2.

Mathematical Formalism of Quantum Mechanics

2.1
2.2

Linear vectors and Hilbert space

Operators
2.2.1 Hermitian operators
2.2.2 Operators and their properties
2.2.3 Functions of operators

Quantum mechanics is a linear theory, and so it is natural that vector spaces play an important role in it. A physical
state is represented mathematically by a vector in a Hilbert space (that is, vector spaces on which a positive-denite
scalar product is dened); this is called the space of states. Physical properties like momentum, position, energy,
and so on will be represented by operators acting in the space of states. We will introduce the essential properties
of Hilbert spaces, mainly in the case of nite dimension, as the mathematical theory of Hilbert spaces of innite
dimension is much more complicated than that of spaces of nite dimension

2.1 Linear vectors and Hilbert space

D: Linear vector space A linear vector space is a set of elements, called vectors, which is closed under addition and

multiplication by scalars.

Using Dirac notation, the vectors are denoted by kets: |k). We can associate to each ket a vector in the dual space

called bra: (|.

If two vectors |) and |) are part of a vector space, then | ) + |) also belongs to the space. If a vector | ) is in

the space, then |) is also in the space (where is a L

complex scalar).

A set of linearly independent vectors {|i )} is such that k ck |k ) = 0 if and only if ck = 0 k (no trivial combination

of them sums to zero).

The dimension of the space N is the maximum number of linearly independent vectors (which is also the smallest

number of vectors that span the space).

D: Basis A maximal set of linearly independent vectors in the space is called a basis. (e.g. {|k )}, k = 1, . . . , N ).

Any vector in the space can be written as a linear superposition of the basis vectors:

|) =

ak |k )

(1)

To any vector we can thus associate a column vector of N complex numbers (a1 , a2 ...an )T . Here we are going to
restrict ourselves to bounded, nite dimension spaces (even if many physical spaces are not: for example energy
spaces can be unbounded and position has innite dimension).

D:

Hilbert space The Hilbert space is a linear vector space over complex numbers with an inner product.

D: Inner product An inner product is an ordered mapping from two vectors to a complex number (for a Hilbert
space a mapping from a ket and a bra to a complex number c = (|)) with the following properties:
positivity: (|) 0. The equality holds only for the zero vector |) = 0.
linearity in the second function: (|(c1 1 ) + c2 |2 )) = c1 (|1 ) + c2 (|2 ).
9

 anti-linearity in the rst function:((c1 1 + (c2 |2 )|(= c1 (1 | ) + c2 (2 | ).

skew symmetry: ( |) = (|)

D:

Norm The norm of a vector is 11 =

(|).
Since the Hilbert space is characterized by its inner product, vectors are dened up to a global phase, that is,
|) = ei |). Relative phase is instead very important: |) + ei |) =
6 |) + |).
The inner product properties allow us to dene two geometric inequalities:
Schwartz inequality: |( |)|2 ( | )(|).
Triangular inequality: 1( + )1 11 + 11.
The equality holds only if the two vectors are in the same direction: |) = c|).
There is also an antilinear correspondence between the dual vectors ket and bra:
c1 |1 ) + c2 |2 ) c1 (1 | + c2 (2 |

D:

Orthonormal set A set of vectors {|k )} is orthonormal if for each pair the inner product (k | j ) = k,j .

2.2 Operators
We can dene a set of operators that acting on the vectors return vectors:

D: Operator An operator A on a vector space is a mapping between two vectors in that space: A|) = |).
A linear operator satises:
A(c1 |1 ) + c2 |2 )) = c1 A|1 ) + c2 A|2 )
To characterize and parametrize A we look at its action on each vector in the space. Because of linearity, it is
however enough to characterize A with its action on the N basis vectors {|)k }. In this way we can associate a
matrix representation to any operator, in the same way we associated arrays of complex numbers with the vectors.
In particular, given an orthonormal basis {|v)k }, the matrix representation of the operator A is an N N square
matrix A whose elements are given by Ak,j = (vk | A |vj ).
LN
Let us consider an orthonormal basis {vi }, then as seen any vector can be written as: |) = i=1 ai |vi ). The action
of an operator A becomes:
N
N
X
X
bi |vi )
A|) = |)
Aai |vi ) =
i=1

i=1

To extract one of the coecients, say bk we multiply by the bra (vk |, obtaining:
N
X
i=1

(vk |Aai |vi ) = bk

Aki ai = bk

The action of an operator can be thus seen as a matrix multiplication (again, here we are restricting to bounded,
nite dimension spaces that support nite operators, hence this simple matrix representation).
? Question: Perform a simple matrix multiplication.

0 1
1 0
0 1
This is equivalent to Rx vz = vy .

0
0
0
1 0 = 1
0
1
0

The domain of an operator is the subspace on which it acts non-trivially (spanned by k N vectors).
10

Two operators A and B are equal if their domains are the same and their action is equal |) in their domains. The
sum and product of operators are then dened as
(A + B)|) = A|) + B|))
(AB)|) = A(B|))
The operators are associative:
A(BC)|) = (AB)C|)
But they are not in general commutative:

D:

AB|) 6= BA|)

Commutator . The commutator of two operators is [A, B] = AB BA. Two operators commute/are commutable

if [A, B] = 0.
2.2.1 Hermitian operators
An important class of operators are self adjoint operators, as observables are described by them.
(

D:

Adjoint The adjoint of an operator A is an operator acting on the dual space with the property: (A ) ) =

D:

Self-adjoint . A self adjoint operator is an operator such that A and A operate on the same domain and with

(|(A)), {|) , |)}.

( We
conjugate) we have (A )
and ((A)|) = (|A |).
transpose of that operator:
the property

can also have other notations. From (|) = (|) (where indicates the complex
) = (|(A)) = (|A ) . Also, we can write the inner product as (|(A)) = (|A|)
In matrix representation, this means that the adjoint of an operator is the conjugate
Ak,j = (k| A |j) = (j| A |k) = Aj,k .

(|A|) = (|A|)

or shortly, A = A. In matrix representation we have then: Aki = Aik .

? Question: Prove that (cA) = c A
We want to prove that (cA) = c A . We can take two strategies:
1) From the adjoint operator denition in the form:
B | = |B ,
with B = cA we obtain:

(cA) | = |cA = c |A = c A | = c A |

2) Alternatively, we can use the adjoint denition in Diracs notation:

|B | = |B| ,
to get:

Note that we can write

|(cA) | = |cA|

= c |A|

= c |A | = |c A |

(B |) = (|B) = (|B|) = (|B |) .

(
The second notation (based on Diracs notation) could be seen as implying (|)) = (| (and thus (A|)) = A .
However, this applies the adjoint operation to a vector, while the adjoint is only properly dened for operators. For
discrete dimensional spaces, which allow a matrix representation, there is no ambiguity since we have the equivalence
of the adjoint with the complex-transpose of an operator (which can be dened also for vectors)4 .
4

See also quant-ph/9907069 page 12, for a subtle dierence between Hermitian and self-adjoint innite-dimensional operators

11

? Question: Prove that (AB) = B A

| i we have |i = (AB)| i h| = h| (AB) . Dene |i = B| i, then |i = A|i, h| = h| B and h| = h| A , so that
h| = h| B A .

A self-adjoint operator is also Hermitian in bounded, nite space, therefore we will use either term. Hermitian
operators have some properties:
1. if A, B are both Hermitian, then A + B is Hermitian (but notice that AB is a priori not, unless the two operators
commute, too.).
2. if A, B are both Hermitian but do not commute, then at least AB + BA is Hermitian.
? Question: Prove property # 2.
(AB + BA) = B A + A B = BA + AB.

Before describing other properties we need the following denition.

D: Eigenvector We dene a right eigenvector as a column vector | )R satisfying A|)R = R | )R , so (A R 11)|)R =

0, which means the right eigenvalues R must have zero determinant, i.e., det(A R 11) = 0. Similarly, a left
eigenvector is such that (|L A = L (|L .

12

The following properties will be very important in QM:

3. if A is Hermitian its eigenvalues are real (eigenvalues: scalar a such that A| i = a|i). It is easy to show this
properties from h |A| i = a = a .
4. distinct eigenvectors of an Hermitian operator are orthogonal: A|1 i = a1 |1 i, A|2 i = a2 |2 i h1 |2 i = 0
unless a1 = a2 .
5. distinct eigenvalues correspond to orthogonal eigenvectors:
6 c2 h1 | 2 i = 0.
Given A |1 i = c1 |1 i and A |2 i = c2 |2 i, if c1 =
As observables are given by Hermitian operators, the first properties will imply that the values that an observable
can take on are only real values (as needed for the observable to have a physical meaning). On the domain of the
operator, the eigenvectors form a complete orthogonal basis set.
? Question: Prove property # 5.
h2 | A1 i = h2 | c1 1 i = hc2 2 | 1 i. For Hermitian operators then c1 h2 | 1 i = c2 h2 | 1 i, which is satisfied only if c1 = c2
or if h1 | 2 i = 0.
? Question: Prove property # 4.
Consider two eigenstates of A |a1 i and |a2 i. We have ha2 | A |a1 i = ha1 | A |a2 i since A is Hermitian. Now ha2 | A |a1 i = a1 ha2 |a1 i
and ha1 | A |a2 i = (a2 ha1 |a2 i) = a2 (ha1 |a2 i) since a2 is real (being an eigenvector of A. We thus have a1 ha2 |a1 i = a2 ha2 |a1 i
which is satisfied iif a1 = a2 (contrary to the hypothesis) or if ha2 |a1 i = 0.

2.2.2 Operators and their properties

D:

The Outer Product |ih| is an operator, since acting on a vector returns a vector: (|ih|) |i = h|i|i.
It defines a projector oP
perator Pi = |vi ihvi |. The sum over all projectors on the space is the identity, therefore, for
any basis set we have: i |vi ihvi | = 11 (closure relation). The product of two projectors is Pj Pk = ik Pj . Projectors
derive their name from thP
e property that they project out a vector component of the related basis vector: given
Pj = |vj i hvj |, Pj |i = Pj k ck |vk i = cj |vj i.
P
D
: Trace - The trace of an operator is the sum of the diagonal elements of an operator Tr {A} = N
j=1 Ajj =
P
j hvj | A |vj i. It is independent of the choice of basis.

D:

Spectral Decomposition - The spectral theorem states that given a self-adjoint operator A on a linear space H,
there exists an orthonormal basis of H consisting of eigenvectors of A. Equivalently, we can state that A can be
written as a linear combination of pairwise orthogonal projections
(which are formed from its eigenvectors). This
P
representation of A is called its spectral decomposition: A = j aj |vj i hvj |, where A |vj i = aj |vj i. In this basis, the
matrix representation of A is diagonal.
 Theorem: If two hermitian operators commute, they share a common set of eigenvectors.

If [A, B] = 0 then AB = BA. Given two eigenvectors of A, we have ha | (AB BA) |a i = a ha | B |a i a ha | B |a i.

This is zero if a = a (and ha | B |a i is a diagonal term of B and it can be anything) or if ha | B |a i = 0 (off-diagonal,
with a 6= a ). Thus B is diagonal in the basis of As eigenvectors, hence As eigenvectors are also eigenvectors of
B.
A simultaneous eigenvector of A and B |a, bi has the property: A |a, bi = a |a, bi and B |a, bi = b |a, bi. The notation
|a, bi is useful when
is degenerate, that is, there exist more than one eigenvector with the same
 the eigenvector

eigenvalue: A a(i) = a a(i) , i = 1, . . . n, where n is the degeneracy. Then the label b serves to distinguish different
eigenvectors.

D:

Unitary operator An operator fulfilling the conditions U U = 11 and U U = 11 is called unitary.

 Theorem: Given two sets of bP

asis kets {|ii } and {|ii } there exist a unitary operator such that |ii = U |ii , i.

(The unitary operator is U =

|k i hk |).

13

2.2.3 Functions of operators

Functions of operators are dened by the corresponding Taylor expansion of the function (if that exists). If f (x) =
1 (n)
1 (n)
f (0)xn + . . . , then f (A) = f (0)11 + f (0)A + + n!
f (0) + f (0)x + + n!
f (0)An + . . . , where the matrix
n
n1
power is dened recursively in terms of products A = A(A
).

? Question: Show that given the

of the
f (A) = a f (a ) |ai ha|.
spectral decomposition

operator A = a a |ai ha|we have

2
2
We can rst prove that A = ( a a |ai ha|)( a a |ai ha|) = a,b a b (|ai ha|)(|bi hb|) = a a |ai ha|. Then show that if the
theorem is valid for n 1 it is also valid for n. Finally, use the Taylor expansion to show its true.

1
n
=
? Question: Consider in particular the exponential function: exp(A) =
n n! (A)
k exp(ak ) |ak i hak |. Prove that
1
1
f (ABA ) = Af (B)A
Its easy to show that (ABA1 )n = AB n A1 by expanding the product and using AA1 = 11. In particular for unitary
matrices U 1 = U f (U AU ) = U f (A)U .

14

MIT OpenCourseWare
http://ocw.mit.edu

22.51 Quantum Theory of Radiation Interactions

Fall 2012

For information about citing these materials or our Terms of Use, visit: http://ocw.mit.edu/terms.