Chapter 3

Unitary Time Evolution

3.1 Introduction
An important part of any physical theory is the dynamics (time-evolution) that it
predicts. According to Postulate 2, we can study the dynamics of isolated quantum
systems by solving the Schrödinger equation

∂|ψ(t)i
i~ = H|ψ(t)i. (3.1)
∂t
where H is the Hamiltonian, the Hermitian observable which represents the
energy of the system. In this chapter of the course, we will study some of the
general properties of solutions to the Schrödinger equation, and learn about some
approximate methods for integrating this equation. We will close the chapter by
introducing the Heisenberg picture, an alternative (and equivalent) formulation of
time-evolution in quantum mechanics, in which the dynamics is captured by the
evolution of observables, rather than states.

3.2 Evolution operator

When we integrate the Schrödinger equation, we need to specify, as a bound-
ary condition, the state of the system at a certain time, e.g. time t = 0. The
Schrödinger equation then allows us to calculate, given |ψ(0)i, the state of the
system at any other time time t, |ψ(t)i. This evolution is a map from one state to
another state, and can therefore be described in terms of an operator U (t), which
92 Chapter 3. Unitary Time Evolution

we call the evolution operator.

|ψ(t)i = U (t)|ψ(0)i (3.2)

We write the evolution operator U (t) since, as we shall show here, it is always
unitary, regardless of the form of the Hamiltonian, and whether or not the Hamil-
tonian is time-dependent. When the Hamiltonian is constant in time, U (t) has a
particularly simple form  
H
U (t) = exp −i t . (3.3)
~

Aside: Exponentiating operators

In equation (3.3), we are exponentiating the linear operator −iHt/~. Exponen-
tiated operators or matrices are defined using the power-series definition of ex ,
i.e.
X xj
ex = (3.4)
j
j!

In analogy with this, we define, for operator (or square matrix) Â,

X Âj
e = . (3.5)
j
j!

where powers of  are defined as follows Â0 = 1, Â1 = Â, Â2 = ÂÂ, etc.
For example, consider the following diagonal matrix B
 
λ1 0
B= . (3.6)
0 λ2

It is easy to take powers of diagonal matrices,

λj1 0
 
j
B = . (3.7)
0 λj2

and thus
X λj /j! 0
  λ
e 1 0

B 1
e = = . (3.8)
j
0 λj2 /j! 0 eλ2

Phas0069 PHASM/G426 Advanced Quantum Theory

3.2. Evolution operator 93

In general, diagonal matrices are easy to exponentiate, however non-diagonal ma-

trices do not share this, e.g. in general
   a b 
a b e e
exp 6= c d (3.9)
c d e e

and the power-series definition must be directly employed.

Exponentiated operators share many, but not all, properties and identities of
exponentiated numbers. In particular, for two operators  and B̂, the identity

ea+b = ea eb (3.10)

which holds for any number a, b does not hold in general for matrices and opera-
tors. One can show, via the power-series definition that this identity does however
hold in the special case that the matrices or operators commute, i.e. if [Â, B̂] = 0

eÂ+B̂ = e eB̂ . (3.11)

If  and B̂ do not commute then this identity usually does not hold.

Time-independent Hamiltonians
Above, we asserted that, if Hamiltonian H is time-independent then the evolution
operator takes the form:
 
H
U (t) = exp −i t (3.12)
~
Let us now verify that this is indeed the case, by verifying that the state
|ψ(t)i = U (t)|ψ(0)i with U (t) defined according to (3.12) satisfies the Schrödinger
equation. We will need to compute

∂ ∂
|ψ(t)i = (U (t)|ψ(0)i) (3.13)
∂t ∂t

Since |ψ(0)i is time invariant, we proceed by calculating

 
∂ ∂ H
U (t) = exp −i t (3.14)
∂t ∂t ~
94 Chapter 3. Unitary Time Evolution

To differentiate an exponentiated operator, we use the power-series definition

  ∞  j
∂ H ∂ X −iH
exp −i t = tj /j!
∂t ~ ∂t j=0 ~
∞  j
X −iH
= jtj−1 /j!
j=1
~
(3.15)
∞  j−1
−iH X −iH
= tj−1 /(j − 1)!
~ j=1 ~
 
−iH H
= exp −i t
~ ~

where the j = 0 term disappeared in the differentiation since it is constant in time.

This implies directly that

∂ −i −i
|ψ(t)i = HU (t)|ψ(0)i = H|ψ(t)i (3.16)
∂t ~ ~
as required.

Unitarity
We can use equation (3.12) directly to show that for time-independent H, U (t)
is unitary. We shall argue below that this unitarity holds for quantum evolution
operators, even if H is time-varying. Taking the Hermitian conjugate of equation
(3.12)
 
† H
U (t) = exp i t (3.17)
~
since H is Hermitian, and multiplying with U (t), we obtain
     
H H H −H
†
U (t)U (t) = exp −i t exp i t = exp i t =1 (3.18)
~ ~ ~

We are allowed to combine the exponentials via equation (3.11) because H and
−H (trivially) commute. Note that U (t)† is therefore the inverse of U (t), which,
is also equal to U (−t).

Phas0069 PHASM/G426 Advanced Quantum Theory

3.2. Evolution operator 95

Tracking evolution from other times

We have defined our evolution operator U (t) as the operator which transforms
state |ψ(0)i to |ψ(t)i, but it is easy to derive evolution operators which track
evolution from states at other times, e.g. |ψ(t1 )i.

|ψ(t1 )i = U (t1 )|ψ(0)i (3.19)

and hence
|ψ(0)i = U (t1 )† |ψ(t1 )i = U (−t1 )|ψ(t1 )i (3.20)
and hence, we can write the state at any time t in terms of |ψ(t1 )i.

|ψ(t)i = U (t)|ψ(0)i = U (t)U (−t1 )|ψ(t1 )i = U (t − t1 )|ψ(t1 )i (3.21)

where we use the fact that U (x)U (y) = U (x + y).

Step-wise changing Hamiltonian

We can use evolution operators to solve the Schrödinger equation for a Hamilto-
nian which evolves step-wise, i.e. one which is constant over finite periods of time
and changes discontinuously at certain points.
For example, consider a Hamiltonian which between times 0 ≤ t ≤ t1 the
Hamiltonian is H1 andbetween
H1
 times t1 ≤ t ≤ t2 the H2
Hamiltonian
 is H2 . Let
us define U1 (t) = exp −i ~ t and U2 (t) = exp −i ~ t . Using the reasoning
in the previous section, for all times 0 ≤ t ≤ t1 , the evolution operator will be
simply U1 (t), but for times t1 ≤ t ≤ t2 , the evolution operator is U2 (t − t1 )U1 (t1 ),
i.e. the first unitary operator applies for the first segment of time, of length t1 , and
the second applies for the segment of length t − t1 .
This approach can be generalised to a Hamiltonian with any number of dis-
crete changes in time. Since the product of two (or any number of) unitary opera-
tors is always unitary (check this), this evolution operator remains unitary.

Continuously changing Hamiltonian

Consider now a Hamiltonian H(t) which varies constantly in time. We can ap-
proximate it by a series of n discrete slices, each applied for time δt = t/n, for
each of which the Hamiltonian takes the constant value H(jδt) = H(jt/n) where
j = 1, 2, . . . , n is an integer which indexes each slice. This is equivalent to when
96 Chapter 3. Unitary Time Evolution

we approximate an integral by a sum of the areas of constant slices. An approxi-

mate evolution operator will be given by
n  
Y H(jt/n)
U (t) ≈ exp −i t (3.22)
j=1
~
This operator is unitary for the reasons given above.
We can obtain an exact evolution operator, in the limit n → ∞ (this is anal-
ogous to obtaining an exact integral by allowing the width of the slices to go to
zero).
n  
Y H(jt/n)
U (t) = lim exp −i t (3.23)
n→∞ ~
j=1

The unitarity of this operator survives in the limit, hence, evolution operators are
unitary also in this general case. Note that there are many similarities between
the limits we have taken here, and the limits you are familiar with for standard
calculus.
In practice, it is often not possible to exactly solve the Schrödinger equation
for time-varying Hamiltonians, and find a closed form for this limit. For that
reason approximation methods are very important. One of the most important of
these is time-dependent perturbation theory, which we introduce in chapter 5 of
this course.
Off-syllabus remark: The limit in equation (3.23) is sometimes known in
the literature as a time-ordered integral and written
n    Z t 0 0

Y H(jt/n) 0 H(t )dt
U (t) = lim exp −i t = T exp −i dt (3.24)
n→∞ ~ 0 ~
j=1

where the limit is to be understood as the definition of the expression on the right.
(The notation is rather unconventional, as the integral sign goes up into the ex-
ponent. We shall not use this notation for this lecture course, but you may come
across it in some advanced textbooks and research papers).

3.3 Suzuki-Trotter Decomposition

In physics, we often encounter Hamiltonians of the form
H = H1 + H2 (3.25)

Phas0069 PHASM/G426 Advanced Quantum Theory

3.3. Suzuki-Trotter Decomposition 97

where the operators H1 and H2 do not commute, E.g.

p̂2
H= + V (x̂) (3.26)
2m

It is also often the case that the individual terms in such Hamiltonians can be
solved individually (e.g. we can calculate a simple expression for exp[−iH1 t/~]
and exp[−iH2 t/~], as in the above example), but the full Hamiltonian cannot be
solved exactly (which is common in this example, unless V (x̂) has a very special
form).
In such cases, we can make use of a so-called Suzuki-Trotter decomposition.
This derives from the following result, first shown by Trotter in 1959 and gen-
eralised by Suzuki in 1976. For any operators (or square matrices) Â or B̂, the
following identity holds:
 n  n
Â+B̂ Â/n B̂/n B̂/n Â/n
e = lim e e = lim e e (3.27)
n→∞ n→∞

which, for quantum evolution operators, implies

−it
U (t) = e ~
(H1 +H2 )
= lim (U1 (t/n)U2 (t/n))n (3.28)
n→∞

where U1 (t) = exp −i H~1 t and U2 (t) = exp −i H~2 t . We can interpret this as
   

follows. The full evolution under H1 + H2 is approximated by splitting time into

discrete slices and then alternating between Hamiltonians H1 and H2 for succes-
sive slices. In the limit that the slices become infinitessimal the approximation
becomes exact.
Equation (3.28) implies that we can construct a family of approximations to
U (t), where instead of taking the infinite limit, we keep a finite n.

Suzuki-Trotter approximations
Here we shall study the validity of Suzuki-Trotter approximations for small n. To
gauge the effectiveness of the approximation, let us first write an expression for
the exact operator. To avoid overloading our calculation with ~s, we shall consider
98 Chapter 3. Unitary Time Evolution

the following operator, and expand it as a power series

X (Â + B̂)j tj
e(Â+B̂)t =
j
j!
(Â + B̂)2 t2 (3.29)
= 1 + (Â + B̂)t + + ···
2
(Â2 + B̂ 2 + ÂB̂ + B̂ Â)t2
= 1 + (Â + B̂)t + + ···
2
The most elementary Suzuki-Trotter approximation is the n = 1 case. In this
case, there are two possible decompositions we could choose eÂt eB̂t or eB̂t eÂt . If
 and B̂ do not commute, these different decompositions are not equal, but one
can show that they both have the same accuracy. We shall study the first one, by
first expanding each exponential as a power series,
X Âj tj X B̂ k tk
eÂt eB̂t = (3.30)
j
j! k k!

and then expressing the full operator as a power series in t,

 
e e = 1 + (Â + B̂)t + Â /2 + B̂ /2 + ÂB̂ t2 + · · ·
Ât B̂t 2 2
(3.31)

By comparing equations (3.29) and (3.31) we see that the n = 1 Trotter approxi-
mation has the correct t0 and t1 terms, but an error in the t2 term. This means that
the n = 1 approximation is accurate to first order in t and that the error is to order
t2 . It can therefore be used as a good approximation if t is sufficiently small.
Now let us consider the n = 2 approximation. Here there are actually four
possible decompositions we can consider. The reason for this is that the individ-
ual pairs of operators can applied in any order. Thus we have eÂt/2 eB̂t/2 eÂt/2 eB̂t/2 ,
eB̂t/2 eÂt/2 eB̂t/2 eÂt/2 , eÂt/2 eB̂t/2 eB̂t/2 eÂt/2 and eB̂t/2 eÂt/2 eÂt/2 eB̂t/2 possible choices.
The latter two operators can be reduced to 3 operators in the product, since eB̂t/2 eB̂t/2 =
eB̂t , so we shall study the third choice of operator here.

X Âj (t/2)j X B̂ k tk X Âl (t/2)l

eÂt/2 eB̂t eÂt/2 =
j
j! k
k! l l!
(3.32)
(Â2 + B̂ 2 + ÂB̂ + B̂ Â)t2
= 1 + (Â + B̂)t + + ···
2

Phas0069 PHASM/G426 Advanced Quantum Theory

3.4. Schrödinger and Heisenberg Pictures 99

We see that this expression is exact to second order in t. If we wrote out further
terms of the series we would see that the error here is of order t3 .
In general, one can show that for a given n, the approximation is valid to nth
order in t and the error is of order tn+1 . Therefore, for many problems, a modest
n suffices to provide a good approximation.

3.4 Schrödinger and Heisenberg Pictures

So far we have embodied the time-evolution of quantum systems in the time-
dependence of their state-vector |ψ(t)i

|ψ(t)i = U (t)|ψ(0)i. (3.33)

In quantum mechanics, the measurable quantities we can predict are statistical

properties such as expectation values, and probabilities, all of which (as men-
tioned above) are described mathematically as an expectation value. For operator
Ô the time-evolving expectation value can be written

hÔ(t)i = hψ(t)|Ô|ψ(t)i (3.34)

We can rewrite this expression

hÔ(t)i = hψ(0)|U (t)† Ô U (t)|ψ(0)i (3.35)

| {z } | {z }
State evolves in time State evolves in time

but now we see that there is another way of interpreting this

hÔ(t)i = hψ(0)| U (t)† ÔU (t) |ψ(0)i. (3.36)

| {z }
Operator evolves in time

Exactly the same expectation value is obtained when we leave the state fixed
as |ψ(0)i and allow the observable to evolve in time as U (t)† ÔU (t).
The two pictures of quantum time evolution represented by equations (3.35)
and (3.36) are called the Schrödinger and Heisenberg pictures respectively.

Schrödinger picture
The Schrödinger picture is the standard approach to quantum evolution, and the
one which you have likely always encountered up to now. State vectors evolve
100 Chapter 3. Unitary Time Evolution

in time as |ψ(t)i = U (t)|ψ(0)i and observables remain constant in time.1 . In

this section, to avoid ambiguity, when I am expressing an operator or state in
Schrödinger picture I will use an S subscript |ψ(t)iS and ÔS . For the Heisenberg
picture, I will use an H subscript.

Heisenberg Picture
At time t = 0, Schrödinger and Heisenberg pictures are equivalent. For all
other times, in the Heisenberg picture, time-evolution is carried by the opera-
tors, which evolve as OH (t) = U (t)† Ô(0)U (t), where Ô(0) = ÔS , the operator
in the Schrödinger picture. Expectation values are computed with respect to states
which are constant in time |ψiH = |ψ(0)iH .
You may wonder why we should introduce a new formalism for time- depen-
dence. There are several reasons for using the Heisenberg picture. Firstly, it can
simplify the calculations we make. Secondly, time-dependence in this picture is
closer, in a sense, to classical physics, where observable quantities x(t), p(t) do
vary in time. Thirdly, equations of motion in the Heisenberg picture are some-
times very reminiscent of the equivalent classical equations, allowing us to bet-
ter pin-point similarities (and hence differences) between quantum and classical
behaviour. Historically, the Heisenberg picture was developed (by Heisenberg)
for his matrix mechanics formulation of quantum theory, developed in parallel to
wave-function quantum mechanics.

An evolution equation in the Heisenberg picture

In the Schrödinger picture Schrödinger’s equation allows us to compute |ψ(t)iS
given the Hamiltonian and the initial state |ψ(0)i. The equivalent in the Heisen-
berg picture is, unsurprisingly, called the Heisenberg equation. This is a differ-
ential equation which allows us to compute ÔH (t) given the Hamiltonian and the
initial operator ÔH (0).
We can derive this equation as follows. Consider the derivative

∂ ∂  
ÔH (t) = U (t)† ÔH (0)U (t) (3.37)
∂t ∂t
1
In certain cases, operators can vary in time in the Schrödinger picture also, for example if
there is a time-varying magnetic field, however, here we shall ignore such cases.

Phas0069 PHASM/G426 Advanced Quantum Theory

3.4. Schrödinger and Heisenberg Pictures 101

We can compute this using the product rule

∂ ∂ ∂   ∂
U (t)† ÔH (0)U (t)+U (t)† ÔH (0) U (t)+U (t)† ÔH (0) (U (t))


ÔH (t) =
∂t ∂t ∂t ∂t
(3.38)
The middle term is zero, since ÔH (0) is constant. Earlier we saw that

∂ i
(U (t)) = − H(t)U (t) (3.39)
∂t ~
and
∂ i
U (t)† = U (t)† H(t)


(3.40)
∂t ~
where H(t) is the Schrödinger picture Hamiltonian (which may be time-varying).
Hence,

∂ i 
ÔH (t) = U (t)† H(t)ÔH (0)U (t) − U (t)† ÔH (0)H(t)U (t)
∂t ~
i † † † †

= U (t) H(t)U (t)U (t) ÔH (0)U (t) − U (t) ÔH (0)U (t)U (t) H(t)U (t)
~
i  
= HH (t)ÔH (t) − ÔH (t)HH (t)
~
i
= [HH (t), ÔH (t)]
~
(3.41)

where HH (t) = U (t)† H(t)U (t) is the Heisenberg picture Hamiltonian.

The final equation we come to is called the Heisenberg equation, which has a
simple and compact form

∂ i
ÔH (t) = [HH (t), ÔH (t)] (3.42)
∂t ~
.

Constants of Motion
It is clear from the Heisenberg equation that if [HH (t), ÔH (t)] = 0 then ÔH (t) is
constant in time. This means that the expectation value of ÔH (t) is a constant of
motion. This well-known result is more self-apparent in the Heisenberg picture.
102 Chapter 3. Unitary Time Evolution

Interaction picture
In addition to the Heisenberg and Schrödinger pictures there is an intermediate
picture of time-evolution called the Interaction picture. We shall study the Inter-
action picture in more detail at the end of chapter 4.

Phas0069 PHASM/G426 Advanced Quantum Theory