Tillmann Berg

[email protected]
Berlin, den 7. Juni 2010

Functional Analysis for Quantum Mechanics

The following text is a transcript of a two hour talk about the mathematical
concepts needed to understand the fundamentals of quantum mechanics. I
gave the talk in March 2010 in the Seminar of the Integrated Research Training
Group of the SFB Raum-Zeit-Materie in Berlin. The intended audience are
Ph. D. students of mathematics and physics. It is not the aim of this script
to give a detailed account of the methods and techniques used or to provide
proofs of the stated theorems. My aim is to explain the motivation and ideas
behind the structures of the mathematics for quantum mechanics.

Functional Analysis
The elementary theory of Linear Algebra describes the structure of finite-
dimensional vector spaces and linear maps on those spaces. While a number
of theorems hold true for infinite-dimensional vector spaces, those theorems
are usually not very useful.
For example, every vector space has a basis and such algebraic bases are
extensively used in finite-dimensional vector spaces. However, most naturally
occurring vector spaces are infinite-dimensional and there are only uncount-
able bases. Those are extremely difficult to find and even if one could describe
such a large basis, it would not be very useful.
To work around those problems without giving up to much of the algebraic
theory and intuition some additional, topological structure is introduced on
the vector space.

Hilbert spaces
In the entire script all vector spaces will be assumed as complex vector spaces.

Definition. A normed vector space (V, k.k) is a vector space with a map k.k : V
R such that

(1) x V : kxk = 0 x = 0,

(2) C, x V : kxk = || kxk and

(3) x, y V : kx + yk kxk + kyk .

The map k.k is called norm. If a normed vector space is complete, meaning that
all Cauchy sequences converge, it is called a Banach space.

Tillmann Berg [email protected]

 A pre-Hilbert space (H, h. , .i) is a vector space H with map h. , .i : H H C
being

(1) Positive: x H\{0} : hx, xi > 0,

(2) Hermitian: x, y H : hx, yi = hy, xi and

(3) Sesquilinear: , C, x, y, z H : hx, y + zi = hx, yi + hx, zi.

The map h. , .i is called scalar product. With the definition kxk B hx, xi each
pre-Hilbert space becomes a normed vector space. If this vector space is a
Banach space, then (H, h. , .i) is called a Hilbert space.

Sesqui means one and a half. The scalar product is sesquilinear in the
sense that it is linear in the second argument and antilinear in the first (meaning
that hx, yi = hx, yi).

Remark. Note first, that there are Banach spaces, which are not Hilbert spaces.
In fact, a Banach space is a Hilbert space, if and only if if satisfies the paral-
lelogram equation kx + yk2 + kx yk2 = 2(kxk2 + kyk2 ). The parallelogram
equation follows easily from the definition of the scalar product, while con-
versely the parallelogram equation allows one to define a compatible scalar
product via polarisation.

Example. The most important Hilbert space in physics is


L2 (Rn ) B f : Rn C | f is measurable and k f kL2 < ,

where the norm is defined by the scalar product

Z
h f , giL2 B f (x) g(x) dx .
Rn

Since the L2 -space is a Hilbert space, it appears more often in mathematics than
the other Lp -spaces. In fact, the L2 -space is the only Hilbert spaces in this class.

In order to preserve some of the methods of linear algebra, while at the

same time using the topological structure of the Hilbert spaces, one defines
the following new concept of basis.

Definition. A topological space is called separable if it contains a countable

but dense subset. Every separable Hilbert space (H, h. , .i) has a Hilbert basis,
which is a countable set {i }iN H with the following properties:

(1) The basis vectors are orthonormal, hi , k i = ik .

P
(2) The set is complete, x H : x = i=1 hi , xii .

In physical literature Hilbert bases are usually called complete orthonormal basis.

Tillmann Berg [email protected]

Remark. It is important to note that a Hilbert basis is not a vector space basis.
Such a basis is defined as a set {i }iI H with the property
X
x H I0 I, #I0 < i C : x = i i .
iI0

While an algebraic basis can be uncountably large, the linear combinations

are always finite. A Hilbert basis, however, is countable but the linear com-
binations may be infinite. To give the term infinite linear combination
P
meaning, one has to use the norm on the Hilbert space: x = i=1 hx, i ii
Pn
means kx i=1 hx, i ii k 0 for n .

Operators
Hilbert spaces correspond roughly to the coordinate or phase spaces of classical
mechanics. In order to construct an entire physical system, one needs some
concept of function or observable.

Definition. Let A and B be two normed spaces. An operator is a linear map

T : A B.

Remark. At this point one should wonder why one uses normed spaces, but
defines operators merely in algebraic terms. For linear maps between normed
spaces the concept of continuity is well-defined. This point is rarely made in
linear algebra courses, since A is assumed to be finite-dimensional and, in this
case, all linear maps are continuous with respect to all norms one could define
on A or B.
However, if A is infinite-dimensional, then linear maps into any normed
space B are in general not continuous.

Lemma. An operator T : A B is continuous, if and only if it is bound,

which means that there is a constant C R with x H : kTxkB CkxkA . The
infimum of those constants is called the norm kTk of the operator T.

Definition. Denote by H0 the set of continuous operators H C and by B(H )

the set of continuous operators H H.

Bra-c-ket
In physics books another notation for vectors and operators is used.

Definition. Physicists like to use the notation h|i B h, i for the scalar
product. Since the scalar product is denoted by brackets, physicists call |i B
H a ket and h| B h, .i H0 a bra. Correspondingly, |ih| B(H ) is
defined via H 3 |i 7 |ih|i H.
While this seems like an unnecessary complication, the notation yields a
simplification if one writes |ii B |i i B i for some Hilbert basis {i }iN .
P P
In particular, one has x H : x = i=1 |iihi|xi and thus id = i=1 |iihi|,
pointwise.

Tillmann Berg [email protected]

Remark. By the Cauchy-Schwarz-inequation, |h, i| kk kk, both h| and
|ih|
n are indeed
o continuous. Also, physicists tend to use the fact that H0 =
h| | H without a comment or proof. This statement is the nonobvious
Riesz representation theorem, which states that the antilinear map H 3 7
h| H0 is an antiisomorphism. (Antilinearity means that h| = h| and
follows from the antilinearity of the scalar product in the first argument.)

Unbounded operators
The Hilbert space most frequently used in physics is the space L2 (R3 ) of
square-integrable functions and very important operators are
 
(1) the position operators: 7 x 7 xi (x)

(2) the momentum operators: 7 xi
and
P 2
(3) the Laplace operator: 7 B .
xi 2

However, it is clear that none of those operators is well-defined on the entire

L2 -space and even on those subspaces where they are well-defined, they are
generally not bounded. To see this consider the momentum operator in one
variable on the interval [0, 1]:
d 2
: L [0, 1] C1 [0, 1] L2 [0, 1] .
dt
The sequence t 7 tn is in C1 [0, 1] and there clearly is no constant C R such
that one has
d n 1
k tn kL2 = C = Cktn kL2 .
dt 2n 1 2n + 1
Thus, the momentum operator is not continuous (with respect to the chosen
norms).
Since a mathematical theory of operators can describe quantum mechanics
only if it can handle those operators, one has to define more general operators
than everywhere defined, bounded operators.
Definition. An unbounded operator (T, dom T ) on a Hilbert space H consist
of a linear subspace dom T H, called the domain of T, and a linear map
T : H dom T H .
An unbounded operator is called densely defined if dom T H is dense. An
unbounded operator is called closed, if its graph is closed as a subset in H H.

The Spectrum
Definition. For a densely defined, unbounded operator T, the resolvent set is
defined as
n o
(T ) B C | ( T )1 exists and is bounded

and (T ) B C\(T ) is called the spectrum of T.

Tillmann Berg [email protected]

To understand the following decomposition of the spectrum in three subsets a
comparison with the finite-dimensional case is helpful. If T is an operator on
a finite-dimensional Hilbert space, then the spectrum is the set of eigenvalues.
 
(T ) = C | T is not injective = C | T is not surjective
However, on infinite-dimensional Hilbert spaces injective endomorphisms
need not be surjective and vice versa. Also, the map T might be invertible
but the inverse could be noncontinuous. Those difficulties suggest a more
detailed analysis of the spectrum.
First, if T is a densely defined operator and if T is not closed, then (T ) = C.
Thus, nonclosed operators do not have an interesting spectrum and it will be
assumed that T is closed. In this case the continuity of ( T )1 follows
automatically from the existence of the inverse (closed graph theorem).
Assuming that T is a densely defined, closed operator one defines the
following decomposition of the spectrum:

(T ) = C | T is not injective (point spectrum)

= C | T is injective, not surjective with dense image
(continuous spectrum)

= C | T is injective, does not have dense image
(residual spectrum)

The spectrum is a closed subset of C. Typically (but not always), the point
spectrum consists of isolated points, the continuous spectrum is a union of
intervals and the residual spectrum is empty. Also, there usually is a Hilbert
basis, which consists of eigenvectors.
If T is bound, then , (T ) { C | || kTk}. But if T is unbounded,
then = (T ) or (T ) = C are possible. In particular, one needs to restrict the
set of unbounded operators under consideration.

Adjoint Operators
Definition. An unbounded operator T is symmetric, if and only if x, y
dom T : hTx, yi = hx, Tyi.

Remark. For symmetric T the point spectrum (i. e. the set of eigenvalues)
consists of real numbers. One might be tempted to conclude that (T ) R.
Indeed, in the physical literature this conclusion is regularly used. For instance,
many physics books assume that the spectrum consists only of eigenvalues
and is therefore real. However, this is false:
Consider the momentum operator
d 2
i : L (R) C 2
0 (R) L (R)
dt
where C 0
(R) is the set of smooth functions R C with compact support.
By partial integration this operator is symmetric with respect to the L2 -scalar
d
product. But the choice of domain implies that i dt is never surjective,
  
d 2 d
i dt C0 (R) C0 (R) , L (R). Therefore, (i dt ) = C.

Tillmann Berg [email protected]

 Obviously, the size of the spectrum depends on the chosen domain. The
example shows that the set of compactly supported, smooth functions is overly
small for the momentum operator. But the example also shows that a more
restrictive property than symmetry is needed.

Definition. For every bound operator T B(H ) there exists one and only one
bound operator T? B(H ) with

x, y H : hTx, yi = hx, T? yi .

T? is called the adjoint of T.

Lemma. For S, T B(H ) and , C one has (S + T ) = S + T ,

kT k = kTk, (S T ) = T S , T = T and (T1 ) = (T )1 , if T is invertible.

Definition. If (T, dom T ) is a densely defined, unbounded operator the adjoint

(T , dom T ) is an unbounded operator defined by

dom T B x H | y H z H : hTz, xi = hz, yi and

T x B y.

Note that the dense domain is crucial at this point. For otherwise y would not
be uniquely determined.

Remark. For unbounded operators the nice formulae of the previous lemma
are generally not true: Even if S and T are densely defined, the sum S + T is
only defined on dom S dom T, which can be {0}. Thus, in general S + T does
not have a well-defined adjoint. Similarly, it is possible that dom T = {0}.

Remark. A densely defined operator T is symmetric, if and only if T T ,

which means that dom T dom T and T | dom T = T.

Definition. An unbounded operator T is self-adjoint, if and only if it is densely

defined and T = T .

Definition. An unbounded operator T is closable, if the closure of the graph

of T is the graph of another operator T. This operator T is called the closure of
T, if it exists. T is essentially self-adjoint, if T exists and if T is self-adjoint.

The concept of self-adjointness is more restrictive and one can prove the
following lemma.

Lemma. If T self-adjoint, then (T ) R.

Remark. Why can physicist (usually) ignore the difference between symmet-
ric and self-adjoint operators? It was shown before that the spectrum of the
d
symmetric momentum operator i dt is the entire complex plane C. However,
the momentum operator is essentially self-adjoint and one may implicitely use
the closure of the momentum operator. This operator then has real spectrum,
as it is required by the axioms of quantum mechanics.

Tillmann Berg [email protected]

Spectral Theorem
From Linear Algebra courses it is well-known that a self-adjoint operator
T B(Cn ) over a finite-dimensional Hilbert space can be decomposed as
X
T= i Ei

where the i are the eigenvalues of T and the Ei the corresponding orthogonal
projections on the eigenspaces. An orthogonal projection is an idempotent
map, E2i = Ei , with ker Ei im Ei .
The aim of this section is to find a similar decomposition for unbounded
operators on infinite-dimensional Hilbert spaces. This can be done indirectly
via the functional calculus, which is of great interest in itself. Again, Linear
Algebra tells us, that for finite-dimensional vector spaces and a polynomial f
one has
X
f (T ) = f ( i ) Ei .

This can be generalised to bounded operators on infinite-dimensional Hilbert

spaces as follows.

Definition (Measurable functional calculus).

Let T B(H ) be a bounded operator and B((T )) the set of measurable,
bounded functions f : (T ) C. One can show, that there is one and only
one map

B((T )) 3 f 7 f (T ) B(H )

defined by:

(1) 1(T ) = id, id(T ) = T

(2) It is an involutive algebra homomorphism. (Meaning that it respects addi-

tion, multiplication and conjugation/adjoint of the two C -algebras B((T ))
and B(H ).)

(3) It is continuous.

(4) fn B((T )), supk fn k < , fn (t) f (t) implies h fn (T )x, yi h f (T )x, yi.

With this functional calculus one can define the necessary generalisation
of the orthogonal projections Ei to the infinite-dimensional case.

Definition. Let denote the Borel--algebra in R. The spectral measure of a

bounded operator T B(H ) is the map

E : B(H ) with A 7 A(T) (T ) .

This is a projection-valued measure as it satisfies:

(1) For each A the operator EA is an orthogonal projection.

(2) E = 0 and ER = id.

Tillmann Berg [email protected]

(3) E is -additive:
P
If A1 , A2 , . . . are pairwise disjoint, then i=1 EAi = E
S .
i=1 Ai

Definition. One can define the integral over bounded measurable functions
B(R) with respect to a projection-valued measure E in the usual way via
step-functions and a completeness argument:
P
Let f = i Ai be step-function with Ai . Define
Z X
f dE B i EAi .
R
R
One can show that k R f dEk k f k , where k.k is the norm on B(R). This
estimate implies that for any approximating
R sequence of step-functions fn
f B(R) the sequence of integrals R fn dE is a Cauchy sequence. Since H is a
Banach space, B(H ) is a Banach space as well, whichR means that this Cauchy
sequence converges to some bounded operator R f dE B(H ).
This implies that there is a well-defined, involutive algebra homomorphism
Z
B(R) 3 f 7 f dE B(H ) .
R

Note that if E Ris the spectral measure of some bounded operator T B(H ),
then the integral R f dE depends only on the values of f on (T ).
Theorem (Spectral Theorem for self-adjoint, bounded operators).
Let T B(H ) be a self-adjoint, bounded operator and f B(R). If E is the
spectral measure of T, then
Z Z
T= dE and f (T ) = f () dE ,
R R
R R
where R dE is a simpler notation for R ( 7 ) dE and where f (T ) is
defined by the measurable functional calculus.

The spectral theorem for unbounded operators is slightly different, since

one has to take domains into account. Also, one cannot use a functional
calculus. Indeed, the spectral theorem is used to define such a calculus.
Note that for each projection-valued measure E the map 3 A 7 hEA x, yi
is a complex measure for all x, y H.
Theorem (Spectral Theorem for self-adjoint, unbounded operators).
If T : H dom T H is a self-adjoint, unbounded operator, then there exists
a uniquely determined projection-valued measure E, such that
Z
x dom T, y H : hTx, yi = dhE x, yi .
R
If f : R C is measurable, then
Z
h f (T )x, yi B f () dhE x, yi with
(R Z )
2
dom f (T ) B x H | | f ()| dhE x, xi <
R

Tillmann Berg [email protected]

defines an unbounded operator. If f is bounded, then this operator is bounded.
If f is real valued, then this operator is self-adjoint.

Example. If A is a self-adjoint, unbounded operator, then for every t R

one has a well-defined, bounded operator eitA with eitA eisA = ei(t+s)A and
(eitA ) = eitA .

Operator Semigroups
Quantum mechanical systems are described by operator equations, usually
partial differential equations. In particular, some results on the existence of
solutions for such equations are important.
The equation

u0 = Au with the initial condition u(0) = u0

has a unique solution u(t) = etA u0 . The aim of this section is to give meaning
to this solution, if A is some operator.

Definition. If X is a Banach space, then a strongly continuous one-parameter

semigroup or C0 -semigroup is a family {Tt : X X}t0 of continuous operators
such that

(1) T0 = id,

(2) Ts+t = Ts Tt and

(3) limt0 Tt x = x for all x X.

Remark. The second axiom implies that all operators Tt commute. Also, the
maps [0, ) X 3 (t, x) 7 Tt x X and u : R+ 3 t 7 Tt u0 X for some u0 X
are continuous.

In order to consider differential equations one needs some concept of differ-

entiation. The following definition gives a first derivative of a C0 -semigroup
at t = 0:

Definition (Infinitesimal generator).

The infinitesimal generator of a C0 -semigroup is the unbounded operator

Th x x
Ax B lim on the domain
h0 h
dom A B {x H | This limes exists.}

Example. If A is some operator such that a functional calculus is defined (for

example A B(Cn )), then the infinitesimal generator of the C0 -semigroup etA
is A.

Lemma. The infinitesimal generator of a C0 -semigroup is always closed,

densely defined and satisfies Tt A = A Tt .

Tillmann Berg [email protected]

Theorem. If A is the infinitesimal generator of a C0 -semigroup {Tt } and u0
dom A, then u(t) B Tt u0 is smooth map R+ X with im u dom A. It is the
unique solution of the differential equation

u0 = Au with u(0) = u0

and the map X dom A 3 u0 7 u(t) X is continuous for all t R+ .

A corollary of this theorem is, for instance, that two C0 -semigroups with the
same infinitesimal generator are equal. Another related theorem is regularly
used in quantum mechanics:

Theorem (Stone).
A C0 -semigroup of unitary operators (meaning Tt = Tt1 for all t R+ ) is
equal to {eitA } for some self-adjoint operator A.

How does one use those theorems in quantum mechanics? Normally, one
is given some (essentially) self-adjoint, differential operator A (such as the
Laplace operator) and seeks a solution to the equation

iu0 = Au .

Such a solution is the map t 7 eitA u0 , where the C0 -semigroup eitA is defined
via the functional calculus.
A very important fact for the physical interpretation is the preservation
of the norm, for all t R one has ku(t)k = ku0 k, since the C0 -semigroup eitA
consists of unitary operators.

Bibliography
[Wer07] Dirk Werner. Funktionalanalysis. Springer, Berlin; 6. Auflage, Septem-
ber 2007.

[RS03] Michael Reed and Barry Simon. Methods of Modern Mathematical

Physics, Volume 1: Functional Analysis. Academic Press; rev. and enl.
ed, 2003.

Tillmann Berg [email protected]