Complex Linear Algebra

The basic mathematical objects in quantum mechanics

are state vectors and linear operators (matrices). Because
the theory is fundamentally linear, and the probability
amplitudes are complex numbers, the mathematics
underlying quantum mechanics is complex linear
algebra. The vectors are members of a complex vector
space, or Hilbert space, with an associated inner
product.
It is important to remember that these abstract
mathematical objects represent physical things, and
should be considered independent of the particular
representations they are given by choosing a particular
basis.

p. 1/31

Dirac notation
State vectors in quantum mechanics are written in Dirac
notation. The basic object is the ket-vector |i, which
(given a particular basis) can be represented as a
column vector. The adjoint of a ket-vector is a bra-vector
h|, represented as a row vector.

) = |i .
|i = ... , h| = (1 N
N

If the vector |i is normalized, that means

h|i =

N
X

|j |2 = 1.

j=1

p. 2/31

Inner and Outer Products

Given two vectors |i and |i,

..
|i = . , |i =
N

1
.. ,
.
N

the inner product between them is written

1
X

) .. =
h|i = (1 N
j j .
j
N

The inner product is independent of the choice of basis.

p. 3/31

h|i is called a bracket. Note that h|i = h|i , and for

a normalized vector h|i = 1. If two vectors are
orthogonal then h|i = 0.

It is also handy to write the outer product (also sometimes

called a dyad):

1
1 1 1 N

..

...
|ih| = ... (1 N
) = ...
.
.
N

N 1

N N

A dyad |ih| is a linear operator. As we shall see, it is

common (and often convenient) to write more general
operators as linear combinations of dyads.

p. 4/31

Linear operators
transforms states to states such that
A linear operator O

O(a|i
+ b|i) = aO|i
+ bO|i

for all states |i, |i and complex numbers a, b. Given a

choice of basis, an operator can be represented by a matrix

a11 a1N
..
.. [a ].
.
=
.
O
.
.
ij
.
aN 1

aN N

The matrix representation depends on the choice of basis.

We will only be dealing with orthonormal bases in this class.

p. 5/31

Expectation values
, we can calculate a
Given a state |i and an operator O
number


h|O|i

hOi
= h| O|i

in the state |i. Given a

which is called the expectation value of O
particular choice of basis, we can express this number in
and |i:
terms of the elements of O
X
=
hOi
i aij j .
i,j

is Hermitian, its expectation value

As we will see, when O
gives the average result of some measurement on a system
in the state |i.
p. 6/31

Matrix elements

Similar to an expectation value is a matrix element h|O|i

where |i 6= |i. If |i and |i are both members of the
orthonormal basis {|ji} then
= aij ,
hi|O|ji
in the
where aij is an element of the matrix representing O
basis {|ji}. The operator can be written as a sum over
outer products,
X
=
O
aij |iihj|.
ij

p. 7/31

Hermitian Conjugation
One of the most important operations in complex linear
algebra is Hermitian conjugation. The Hermitian conjugate
is the complex conjugate of the transpose of an operator
O
. If in a particular basis O
= [aij ] then O
= [a ].
O
ji
Hermitian conjugation works similarly to transposition in
=B
A . When applied to
real linear algebra: (AB)
state vectors, (|i) = h|. Similarly, for dyads
(|ih|) = |ih|.
Note that Hermitian conjugation is not linear, but rather
is antilinear:
= a O
,
(aO)

(a|i) = a h|.

p. 8/31

Orthonormal bases and the trace

It is normally most convenient to choose a particular orthonormal basis {|ji} and work in terms of it. As long as one
works within a fixed basis, one can treat state vectors as
column vectors and operators as matrices. An orthonormal
basis for an N -dimensional space has N vectors that satisfy
hi|ji = ij ,

N
X

|jihj| = I.

j=1

The trace of an operator is the sum of the diagonal elements:

X
X
=
=
Tr{O}
hj|O|ji
ajj .
j

= 0.
A traceless operator has Tr{O}
p. 9/31

The trace is independent of the choice of basis. If {|ji} and

{|k i} are both orthonormal bases, then
X
X
=
=
k i.
Tr{O}
hj|O|ji
hk |O|
j

The trace also has the useful cyclic property

= Tr{B
A}.

Tr{AB}

This applies to products of any number of operators:

C}
= Tr{C AB}
= Tr{B
C A}.

Tr{AB

This invariance implies that Tr{|ih|} = h|i.

p. 10/31

Normal operators
O
=O
O
. Operators are
A normal operator satisfies O
diagonalizable if and only if they are normal. That is, for
we can always find an orthonormal basis
normal O
{|j i} such that
X
X
=
=
O
j |j ihj |, Tr{O}
j ,
j

and any diagonalizable operator must be normal.

and {|j i} the
These values j are the eigenvalues of O
j i = j |j i. If O
is
corresponding eigenvectors, O|
nondegeneratei.e., all the j are distinctthen the
eigenvectors are unique (up to a phase). Otherwise
there is some freedom in choosing this eigenbasis.
p. 11/31

Hermitian operators
One very useful class of operators are the Hermitian
that satisfy H
=H
. These are the complex
operators H
analogue of symmetric matrices. They are obviously
H
=H
2 = H
H
. The eigenvalues of a
normal: H
Hermitian matrix are always real. We will look at this in
more detail later.
We have already seen an example: the Pauli matrices.
!
!
!
0 1
0 i
1 0

X=
, Y =
, Z=
.
1 0
i 0
0 1
These matrices are obviously Hermitian. It is easy to
see that any 2 2 Hermitian matrix can be written
+ cY + dZ for some real values a, b, c, d.
aI + bX
p. 12/31

The Commutator
Matrix multiplication is noncommutative, in general. That
is, in general AB 6= B A. Given two operators A and B ,
B]
AB
B
A. [A,
B]
= 0 if and
their commutator is [A,
only if A and B commute.
Occasionally, one will encounter matrices that
B
= B
A. For example, the Pauli
anticommute: A
matrices anticommute with each other. In these cases,
it is sometimes helpful to define the anticommutator:
B}
AB
+B
A.

{A,

If two normal operators A and B commute, it is possible

to find an eigenbasis which simultaneously diagonalizes
both of them. (The converse is also true.)
p. 13/31

Orthogonal projectors
An orthogonal projector P is an Hermitian operator that
obeys P 2 = P . All the eigenvalues of P are either 0 or
1. The complement of a projector I P is also a projector.
Note that |ih| is a projector for any normalized state
vector |i. Such a projector is called one-dimensional; a
.
projector more generally has dimension d = Tr{P}
In dimension 2, any projector can be written in the form


/2 = |~n ih~n |,
P = I + ~n ~

where ~n is a (real) unit three-vector (nx , ny , nz ) (i.e., with

= (X,
Y , Z)
. In the Bloch sphere
n2x + n2y + n2z = 1) and ~
picture, ~n is the direction in space, and P is the projector
onto the state |~n i that is spin up along that axis.
p. 14/31

The Spectral Theorem

is a normal operator, and j are its distinct
Suppose O
eigenvalues (i.e., not including any repetition) for
j = 1, . . . , M N . There is a unique set of orthogonal
projectors Pj such that
=
O

M
X

j Pj ,

j=1

and Pj Pk = jk Pj , j Pj = I. Such a set of projectors

is called a decomposition of the identity. For a Hermitian
operator, the eigenvalues j will all be real.
that is not repeated, then
If j is an eigenvalue of O
.
Pj = |j ihj | where |j i is an eigenvector of O
p. 15/31

Unitary Operators
A unitary operator satisfies U U = U U = I. It is clearly a
normal operator. All of its eigenvalues have unit norm;
that is, |j | = 1 for all j . This means that
j = exp(ij )

for real 0 j < 2 .

There is a correspondence between Hermitian and
unitary operators: for every unitary operator U there is
such that
an Hermitian operator H
= exp(iH).

U
means shortly.)
(We will clarify what exp(iH)
p. 16/31

We have already seen that a unitary operator is

equivalent to a change of basis. This is easy to check: if
{|ji} is an orthonormal basis, then so is {U |ji}:
(U
|ji) = hi|U
U
|ji = hi|ji = ij .
(U|ii)

For spin-1/2, the most general 2 2 unitary can be

written (up to a global phase)
,
U = cos(/2)I + i sin(/2)~n ~

where ~n is again a real unit three-vector (nx , ny , nz ) and

= (X,
Y , Z)
. In the Bloch sphere picture, U is a
~
rotation by about the axis ~n.

p. 17/31

Operator space
The space of all operators on a particular Hilbert space
of dimension N is itself a Hilbert space of dimension

N 2 ; sometimes this fact can be very useful. If A and B

are operators, so is aA + bB for any complex a, b.
One can define an inner product on operator space.
B)
Tr{A B}
(the
The most commonly used one is (A,
Frobenius or Hilbert-Schmidt inner product). It is easy to see
B)
= (B,
A)
, and (A,
A)
0 with equality only
that (A,
for A = 0. With respect to this inner product, the Pauli
matrices together with the identity form an orthogonal
basis for all operators on 2D Hilbert space.
A linear transformation on operator space is often
referred to as a superoperator.
p. 18/31

Functions of Operators
(where
It is common to write a function of an operator f (O)
f is ordinarily a function on the complex numbers) which is
itself an operator. Usually f (x) is defined by a Taylor series:
f (x) = c0 + c1 (x x0 ) + c2 (x x0 )2 + . For the operator
version, we write
2 +
+ c2 ( O
x0 I)
x0 I)
= c0 I + c1 (O
f (O)

For particular functions and operators, this series can

sometimes be summed explicitly. For instance, for a nilpotent
2 = 0, the series obviously truncates after the
operator O
first-order term (taking x0 = 0). For projectors, P n = P for all
2 = I (such as the
n 1. For idempotent operators obeying O
Pauli matrices), one can sum the even and odd terms
separately.
p. 19/31

 is normal, we simplify by writing O

in diagonal form:
If O
X
X
=
=
O
j |j ihj |, f (O)
f (j )|j ihj |.
j

The most commonly-used function is the exponential

exp(x) = 1 + x + x2 /2! + ,

but others also occur from time to time:

cos(x) = 1 x2 /2 + x4 /4!
sin(x) = x x3 /3! + x5 /5!
log(1 + x) = x + x2 /2 + x3 /3 +
p. 20/31

Polar Decomposition
There are certain special forms in which operators can
always be written. One of these is the polar decomposition. For
, there is a unitary operator U and
any linear operator O
and B
such that
positive operators A
= U A = B
U ,
O
.
which we call the left and right polar decompositions of O
A positive operator is an Hermitian operator with all
nonnegative eigenvalues. In this case, the positive
operators A and B are uniquely given by
p
p
B
O
O,
O
.
A O

is invertible then U is also unique, otherwise not.

If O
p. 21/31

Singular Value Decomposition

we can find unitary operators U and V
For any operator O
with all nonnegative entries,
and a diagonal real matrix D
such that
= U D
V .
O
is unique up to a permutation of the order of
The operator D
are all
the diagonal entries. If the diagonal elements of D
is invertible.
nonzero, then O

p. 22/31

Tensor products
The tensor (or Kronecker) product is a way of combining
two Hilbert spaces to produce a higher dimensional space.
Let |i be a state in a D1 -dimensional Hilbert space H1 and
|i be a state in a D2 -dimensional Hilbert space H2 . Then
we define |i |i to be a state in the D1 D2 -dimensional
space H1 H2 . Such a state is called a product state. Any
state in this larger space can be written as a linear
combination of product states
X
|i =
| i | i

where | i H1 and | i H2 .

p. 23/31

What are the properties of this product?

(a|i + b| i) |i = a|i |i + b| i |i.
|i (a|i + b| i) = a|i |i + b|i | i.

We need also to define bra-vectors, and the inner product:

(|i |i) = h| h|.
(h | h |)(|i |i) = h |ih |i.

p. 24/31

If {|ji1 } is a basis for H1 and {|ki2 } is a basis for H2 then

{|ji1 |ki2 } is a basis for H1 H2 . Given two states in H1
and H2
D1
D2
X
X
|i =
j |ji1 , |i =
k |ki2 ,
j=1

k=1

in terms of this basis

|i |i =

j k |ji1 |ki2 .

j,k

Generic states in H1 H2 are not product states:

X
|i =
tjk |ji1 |ki2 .
j,k

p. 25/31

Operator Tensor Products

If A is an operator on H1 and B on H2 , we construct a
similar product to get a new operator A B on H1 H2 . Its
properties are similar to tensor products of states:
= aA B
+ bA B.

(aA + bA ) B
+ bB
) = aA B
+ bA B
.
A (aB
= A B
.
(A B)
A B
) = AA B
B
.
(A B)(
= Tr{A}Tr{

Tr{A B}
B}.

p. 26/31

We can also apply these tensor product operators to tensor

product states:

(A B)(|i
|i) = A|i
B|i.
= h|A h|B.

(h| h|)(A B)

A general operator on H1 H2 is not a product operator, but

can be written as a linear combination of product operators:
X
=
.
O
A B

Tensor products play an important role in quantum

mechanics! They describe how the Hilbert spaces of
subsystems are combined.
p. 27/31

Matrix representation
What does the matrix representation of a tensor product
look like? If |i has amplitudes (1 , , D1 ) and |i has
amplitudes (1 , , D2 ), the state |i |i in H1 H2 is
represented as a D1 D2 -dimensional column vector:

1 1

1 2

1 |i
.

..
2 |i

|i |i =
=
.

1
D
.
2
..

2 1

D1 |i
..

D1 D2
p. 28/31

Similarly, if A = [aij ] and B = [bij ] then

a1D B
a11 B
1

..
..
.
=
.
A B
.

.
.
.

aD D B
aD1 1 B
1 1

For brevity, |i |i is often written |i|i or even |i. One

can similarly condense the notation for operators; but one
should be very careful not to confuse A B with AB .

p. 29/31

Example: Two Spins

Given two states in the Z basis, |i = 1 | i + 2 | i and
|i = 1 | i + 2 | i, we can write
|i |i = 1 1 | i + 1 2 | i + 2 1 | i + 2 2 | i.

A general state of two spins would be

|i = t11 | i + t12 | i + t21 | i + t22 | i.

As a column vector this is

|i =

t11
t12
t21
t22

p. 30/31

Similarly, operators on two spins can be written as linear

combinations of tensor products of the Pauli matrices and
the identity. For instance,

1 0 0
0
1 0 0 0

0 1 0
0 1 0 0
0

I I =
,
, Z I =
0 0 1 0
0 0 1 0
0 0 0 1
0 0 0 1

=
I X

0
1
0
0

1
0
0
0

0
0
0
1

0
0
1
0

,
X

Y
=

0 0 0 i

0 0 i 0
.
0 i 0 0
i 0 0 0
p. 31/31