Вы находитесь на странице: 1из 10

Quantum Mechanics: bits and pieces 4

Notes by Sergei Winitzki


DRAFT July 17, 2005
Contents
1 Introducing the density operator 1
1.1 Motivation . . . . . . . . . . . . . . . . . . . . 1
1.1.1 Classical and quantum probability . . 1
1.1.2 Superposition and interference . . . . 1
1.1.3 Composite systems and tensor prod-
ucts . . . . . . . . . . . . . . . . . . . . 2
1.1.4 Life without the state vector . . . . . 3
1.1.5 Calculations without density operators 4
1.2 Denition of density operator . . . . . . . . . 4
1.2.1 Using the density operator . . . . . . 5
1.2.2 Examples . . . . . . . . . . . . . . . . 5
1.2.3 Equivalence of Cases 1 and 2 . . . . . 6
1.3 Further properties . . . . . . . . . . . . . . . . 6
1.3.1 Pure states and mixed states . . . . . . 6
1.3.2 Observability of phases . . . . . . . . 6
1.3.3 Algebraic properties . . . . . . . . . . 7
1.3.4 Partial trace . . . . . . . . . . . . . . . 7
1.3.5 Basis-free denition of trace . . . . . 7
2 Evolution of the density operator 8
2.1 Effect of measurement . . . . . . . . . . . . . 8
2.1.1 Pure states after measurement . . . . 8
2.1.2 Mixed states after measurement . . . 8
2.2 The Liouville-von Neumann equation . . . . 9
3 Applications 9
3.1 Thermal ensembles . . . . . . . . . . . . . . . 9
3.2 Entropy . . . . . . . . . . . . . . . . . . . . . . 9
3.3 Decoherence . . . . . . . . . . . . . . . . . . . 10
1 Introducing the density operator
1.1 Motivation
1.1.1 Classical and quantum probability
In a classical theory, the result of any measurement is cer-
tain as long as the state of the system is precisely known.
(For example, the state of an oscillator is uniquely specied
by its coordinate q (t) and velocity q (t) at some time t.) If
the state of the system is known probabilistically, i.e. one
knows only the probability p
k
for the system to be in a
certain state, then only probabilities for the results of mea-
surements can be predicted. The loss of predictive power is
clearly related to the lack of information about the state of
the system.
Quantum theory always describes probabilities for mea-
surement outcomes and generally does not predict the ac-
tual outcomes with certainty. This is not because one has
insufcient information about the state of the quantum sys-
tem, but because the outcome is considered to be inherently
random and unpredictable. In quantum theory, the knowl-
edge of a quantum state vector | is the most complete pos-
sible information about the state of a quantum system.
If the quantum state is known only probabilistically, the
measurement outcomes are even more uncertain. Thus in
quantum mechanics there are two possible sources of ran-
domness: the inherent unpredictability of quantum phe-
nomena and incomplete information about the state of the
system.
Let us compare the classical and the quantum probabilis-
tic descriptions in more detail. As an example of a classical
system, consider
a particle having energy E
1
or E
2
with probability
1
2
. (1)
For instance, we could imagine a device that prepares the
particle with energy E
1
or E
2
according to an internal ran-
dom number generator to which we have no access. Our
quantum example is a particle prepared in the quantum
state
| =
1

2
|E
1
+
1

2
|E
2
, (2)
where |E
1,2
are quantum states with energy E
1,2
. In both
cases a measurement of energy will yield E
1
or E
2
with
probability
1
2
each. Despite a supercial similarity, there
is a crucial physical difference between these two systems.
This difference is not seen when one measures the energy of
the particle, but is manifested when one uses experiments
involving interference, such as the famous double-slit and
Stern-Gerlach experiments.
1.1.2 Superposition and interference
The principle of superposition says that whenever a quan-
tum system can be prepared (with help of appropriate
physical devices) in two states |
1
and |
2
, there exists
also a physical process that prepares the system in the state
|
1
+ |
2
(this state is called a superposition of |
1

and |
2
), where and are some complex numbers. This
process does not depend on the states |
1,2
but only on
the numbers and . Interference is the physical phe-
nomenon underlying the superposition principle; any de-
vice that prepares a superposition of states is based on in-
terference. The principle of superposition is considered to
be valid for all sufciently simple quantum systems.
1
1
There exist superselection rules that prohibit superpositions of states
with different charge or with different parity symmetry. These issues are
relevant for quantum eld theory and we shall ignore superselection rules
at this point.
1
Consider two quantum states
|
+
=
1

2
(|E
1
+|E
2
) , |

=
1

2
(|E
1
|E
2
) . (3)
For both states, the probability for observing the energy E
1
or E
2
is
1
2
, but nevertheless these states are physically differ-
ent, in other words there exist physical processes leading to
different observable results when operating on these states.
An example of such a process is a device that prepares a
superposition
1

2
|
+
+
1

2
|

= |E
1
, (4)
the result being a state of denite energy E
1
. The same de-
vice operating on two copies of |
+
will yield again |
+

as the resulting state. Since the states |


+
and |E
1
can be
observed to have different distributions of energy, it is clear
that the states |
+
and |

are different.
In contrast to a quantum particle, a classical particle in
the state (1) does not exhibit any interference. It is impossi-
ble to superimpose two such states to obtain a state with
a denite energy E
1
.
1.1.3 Composite systems and tensor products
We now make a digression to consider the quantum-
mechanical description of composite systems.
In general, a quantumsystemis described by observables

A which are operators in a Hilbert space H. We may also


consider a system that is a union of two systems with state
spaces H
1
and H
2
. In that case, the composite system is
described by the state space H = H
1
H
2
, the tensor product
of the two state spaces. The composite system may be in a
state such as | |, showing that the rst subsystem is
in the state | while the second one is in the state |.
The motivation for using the tensor product as the state
space of the composite system (and not e.g. the direct sum
H
1
H
2
) is found in the superposition principle.
Let us rst suppose that the state of the composite system
is simply a symbolic pair {| , |} indicating the states
| , | of the respective subsystems. In other words, we
introduce a map
{, } : | , | {| , |} H (5)
that converts pairs of states from H
1
and H
2
into states of
the composite system that belong to some yet unknown
space H. Now we shall identify the space H. According
to the principle of superposition, we should be able to pre-
pare arbitrary superpositions of pairs, e.g.

1
{|
1
, |
1
} +
2
{|
2
, |
2
} +... (6)
Therefore the space H must be the set of all linear com-
binations of the form (6), and not just the set of all pairs
{| , |}. Further, we can consider a superposition of
states {|
1
, |} and {|
2
, |}, e.g.

1
{|
1
, |} +
2
{|
2
, |} . (7)
This superposition is prepared by a physical process that
leaves the state | of the second subsystem unchanged.
It is clear that such a process does not need to involve
the second subsystem at all; this process merely prepares
the superposition of states of the rst subsystem. There-
fore the resulting state must be the same as the pair
{
1
|
1
+
2
|
2
, |}. We have found the identity

1
{|
1
, |} +
2
{|
2
, |} = {
1
|
1
+
2
|
2
, |} .
(8)
A similar identity must hold with respect to the second
subsystem,

1
{| , |
1
} +
2
{| , |
2
} = {| ,
1
|
1
+
2
|
2
} .
(9)
Thus we have derived the bilinearity of the map {, } as a
necessary consequence of the superposition principle.
By denition, the space H of all linear combinations of
the form (6), where the symbol {, } is understood as a bi-
linear map according to Eqs. (8)-(9), is the tensor product
space of H
1
and H
2
; this new space is denoted by H
1
H
2
.
Instead of the braces {, }, one writes the tensor product
symbol between the vectors, i.e.
{| , |} | | . (10)
In physics literature, the tensor product symbol is fre-
quently omitted for brevity, and one writes
| | | | (11)
when this does not cause confusion. (We shall write the
symbol explicitly in these notes.) A state |
total
H of
the composite system is in general a linear combination of
such tensor products,
|
total
=

n=1
|
n
|
n
. (12)
The identities (8)-(9) allow one to perform calculations in
the space Hby treating the tensor product symbol as if it
were a multiplication operation. For instance,
_
|
1
+ 2 |
2

_
|
1
|
2

_
|
1
|
1

= 2 |
2
|
1
|
1
|
2
2 |
2
|
2
. (13)
Note that a general state such as |
2
|
1
+|
1
|
2
is
not equal to any single tensor product | | unless the
vectors |
1,2
or |
1,2
are linearly dependent.
The tensor product space H
1
H
2
can also be visualized
using explicit bases {|e
n
} in H
1
and {|f
n
} in H
2
. Namely,
a basis in H
1
H
2
is the set of tensor products |e
m
|f
n

for all m, n. An arbitrary vector in the space H


1
H
2
is a
linear combination
| =

m,n

mn
|e
m
|f
n
, (14)
where
mn
are complex coefcients, the components of the
vector | in the basis {|e
m
|f
n
} .
To make the description of a composite system complete,
we need to specify operators acting in the space Hthat cor-
respond to various observables. First consider an observ-
able that belongs to the rst subsystem and is described by
an operator

A
1
acting in the space H
1
. The operator

A
1
can
be naturally extended to act in the tensor product space H
by the formula

A
1
(| |) = (

A
1
|) | . (15)
2
One can easily verify that this formula agrees with the
denining properties of the tensor product when

A
1
is a
linear operator. This extension of the operator

A
1
is de-
noted by

A
1

1. More generally, if

A
1
and

A
2
are operators
in H
1
and H
2
respectively, then

A
1


A
2
is an operator in
H = H
1
H
2
dened by the natural formula
_

A
1


A
2
_
(| |) =
_

A
1
|
_

A
2
|
_
. (16)
The action of

A
1


A
2
on linear combinations of the
form (12) is the linear combination of the respective terms,
_

A
1


A
2
_
_

n=1
|
n
|
n

_
=

n=1
_

A
1
|
n

A
2
|
n

_
.
(17)
Thus we can promote arbitrary observables

A
1
and

A
2
of
each of the subsystems to observables of the composite sys-
tem.
The bra-ket notation is extended to tensor product spaces
by the convention that
(
1
|
1
|) (|
2
|
2
)
1
|
2

1
|
2
. (18)
Note that the RHS above contains the product of ordinary
numbers. A dual vector
1
| may be viewed as a linear map
fromH = H
1
H
2
to H
2
, acting as

1
| (|
2
|
2
) =
1
|
2
|
2
H
2
. (19)
In other words,
1
| acts on |
2
but not on |
2
since |
2

belongs to a different space. Similarly,


1
| may be viewed
as a linear map fromH = H
1
H
2
to H
1
.
Specic operators acting in H can be easily written us-
ing the bra-ket notation. For example, consider projectors

P
|
and

P
|
onto some states | H
1
and | H
2
; the
projector onto the product state | | is

P
||
=
_

P
|
_

P
|
_
= (| |) (| |)
= (| |) (| |) . (20)
The projector

P
|
onto a state
| |
1
|
1
+|
2
|
2
(21)
is expressed by

P
|
= | |
= (|
1
|
1
+|
2
|
2
) (
1
|
1
| +
2
|
2
|) .
Since an arbitrary operator

A
1
in the space H
1
can be
written as

A
1
=

n=1
|
n

n
| , (22)
where |
n
and |
n
are suitable vectors (not necessarily or-
thogonal or a basis), it is clear that a general operator

A act-
ing in H can be written as

A =

n=1
(|
n
|
n
) (
n
|
n
|) , (23)
with suitable vectors |
n
, |
n
H
1
and |
n
, |
n
H
2
,
or equivalently as a linear combination

A =

n=1

A
n


B
n
, (24)
where

A
n
and

B
n
are some suitable operators in H
1
and H
2
respectively.
1.1.4 Life without the state vector
The correct quantum counterpart of the state (1) is not the
state (2) but the situation when the quantum state is not
known with certainty, namely
a particle is in |E
1
or in |E
2
with probability
1
2
. (25)
The particle prepared in this way does not show any inter-
ference phenomena. For example, it is impossible to pre-
pare a superposition with another state such that the result-
ing energy of the particle is denite. Therefore a particle in
the situation (25) is physically different from a particle pre-
pared in the state (2). In fact, there is no single state vector
| that correctly describes the situation (25). We shall now
consider such situations in more detail.
There are two cases when the precise information about
the quantum state is unavailable:
Case 1: The state of a system is known only probabilisti-
cally. For example, our information might be limited to the
knowledge that the system is in a state |
1
with probabil-
ity p
1
and in a state |
2
with probability p
2
. The system
is actually in a denite quantum state, but we just do not
know that state with certainty.
Case 2: A quantum system consists of several parts (sub-
systems) and we are performing measurements only on
one of those subsystems. We would like to describe the re-
sults of these measurements entirely in terms of the state of
the interesting subsystem. However, quantum mechan-
ics ascribes a state | only to the entire system and not to
its constituent parts. Thus a subsystem usually does not
have a denite quantum state. For example, if a system of
two harmonic oscillators is in the quantum state
|
total
= |0 |1 + |1 |2 , (26)
where , = 0 are complex constants, then neither of the
oscillators (taken separately) has a denite quantum state.
In both these cases, it is possible to describe the system
in terms of the density operator (also called the density
matrix and the statistical operator) which we shall dene
in the next subsection. The density operator is not a new
postulate of quantum mechanics but merely a convenient
mathematical construction that simplies calculations in
situations described by Cases 1 and 2. The ordinary for-
malism of quantum states is already capable of performing
computations in all these situations, without introducing
the density operator. To demonstrate the advantage of us-
ing the density operator, we shall rst show some calcula-
tions performed using only the quantum state formalism.
3
1.1.5 Calculations without density operators
There are two main tasks that we now consider: (1) to com-
pute the probability for a measurement outcome

A a, as-
suming that |a is a nondegenerate eigenvector; (2) to com-
pute the expectation value of

A.
Case 1: The system is in the state |
k
with probability
p
k
, where k = 1, 2. In other words, there is a stochastic
ensemble of many identical systems, where a fraction p
k
of
systems are in the state |
k
. For those systems that are in
the state |
1
, we can apply the measurement postulate to
determine the probability for the outcome

A a as
Prob(

A a, |
1
) = ?? |a|
1
|
2
. (27)
In other words, the fraction |a|
1
|
2
of systems that are in
the state |
1
exhibit the value a. The expectation value of

Afor these systems is


1
|

A|
1
. Analogous results are ob-
tained for systems that are in the state |
2
. Now we need
to compute the total probability for

A a; this is the prob-
ability that a system selected at random from the stochastic
ensemble will exhibit the value a. Suppose that the total
number of systems in the ensemble is N. The subset of sys-
tems that exhibit the value a is comprised of: rstly, the Np
1
systems that were in the state |
1
, of which only a fraction
|a|
1
|
2
exhibits

A a; and secondly, the Np
2
systems
that were in the state |
2
, of which only a fraction |a|
2
|
2
exhibits

A a. Therefore the total number of systems that
exhibit

A a is
Np
1
|a|
1
|
2
+Np
2
|a|
2
|
2
, (28)
and the total probability for

A a is
p
1
|a|
1
|
2
+p
2
|a|
2
|
2
. (29)
After similar arguments involving the two groups of sys-
tems, we nd that the total expectation value of

A (aver-
aged over the entire ensemble) is

A = p
1

1
|

A|
1
+p
2

2
|

A|
2
. (30)
Case 2: The system consists of two oscillators, is in the
state (26), and a measurement of energy is performed only
on the rst oscillator. We need to predict the results of this
measurement. The energy of the rst oscillator is described
by the operator

H
1
which acts in the space H
1
. However,
the measurement postulate requires us to consider opera-
tors in the entire two-oscillator state space H
1
H
2
. There-
fore we imagine that the measurement of the energy

H
2
is
also performed on the second oscillator, but the results of
that measurement are ignored. Now we apply the mea-
surement postulate to the observable

H
1


H
2
which is
measured in the state |
total
. It is easy to verify that the
operator

H
1


H
2
has eigenvectors of the form |m |n,
where m, n 0 are integers labeling the occupation num-
ber eigenstates of each oscillator. A measurement outcome
can be described by a pair (E
m
, E
n
), meaning that the rst
oscillator has energy E
m
and the second one has energy
E
n
. According to the measurement postulate, the probabil-
ity for an outcome (E
m
, E
n
) is
Prob
_

H
1


H
2
(E
m
, E
n
) , |
total

_
= |(m| n|) |
total
|
2
= |(m| n|) (|0 |1 + |1 |2)|
2
= ||
2

m0

n1
+||
2

m1

n2
. (31)
However, we are only interested in the probability to ob-
serve an energy E
m
in the rst oscillator. This probability
is equal to a sum over all possible outcomes of the measure-
ment of the second oscillators energy,
Prob
_

H
1
E
m
, |
total

_
=

n=0
Prob
_

H
1


H
2
(E
m
, E
n
) , |
total

_
= ||
2

m0
+||
2

m1
. (32)
Thus (assuming ||
2
+||
2
= 1) the expectation value of

H
1
in the state |
total
is

H
1
=

m=0
E
m
Prob
_

H
1
E
m
, |
total

_
=
1
2
+ ||
2
. (33)
The calculations without the density operator are
straightforward but cumbersome. The advantage of in-
troducing the density operator is that both cases 1 and 2
are described within a single framework, and the resulting
equations are simpler.
1.2 Denition of density operator
The denition of the density operator has two parts, cor-
responding to the two situations described above:
1. If it is known that a quantum system is in the quantum
state |
k
with probability p
k
, where k = 1, 2, ... and all
states are normalized to
k
|
k
= 1, then the density
operator is dened by
=

k=1
p
k
|
k

k
| . (34)
2. If a quantum system consists of two subsystems de-
scribed by Hilbert spaces H
1
and H
2
respectively, and
if the state of the total system is expressed as
|
total
=

k=1
|
k
|
k
, |
k
H
1
, |
k
H
2
,
(35)
where
total
|
total
= 1, then we can dene the density
operator for the rst subsystem as

(1)
=

k=1

l=1

l
|
k
|
k

l
| . (36)
The density operator
(2)
for the second subsystem is
dened similarly (by exchanging and in the above
denition).
4
Remarks:
1. The density operator (36) acts in the space H
1
; the anal-
ogously dened density operator
(2)
acts in H
2
.
2. The states |
k
involved in Eq. (34) are normalized but
do not need to be orthogonal to each other and are not
necessarily a basis. The vectors |
k
and |
k
in Eq. (36)
are not necessarily a basis in their respective spaces H
1
and H
2
and are not necessarily normalized; only the
total state |
total
is normalized.
3. It follows that the vectors |
k
involved in Eqs. (34) and
(36) are not necessarily eigenvectors of the density op-
erator. (They are eigenvectors only if they are all or-
thogonal to each other.)
4. When the vectors |
k
are orthonormal, the second
denition is reduced to the rst one if we assume that
the rst subsystem is in the state |
k
with probabil-
ity
k
|
k
. Below we shall show that a reduction of
the second case to the rst is also possible for arbitrary
|
k
.
1.2.1 Using the density operator
The density operator allows one to compute the probability
distribution for measurement of any observable

A and the
expectation value of

A directly through the formulae
Prob
_

A a,
_
= a| |a , (37)

A = Tr (

A). (38)
(The formula (37) assumes that |a is a non-degenerate
eigenvector of

A with eigenvalue a.) Therefore, if we know
the density operator that describes either of the cases dis-
cussed in Sec. 1.1.4-1.1.5, then Eqs. (37)-(38) yield the cor-
rect answers directly, without going through the long argu-
ments of Sec. 1.1.5. The density operator summarizes the
known properties of a quantum system in all cases where
the state vector cannot be unambiguously dened.
Derivation of Eqs. (37)-(38). The relation (37) follows
from Eq. (38) if we consider the observable

A = |a a|. To
derive Eq. (38), we shall treat each case in the denition of
separately.
Case 1: The system is in the state |
n
with probability p
n
.
For systems that are in the state |
n
, the expectation value
of

A is
n
|

A|
n
. Therefore the total expectation value is

A =

n
p
n

n
|

A|
n
=

n
Tr (

Ap
n
|
n

n
|) = Tr (

A).
(39)
Case 2: The system belongs to a larger total system which
is in a state
|
total
=

k=1
|
k
|
k
. (40)
We consider the hypothetical measurement of the observ-
able

A

1 of the total system. The expectation value of this


observable is equal to the expectation value of

A since any
measurement of

1 always yields the value 1. Therefore

A =
total
|
_

1
_
|
total

=
_

l=1

l
|
l
|
_
_

1
_
_

k=1
|
k
|
k

_
=

l=1

k=1

l
|

A|
k

l
|
k
= Tr (
(1)

A), (41)
where we have used the denition (36) in the last line.
1.2.2 Examples
1. If a system is denitely (with probability 1) in a state
|, the corresponding density operator is
= | | . (42)
This is the projector onto the one-dimensional sub-
space spanned by |. [Check that
2
= in this case.]
2. For a harmonic oscillator which is in the quantum
states |0 or |1 with probabilities
1
2
, the density opera-
tor is
=
1
2
|0 0| +
1
2
|1 1| . (43)
In the basis {|0 , |1 , ...}, this operator can be written
in matrix form as
=
_
_
_
_
_
1
2
0 0
0
1
2
0
0 0 0
.
.
.
.
.
.
_
_
_
_
_
. (44)
3. If the oscillator is in the state |0 with probability
1
2
and
in the state
1

2
(|0 +i |1) with probability
1
2
, the den-
sity operator is
=
1
2
|0 0| +
1
2
1

2
(|0 +i |1)
1

2
(0| i 1|)
=
3
4
|0 0| +
i
4
(|1 0| |0 1|) +
1
4
|1 1| . (45)
In the matrix form,
=
_
_
_
_
_
3
4

i
4
0
+
i
4
1
4
0
0 0 0
.
.
.
.
.
.
_
_
_
_
_
. (46)
4. Suppose that the system of two harmonic oscillators is
in the state (26), where ||
2
+ ||
2
= 1. The density
operator for the rst oscillator is

(1)
= 1|1 |0

0| +2|2 |1

1|
= ||
2
|0 0| +||
2
|1 1| . (47)
For the second oscillator,

(2)
= ||
2
|1 1| +||
2
|2 2| . (48)
5
1.2.3 Equivalence of Cases 1 and 2
The situations we called Cases 1 and 2 are physically
equivalent. The precise meaning of this term is the follow-
ing. Suppose that in a rst experiment a quantum system
X is prepared in a mixed state by specifying the probabil-
ities p
n
for pure states |
n
. According to the denition of
Case 1, this gives rise to a density operator

1
=

n
p
n
|
n

n
| . (49)
Further suppose that in another experiment, the same
quantum system X belongs to a composite system (X, Y )
which is prepared in a state |
total
, yielding a density op-
erator
(1)
according to the prescription of Case 2. Now,
if in both experiments the density operator is the same,

1
=
(1)
, then there is no physical process operating only
on the system X whereby one could distinguish between
the two preparations.
To demonstrate this, it is sufcient to show that the prob-
abilities for measured values of all observables are identi-
cal in both cases. This follows from the formula (37): if two
systems are described by the same density operator , then
the probabilities for all observables will always coincide.
Therefore there is indeed no experimental means of distin-
guishing the cases 1 and 2, as long as the density operator is
the same. The density operator is the only information we
need to describe the state of the system; it is not necessary
to specify whether the system X in the situation of Case 1
or Case 2.
1.3 Further properties
1.3.1 Pure states and mixed states
As we discussed in Sec. 1.1.4, there are situations where a
quantum system cannot be described as being in a denite
quantum state. In that case, the system must be described
by a density operator rather than by a vector |, and one
says that the system is in a mixed state . In contrast, one
says that the system is in a pure state | when there ex-
ists a vector | that describes all measurements performed
on the system. It is important to realize that a mixed state is
not a special mixed vector | but a situation where there
is no state vector | at all. Although one uses the word
state when one says mixed state, this does not mean
that the system is described by a state vector. A phrase such
as the oscillator is in a mixed state is a conventional de-
scription of the situation when no denite state vector is
available. Do not get confused by this terminology.
Since the entire physically measurable information is
contained in the density operator, a mixed state can always
be thought of as a probabilistic mixture of different pure
states (Case 1). To contrast mixed states and pure states,
one says that e.g.
1

2
|a +
1

2
|b (50)
is a coherent superposition of states |a and |b, whereas
=
1
2
|a a| +
1
2
|b b| (51)
is a statistical (or incoherent ) mixture of these states.
Pure states | are represented by density operators
| | and can be thought of as a special kind of mixed
states. However, the following statement shows that in
general a mixed state cannot be adequately represented by
a state vector.
Statement: Some mixed states are not pure states.
Proof: It sufces to present an example of a mixed state
that can be experimentally distinguished from any pure
state. The example is a mixed state with the density
operator
=
1
2
|0 0| +
1
2
|1 1| , (52)
where |0 and |1 are two linearly independent and normal-
ized states. For simplicity we may assume that {|0 , |1} is
a basis of the entire space of quantum states, so that
=
1
2
(|0 0| +|1 1|) =
1
2

1. (53)
We shall now demonstrate that there exists no pure state
| that could correctly predict probabilities for all mea-
surements performed on the system in the state (52).
Assume on the contrary that there exists such a pure state
|, and consider all projection operators

P
|v
= |v v| , (54)
where |v is an arbitrary normalized vector. We shall now
investigate whether the expectation values for measure-
ments of

P
|v
can be reproduced by the state vector |.
The expectation value of

P
|v
in the state (52) is
_

P
|v
_
= Tr
_

P
|v
_
= Tr
_
1
2

1

P
|v
_
=
1
2
Tr

P
|v
=
1
2
. (55)
On the other hand, the expectation value of

P
|v
in the state
| is
|

P
|v
| = |v||
2
. (56)
This expectation value would coincide with Eq. (55) if the
vector | is such that
|v||
2
=
1
2
for all |v . (57)
However, this cannot be true: for any | there always ex-
ists a normalized vector |v such that v| = 0. Therefore
Eq. (57) cannot be satised by any vector |.
It is not necessary to consider innitely many observ-
ables

P
|v
to prove the nonexistence of |. In fact, it is
sufcient to consider only three projectors

P
|v
correspond-
ing to three appropriately chosen vectors |v, for instance
|v
1
= |1, |v
2
=
1

2
(|0 +|1), and |v
3
=
1

2
(|0 +i |1).
One can show that there exists no vector | such that
|v
k
|| =
1

2
for k = 1, 2, 3. Q.E.D.
1.3.2 Observability of phases
Since the probability formula (27) involves only the square
modulus, a measurement of an observable

A yields equal
probabilities for all states e
i
|, where is a real constant.
(The complex number e
i
is called a phase in this context.)
6
However, this does not mean that | and e
i
| are two
identical quantum states. If a quantum system is in a state
|
1
and a superposition is prepared with a state |
2
and
with equal coefcients, then the resulting state is
| =
1

2
|
1
+
1

2
|
2
. (58)
The resulting state | depends on the phase of |
1
because
|

=
1

2
e
i
|
1
+
1

2
|
2
= | . (59)
The difference between the states |
1
and e
i
|
1
cannot
be observed by an immediate measurement of some quan-
tity, but can be found in an experiment involving interfer-
ence of |
1
with another state. Thus, both coefcients
1

2
in the state (58) are in principle observable. The state |
can be experimentally distinguished from another super-
position such as
|

=
1

2
e
i
|
1
+
1

2
e
i
|
2
, (60)
which differs from| only by a choice of phases (assuming
that = ).
This situation should be contrasted with the mixed state
=
1
2
|
1

1
| +
1
2
|
2

2
| . (61)
This incoherent mixture of |
1
and |
2
resembles the
state (58) with all phase information removed. Since the
coefcients in the density operator are always real, a mixed
state has no place for storing the phase information. This
is yet another way to visualize the fact that mixed states
represent a lack of information about the quantum state.
1.3.3 Algebraic properties
It follows (more or less straightforwardly) from the de-
nition (34) that any density operator always satises the
following properties:
1. The operator is Hermitian and diagonalizable.
2. Tr = 1.
3. For any state |, one has 0 | | |.
4. All eigenvalues of are real and belong to the interval
[0, 1].
5. If
2
= , then the system is in a pure state, i.e. there
exists | such that = | |.
6. Tr (
2
) 0 and Tr ( ln ) 0; equality occurs
only when
2
= . (To show this, rst diagonalize .)
1.3.4 Partial trace
As a generalization of the Case 2 above, we may consider
the situation when a composite system is not in a pure state
|
total
but in a mixed state described by a density operator

total
. In that case, the density operators
(1)
,
(2)
for the
rst and the second subsystems can be found directly from
the density operator
total
. This is achieved by an operation
called the partial trace.
The partial trace operation is always applied to an opera-
tor acting in a tensor product space; the result is an operator
acting in one of the spaces.
Denition: If

A is a linear operator in a space H
1
H
2
,
then the partial trace with respect to H
2
is the operator
Tr
H
2

A acting in H
1
that is dened by its matrix elements
with respect to arbitrary vectors |v , |w H
1
as follows,
w|
_
Tr
H
2

A
_
|v =

k=1
(w|
k
|)

A(|v |
k
) , (62)
where {|
k
} is a basis in H
2
. In effect, the trace is taken
only in the space H
2
, while the part of the operator

A that
acts in H
1
is left untouched. This operation is also called
integrating out the H
2
degrees of freedom.
Similarly, one denes the partial trace with respect to H
1
;
the result is an operator acting in H
2
.
For example, a projection onto a vector |a |b is

P
|a|b
= (|a |b) (a| b|) ; (63)
after some calculations, one obtains the partial trace with
respect to H
2
as
Tr
H
2

P
|a|b
= |a a| . (64)
The partial trace with respect to H
1
is |b b|.
If the composite system is in a pure state |
total
, the par-
tial trace can be used to dene the density operator
(1)
as

(1)
= Tr
H
2
(|
total

total
|) . (65)
In the general case when the composite system is described
by a density operator
total
, the density operators for the
subsystems 1 and 2 are expressed directly as

(1)
= Tr
H
2

total
,
(2)
= Tr
H
1

total
. (66)
1.3.5 Basis-free denition of trace
The usual denition of trace of an operator is
Tr

A =

k=1

k
|

A|
k
, (67)
where {|
k
} is a basis. This denition is independent of the
choice of the basis {|
k
}; in particular, {|
k
} does not have
to be an orthonormal basis. The trace can be redened in a
manifestly basis-free manner if we rst rewrite the operator

A as

A =

k=1
|v
k
w
k
| , (68)
where |v
k
and |w
k
are suitable vectors (they do not need
to be orthogonal or normalized). Any operator can always
be written in this form, and there are of course innitely
many possible choices of the vectors |v
k
and |w
k
. Then
the trace of

A is
Tr

A =

k=1
w
k
|v
k
, (69)
provided that this series converges.
The denition (69) does not depend on the way the oper-
ator

A is expressed as a linear combination of the form (68).
(We omit the proof of this statement.)
7
The partial trace can be similarly redened without ref-
erence to a basis. If

A is an operator in the space H
1
H
2
,
it is always possible to nd vectors |t
i
, |u
i
H
1
and
|v
i
, |w
i
H
2
such that

A =

k=1
(|t
i
|v
i
) (u
i
| w
i
|) . (70)
(The vectors do not need to be orthogonal or normalized,
and such a representation can be found in innitely many
ways.) Then the partial trace of

A with respect to H
2
is
Tr
H
2

A =

k=1
w
i
|v
i
|t
i
u
i
| . (71)
A similar formula denes the partial trace with respect to
the space H
1
.
One can show that this denition of the partial trace is
independent of the choice of vectors |t
i
, |u
i
H
1
and
|v
i
, |w
i
H
2
(we omit the proof).
Comparing Eqs. (71) and (36), it is clear that the den-
sity operator
(1)
is the partial trace of |
total

total
| with
respect to H
2
.
2 Evolution of the density operator
2.1 Effect of measurement
2.1.1 Pure states after measurement
According to the quantum-mechanical measurement pos-
tulate, a measurement of an observable

A is always a par-
ticular eigenvalue a of

A. In general, the eigenvalue a
may be degenerate, so that there exist several linearly in-
dependent eigenvectors |a
1
, |a
2
, ..., with eigenvalue a. All
eigenvectors of

A with eigenvalue a belong to a linear sub-
space of H which is called the eigenspace of

A with eigen-
value a. For any eigenspace with eigenvalue a, there ex-
ists a Hermitian projection operator

P
a
called the eigenpro-
jector that satises

P
2
a
=

P
a
and

P
a
|a = |a for all |a
from this a-eigenspace. The operator

P
a
is unique for each
eigenspace of

A and projects the space H onto the corre-
sponding eigenspace (in other words,

P
a
| belongs to the
a-eigenspace for any vector | check this!). If |a
1
, ..., |a
n

is an orthonormal basis in the eigenspace, then the eigen-


projector is expressed by

P
a
=
n

k=1
|a
k
a
k
| . (72)
In the case of a non-degenerate eigenvector |a, the eigen-
projector is simply

P
a
= |a a|.
We now need to generalize the measurement postulate to
the case of degenerate eigenvalues:
1. If the quantum system is in a state |, the probability
to measure a value a of an observable

A is
Prob
_

A a; |
_
=
|

P
a
|
|
, (73)
where

P
a
is the Hermitian projector onto the
eigenspace of

A with eigenvalue a.
2. The state after measurement is

_
=
1
N

P
a
| , N
_
_
_

P
a
|
_
_
_ , (74)
where N is a real-valued normalization constant,
N =
_
|

P
a

P
a
| =
_
|

P
a
|. (75)
2.1.2 Mixed states after measurement
If a quantum system is in a mixed state and a measurement
has yielded a value a of an observable

A, then after the mea-
surement the system is in general again in a mixed state.
The density operator

of the system after measurement is
related to the initial density operator by

=
1
N

P
a

P
a
, N Tr(

P
a

P
a
), (76)
where N is a normalization constant. We shall now derive
this formula.
It is sufcient to consider the initial density operator of
the form
=

n=1
p
n
|
n

n
| , (77)
describing an ensemble of systems in different quantum
states (normalized to
n
|
n
= 1,

n=1
p
n
= 1 ). Before
measurement, the fraction of systems in the state |
n
in
the ensemble is p
n
, and the measurement outcome a oc-
curs with probability
n
|

P
a
|
n
. Among those systems
that exhibit the outcome a, the fraction p
n
of systems in the
state |
n
is
p
n
=
1
N
p
n

n
|

P
a
|
n
, N

n=1
p
n

n
|

P
a
|
n
, (78)
where N is a normalization constant chosen so that

n=1
p
n
= 1. In other words, p
n
is the conditional prob-
ability for a system to be in the n-th sub-ensemble given
that the measurement has yielded the value a.
Note that there may be n for which

P
a
|
n
= 0; since
p
n
= 0 for those n, those systems cannot show the value
a in the experiment and therefore do not contribute to the
resulting ensemble after the measurement. Therefore we
may simply ignore these states |
n
in the calculation. We
shall temporarily write

n
for the summation restricted to
those n for which

P
a
|
n
= 0. The nal formula will be
valid without this technical complication.
For those systems that are in a state |
n
, the resulting
state after measurement according to Eqs. (74)-(75) is

n
_
=

P
a
|
n

n
|

P
a
|
n

. (79)
Therefore the resulting density operator is

n
p
n

n
__

=
1
N

n
p
n

n
|

P
a
|
n

n
__

=
1
N

P
a
_

n
p
n
|
n

n
|
_

P
a
=
1
N

P
a

P
a
. (80)
Already at this point we can replace the restricted summa-
tion

n
by the full summation

n=1
since the terms for
8
which

P
a
|
n
= 0 do not contribute to the sum. Finally,
we check that the normalization constant N is expressed
by Eq. (76),
N =

n=1
p
n

n
|

P
a
|
n
=

n=1
p
n

n
|

P
a

P
a
|
n

= Tr
_

n=1
p
n

P
a
|
n

n
|

P
a
_
= Tr(

P
a

P
a
). (81)
Q.E.D.
2.2 The Liouville-von Neumann equation
In the Schrdinger picture, quantum states | (t) of an iso-
lated system evolve with time according to the Schrdinger
equation. The analogous equation for the density operator
(t) is the Liouville-von Neumann equation,
i
d
dt
= [

H, ]. (82)
It follows that the time dependence of the density operator
can be expressed through the evolution operator

U (t, t
0
) as
(t) =

U (t, t
0
) (t
0
)

U

(t, t
0
) . (83)
To derive Eq. (82), one assumes that the system was pre-
pared at an initial time t = t
0
in the mixed state described
by the initial density operator
(t
0
) =

n=1
p
n
|
n
(t
0
)
n
(t
0
)| . (84)
The mixed state can be interpreted as an ensemble of
systems of which a fraction p
n
are in the quantum state
|
n
(t
0
). Each system of the ensemble is in a denite (pure)
state |
n
(t
0
) satisfying the Schrdinger equation,
i
d
dt
|
n
(t) =

H|
n
(t) , i
d
dt

n
(t)| =
n
(t)|

H.
Since the probabilities p
n
are time-independent (they are
specied when we dene the ensemble at the initial time),
the density operator at time t is
(t) =

n=1
p
n
|
n
(t)
n
(t)| . (85)
Differentiating Eq. (85) with respect to t, we obtain Eq. (82).
3 Applications
3.1 Thermal ensembles
In statistical mechanics, the Boltzmann probability dis-
tribution means that the energy of a physical system
has random values E with probabilities proportional to
exp (E), where T
1
is the inverse temperature. This
distribution describes a system coupled to a thermal reser-
voir at temperature T.
If E
1
, E
2
, ... are the allowed values of energy, the Boltz-
mann distribution can be written as
Prob (E = E
k
) =
1
Z
exp (E
k
) , Z

k=1
exp (E
k
) ,
(86)
where Z is a normalization factor (which is a function of )
called the statistical sum.
Now we consider a quantum-mechanical system that has
an energy spectrum{E
k
} and a basis of energy eigenstates
|E
k
. The Boltzmann distribution becomes an instance of
Case 1 described by the density operator
=
1
Z

k=1
e
E
k
|E
k
E
k
| . (87)
Since by assumption

H =

k=1
E
k
|E
k
E
k
| , (88)
the density operator can be written more concisely as
=
1
Z
exp(

H), Z Tr
_
exp(

H)
_
. (89)
To compute the thermal density operator (89), it is con-
venient to rst calculate the un-normalized operator

u
() = exp(

H), (90)
which differs from the density operator only by the nor-
malization factor Z = Tr
u
. It is easy to see that the opera-
tor
u
satises the differential equation


u
=

H
u
,
u
()|
=0
=

1. (91)
In the coordinate representation, the operator
u
is repre-
sented by a function
u
(x, x

; ) and the Hamiltonian



H is
usually a differential operator in the space variable x. Thus,
Eq. (91) becomes a partial differential equation for the func-
tion
u
(x, x

; ) that can be solved by a variety of methods.


3.2 Entropy
If a quantum system is in a mixed state described by a den-
sity operator , one denes the entropy of the state as
S = Tr ( ln ) . (92)
For example, if a harmonic oscillator is in a mixed state
=

n=0
p
n
|n n| , (93)
where p
n
are probabilities for the occupation number n
such that

n=0
p
n
= 1, (94)
then the entropy of this state is
S =

n=0
p
n
ln
1
p
n
. (95)
The entropy is always non-negative, S 0; it is equal to
zero only if corresponds to a pure state, = | |. The
entropy is interpreted as a measure of our lack of knowl-
edge about the quantum state of the system: our knowl-
edge is complete if the system is in a pure state and then
9
S = 0. Note that the entropy is not an operator but a num-
ber, even in quantum theory, and thus we cannot have an
eigenstate of entropy. The statistical ideas underlying
the notion of entropy are completely independent of the
quantum-mechanical language of operators.
In general, the entropy is unbounded from above. It is
possible to dene (rather contrived and articial) mixed
states that have innite entropy, for instance by choosing
the sequence {p
n
} such that the series (95) does not con-
verge despite the condition (94).
2
A simple calculation shows that the entropy of a mixed
state is time-independent (for an isolated system). In fact,
it is easy to prove that the trace Tr f( ) of any operator-
valued function of is time-independent, as long as (t)
satises Eq. (82) and f

(x) is bounded for x [0, 1]. Evalu-


ating the time derivative, we nd
i
d
dt
Tr f( ) = Tr
_
f

( )[

H, ]
_
= Tr
_
f

( )

H f

( )

H
_
.
Since f

( ) commutes with and Tr(



A

B) = Tr(

B

A) when
both traces are well-dened, we have
Tr
_
f

( )

H f

( )

H
_
= Tr
_
f

( )

H f

( )

H
_
= 0.
Here we have implicitly assumed that the trace of opera-
tors such as f

( )

H is well-dened, which is the case for
bounded f

(x).
3.3 Decoherence
We have derived the conservation of entropy for a closed
quantum system. In the process of interaction of a quan-
tum system with other systems, the density operator usu-
ally changes in such a way that its entropy grows. This
phenomenon is called decoherence. We shall now discuss
decoherence on a qualitative level.
Suppose that a quantum particle with a state space H
1
interacts with another system (the environment ) described
by a state space H
2
. If

H
total
is the Hamiltonian of the total
system acting in the space H = H
1
H
2
, the evolution of
the total state |
total
H is described by the Schrdinger
equation
i
t
|
total
=

H
total
|
total
. (96)
Suppose that initially the total state is a tensor product,
|
total
(t
0
) = | | , | H
1
, | H
2
, (97)
indicating that the particle is initially in a pure state |.
Since Hamiltonian

H
total
involves interactions between the
particle and the environment, at a later time t > t
0
the total
state |
total
will be of the form
|
total
(t) = |
1
|
1
+|
2
|
2
+..., (98)
which is not reducible to a single tensor product. Thus at
later times the particle will generally be in a mixed state,
although the total system (particle plus environment) is at
all times in a pure state |
total
(t).
2
To describe states that have innite entropy, one can introduce the
notion of relative entropy S [
1
;
2
] between two mixed states
1
and
2
,
which is a measure of the information one loses if the system were to pass
from the state
1
to
2
. We shall not consider this issue in detail.
At any time t one can compute the density operator
(1)
describing the particle,

(1)
(t) = Tr
H
2
(|
total
(t)
total
(t)|) . (99)
The resulting density operator
(1)
(t) does not satisfy the
Liouville-von Neumann equation (82) because the particle
is not an isolated system. In other words, there exists no
effective Hamiltonian that could describe the evolution
of
(1)
(t) alone. To compute the density operator
(1)
(t),
one needs to consider the full quantum dynamics of the to-
tal system according to Eq. (96). Then in some cases it is
possible to derive a (rather complicated) equationfor
(1)
(t)
which we cannot consider here.
Since
(1)
(t) does not satisfy Eq. (82), the entropy of the
mixed state
(1)
(t) is in general time-dependent, unlike that
of the isolated system. If the particle is in a pure state at
t = t
0
and in a mixed state at a later time t
1
, it follows that
S(t
0
) = 0 while S(t
1
) > 0. The growth of entropy is a man-
ifestation of decoherence. The information about the state
of the particle is theoretically present in the environment,
but this information is unavailable since we can only per-
form measurements on the particle. For example, a realistic
environment for a charged particle is the electromagnetic
eld. We cannot recover the complete information about
the quantum state of the particle because electromagnetic
waves carrying some information will quickly move away
to spatial innity. Therefore the information is effectively
lost, which is reected by the increased entropy.
However, decoherence is not always irreversible: in prin-
ciple one cannot exclude the possibility that the particle
eventually returns to a pure state (recoherence). One
needs to analyze a particular Hamiltonian

H
total
to reach
denite conclusions as to whether decoherence is irre-
versible in each particular case. (Most often it is irre-
versible.)
10

Вам также может понравиться