
1
, v
2
. Since u
1
, u
2
, v
1
, v
2
were arbitrary, in general
we must have
K u, v = u v v u ()
for all u, v polynomials in x, p.
Classical mechanics is obtained in the case K = 0. In quantum mechanics, based on experiments we have
the following:
Physical Assumption 2 Quantum mechanics is consistent with the choice K = i (where is Planks
constant) in equation ().
So, in quantum mechanics we have
[

f, g] = i
f, g
which leads immediately to Heisenbergs Canonical Commutation Relations (CCRs):
[ x
(r)
i
, p
(s)
j
] = i
i,j
r,s
[ x
(r)
i
, x
(s)
j
] = [ p
(r)
i
, p
(s)
j
] = 0
and a representation of the Poisson bracket for QM:
f, g =
1
i
[

f, g]
2.3 Heisenberg Equations
Using the above representation for the Poisson bracket, in QM Hamiltons equation is represented as
d
dt
f =
1
i
[

f,

H] +
f
t
.
The Heisenberg equations are the Hamilton equation for the cases

f = x and

f = p:
i
x = [ x,

H]
i
p = [ p,

H]
The physics will also require x = x
, and p = p
(Hermiteicity, or reality conditions). We say

f is Hermitean
when

f =

f
. We come back to exactly what means when we look at representation theory.

Before proceeding, we record a fundamental physical assumption underlying our development of quantum
mechanics.
Physical Assumption 3 Any physically observable quantity can be expressed as a linear, Hermitean oper-
ator

f which acts on a complex vector space ( Hilbert space, dened later). Such an

f is appropriately called
an observable of the system.
2.4 A Representation for x, p
The x and p can be represented in a way consistent with the development above, using innite-dimensional
square matrices.
Denition 1
a :=
_
_
_
_
_
_
_
_
0

1 0 0
0 0

2 0
0 0 0

3
0 0 0 0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
_
_
_
_
_
_
_
_
Then let
x(t
0
) = L(a
+a)
p(t
0
) =
i
2L
(a
a)
where L R.
This gives x(t
0
) = x
(t
0
) , p(t
0
) = p
(t
0
) and [ x(t
0
), p(t
0
)] = i. As we shall see, it turns out that these
properties are preserved under the time-evolution of a quantum system, and so they will hold in general for
x(t) and p(t).
2.5 Dynamics
2.5.1 Time-Evolution Preserves the CCRs and Hermiteicity Requirement
Theorem 1 The CCRs are preserved under time-evolution.
Proof: We need to show
d
dt
[ x, p] = 0.
d
dt
[ x, p] = [
x, p] + [ x,

p] product rule
=
1
i
_
[ x,

H], p
_
+
1
i
_
x, [ p,

H]
_
=
1
i
_
p, [

H, x]
_
+
1
i
_
x, [ p,

H]
_
=
1
i
_
H, [ x, p]
_
(Jacobi Identity)
= 0
Theorem 2 If

H satises

H
=

H, then the Hermiteicity of x and p is preserved under time-evolution.
Proof: We show that
d x(t)
dt
is Hermitean.
_
d x(t)
dt
_
=
_
1
i
[ x(t),

H]
_
=
1
i
[

H, x(t)] (use (AB)
= B
)
=
1
i
[ x(t),

H]
=
d x(t)
dt

The proof that
d p(t)
dt
is Hermitean is analogous.
2.5.2 The Time-Evolution Operator
Denition 2 The time-evolution operator

U(t) is the solution to
i
d
dt
U(t) =

U(t)

H(t) ,

U(t
0
) = 1
which is
U(t) = Te
1
i
t
t
0
H(t)dt
where T is the time-ordering symbol, needed to resolve ordering ambiguities arising from noncommutivity.
Note:

U(t) is unitary (
(t) =

U
1
(t)).
The time-evolution operator is simpler when

H
t
= 0:
U(t) = e
1
i
(tt0)

H
.
Once we have the time-evolution operator, it gives the solution to the equations of motion.
Theorem 3 Assume x
(r)
i
(t
0
), p
(r)
i
(t
0
) obey the CCRs and are Hermitean. Then
x
(r)
i
(t) =

U
(t) x
(r)
i
(t
0
)
U(t) ()
p
(r)
i
(t) =

U
(t) p
(r)
i
(t
0
)
U(t)
are the full solutions to the equations of motion.
Proof: First, we will need the following expressions for

U and

, which follow immediately

from the denition of

U, and using the fact that

H is Hermitean, and that (AB)
= B
U(t) =
1
i
U

H
(t) =
1
i
H

U
Now, we check that x

(r)
i
(t) obeys the equation of motion.
i
d
dt
x
(r)
i
(t) = i
d
dt
_
(t) x
(r)
i
(t
0
)
U(t)
_
= i
_

U
(t) x
(r)
i
(t
0
)
U(t) +

U
(t) x
(r)
i
(t
0
)

U(t)
_
=
H(t)
(t) x
(r)
i
(t
0
)
U(t) +

U
(t) x
(r)
i
(t
0
)
U(t)

H(t)
=
H(t) x
(r)
i
(t) + x
(r)
i
(t)

H(t)
= [ x
(r)
i
(t),

H(t)]
The proof that p
(r)
i
(t) obeys the equation of motion is similar.
For any observable

f (polynomial in x, p) which is Hermitean and obeys the CCRs at time t
0
, Theorem 3
generalizes to give the equation for

f(t):
f(t) =

U
(t)

f(t
0
)
U(t).
We can also recover the Hamiltonian from the time-evolution operator:

H = i
(t)
d
dt
U(t).
2.6 Measurement of Observables
We have used a matrix representation of x, p. We would like measurements of these quantities to yield
real numbers. How do we compute these? In QM, we can compute the expected values x, p of position and
momentum measurements.
Physical Assumption 4 A unit vector [) (in a complex Hilbert space) can be chosen which contains
enough information about the state of a quantum system at time t
0
to predict measurement expecation values.
The expectation values are given by
x(t) = [ x(t)[)
p(t) = [ p(t)[)
f(t) = [
f(t)[) for any observable

f satisfying

f
=

f.
Notice that in this picture, the state-vector [) describes the system at the initial time t
0
, and it is the
operators

f describing observable quantities that vary with time.
Note: Observables are required to be Hermitean (

f
=

f). The Hermiteicity requirement ensures that
measurement outcomes are real numbers.
The time-evolution of x(t), p(t) does not in general follow the equations of motion of classical mechanics.
Theorem 4 ( Ehrenfest Theorem) If

H is a polynomial in the x and p variables of degree 2, then x(t)
and p(t) obey the classical equations of motion.
Proof: Suppose

H is a polynomial in the x and p variables of degree 2. Then
x = x,

H =: g( x, p) is linear in x, p, and
p = p,

H =: h( x, p) is linear in x, p
= g( x, p) = g(x, p) and h( x, p) = h(x, p)
We also dene the variance of

f(t)
(

f(t)) =
_
f(t) f(t)
_2
=

f
2
(t)
_
f(t)
_
2
= [
f
2
(t)[) [
f(t)[)
2
and the standard deviation f =
_
(

f(t)).
2.7 The General Uncertainty Principle
Theorem 5 Assume

f, g are Hermitean, and that a system has state-vector [). Then fg
1
2
[[

f, g][)
.
Proof: Recall from linear algebra: [) 0. Choose
[) =
_
f f1 +i( g g1)
_
[),

R.
Then [) 0 gives:
[
_
f f1 i( g g1)
__
f f1 +i( g g1)
_
[) 0
= [(

f f1)
2
[) +
2
[( g g1)
2
[) +[i(

f g g

f)[) 0
= (f)
2
+
2
(g)
2
+[i[

f, g][) 0
= (f)
2
+ (g)
2
_
+
[i[

f, g][)
2(g)
2
_
2
_
[i[

f, g][)
_
2
(2(g)
2
)
2
(g)
2
0
The above inequality is most stringent if we choose =
]i[

f, g]])
2(g)
2
. Then we have
(f)
2
_
[i[

f, g][)
_
2
4(g)
2
0
= fg
1
2
[[

f, g][)

Alternative Proof of the Uncertainty Principle: The uncertainty principle is often stated
in the (equivalent) form:
(f)
2
(g)
2
_
1
2i
[

f, g]
_
2
which has the advantage that it allows the Cauchy-Schwartz inequality to be exploited, making
the proof appear somewhat simpler. We have
(f)
2
=
_
f f
_2
= [
_
f f
_
2
[) = F[F)
where [F)
_
f f
_
[) (note that since

f is Hermitean, so is

f f. Similarly we have (g)
2
=
G[G). Now, apply the Cauchy-Schwartz inequality:
(f)
2
(g)
2
= F[F)G[G) [f[g)[
2
.
For any complex number z,
[z[
2
= (Re(Z))
2
+ (Im(z))
2
(Im(z))
2
=
_
1
2i
(z z
)
_
2
So, letting z = F[G), we have
(f)
2
(g)
2
_
1
2i
(F[G) G[F))
_
2
.
Now,
F[G) = [
_
f f
_
( g g) [)
= [
_
f g

fg gf +fg
_
[)
= [
f g[) g[
f[) f[ g[) +fg[)

= fg fg gf +fg
= fg fg.
Similarly,
G[F) = gf fg.
Therefore
F[G) G[F) = fg gf = [

f, g].
So we have
(f)
2
(g)
2
_
1
2i
[

f, g]
_
2
An important special case of the general uncertainty principle is position-momentum uncertainty:

x(t)p(t)

2
t.
2.8 The Time-Energy Uncertainty Relation
There is a time-energy uncertainty relation
t

H

2
which is not a consequence of the general uncertainty principle above. What does the time-energy uncertainty
mean? Time t is not an observable quantity like

f, so how do we dene t? Any uncertain entity

f takes
time for a change to be noticeable. A change is not noticeable until it is about one standard deviation f.
So the time it takes is t where t
df(t)
dt
= f. That is
t =
f(t)
df(t)
dt
.
The time-energy uncertainty relation is then derived as follows. For Hermitean

f we have
i
d
dt
f(t) = [

f,

H] (1)
f(t)

H
1
2
[[

f,

H][)
(2)
Sub (1) into (2):
f(t)

H
1
2
[i
d
dt
f(t)[)
=

2
d
dt
[
f(t)[)
=

2
df(t)
dt
=

2
f(t)
t
=t

H

2
2.9 Pictures of Quantum Mechanics
We have used a matrix representation. Matrix multiplication is noncommutative, but it is associative.
Depending on how we bracket the fundamental equations, we can get dierent pictures. The pictures
dier in which dierent quantities are considered constant, and which are time-dependent.
2.10 The Heisenberg Picture
What we have considered so far is the Heisenberg Picture. In this picture, the state vector [) is constant, and
the observables

f(t) are time-dependent. We solve i
d
dt
f(t) = [

f(t),

H(t)] +

f
t
directly and then calculate
measurement-expectations f(t) = [
f(t)[).
2.11 The Schr odinger Picture
This is the picture most often used in practise for quantum mechanics. Recall the time-evolution operator
method:

f(t) =

U
(t)

f(t
0
)
U(t). Substituting this into the equation for measurement-expectations gives

f(t) = [
(t)

f(t
0
)
U(t)[) ().
Recall that in this expression

f(t
0
) is a constant matrix, and [) is the (constant) initial state vector. In the
Schrodinger picture we dene time-dependent Schrodinger states:
[
s
(t)) =

U(t)[)
which, when substituted in () gives:
f(t) =
s
(t)[
f(t
0
)[
s
(t)).
An important point is that when

H
t
= 0, we can actually calculate [
s
(t)) without rst having to solve for
U(t). This is done by solving the Schr odinger Equation:

i
d
dt
[
s
(t)) =

H
s
(t)[
s
(t)).
where
H
s
(t) =

U(t)

H(t)
(t)
is called known as the Schr odinger Hamiltonian.
This formulation is especially simple for time-independent Hamiltonians.
Claim 1 If

H
t
= 0 then

H
s
(t) =

H(t).
Proof: Recall that when

H
t
= 0, the time-evolution operator has the simple form
U(t) = e
1
i
(tt0)

H
=
m=0
1
i
(t t
0
)
m
m!

H
m
and so
[
U(t),

H] = 0
=

H
s
(t) =

U(t)

H(t)
(t) =

H

U(t)
(t) =

H(t)
Derivation of the Schrodinger equation: Recall the denition of the time-evolution op-
erator i
d
dt
U(t) =

U(t)

H(t). From the denition of the Schrodinger Hamiltonian we have
H(t) =

U
(t)

H
s
(t)
U (t). Putting these together we get

i
d
dt
U(t) =

U(t)
(t)

H
s
(t)
U(t)
= i
d
dt
U(t) =

H
s
(t)
U(t) (since U is unitary)

= i
d
dt
U(t)[) =

H
s
(t)
U(t)[)
= i
d
dt
[
s
(t)) =

H
s
(t)[
s
(t))
2.12 The Dirac Picture
This is the picture most often used in practise for quantum eld theory. It is sometimes called the interaction
picture. Assume

H
s
(t) is given, and suppose we can write
H
s
(t) =

H
0
(t) +

H
t
(t)
such that

H
0
(t) is an easily solvable Hamiltonian (for example in quantum computing,

H
0
(t) is that of a
quantum gate). By assumption, we can easily nd the solution to i
d
dt
U
0
(t) =

H
0
(t)
U
0
(t). So assume we
have

U
0
(t) explicitly. Now, dene

U
t
(t) :=

U
0
(t)
U(t) so we have

U(t) =

U
0
(t)
U
t
(t). Then the equation for
measurement-expectations can be written as
f(t) = [
(t)

f(t
0
)
U(t)[)
= [
U
t
(t)
0
(t)

f(t
0
)
U
0
(t)
U
t
(t)[)
=
_
[
U
t
(t)
__
0
(t)

f(t
0
)
U
0
(t)
__
U
t
(t)[)
_
So, we dene operators in the Dirac picture as:
f
D
(t) =

U
0
(t)

f(t
0
)
U
0
(t)
and states in the Dirac picture as:
[
D
(t)) =

U
t
(t)[)
So, since

U
0
(t) is assumed easy, if we can calculate the Dirac states [
D
(t)) then we get expectation values
f(t). The dynamics of [
D
(t)) is captured in the following theorem.
Theorem 6 [
D
(t)) is the solution to
i
d
dt
[
D
(t)) =

H
D
(t)[
D
(t))
with initial condition [
D
(t
0
)) = [), where
H
D
(t) =

U
0
(t)

H
t
(t)
U
0
(t).
Proof:
i
d
dt
U(t) =

H
s
(t)
U(t)
i
d
dt
_
U
0

U
t
_
=
_
H
0
+

H/
_

U
0

U
t
i
U
0

U
t
+i
U
0
U
t
= i
U
0

U
U
0

U
t
+

H
t

U
0

U
t
i
U
0
U
t
=

H
t

U
0

U
t
i
U
t
=

U
H
t

U
0

U
t
i
U
t
=

H
D

U
t
i
d
dt
U
t
[) =

H
D

U
t
[)
i
d
dt
[
D
(t)) =

H
D
(t)[
D
(t))
Once we have [
D
(t)),
f(t) =
D
(t)[
f
D
(t)[
D
(t)).
3 Representation Theory
We need to represent x and p as mathematical objects which are Hermitean and which obey the CCRs. Any
mathematical object can be viewed as a map. In Quantum Mechanics the physically successful representa-
tions are as maps which are linear. We will look at linear maps in complex and unitary vector spaces in
nite dimensions, and then consider what changes for these in innite dimensions. A number of basic results
in linear algebra are recalled; many without proof.
3.1 Dual Spaces
In the following, assume H is a vector space over the complex numbers C, and vectors in this space are
written in the Dirac notation (e.g. [)).
Denition 3 Assume H is given. H
is dened as the set of linear maps H C. We denote elements of

H
by [, where the action of [ is:

[ : [) [) C
Theorem 7 H
, (with the natural denitions for + and scalar multiplication) is a complex vector space,
called the dual vector space of H.
Denition 4 Assume H, H
possess a map , called Hermitean conjugation:

: H H
and the inverse operation : H
H
so that
[)
= [ , [
= [) [) H.
Assume that obeys the axioms:
([) +[))
= [ +[
([))
[
[)
= [)
[) = 0 i [) = 0
Then we say that H is a unitary vector space.
3.2 Bases
Assume that H is nite-dimensional.
Denition 5 A set of vectors [b
m
)
m
H, where is some index set, is called an orthonormal basis
b
n
[b
m
) =
n,m
m, n
and every [) H can be written as
[) =
n
[b
n
) ,
n
C.
n
= b
n
[) are called the coecients of [) in basis [b
n
).
Theorem 8 The set b
n
[ is an O.N. basis for H
called the dual basis.

Denition 6 The map : C which maps n
n
is called the wavefunction of [) in the basis [b
n
).
Remark: The same [) has a dierent wavefunction for each choice of O.N. basis.
Theorem 9 Choose an O.N. basis b
n
n
. Then every linear map

T can be written as
T =
n,m
T
n,m
[b
n
)b
m
[
The action of

T is then
T : [)
n,m
T
n,m
[b
n
)b
m
[) where

T
n,m
= b
n
[
T[b
m
).
The identity operator can be written as
1 =
n
[b
n
)b
n
[ ().
() is called the resolution of the identity in the basis [b
n
)
n
.
Consider [
f[) and insert the resolution of the identity twice:

[
f[) =
n,m
[b
n
)b
n
[
f[b
m
)b
m
[)
=
n,m
n

f
n,m
m
Theorem 10 All matrix representations of the CCRs give the same physics, because they are all obtained
in this way, by choosing an O.N. basis.
3.3 for Matrix Representations
[
A[) means [
A([)).
Denition 7

A
: H H is dened as that linear map which, if acting rst on the left, will give the same
scalar as A acting on the right:
_
[)
_
[) = [
A[) , [), [) H.
That is,
[
[)
= [
A[) ().
If we insert the identity in (), we obtain

A
m,n
=

A
n,m
(i.e. is matrix Hermitean conjugation).
3.4 Eigenvectors in Quantum Mechanics: State Collapse
Theorem 11 If [) is an eigenvector of a Hermitean

f (an observable) with eigenvalue , at time t the
measurement of

f will surely yield .
Suppose

f[) = [). Then
f = [
f[)
= [[)
= [)
=
and
(f) = [
f
2
[) [
f[)
2
=
2
2
= 0
The following theorem guarantees that the measurement results will be real numbers.
Theorem 12 If

f =

f
and if

f[) = [) then R.
Suppose we have a system initially in state [) and at time t we perform measurements of several observables
f
(1)
,

f
(2)
, . . . ,

f
(s)
at once. Recall the uncertainty principle
f
(i)
f
(j)
1
2
[[

f
(i)
,

f
(i)
][)
. ()
To allow for reproducibility of measurements we want f
(i)
, f
(j)
= 0 i, j. So we need the RHS of ()
to be 0. This means we can only measure simultaneously commuting sets of observables. The maximum
number is s = 3r, (where r is the number of particles).
Suppose we prepare a state of r particles at time t
0
, and perform 3r measurements at time t
1
obtaining the
results v
i
R for

f
(i)
. We know (from experiment) that an immediate repeat measurement of the

f
(i)
at
time t
2
> t
1
will yield the same numbers v
i
. Starting from time t
1
, the system has the initial state vector
[) which is a joint eigenvector to all

f
(i)
with eigenvalues v
i
. The old state [) collapses to become the new
state [) at time t
1
. We record this fundamental observation here.
Physical Assumption 5 If a measurement of an observable

f of a quantum system in state [) is made
at time t, then the result of the measurement is some eigenvalue of

f, and the new state immediately after
the measurement is the eigenvector of

f corresponding to eigenvalue .
3.5 The Spectral Theorem
The eigenvectors of a Hermitean operator corresponding to dierent eigenvalues are mutually orthogonal.
Theorem 13 If

f =

f
and if

f[
1
) =
1
[
1
) and

f[
2
) =
2
[
2
), and
1
,=
2
, then
1
[
2
) = 0.
Hermitean f may possess many non-orthogonal eigenvectors with the same eigenvalue. The set of eigenvectors
with a common eigenvalue is called the eigenspace of

f corresponding to eigenvalue .
Theorem 14 The eigenvalues of a Hermitean operator are all real.
Theorem 15 Any eigenspace is a unitary vector space.
Denition 8 Eigenvalues with eigenspaces larger than 1 are called degenerate eigenvalues.
Denition 9 We denote the set of eigenvectors of a Hermitean

f by [
f
n
)
nG
, G Z.
Theorem 16 (Spectral Theorem) For every Hermitean

f there is at least one O.N. basis consisting of
eigenvectors of

f.
f is diagonal in its own eigenbasis:

f =

n
f
n
[f
n
)f
n
[.
Theorem 17 (Extended Spectral Theorem) For every maximal set of commuting Hermitean operators
f
(1)
,

f
(2)
, . . . ,

f
(n)
there is a unique O.N. basis of H consisting of joint eigenvectors of all

f
(1)
, . . . ,

f
(n)
.
We denote a joint eigenvector by [k
1
, k
2
, . . . k
3r
) so,

f
(i)
[k
1
, k
2
, . . . k
3r
) = k
i
[k
1
, k
2
, . . . k
3r
).
We have a resolution of the identity:
1 =
n1,n2,...,n3r
[k
n1
, k
n2
, . . . k
n3r
)k
n1
, k
n2
, . . . k
n3r
[.
3.6 Whats Dierent in -Dimensions
Much of the theory developed above does not apply for innite dimensional vector spaces. In nite-
dimensional spaces, we found that any Hermitean operator has a complete (and orthonormal) set of eigen-
vectors in the space (spectral theorem). This is no longer true in innite-dimensional spaces. We will develop
a dierent version of the spectral theorem for innite-dimensional spaces, in which Hermitean operators will
have a complete set of eigenvectors which do not lie in the space, but which can be orthonormalised in a
dierent sense.
The reason we investigate innite-dimensional representations is that they arise naturally when we consider
the eigenvectors of the position and momentum operators. We would like to use these sets of eigenvectors as
orthonormal bases for our Hilbert space but, as we shall see later, this can only be done in the continuous
sense. Also, while matrix equations are very intuitive, they are usually harder to solve than the dierential
equations that arise in continuous bases.
3.6.1 Vectors [) in -dimensions
For -dimensional spaces,

n=1
n
in general. So far we considered the unitary vector space of
vectors which contain only a nite number of nonzero components. This is called a pre-Hilbert space, denoted
H
p
.
Denition 10 The completion H of a pre-Hilbert space H
p
(obtained by adding the limit of every Cauchy
sequence in H
p
) is called a Hilbert Space
Note: In practise we take H to be the set of all vectors [) such that

n=1
n
< .
Denition 11 A Hilbert space H is called separable if it has a countable O.N. basis [b
n
)
nG
(so that
every [) has a unique expansion [) =

nG
n
[b
n
)). Such a basis is called a Hilbert basis for H.
Note: We know that the CCRs in QM can be represented using a separable Hilbert Space.
Theorem 18 All separable Hilbert Spaces are isomorphic.
Theorem 19 There exist families of non-normalisable vectors [b
parameterized by a parameter
which takes values in a continuous subset of R (e.g. = (1, 4]), which form an improper O.N. basis for
H, so that b
[b
) = (
t
), and every [) H can be expanded uniquely as: [) =
_
()[b
)d.
We have a resolution of the identity with respect to an improper O.N. basis: 1 =
_
[b
)b
[d. A word
about what the above theorem is saying. That the [b
is an improper basis means that the vectors

[b
) are not in the Hilbert space H. That these vectors are non-normalisable means that

n
b
[b
n
)b
n
[b
)
is divergent. These vectors can, however, be continuously orthonormalised so that b
[b
) = (
t
).
Remark: We can also dene a mixed O.N. basis [b
n
)
nG
[b
so that
1 =
nG
[b
n
)b
n
[ +
_
[b
)b
[d.
3.6.2 Operators

A[) in -dimensions
Even if

nG
n
< , it does not necessarily follow that =

A[) H (because

n=1
n
might be
divergent).
Denition 12 The maximal domain D
A
for an operator

A is :
D
A
= [) H :

A[) H
A is called unbounded if D
A
H
A is called bounded if D
A
= H
3.6.3 Scalars [
A[) in -dimensions
In general:
lim
N
lim
M
N
n=1
M
m=1
n

A
n,m
m
,= lim
M
lim
N
N
n=1
M
m=1
n

A
n,m
m
So we must be careful with our denition of

A
.
Denition 13 Assume

A and D
A
are given. Then

A
and D
are dened so that

_
[
_
[) = [
[) [) D
A
Note: In general D
A
,= D
.
3.6.4 Spectral Theorem in -dimensions
Denition 14

U is called unitary if

U
=

U
1
.
Denition 15

f is called self-adjoint if

f =

f
.
Denition 16

f is called symmetric if [
f[) R [) D
f
.
Theorem 20 If H is nite dimensional, then
f symmetric

f self-adjoint.
Theorem 21 If H is -dimensional, then
f self-adjoint =

f symmetric.
Note: There is no universally agreed denition for

f is Hermitean.
Note: In linear algebra,

A symmetric is dened as

A
n,m
=

A
m,n
, rather than as

A
n,m
=

A
m,n
.
Note: For

f to be an observable, we need

f is symmetric. Most observables are also self-adjoint.
Theorem 22 (Spectral Theorem) Assume

f =

f
. Then

f possess a (possibly improper) O.N. basis
[f
n
)
nG
[f
, ( R) such that
f[f
n
) = f
n
[f
n
) and

f[f
) = f
[f
)
and
f
n
[f
m
) =
n,m
, f
[f
) = (
t
) , f
n
[f
) = 0.
Then, we can expand every [) H uniquely as:
[) =
nG
n
[f
n
) +
_
[f
)d
and
n
= f
n
[) ,
= f
[).
Denition 17 The set of proper and improper eigenvalues of

f is called the spectrum of

f, denoted spec(

f).
We can write

f in its own eigenbasis as
f =
nG
f
n
[f
n
)f
n
[ +
_
[f
)d
and, of course, we have the resolution of the identity
1 =
nG
[f
n
)f
n
[ +
_
[f
)d.
3.7 Sharpness of Measurement Predictions
If

f is self-adjoint then by the spectral theorem we can prepare a system so that f = 0 by choosing an
eigenstate (state vector which is an eigenvector of

f).
For measurement-expectations to be real, however, only requires that

f be symmetric. There are important
examples of observables which are symmetric, but not self-adjoint (e.g. in quantum optics). For such
observables, spec(

f) may be the empty set. In this case f would be bounded below by some positive
number f
min
.
3.8 The Position and Momentum Eigenbases
We know from Physical Assumption 5 that if we measure observable x and obtain the result x, the state
collapses to an eigenstate [x) of x, with corresponding eigenvalue x. A similar thing happens for measure-
ments of p. If we make the reasonable assumption that a measurement of x or p can yield any real number,
this suggests that every real number is an eigenvalue of x and of p. So
spec( x) = R , spec( p) = R.
Since x and p are self-adjoint, the spectral theorem applies. It turns out that the eigenvectors of x are
improper, and nondegenerate. So there is an improper orthonormal position basis
[x)
xR
so that
x[x
t
) = (x
t
x) , 1 =
_
R
[x)x[dx.
Similarly, there is an improper orthonormal momentum basis
[p)
pR
.
We can expand every vector in the Hilbert Space H in the position basis:
[) =
_
R
(x)[x),
where (x) = x[) is called the wavefunction of [) in the position basis, or simply the position wavefunction
of [).
We have [) H [) < which, in the position basis, means
_
R
[x)x[)dx =
_
R
(x)(x)dx < .
It is natural to ask how the operator x acts on a position wavefunction (x). This is easy to calculate,
starting with the action of x on the state vector :
x[) = [)
=x[ x[) = x[)
insert the identity on the LHS:
=
_
R
x[ x[x
t
)x
t
[)dx
t
= x[)
=
_
R
x
t
(x x
t
)(x
t
)dx
t
= (x)
=x(x) = (x).
So we nd that x acts on position wavefunctions (x) as multiplication by x. What about p? The compu-
tation is analogous, but relies on the following result (the proof of which we omit as it involves distribution
theory):
Theorem 23
x[ p[x
t
) = i
d
dx
t
(x x
t
).
Note that, even though operators can not be described in an improper basis by a discrete matrix, we will
refer, for example, to x[ p[x
t
) as a matrix element of the p operator in the position basis.
Now,
p[) = [)
=x[ p[) = x[)
=
_
R
x[ p[x
t
)x
t
[)dx
t
= x[)
=
_
R
i
d
dx
t
(x x
t
)(x
t
)dx
t
= (x)
integration by parts gives
=i
d
dx
(x) = (x).
So p acts on position wavefunctions (x) as i
d
dx
.
3.9 Hermite and Fourier Transforms
We claimed that every real number x is an eigenvalue of x, and that the corresponding eigenvectors satisfy
x[x
t
) = (xx
t
). We can prove that this is the case, by solving x[x) = x[x) explicitly in a convenient basis.
Choose the orthonormal basis [n)
nN
in which
n[ x[m) = L
_
_
_
_
_
_
_
_
0

1 0 0
1 0

2 0
0

2 0

3
0 0

3 0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
_
_
_
_
_
_
_
_
n,m
and
n[ x[m) = L
_
_
_
_
_
_
_
_
0
1 0 0
1 0
2 0
0

2 0
3
0 0

3 0
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
.
_
_
_
_
_
_
_
_
n,m
We have n[ x[n) = xn[x). Inserting the identity,
m
n[ x[m)m[x) = xn[x).
If we dene u
n
(x) n[x) (so that [x) =

n
u
n
(x)[n)), this becomes
m
n[ x[m)u
m
(x) = xu
n
(x).
This is a matrix equation, from which we can read-o equations:
L
1u
2
(x) = xu
1
(x)
L
1u
1
(x) +L
2u
3
(x) = xu
2
(x)
.
.
.
L
n 2u
n2
(x) +L
n 1u
n
(x) = xu
n1
(x) n N
which leads to the recursion relation
u
n
(x) =
1
L
n 1
_
xu
n1
(x) L
n 2u
n2
(x)
_
.
This recursion relation can be solved (we will not do it) to give
u
n
(x) =
1
_
2
n1
(n 1)!
e
x
2
4L
2
H
n1
_
x
2L
_
where
H
n
(z) = (1)
n
e
z
2 d
n
dz
n
e
z
2
.
So, we have solved explicitly for the eigenvectors [x) =

nN
u
n
(x)[n). It can be shown that
x[x
t
) =
nN
x[n)n[x
t
) =
nN
u
n
(x)u
n
(x
t
) = (x x
t
).
A natural question is how we convert the wavefunctions between the position basis and the discrete bases;
i.e. how do we convert between (x) and
n
? The answer is found (not surprisingly) by the trick of
inserting the identity, together with the fact that u
n
(x) is self-adjoint:
(x) = x[) =
n
x[n)n[) =
n
u
n
(x)
n
(A)
and conversely
n
= n[x) =
_
R
n[x)x[)dx =
_
R
u
n
(x)(x)dx (B).
(A) and (B) are known as the Hermite Transforms.
Consider the momentum eigenbasis [p)
pR
.
Denition 18

(p) = p[) is known as the momentum wavefunction of [).
So we have that [) H i
_
R

(p)

(p)dp < , and every state [) can be written as
[) =
_
R
[p)p[dp =
_
R
(p)[p)dp.
We have the following result
Theorem 24
x[p) =
1
2
e
ixp
.
A natural question is how to convert between position wavefunctions and momentum wavefunctions. Again,
the answer is calculated by inserting the identity, and using the theorem stated above.
(p) = p[) =
_
R
p[x)x[)dx =
_
R
1
2
e
ixp
(x)dx (C)
and conversely
(x) = x[) =
_
R
x[p)p[)dp =
_
R
1
2
e
ixp

(p)dp (D)
(C) and (D) are known as the Fourier Transform and the Inverse Fourier Transform respectively.
3.10 Equivalence of Representations: the Stone/Von-Neumann Theorem
Any representation we choose must satisfy the CCRs; i.e. must satisfy [ x, p] = i. So, for all [) we need
x p[) p x[) = i[) ().
In the discrete basis [n), we have x, p as matrices L(a
+a) and
i
2L
(a
a) acting on vectors (
1
,
2
, . . .)
T
,
and it is an easy matrix calculation to verify that () is satised. In the position basis [x) we verify (),
by rst noting that
() x[ x p[) x[ p x[) = ix[)
. .
=(x)
()
for any position eigenvector [x). So it suces to check (). We use the (now familiar) trick of inserting
the identity a total of 4 times on LHS of (), and use integration by parts.
x[ x p[)x[ p x[)
=
_
R
_
R
x[ x[x
t
)x
t
[ p[x
tt
)x
tt
[)dx
t
dp
t
_
R
_
R
x[ p[x
t
)x
t
[ x[x
tt
)x
tt
[)dx
t
dp
t
= x
_
i
d
dx
_
(x) +i
d
dx
x(x)
= x
_
i
d
dx
_
(x) +i(x) +x
_
i
d
dx
_
(x)
= i(x)
= ix[)
So () is indeed satised.
We can do computations like the above (much) more quickly, without going through the tedious exercise of
inserting the identity everywhere, by using the rule
x acts on position wavefunctions (x) as multiplication by x, and
p acts on position wavefunctions (x) as i
d
dx
which we proved in Section 3.8.
The representations of [ x, p] = i in terms of matrices, and in terms of operators x, p, dier only in a choice of
basis for the Hilbert space H. The following theorem says that this is so for all representations of [ x, p] = i.
Theorem 25 (Stone/von-Neumann) Every representation of x, p obeying [ x, p] = i as linear maps is ob-
tainable as a choice of basis in the Hilbert space H that we have been working with.
As another example, consider the representations of x and p in the momentum basis [p). Here
x maps momentum wavefunctions

(p) to i
d
dp
(p), and
p maps momentum wavefunctions

(p) to p
(p).
4 Quantum Mechanics in the Schr odinger Picture, and the Posi-
tion Basis
4.1 The Schr odinger Equation Revisited
We are now in a position to view the Schrodinger equation as it is often seen in introductory quantum
mechanics books as the more general equation expressed in a particular choice of basis for H. We will
assume that our Hamiltonians are time-independent (i.e.

H
t
= 0) so that the

H
s
=

H, and we will denote
the Hamiltonians simply by

H.
Recall the Schrodinger equation
i
d
dt
[(t)) =

H[(t)).
For a free particle, with

H =
p
2
2m
, this reads
i
d
dt
[(t)) =
p
2
2m
[(t)).
To write this in the position basis, replace [(t)) with the position wavefunction (x, t), and use the rule for
how x and p acts on position wavefunctions:
i

t
(x, t) =

2
2m
2
x
2
(x, t).
Do the same thing for general Hamiltonian
H =
p
2
2m
+V ( x, p)
where V is a potential energy function. The Schrodinger equation for this Hamiltonian in the position basis
is
i

t
(x, t) =
_

2m
2
x
2
+V
_
x, i

x
__
(x, t).
This is the form of quantum mechanics as Schrodinger found it (Heisenberg found quantum mechanics in
the discrete [n) basis.
4.2 Predicting Observables
Recall that in the Schrodinger picture, measurement-expectations obey
f(t) = (t)[
f( x, p)[(t)).
Using the rule for how x and p acts on position wavefunctions, we write this in the position basis as
f(t) =
_
R
(x, t)

f
_
x, i

x
_
(x, t)dx.
Recall that physical states [) H must obey [) = 1. In the position basis, this is expressed as
_
R
(x)(x)dx = 1.
Example: Consider a particle in 3 dimensions. We have [ x
i
, p
j
] = i
i,j
, i, j 1, 2, 3. We can
use x
1
, x
2
, x
3
as a maximal set of commuting observables and nd a joint eigenbasis [x
1
, x
2
, x
3
).
According to the continuous normalization, we have
x
1
, x
2
, x
3
[x
t
1
, x
t
2
, x
t
3
) = (x
1
x
t
1
)(x
2
x
t
2
)(x
3
x
t
3
)
1 =
___
[x
1
, x
2
, x
3
)x
1
, x
2
, x
3
[dx
1
dx
2
dx
3
.
Any [) can be expanded as
[) =
___
[x
1
, x
2
, x
3
)x
1
, x
2
, x
3
[)dx
1
dx
2
dx
3
=
___
(x
1
x
2
x
3
)dx
1
dx
2
dx
3
where (x
1
x
2
x
3
) is the position wavefunction. Suppose the particle is an electron orbiting a
nucleus. Then the Hamiltonian is
H =
p
2
2m

_
x
2
.
In the Schrodinger picture, and the position basis, i

t
[(t)) =

H[(t)) reads
i

t
(x, t) =
_

2
2m
_

2
x
2
1
+

2
x
2
1
+

2
x
2
1
_

[ x[
_
(x, t). ()
To obtain the dynamics of the electron, we would need to solve () with some initial condition
(x, t
0
) =
0
(x). Once (x
1
, x
2
, x
3
, t) is found, we can predict all expectation values! For
example
p
2
(t) =
___

(x
1
, x
2
, x
3
, t)(i)

x
2
(x
1
, x
2
, x
3
, t)dx
1
dx
2
dx
3
.
4.3 The Green Function Method for solving Schr odinger Equations
The Green Method prevents us from having to solve the Schrodinger equation separately for each initial
condition (x, t
0
) =
0
(x). The method is:
1. Solve
i

t
G(x, x
t
, t) =

H
_
x, i

x
_
G(x, x
t
, t)
with initial condition
G(x, x
t
, t
0
) = (x x
t
).
2. For any initial condition
0
(x) we obtain
(x, t) =
_
R
G(x, x
t
, t)
0
(x
t
)dx
t
.
The function G(x, x
t
, t) is called the Green function, and sometimes called the propagator.
Proof that Greens Method Works:
i

t
_
R
G(x, x
t
, t)
0
(x
t
)dx
t
=
_

H
_
x, i

x
_
G(x, x
t
, t)
0
(x
t
)dx
t
=

H
_
x, i

x
__
G(x, x
t
, t)
0
(x
t
)dx
t
=

H
_
x, i

x
_
(x, t)
Also,
(x, t
0
) =
_
G(x, x
t
, t
0
)
. .
(xx
0
(x
t
)dx
t
=
0
(x)
Now, recall the time-evolution operator

U(t). The problem of solving the Schrodinger equation
i
d
dt
[(t)) =

H[(t))
[(t
0
)) = [
0
)
becomes that of solving
i
d
dt
U(t) =

H

U(t)
U(t
0
) = 1
Then we have [(t)) =

U(t)[
0
). Inserting everywhere the identity resolved in the position basis we would
obtain
G(x, x
t
, t) = x[
U(t)[x
t
).
So we see that G(x, x
t
, t) are the matrix elements of the time-evolution operator

U(t) in the position basis.
It is natural to wonder what is the physical meaning of the propagator. This question takes us to a formulation
of quantum mechanics due to Feynman.
5 Feynmans Formulation of Quantum Mechanics
For convenience, choose the Schrodinger picture. At time t
i
suppose the initial state of a quantum system
is [). Then [(t)) =

U(t)[).
Suppose at time t
f
we use a measurement device which outputs 1 if the system is found in state [) and
outputs 0 otherwise. This device is described by a self-adjoint operator

P
])
having eigenvalues 0 and 1. The
eigenspace corresponding to eigenvalue 1 is spanned by [). All vectors [) which are orthogonal to [) are
in the eigenspace corresponding to eigenvalue 0. The above observations tell us that

P
])
is the projector
P
])
= [)[.
Decompose the nal state as [(t
f
)) = [) +[) such that [)[). We need
[) = 0 = [(t
f
)) [)
. .
=1
= = [(t
f
))
So
[(t
f
)) = [) +[)
Now consider the wavefunction collapse resulting from the measurement. If the measurement outputs 1,
then the new state immediately after t
f
is [). If the measurement outputs 0, the state immediately after
t
f
is k[), where k is chosen so the state is normalised.
Now consider an ensemble of identically prepared systems, and running the experiment separately on each
system. In general, we will have many 0s and many 1s in the collection of measurement outputs. The mean
of these numbers is equal to the probability of nding [) at time t
f
on one particular run of the experiment.
But, from the theory we have developed, we know this probability can be calculated as
P
])
(t
f
) = (t
f
)[
P
])
[(t
f
)) = (t
f
)[[)[[(t
f
)) = [[(t
f
))[
2
.
The quantity [(t
f
)) is called a probability amplitude or transition amplitude. If [) is improper (for
example [) = [x)), then x[(t)) = (x, t) is a probability density function and, for example,
_
x
b
xa
[(x, t)[
2
dx =
_
x
b
xa
(x, t)(x, t)dx

= prob. of nding the particle in the interval (x
a
, x
b
).
Recall the time-evolution operator

U(t). We have [(t)) = [U(t)[), and so we see that the probability
amplitudes are just the matrix elements of the time-evolution operator.
Recall the Green function G(x, x
t
, t) = x[
U(t)[x
t
). In light of the above, we see that G(x, x
t
, t) is the
probability amplitude for a particle to propagate from x
t
to x.
Now, Feynman compared classical probability theory to the laws of probability dictated by quantum mechan-
ics. Consider a classical system going from state A to state B with some probability P(A B). Suppose
there are n possible paths from A to B, each passing through one of the n states C
1
, . . . C
n
. Then the
(classical) law of conditional probability dictates that
P(A B) =
n
i=1
P(A C
i
)P(C
i
B).
Now consider an isolated, unobserved quantum system with Hamiltonian

H, starting in state [
0
) at time
t
0
and evolving until time t
f
. Consider the probability amplitude [(t
f
)) for the state being [) at time
t
f
. Suppose that to get to state [), the system would have to evolve through one of n states [b
1
), . . . , [b
n
)
at time t
1
(t
0
< t
1
< t
f
). Then we have
[(t
f
)) = [
U(t
f
, t
0
)
. .
e
t
f
t
0
i

H
[
0
) = [
U(t
f
, t
1
)
. .
e
t
f
t
1
i

H
U(t
1
, t
0
)
. .
e
t
1
t
0
i

H
[
0
)
=
n
i=1
[
U(t
f
, t
1
)[b
i
)b
i
[
U(t
1
, t
i
)[
0
).
Note that [
U(t
f
, t
1
)[b
i
) is the probability amplitude for the evolution [b
i
) [) and b
i
[
U(t
f
, t
1
)[
0
) is
the probability amplitude for the evolution [
0
) [b
i
). Feynman observed that in quantum mechanics,
the law of conditional probabilities applies to probability amplitudes, instead of simple probabilities. This is
signicant, because probability amplitudes can have phase factors (so in particular, the could be negative!).
This is why quantum interference can happen.
6 Density Matrices and Mixed States
6.1 Density Matrices and Mixed States
So far, we have assumed that a state [) of a system can be known with certainty. Then our expectation
values are f = [
f[). A more realistic situation is that we do not know the state of the system with
certainty, but rather know that it is in one of the states [b
1
), . . . , [b
n
) with respective probabilities p
1
, . . . , p
n
.
Without loss of generality we can assume that the [b
i
) form an O.N. basis (otherwise we could extend the
set by adding new states with corresponding probabilities equal to zero), and that

n
i=1
p
i
= 1 (we know
for certain the system is in one of the n possible states). In this case, the expectation values are computed
through
f =
n
i=1
p
i
b
i
[
f[b
i
).
Note that the uncertainty in f now has two sources:
a) our uncertainty about which state the system is in, and
b) the quantum-mechanical uncertainty.
Denition 19 We say that a system is in a Mixed State if more than one p
i
is nonzero. Otherwise the
system is said to be in a Pure State.
Denition 20
=
i
p
i
[b
i
)b
i
[
is called the Density Matrix for the system.
Denition 21 The trace of an operator

A is
Tr(

A) =
n
c
n
[
A[c
n
)
where [c
n
) is any O.N. basis for the space of

A.
Note: is self-adjoint, and Tr( ) = 1 (proof follows from cyclicity of trace).
Now, the equation f =

i
p
i
b
i
[
f[b
i
) becomes
f = Tr(
f).
Suppose we choose a basis [c
n
). Then f = Tr(
f) becomes
f =
n,m
c
n
[ [c
m
)c
m
[
f[c
n
).
We can dene coecient matrices for and

f in the [c
n
) basis:

n,m
= c
n
[ [c
m
) ,

f
n,m
= c
n
[
f[c
m
).
Then f =

n,m

n,m

f
n,m
.
Remarks:
We can choose, equivalently, improper or mixed O.N. bases in all of the above.
For a mixed state, is non-diagonal in almost all bases. But, because is self-adjoint, there always
exists a basis in which it is diagonal (the eigenbasis). Only then does have the form
=
n
p
n
[b
n
)b
n
[.
In general bases we have
=
n,m

n,m
[c
n
)c
m
[.
We know that the time-evolution of Schrodinger states [(t)) obeys the Schrodinger equation. This leads us
to a result for the time-evolution of density matrices.
Theorem 26 The time-evolution of the density matrix for a closed system follows the von-Neumann eqution:
i
d
dt
(t) =
_
(t),

H
_
.
Proof: Write
(t
0
) =
n
p
n
[b
n
)b
n
[.
From the Schrodinger equation:
i
d
dt
[b
n
(t)) =

H(t)[b
n
(t))
and thus
i
d
dt
b
n
(t)[ = b
n
(t)[
H(t) (since

H
(t) =

H(t)).
At time t, each [b
n
(t
0
)) has evolved to [b
n
(t)). So (t
0
) has evolved to
(t) =
n
p
n
[b
n
(t))b
n
(t)[ ().
Apply i
d
dt
to both sides of ():
i
d
dt
(t) = i
d
dt
n
p
n
[b
n
(t))b
n
(t)[
=
n
p
n
_
i
d
dt
([b
n
(t))b
n
(t)[)
_
=
n
p
n
_
i
_
d
dt
[b
n
(t))
_
b
n
(t)[ +i[b
n
(t))
_
d
dt
b
n
(t)[
__
=
n
p
n
_
i
_
d
dt
[b
n
(t))
_
b
n
(t)[ [b
n
(t))
_
i
d
dt
b
n
(t)[
__
=
n
p
n
_
H(t)[b
n
(t))b
n
(t)[ [b
n
(t))b
n
(t)[
H(t)
_
=

H(t) (t) (t)

H(t)
=
_
(t),

H
_

Note: We also have (t) = U(t)
U(t)
.
Review: changing bases for operators
How do we convert from the [c
n
) basis to a [d
n
) basis? We use the old trick of inserting
the identity resolved in the new basis:
=
n,m,r,s

n,m
[d
r
)d
r
[c
n
)c
m
[d
s
)d
s
[
=
r,s

r,s
[d
r
)d
s
[
where
r,s
=

n,m

n,m
d
r
[c
n
)c
m
[d
s
).
6.2 Example: quantum system S in a heat bath
Suppose a quantum system o is in a heat bath at temperature T. According to statistical physics, the
probability for a subsystem of a heat bath to be at energy E is proportional to
e
E
kT
where k = 1.38 10
23
J/Kelvin is the Boltzman Constant.
Assume the system o has nondegenerate energy eigenvalues E
n
. Then
prob(E
n
) = Ne
En
kT
,
where N =
1
m
e
Em
kT
. In other words, the probability is
p
n
= prob(E
n
) =
e
En
kT
m
e
Em
kT
.
Choose the energy eigenbasis of the system. Then the density matrix is
=
n
p
n
[E
n
)E
n
[.
Now, since E
m
are the eigenvalues of the Hamitonian

H, we have that

H is diagonal in the basis of its
eigenvectors [E
m
), and so
e

H
kT
=
m
e
Em
kT
[E
m
)E
m
[
Using this observation, and the denition of Trace, we get
Tr(e

H
kT
) =
n
E
n
[e

H
kt
[E
n
)
=
n
E
n
[
m
e
Em
kT
[E
m
)E
m
[E
n
)
=
n,m
E
n
[E
m
)e
Em
kT
E
m
[E
n
)
=
n
e
En
kT
.
Therefore we can write the density matrix for the system in the Heat bath as
=
1
Tr(e

H
kT
)
e

H
kT
.
6.3 Decoherence and Mixed States
Consider a quantum system S
c
in a pure state, evolving in time. If S
c
is completely isolated from its
environment, then the time-evolution is unitary, and the state of S
c
remains pure. If S
c
is allowed to interact
with the environment at time t
1
, then the interaction carries away some information about the state of
S
c
, and in this sense constitutes a measurement of S
c
(whether or not this interaction was intended as a
controlled measurement). Thus the state of S
c
collapses as a result of the interaction. Since we dont learn
the outcome of this measurement, we dont know which pure state S
c
is in, and so S
c
is left in a mixed
state
Sc
at time t
1
. Suppose we know that the interaction with the environment was, in particular, with
a subsystem of the environment which we will label S
b
. Imagine S
b
is a photon which comes in from the
environment, bounces o S
c
, and returns to the environment, carrying with it some information about the
state of S
c
. Let us investigate how to compute the mixed state
Sc
.
Assume S
c
has N particles. Assume S
c
is in a pure state [
(Sc)
) at time t
0
. We have x
(j)
i
, p
(j)
i
, where
i 1, 2, 3, j 1, 2, . . . , N. The system S
c
has some Hamiltonian

H
(Sc)
_
x
(j)
i
, p
(j)
i
_
. Choose a maximal
set of 3N commuting observables (self-adjoint operators), say

Q
(1)
, . . . ,

Q
(3N)
. Find a joint eigenbasis of all
the

Q
(j)
spanning H
Sc
:
[q
(1)
n1
, q
(2)
n2
, . . . , q
(3N)
n3N
),
where q
(j)
ni
are the eigenvalues of

Q
(j)
. The pure state of S
c
can be expanded in this basis as
[
(Sc)
) =
n1,...,n3N
(Sc)
n1,...,n3N
[q
(1)
n1
, q
(2)
n2
, . . . , q
(3N)
n3N
)
and any observable

f can be expanded as
f =
n
1
,...,n
3N
,
n
1
,..., n
3N
f
n1,...,n3N, n1,..., n3N
[q
(1)
n1
, q
(2)
n2
, . . . , q
(3N)
n3N
)q
(1)
n1
, q
(2)
n2
, . . . , q
(3N)
n3N
[.
Expectation values are calculated as
f =
(Sc)
[
f[
(Sc)
)
=
n
1
,...,n
3N
,
n
1
,..., n
3N
(Sc)
n1,..., n3N
f
n1,...,n3N, n1,..., n3N
(Sc)
n1,...,n3N
.
Now, consider the subsystem of the environment, S
b
. Assume S
b
has M particles. We have operators
x
(j)
i
, p
(j)
i
, where i 1, 2, 3, j N + 1, N + 2, . . . , N +M, and a Hamiltonian

H
(S
b
)
_
x
(j)
i
, p
(j)
i
_
. Choose
a maximal set of 3M observables, say

R
(1)
, . . . ,

R
(3M)
. Obtain a joint orthonormal eigenbasis of all

R
(i)
:
[r
(1)
m1
, r
(2)
m2
, . . . , r
(3M)
m3M
).
Assume S
b
is in a pure state [
(S
b
)
) H
S
b
, and expand as
[
(S
b
)
) =
m1,...,m3M
(S
b
)
m1,...,m3M
[r
(1)
m1
, r
(2)
m2
, . . . , r
(3M)
m3M
).
Now, consider the total system S
tot
= S
c
S
b
. We have operators x
(j)
i
, p
(j)
i
, where i 1, 2, 3, j
1, . . . , N + M. Now all j 1, . . . , N + M can occur in the Hamiltonian

H
(tot)
_
x
(j)
i
, p
(j)
i
_
. Choose a
maximal set of 3(N +M) commuting self-adjoint operators

Q
(1)
, . . . ,

Q
(3N)
,

R
(1)
, . . . ,

R
(3M)
. Again, obtain
a joint orthonormal eigenbasis
[q
(1)
n1
, . . . , q
(3N)
n3N
, r
(1)
m1
, . . . , r
(3M)
m3M
).
A general pure state of S
tot
is
[
(tot)
(t)) =
n1,...,m3M
(tot)
n1,...,m3M
(t)[q
(1)
n1
, . . . , r
(3M)
m3M
).
Now, given that at time t
0
, S
c
and S
b
are respectively in pure states [
(Sc)
) and [
(S
b
)
), then the state of
S
tot
has the components
(tot)
n1,...,m3M
(t
0
) =
(Sc)
n1,...,n3N
(S
b
)
m1,...,m3M
.
Consider the expectation of an observable

f of S
c
at time t
0
:
f =
(tot)
(t
0
)[
f[
(tot)
(t
0
))
=
n
1
,...,m
3M
,
n
1
,..., m
3M
(tot)
n1,..., m3M
(t
0
)

f
n1,..., n3N,n1,...,n3N
(tot)
n1,...,m3M
(t
0
)
=
_
_
_
n
1
,...,n
3N
,
n
1
,..., n
3N
(Sc)
n1,..., n3N
f
n1,..., n3N,n1,...,n3N
(Sc)
n1,...,n3N
_
_
_
_
_
_
m
1
,...,m
3M
,
m
1
,..., m
3M
(S
b
)
m1,..., m3M
(S
b
)
m1,...,m3M
_
_
_
. .
=1
=
(Sc)
[
f[
(Sc)
).
The evolution according to the joint Hamiltonian

H
(tot)
will destroy the product form of the state [
(tot)
(t))
and we obtain, at time t
1
, the general components
(tot)
n1,...,m3M
(t
1
).
Denition 22 If
(tot)
is a product, it is called an unentangled, or separable state. Otherwise, it is called
an entangled state.
Consider an observable

f of S
c
. We have
f
n1,..., m3M,n1,...m3M
=

f
n1,..., n3N,n1,...n3N
. .
action on 1c
m1,m1
m2,m2

m3M,m3M
. .
action on 1
b
.
At time t
1
, the expectation value of the observable f of S
c
is:
f =
(tot)
(t
1
)[
f[
(tot)
(t
1
))
. .
pure-state calculation in S
(tot)
=
n
1
,...,n
3N
,m
1
,...,m
3M
,
n
1
,..., n
3N
, m
1
,... m
3M
(tot)
n1,..., m3M
(t
1
)

f
n1,..., n3N,n1,...,n3N
(
m1,m1

m3M,m3M
)
(tot)
n1,...,m3M
(t
1
)
=
n
1
,...,n
3N
,
n
1
,..., n
3N
f
n1,..., n3N,n1,...,n3N

(c)
n1,...,n3N, n1,..., n3N
(t
1
)
. .
mixed-state calculation in S
(c)
where
(c)
n1,...,n3N, n1,..., n3N
(t
1
) =
m1,...,m3M
(tot)
n1,..., n3N,m1,...m3M
(t
1
)
(tot)
n1,...,n3N,m1,...m3M
(t
1
).
7 Compositions of Several Systems
Consider two quantum systems S
1
and S
2
. Observables of S
1
contain only x and p from S
1
acting on
[) H
1
. Observables of S
2
contain only x and p from S
2
acting on [) H
2
. Observables of the total
system S
tot
= S
1
S
2
may contain all x, p of S
1
and S
2
, acting on H
tot
. For example
H
(tot)
=
_
p
(1)
_
2
2m
1
+
_
p
(2)
_
2
m
2
+V
_
x
(1)
x
(2)
_
.
We call V the interaction term.
A natural question is how do we construct H
tot
from H
1
and H
2
? A partial answer is that we often have
H
tot
= H
1
H
2
, where denotes the tensor product, dened as follows.
Denition 23 As a set, dene H
1
H
2
= span[) [) : [) H
1
, [) H
2
with the additional laws:
I)
([
1
) +[
2
)) [) = [
1
) [) +[
2
) [)
[) ([
1
) +[
2
)) = [) [
1
) +[) [
2
)
II)
([)) [) = ([) [)) = [) ([))
and with conjugation map :
: [) [) [ [ (H
1
H
2
)
(extended linearly for linear combinations).

It can be shown that H
1
H
2
is a Hilbert space. In particular, if [b
n
)
n
and [c
m
)
m
are orthonormal
bases of H
1
and H
2
respectively, then [b
n
) [b
m
)
n,m
is an orthonormal basis of H
1
H
2
.
Assume

f is polynomial in the x and p of S
1
. Dene

f to act on [) [) H
1
H
2
as
f : [) [) (

f[)) [).
We write

f on the acting on the whole system as

f 1.
Note: All the CCRs hold on H
1
H
2
, and respects Hermiteicity.
By the Stone/Von-Neumann uniqueness theorem, H
tot
H
1
H
2
. Thus every [) H
(tot)
can be written
as
[) =
n,m
n,m
[b
n
) [c
m
).
Denition 24 [) H
tot
is called unentangled (or separable) if it can be written as
[) = [
1
) [
2
)
for [
1
) H
1
, [
2
) H
2
. Otherwise, [) is called entangled.
Assume [) = [
1
) [
2
) (i.e. [) is unentangled). Now let

f be any observable on the rst subsystem.
Consider the expectation of this observable on the whole system; i.e. the expectation of

f 1.
f = [(

f 1)[)
= (
1
[
2
[)(

f 1)([
1
) [
2
))
=
1
[
f[
1
)
2
[
2
)
=
1
[
f[
1
).
But
1
[
f[
1
) is exactly the expectation we would predict for

f acting on system S
1
alone. Since this is true
for any observable

f, system S
1
is in the pure state [
1
). We have shown the following:
Fact 1 [) H
tot
is unentangled i the states of subsystems S
1
and S
2
considered on their own are pure
states.
If, however, [) is entangled, then the state of system S
1
is a mixed state whose density operator is obtained
by the partial trace (i.e. we trace over the basis of the second system; sometimes referred to as tracing-out
the second system). For the remainder of this section, we show that the partial trace operation gives the
right result.
Let [b
n
) be an orthonormal basis of subsystem S
1
, and let [c
m
) be an orthonormal basis of subsystem
S
2
. Suppose [) is a general (possibly entangled) pure state on the total system:
[) =
n,m
n,m
[b
n
) [c
m
).
The subsystem S
1
is in a mixed state with density operator
(1)
i
f = Tr(

f
(1)
) ()
for arbitrary Hermitean observables

f of S
1
.
We compute the left-hand side (LHS) and right-hand side (RHS) of () separately, and make sure they are
equal.
RHS:
By denition of the reduced density operator and partial trace, we have

(1)
= Tr
(2)
([)[)
= Tr
(2)
_
_

n,m, n, m
n,m
n, m
([b
n
) [c
m
))(b
n
[ c
m
[)
_
_
=
n,m,
n, m
n,m
n, m
[b
n
)b
n
[c
r
[c
m
)c
m
[c
r
)
=
n,m,
n, m
n,m
n, m
[b
n
)b
n
[
r,m
m,r
=
n, n,m
n,m
n,m
[b
n
)b
n
[.
Now, (

f
(1)
) is an operator on system S
1
alone, so by denition of Trace we have
Tr(

f
(1)
) =
u
b
u
[
_
_
f
n, n,m
n,m
n,m
[b
n
)b
n
[
_
_
[b
u
)
=
u,m,n, n
n,m
n,m
b
u
[
f[b
n
)b
n
[b
u
)
=
u,m,n, n
n,m
n,m
b
u
[
f[b
n
)
n,u
=
m,n, n
n,m
n,m
b
n
[
f[b
n
). (1)
LHS:
To compute the LHS, we consider

f acting on the total system as

f 1.
f = [(

f 1)[)
=
n,m, n, m
n, m
n,m
(b
n
[ c
m
[)(

f 1)([b
n
) [c
m
))
=
n,m, n, m,r,s,u,v
n, m
n,m
(b
n
[ c
m
[)([b
r
) [c
s
))(b
r
[ c
s
[)(

f 1)([b
u
) [c
v
))(b
u
[ c
v
[)([b
n
) [c
m
))
=
n,m, n, m,r,s,u,v
n, m
n,m
(b
n
[b
r
)b
r
[
f[b
u
)b
u
[b
n
)) (c
m
[c
s
)c
s
[c
v
)c
v
[c
m
))
=
n,m, n, m,r,s,u,v
n, m
n,m
(
n,r
b
r
[
f[b
u
)
u,n
) (
m,s
s,v
v,m
)
=
m,n, n
n,m
n,m
b
n
[
f[b
n
). (2)
Since (1) = (2), we have LHS=RHS, and so we are done. We have thus shown the following.
Fact 2 For a general pure state [) of the joint system S
1
S
2
, the subsystem S
1
considered alone is in a
(generally mixed) state whose density operator is

(1)
= Tr
(2)
([)[).
8 Identical Particles
8.1 Bosons and Fermions
Consider a systemS. Choose a maximum set of commuting observables

Q
(1)
, . . . ,

Q
(N)
with joint orthonormal
eigenbasis [q
(1)
n1
, . . . , q
(N)
nN
). We will write

Q and [q
n
) for short. Every [) H can be written [) =
n
[q
n
).
Consider having two identical such systems. The total system is in a state [) H
tot
. We know from the
previous section that H
tot
H H. Suppose we measure

Q for both systems and obtain measurement
outcomes q
n
, q
m
. If we assume that the systems are truly indistinguishable, there is no way of knowing
which system was measured in q
n
and which was measured in q
m
. All that we can know is that one system
was measured in each of the two. In fact, it is not even meaningful to ask which system was measured in
q
n
?, since the measurement was of a joint observable of the overall system. Now consider the state of the
system immediately after the collapse. Any state of the form
[) = [q
n
) [q
m
) +[q
m
) [q
n
)
would yield q
n
, q
m
on an immediate remeasurement. So the question is, which one of these states does
nature select? Does nature conne itself to a xed choice of and ? Before answering the question, note
that an overall phase e
i
never matters, since
f = [e
i

fe
i
[) = [
f[)
and so [) and e
i
[) describe the same physical state. Now, to answer the question above, it turns out that
nature always makes one of two choices for the relative phase:
(I) = (Bosonic case)
(II) = (Fermionic case).
So, two identical Bosons collapse into
[) =
1
2
([q
n
) [q
m
) +[q
m
) [q
n
))
and two identical Fermions collapse into
[) =
1
2
([q
n
) [q
m
) [q
m
) [q
n
)).
Bosons are particles which have integer values of angular momentum, and Fermions are particles which have
half-integer values of angular momentum.
Consider carefully the situation for Fermions. Suppose the outcome of the measurement was q
n
, q
n
. Then
the two identical Fermions will collapse into the state
[) =
1
2
([q
n
) [q
n
) [q
n
) [q
n
)) = 0.
But 0 is not a physical state, and so we see that identical Fermions can never be in the same state.
Note: If space had fewer than 3 dimensions, then any relative phase would be possible. Such a space would
give rise to Anyons (e.g. quantum Hall eect).
Remark:

H treats identical subsystems identically. One can show that symmetry (or anti-symmetry) is
preserved under time-evolution. This means that H
tot
HH for bosons and fermions.
8.2 Special Behaviour of Bosonic and Fermionic Systems
Consider a system S. Suppose a maximum set of commuting observables is

H,

Q. Assume the eigenvalues of
H form a discrete set E

m
, and that

Q has only two eigenvalues q
1
, q
2
. The joint orthonormal eigenbasis
is [E
n
, q
i
)
nN,i1,2]
. Now, suppose we put the system into a heat bath. The density matrix of the system
is
=
nN,i1,2]
n
[E
n
, q
i
)E
n
, q
i
[ ,
n
=
1
N
e
En
kT
.
Suppose at time t
1
we measure

H and

Q. The probability for nding state [E
n
, q
1
) is proportional to e
En
kT
,
and the probability for nding state [E
n
, q
2
) is also proportional to e
En
kT
. In both cases, the probability is
independent of i. If we nd E
n
, then the probability of nding q
1
is
1
2
.
Now, suppose we have two similar systems in the heat bath together, and at time t
1
measure

H and

Q
of each system. Suppose we obtain energy E
n
for both systems (i.e. the measurement result is E
n
, E
n
).
Consider the probability for nding the q
i
to be dierent, verses the probability for nding the q
i
the same.
There are three cases.
Case 1: Distinguishable systems
We are sure to nd one of 4 states:
[E
n
, q
1
) [E
n
, q
1
) , [E
n
, q
1
) [E
n
, q
2
)
[E
n
, q
2
) [E
n
, q
1
) , [E
n
, q
2
) [E
n
, q
2
)
All four states have equal energy 2E
n
, and so are equally probable. So,
the probability to nd the q
i
equal is
1
2
, and
i
unequal is
1
2
.
Case 2: Identical Bosons
We are sure to nd one of 3 states:
[E
n
, q
1
) [E
n
, q
1
) , [E
n
, q
2
) [E
n
, q
2
)
1
2
([E
n
, q
1
) [E
n
, q
2
) +[E
n
, q
2
) [E
n
, q
1
))
All three states have equal energy, and so are equally probable. So,
i
equal is
2
3
, and
i
unequal is
1
3
.
The probability for Bosons to be in the same state is enhanced. This is what makes Bose-Einstien
condensates (many Bosons in the same state) possible.
Case 3: Identical Fermions
We are sure to nd the state:
1
2
([E
n
, q
1
) [E
n
, q
2
) [E
n
, q
2
) [E
n
, q
1
))
So,
i
equal is 0, and
i
unequal is 1.
This is known as the Pauli exclusion principle.
9 Angular Momentum
9.1 Generators of Symmetries
We begin by examining how which the (linear) momentum operator p can be thought to generate trans-
lations of the position operator. It can be shown that
e
ib p
xe
ib p
= x +b b R
by expanding the second exponential in terms of a power series, and then commuting x through p
n
.
Now, consider the action of x on the state e
ib p
[x).
x
_
e
ib p
[x)
_
= e
ib p
ib p
x
_
e
ib p
[x)
_
= e
ib p
_
e
ib p
xe
ib p
_
[x)
= e
ib p
( x +b)[x)
= ( x +b)
_
e
ib p
[x)
_
and so we see that e
ib p
[x) is an eigenvector of x, with eigenvalue ( x +b). This eigenvector is [x +b). So we

have found
e
ib p
[x) = [x +b).
In the position basis,
e
ib(i)
d
dx
(x) = (x +b)
= e
b
d
dx
(x) = (x +b)
= (x +b) = 1 +
_
d
dx
(x)
_
b +
1
2
_
d
2
dx
2
(x)
_
b
2
+. . . (Taylor series).
Consider x
2
. We have
e
ib p
x xe
ib p
= e
ib p
xe
ib p
e
ib p
x
e
ib p
= ( x +b)
2
,
and by induction, e
ib p
x
n
e
ib p
= ( x +b)
n
.
Since [ p, p] = 0, we clearly have e
ib p
pe
ib p
= p. Putting these observations together, for any Hamiltonian

which is polynomial in x, p we have e
ib p

H( x, p)e
ib p
=

H( x +b, p).
For a Hamiltonian

H to be translation invariant means
H( x, p) =

H( x +b, p) b R.
We have
H( x +b, p) =

H( x, p) = e
ib p
ib p

H( x, p).
We also know from above that
H( x +b, p) = e
ib p

H( x, p)e
ib p
.
Putting these together we get
e
ib p
ib p

H( x, p) = e
ib p

H( x, p)e
ib p
,
which implies that [ p,

H] = 0. By Heisenbergs equation, this implies that

p = 0. So for a Hamiltonian
H which is translation invariant, the generator p of the translation symmetry is conserved. This is an
instance of a fact sometimes referred to as the Noether Theorem; namely, that every symmetry of the
Hamiltonian implies a conserved quantity.
Now, keeping in mind this picture of operators behaving as generators of symmetries, we move onto angular
momentum.
9.2 Orbital Angular Momentum
Denition 25 Orbital angular momentum in three dimensions is dened as follows.
L
1
= x
2
p
3
x
3
p
2
,

L
2
= x
3
p
1
x
1
p
3
,

L
3
= x
1
p
2
x
2
p
1
.
The following notation is often used.
L
i
=
3
i,j,k=1
i,j,k
x
j
p
k
where
1,2,3
= 1, and is totally antisymmetric (so, for example,
2,1,3
= 1).
In the same sense that p generates translations of x, the orbital angular momentum

L generates rotations
of x and p. Consider, for example, a rotation

L = (0, 0,

L
3
) by angle about the z-axis:
x
1
x
1
cos + x
2
sin
x
2
x
1
sin + x
2
cos
x
3
x
3
x Rx.
One can show that
e
i
L
3
x
1
e
L
3
= x
1
cos + x
2
sin
e
i
L
3
x
2
e
L
3
= x
1
sin + x
2
cos
e
i
L
3
x
3
e
L
3
= x
3
and therefore
e
i
L
3
[x) = [Rx)
where R is the rotation matrix.
Similarly,
e
i
L
3
[p) = [Rp).
Note that this gives the Taylor expansion analogue:
e
x1

x
2
x2

x
1
(x) = (Rx).
For

H to be rotation invariant means
H( x, p) =

H(R x, R p).
This means
e
i
L
3

H( x, p)e
L
3
=

H( x, p)
which gives [
L
3
,

H] = 0, and so

L
3
= 0. So for a Hamiltonian which is rotation invariant, the generator

L
3
of the rotation symmetry is conserved.
For example, consider the Hamiltonian for the Hydrogen atom
H =
p
2
2m
+

_
x
2
.
This Hamiltonian is rotation invariant, because x
2
, p
2
are scalars for rotations.
9.3 Angular Momentum and Identical Particles
Recall that when we studied identical particles we said that bosons are particles having integer values for
angular momentum, and fermions are particles having half-integer values for angular momentum. Suppose
we have a boson, and assume that

L
3
[) = m[), where m Z. Then
2 rotated [) is e
2i
L
3
[) = e
2im
[) = [).
On the other hand, suppose we have a fermion, and assume that

L
3
[) =
m
2
[), where m Z. Then
2 rotated [) is e
2i
L
3
[) = e
2im
2
[) = e
im
[) = [).
This is the underlying reason why swapping two bosons leaves the state unchanged, but swapping two
fermions introduces a (-1) factor.
9.4 Orbital Angular Momentum v.s. Spin
We proceed by nding joint eigenvectors of

L
2
and L
3
:
L
3
[, , ) = [, , ) (I)
L
2
[, , ) =
2
[, , ) (II)
where are the eigenvalues with respect to the rest of a maximal set of commuting observables. For simplicity,
we will omit writing .
One approach to solving (I) and (II) is to convert to a position representation:
,
(x) = x[, ). From
Denition 25, and the rule for converting to the position basis, (I) and (II) become
_
ix
1
x
2
+ix
2
x
1
_
,
(x) =
,
(x) (I
t
)
3
i=1
( )
2
,
(x) =
2
,
(x) (II
t
)
These two partial dierential equations can be solved, giving the result of spherical harmonics:
,
(x) = Y
(x)
with eigenvalues
= j(j + 1) where j 0, 1, 2, . . .
= j, j + 1, . . . , 0, 1, 2, . . . , j
Notice that we only get integer solutions. This means we have only found solutions corresponding to bosons.
But we know that fermions exist, so our solution must have missed something. The calculation didnt miss
any solutions... what went wrong was that we used the denition for orbital angular momentum when we
set up (I
t
), (II
t
). That is, we assumed
L
i
=
3
j,k=1
i,j,k
x
j
p
k
.
It turns out that if we only make the weaker assumption that
[
L
i
,

L
j
] = i
3
k=1
i,j,k
L
k
()
the resulting

L
i
can be shown to generate of rotations of the vector (
L
1
,

L
2
,

L
3
). For example, the rotation
generated by

L
3
is:
e
i
L
3

L
1
e
L
3
=

L
1
cos +

L
2
sin
e
i
L
3

L
2
e
L
3
L
1
sin +

L
2
cos
e
i
L
3

L
3
e
L
3
=

L
3
So, to be general, we should study (2) directly.
Note: A direct consequence of () is
_
L
2
,

L
i
_
= 0.
Remark: (2) is a special case of a Lie Algebra.
Denition 26
L
+
=

L
1
+i
L
2
=

L
1
i
L
2
L
has the following easily veriable properties.

[
L
+
,

L
] = 2
L
3
[
L
3
,

L
] =
L,

L
] = 0 (I)
L
2
=

L
L
+
+

L
3
+
L
3
(II)
Now assume [, ) is a joint eigenvector of L
2
and

L
3
.
Proposition 1

L
[, ) is proportional to [, 1).
Check:
L
2
(
[, )) =

L
L
2
[, )) by (I)
=
2
(
[, ))
L
3
(
[, )) = (
L
3
)[, ) = ( 1)(
[, ))
Now, we calculate |
[, )| (we set = 1 for simplicity).

|
[, )|
2
= , [
[, )
= , [(
L
2

L
2
3

L
3
[, )
=
2
(because , [, ) = 1)
In general, we have
|(
)
n
[, )|
2
=
n1
=0
( ( )
2
( )).
Clearly, this may become negative for suciently large choices of n. Since any vectors norm is zero or
positive, the assumption that [, ) is a joint eigenvector of

L
2
and

L
3
can be true only if and are such
that the product on the RHS never becomes negative. So and must be such that for some n, a factor
in the product equals zero. In this way, the application of high powers on

L
on [, ) merely leads to the

zero vector in the Hilbert space for suciently large n.
This means that the possible eigenvalues , are such that there exist nonnegative integers ,
t
so that
( +)
2
( +) = 0 (1)
(
t
)
2
+ (
t
) = 0 (2)
Subtracting the two equations yields:
2 2
t
2
+
t
2
2 +
t
= 0
i.e.
2( +
t
+ 1) + ( +
t
+ 1)(
t
) = 0.
Clearly, for this equation to hold, the eigenvalues of must be of the form
=

t
2
and substituting into (2), the eigenvalues must be of the form
=

t
+
2
_
t
+
2
+ 1
_
.
We allowed and
t
to be any two nonnegative integers. Dene j =

+
2
. We can now read-o that joint
eigenvectors [, ) of

L
2
and

L
3
have these (and only these) eigenvalues:
= j(j + 1) where j 0,
1
2
, 1,
3
2
, 2,
5
2
, . . .
and
j, . . . , 1, 0, 1, . . . , j.
Note: Since j is almost the root of , j is essentially the length of the angular momentum vector. The
condition for says that the component of the angular momentum vector in the z-direction cannot be longer
than the total length of the angular momentum vector.
Note: It would seem we have only shown that all eigenvalues other than the above mentioned ones are
excluded. In fact, we have shown their (mathematical) existence by explicitly constructing their eigenvectors
from one vector [, ) by the repeated application of

L
.
Note: An often used notation for the state [, ) is [j, ).
9.5 Spin
We studied the commutation relation
[
L
i
,

L
j
] = i
3
k=1
i,j,k
L
k
()
and found that for any

L satisfying this the eigenvalues corresponding to
L
2
[, ) =
2
[, )
L
3
[, ) = [, )
are
= j(j + 1) where j 0,
1
2
, 1,
3
2
, 2,
5
2
, . . .
= j, j + 1, . . . , 0, 1, . . . , j.
We saw that an example of

L which obeys () is the orbital angular momentum:
L
(orb)
i
=
3
j,k=1
i,j,k
x
j
p
k
.
We saw that the possible eigenvalues for orbital angular momentum are only those corresponding to j
0, 1, 2, . . .. But we found that for operators obeying () the mathematics admits the possibility of those
having half-integer eigenvalues. We cannot interpret these as orbital angular momentum vectors. But the
presence of Fermions in the universe indicates that half-integer angular momentum values do occur in nature.
Recall we saw that for Fermions, a 2-rotation introduced a relative phase factor of (1). We know that
all our usual spatial degrees of freedom are unaected by such a 2-rotation. So it must be that an angular
momentum operator having half-integer eigenvalues acts on something other than the Hilbert space that we
have dealt with so far. This means we have a new degree of freedom for these operators. We call this the
spin degree of freedom. We introduce a new, nite-dimensional, Hilbert space for the spin, which we attach
to our familiar Hilbert space using a tensor product.
Physical Assumption 6 Quantum mechanical systems have a physically meaningful spin degree of freedom
which is independent of the position/momentum degrees of freedom. The corresponding spin operators

L
(spin)
act on a (new) nite dimensional Hilbert space. A description of a system including its position/momentum
and spin degrees of freedom is given by a unit vector in H = H
0
H
s
, where H
0
is the Hilbert space on
which the x and p act, and H
s
is the new nite-dimensional Hilbert space on which the spin operators act.
Note: The total angular momentum operator for a system is the sum of all its angular momentum operators;
i.e. the sum of all the orbital angular momentum operators of its constituent particles and all the spin
operators of its constituent particles.
The following is also known from physical experiment.
Physical Assumption 7 Each species of particle has a xed eigenvalue for its spin.
Some examples:
e
1
, quarks, neutrinos: =
1
2
(1 +
1
2
) (i.e. j =
1
2
) spinor
Higgs particles: = 0 (i.e. j = 0) scalar
Photons, gluons, w, z: = 1(1 + 1) (i.e. j = 1) vector
Since the spin operators act nontrivially only on these new degrees of freedom, they commute with all the
operators which act on the old Hilbert space (such as the x, p, orbital angular momenta, etc). This means
we can add, for example,

L
(spin)
2
,

L
(spin)
3
to a maximal set of commuting observables. So a new maximal
set of observables for a single particle could be
Q
(1)
,

Q
(2)
,

Q
(3)
,

L
(spin)
2
,

L
(spin)
3
where the

Q
(i)
are commuting observables which only contain the x and p (i.e. which do not contain any of
the spin operators

S
i
). The operators x 1, p 1 and 1

S
i
act on the full Hilbert space H = H
0
H
s
. In
general, the Hamiltonian may contain any polynomial (or other functions) of these operators.
Note that, in practise, it is often a good approximation to assume that the Hamiltonian is of the form
H =

H
0
+

H
s
, where

H
0
contains only the x and p operators, and where

H
s
contains only the spin operators
S
i
. (This is completely analogous to the case of two noninteracting subsystems of a larger system, the total
Hamiltonian being the sum of the subsystems Hamiltonians without any additional interaction Hamiltonian.)
We say in this case that the spin degree of freedom is decoupled from the usual motion degrees of freedom
of the particle.
9.6 Working With Spin-
1
2
Particles
For a spin-
1
2
particle, we choose as a maximal set of commuting observables
Q
(1)
,

Q
(2)
,

Q
(3)
,

L
(spin)
2
,

L
(spin)
3
with joint eigenbasis
[q
n1
, q
n2
, q
n3
, , )
where =
1
2
(1 +
1
2
).
Sometimes these basis vectors are denoted
[q
n1
, q
n2
, q
n3
) [, ).
Since is xed, [, ) is often shortened to [) (i.e. [
1
2
), [
1
2
) or also as [ ), [ )).
Let

S =

L
(spin)
. Consider the matrix elements of

S
i
in the [
1
2
), [
1
2
)-basis. We have
S
3
[
1
2
) =
1
2
[
1
2
)
S
3
[
1
2
) =
1
2
[
1
2
).
So
[
S
3
[
t
) =
_
1
2
0
0
1
2
_
,
.
Also, using the operators

L
we saw in Section 9.4 we can derive

[
S
1
[
t
) =
_
0
1
2
1
2
0
_
,
S
2
[
t
) =
_
0
i
2
i
2
0
_
,
.
These are sometimes more compactly represented as
S
i
=

2
i
where
i
are the Pauli Matrices:
1
=
_
0 1
1 0
_
2
=
_
0 i
i 0
_
3
=
_
1 0
0 1
_
.
Suppose we choose x
1
, x
2
, x
3
,

S
3
as a maximal set of commuting observables, with joint orthonormal basis
[ x
1
, x
2
, x
3
, ).
We have a resolution of the identity:
1 =
1
2
,
1
2
]
_
R
3
[x, )x, [d
3
x
and then
[) =
1
2
,
1
2
]
_
R
3
(x)[x, )d
3
x where
(x) = x, [).
Scalar products are
[) =
1
2
,
1
2
]
_
R
3
(x)
(x)d
3
x.
Matrix elements of operators are
x, [
S
3
[x
t
,
t
) =
3
(x x
t
)
2
_
1 0
0 1
_
,
x, [ x
i
[x
t
,
t
) = x
i
3
(x x
t
)
,
= x
i
3
(x x
t
)
_
1 0
0 1
_
,
x, [ p
j
[x
t
,
t
) = i

x
j
3
(x x
t
)
_
1 0
0 1
_
,
.
Wavefunctions now become 2-vectors
_
_
1
2
(x)
1
2
(x)
_
_
and the scalar product in terms of these is
[) =
_
R
3
_
1
2
(x),
1
2
(x)
_
_
_
1
2
(x)
1
2
(x)
_
_
d
3
x.
The action of operators

Q[) = [) in the x,

S
3
basis becomes
Q = x
j
: x, [ x
j
[) = x
j
_
1 0
0 1
_
_
_
1
2
(x)
1
2
(x)
_
_
Q = p
j
: x, [ p
j
[) = i

x
j
_
1 0
0 1
_
_
_
1
2
(x)
1
2
(x)
_
_
Q =

S
1
: x, [
S
1
[) =

2
_
_
_
0 1
1 0
_
_
_
1
2
(x)
1
2
(x)
_
_
_
_
.
9.7 Example: e
in a B-Field
H =
p
2
2m
+
L
(orb)
B+
S B.
Suppose, for example, the B-eld is constant in the x-direction. Then Schrodingers equation in the x,
S
3
-basis is:
i
d
dt
_
_
1
2
(x)
1
2
(x)
_
_
=
_

2
2m
_

2
x
2
1
+

2
x
2
2
+

2
x
2
3
__
1 0
0 1
_
+
_
ix
2
x
3
+ix
3
x
2
_
b
_
1 0
0 1
_
+b
2
_
0 1
1 0
__
_
_
1
2
(x)
1
2
(x)
_
_
.
10 Time-Independent Perturbation Theory
As we have just seen in the example above, Hamiltonians can very quickly become extremely complicated,
even for relatively simple physical systems. The can give rise to systems of many coupled dierential
equations, which could be very dicult to solve, in general. Perturbation theory is a systematic procedure
for obtaining approximate solutions by building on the known exact solutions to a simplied case.
Suppose the Hamiltonian

H can be split into the sum of an easy-to-diagonalise term

H
0
, and the remaining
term is

W, where is some small number:
H =

H
0
+

W
and

H
0
[E
(0)
n
) = E
(0)
n
[E
(0)
n
) , where E
(0)
n
, [E
(0)
n
) are assumed easy to calculate.
The goal is to nd E
n
, [E
n
) which solve

H[E
n
) = E
n
[E
n
). (1)
Assume spec(

H
0
) is nondegenerate and discrete (the more general case is just more complicated). Treat
H() as a function of . Then the eigenvalues E

n
() and the eigenvectors [E
n
()) are functions of as well.
Taylor expand E
n
and [E
n
) in powers of , and use the fact that is small to ignore terms of -degree 3
in the expansion:
E
n
= E
(0)
n
+E
(1)
n
+
2
E
(2)
n
+O(
3
) (2)
[E
n
) = [E
(0)
n
) +[E
(1)
n
) +
2
[E
(2)
n
) +O(
3
) (3)
(assume the [E
(i)
n
) have been normalised). Substituting (2) and (3) into (1) we get:
_
H
0
+

W
__
[E
(0)
n
) +[E
(1)
n
) +
2
[E
(2)
n
) +O(
3
)
_
=
_
E
(0)
n
+E
(1)
n
+
2
E
(2)
n
+O(
3
)
__
[E
(0)
n
) +[E
(1)
n
) +
2
[E
(2)
n
) +O(
3
)
_
.
Now compare terms of like powers of in the above equation to get:
0
-terms:

H
0
[E
(0)
n
) = E
(0)
n
[E
(0)
n
) (4)
1
-terms:

W[E
(0)
n
) +

H
0
[E
(1)
n
) = E
(0)
n
[E
(1)
n
) +E
(1)
n
[E
(0)
n
) (5)
2
-terms:

W[E
(1)
n
) +

H
0
[E
(2)
n
) = E
(2)
n
[E
(0)
n
) +E
(1)
n
[E
(1)
n
) +E
(0)
n
[E
(2)
n
) (6).
Equation (4) was assumed to be easy to solve. Now, multiply equation (5) on the left by E
(0)
n
[:
E
(0)
n
[
W[E
(0)
n
) +E
(0)
n
[
H
0
[E
(1)
n
) = E
(0)
n
E
(0)
n
[E
(1)
n
) +E
(1)
n
E
(0)
n
[E
(0)
n
).
This gives the rst-order correction to the eigenvalues:
E
(1)
n
= E
(0)
n
[
W[E
(0)
n
).
Since, by assumption, the eigenvectors [E
(0)
n
) are explicitly known in some basis, the calculation of the rst-
order correction in practise boils down to some explicit integral or sum (depending on whether one uses a
discrete or continuous basis).
To compute the rst-order correction [E
(1)
n
) to the eigenstate, multiply equation (5) on the left by E
(0)
n
[:
E
(0)
n
[
W[E
(0)
n
) +E
(0)
n
E
(0)
n
[E
(1)
n
) = E
(0)
n
E
(0)
n
[E
(1)
n
) +E
(1)
n
E
(0)
n
[E
(0)
n
).
Therefore:
E
(0)
n
[E
(1)
n
) =
E
(0)
n
[
W[E
(0)
n
)
E
(0)
n
E
(0)
n
. (7)
Note that substituting the expansion equation (3) into E
n
[E
n
) = 1 yields:
E
(0)
n
[E
(1)
n
) = 0 (8).
Equation (8) immediately gives:
[E
(1)
n
) =
,=n
[E
(0)
n
)E
(0)
n
[E
(1)
n
)
which together with equation (7) gives the rst order correction to the eigenstate:
[E
(1)
n
) =
,=n
E
(0)
n
[
W[E
(0)
n
)
E
(0)
n
E
(0)
n
[E
(0)
n
).
Finally, we calculate the second-order correction E
(2)
n
to the energy eigenvalues. Multiply equation (6) on
the left by E
(0)
n
[:
E
(0)
n
[
W[E
(1)
n
) +E
(0)
n
[
H
0
[E
(2)
n
) = E
(2)
n
E
(0)
n
[E
(0)
n
) +E
(1)
n
E
(0)
n
[E
(1)
n
) +E
(0)
n
E
(0)
n
[E
(2)
n
).
Therefore
E
(2)
n
= E
(0)
n
[
W[E
(1)
n
)
=
,=n
E
(0)
n
[
W[E
(0)
n
)E
(0)
n
[
W[E
(0)
n
)
E
(0)
n
E
(0)
n
and the second-order energy eigenvalue correction is thus

E
(2)
n
=
,=n
[E
(0)
n
[
W[E
(0)
n
)[
2
_
E
(0)
n
E
(0)
n
_ .
11 Bells Theorem
A profound question is the following: why is quantum mechanics probabilistic?. In the 1930s, Einstein,
Podolsky and Rosen (EPR) put forth a hypothesis to answer this question. They believed that the proba-
bilistic nature of quantum mechanics was due to incompleteness in the theory. The hypothesis was that the
state of any system is completely specied by giving
[),
1
, . . . ,
N
where
1
, . . . ,
N
are some hidden variables not accounted for in quantum mechanics. The hypothesis
says that every outcome of any observable

Q is exactly predictable when the hidden variables are taken into
account. That is, the outcome q of measurement of

Q is given by
q = f([), ).
Since quantum mechanics doesnt account for the hidden variables, we have to assign a probability distribu-
tion to the possible values of :
w(
1
, . . . ,
N
) = probability the hidden variables take values
1
, . . . ,
N
.
These probabilities must sum to 1:
_
w()d = 1.
Then along came John Bell. He proved that if EPR were right about the hidden variables, then there is
an experimentally testable inequality for correlations. In fact, the experiments have been done extensively
since Bells time, and they have conrmed beyond any doubt that EPR were wrong, and there are no such
hidden variables.
Consider two spin-
1
2
particles in a maximally entangled state:
[) =
1
2
([ ) [ ) [ ) [ ))
where [ ) is an eigenvector to

S
(i)
3
with eigenvalue +
2
and [ ) is an eigenvector to

S
(i)
3
with eigenvalue
2
.
Now consider two observables:
Q
(1)
a
=
2
S
(1)
a

Q
(2)
b
=
2
S
(2)
b where a, b are unit vectors.
The following are easy consequences:
Q
(1)
a
= [
Q
(1)
a
[) = 0
Q
(2)
b
= [
Q
(2)
b
[) = 0.
An ensemble of n runs of measurements of these observables on (identical copies of) the given state would
some sequence of pairs of 1s and -1s: (q
(1)
1
, q
(2)
1
), (q
(1)
2
, q
(2)
2
), . . . , (q
(1)
n
, q
(2)
n
). Consider the expectation for
the correlation between the outcomes of the two observables.
Denition 27
K(a, b) = lim
N
1
N
n=1
q
(1)
n
q
(2)
n
.
The quantum mechanical prediction for K is
K(a, b) = [
Q
(1)
a

Q
(2)
b
[).
For example, suppose we choose
a = (0, 0, 1)
T
, b = (sin(), 0, cos())
T
.
Then
K(a, b) =
4
2
[
S
(1)
3
_
sin()
S
(2)
1
+ cos()
S
(2)
3
_
[)
= cos()
In general, we have
K(a, b) = a b.
This is experimentally conrmed.
Remark: Choose for example a = b. Then K(a, b) = 1, and so spins are always found anti-aligned in
this state if a = b is chosen.
If EPR were correct, then all outcomes would be predetermined:
q
(1)
= A(a, ) , q
(2)
= B(b, ).
The have a distribution w() with
_
w()d = 1. So EPRs claim demands

K(a, b) =
_
w()A(a, )B(b, )d.

We know already that K(a, a) = 1. So A(a, ) = B(a, ) (except possibly for aset of of measure 0).
So this means
K(a, b) =
_
w()A(a, )A(b, )d.

Now, we eliminate w() from the equation:
K(a, b) K(a, c) =
_
w()A(a, ) (A(b, ) A(c, )) d

=
_
w()A(a, )A(b, ) (1 A(b, )A(c, )) d.

Since A(b, )
2
= 1, and A(a, )A(b, ) 1, we get
[K(a, b) K(a, c)[
_
w() (1 A(b, )A(c, )) d

which leads immediately to Bells Inequality:
[K(a, b) K(a, c)[ 1 K(b, c).
Now, recall the experimentally conrmed quantum mechanical prediction that K(a, b) = a b, and choose
=

4
. Then, in Bells inequality we would have
2

1
0
which is clearly false. So Bells inequality is violated, proving that EPR must have been wrong. This results
is called Bells Theorem.
Remark: Bells Theorem does not imply information transmission faster than the speed of light, since each
subsystem of [) is in a maximally mixed state whose density matrix is the identity, and the identity is the
same in any basis.
12 The Adiabatic Theorem
12.1 Dynamical Phase for a Time-Independent Hamiltonian
First consider a system with a time-independent Hamiltonian

H. Suppose at time t
0
we have a system
initially in an eigenstate [E
n
(t
0
)) = [E
n
) of

H, with energy eigenvalue E
n
. Now, let U(t) be the time-
evolution operator for the system. Since

H is time-independent, we have
U(t) = e
1
i
(tt0)

H
, U(t
0
) = 1.
Consider the eigenstate [E
n
(t
0
)) time-evolved for time t.
[E
n
(t)) =

U(t)[E
n
(t
0
))
= e
1
i
(tt0)

H
[E
n
(t
0
))
=
r=1
_
1
i
(t t
0
)

H
_
r
r!
[E
n
(t
0
))
=
r=1
_
1
i
(t t
0
)
_
r
r!

H
r
[E
n
(t
0
))
=
r=1
_
1
i
(t t
0
)
_
r
r!
E
r
n
[E
n
(t
0
))
=
r=1
_
1
i
(t t
0
)E
n
_
r
r!
[E
n
(t
0
))
= e
(tt0)En
[E
n
(t
0
))
So we see that for any time t the system remains in the eigenstate [E
n
) = [E
n
(t
0
)) of

H, only picking-up a
phase factor of
1
(t t
0
)E
n
. As we know, such phase factors cancel out in the calculation of measurement
expectations, and so are physically insignicant. A phase like this, which arises as a result of the time-
evolution of the system, is called a dynamical phase.
12.2 The Adiabatic Theorem
Consider now the case of a time-dependent Hamiltonian

H(t), with eigenvectors [E
n
) and eigenvalues E
n
.
We make the following assumptions about

H(t).
Assumptions
1)

H(t) is varying continuously over the time interval of interest, so E
1
(t), E
2
(t), . . . and [E
1
(t)), [E
2
(t)), . . .
are continuous functions of t.
2) The spectrum of

H(t) is discrete and nondegenerate throughout the time 0 t 1 during which the
Hamiltonian is changing.
Consider assumption (2). If there is a degeneracy at time t
c
, then at that time there is an ambiguity in the
ordering of the eigenvalues. It is to resolve such ambiguities that we must use the time-ordering symbol T
in the time-evolution operator for a general time-dependent Hamiltonian. If, however, we assume that the
spectrum remains nondegenerate for all t
i
t t
f
, we have a well-dened ordering of the eigenvalues and
we can write the time-evolution operator as
U(t) = e
1
i
t
t
0
H(t
)dt
, U(t
0
) = 1
without the need for the time-ordering symbol.
Note that assumption (2) may be relaxed, but it makes the analysis quite a bit more complicated, as we
need something like the time-ordering symbol to allow us to track the eigenstates.
Now, suppose

H is changing in compliance with above assumptions over the time interval t = t
f
t
i
.
Suppose the initial Hamiltonian is

H(t
i
) =

H
i
and the nal Hamiltonian is

H(t
f
) =

H
f
. Note that (t)
provides a measure of how fast the Hamiltonian varies between t
i
and t
f
. We introduce a new time variable
s, which is oset to s = 0 at time t
i
, and scaled by t:
s =
t t
i
t
.
Note the time-correspondences
s = 0 t = t
i
, and
s = 1 t = t
f
.
Notice that the assumptions (1) and (2) above for the functions of t imply the equivalent conditions for
the corresponding functions of s. There is one more technical assumption that must be made; that is, we
assume that the rst and second derivatives of [E
j
(s))E
j
(s)[ with respect to s are well-dened and piecewise
continuous throughout the time 0 s 1. We express

H(s) and

U(s) as functions of the new time variable
s. Our present goal is to determine the dependence of U(s) on t. To emphasize this, we write U
t
(s). In
the limiting case t 0 (i.e. a sudden change in the Hamiltonian) we have
lim
t0
U
t
(1) = 1.
So the state of the system remains unchanged between t
i
and t
f
.
We are interested in very gradual changes in the Hamiltonian, which are called adiabatic changes. These
correspond to the limiting case t . We consider this case in more detail. Suppose the eigenvalues of

H
are E
1
, E
2
, . . ., and the eigenvectors are [E
1
), [E
2
), . . .. These are, of course, all functions of t, and therefore
also of s.
Theorem 27 (Adiabatic Theorem)
lim
t
U
t
(s)[E
j
(0))E
j
(0)[ = [E
j
(s))E
j
(s)[ lim
t
U
t
(s) j = 0, 1, 2, . . .
Now, to see consider what the theorem is saying, multiply on both sides by the j
th
eigenstate of

H
i
(namely
[E
j
(0))).
lim
t
U
t
(s)[E
j
(0))E
j
(0)[E
j
(0)) = [E
j
(s))E
j
(s)[ lim
t
U
t
(t)[E
j
(0))
= lim
t
U
t
(s)[E
j
(0)) =

P
j
(s) lim
t
U
t
(s)[E
j
(0)) ()
The operator

P
j
(t) = [E
j
(s))E
j
(s)[ is the projector onto the j
th
eigenspace of

H(s). If we think in the
Scrodinger picture of the eigenstates as time-dependent Schrodinger states, then

U
t
(s)[E
j
(0)) is the j
th
eigenstate of

H
i
time-evolved to time s. So what equation () is saying is that, in the limit, once we evolve
the eigenstate forward to time s, the projector onto the j
th
eigenspace of

H(s) acts on the new state as
the identity. This means that the new state is the j
th
eigenstate of

H(s). So the adiabatic theorem can be
stated more informally as follows.
Theorem 28 (Adiabatic Theorem (restated)) For a slowly varying Hamiltonian, a system in the j
th
eigenstate of the initial Hamiltonian follows to the j
th
eigenstate of the new Hamiltonian.
Loosely speaking, what is meant by slowly varying Hamiltonian is that the characteristic time for a no-
ticeable change in the Hamiltonian is very large compared to the characteristic time required for a transition
from one eigenstate to the next. The adiabatic theorem, when used in practice, is often called the adiabatic
approximation.
12.3 Example
Consider an electron e
in a magnetic eld B = (B
x
, B
y
, B
z
). Considering only the spin degree of freedom,
the Hamiltonian is

H =
S B, where S = (
S
x
,

S
y
,

S
z
). Suppose the magnetic eld has constant magnitude
[B[ = B
0
, and is precessing about the z-axis with a constant angular-momentum , sweeping out a cone
with constant opening angle .
The time-dependent magnetic eld vector is
B(t) = B
0
(sin() cos(t), sin() sin(t), cos()) .
So the Hamiltonian in the eigenbasis of

S
z
(spin operator in the z-direction) is
H(t) =
S B =
B
0
2
(sin() cos(t)
x
+ sin() sin(t)
y
+ cos()
z
) ,
where
x
,
y
,
z
are the Pauli matrices. As a matrix (in the

S
z
-eigenbasis), this is
H(t) =
1
2
_
cos e
it
sin
e
it
sin cos
_
where
1
= B
0
. Its normalised eigenvectors (still in the

S
z
-eigenbasis) are
[
(t)) =
_
cos
_
2
_
e
it
sin(
2
)
_
, [
(t)) =
_
sin
_
2
_
e
it
cos(
2
)
_
with corresponding eigenvalues
1
2
,
=

1
2
respectively. These eigenvectors represent spin-up and spin-down along the instantaneous direction of B(t).
In the language we used in the discussion of the adiabatic theorem above, [
(t)) is the j = 1
st
eigenstate
of

H(t), and [
(t)) is the j = 2
nd
eigenstate of

H(t). Suppose the electron is initially in the rst (spin-up)
eigenstate of

H(t
0
) (to simplify the expression, let t
0
= 0):
[(0)) =
_
cos
_
2
_
sin
_
2
_
_
.
In general, there will be some positive probability that a transition occurs to the second eigenstate (spin-
down) of

H(t), but in the adiabatic limit, we should nd that this probability goes to zero.
The exact solution to the Schrodinger equation can be found to be
[(t)) =
_
_
_
cos
_
t
2
_
+i
(1+)
sin
_
t
2
_
_
cos(
2
)e
it
2
_
cos
_
t
2
_
+i
(1)
sin
_
t
2
_
_
sin(
2
)e
it
2
_
_
,
where =
_
2
+
2
1
+ 2
1
cos . In terms of the eigenvectors [
(t)), [
(t)), this is
[(t)) =
_
cos
_
t
2
_
+i
(
1
+ cos )
sin
_
t
2
__
e
it
2
[
(t))
+i
_
sin sin
_
t
2
__
e
it
2
[
(t)).
So the probability of a transition to spin-down at any time t (along the current direction of B(t)) is
[(t)[
(t))[
2
=
_
sin sin
_
t
2
__
2
.
The adiabatic approximation is valid when the characteristic time for changes in the Hamiltonian is much
greater than the characteristic time for a state-transition from one eigenstate to another. The characteristic
time for changes in the Hamiltonian in the present example is 1/ (since the B-eld is changing at the
rate ). The characteristic time for a phase change from spin-up to spin-down is 1/
1
. So the adiabatic
approximation is valid when
1
. In this case
1
, and so
[(t)[
(t))[
2
=
_

1
sinsin
_
t
2
__
2
0
and so we see that in the adiabatic limit the probability of a transition from spin-up to spin-down is 0. This
is exactly what the adiabatic theorem tells us; the system will remain in the rst (spin-up) eigenstate [
(t))
as long as the Hamiltonian is changing adiabatically.
13 The Berry Phase
13.1 Geometric Phase
Suppose we are standing at the North Pole holding a (ideal, frictionless) pendulum. Say we set the pendulum
in motion at the north pole, so that it swings directly along the longitudinal line passing through Waterloo.
If we want to carry the pendulum without disturbing its natural oscillation, we must do so adiabatically. Any
quick or sudden movements could disturb the orientation of the plane in which the pendulum is swinging.
Now suppose we (adiabatically) carry the Pendulum down the longitudinal line, keeping the pendulum
swinging north-south. We walk through Waterloo, and keep heading south, until we reach the equator.
Then, we walk (still adiabatically) along the equator, still keeping the pendulum swinging north-south. We
walk some distance along the equator (say 1000 km) and nally walk north again directly back to the north
pole. During the entire journey, the motion has been adiabatic, and the pendulum has been swinging north-
south. It is fairly easy to see that when we return to the north pole, the pendulum is swinging in a dierent
plane than it was originally; now it swings in the direction of the longitudinal line along which we travelled
back to the north pole. Let be the angle between the original plane of the pendulums oscillation, and the
plane of its oscillation after the motion.
The movement of the pendulum around the path while keeping the plane of its oscillation oriented in the
north/south direction is an example of parallel transport. Consider the area A bounded by the path along
which the pendulum was carried. It is a fraction

2
of the northern hemisphere of the earth:
A =
_
4r
2
2
__

2
_
= r
2
where r is the earths radius. So we can write the angle by which the plane of the pendulums motion has
rotated in terms of the area bounded by the path of the parallel transport:
=
A
r
2
(notice that this expression for is independent of the shape of the path, and depends only on the area
enclosed by the path). This angle is an example of a geometric phase. In this (classical) example, it arises
because of the intrinsic curvature of the sphere (the geometric phase would not occur for parallel transport
on the surface of a at manifold).
13.2 Development of the Berry Phase
It is crucial to note that the adiabatic theorem leaves open the possibility that the eigenstate may accumulate
an additional phase factor, on top of the dynamical phase we saw earlier. This is because the j
th
eigenstate
multiplied by some extra phase-factor is still the j
th
eigenstate. Let [E
j
(t)) denote the j
th
eigenstate of
H(t), with eigenvalue E

j
(t). For a system initially in state [E
j
(t
0
)), the adiabatic theorem tells us that at
time t the state is approximated by (and in the adiabatic limit is equal to)
[(t)) = e
ij(t)
e
t
t
0
Ej(t
)dt
[E
j
(t)). (1)
Here
1
_
t
t0
E
j
(t
t
)dt
t
is the dynamical phase. It is obtained in a manner analogous to what we did in Section
12.1 to nd the dynamical phase
1
(t t
0
)E
n
for time-independent

H. Where in that derivation we used
U(t) = e
1
i
(tt0)

H
, we now have

U(t) = e
1
i
t
t
0
H(t
)dt
(recall that our assumptions allow us to ignore the

time-ordering symbol in the expression for

U(t)). The factor
j
(t) in equation (1) is the additional phase
allowed by the adiabatic theorem. This
j
(t) is a geometric phase like the type we saw in Section 13.1, and
is called the Berry phase. We wish to obtain an expression for
j
(t). To simplify the calculations, let
j
(t)
denote the dynamical phase:
j
(t) :=
1
_
t
t0
E
j
(t
t
)dt
t
(2).
To evaluate the Berry phase, plug equation (1) into the Schrodinger equation, and use the eigenvalue equation,
obtaining
i

t
_
e
ij(t)
e
ij(t)
[E
j
(t))
_
=

H(t)
_
e
ij(t)
e
ij(t)
[E
j
(t))
_
= i
_
i
d
j
dt
e
ij(t)
e
ij(t)
[E
j
(t))
i
E
j
(t)e
ij(t)
e
ij(t)
[E
j
(t)) +e
ij(t)
e
ij(t)

t
[E
j
(t))
_
= e
ij(t)
e
ij(t)
E
j
(t)[E
j
(t))
which simplies to
t
[E
j
(t)) +i[E
j
(t))
j
t
= 0.
Multiply this equation by E
j
(t)[ to obtain
j
t
= iE
j
(t)[
t
[E
j
(t)),
and an expression for the Berry phase
j
(t) = i
_
t
t0
E
j
(t
t
)[

t
t
[E
j
(t
t
))dt
t
(3).
13.3 When is the Berry Phase Nonzero?
The Hamiltonian

H is dependent on time through some parameters Q
i
(t) which we assume are changing
(adiabatically) in time. First, suppose there is only one such parameter Q which changes from Q
0
at time
t
0
to Q
f
at some later time t
f
. Then E
j
and [E
j
) are both dependent on time through Q, and we have
t
[E
j
(t)) =

Q
[E
j
(Q))
dQ
dt
Plugging this expression into equation (3) we get
j
(t
f
) = i
_
t
f
t0
E
j
(Q)[

Q
[E
j
(Q))
dQ
dt
dt
= i
_
Q
f
Q0
E
j
(Q)[

Q
[E
j
(Q))dQ
In particular, if the Hamiltonian returns to its original form at time t
f
, then Q
0
= Q
f
and we have
j
(t
f
) = i
_
Q0
Q0
E
j
(Q)[

Q
[E
j
(Q))dQ = 0
and so the Berry Phase is 0. So in the case that only a single parameter in the Hamiltonian is changing in
time, the Berry phase equals 0.
Now suppose that there are N such adiabatically-changing parameters
Q
1
(t), Q
2
(t), . . . , Q
n
(t). Then

H, E
j
, and [E
j
) are dependent on t through
Q = (Q
1
(t), Q
2
(t), . . . , Q
n
(t)), and we have
t
[E
j
(t)) =

Q
1
[E
j
(Q))
dQ
1
dt
+

Q
2
[E
j
(Q))
dQ
2
dt
+ +

Q
N
[E
j
(Q))
dQ
N
dt
= (
Q
[E
j
(Q)))
dQ
dt
,
where
Q
is the gradient with respect to these parameters. So now we have
j
(t
f
) = i
_
Q
f
Q0
E
j
(Q)[
Q
[E
j
(Q)) dQ
for the Berry phase. If the parameters of the Hamiltonian return to their initial form after time t
f
, then this
becomes
j
(t
f
) = i
_
E
j
(Q)[
Q
[E
j
(Q)) dQ (4)
which is a line integral around a closed loop in this parameter space. From complex analysis, we know that
such an integral is not in general equal to zero. Note that
j
(t
f
) depends only on the path taken, and not on
the speed at which the parameters changed along that path. This is in contrast with the dynamical phase
j
(t
f
) =
1
_
t
f
t0
E
j
(t
t
)dt
t
which depends on the elapsed time t
f
t
0
.
13.4 The Berry Phase is Real
Note also that we can show that the Berry phase is real. If it were purely imaginary, then e
ij(t)
would be
an exponential factor and would be lost in the normalisation of [E
j
(t)).
Proof that
j
(t) R:
Q
(E
j
(Q)[E
j
(Q))) = 0
= (
Q
E
j
(Q)[) [E
j
(Q)) +E
j
(Q)[ (
Q
[E
j
(Q)))
= E
j
(Q)[
Q
[E
j
(Q))
+E
j
(Q)[
Q
[E
j
(Q)) = 0
= E
j
(Q)[
Q
[E
j
(Q)) is imaginary
=
j
(t) = i
_
Q
f
Qi
E
j
(Q)[
Q
[E
j
(Q)) dQ is real.
13.5 Alternative Expression for
j
(t)
Recall equation (4) expresses
j
(t) after a complete cycle as a line-integral around a closed loop C in the
parameter space:
j
(t
f
) = i
_
E
j
(Q)[
Q
[E
j
(Q)) dQ.
The assumption we made in Section 12.2 about the piecewise continuity of the rst and second derivatives
of [E
j
(s))E
j
(s)[ with respect to s will ensure that the curve C is piecewise-smooth. If we assume that the
parameter space is three-dimensional (so only parameters Q
1
, Q
2
, Q
3
of

H are changing in time) then C is
the boundary of some orientable surface S in the parameter space. Recalling vector calculus we can apply
Stokes theorem, and write
j
(t
f
) as a surface integral
j
(t
f
) = i
_
S
[
Q
E
j
(Q)[
Q
[E
j
(Q))d] a,
where
Q
E
j
(Q)[
Q
[E
j
(Q)) is the curl of E
j
(Q)[
Q
[E
j
(Q)).
Analogous results can be stated for parameter spaces with higher dimensions, but we will not pursue these
here.
13.6 Example
Recall the example in Section 12.3, illustrating the adiabatic evolution of a spin-
1
2
particle (an electron) in
an adiabatically rotating magnetic eld. We assumed the electron was initially in the spin-up eigenstate at
time t
0
= 0, with the corresponding eigenvalue
=
1
2
. Recall equation (2) for the dynamical phase:
j
(t) :=
1
_
t
t0
E
j
(t
t
)dt
t
.
For the present example, the state is the j = 1
st
eigenstate (the spin-up eigenstate), and the eigenvalue E
j
(t)
is a constant value
1
2
. So the dynamical phase is found to be
(t) =
1
_
t
0
_
1
2
_
dt
t
=

1
t
2
Now recall equation (3) for the Berry phase:
j
(t) = i
_
t
t0
E
j
(t
t
)[

t
t
[E
j
(t
t
))dt
t
.
Recalling the development of the example in Section 12.3, we have the j = 1
st
eigenstate written in the
S
3
-eigenbasis:
[
(t)) =
_
cos
_
2
_
e
it
sin
_
2
_
_
,
and so
t
[
(t)) =
_
0
ie
it
sin
_
2
_
_
and
(t)[
t
[
(t)) = i sin
2
_
2
_
.
A complete cycle of the magnetic eld takes time t = 2/, and so equation (3) now becomes
(t) = i
_ 2
0
i sin
2
_
2
_
= (cos 1)
_
using the identity 2 sin
2
_
2
_
= 1 cos
_
.
Now, consider the more general case in which the magnetic eld vector B traces out an arbitrary closed
path C (while [B[ = B
0
remains constant), returning to its starting position at time t
f
. Now the angular
momentum is not assumed to be constant; nor is there a constant opening-angle . The spin-up
eigenstate along B(t) is now
[
(t)) =
_
cos
_
2
_
e
i
sin
_
2
_
_
,
where (t) and (t) are functions of time. From vector calculus, we recall how to write the gradient in
spherical coordinates:
[
) =

r
[
) r +
1
r
+
1
r sin
)
=
1
r
_

1
2
sin
_
2
_
1
2
e
i
cos
_
2
_
_
+
1
r sin
_
0
ie
i
sin
_
2
_
_
.
So we get
[[
) =
1
2r
_
sin
_
2
_
cos
_
2
_
+ sin
_
2
_
cos
_
2
_
+ 2i
sin
2
_
2
_
sin

_
= i
sin
2
_
2
_
r sin
.
The curl is
[[
) =
1
r sin
_
sin
_
i sin
2
_
2
_
r sin
__
r
=
i
2r
2
r.
Recalling Section
[[
), we can write
(t
f
) as a surface integral over the surface S bounded by the
curve C swept-out by the the magnetic eld vector B:
(t
f
) =
1
2
_
1
r
2
r da.
By the geometry of this example da = r
2
d r, where is the angle subtended at the origin, and so our
integral becomes
(t
f
) =
1
2
_
d =
1
2
.
14 APPENDIX: The -Function and Fourier Transforms
Here we give a brief summary of some aspects of -functions and Fourier Transforms which are relevant to
quantum mechanics.
14.1 Denition of (x)
The Dirac delta function (x) is a continuous generalization of the Kronecker delta symbol:
n,m
=
_
0 if m ,= n
1 if m = n
.
Suppose f(x) is a function which is well-dened at the point x = x
0
. Then the dirac delta function (x) is
dened by
_

f(x)(x x
0
)dx = f(x
0
).
Thus (x) is a function which satises
(x x
0
) =
_
0 if x ,= x
0
if x = x
0
and
_

(x x
0
)dx = 1.
14.2 Properties of (x)
The following properties can be proved rigorously by means of distribution theory. We just state the prop-
erties without proof.
(x) = (x)
(ax) =
1
]a]
(x) a ,= 0
[g(x)] =

n:g(xn)=0,g
(xn),=0
1
]g
(xn)]
(x x
n
)
x(x) = 0
f(x)(x a) = f(a)(x a)
_
(x y)(y a)dy = (x a)
(x) =
a
2
_
e
ikx
dx
Also, the rst derivative
t
(x) of (x) has the following properties:
_
t
(x)f(x)dx = f
t
(0)
t
(x) =
t
(x)
_
t
(x y)(y a)dy =
t
(x a)
x
t
(x) = (x)
x
2
t
(x) = 0
t
(x) =
i
2
_
ke
ikx
dk
14.3 The Fourier Transform
If the Fourier Transform of a function f(x) exists, it is the function
F(u) T[f] =
_

2
_

e
iux
f(x)dx
where is a xed constant. In quantum mechanics, is taken to be
1
, and so the Fourier transform for us

is
F(u) T[f] =
_
1
2
_

e
iux
f(x)dx.
The inverse fourier transform is
f(x) T
1
[F] =
_
1
2
_

e
iux
F(u)du.
Generally, if f(x
1
, . . . , x
n
) is a function of n variables, its Fourier Transform is
F(u
1
, . . . , u
n
) T[f] =
_
1
2
_n
2
_

e
i(u
1
x
1
+...+unxn)
f(x
1
, . . . , x
n
)dx
1
. . . dx
n
and the inverse Fourier Transform is
f(x
1
, . . . , x
n
) T[F] =
_
1
2
_n
2
_

e
i(u
1
x
1
+...+unxn)
F(u
1
, . . . , u
n
)du
1
. . . du
n
.
14.4 Properties of the Fourier Transform
An integrable function f(x) is one satisfying
_

[f(x)[dx < .
Any integrable function has a Fourier transform.
In the following, we assume the Fourier Transforms of f and g exist, and F(u) = T[f], G(u) = T[g].
14.4.1 Important Limit Property
lim
]u]
F(u) = 0
14.4.2 Linearity
T[af(x) +bg(x)] = aF(u) +bG(u)
14.4.3 Fourier Transform of Derivatives
If f(x) is dierentiable, and its derivative is integrable, then
T[f
t
(x)] =
_
iu
_
F(u).
Suppose f(x) is m-times continuously dierentiable, and its derivatives are integrable, then
T
_
d
m
f
dx
m
_
=
_
iu
_
m
F(u).
The property is essentially that derivatives of f(x) get transformed into multiplication by
iu
. The converse
is essentially the same:
T[x
m
f(x)] =
_
i
_
m
d
m
du
m
[F(u)].
14.4.4 Multiplication Convolution
The convolution of two functions f(x), g(x) is dened as
(f g)(x) =
_

f(t)g(x t)dt =
_

f(x t)g(t)dt
The following properties essentially say that multiplication of functions Fourier Transforms to convolution
of functions (and vice-versa).
T[(f g)(x)] =
2F(u)G(u)
T[f(x)g(x)] =
1
2
(F G)(u)
14.4.5 Scaling and Shifting Properties
For a constant c, we have
T[f(cx
1
, . . . , cx
n
)] =
1
[c[
n
F(
u
1
c
, . . . ,
u
n
c
)
and
T
1
[F(cu
1
, . . . , cu
n
)] =
1
[c[
n
f(
x
1
c
, . . . ,
x
n
c
)
We give the shifting properties for the case of a function in one variable:
T[f(x a)] = e
iua
F(u)
and
T[e
ix
f(x)] = F(u )
14.4.6 Fourier Transform of (x)
T[(x a)] =
1
2
_

(x a)e
iux
dx =
1
2
e
iua
.

Lecturenotes PDF

Caricato da

Informazioni sul documento

Titolo originale

Copyright

Formati disponibili

Condividi questo documento

Condividi o incorpora il documento

Opzioni di condivisione

Hai trovato utile questo documento?

Questo contenuto è inappropriato?

Copyright:

Formati disponibili

Lecturenotes PDF

Caricato da

Copyright:

Formati disponibili

Advanced Quantum Mechanics

Notes by P.R. Kaye based on the lectures of A. Kempf

where K is some entity that is independent of the u

(Hermiteicity, or reality conditions). We say

. We come back to exactly what means when we look at representation theory.

, which follow immediately

Now, we check that x

f(t)[) for any observable

f[) f[ g[) +fg[)

An important special case of the general uncertainty principle is position-momentum uncertainty:

U(t). Substituting this into the equation for measurement-expectations gives

U(t). This is done by solving the Schr odinger Equation:

U (t). Putting these together we get

U(t) (since U is unitary)

is dened as the set of linear maps H C. We denote elements of

by [, where the action of [ is:

possess a map , called Hermitean conjugation:

and the inverse operation : H

called the dual basis.

f[) and insert the resolution of the identity twice:

f is diagonal in its own eigenbasis:

is an improper basis means that the vectors

are dened so that

(x, t)(x, t)dx

(extended linearly for linear combinations).

H form a discrete set E

[x) is an eigenvector of x, with eigenvalue ( x +b). This eigenvector is [x +b). So we

= p. Putting these observations together, for any Hamiltonian

has the following easily veriable properties.

[, )| (we set = 1 for simplicity).

on [, ) merely leads to the

we saw in Section 9.4 we can derive

H() as a function of . Then the eigenvalues E

and the second-order energy eigenvalue correction is thus

w()d = 1. So EPRs claim demands

w()A(a, )B(b, )d.

w()A(a, )A(b, )d.

w()A(a, ) (A(b, ) A(c, )) d

w()A(a, )A(b, ) (1 A(b, )A(c, )) d.

w() (1 A(b, )A(c, )) d

H(t), with eigenvalue E

(recall that our assumptions allow us to ignore the

, and so the Fourier transform for us

Potrebbero piacerti anche