Summary of Lecture #3: quantum dynamics

Goals and Requirements:

• Get a clear understanding about the Schrödinger { ih̄ dt

d
|ψ, ti = ĤS (t)|ψ, ti }
and the Heisenberg { h̄ ddt ÔH (t) = i [ĤH (t), ÔH (t)] }
pictures about time evolution.

• Get some basic understanding about propagators (Green’s functions):

matrix element of time-evolution operator.

• Get some basic understanding about path integrals.

• Get some taste about geometric phase: Berry’s phase.

• Get some basic understanding about gauge invariance in quantum mechanics.

• Optional references:
J.J. Sakurai, Modern Quantum Mechanics, Chapter 2.
P.A.M. Dirac, The Principle of Quantum Mechanics, Chapter V.
R.P. Feynman, A.R. Hibbs, Quantum Mechanics and Path Integral, Chapter 2.
A. Altland, B.D. Simons, Condensed Matter Field Theory, Chapter 3.
M.E. Peskin, D.V. Schroeder, An Introduction to Quantum Field Theory, Section 9.1

• Note: for simplicity I will frequently assume the space is one-dimensional,

generalization to higher spatial dimensions should be obvious in most cases.

Advanced Quantum Mechanics, Fall 2018 1/14

 I TIME EVOLUTION

I. TIME EVOLUTION

A. Unitary Time Evolution

• The basic assumption of quantum dynamics:

time evolution of a closed system is unitary.

– State at t is related to state at t0 by a unitary operator Û (t, t0 ):

|ψ, ti = Û (t, t0 )|ψ, t0 i, Û † (t, t0 )Û (t, t0 ) = 1.

– Û (t, t1 ) · Û (t1 , t0 ) = Û (t, t0 ), and Û (t0 , t0 ) = 1.

– The time evolution is usually ‘continuous’, and reversible, Û † (t, t0 ) = [Û (t, t0 )]−1 .

• The infinitesimal time evolution: Û (t0 + dt, t0 ) = 1 − h̄i Ĥ(t0 ) dt + O(dt2 ).

The Hamiltonian Ĥ(t0 ) = ih̄ d
dt
Û (t, t0 ) = ih̄ d
dt
Û (t, t0 ) · Û † (t, t0 ) .
t=t0 t=t0

– Ĥ(t) = ih̄ dt
d
Û (t, t0 ) · Û † (t, t0 ), independent of the choice of t0 :
i dt
d
Û (t, t00 ) · Û † (t, t00 ) = i dt
d
Û (t, t0 )Û (t0 , t00 ) · [Û (t, t0 )Û (t0 , t00 )]†
= i dtd
Û (t, t0 )Û (t0 , t00 ) · Û † (t0 , t00 )Û † (t, t0 ) = i ddt Û (t, t0 ) · Û † (t, t0 ).

– Ĥ is Hermitian:
Ĥ † = −ih̄Û (t, t0 ) dt
d
Û † (t, t0 ) = −ih̄{ dt
d
[Û (t, t0 ) · Û † (t, t0 )] − d
dt
Û (t, t0 ) · Û † (t, t0 )}
= −ih̄{ ddt (1) − d
dt
Û (t, t0 ) · Û † (t, t0 )} = ih̄ dt
d
Û (t, t0 ) · Û † (t, t0 ) = Ĥ

B. The Schrödinger Equation and The Schrödinger Picture

• From the definition of Hamiltonian, we have the

Schrödinger equation for the time evolution operator: ih̄ dt
d
Û (t, t0 ) = Ĥ(t) · Û (t, t0 ).

• This translates into the

Schrödinger equation for the quantum state: ih̄ ddt |ψ, ti = Ĥ(t)|ψ, ti,
and the Schrödinger equation for the ‘bra’: −ih̄ dt
d
hψ, t| = hψ, t|Ĥ(t).

• Explicit form of the time evolution operator in terms of Ĥ:

– If Ĥ is independent of t, Û (t, t0 ) = exp[−i(t − t0 )Ĥ/h̄]. (IMPORTANT)

Conversely, if Û (t, t0 ) depends only on (t − t0 ), then Ĥ is independent of time.

Advanced Quantum Mechanics, Fall 2018 2/14

 I TIME EVOLUTION

Rt
– In general, Û (t, t0 ) = T (exp[− h̄i Ĥ(t0 )dt0 ]), where T means time-ordering.
t0 =t0

 Â(t)B̂(t0 ), t > t0 ,
0
∗ Time-ordering for bosonic  & B̂, T [Â(t)B̂(t )] =
 B̂(t0 )Â(t), t0 > t.

– This is equivalent to the Dyson series,

Û (t, t0 ) = 1 + −h̄i t1 =t0 Ĥ(t1 )dt1 + ( −h̄i )2 t2 =t0 t12=t0 Ĥ(t2 )Ĥ(t1 )dt2 dt1 + · · · .
Rt Rt Rt

• Stationary states: eigenstates of Ĥ, |Ĥ = E, ti = e−iE(t−t0 ) |Ĥ = E, t0 i.

Expectation value of any operator does not change over time (stationary).
Density matrix does not change over time.

• This is the Schrödinger picture (subscript S hereafter):

time evolution is implemented on the states.

C. The Heisenberg picture

• The Heisenberg picture (subscript H hereafter):

time evolution is encoded in operators, while the states have no evolution.

• Consider the time evolution of matrix elements of an operator ÔS :

hφ, t|ÔS |ψ, ti = hφ, t0 = 0|Û † (t)ÔS Û (t)|ψ, t0 = 0i.

• Define the time-dependent operator ÔH (t) in the Heisenberg picture:

(IMPORTANT) ÔH (t) = Û † (t)ÔS Û (t).
The time-dependent matrix element is simply hφ|ÔH (t)|ψi,
where the states φ & ψ do not evolve over time in Heisenberg picture.

d
• The Heisenberg equation of motion: h̄ dt ÔH (t) = i [ĤH (t), ÔH (t)].
NOTE: the Hamiltonian ĤH (t) here is also in the Heisenberg picture, ĤH (t) =
ih̄ Û † (t) ddt Û (t) = Û † (t)ĤS (t)Û (t).
d
– Proof: use h̄ dt Û (t) = −iÛ (t) · ih̄Û † (t) ddt Û (t) = −iÛ (t) · ĤH (t), and
h̄ ddt Û † (t) = h̄ dt
d
Û † (t) · Û (t)Û † (t) = −h̄Û † (t) dt
d
Û (t) · Û † (t) = iĤH (t) · Û † (t).
Then h̄ ddt ÔH (t) = h̄ dt
d
Û † (t) · ÔS · Û (t) + Û † (t) · ÔS · h̄ dt
d
Û (t)
= iĤH (t) · Û † (t) · ÔS · Û (t) − iÛ † (t) · ÔS · Û (t) · ĤH (t) = i [ĤH (t), ÔH (t)].

– If ĤS (t) is independent of time, then Ĥ commutes with Û (t), and ĤH = ĤS .

Advanced Quantum Mechanics, Fall 2018 3/14

 I TIME EVOLUTION

D. Some Applications

• The Schrödinger picture: quantum Liouville equation,

time evolution of density matrix: ih̄ dt
d
ρ̂(t) = [ĤS (t), ρ̂(t)].

p̂2
• Consider time-independent Ĥ = 2m
+ V (x̂).

d
– The Schrödinger picture: continuity equation, dt
[ρ(x, t)] + ∇ · J (x, t) = 0,
where the probability density ρ(x, t) = ψ ∗ (x, t)ψ(x, t) = hx|ρ̂(t)|xi,
probability current J (x, t) = Re[ψ ∗ (x, t) −im
h̄∂x
ψ(x, t)].

– The Heisenberg picture: Ehrenfest theorem:

the equation of motion of position x̂ and momentum p̂ are (subscript H omitted)
d i
dt
x̂(t) = h̄
[Ĥ, x̂(t)] = p̂/m,
d
dt
p̂(t) = i
h̄
[Ĥ, p̂(t)] = i[V (x̂), p̂] = − ∂x
∂
V (x).
2 2
∂
Combine them, m ddt2 x̂(t) = − ∂x V (x), or equivalently, m ddt2 hx̂i = −h ∂x
∂
V (x)i,
which looks like the classical equation of motion.

p̂ 2 2
• 1D harmonic oscillator: time-independent Ĥ = 2m + mω 2
x̂2 = h̄ω(↠â + 1/2),
q q
ip̂ † †
i (â − ↠).
p mω h̄ mωh̄
where â = 2h̄
(x̂ + mω
), [â, â ] = 1. x̂ = 2mω
(â + â ), p̂ = − 2

Normalized ground state |0i ( â|0i = 0 ), and excited states |ni ≡ √1 (↠)n |0i, energy
n!
eigenvalues En = h̄ω · (n + 12 ).

– Equation of motion: d
dt
â(t) = h̄i [Ĥ, â(t)] = −iω â(t). Then
â(t) = e−iωt â(0), ↠(t) = eiωt ↠(0).

– Solution to the equation of motion of x̂ and p̂: ellipse on hxi-hpi plane,

1
x̂(t) = cos(ωt) x̂(0) + mω
sin(ωt) p̂(0),
p̂(t) = −mω sin(ωt) x̂(0) + cos(ωt) p̂(0).
2 /2 †
– Example: t = 0 state is coherent state, |ψ(0)i = e−|z| ezâ |0i, with hâ(0)i = z.
q
2h̄
√
Then hx̂(0)i = mω Re(z), hp̂(0)i = 2h̄mω Im(z).
The expectation value of x̂ at time t is hx̂(t)i = cos(ωt)hx̂(0)i + sin(ωt)
mω
hp̂(0)i
q
i
= mω 2h̄
Re(ze−iωt ). You do not need to solve |ψ(t)i = e− h̄ Ĥ·t |ψ(0)i.

– The coherent state satisfies the minimal uncertainty relation for x̂ and p̂:
h̄ h̄mω h̄2
hx̂2 i − (hx̂i)2 = 2mω
, hp̂2 i − (hp̂i)2 = 2
, so h(∆x̂)2 i h(∆p̂)2 i = 4
= 14 |h[x̂, p̂]i|2 .
Exercise: check the proof of uncertainty relation to see why.

Advanced Quantum Mechanics, Fall 2018 4/14

 I TIME EVOLUTION

1 2 1
• Landau level: time-independent Ĥ = 2m
P̂ = 2m
[p̂ − qA(r)]2 ,
charge-q particle in xy-plane under uniform magnetic field B = Bez along z (B > 0),
r = (x, y), p̂ = −i( ∂∂x , ∂y
∂
), A is vector potential with ∇×A = B or ∂ ∂
A − ∂y
∂x y
Ax = B.

– Note: [r̂a , r̂b ] = 0, [r̂a , P̂b ] = ih̄δab ,

[P̂a , P̂b ] = [p̂a − qAa , p̂b − qAb ] = ih̄q (∂a Ab − ∂b Aa ) = ih̄q
P
c abc Bc .

– Equation of motion:
d
dt
r̂(t) = h̄i [Ĥ, r̂] = 1
m
P̂ , ddt P̂ (t) = h̄i [Ĥ, P̂ ] = 1
m
P̂ × qB.
2
Combine these, m ddt2 r̂(t) = ddt r̂(t) × qB, Lorentz force & cyclotron motion.
q
1
– Define b̂ = 2h̄qB (P̂x + iP̂y ), then [b̂, b̂† ] = 1, and Ĥ = h̄ωc (b̂† b̂ + 1/2).
ωc = qB/m is the cyclotron frequency, energy levels are En = h̄ωc · (n + 1/2) for
non-negative integer n. Exercise: check these statements.
ez P̂y P̂x
– Guiding center coordinates: R̂ = (X̂, Ŷ ) = r̂ − qB
× P̂ = (x̂ + qB
, ŷ − qB
).
d
R̂ is conserved: h̄ dt R̂(t) = i[Ĥ, R̂] = 0.
i
NOTE: [X̂, Ŷ ] = − qB 6= 0, indicates degeneracy of Landau level.
Exercise: eiX̂ commutes with Ĥ, but changes eigenvalue of Ŷ by −1/qB.

• The adiabatic theorem: roughly speaking, if a system starts at (one of) the instanta-
neous ground state(s), and the Hamiltonian changes slowly with time, then the system
will remain to be (one of) the instantaneous ground state(s) at later times.

– Sketch of a proof (for non-degenerate case):

Denote the instantaneous eigenstates of ĤS (t) by |ψn (t)i, and corresponding
eigenvalues by En (t), where n = 0, 1, 2, . . . , and E0 (t) < E1 (t) < E2 (t) < . . . .
Suppose |ψ, ti satisfies ih̄ ∂t
∂
|ψ, ti = ĤS (t)|ψ, ti, and |ψ, t = 0i = |ψ0 (t = 0)i.
Rt
Expand |ψ, ti, |ψ, ti = n cn (t)e−iθn (t) |ψn (t)i, where θn (t) ≡ h̄1 0 En (t)dt. The
P

Schrödinger equation becomes differential equations for coefficients cn (t),

∂ ∂ iθn (t)−iθm (t)
P
∂t
c n (t) = − m hψn (t)|( ∂t |ψm (t)i)·e ·cm (t), and c0 (t = 0) = 1, cn6=0 (t =
0) = 0. Take t-derivative on hψn (t)|ψm (t)i = δmn and hψn (t)|ĤS (t)|ψm (t)i =
∂
δmn En (t), we have hψn (t)|( ∂t |ψ (t)i) = Em (t)−E 1
· hψn (t)| ∂ Ĥ∂tS (t) |ψm (t)i for
h m n (t) i
1 ∂ ĤS (t)
m 6= n. So if the magnitude of E0 (t)−E n (t)
· hψ n (t)| ∂t
|ψ0 (t)i · (time duration)
is small, the final cn6=0 (t) will be small, the system will approximately remain to
be the instantaneous ground state (up to an overall phase).

Advanced Quantum Mechanics, Fall 2018 5/14

 II PROPAGATOR AND PATH INTEGRAL

II. PROPAGATOR AND PATH INTEGRAL

A. Brief Review of Gaussian Integrals

R∞ 2 /2a √
• One dimensional case: −∞
e−x dx = 2π a.
2
x2 e−x /2a dx
R
– hx2 i = R 2
e−x /2a dx
= a.
R 2n −x2 /2a
x e dx
– hx2n i = R 2
e−x /2a dx
= an · (2n − 1)!!, satisfies the ‘Wick expansion’ [(2n − 1)!! =
(2n − 1)(2n − 3) · · · (1) ways of pairing up xs].
R∞ 2 √ 2
– −∞ e−x /2a+yx dx = 2π a · eay /2 .
−1
·x
√
exp(− x·A2
R
• Higher dimensional Gaussian integral: ) dm x = (2π)m/2 det A, where
x = (xi ) is m-component real vector, A = (Aij ) is m × m real symmetric positive-
definite matrix, the integral is over all components of x from −∞ to +∞.

xi xj exp(−x·A−1 ·x/2) dm x
R
– hxi xj i = R
exp(−x·A−1 ·x/2) dm x
= Aij .

– ‘Wick expansion’: all possible ways of pairing up xs,

example, hxi xj xk x` i = Aij Ak` + Aik Aj` + Ai` Ajk .
√
– exp(−x · A−1 · x/2 + y · x) dm x = (2π)m/2 det A · exp(y · A · y/2).
R

∗ z/a
e−z
R
• Complex Gaussian integral: d2 z = πa,
where d2 z = dRez dImz, and the integral is over Rez and Imz from −∞ to +∞.
∗
z z ∗ e−z z/a d2 z
R
– hz z ∗ i = R ∗
e−z z/a d2 z
= a.

– hz n (z ∗ )m i = 0 if n 6= m. Consider z → eiθ z.

– hz n (z ∗ )n i = an · n! (‘Wick expansion’: n! ways of pairing up z ∗ and z).

∗ ∗ ∗ ∗
– e−z z/a+y z+z y d2 z = πa eay y .
R

exp(−z ∗ ·A−1 ·z) d2m z = π m det(A),

R
• Higher dimensional complex Gaussian integral:
where z = (zi ) is m-component complex vector, A = (Aij ) is a m × m Hermitian
positive-definite matrix.

– hzi zj∗ i = Aij .

– Non-vanishing ‘correlators’ must contain the same number of z ∗ and z.

Advanced Quantum Mechanics, Fall 2018 6/14

 II PROPAGATOR AND PATH INTEGRAL

– ‘Wick expansion’: all possible ways of pairing up z ∗ and z,

example, hzi zj zk∗ z`∗ i = Aik Aj` + Ai` Ajk .

– exp(−z ∗ · A−1 · z + y ∗ · z + z ∗ · y) d2m z = π m det(A) exp(y ∗ · A · y).

R

B. Propagator

• The propagator is the time-evolution operator represented in coordinate basis:

K(x0 , t; x, t0 ) = hx0 |Û (t, t0 )|xi, then K(x0 , t; x, t) = δ(x0 − x).

– By definition, ψ(x0 , t) = K(x0 , t; x, t0 )ψ(x, t0 ) dx.

R

– It is the transition probability amplitude for the particle (the system)

to start at x at time t0 and end up at x0 at time t.

• Customarily, when t < t0 , K(x0 , t; x, t0 ) ≡ 0, then K is the Green’s function satisfying

[H 0 − ih̄ ddt ]K(x0 , t; x, t0 ) = −ih̄δ(x0 − x)δ(t − t0 ), where H 0 is the Hamiltonian acting
on x0 .

• For time-independent Ĥ with energy eigenstates |Ei,

K(x0 , t; x, t0 ) = E e−iE (t−t0 )/h̄ hx0 |EihE|xi, for t > t0 .
P

p̂2
– For Ĥ = 2m
,
dp − h̄i p2 0
1 h̄i p (x0 −x) 2
q
K(x0 , t; x, t0 ) = m
exp[ im (x −x)
R (t−t0 )
h̄
e 2m
2π
e = 2πh̄(t−t0 )i 2h̄(t−t0 )
].
Exercise: draw qualitatively the shape of real/imaginary part of K.

R
• Trace of time-evolution operator G(t, t0 ) = Tr[Û (t, t0 )] = K(x, t; x, t0 ) dx.
For time-independent Ĥ with eigenvalues Ei , G(t) = i exp[−iE t/h̄], for t > 0,
P

similar to finite temperature partition function (with β replaced by it/h̄).

1. Brief Notes on Causal Functions

• The propagator and G(t) are “causal functions”: nonzero only for later (t > 0) times.

• You especially experimentalists will frequently encounter/measure such functions (re-

sponse functions): perturb the system and measure the response at later times.

Advanced Quantum Mechanics, Fall 2018 7/14

 II PROPAGATOR AND PATH INTEGRAL

• Consider the Fourier transform G̃(ω) = −i G(t) eiωt dt.

R

Because G(t < 0) = 0, G̃(ω 0 + iω 00 ) is non-singular(analytic) for ω 00 > 0 for all real ω 0 ,
and tends to zero fast enough at infinity with ω 00 > 0.

• Kramer-Kronig relation (Hilbert transform): for such functions G̃(ω),

R +∞ 0) R +∞ 0)
ReG̃(ω) = π1 P −∞ ImωG̃(ω
0 −ω dω 0 , ImG̃(ω) = − π1 P −∞ ReωG̃(ω
0 −ω dω 0 .
Or equivalently, iπ G̃(ω) = P −∞ G̃(ω
R +∞ 0)

ω 0 −ω
dω 0 .
Here P means Cauchy principal value.
G̃(ω 0 )
For a proof, consider the integral of ω 0 −ω
over ω 0 contour on the right.

• Measurements usually observe the imaginary part (dissipation, absorption, etc.) of

the response functions. Kramer-Kronig can be used to get the real part.

• Example: G(t > 0) = e−iEt , then G̃(ω) = 1

ω−E
− i π δ(ω − E).
The imaginary part (poles) can be used to identify the energy spectrum.

C. Reminder about Classical Mechanics

• The Lagrangian for dynamical system is L(q, q̇) = T − V (kinetic − potential energy),
d
which is a function of generalized coordinate q and generalized velocity q̇ = dt
q.

• Dynamics follows the principle of least action: classical trajectory ‘minimizes’ the
R
action, S = L(q, q̇) dt, among all trajectories with the same boundary condition.

d ∂L ∂L
• Euler-Lagrange equation: dt ∂ q̇
− ∂q
= 0, from δS = 0.

• The Hamiltonian is H(p, q) = pq̇ −L(q, q̇), the Legendre transformation of Lagrangian,
∂L
where the generalized momentum p = ∂ q̇
, and q̇ should be solved in terms of p and q.

• The Hamilton’s equation: ṗ = − ∂H

∂q
and q̇ = + ∂H
∂p
.

∂A ∂B ∂A ∂B
• Poisson bracket: {A, B} = ∂q ∂p
− ∂p ∂q
.
Corresponds to quantum commutator − h̄i [Â, B̂] of corresponding observables.

d
• Equation of motion: dt
A(p, q) = −{H, A}.
Corresponds to the Heisenberg equation of motion, d
dt
Â(t) = h̄i [Ĥ, Â].

Advanced Quantum Mechanics, Fall 2018 8/14

 II PROPAGATOR AND PATH INTEGRAL

R tf
• Hamilton’s principal function: S(qf , tf ; qi , ti ) = ti
L(q, q̇) dt, with q(ti ) = qi & q(tf ) =
∂S ∂S
qf , integrated over a classical trajectory. ∂qf
= p(t = tf ), ∂tf
= −H(t = tf ).

∂S ∂S
• Hamilton-Jacobi equation: ∂tf
+ H( ∂qf
, qf ) = 0.

D. Path Integral in Quantum Mechanics

• The goal: try to describe the quantum dynamics from a ‘classical’ point of view,
as particle moving in coordinate space (or coordinate-momentum phase space).
Then quantum interference between paths must be considered.

• Path integral version #1:

Rt
K(x0 , x, t) = D[x(τ )] exp[ h̄i 0 L(x(τ ), ẋ(τ )) dτ ] = D[x] exp[ h̄i S].
R R

D[x(τ )]: functional integral over all path x(τ ) with x(0) = x and x(t) = x0 .
R
–
The measure of paths is very difficult to define.

– L(x, ẋ): the Lagrangian.

d
ẋ = dτ
x:
the ‘velocity’ on the path x(τ ).
Rt
S: the action of path x(τ ), S[x(τ )] ≡ 0 L(x(τ ), ẋ(τ )) dτ .

• Path integral version #2:

Rt
K(x0 , x, t) = D[x(τ )]D[p(τ )] exp[ h̄i 0 [pẋ − H(p, x)] dτ .
R

R
– D[p(τ )]: integral over all path in momentum space, with proper measure.
Path of p has no boundary condition.

– H(p, x): classical Hamiltonian.

p̂2 i
• An example: time-independent Ĥ = 2m
+V (x). Propagator K(x0 , x, t) = hx0 |e− h̄ tĤ |xi.

– Divide this propagation over time t into N steps, each of time  = t/N ,
i
K(x0 , x, t) = hx0 |(e− h̄ Ĥ )N |xi, insert N − 1 resolution of identities. K(x0 , x, t) =
i i i
dxN −1 · · · dx2 dx1 hx0 |e− h̄ Ĥ |xN −1 i · · · hx2 |e− h̄ Ĥ |x1 ihx1 |e− h̄ Ĥ |xi.
R R R

i i p̂2 i
– Approximation (Trotter-Suzuki): e− h̄ Ĥ = e− h̄  2m e− h̄ V (x) + O(2 ).
i i p̂
2 i p m i  [ m ( xi+1 −xi )2 −V (x )]
hxi+1 |e− h̄ Ĥ |xi i ≈ hxi+1 |e− h̄  2m |xi i e− h̄ V (xi ) = 2πh̄ i
e h̄ 2  i
.

Advanced Quantum Mechanics, Fall 2018 9/14

 II PROPAGATOR AND PATH INTEGRAL

N PN −1
– K(x0 , x, t) ≈ m
) 2 exp( h̄i  [ m2 ( xi+1−xi )2 − V (xi )]),
R
dxN −1 · · · dx1 ( 2πh̄i i=0

where x0 = x & xN = x0 . The integral is over the discretized path

x0 (τ = 0) = x, x1 (τ = ), · · · , xN (τ = N  = t) = x0 .
P −1
– Take N → ∞ ( → 0) limit, sum in the exponent ( N i=0 ) becomes integral
Rt t 2
dτ , then K(x0 , x, t) = D[x(τ )] exp( h̄i 0 [ m2ẋ − V (x)]dτ )
R R
0
Rt
= D[x(τ )] exp( h̄i 0 L(x, ẋ) dτ ).
R

m N
∗ NOTE: there is an ugly normalization factor ( 2πh̄i
) 2 hidden in D[x(τ )].

p̂2
• The example again: Ĥ = 2m
+ V (x) = T (p̂) + V (x).
i i
K(x0 , x, t) = hx0 |e− h̄ tĤ |xi = hx0 |(e− h̄ Ĥ )N |xi.

i i i
– Trotter-Suzuki: e− h̄ Ĥ = e− h̄ T (p̂) e− h̄ V (x) + O(2 ).

– Insert N − 1 resolution of identity in terms of x eigenstates,

and N resolution of identity in terms of p̂ eigenstates,
K(x0 , x, t) = h̄−N dpN −1 · · · dp0 dxN −1 · · · dx1
R
i i i i
× hxN |e− h̄ T (p̂) |pN −1 ihpN −1 |e− h̄ V (x) |xN −1 i · · · hx1 |e− h̄ T (p̂) |p0 ihp0 |e− h̄ V (x) |x0 i
P −1
= h̄−N dpN −1 · · · dp0 dxN −1 · · · dx1 exp{− h̄i N
R
i=0  [T (pi ) + V (xi )]}

× hxN |pN −1 ihpN −1 |xN −1 i · · · hx1 |p0 ihp0 |x0 i

h̄−N dpN −1 · · · dp0 dxN −1 · · · dx1 (2π)−N
R
=
PN −1
× exp{ h̄i i=0  [pi ( xi+1−xi ) − T (pi ) − V (xi )]}
Rt
– In the N → ∞ limit, K(x0 , x, t) = D[x]D[p] exp( h̄i 0 [pẋ − H(p, x)] dτ ).
R

∗ The measure of paths D[x]D[p] contains the (2πh̄)−N normalization factor.

– If the Hamiltonian contains terms like p̂x̂ or x̂p̂, special care is needed.
See e.g. Peskin&Schroeder, Section 9.1.

E. Equivalence of Path Integral to the Schrödinger Equation

p̂2
• The example: Ĥ = 2m
+ V (x).

• Consider K(x0 , x, t + ), add one last step (xN +1 = x0 ) to the path integral.
i
K(x0 , x, t + ) = dxN hx0 |e− h̄ Ĥ |xN i K(xN , x, t)
R
R p m i  [ m ( x0 −xN )2 −V (x )]
≈ dxN 2πh̄ i
e h̄ 2  N
K(xN , x, t).

Advanced Quantum Mechanics, Fall 2018 10/14

 III GEOMETRIC PHASE

• Expand K(xN , x, t) around xN ∼ x0 , K(x0 , x, t + )

p m i  [ m ( x0 −xN )2 −V (x )] h (xN −x0 )2 ∂ 2
i
1 + (xN − x0 ) ∂x
∂
+ · · · K(x0 , x, t).
R
≈ dxN 2πh̄i e h̄ 2  N
0 + 2 ∂x02

• Do the Gaussian integral, keep terms up to O(),

i  ∂2
K(x0 , x, t + ) ≈ [1 + h̄ 2m ∂x02
− h̄i V (x0 ) + O(2 )]K(x0 , x, t).

• Finally, taking limit of  → 0,

2
∂
∂t
K(x0 , x, t) = − h̄i [− 2m
1 ∂
∂x02
+ V (x0 )]K(x0 , x, t) = − h̄i Ĥ 0 K(x0 , x, t),
which is the Schrödinger equation for propagator.

F. Stationary Phase Approximation

eik f (x) dx with large k, most contribution comes from xs where f is ‘stationary’
R
• For
[f 0 (xs ) = 0], where the integrand has no rapid oscillation. Expand f around xs ,
f (x) ≈ f (xs ) + 21 f 00 (xs ) (x − xs )2 , do the Gaussian integral, sum over all stationary xs ,
R ik f (x) P q i ik f (xs )
e dx ≈ xs k f2π 00 (x ) e
s
.

eik f (x) dn x, where x = (xi ) is n-component vector,

R
• For n-dimensional integral,
1 ∂ 2 f (xs )
expand f around stationary point xs , f (x) ≈ f (xs ) + 2 ∂xi ∂xj
(x − xs )i (x − xs )j ,
do the Gaussian integral, sum over all stationary points xs ,
P 2πi n/2  −1/2
R ik f (x) n ∂ 2 f (xs )
e d x ≈ xs ( k ) det ∂xi ∂xj eik f (xs ) .

• In path integral formulation of quantum mechanics, the large number k is 1/h̄.

The stationary phase approximation is the semi-classical approximation (h̄ → 0).
R i
– The stationary phase condition for D[x] e h̄ S is
δS
δx
= 0, or the classical equation of motion (Euler-Lagrange equation).

– (Not required) van Vleck formula: (see e.g. Prof. Littlejohn’s lecture notes #9)
i 2
K(x0 , x, t) = D[x] e h̄ S ≈ ( ∂ S )1/2 exp[ h̄i S(x0 , x, t)],
R P 1
√
2πh̄i ∂x0 ∂x
the sum is over all classical trajectories from x to x0 in time t.

III. GEOMETRIC PHASE

• Consider an adiabatic periodic evolution of a Hamiltonian Ĥ(t) with Ĥ(T ) = Ĥ(0).

Suppose the Hamiltonian always has a unique ground state |E0 (t)i of energy E0 (t).

Advanced Quantum Mechanics, Fall 2018 11/14

 III GEOMETRIC PHASE

After the periodic evolution, what is the phase acquired by the ground state?
RT
• The phase factor is hE0 (0)|Û (T )|E0 (0)i, where U (T ) = T (exp[− h̄i t0 =0
Ĥ(t0 )dt0 ]).

– Note that |E0 (t)i is not a “trajectory” of time evolution, t here is just a parameter
of these states. Û (t)|E0 (0)i is not exactly |E0 (t)i, but by the adiabatic theorem
they will only differ by a complex phase.

• Divide T into N intervals of  = T /N , define tn = n. Then up to O(2 )

i i
error, hE0 (0)|Û (T )|E0 (0)i ≈ hE0 (0)|e− h̄ Ĥ(tN −1 ) |E0 (tN −1 )i · · · hE0 (t1 )|e− h̄ Ĥ(t0 ) |E0 (0)i
i PN −1
≈ e− h̄  i=0 E(ti )
hE0 (0)|E0 (tN −1 )i · · · hE0 (t1 )|E0 (0)i.
i
RT
• With  → 0 (N → ∞) limit, the first factor becomes e− h̄ 0 E(t)dt
, which is the expected
dynamic phase acquired from time-evolution.

• The second factor is hE0 (0)|E0 (tN −1 )i · · · hE0 (tn )|E0 (tn−1 i · · · hE0 (t1 )|E0 (0)i.
IF |E0 (0)i = |E0 (tN = T )i, this is N
Q
n=1 hE0 (tn )|E0 (tn−1 )i

i i
QN ∂
PN RT
≈ n=1 [1 − hE0 (tn )| ( ∂tn
|E0 (tn )i)] ≈ exp[ n=1  At (tn )] ≈ exp[ 0
At (τ )dτ ],
where At (t) = ihE0 (t)| ( ∂t
∂
|E0 (t)i).
R
• The Berry’s phase: At (t)dt, where t parametrizes a periodic evolution |ψ(t)i,
the Berry connection (with respect to t) is At = ihψ(t)|( ∂t
∂
|ψ(t)i).

– Periodicity requirement: ψ(tfinal ) = ψ(tinitial ).

R
Otherwise At (t)dt is not the total Berry’s phase accumulated.

– Here t is just a parameter describing the path in the Hilbert space.

|ψ(t)i is not a “trajectory” of time-evolution, |ψ(t)i is not exactly
Û (t, tinitial )|ψ(tinitial )i.

• NOTE: The Berry’s phase does not depend on the speed of evolution,
it only depends on the closed path (geometry) in Hilbert space.

– Consider another evolution parametrized by u = f (t), then

Au = ihψ|( ∂∂u |ψi) = (f 0 )−1 · ihψ|( ∂t
∂
|ψi) = (f 0 )−1 At ,
the Berry’s phase Au (u)du = (f 0 )−1 At · f 0 dt = At dt.
R R R

• The Berry’s phase does not depend on h̄.

Advanced Quantum Mechanics, Fall 2018 12/14

 IV GAUGE INVARIANCE AND ELECTROMAGNETIC FIELD

• ‘Gauge transformation’ of the Berry connection:

add (t-dependent) complex phases to the wavefunctions |ψ(t)i → eiθ(t) |ψ(t)i,
note that θ(T ) − θ(0) ≡ 0 mod 2π for periodicity.

– The Berry connection becomes,

At → i hψ(t)|e−iθ(t) ∂∂t eiθ(t) |ψ(t)i = At − dθ


dt
.
dθ
R R R R
– The Berry’s phase becomes At dt → At dt − dt
dt = At dt − [θ(T ) − θ(0)] =
R
At dt mod 2π.

• Example: spin-1/2 under a Zeeman field rotating in xy-plane.

 
−iφ
0 e
– Ĥ(φ) = −(gµB B/2)(cos φ · σx + sin φ · σy ) = −(gµB B/2)  .
iφ
e 0
p p
– The ground state is ( 1/2, 1/2eiφ )T , periodic with φ from 0 to 2π.

– The Berry connection Aφ = i( 1/2, 1/2e−iφ ) · (0, i 1/2eiφ )T = −(1/2).

p p p

R 2π
– The Berry’s phase is 0 Aφ dφ = π.
The state acquires an additional minus sign after this periodic evolution.
p p
– Another choice of eigenvector ( 1/2e−iφ/2 , 1/2eiφ/2 )T is not good for comput-
ing Berry’s phase (not explicitly periodic for φ from 0 to 2π).

IV. GAUGE INVARIANCE AND ELECTROMAGNETIC FIELD

• Consider non-relativistic particle described by normalized wavefunction ψ(r, t).

∂ 2 + V (r)]ψ = ih̄∂t ψ.
2
– The Schrödinger equation is [ −h̄
2m r

– The probability density is ρ(r, t) = |ψ|2 .

p̂
– The probability current density is J (r, t) = Re[ψ ∗ m ψ] = −i 2m
h̄
(ψ ∗ ∂r ψ − ψ∂r ψ ∗ ).
d
– The continuity equation for probability is dt
[ρ(r, t)] + ∇r · J (r, t) = 0

• Adding a global phase factor ψ → eiθ ψ with real θ independent of r and t will not
change the above results.

Advanced Quantum Mechanics, Fall 2018 13/14

 IV GAUGE INVARIANCE AND ELECTROMAGNETIC FIELD

• If θ depends on r,t, then ∂t ψ → eiθ ∂t ψ + (i∂t θ)eiθ ψ, ∂r ψ → eiθ ∂r ψ + (i∂r θ)eiθ ψ,

h̄
ρ → ρ, J → J + m r
(∂ θ)ρ. It seems that the Schrödinger equation is not preserved,
the probability current density changes, and the continuity equation is violated.

• To make the theory formally “gauge invariant” under arbitrary ψ → eiθ ψ, we need to
absorb the ∂t θ and ∂r θ terms into the transformation of a “gauge field”.

– Define a 4-component space-time-dependent real-valued “gauge field” (a0 , a). De-

fine the canonical momentum P̂ = p̂ − h̄a.

– Demand the gauge transform to be: ψ → eiθ ψ, a → a + ∂r θ, a0 → a0 + ∂t θ.

Then it is easy to see that P̂ ψ → eiθ P̂ ψ.
2
P̂
– Modify the Schrödinger equation as [ 2m + V (r)]ψ = h̄(i∂t + a0 )ψ.
This will be invariant under the above gauge transformation.

– Modify the definition of probability current density as J (r, t) = Re[ψ ∗ m

P̂
ψ] =
−i 2m
h̄
(ψ ∗ ∂r ψ − ψ∂r ψ ∗ ) − m
h̄
aρ. This is invariant under the gauge transformation.
So continuity equation is preserved.

• For particle with electric charge q, the above “gauge field” is the electromagnetic 4-
potential, (a0 , a) = h̄q (−φ, A), (under SI units), where φ is the electrostatic potential
(electric field E = −∇φ − ∂t A), A is the vector potential (magnetic field B = ∇× A).
The gauge transformation of wavefunction is related to the gauge transformation of
electromagnetic 4-potential.

Advanced Quantum Mechanics, Fall 2018 14/14