Vol. XX (XXXX) No.

CHANGE OF REPRESENTATION AND THE RIGGED HILBERT

SPACE FORMALISM IN QUANTUM MECHANICS

Nadia Boudi
CeReMAR Center, Laboratory LMSA,
Faculty of Sciences P.O. Box 1014, Mohammed V University in Rabat,
Rabat, Morocco
nadia [email protected]

Zakariae Ennadifi ∗
CeReMAR Center, Laboratory LMSA,
Faculty of Sciences P.O. Box 1014, Mohammed V University in Rabat,
Rabat, Morocco
[email protected]
(Received 2019)

Generalized eigenvectors are key tools in the theory of rigged Hilbert spaces.
Let H be a Hilbert space and let Φ be a dense subspace of H. Let A be a densely
defined linear operator in H such that Φ ⊂ DA and AΦ ⊂ Φ. The generalized
eigenvectors of A are the eigenvectors of the algebraic dual of A |Φ . In the case
where Φ is endowed with a topology τ finer than the norm topology inherited from H,
generalized eigenvectors that are τ -continuous may be of great interest. We discuss
conditions which ensure the existence of representations associated to generalized
eigenvectors of A. As an application, we review and refine Böhm’s study of the
algebra of the quantum harmonic oscillator.

Keywords: Generalized eigenvectors; rigged Hilbert space; change of representation;

Weyl algebra; quantum harmonic oscillator

1. Introduction
The rigged Hilbert space (RHS) is a generalization of the following construction from
Distribution theory
S(Rn ) ⊂ L2 (Rn ) ⊂ S × (Rn ),
where S(Rn ) is the Schwarz space and S × (Rn ) is the space of tempered distributions.
The RHS theory was developed in the 1950s, mainly by Gel’fand, Kostutchenko, Vilenkin,
∗
Supported by the National Center for Scientific and Technical Research (CNRST, Morocco) as part
of the research excellence scholarship program

[1]
 [Author and title] 2

Shilov and Maurin [2, 3, 1]. A cornerstone of the theory is the nuclear spectral theorem
[4, 2, 5], which refines von Neumann’s spectral theorem [6]. The use of the RHS for a
rigorous mathematical formulation of Dirac’s formalism was suggested and investigated
by several authors, in many papers. Among them we mention the works of Foias [7],
Roberts [8, 9], Böhm [10] and Antoine [11]. The RHS theory allows us to associate to
every quantum system a Hilbert space H, a construction (or more [12, 13]) of the form
(Φ ⊂ H ⊂ Φ× , A) where Φ is a dense linear subspace of H endowed with a locally con-
vex topology τ finer than the topology induced by the norm of H, Φ× is its topological
antidual space and A is an algebra of continuous operators on Φ. For more details on
the rigged Hilbert space formalism see for instance [14, 12, 16, 10, 17, 18, 15, ?] and
references therein. The basic tools of the theory are the generalized eigenvectors of self-
adjoint or unitary operators which leave the space Φ invariant. Recall that if A is a linear
operator in H that leaves Φ invariant, a generalized eigenvector of A is an eigenvector of
the algebraic dual of the operator A |Φ , which is continuous with respect to the topology
τ . We refer the reader to [19, 20] for conditions that ensure tight rigging of the operator
A, where the spectrum of the operator A coincides with an adequate set built from gen-
eralized eigenvalues of A. For the study of other constructions that generalise the RHS,
see, e.g., [14, 21, 22]. Finally, let us mention the work of Bergeron [23], where another
explanation of Dirac’s formalism is proposed.
In practice, in the study of a quantum system, we may consider observables and other
fundamental operators. To every fundamental operator (or a commuting family of fun-
damental operators) we may associate an Hilbert space and a representation. The goal
of any change of representation is to facilitate calculations and understanding of the
system. The following two problems arise naturally: Given an initial representation of
the system and an associated algebra of operators containing fundamental observables
of the system, how to construct new useful representations? How to identify all the
fundamental operators of the system? Important examples of representations are the so
called ”orthonormal representations” which are associated to self-adjoint operators, like
the well known discrete basis {|ni}n , or continuous bases {|xi}x and {|pi}p [24, 26, 25]
or the representations associated to coherent states, these are representations associated
somehow to the eigenvectors of some lowering operator [25, 27].
In this paper we are concerned with the existence of representations realized with a fam-
ily of generalized eigenvectors of an operator. Given a Hilbert space H, we mainly deal
with constructions of the form Λ ⊂ H ⊂ Λ∗ , where Λ is a dense linear subspace of H
and Λ∗ is its algebraic antidual space, and we consider linear operators that leave Λ
invariant. Observe that this may allow a greater generality since Λ can be constructed
using some physical states, and we can define a topology on Λ that makes some general-
ized eigenvectors continuous. At this stage, let us mention that the topology cannot be
obtained from physical data, see for instance [10, p. 4]. We first show some simple and
basic results on algebraic generalized eigenvectors of certain unbounded linear operators
in H. Then we discuss conditions that ensure the existence of associated representations.
This may be considered as a practical method for finding new representations, and more
fundamental operators. We may classify elements of the algebra A of operators on Λ by
the behavior of their generalized eigenvectors. If Λ is endowed with a topology τ , we may
 [Author and title] 3

define coherence like properties in the sense of Klauder-Skagerstam [27] with respect to
τ . In this paper we touch only a few aspects of the subject, the idea we wish to convey
is that generalized eigenvectors arise naturally in many situations. On the other hand,
we believe that the use of generalized eigenvectors may enrich our understanding of the
expansion of a state into eigenfunctions of hermitian or non-hermitian operators, see for
instance [28, 25, 29].
As an application, we review and refine Böhm’s algebraic study of the quantum harmonic
oscillator [10] (see also [30, 13]). This is a basic, theoretical and elementary quantum
system, without boundary conditions. We focus on the study of the change of represen-
tations. Hence we use the number representation, which can be constructed using the
algebraic span of the eigenstates of the Hamiltonian in any representation and construct
other representations, by using the generalized eigenvectors of suitable operators. As it
was shown by Böhm, the rigged Hilbert space structure Φ ⊂ H ⊂ Φ× arise naturally. In
particular, we recover well known results by using elementary methods, without using
the nuclear spectral theorem. Indeed, the concept of generalized eigenvectors allows a
rigorous and elementary mathematical study of this basic example. The position, the
momentum and the Fock-Bargmann representation are respectively associated to gener-
alized eigenvectors of the position, momentum and creation operators. The |xi and the
|pi vectors do not lie in the Hilbert space. However, with respect to the topology of Φ,
we see that they exhibit coherence like properties in the sense of Klauder-Skagerstam.

2. Preliminaries
Let Λ be a complex vector space and let T : Λ → Λ be a linear map. The spectrum
of T is
σ(T, Λ) = {λ ∈ C : T − λ is not bijective}.
We denote by Λ∗ the dual space of Λ, elements of Λ∗ are the linear functionals F : Λ → C.
Denote by T ◦ : Λ∗ → Λ∗ the dual of T , which is defined by T ◦ F = F T . It is easy to
see that if the range of T has finite codimension n, then the dimension of Ker T ◦ is
equal to n. If F is an eigenvector of T ◦ with respect to the eigenvalue λ, we say that
(F, λ) (or F ) is a generalized eigenvector of T and λ is a generalized eigenvalue of T .
Suppose that Λ is endowed with a topology τ , then we denote by Λ∝ its topological dual.
Suppose moreover that Λ is endowed with a scalar product which is τ -continuous in each
variable. Then Λ ⊆ Λ∝ via u 7→ hu, .i (we assume that our scalar products are antilinear
in the first argument). Denote by H the Hilbert space completion of (Λ, h., .i). Clearly
we have H ⊆ Λ∝ . Since we want to deal with linear embeddings, we shall also deal with
antidual spaces. We denote by Λ∗ the algebraic antidual space of Λ, and by Λ× the
topological antidual space (with respect to τ ). Clearly, elements of Λ∗ (resp. Λ× ) are of
the form F where F ∈ Λ∗ (resp. Λ∝ ) and F u = F u for every u. Analogously, we define
T n : Λ∗ → Λ∗ by T n F = F T and T × : Λ× → Λ× by T × F = F T . We have
Λ ⊂ H ⊂ Λ× ⊂ Λ∗ .
Moreover, let T ∗ be the the Hilbert adjoint of T , then it is easy to see that
DT ∗ = {u ∈ H : T n u ∈ H} = (T n )−1 (H) ∩ H, T ∗ = T n |DT ∗ .
 [Author and title] 4

For F ∈ Λ∗ and ϕ ∈ Λ, we shall use the notations F (|ϕ)) = (ϕ, F i = hF |ϕ). We shall
deal with topological, algebraic, linear or antilinear generalized eigenvectors. The words
”linear, antilinear, topological and algebraic” can be omitted when there is no risk of
ambiguity.
Let Λ, Ω be complex linear spaces and let T : Λ → Ω be a bijective linear map. Then it
is clear that Ω∗ = {F T −1 : F ∈ Λ∗ }. In particular, if Λ and Ω are topological spaces
and if T is an homeomorphism, then Ω∝ = {F T −1 : F ∈ Λ∝ }. Let H be a complex
1
separable Hilbert space, we write k.k = (., .) 2 for its norm. We may also use the notation
h.|.i for the scalar product. By a linear operator in H we mean an unbounded linear
operator A : DA → H where DA is a dense linear subspace of H. Suppose moreover
that A is closed. The spectrum σ(A, H) of A is the complement in C of the resolvent set
ρ(A), where
ρ(A) = {λ ∈ C; A − λ : DA → H is bijective and (A − λ)−1 is bounded}.
Next we fix some general notational conventions. If J is a subset of a topological space,
its closure is denoted by J. Let (X, µ) be a measure space, as usual L2 (X, µ) denotes
the space of functions
Z
L2 (X, µ) = {ψ : X → C, ψ measurable and |ψ(x)|2 dµ < ∞},
X

while L2 (X, µ) denotes the corresponding Hilbert space. It consists of equivalence classes
of measurables functions in L2 (X, µ) which are equal almost everywhere. We shall denote
the Lebesgue measure by dx, and we shall write k.k2 for the norm of L2 spaces. Since
there is no risk of ambiguity, we shall use the same notation for f ∈ L2 (X, µ) and its
corresponding equivalence class. Finally, for any linear map T defined on a linear space
having a countable basis {en }n , we denote by M (T, {en }n ) the matrix of T with respect
to the basis {en }n .

3. Generalized eigenvectors and associated representations

Throughout this section, X is a subset of C and H is a Hilbert space. To deal freely
with evaluation maps, we first consider spaces of functions in a general sense. We shall
say that a complex linear space Ω is a space of functions on X if all elements of Ω are
functions f : X → C; in particular, f = 0 if and only if f (x) = 0 for every x ∈ X.
Our discussion is based on the following elementary lemma.
Lemma 1. Let Ω be a space of functions on X, f : X → C a function and let
A : Ω → Ω be a linear map. Then A is defined by Aψ(x) = f (x)ψ(x) for all ψ ∈ Ω if
and only if for every x ∈ X, either δx is a generalized eigenvector for A with respect to
f (x) or δx = 0.
Proof. This is straightforward. Just observe that for every ψ ∈ Ω and x ∈ X,
Aψ(x) = f (x)ψ(x) ⇔ δx (Aψ) = f (x)δx (ψ) ⇔ (A◦ δx )(ψ) = f (x)δx (ψ).
 [Author and title] 5

3.1. Completeness
From now on, in this section we suppose that A is a linear operator in H, Λ is a dense
linear subspace of H such that Λ ⊂ DA and AΛ ⊂ Λ. Let {(Fx , f (x))}x∈X be a family
of generalized eigenvectors of A |Λ . For every ϕ ∈ Λ, consider the function ϕ e defined on
X by ϕ(x)
e = Fx (ϕ) and put Ω = {ϕ e : ϕ ∈ Λ}. Then it is easy to see that Ω is a space of
functions on X and that for every x ∈ X, δx 6= 0. Moreover, the map T : Λ → Ω defined
by T ϕ = ϕe is linear and surjective by construction.
Here and subsequently, we suppose that δx 6= 0 for all x ∈ X in every space of functions
on X. Let us consider the following general definition of completeness (see [2, p.120]).
A family {Fλ }λ of elements of Λ∗ (or Λ∗ ) is said to be complete if for every ϕ ∈ Λ,
Fλ (ϕ) = 0 for every λ implies that ϕ = 0.
Lemma 2. The following assertions are equivalent:
(i) The family {Fx }x∈X is complete.
(ii) The map T : Λ → Ω defined by T ϕ = ϕ e is injective.
In this case, for every ψ(x) ∈ Ω, we have T AT −1 (ψ(x)) = f (x)ψ(x). Moreover, if Λ is
a topological space, and Ω is endowed with the topology induced by T , then for every x,
the evaluation map on Ω is continuous if and only if Fx is continuous.

Proof. The implication (ii) ⇒ (i) is obvious. Let us show that (i) ⇒ (ii): Suppose that
for some ϕ ∈ Λ, T ϕ = 0. Then ϕ(x)
e = 0 for every x. That is, Fx (ϕ) = 0 for every x. By
completeness of the family {Fx }x , we get ϕ = 0. Hence T is injective.
Next suppose that Λ is a topological space and that Ω is equipped with the topology
induced by T . Then the map T is continuous. Since δx = Fx T −1 , we obtain the desired
conclusion.

3.2. Coherence-like conditions

In [27], Klauder and Skagerstam observed that coherent states share two basic prop-
erties, strong continuity and completeness. Next we generalise those two properties to
the case of a family of generalized eigenvectors. We shall say that {Fx }x∈X is strongly
continuous if for every sequence {xn } ⊆ X satisfying limn xn = x where x ∈ X, we have
limn Fxn (ϕ) = Fx (ϕ) for every ϕ ∈ Λ. On the other hand we shall say that {Fx }x∈X
is measure-complete if there exists a positive measure µ on X, a closed subspace K of
L2 (X, µ) and a unitary map U : H → K such that (U ϕ)(x) = Fx (ϕ) for every ϕ ∈ Λ.
Observe that this is equivalent to the fact that there exists a Hilbertian basis {en }n of
H contained in Λ and a positive measure µ such that the functions {Fx (en )}n form an
orthonormal family of L2 (X, µ). In the case that Fx ∈ H for every x ∈ X, we recover
Klauder-Skagerstam’s conditions.

The next proposition provides a criterion of completeness for a family of generalized

eigenvectors of A. It may be used to construct new representations using generalized
eigenvectors of certain operators.
 [Author and title] 6

Proposition 1. Suppose that {Fx }x∈X is measure-complete. Then it is complete.

Suppose moreover that Λ is endowed with a metric topology τ finer than that of the
norm of H such that the maps Fx are τ - continuous, and let Λ0 be the completion of
(Λ, τ ). Then the family {F
fx }x∈X is complete, where F
fx is the continuous extension of
0
Fx to Λ .
Proof. Let ϕ be an element of Λ and suppose that for every x, Fx (ϕ) = 0. Let µ be a
positive measure on X associated to the measure-complete family {Fx }x and let K be a
closed subspace of L2 (X, µ) such that there exists a unitary map U : H → K satisfying
(U ϕ)(x) = Fx (ϕ) for every ϕ ∈ Λ. Then (U ϕ)(x) = 0 for every x. That is, U ϕ = 0.
Since U is unitary, we get ϕ = 0. Next let us prove the second assertion. Let ϕ ∈ Λ0 and
suppose that for every x, Ffx (ϕ) = 0. In the case that ϕ ∈ Λ, we have F fx (ϕ) = Fx (ϕ)
0
and we are done. Next suppose that ϕ ∈ Λ \ Λ and consider a sequence {ϕn }n ⊆ Λ such
that limn ϕn = ϕ for the topology τ . Since τ is finer than the topology induced by the
norm k.k on Λ, the sequence {ϕn }n converges to ϕ in H. We have (U ϕn )(x) = Fx (ϕn )
for every x and in L2 (X, µ), the sequence {U ϕn }n converges to U ϕ, in particular, a
subsequence of {U ϕn } converges pointwise almost everywhere on X to U ϕ. On the
other hand, since the maps Fx are continuous, limn Fx (ϕn ) = F fx (ϕ) = 0. This entails
that the map (U ϕ)(x) = 0 almost everywhere, that is U ϕ = 0. Finally, we infer that
ϕ = 0.

3.3. Construction of the representation

Let A be a linear operator in H and suppose that Λ is a dense linear subspace of
H such that Λ ⊂ DA and AΛ ⊆ Λ. Suppose moreover that there exists a hilbertian
basis {en }n of H contained in Λ such that AΛ0 ⊂ Λ0 where Λ0 = span{en : n ∈ N}. Let
{(Fx , f (x))}x∈X be a measure complete family of antilinear generalized eigenvectors of
A |Λ0 . Denote by µ the corresponding measure. Put Fx (en ) = ψn (x). It follows from
Lemma 2 that the linear map U : Λ0 → span{ψ P n (x) : n ∈ N}
P defined by U (en ) = ψn (x)
2
is bijective. Consider the abstract set K = { n cn ψn (x) : |c
n n | < ∞}. Then K may
be seen as a Hilbert space, where the scalar product is the completion of
Z
(ϕ, ψ) = ϕ(x)ψ(x)dµ(x), ϕ, ψ ∈ U Λ0 , (1)
X

and the map U can be extended to a unitary map U : H → K. We have Fx = δx U |Λ0

for every x ∈ X, which entails that U AU ∗ ψ(x) = f (x)ψ(x) for every ψ(x) ∈ U Λ0 . We
shall say that the representation K is realized with the family of antilinear generalized
eigenvectors {(Fx , f (x))}x∈X . Moreover, K is the completion of the space of functions
U Λ0 = span{ψn (x) : n ∈ N}. Since U Λ0 ⊆ L2 (X, µ), by using classical arguments, we
can either suppose that the completion K of U Λ0 is contained in L2 (X, µ), and K is a
function space (see for instance [31, Proof of Theorem 3.4.1]), or we may suppose that
K \ U Λ0 ⊆ L2 (X, µ). In particular, in the first case, evaluation maps are defined on K,
but certainly, they are not in general continuous on K. Put Fx U ∗ = |xi. Then
Z Z
(ϕ, ψ) = dµ Fx U ∗ (ϕ)Fx U ∗ (ψ) = dµ (ϕ|xihx|ψ), ϕ, ψ ∈ U Λ0 . (2)
X X
 [Author and title] 7

Hence the notation Z

I |U Λ0 ≡ dµ |xihx|.
X

On the other hand, we may put as a notation, ψ(x) = hx|ψ) for every ψ ∈R K. Then the
equality (2) has a meaning in K and we may also use the notation I ≡ dµ |xihx| on
the space K.
Next suppose that we endow Λ0 with a metric topology τ finer than the one induced by
the norm of H that makes the maps Fx continuous and such that Λ coincides with the
completion of (Λ0 , τ ), then by Lemma 2, evaluation maps are continuous on U Λ with
respect to the topology U τ induced by τ . On the other hand, observe that for every
measurable function g : X → C which does not vanish, (g(x)Fx , f (x)) is a generalized
eigenvector of A |Λ0 and the representation realized with respect to the family {g(x)|xi}x
corresponds to a Hilbert space contained in L2 (X, |g(x)|−2 dµ).
Finally, we distinguish three particular cases:
(1) All the maps Fx are continuous with respect to the norm of H, hence |xi can be
seen as an element of K and evaluation maps are continuous on K. That is, K is a
reproducing kernel Hilbert space (RKHS). Put δx = hkx , .i, then its reproducing kernel
is K(y, x) = hky , kx i.
(2) For every x, Fx is continuous with respect to τ . Hence evaluation maps are continuous
on U Λ with respect to U τ .
(3) For every x, Fx is not continuous with respect to τ . Hence evaluation maps are not
continuous on U Λ with respect to U τ .
We shall use the notation Fx ≡ |xi, when there is no risk of ambiguity.

3.4. Change of bases formulas

Recall that if T : K1 → K2 is a unitary map, where K1 and K2 are two finite-
dimensional Hilbert spaces of dimension n, and if {uji }i is an orthonormal basis of Kj ,
then for every u ∈ K1 , we have
n
X X
|ui = hu1k |ui|u1k i, |u1i i = δik |u1k i.
k=1 k
Pn
Hence we
P may1write I2 = 1 i=0 |u1i ihu1i | and hu1i |u1k i = δik . On the other hand, we have
2
|T ui = i,k huk |ui hui |T uk i|ui i, hence we have
X
hu2i |T ui = hu1k |uihu2i |T u1k i.
k

Analogously, we get X
hu1k |T −1 vi = hu2i |vihu1k |T −1 u2i i.
i

In particular, observe that hu2i |T u1k i = hu1k |T −1 u2i i. In Dirac’s formalism, analogous
formulas are used for the so called continuous bases, the symbol sum is replaced by the
 [Author and title] 8

integral [26]. Next we explain a rigorous form. let K1 , K2 be two infinite dimensional
Hilbert spaces, where Kj is a closed subspace of L2 (Xj , µj ). Suppose that Kj is realized
with a measure complete family of antilinear generalized eigenvectors {(|xj i, f j (xj ))}x∈Xj
of the operator Aj : Λ → Λ, where Λ is a dense linear subspace of H. Let Uj : H → Kj
be the corresponding unitary map, that is, Uj Aj Uj∗ ψ(.) = f j (.)ψ(.), for ψ(.) ∈ Uj Λ. Put
V = U2 U1∗ .
Let x, x0 ∈ Xj . If the evaluation maps lie in Kj , then we may put hx0 , xi = K(x0 , x),
the reproducing kernel of Kj . In the general case, observe that if weP put as in the finite
case hx, xi = 1 and hx, x0 i = 0 if x 6= x0 , then the formula ψ(.) = x0 ∈X h., x0 iψ(x0 ) has
a rigorous meaning. Here the sum is trivial. For a general definition of the sum (over an
uncountable set) in a normed space see for instance [32, p. 113].
Now for the change of bases formulas between (K1 , {|x1 i}x1 ∈X1 ) and (K2 , {|x2 i}x2 ∈X2 ),
we can ask the following question:
Question. Is there a family of scalars hx1 , x2 i such that
Z
2
V Ψ(x ) = dµ1 hx2 , x1 i ψ(x1 ), ψ ∈ U1 Λ,
X1

In other words, is V |U1 Λ an integral transform? What about V ∗ and if V ∗ |U2 Φ is an

integral transform, do we have hx2 , x1 i = hx1 , x2 i?
Observe that if an integral transform exists between (K1 , {|x1 i}) and (K2 , {|x2 i}), then we
can use it to deduce an integral transform between (K1 , {g1 (x)|x1 i}) and (K2 , {g2 (x)|x2 i}),
where gj : Xj → C is a measurable function which does not vanish.
Let {en }n be a Hilbertian basis of H contained in Λ. Let F ∈ Λ∗ . We shall say that F
Pk
is continuous with respect toP {en }n if for every ϕ ∈ Λ, F (ϕ) = limk n=0 F (en )(ϕ|en ).
In this case, we write F = F (en )|en ). Next we provide a result which is in the same
spirit of the construction of integral transforms described in [25, ch. 5].
Lemma 3. Suppose that there exists an Hilbertian basis {en }n of H contained in Λ
such that for every xj , |xj i is continuous with respect to {en }n , (j = 1, 2). Suppose
moreover that for every x2 , the sum n |(en |x1 ihx2 |en )| converges in L2 (X1 , µ1 ). Then
P
for every ψ(x1 ) ∈ U1 Λ
Z
(V Ψ(.))(x2 ) = dµ1 (x1 ) hx2 , x1 iψ(x1 ),
X1

where hx2 |x1 i = 1

ihx2 |en ) ∈ L2 (X1 , µ1 ).
P
n (en |x
Pk
Proof. Fix x2 ∈ X2 . Put Gk = n=0 (en |x2 i|en ). Then Gk ∈ Λ. RLet ψ ∈ Λ. Then
[U2 ψ](x2 ) = hx2 |ψ) = limk (Gk , ψ) = limk (U1 Gk , U1 ψ). Since I |U1 Λ = X1 dµ1 (x1 )|x1 ihx1 |,
we get Z
[U2 ψ](x2 ) = [V ψ(x1 )](x2 ) = lim dµ1 (x1 )(Gk |x1 ihx1 |ψ).
k X1

On the other hand, it follows from our assumption on the sum that n |(en |x1 ihx2 |en )hx1 |ψ)|
P
lies in L1 (X1 , µ1 ). Hence we can interchange the integral and the sum in the above equal-
 [Author and title] 9

ity and we get Z

[V ψ(.)](x2 ) = dµ1 (x1 )hx2 |x1 ihx1 |ψ),
X1

where hx2 |x1 i = 1

ihx2 |en ), as desired.
P
n (en |x

At this stage, let us mention that Antoine and Grossmann initiated the study of vector
spaces with an inner product defined partially, see for instance [14, 33, 21].

3.5. The spectral theorem and generalized eigenvectors

The nuclear spectral theorem asserts that any unitary or self-adjoint operator of the
RHS Φ ⊂ H ⊂ Φ× admits a complete family of continuous generalized eigenvectors (see
the proof in [2] and a revised and corrected version in [5]). We shall say that a linear
operator A in H is cyclic if there exists ϕ ∈ ∩n DAn such that span{An ϕ : n ∈ N} = H
[2], here we shall call such a vector ϕ a cyclic vector. For a detailed study of cyclic
operators see [34, § 83].
Next we provide an elementary construction of a complete family of algebraic generalized
eigenvectors for unitary or self-adjoint operators which are cyclic in a Hilbert space. In
the case of a RHS Φ ⊂ H ⊂ Φ× , our construction is suitable for cyclic unitary or self-
adjoint operators having a cyclic vector lying in Φ. Observe that this is the case for many
physical operators (see the next section).
Theorem 1. Let A be a cyclic linear operator in H. Suppose that there exists a set
X ⊆ C, a positive measure µ on X and a unitary map U : H → L2 (X, µ) such that
U AU ∗ (ψ(x)) = xψ(x) for every ψ(x) ∈ U DA . Let ϕ be a cyclic vector of A and put
Λ = span{An ϕ : n ∈ N}. Then there exists a complete family {Fx }x∈X of generalized
eigenvectors of A |Λ such that (U ψ)(x) = Fx (ψ) for every ψ ∈ Λ and x ∈ X.
Proof. The linear operator U AU ∗ is cyclic and it is easy to see that U ϕ is a cyclic
vector. Hence L2 (X, µ) = span{xn (U ϕ)(x) : n ∈ N}. This entails that for every non-
negligible subset Y of X, we must have U ϕ |Y 6= 0. Hence we can choose a true function,
ψ0 : X → C representative of U ϕ such that ψ0 (x) 6= 0 for every x ∈ X. Consider the
functions ψn : X → C defined by ψn (x) = xn ψ0 (x). We have U AU ∗ (ψn (x)) = ψn+1 (x).
We distinguish two cases, either H is infinite-dimensional, hence the set {Ak ϕ : k ∈ N}
is linearly independent; or H has finite dimension, hence there exists N ∈ N such that
{Ak ϕ : 0 ≤ k ≤ N } is a basis of H and dim H = N + 1. For every x ∈ X, consider the
linear form Fx ∈ Λ∗ defined by Fx (An ϕ) = ψn (x). Then Fx 6= 0 and Fx A |Λ = xFx . Next
Pl
let us show that the family {Fx }x∈X is complete. Let ψ = n=0 cn An ϕ ∈ Λ, where l ∈ N
and suppose that Fx (ψ) = 0 for every x ∈ X. Since ψ0 does not vanish on X, we infer
Pl
cn xn = 0 for every x ∈ X. Consider the complex polynomial P = k=0 ck Xk .
P
that
Either P = 0, or it has at most l roots. Suppose for a moment that P 6= 0 and has
degree l. Then X must be finite, with cardinality at most l. This implies that dim H ≤ l.
On the other hand, N ≥ l, that is dim H ≥ l + 1, a contradiction. This completes the
proof.
 [Author and title] 10

Remark 1. Using the notations of the above proposition, let N ∈ N ∪ {∞} be

maximal such that {An ϕ}0≤n≤N is linearly independent. Then it is easy to see that the
PN
norm defined on Λ by k n=0 cn An ϕk = |cn |(n + 1)n makes all the functionals Fx
P
continuous.

Now let us treat the general case and the direct integral form of the spectral theorem.
Let A be a unitary or self-adjoint operator in H. We shall R ⊕ need some details from the
construction of the direct integral representation K = X Hx dµ(x) of A [2, p. 130],
and some notations. Let {ϕj }j∈J be a spectral basis of H associated to A. By using
[34, Theorem 3, p. 281], we may suppose that H = ⊕j Hj , where Hj = span{Ak ϕj }.
Denote by Λ = Λ({ϕj }j ) the linear span of Ak ϕj , where k ∈ N and j ∈ J and let
πj : Λ → span{Ak ϕj : k ∈ N} be the natural projection. Consider the representation
of H defined by K1 = ⊕j∈J L2 (Xj , µ) where Xj ⊆ σ(A) for every j, there exists a
unitary map U j : Hj → L2 (Xj , µ) satisfying U j (Ak ϕj ) = xk ϕj (x). For every x ∈ σ(A),
denote by n(x) the cardinality of Jx = {j ∈ J : x ∈ Xj }. Then n(x) ∈ N ∪ {∞}.
Hx = {((U i1 ψi1 )(x), . . . , (U in(x) ψin (x) )(x))}, where i1 < i2 . . . < in(x) and ψj ∈ Hj .

Theorem 2. Let A be a unitary or self-adjoint linear operator in H Rand let Φ be

⊕
a dense linear subspace of H such that AΦ ⊆ Φ. Let U : H → K = X Hx dµ(x)
be the direct integral representation associated to A, such that U AU ∗ ψ(x) = xψ(x) for
every ψ(x) ∈ U DA . Then there exists a dense linear subspace Λ of H with Λ ⊆ Φ,
AΛ ⊂ Λ, and a family {Fx }x∈σ(A) of generalized eigenvectors of A |Λ such that (U ψ)(x) =
(Fx (πi1 ψ), . . . , Fx (πin(x) ψ)) for every ψ ∈ Λ. Moreover, for every x and j ∈ Jx , Fx πj is
a generalized eigenvector of A |Λ and the family {Fx πj : x ∈ σ(A), j ∈ Jx } is complete.

Proof. Since Φ is dense in H, it contains a spectral basis {ϕj }j∈J , where J ⊆ N. Consider
the representation of H defined by K1 = ⊕j∈J L2 (Xj , µ) where Xj ⊆ σ(A), L2 (Xj , µ) is
associated to ϕj as described above. As in the proof of Theorem 1, for every j, choose
a true function ϕj0 (x) : Xj → C which does not vanish and such that ϕj0 (x) = ϕj (x)
in L2 (Xj , µ). Fix x ∈ σ(A) and j. Consider the linear functional Fx defined on Λ =
Λ({ϕj }j ) by Fx (Ak ϕj ) = 0 if x 6∈ Xj and Fx (Ak ϕj ) = xk ϕj0 (x) if x ∈ Xj . Since the
family {Ak ϕj : k ∈ N, j ∈ J} is linearly independent, then Fx is well defined. Fix j ∈ J
and let ψ P ∈ span{Ak ϕj : k ∈ N}. Then (U j ψ)(x) = Fx (ψ) for every x ∈ Xj . Let ψ ∈ Λ,
then ψ = j πj ψ, and

(U ψ)(x) = (Fx (πi1 ψ), . . . , Fx (πin(x) ψ)).

Next let us check that the family {Fx πj }x,j is complete. Let ψ ∈ Λ and suppose that
Fx πj ψ = 0 for every x and j. Then, by the above result, for every j, πj ψ = 0. Thus,
ψ = 0 as desired.

4. The algebra of the Harmonic oscillator

We describe in this section the number representation and the associated RHS. Let
Ψ be a complex linear space and let P, Q :Ψ → Ψ be linear operators satisfying the
 [Author and title] 11

canonical commutation rule (CCR), that is [P, Q] = −i~I. Put

1 2 mω 2 2
r
mω i
H= P + Q , a= Q+ √ P,
2m 2 2~ 2mω~
r
mω i
a+ = Q− √ P, N = a+ a.
2~ 2mω~
The following properties, which are well known, are easy to prove
Lemma 4.
(1) H = ~ω(N + 21 I) = ~ω
2 (a+ a + aa+ ), [a, a+ ] = I and [a, (a+ )n ] = n(a+ )n−1 .
(2) If α is an eigenvalue of N and v is an α-eigenvector, then N a+ v = (α + 1)a+ v and
N av = (α − 1)av.

Remark 2. Suppose that a scalar product is defined on Ψ. Then P and Q are

symmetric if and only if a∗+ = a.

Next we state the basic properties of irreducible algebraic representations of the Heisen-
berg algebra generated by I, P, Q, such that the Hamiltonian has a minimal real eigen-
value. This is slightly more general than [10] (see also [13]), since we do not suppose that
Ψ is endowed with a scalar product (., .). We provide a proof for completeness.
Theorem 3. Suppose that H admits a minimal real eigenvalue λ. Let u be a λ-
eigenvector and suppose that Ψ coincides with the minimal complex linear subspace con-
taining u and invariant under P and Q. Then
(1) λ = ω~
2 and 0 is the minimal eigenvalue of N . Set ϕ0 = u and for every k ≥ 1, set
ϕk := (a+ )k ϕ0 . Then ϕk 6= 0 and N ϕk = kϕk .
(2) Ψ coincides with span{ϕ0 , ϕ1 , ..} and σ(N, Ψ) = N.
(3) There exists a unique scalar product (., .) on Ψ for which P and Q are symmetric
and kϕ0 k = 1. In this case, the states {ϕk }k are orthogonal.
(4) For all k ≥ 0 let φk = √1k! ϕk . Then the family {φk }k is an orthonormal basis of Ψ.
Moreover, aφ0 = 0 and we have
√ √
a+ φk = k + 1φk+1 , aφl = lφl−1 , N φk = kφk , k ∈ N, l ∈ N∗ .

Proof. For (1), let α be the corresponding minimal real eigenvalue of N . Then, α−1 is not
an eigenvalue of N and it follows from Lemma 4 that au = 0. Therefore αu = N u = 0,
hence α = 0. Now for the second assertion it is easy to check, using once again Lemma
4, that for every k, N ϕk = kϕk . Suppose that for some k ≥ 0, ϕk 6= 0, then since

ϕk = [a, a+ ]ϕk = aϕk+1 − N ϕk = aϕk+1 − kϕk ,

we infer that ϕk+1 6= 0. Thus by induction, ϕk 6= 0 for all k ≥ 0.

Next to prove (2), set V = span{ϕ0 , ϕ1 , ...}. Then V ⊆ Ψ and u ∈ V . On the other
 [Author and title] 12

hand, we show by induction on k that aϕk = kϕk−1 for every k ≥ 1. Thus Λ is invariant
under a and a+ , hence it is invariant under P and Q. Moreover we get V = Ψ. For
the second assertion, we already know that N ⊂ σ(N, Ψ). If λ ∈ / N, we check easily that
N − λ is bijective and hence λ ∈ / σ(N, Ψ).
For (3), suppose that there is a scalar product (., .) on Ψ for which P and Q are symmetric.
We have for every k ∈ N and l ∈ N∗ ,
l l
(ϕk+1 , ϕl ) = (ϕk , aϕl ) = (aϕk+1 , ϕl−1 ) = (ϕk+1 , ϕl ).
k+1 k+1
Moreover, (a+ ϕk , ϕ0 ) = 0 = (ϕk+1 , ϕ0 ). Thus, the family {ϕk }k is orthogonal. On the
other hand, we have

(a+ ϕk , ϕk+1 ) = (ϕk+1 , ϕk+1 ), (ϕk , aϕk+1 ) = (k + 1)(ϕk , ϕk ).

Hence we must have (ϕk+1 , ϕk+1 )√= (k + 1)(ϕk , ϕk ). Suppose that kϕ0 k = 1. Then, by
induction, we see that kϕk+1 k = k + 1kϕk k for all k ≥ 0. Finally, it is clear that the
formulas
√
(ϕk , ϕl ) = δkl kϕk k2 , kϕk+1 k = k + 1kϕk k, kϕ0 k = 1,

define a scalar product on Ψ for which a∗+ = a.

For (4), we prove easily that the family {φk } is orthonormal and the last assertion is
easy to check.

Remark 3. It is easy to see (from the proof) that the condition H admits a minimal
real eigenvalue in the above theorem may be replaced by the condition there exists
u ∈ Ψ\{0} such that au = 0.

From now on, we suppose that Ψ is endowed with the scalar product described in the
above theorem. Denote by H the Hilbert space completion of (Ψ, (., .)). We shall use the
notation φn = |n).
Let us now turn to the algebraic relation of symmetry between P and Q. Put | − 1) = 0.
Then we have for every n ∈ N
r
~ √ √
Q|n) = ( n|n − 1) + n + 1|n + 1)), (3)
2mω
r
~mω √ √
P |n) = −i ( n|n − 1) − n + 1|n + 1)). (4)
2
Put {ni = in |n). Then
r
~mω √ √
P {ni = ( n{n − 1i + n + 1{n + 1i).
2
 [Author and title] 13

1
That is, if c = mω , then M (cP, {{ni}n ) = M (Q, {|n)}n ). Analogously, if c0 = −mω, then
0
M (c Q, {{ni}n ) = M (P, {|n)}n ). Let Γ : H → H be the unitary operator that satisfies
Γ(|n)) = {ni, then cΓ∗ P Γ = Q.
Now for completeness, we provide an elementary proof of the fact that the operators
P, Q, H are essentially self-adjoint.
Proposition 2. The operators P, Q, H : Ψ ⊂ H → H are essentially self-adjoint.
∗
P∞ ∗
Proof.
P∞ Let us first calculate N . Let ψ = 0 ck |k) ∈ DN and put η = N ψ =
∗

0 dk |k). Then it follows from the equality (η, n) = (ψ|N |n) for every |n) that dn = ncn .
Hence
X∞ ∞
X ∞
X ∞
X
DN ∗ = { ck |k) ∈ H : k 2 |ck |2 < ∞}, N∗ ck |k) = kck |k).
0 0 0 0

Now it is clear that N ∗ coincides with the closure N of H, hence N is essentially self-
adjoint. P∞
Next, for Q, we assume that 2mω~
= 1. Let ψ = 0 ck |k) ∈ DQ∗ and put η = Q∗ ψ. Put
c−1 = 0. It follows from equation (3) that
√ √
(η, n) = (ψ|Q|n) = n cn−1 + n + 1 cn+1 (n ≥ 1), (ψ|Q|0) = c1 .

Hence
∞ ∞
X X √ √
η= (n|η)|n) = ( ncn−1 + n + 1cn+1 ) |n).
0 0
Observe that
k k−1 √
X X √ √ √
cn Q|n) = [ ncn−1 + n + 1cn+1 ]|n) + kck−1 |k) + k + 1ck |k + 1).
0 0

On the other hand, since n |cn |2 P

P
< ∞,√ we can choose
√ a subsequence {cnk }k of {cn }n
√
such that limk nk cnk −1 = 0. But n | ncn−1 + n + 1cn+1 |2 < ∞, hence
√
lim nk + 1cnk = 0.
k
Pnk
This entails that limk 0 cn Q|n) = η in H. That is, DQ = DQ∗ and Q is essentially
self-adjoint. Finally, P = c−1 ΓQΓ∗ is essentially self adjoint.
Since H is essentially self-adjoint, it follows from [35, Theorem 5] that the realization of
the (CCR) associated to Ψ is integrable (see also [36]).
Next we endow the linear space Ψ with a topology that makes all the powers of H contin-
uous. Let τ be the topology defined on Ψ by the family of scalar products {h., .in }n∈N∗
where for every n, the norm k.kn associated to h., .in is defined by kϕkn = k(N + I)n/2 ϕk
for every ϕ ∈ Ψ. Then it is easy to see that N, a+ and a are continuous on Ψ and if
S ∈ {N, a, a+ }, then
kSϕkn ≤ kϕkn+2 , (n ∈ N∗ ),
 [Author and title] 14

Denote by Φ the completion of Ψ with respect to the topology τ , then

X X
Φ={ cn |n) ∈ H : cn (N + I)k/2 |n) ∈ H, ∀ k ∈ N}. (5)
n n

In particular, the τ topology defines a metric on Φ [2]. Denote by A(Ψ) the unital asso-
ciative complex algebra of operators on Ψ generated by a, a+ and the relation [a, a+ ] = I.
Then A(ψ) is the first Weyl algebra (also called the Heisenberg-Weyl algebra), it was
introduced by Weyl and a systematic study was initiated in [37]. Clearly, every element
A ∈ A(Ψ) can be extended to a continuous operator on Φ. Moreover, if A is symmetric,
its extension remains symmetric. Denote by A(Φ) the algebra formed by all those exten-
sions. We shall keep the same notations for A ∈ A(Ψ) and its continuous extension, and
we shall write the symbol A for either A(Φ) or A(Ψ), when there is no risk of ambiguity.
Denote by k.k0 = k.k the initial norm of Ψ. Then k.kn ≤ k.km for all n ≤ m in N. For
every n, denote by Φn the completion of Φ with respect to k.kn . It is easy to see that
X X
Φn = { ck φk ∈ H : |ck |2 (k + 1)n < ∞} and Φ = ∩n∈N Φn .
k k

The assumption of nuclearity is crucial in the proof of the nuclear spectral theorem, see
the proofs of the theorem in [2, 5]. As it was pointed out by Roberts [9, p. 107], it is
not difficult to prove Nuclearity for this system. Though we shall not use explicitely the
nuclear spectral theorem, we provide here an elementary proof of the nuclearity of Φ.
Indeed, we have a stronger property
Lemma 5. Let n, m ∈ N∗ such that m ≤ n and let i : Φn → Φm be the natural
injection. Then if n ≥ 2, i is a trace-class operator. In particular, Φ is nuclear.
P∞ 1
Proof. Let ϕ = j=0 cj |j) be an element of Φn . For all j ≥ 0, put λj = (j+1) n . Then
P∞ n n
ϕ = j=0 λj (j + 1) cj |j) and (k|ϕ)n = (k + 1) ck for every k ∈ N. This entails that
P∞
ϕ = j=0 λj (j|ϕ)n |j). Now it is clear that if n ≥ 2, i is of trace-class.

Let Φ× be the topological antidual space of Φ. It is now easy to conclude that Φ ⊂ H ⊂

Φ× is a RHS [2].

5. Eigenstates of the Hamiltonian in function spaces

Next we pursue our study of the algebra of the quantum harmonic oscillator, and we
see how the three other fundamental representations (position, momentum and Bargmann-
Fock) arise naturally, using fundamental operators of the system and their general-
ized eigenvectors. The three spaces are spaces of functions, associated to pairs (A, B)
of operators in the first Weyl algebra that satisfy [A, B] = I, Aψ(x) = xψ(x) and
Bψ(x) = − dψ(x)
dx . At this stage, a natural question presents itself: Are there other im-
portant examples of representations overlooked? In other words, can we, by using the
generalized eigenvectors and associated representations, find new important representa-
tions physically? From our study, it seems that the first step to finding an answer is the
 [Author and title] 15

characterization of A ∈ A that admits a pair B such that [A, B] = I? This condition

was partially investigated in [38]. They showed that in this case the centralizer of A is
C[A]. Other necessary conditions were given in [39]. But it seems that this question is
still open [40].
We shall use the following form of the real and complex Hermite polynomials (see [41]
and [42] for the complex case)

H0 (x) = 1, H1 (x) = 2x, and Hn (x) = 2xHn−1 (x) − 2(n − 1)Hn−2 (x).

That is
[n/2]
X n!(−1)k
Hn (x) = (2x)n−2k (6)
k!(n − 2k)!
k=0
2
As usual, we denote by hn (x) = e−x /2 Hn (x) the Hermite functions.
Let z ∈ C. It is easy to see that |0) 6∈ (Q − z)Ψ and (Q − z)Ψ + C|0i = Ψ. Hence
z is a generalized eigenvalue and the associated space of generalized eigenvectors has
dimension one. With respect to the Hilbert space H, Hermite polynomials correspond to
the algebraic generalized eigenvectors of Q and P . Indeed, more exactly every solution
|F i of the equation Qn F = zF in Ψ∗ is defined by the relations F Q|n) = zF |n) for every
n. By solving the associated recurrence relations, see for instance [43] or [10], we get
1 p
(n, Fz i = √ (0|Fz i Hn ( mω/~ z). (7)
n
2 n!

On the other hand, for G ∈ Ψ∗ ,

P n G = zG ⇔ GΓQ = czGΓ ⇔ G = Fcz Γ∗ .

Thus every solution of the equation P n G = zG in Ψ∗ is defined by

in p
(n, Gz i = √ (0|Gz i Hn ( 1/~mω z).
2n n!
Next we use properties of Hermite polynomials to show the continuity of Fx and Gx , for
x ∈ R.
Lemma 6. Let x ∈ R and let |F i be the element of Ψ∗ defined by (n|F i = √21n n! Hn (x).
Then |F i is continuous for the topology induced by the norm k.kn , for every n ∈ N∗ . In
particular F ∈ Ψ× .

Proof. First observe that if |F i is continuous for k.kn and n ≤ m, then |F i is continuous
Pl
for k.km . Let ϕ = n=0 cn |ni be an element of Ψ and let k ∈ N∗ . Then
l
X 1
(ϕ|F i = cn (n + 1)k/2 p Hn (x).
n=0 2n n!(n + 1)k
 [Author and title] 16

Thus
l l
X X 1
|(ϕ|F i| ≤ [ |cn |2 (n + 1)k ]1/2 [ H (x)2 ]1/2
n n!(n + 1)k n
n=0 n=0
2
∞
X 1
≤ kϕkk [ H (x)2 ]1/2 ,
n n!(n + 1)k n
n=0
2
where the convergence of the second series can be deduced by using for instance the
following inequality (see [44])
2
Cex 2n n! √
Hn2 (x) ≤√ , |x| < 2n + 1.
2n + 1 − x2
for some positive constant C.

Thus, for every x ∈ R, by considering the continuous extensions, we may suppose that
Fx and Gx lie in Φ× . Moreover, observe that Fx and Gx are continuous with respect to
{|n)}n . In particular, we shall write
∞
X 1 p
|Fx i = √ (0|Fx i Hn ( mω/~ x)|n),
n
2 n!
n=0

∞
X in p
|Gx i = √ (0|Gx i Hn ( 1/~mω x)|n).
n
2 n!
n=0

Proposition 3. Suppose that the map x 7→ (0|Fx i is continuous on R, then the

family {Fx }x∈R is strongly continuous. Analogously, if the map x 7→ (0|Gx i is continuous
on R, then the family {Gx }x∈R is strongly continuous.
Proof. Put f (x) = (0|Fx i for every
P x ∈ R. Let {xn }n be
pa sequence of real numbers
converging to x ∈ R and let φ = n cn |ni ∈ Φ. Put α = µω/~. Then, since the map
f is continuous, we get
N
X cn
lim(φ|Fxk i = f (x) lim lim √ Hn (αxk ).
k k N
n=0 2n n!

√ cn Hn (αx)
P
But the series n 2n n!
converges uniformly in a neighborhood of x, thus

N
X cn
lim(φ|Fxk i = f (x) lim lim √ Hn (αxk ) = (φ|Fx i,
k N k
n=0 2n n!

as desired.
 [Author and title] 17

Now from the well known orthogonality relation satisfied by Hermite polynomials, we
deduce the following relation for every α ∈ R
Z
α 2 2
√ Hn (αx)Hm (αx)e−α x dx = δnm , ∀ n ∈ N. (8)
2n n! π R
Put α = mω
p
~ . Next observe that for every function g : R → C which does not vanish,
the family {g(x)|Fx i}x∈R is measure-complete with respect to the measure
α 2 2
dµ = √ e−α x |g(x)|−2 dx.
π
Hence it is complete by Proposition 2 and the family {g(x)hFx |n)i}x∈R is orthonormal.
By using the well known fact that Hermite functions constitute an orthogonal basis of
L2 (R, dx), we conclude that { √g(x) H (αx)} is an orthonormal basis of L2 (R, dµ).
2n n! n
Next we see that Hermite functions are associated to the pairs of operators (Q, P ) and
(P, Q).
Proposition 4. Let K be a Hilbert space, and let U : H → K be unitary such
that U Ψ is a space of C ∞ -functions on R. Then U QU ∗ ψ(x) = xψ(x) and U P U ∗ ψ(x) =
−i~ψ 0 (x) for every ψ(x) ∈ U Ψ if and only if there exists λ ∈ C such that
r
1 mω
U |n) = λ √ hn ( x).
n
2 n! ~

Moreover, for λ = (mω/~π)1/4 , the space K is isomorphic to L2 (R, dx).

Proof. Suppose that U QU ∗ ψ(x) = xψ(x) and U P U ∗ ψ(x) = −i~ψ 0 (x) for every ψ(x) ∈
U Ψ. Let x ∈ R. It follows from Lemma 1 and the equality (7) that
r
1 mω
δx U |ni = √ f (x)Hn ( x),
2n n! ~
: X → C is a map which does not vanish. Put U |n) = ψn . Then ψ0 (x) = f (x),
where f q
∗
ψ1 (x) = 2mω~ xf (x), and U P U ψ0 (x) = imωxf (x). On the other hand, observe that

U P U ∗ ψ0 (x) = −i~ψ00 (x) ⇔ −i~f 0 (x) = imω xf (x).

mω 2
Hence there exists λ ∈ C such that f (x) = λe− 2~ x . Now for the converse, it is enough
to calculate U P U ∗ ψn (x) for every n. With no loss of generality, we may put λ = 1.
Recall the following recurrence formula satisfied by Hermite functions [45, p. 187]
1
h0n (x) = nhn−1 (x) − hn+1 (x).
2
Hence
1 p p
√ h0n (x) = n/2 ψn−1 (x) − (n + 1)/2 ψn+1 (x).
2n n!
 [Author and title] 18

Thus r
~mω √ √
−i~ψn0 (x) = −i ( n ψn−1 (x) − n + 1 ψn+1 (x)),
2
as desired.
Now for the last assertion, we apply once again the fact that Hermite functions form an
orthogonal basis of L2 (R, dx). Since the vector U |n) must be unitary, we use equation
(8) and we deduce the exact value of λ.

Denote by U : H → L2 (R, dx) the unitary map between the number representation
and position space which is described in the above proposition. By using the induced
topology from Φ, and the fact that Hermite functions lie in the Schwartz space S(R), we
check easily that U Φ coincides with S(R). Now we may use the position representation
to check that σ(Q, H) = σ(U QU ∗ , L2 (R, dx)) = R. A detailed study can be found in [34,
§ 54], or we can check it directly. Indeed, observe first observe that

DU QU ∗ = {ψ(x) ∈ L2 (R, dx) : xψ(x) ∈ L2 (R, dx)}.

On the other hand, let λ ∈ R and set θn (x) = n/2 1[λ−1/n,λ+1/n] , then kθn k2 = 1,
p

while limn k(U QU ∗ − λ)θn k2 = 0. Hence (U QU ∗ − λ)−1 cannot be bounded. Consider

the trivial (true) function λ defined on R by λ (λ) = 1 and λ (x) = 0 if x 6= λ. It is
clear that λ is a λ-eigenvector of the operator ψ(x) 7→ xψ(x) defined
p on the linear space
of functions on R. Moreover, observe that the family of maps { 2/n θn (x)}n converges
simply to λ . Finally, let z ∈ C \ R. It is well known and easy to see that by basic
distribution theory, Fz U ∗ cannot be a tempered distribution. Thus, the functional Fz is
not continuous on Φ.
Analogously to the position space, we obtain the following characterization of the mo-
mentum space
Proposition 5. Let K be a Hilbert space, and let U : H → K be unitary such
that U Ψ is a space of C ∞ -functions on R. Then U P U ∗ ψ(x) = xψ(x) and U QU ∗ ψ(x) =
i~ψ 0 (x) for every ψ(x) ∈ U Ψ if and only if there exists λ ∈ C such that

(−i)n p
U |n) = λ √ hn ( 1/~mω x).
n
2 n!

Moreover, for λ = (1/~mωπ)1/4 , the space K is isomorphic to L2 (R, dx).

Put r
mω 1/4 X 1 mω
|xi = ( ) √ hn ( x)|n),
~π 2n n! ~
n

and
1 X in √
|pi = ( )1/4 √ hn (1/ ~mω x)|n).
~mωπ n 2n n!
 [Author and title] 19

To find hp, xi, one may use the fact that F(hn (x)) = (−i)n hn (p) (see for instance [46]),
where F is the unitary map defined on L2 (R, dx) as a completion of the Fourier transform
on the Schwartz space
Z
F{f }(p) = (2π)−1/2 e−ipx f (x)dx, f ∈ S(R).
R

By rescaling, one gets the desired formulas (see [10]). Here we shall use Lemma 3.
Observe that all the maps |xi and |pi are continuous with respect to the basis {|n)}n ,
and it follows from Mehler’s formula for Hermite polynomials that
X 1
(en |xihp|en ) = √ e−ipx/~ = hp, xi.
n 2π~

Arguing analogously, we show that hx, pi = hp, xi.

The canonical coherent states, eigenvectors of the annihilation operator are well known
and easy to compute. If z ∈ C, every solution Bz of the equation an ∗
+ B = zB in Ψ
P∞ n n
z z
has the form Bz = n=0 (0|Bz i n! |n), and Bz ∈ H. It is well known that { n! }n
√ √
2
is an orthonormal family with respect to the measure e−|z| dz, the closed subspace of
2
L2 (C, e−|z| dz) generated by this family is the Bargmann-Fock space
Z
2
F = {ψ : C → C, ψ is analytic and |ψ(z)|2 e−|z| dz < ∞},

In particular, the family {Bz }z∈C is measure-complete. By Section 3., for every mea-
surable map g : C → C which does not vanish, we can associate a representation to the
family of generalized eigenvectors {g(z)|Bz i}z∈C . Next we see that to get a unitary rep-
resentation associated to the pair (a, a+ ), the map g must be a constant. In particular,
the representation must be isomorphic to F.
Proposition 6. Let K be a Hilbert space and let U : H → K be unitary such
that U Ψ is a space of holomorphic functions on C. Then U a+ U ∗ ψ(z) = zψ(z) and
U aU ∗ ψ(z) = ∂z ψ(z) for every ψ(z) ∈ U Ψ if and only if there exists λ ∈ C such that
zn
(U |n))(z) = λ √ .
n!
0
Furthermore, H can be seen as a RKHS, its reproducing kernel is K(z 0 , z) = λez z and
for λ = 1, K is isomorphic to F.
Proof. Put ψn (z) = (U |n))(z). Since U a+ U ∗ ψ(z) = zψ(z) for every ψ(z) ∈ U Ψ, by
Lemma 1, δz is a generalized eigenvector for a+ with respect to z, that is, δz = Bz .
n
Hence ψn (z) = hBz |n) = hBz |0) √z n! . On the other hand, it follows from the equality
U aU ∗ ψ0 = ∂z ψ0 = 0 that ∂z hBz |0) = 0 for all z ∈ C. Thus the map z 7→ hBz |0) is a
constant λ. The converse is immediate. Now it is clear that we can assume that the
completion K of U Ψ is constructed with holomorphic functions.
 [Author and title] 20

For the rest of the proof, we suppose that λ = 1. To get the reproducing kernel, we use
the fact that evaluation maps are continuous with respect to the norm on U Φ. Hence
for every ψ(.) ∈ U Φ,
X X zn
δz ψ(.) = δz ( hψn (.), ψ(.)iψn (.)) = hψn (.), ψ(.)i √ .
n n n!
P n P n
Hence δz = h n √z n! ψn (.), .i. This implies that K(z, z 0 ) = n √z n! ψn (z 0 ), as desired.
The rest of the proof is immediate.
P n
Put |zi = n √z n! |n). Then |zi ∈ Φ and the generalized eigenvector |xi is well defined
on |zi. Moreover, we have
mω 1/4 −mωx2 /2~ X 1 z n p
hz|xi = ( ) e ( √ ) Hn ( mω/~ x)
~π n
n! 2

By using the exponential generating function of the Hermite polynomials, we get

mω 1/4 p 1
hz|xi = ( ) exp( 2mω/~xz − (z 2 + mω/~ x2 )).
~π 2
Now it follows from Lemma 3 that the unitary map corresponding to the change from
position space to F is given by
Z
Υ : L2 (R, dx) → F, Υ(ψ(x))(z) = dx hz, xiψ(x).
R

This corresponds to the well known Bargmann transform [47].

On the other hand, since a|0) = 0, there is no Hilbert space K, completion of a space
of functions and unitary map U : H → K such that U aU ∗ ψ(x) = xψ(x). This can be
also deduced from the fact that the operator a does not admit an algebraic generalized
eigenvector.

Aknowledgements. We thank the referee for various useful comments and for
drawing our attention to papers [17, 18].

REFERENCES
[1] I.M. Gel’fand and A. G. Kostutchenko, Eigenfunction expansions for differential operators,
Doklady. Akad, Nauk SSSR, 103, 345–352 (1955).
[2] I. M. Gel’fand and Y. Vilenkin, Generalized Functions, vol. IV, Academic Press, New York
1964.
[3] K. Maurin, Generalized Eigenfunctions Expansions and Unitary Representations of Topo-
logical Groups, Polish Scientific Publishers, Warsaw 1968.
[4] M. Gadella and F. Gomez, On the mathematical basis of the Dirac formulation of quantum
mechanics, Int. J. Theor. Phys. 42, 2225–2254 (2003).
 [Author and title] 21

[5] G. G. Gould, The spectral representation of normal operators on a rigged Hilbert space, J.
London Math. Soc. 43, 745–754 (1968).
[6] J. von Neumann, Mathematical Foundations of Quantum Mechanics, Princeton University
Press, Princeton, 1955.
[7] C. Foias, Décompositions intégrales des familles spectrales et semi-spectrales en opérateurs
qui sortent de l’espace Hilbertien, Acta Scient.-Math. (Szeged), 20, 117–155 (1959).
[8] J. E. Roberts, The Dirac bra and ket formalism, J. Math. Phys. 7, 1097–1104 (1966).
[9] J. E. Roberts, Rigged Hilbert spaces in quantum mechanics, Commun. Math. Phys. 3,
98–119 (1966).
[10] A. Böhm, The Rigged Hilbert Space in Quantum Mechanics, in Boulder Lectures in Theo-
retical Physics, 9A, Interscience, NewYork, 1966.
[11] J.-P. Antoine, Dirac formalism and symmetry problems in quantum mechanics. I. General
Dirac formalism, J. Math. Phys. 10, 53–69 (1969).
[12] J.-P. Antoine, R. Bishop, A Böhm and S. Wickramasekara, Rigged Hilbert Spaces in Quan-
tum Physics, in D. Greenberger, K. Hentschel and F. Weinert, eds., Compendium of Quan-
tum Physics, Springer, Berlin, Heidelberg 2009.
[13] R. de la Madrid: Quantum Mechanics in Rigged Hilbert Space Language. PhD Thesis,
Universidad de Valladolid, Valladolid (2001).
[14] J.-P. Antoine, Quantum Mechanics Beyond Hilbert Space, in Irreversibility and Causality,
Semigroups and Rigged Hilbert Spaces, A. Böhm, H.-D. Doebner, and P. Kielanowski, eds.,
Springer, Berlin, 1998.
[15] J.-P. Antoine, A Böhm and S. Wickramasekara, Rigged Hilbert Spaces for the Dirac Formal-
ism of Quantum Mechanics and Time Symmetric Quantum Mechanics, in D. Greenberger,
K. Hentschel and F. Weinert, eds., Compendium of Quantum Physics, Springer, Berlin,
Heidelberg 2009.
[16] M. Gadella and F. Gomez, Dirac formulation of quantum mechanics, recent and new results,
Rep. Math. Phys. 59, 127–143 (2007).
[17] G. Lindblad and B. Nagel, Continuous bases for unitary irreducible representations of
SU (1, 1), Ann. Inst. H. Poincaré 13, 27–56 (1970).
[18] B. Nagel, Generalized eigenvectors in group representatiuons, in Studies in Mathematical
Physics, A. O Barut (ed), Reidel Dordrecht and Boston, 1970.
[19] D. Babbitt, Rigged Hilbert spaces and one-particle Schrödinger operators, Rep. Math. Phys.
3, 37–42 (1972).
[20] D. Fredricks, Tight riggings for a complete set of commuting observables, Rep. Math. Phys.
8, 277–293 (1975).
[21] J.-P. Antoine and C. Trapani, Partial Inner Product Spaces, Theory and Applications,
Springer, Berlin, Heidelberg 2009.
[22] S. J. L. Van Eijndhoven and J. de Graaf, A Mathematical Introduction to Dirac’s Formalism,
North Holland, Amsterdam, 1986.
[23] H. Bergeron, Rigorous bra-ket formalism and wave function operator for one particle quan-
tum mechanics, J. Math. Phys. 47, 022105 (2006).
 [Author and title] 22

[24] P. A. M. Dirac, The principles of Quantum Mechanics, 3nd ed., Clarendon Press, Oxford
1947.
[25] J. P. Gazeau, Coherent states in quantum physics, Wiley-VCH, Berlin 2009.
[26] C. Cohen-Tannoudji, B. Diu, F. Laloë, Quantum Mechanics, Wiley, New York 1977.
[27] J. R. Klauder and B. S. K. Skagerstam, A Coherent-State Primer, in J. R. Klauder and
B-S. Skagerstam (eds), Coherent states, World Scientific, Singapore, 1985.
[28] J. M. Berezanskii, Expansions in Eigenfunctions of Selfadjoint Operators. Transl. Math.
Monogr., vol. 17. Providence, American Mathematical Society 1968.
[29] T. Poerschke, G. Stoltz and J. Weidmann, Expansions in generalized eigenfunctions of
selfadjoint operators, Math. Z. 202, 397–408 (1989).
[30] A. Böhm, M. Gadella, Dirac kets, Gamow Vectors and Gelfand Triplets, Lecture Notes in
Physics, vol. 348. Springer, New York 1989.
[31] D. L. Cohen, Measure Theory, Birkhauser, Boston 1980.
[32] J. Dixmier, General Topology, Springer, New York 1984.
[33] J-P. Antoine and A. Grossmann, Partial inner product spaces I. General properties, J.
Funct. Anal. 23, 369–378 (1976).
[34] N. I. Akhiezer and I. M. Glazman, Theory of Linear Operators in Hilbert Space I, Pitman,
Boston 1981.
[35] E. Nelson, Analytic vectors, Ann. of Math. 70, 572–615 (1959).
[36] C. Trapani, Remarks on Infinite-dimensional Representations of the Heisenberg Algebra, in
Lie Groups, Differential Equations, and Geometry, Advances and Surveys, 2017.
[37] J. Dixmier, Sur les algèbres de Weyl, Bull. Soc. Math. France, 96, 209–242 (1968).
[38] J. A. Guccione, J. J. Guccione and C. Valqui, On the centralizers in the Weyl algebra,
Proc. Amer. Math. Soc.140, 1233–1241 (2012).
[39] V. V. Bavula, On the eigenvector algebra of the product of elements with commutator one
in the first Weyl algebra, Math. Proc. Camb. Phil. Soc. 151, 245–262 (2011).
[40] C. Zhang, On solvable elements in the Weyl algebra, arXiv: 1703.04076v1
[41] G. Sansone, Orthogonal Functions, revised english ed., Interscience, New York, 1959.
[42] S. J. L. Van Eijndhoven and J. L. H. Meyers, New orthogonality relations for the Hermite
polynomials and related Hilbert spaces, J. Math. Anal. App. 146, 89–98 (1990).
[43] L. E. Ballentine, Quantum Mechanics, Prentice-Hall International, New Jersey 1990.
[44] S. Bonan and D. S. Clark, Estimates of the Hermite and the Freud polynomials, J. Approx
Theory 63, 210–224 (1990).
[45] G. B. Folland, Fourier Analysis and its Applications, Wadsworth, California 1992.
[46] E. C. Titchmarsh, Introduction to the Theory of Fourier Integrals, 2nd. ed., Oxford Uni-
versity Press, 1948.
[47] V. Bargmann, On a Hilbert space of analytic functions and an associated integral transform
I, Comm. Pure Appl. Math. 14, 187–214 (1961).