Time evolution operator

In quantum mechanics
unlike position, time is not an observable.
there is no Hermitean operator whose eigenvalues
were the time of the system.
time appears only as a parameter, not as a
measurable quantity.
So, contradictory to teachings of the relativity theory,
time and position are not on equal standing. In
relativistic quantum field theories the equality is restored
by degrading also the position down to the parameter
level.
We consider a system which at the moment t0 is in the
state |i. When time goes on there is no reason to expect
it to remain in this state. We suppose that at a later
moment t the system is described by the state
|, t0 ; ti,

(t > t0 ),

where the parameter t0 reminds us that exactly at that

moment the system was in the state |i. Since the time is
a continuous parameter we obviously have
lim |, t0 ; ti = |i,

tt0

and can use the shorter notation

external constant magnetic field parallel to the z-axis, it

will precess in the xy-plane: the probability for the result
x oscillates between 0 and 1
h/2 in the measurement SG
as a function of time. In any case, the probability for the
result h/2 or h/2 remains always as the constant 1.
Generalizing, it is natural to require that
X
X
|ca0 (t0 )|2 =
|ca0 (t)|2 .
a0

In other words, the normalization of the states does not

depend on time:
h, t0 |, t0 i = h, t0 ; t|, t0 ; ti
= h, t0 |U (t, t0 )U(t, t0 )|, t0 i.
This is satisfied if we require U(t, t0 ) to be unitary, i.e.
U (t, t0 )U(t, t0 ) = 1.
2. Composition property

The evolution from the time t0 to a later time t2 should

be equivalent to the evolution from the initial time t0 to
an intermediate time t1 followed by the evolution from t1
to the final time t2 , i.e.
U(t2 , t0 ) = U(t2 , t1 )U(t1 , t0 ),

|, t0 i

|, t0 ; t0 + dti = U(t0 + dt, t0 )|, t0 i.

Due to the continuity condition

evolution
|, t0 ; ti.

We work like we did with translations. We define the

time evolution operator U(t, t0 ):

(t2 > t1 > t0 ).

Like in the case of the translation operator we will first

look at the infinitesimal evolution

|, t0 ; t0 i = |, t0 i.
Lets see, how state vectors evolve when time goes on:

a0

lim |, t0 ; ti = |i

tt0

we have
lim U(t0 + dt, t0 ) = 1.

dt0

|, t0 ; ti = U(t, t0 )|, t0 i,
which must satisfy physically relevant conditions.
1. Conservation of probability

We expand the state at the moment t0 with the help of

the eigenstates of an observable A:
X
|, t0 i =
ca0 (t0 )|a0 i.
a0

At a later moment we get the expansion

X
|, t0 ; ti =
ca0 (t)|a0 i.
a0

In general, we cannot expect the probability for the

system being in a specific state |a0 i to remain constant,
i.e. in most cases
|ca0 (t)| 6= |ca0 (t0 )|.
For example, when a spin 21 particle, which at the
moment t0 is in the state |Sx ; i, is subjected to an

So, we can assume the deviations of the operator

U(t0 + dt, t0 ) from the identity operator to be of the order
dt. When we now set
U(t0 + dt, t0 ) = 1 idt,
where is a Hermitean operator, we see that it satisfies
the composition condition
U(t2 , t0 ) = U(t2 , t1 )U(t1 , t0 ),

(t2 > t1 > t0 ),

is unitary and deviates from the identity operator by the

term O(dt).
The physical meaning of will be revealed when we
recall that in classical mechanics the Hamiltonian
generates the time evolution. From the definition
U(t0 + dt, t0 ) = 1 idt
we see that the dimension of is frequency, so it must be
multiplied by a factor before associating it with the
Hamiltonian operator H:
H = h,

or

iH dt
U(t0 + dt, t0 ) = 1
.
h

The factor h here is not necessarily the same as the factor

h in the case of translations. It turns out, however, that

in order to recover Newtons equations of motion in the

classical limit both coefficients must be equal.
Applying the composition property
U(t2 , t0 ) = U(t2 , t1 )U(t1 , t0 ),

(t2 > t1 > t0 )

we get
U(t + dt, t0 )

= U(t + dt, t)U(t, t0 )



iH dt
U(t, t0 ),
=
1
h

where the time difference t t0 does not need to be

infinitesimal. This can be written as
 
H
U(t + dt, t0 ) U(t, t0 ) = i
dt U(t, t0 ).
h
Expanding the left hand side as the Taylor series we end
up with

ih U(t, t0 ) = HU(t, t0 ).
t
This is the Schr
odinger equation of the time evolution
operator. Multiplying both sides by the state vector
|, t0 i we get
ih

U(t, t0 )|, t0 i = HU(t, t0 )|, t0 i.

t

Since the state |, t0 i is independent on the time t we can

write the Schrodinger equation of the state vectors in the
form

ih |, t0 ; ti = H|, t0 ; ti.
t
In fact, in most cases the state vector Schr
odinger
equation is unnecessary because all information about the
dynamics of the system is contained in the time evolution
operator U(t, t0 ). When this operator is known the state
of the system at any moment is obtained by applying the
definition
|, t0 ; ti = U(t, t0 )|, t0 i,

(ii) The Hamiltonain H depends on time but the operators

H corresponding to different moments of time commute.
For example, a spin 12 particle in the magnetic field whose
strength varies but direction remains constant as a
function of time. A formal solution of the equation
ih

U(t, t0 ) = HU(t, t0 )
t

is now

  Z t
i
0
0
dt H(t ) ,
U(t, t0 ) = exp
h
t0
which, again, can be proved by expanding the exponential
function as the series.
(iii) The operators H evaluated at different moments of
time do not commute For example, a spin 21 particle in a
magnetic field whose direction changes in the course of
time: H is proportional to the term S B and if now, at
the moment t = t1 the magnetic field is parallel to the
x-axis and, at the moment t = t2 parallel to the y-axis,
then H(t1 ) BSx and H(t2 ) BSy , or
[H(t1 ), H(t2 )] B 2 [Sx , Sy ] 6= 0. It can be shown that the
formal solution of the Schrodinger equation is now
U(t, t0 ) =
n Z t
Z t1

X
i
dt1
dt2
1+
h
t0
t0
n=1
Z tn1
dtn H(t1 )H(t2 ) H(tn ).
t0

This expansion is called the Dyson series. We will assume

that our Hamiltonians are time independent until we
start working with the so called interaction picture.
Suppose that A is an Hermitean operator and
[A, H] = 0.
Then the eigenstates of A are also eigenstates of H, called
energy eigenstates. Denoting corresponding eigenvalues of
the Hamiltonian as Ea0 we have

H|a0 i = Ea0 |a0 i.

We consider three cases:
(i) The Hamiltonian does not depend on time. For example,
The time evolution operator can now be written with the
a spin 12 particle in a time independent magnetic field
help of these eigenstates. Choosing t0 = 0 we get
belongs to this category. The solution of the equation




XX
iHt
iHt
00
00

=
|a
iha
|
exp

|a0 iha0 |
exp

ih U(t, t0 ) = HU(t, t0 )
h

t
a0 a00


X
iEa0 t
is


0
=
|a
i
exp

ha0 |.
iH(t t0 )
h

U(t, t0 ) = exp
a0
h

as can be shown by expanding the exponential function as

the Taylor series and differentiating term by term with
respect to the time. Another way to get the solution is to
compose the finite evolution from the infinitesimal ones:

N


(iH/h(t t0 )
iH(t t0 )
lim 1
= exp
.
N
N
h

Using this form for the time evolution operator we can

solve every intial value problem provided that we can
expand the initial state with the set {|a0 i}. If, for
example, the initial state can be expanded as
X
X
|, t0 = 0i =
|a0 iha0 |i =
ca0 |a0 i,
a0

a0

we get



iHt
|, t0 = 0; ti = exp
|, t0 = 0i
h



X
iEa0 t
0
0
=
|a iha |i exp
.
h

in an eigenstate |a0 i of an operator A commuting with the

Hamiltonian H. Suppose, we are interested in the
expectation value of an operator B which does not
necessarily commute either with A or with H. At the
moment t the system is in the state
|a0 , t0 = 0; ti = U(t, 0)|a0 i.

In other words, the expansion coefficients evolve in the

course of time as


iEa0 t
0
0
0
ca (t = 0) ca (t) = ca (t = 0) exp
.
h

So, the absolute values of the coefficients remain

constant. The relative phase between different
components will, however, change in the course of time
because the oscillation frequencies of different
components differ from each other.
As a special case we consider an initial state consisting of
a single eigenstate:
|, t0 = 0i = |a0 i.
At some later moment this state has evolved to the state


iEa0 t
0
.
|, t0 = 0; ti = |a i exp
h

Hence, if the system originally is in an eigenstate of the

Hamiltonian H and the operator A it stays there forever.
Only the phase factor exp(iEa0 t/
h) can vary. In this
sense the observables whose corresponding operators
commute with the Hamiltonian, are constants of motion.
Observables (or operators) associated with mutually
commuting operators are called compatible. As mentioned
before, the treatment of a physical problem can in many
cases be reduced to the search for a maximal set of
compatible operators. If the operators A, B, C, . . . belong
to this set, i.e.
[A, B] = [B, C] = [A, C] = = 0,
and if, furthermore,
[A, H] = [B, H] = [C, H] = = 0,
that is, also the Hamiltonian is compatible with other
operators, then the time evolution operator can be
written as

 X


iHt
iEK 0 t
0
exp
=
|K i exp
hK 0 |.
h
h

0
K

Here K 0 stands for the collective index:

In this special case we have

hBi = ha0 |U (t, 0)BU(t, 0)|a0 i




iEa0 t
iEa0 t
0
B exp
|a0 i
= ha | exp
h
h
= ha0 |B|a0 i,
that is, the expectation value does not depend on time.
For this reason the energy eigenstates are usually called
stationary states
We now look at the expectation value in a superposition
of energy eigenstates, in a non stationary state
X
|, t0 = 0i =
ca0 |a0 i.
a0

It is easy to see, that the expectation value of B is now



XX
i(Ea00 Ea0 )t
.
ca0 ca00 ha0 |B|a00 i exp
hBi =
h
00
0
a

This time the expectation value consists of terms which

oscillate with frequences determind by the Bohr
frequency condition
a00 a0 =

As an application we look at how spin 21 particles behave

in a constant magnetic field. When we assume the
magnetic moments of the particles to be eh/2me c (like
electrons), the Hamiltonian is


e
H=
S B.
me c
, we have
If we choose B k z


eB
H=
Sz .
me c
The operators H and Sz differ only by a constant factor,
so they obviously commute and the eigenstates of Sz are
also energy eigenstates with energies
ehB
2me c
ehB
+
2me c

for state |Sz ; i

for state |Sz ; i.

A|K 0 i = a0 |K 0 i, B|K 0 i = b0 |K 0 i, C|K 0 i = c0 |K 0 i, . . .

Thus, the quantum dynamics is completely solved (when
H does not depend on time) if we only can find a
maximal set of compatible operators commuting also with
the Hamiltonian.
Lets now look at the expectation value of an operator.
We first assume, that at the moment t = 0 the system is

Ea00 Ea0
.
h

We define the cyclotron frequency c so that the energy

difference between the states is h
c :
c

|e|B
.
me c

The Hamiltonian H can now be written as

H = c Sz ,
when we assume that e < 0.
All information about the evolution of the system is
contained in the operator


ic Sz t
.
U(t, 0) = exp
h

If at the moment t = 0 the system is in the state

|i = c |Sz ; i + c |Sz ; i,
it is easy to see that at the moment t it is in the state


ic t
|, t0 = 0; ti = c exp
|Sz ; i
2


ic t
+c exp +
|Sz ; i.
2
If the initial state happens to be |Sz ; i, meaning that in
the previous equation
c = 1,

c = 0,

we see that the system will stay in this state at all times.
This was to be expected because the state is stationary.
We now assume that the initial state is |Sx ; i. From the
relation
1
1
|Sx ; i = |Sz ; i + |Sz ; i
2
2
we see that

Lastly we look at how the statevectors corresponding to

different times are correlated. Suppose that at the
moment t = 0 the system is described by the state vector
|i, which in the course of time evolves to the state
|, t0 = 0; ti. We define the correlation amplitude C(t) as
= h|, t0 = 0; ti
= h|U(t, 0)|i.

C(t)

The absolute value of the correlation amplitude tells us

how much the states associated with different moments of
time resemble each other.
In particular, if the initial state is an energy eigenstate
|a0 i, then


iEa0 t
,
C(t) = exp
h
and the absolute value of the correlation amplitude is 1 at
all times. When the initial state is a superposition of
energy eigenstates we get


X
iEa0 t
|ca0 |2 exp
C(t) =
.
h
0
a

When t is relatively large the terms in the sum oscillate

rapidly with different frequencies and hence most
probably cancel each other. Thus we expect the
correlation amplitude decreasing rather rapidly from its
initial value 1 at the moment t = 0.
We can estimate the value of the expression


X
iEa0 t
2
|ca0 | exp
C(t) =
h
0
a

1
c = c = .
2

For the probabilities that at the moment t the system is

in eigenstates of Sx we get
|hSx ; |, t0 = 0; ti|2

|hSx ; |, t0 = 0; ti|2

c t
2
2 c t
sin
.
2
cos2

more concretely when we suppose that the statevectors of

the system comprise so many, nearly degenerate, energy
eigenvectors that we can think them almost to form a
continuum. Then the summation can be replaced by the
integration

Z

X

dE (E), ca0 g(E)
,

0
a

EEa0

Even if the spin originally were parallel to the positive

x-axis a magnetic field parallel to the z-axis makes the
direction of the spin to rotate. There is a finite
probability for finding the system at some later moment
in the state |Sx ; i. The sum of probabilities
corresponding to different orientations is 1.
It is easy to see that the expectation values of the
operator S satisfy
 
h

hSx i =
cos c t
2
 
h

hSy i =
sin c t
2
hSz i = 0.

where (E) is the density of the energy eigenstates. The

expression


X
iEa0 t
C(t) =
|ca0 |2 exp
h
0

Physically this means that the spin precesses in the

xy-plane.

In many realistic physical cases |g(E)|2 (E) is

concentrated into a small neighborhood (size E) of a

can now be written as



Z
iEt
C(t) = dE |g(E)|2 (E) exp
,
h
which must satisfy the normalization condition
Z
dE |g(E)|2 (E) = 1.

point E = E0 . Rewriting the integral representation as



iE0 t
C(t) = exp
h


Z
i(E E0 )t
dE |g(E)|2 (E) exp
,
h

we see that when t increases the integrand oscillates very

rapidly except when the energy interval |E E0 | is small
as compared with h
/t. If the interval, which satisfies
|E E0 | h/t, is much shorter than E the interval
from which the integral picks up its contribution, the
correlation amplitudes practically vanishes. The
characteristic time, after which the absolute value of the
correlation amplitude deviates significantly from its initial
value 1, is
h

.
t
E
Although this equation was derived for a quasi continuous
energy spectrum it is also valid for the two state system
in our spin precession example: the initial state |Sx ; i
starts to lose its identity after the time
1/c = h
/(E E ) as we can see from the equation
|hSx ; |, t0 = 0; ti|2 = cos2

c t
.
2

As a summary we can say that due to the evolution the

state vector describing the initial state of the system will
not any more describe it after a time interval of order
h/E. This property is often called the time and energy
uncertainty relation. Note, however, that this relation is
of completely different character than the uncertainty
relation concerning position and momentum because time
is not a quantum mechanical observable.