Annals of Physics 335 (2013) 78–85

Contents lists available at SciVerse ScienceDirect

Annals of Physics
journal homepage: www.elsevier.com/locate/aop

Coherent orthogonal polynomials

E. Celeghini a , M.A. del Olmo b,∗
a
Dipartimento di Fisica, Università di Firenze and INFN–Sezione di Firenze, I50019 Sesto Fiorentino,
Firenze, Italy
b
Departamento de Física Teórica and IMUVA, Universidad de Valladolid, E-47005, Valladolid, Spain

highlights
• Fundamental characteristic of orthogonal polynomials (OP): existence of a Lie algebra.
• Differential recurrence relations of OP determine a unitary representation of a non-compact Lie group.
• 2nd order Casimir originates a 2nd order differential equation that defines the corresponding OP family.
• Generalized coherent polynomials are obtained from OP.

article info abstract

Article history: We discuss a fundamental characteristic of orthogonal polynomi-
Received 27 June 2012 als, like the existence of a Lie algebra behind them, which can be
Accepted 27 April 2013 added to their other relevant aspects. At the basis of the complete
Available online 6 May 2013
framework for orthogonal polynomials we include thus – in addi-
tion to differential equations, recurrence relations, Hilbert spaces
Keywords:
and square integrable functions – Lie algebra theory.
Orthogonal polynomials
We start here from the square integrable functions on the open
Group representation theory
Quantum mechanics connected subset of the real line whose bases are related to orthog-
Coherent states onal polynomials. All these one-dimensional continuous spaces al-
low, besides the standard uncountable basis {|x⟩}, for an alternative
countable basis {|n⟩}. The matrix elements that relate these two
bases are essentially the orthogonal polynomials: Hermite polyno-
mials for the line and Laguerre and Legendre polynomials for the
half-line and the line interval, respectively.
Differential recurrence relations of orthogonal polynomials al-
low us to realize that they determine an infinite-dimensional irre-
ducible representation of a non-compact Lie algebra, whose second
order Casimir C gives rise to the second order differential equation
that defines the corresponding family of orthogonal polynomials.
Thus, the Weyl–Heisenberg algebra h(1) with C = 0 for Hermite

∗ Corresponding author. Tel.: +34 983423146; fax: +34 983423013.

E-mail addresses: [email protected] (E. Celeghini), [email protected], [email protected] (M.A. del Olmo).

0003-4916/$ – see front matter © 2013 Elsevier Inc. All rights reserved.
http://dx.doi.org/10.1016/j.aop.2013.04.017
 E. Celeghini, M.A. del Olmo / Annals of Physics 335 (2013) 78–85 79

polynomials and su(1, 1) with C = −1/4 for Laguerre and Legen-

dre polynomials are obtained.
Starting from the orthogonal polynomials the Lie algebra is ex-
tended both to the whole space of the L2 functions and to the
corresponding Universal Enveloping Algebra and transformation
group. Generalized coherent states from each vector in the space
L2 and, in particular, generalized coherent polynomials are thus
obtained.
© 2013 Elsevier Inc. All rights reserved.

1. Introduction

Orthogonal polynomials are relevant in many fields of mathematics: differential equations, alge-
bras and Lie groups, Hilbert spaces and generalized Fourier series are, perhaps, the principal ones.
They are essentially orthogonal bases of square integrable functions L2 (E ). We consider in this
paper the cases discussed by Cambanis in [1], where E is a connected open subset of the real line,
i.e. E = (a, b) ⊂ R with −∞ ≤ a < b ≤ +∞. We will deal with the Hermite functions when E =
(−∞, +∞), the Laguerre functions for E = (a, +∞) with −∞ < a or E = (−∞, b) with b < +∞
(both with an appropriate change of variables reduce to E = (0, +∞)) and the Legendre polynomials
for E = (a, b) with −∞ < a < b < +∞ (that, with a suitable change of variables, reduces to E =
(−1, +1)). We will see that the operators acting on each vector space have a Lie algebra structure.
The framework we are dealing with is related to the intertwining of the basis that defines the space
and the alternative countable basis introduced by the orthogonal polynomials. The generators of the
Lie algebra are shown to be related to first order differential recurrence relations while C , the second
order invariant of the algebra, originates the second order differential equation that defines the family
of orthogonal polynomials. From the generators, the full set of operators on the space L2 (E ) is built
inside the Universal Enveloping Algebra (UEA), i.e. the closure of the underlying vector space of a Lie
algebra that has the ordered monomials constructed on the generators as a basis. In particular, as the
group elements belong to the UEA, the group transformations on L2 (E ) are in deep relationship with
the orthogonal polynomials. Specifically, each type of orthogonal polynomials is related to a unitary
irreducible representation (UIR) of a non-compact Lie group: Hermite functions (as it is well known
from the quantum harmonic oscillator) support a UIR of the Weyl–Heisenberg group H (1) and the
Laguerre and Legendre functions belong to the fundamental representation of the discrete series of
SU (1, 1).
As algebra and group operators are defined on the whole L2 (E ) space we can obtain in an op-
eratorial way peculiar combinations of vectors of L2 (E ) such as, for instance, generalized coherent
orthogonal polynomials.
Our starting point is the separable Hilbert space L2 (E ) equipped with the basis {|x⟩} on the con-
nected open subset E = (a, b) ⊂ R. Orthonormality and completeness are
 b
⟨x | y⟩ = δ(x − y), |x⟩ dx ⟨x| = I. (1)
a

For each type of orthogonal polynomials we will write first the vector space – essentially to fix the
notations – and then we will discuss the operators acting on it.

2. Hermite polynomials

We start our discussion from the Hilbert space constructed by means of the Hermite polynomials
Hn (x), defined on the full real line (a, b) = (−∞, ∞), because in this case both vector space and
Lie algebra structure are well known and we shall limit ourselves to put together, in a compact way,
known results.
80 E. Celeghini, M.A. del Olmo / Annals of Physics 335 (2013) 78–85

To begin with, some factors are attached to the Hn (x), writing the usually named Hermite functions
Kn (x):
2/2
e −x
Kn (x) :=  √ Hn (x), n ∈ N. (2)
2n n! π
In terms of them orthonormality and completeness are [2]
 ∞
Kn (x) Km (x) dx = δn,m ,
−∞
∞
 (3)
Kn (x) Kn (y) = δ(x − y).
n =0

Hence, the functions {Kn (x)} determine an orthonormal basis for the real square integrable functions
on the line, L2 ((−∞, ∞)) = L2 (R) [3]. We can now define
 ∞
|n⟩ := |x⟩ Kn (x) dx. (4)
−∞
So, the set {|n⟩}n∈N is in one-to-one correspondence to {Kn (x)}, and is an orthonormal basis since from
(1) to (3)
∞

⟨n|m⟩ = δn,m , |n⟩ ⟨n| = I.
n =0

Note that, instead of the usual heavy notation |Kn (x)⟩, we use the physical notation |n⟩ to emphasize
the two bases {|x⟩}x∈R and {|n⟩}n∈N of the Hilbert space L2 (R). Orthogonal functions play the role of
transition matrices and, as the space is real, are written as
Kn (x) = ⟨x|n⟩ = ⟨n|x⟩.
Now, in complete analogy with the Fourier analysis on the circle, an arbitrary vector |f ⟩ of L2 (R) can
be written
 +∞ ∞

|f ⟩ = |x⟩ dx ⟨x|f ⟩ = |n⟩⟨n|f ⟩, (5)
−∞ n =0

where the wave function f (x) and the sequence {fn }n∈N describe the vector |f ⟩ in the two bases:
∞
  +∞
f (x) := ⟨x|f ⟩ = Kn (x)fn fn := ⟨n|f ⟩ = Kn (x)f (x)dx. (6)
n=0 −∞

In particular, the completeness of the two bases determines the inner product as well as the Parseval
identity:
∞
  +∞ ∞
  +∞
⟨g |f ⟩ = gn · fn = g (x) · f (x) dx, fn 2 = f (x)2 dx. (7)
n =0 −∞ n =0 −∞

Let us introduce now the operators defined on the line that, as is well known, are functions of
creation, annihilation and number operators. Following the factorization method [4,5] we consider,
among the recurrence relations of Hermite polynomials, those that include first order derivatives [2]:

Hn′ (x) = 2 n Hn−1 (x), Hn′ (x) − 2 x Hn (x) = −Hn+1 (x). (8)
The second order differential equation, that defines the Hermite polynomials, is obtained by subse-
quent applications of recurrence relations. However, the fundamental limitation of such an approach
is that in the factorization method the problem has been considered from the point of view of differ-
ential equations theory and, hence, the labels are assumed to be parameters.
 E. Celeghini, M.A. del Olmo / Annals of Physics 335 (2013) 78–85 81

We shall show here that a consistent Hilbert space framework, where the operators are well de-
fined, allows us to reformulate special functions in the Lie algebra representation scheme, such that,
in particular, Hermite functions support a well defined IR of the Weyl–Heisenberg algebra h(1). Effec-
tively, we can rewrite Eqs. (8) in terms of Hermite functions obtaining
√ √
a Kn (x) = n Kn−1 (x), aĎ Kn (x) = n + 1 Kn+1 (x), (9)

where a and aĎ are the annihilation and creation operators of the harmonic oscillator:

1 1
a := √ (X + Dx ) , aĎ := √ (X − Dx ) . (10)
2 2
In addition to the two previous operators X and Dx

X f (x) := x f (x), Dx f (x) := f ′ (x), (11)

we introduce two other operators N and I defined by

N Kn (x) := n Kn (x), I Kn (x) := Kn (x), (12)

in order to complete our set of operators. Note that the change introduced by substituting the label n
of the Hermite polynomials for the operator N is far from irrelevant since N does not commute with
a and aĎ :

[N , a] = −a, [ N , aĎ ] = aĎ .
The whole Weyl–Heisenberg algebra follows

[N , a] = −a, [ N , aĎ ] = aĎ , [ a, aĎ ] = I , [I , •] = 0.
Expression (4) allows us to connect to the operators a, aĎ , N and I, defined on the set {Kn (x)}, four
related operators defined on the line Hilbert space that we denote with the same symbols:
√ √
a |n⟩ = n |n − 1⟩, aĎ |n⟩ = n + 1 |n + 1⟩, N |n⟩ = n|n⟩, I |n⟩ = |n⟩.

The quadratic invariant operator C can be written [6]

C = {a, aĎ } − 2(N + 1/2) I .

As, by inspection, we are dealing with the representation C = 0, the Schrödinger equation of the
quantum harmonic oscillator is obtained, by means of Eqs. (9)–(12), i.e.

C Kn (x) ≡ X 2 − D2x − (2N + 1) Kn (x) = 0,

 

that, rewritten in terms of the Hermite polynomials allows us to recover the equation defining the
Hermite polynomials

Hn′ (x) − 2 x Hn′ (x) + 2 n Hn (x) = 0.

Thus, we can conclude that the Hermite functions constitute a basis of the infinite-dimensional IR
C = 0 of the Weyl–Heisenberg algebra h(1) and the whole UEA of h(1) is defined on the {Kn (x)}. In
particular, as all elements of the Weyl–Heisenberg group H (1) are contained inside the UEA of h(1),
all the transformations of H (1) can be realized in the space {Kn (x)}. Incidentally, the IR C = 0 of h(1)
originates a UIR of H (1). As an example, Eqs. (9) allow to obtain the simplest normalized coherent
Hermite function as [7]:
∞
1 Ď 1  zn
Kz (x) :=  e z a K0 ( x ) =  √ Kn (x), z ∈ C.
2 2
e|z | e|z | n =0 n!
82 E. Celeghini, M.A. del Olmo / Annals of Physics 335 (2013) 78–85

3. Laguerre polynomials

The Laguerre functions differ from Hermite ones in the relevant fact that recurrence relations
depend not only on the variable x (as it happens for a and aĎ in Eqs. (10)) but also on the parameter n.
The discussion is similar to the previous one. We shall start renormalizing the Laguerre polynomials
Ln (x) by defining in this way the Laguerre functions
Mn (x) := e−x/2 Ln (x).
Orthonormality and completeness relations are [1]:
 ∞
Mn (x) Mm (x) dx = δn,m ,
0
∞
 (13)
Mn (x) Mn (y) = δ(x − y).
n =0

In analogy with expression (4), expressions (1) allow us to define

 ∞
|n⟩ := |x⟩ Mn (x) dx,
0

and to obtain a countable basis {|n⟩}n∈N of the half-line, in addition to the standard basis {|x⟩}x∈(0,∞) .
The functions Mn (x) are the transformation matrix elements that relate the two orthonormal bases
Mn (x) = ⟨x | n⟩ = ⟨n | x⟩,

and for an arbitrary vector |f ⟩ of the Hilbert space L2 ((0, ∞)) all formulas (5)–(7) are valid after the
changes (−∞, +∞) → (0, +∞) and {Kn (x)} → {Mn (x)}.
In order to obtain the operators acting on L2 ((0, ∞)), let us write the recurrence formulas
involving derivatives for the Laguerre polynomials [8]
x L′n (x) − n Ln (x) = −n Ln−1 (x), x L′n (x) + (1 + n − x) Ln (x) = (n + 1) Ln+1 (x),
that can be rewritten on terms of the functions Mn (x) in a slightly more symmetrical form
 
d x
J− Mn (x) := −x +n− Mn (x) = n Mn−1 (x),
dx 2
  (14)
d x
J+ Mn (x) := x +n+1− Mn (x) = (n + 1) Mn+1 (x).
dx 2
Because the operator XDx is zero for x = 0, we can properly define J± as true operators on the half-line
Hilbert space
X X
J− := −XDx + N − , J+ := XDx + N + 1 − . (15)
2 2
Like in the Hermite case, the introduction of these full operatorial structures is not trivial since the
operators N and J± do not commute
[N , J± ] = ±J± .
The operators acting on the basis {|n⟩} are equivalent to the ones acting on the basis {Mn (x)}
J− |n⟩ = n |n − 1⟩, J+ |n⟩ = (n + 1) |n + 1⟩, N |n⟩ = n |n⟩. (16)
Defining J3 := N + 1/2, commutators and anticommutators of J± are
[J+ , J− ] = −2 J3 , { J+ , J− } = 2 J32 + 1/2
and, {Mn (x)} and {|n⟩} are found to be representations of the su(1, 1) algebra:
[J3 , J± ] = ±J± [J+ , J− ] = −2J3 .
 E. Celeghini, M.A. del Olmo / Annals of Physics 335 (2013) 78–85 83

The differential equation defining the Laguerre polynomials can be obtained, following the factor-
ization method, from

(J+ J− − N 2 ) Mn (x) = 0,
or

(J− J+ − (N + 1)2 ) Mn (x) = 0.

Both expressions, using Eqs. (15), give

−X XD2x + Dx + N + 1/2 − X /4 Mn (x) = 0,

 

which is equivalent, since X in never the null operator on (0, +∞), to

XD2x + Dx + N + 1/2 − X /4 Mn (x) = 0,

 
(17)

that, rewritten in terms of the {Ln (x)}, is the definition equation of the Laguerre polynomials, i.e.

x L′′n (x) + (1 − x) L′n (x) + n Ln (x) = 0.

Alternatively, following the algebraic approach, the anticommutator { J+ , J− } is related to the

Casimir operator C . From Eqs. (16)
 
1 1
C= J32 − { J+ , J− } = − ,
2 4
that, introducing expressions (15), gives the operational form of Eq. (17), i.e. the definition of Laguerre
polynomials.
Ď
Because C = −1/4 and J± = J∓ we are dealing, at the Lie algebra level, with the fundamental
representation of the discrete series of UIR of SU (1, 1). Since C = k (k − 1) with k the maximum
weight of the representation, we find k = 1/2 in agreement with the minimum eigenvalue 0 of the
operator N of the Laguerre polynomials. Hence, the spectrum of the operator J3 is 1/2, 3/2, 5/2, . . . .
Again, from the algebra su(1, 1) we can move to its UEA and to the group SU (1, 1). In this way, a co-
herent Laguerre polynomial, Lα (x), could thus be defined à la Perelomov [7]. These coherent states are
in correspondence with the points of the upper sheet of the two-sheet hyperboloid H = {(y0 , y1 , y2 ) |
y20 − y21 − y22 = 1, y0 > 0}, and we can parametrize the points of H by the coordinates (ξ , θ ), with
ξ ∈ R+ and θ ∈ [0, 2π ), as follows: y0 = cosh ξ , y1 = sinh ξ cos θ , y2 = sinh ξ sin θ . Then

Lα (x) := ex/2 eξ̂ J+ −ξ̂ J− M0 (x) = ex/2 eα J+ (1 − |α|2 )J3 e−α J− M0 (x)
∗ ∗

= ex/2 (1 − |α|2 )1/2 eαJ+ M0 (x)

∞
= (1 − |α|2 )1/2

α n Ln (x), (18)
n =0

where ξ̂ = ξ eiθ and α = eiθ tanh ξ .

4. Legendre polynomials

The approach is similar to that of Laguerre polynomials, with the difference that the interval E =
(a, b) ⊂ R is now (−1, 1). We have [2]:
 1
Pn (x) (n + 1/2) Pm (x) dx = δn,m ,
−1
∞ (19)

Pn (x) (n + 1/2) Pn (y) = δ(x − y).
n =0
84 E. Celeghini, M.A. del Olmo / Annals of Physics 335 (2013) 78–85

As in previous cases, expressions (19) allow us to define a new basis {|n⟩}n∈N in the interval (−1, 1)
in addition to the standard basis {|x⟩}x∈(−1,1) :
 +1
n + 1/2 Pn (x) dx.

|n⟩ := | x⟩
−1
The Pn (x) polynomials are thus the transformation matrix elements that relate both two orthonormal
bases:
n + 1/2 Pn (x) = ⟨x | n⟩ = ⟨n | x⟩,


and, in complete analogy with the previous cases, for an arbitrary vector |f ⟩ of the Hilbert space
L2 ((−1, +1)) all formulas√(5)–(7) are valid for the Legendre case after the changes (−∞, +∞) →
(−1, +1) and {Kn (x)} → { n + 1/2 Pn (x)}.
The vector space structure described, let us now study the operators. The recurrence formulas [8]:

(x2 − 1) Pn′ (x) = (n + 1) Pn+1 (x) − x Pn (x) ,

 
′ (20)
(x2 − 1) Pn (x) = n (x Pn (x) − Pn−1 (x)),
can be rewritten in terms of differential operators as follows:

J− Pn (x) ≡ (1 − X 2 )Dx + X N Pn (x) = n Pn−1 (x),

 
(21)
J+ Pn (x) ≡ −(1 − X 2 )Dx + X (N + 1) Pn (x) = (n + 1) Pn+1 (x).
 

Again, the change we introduced into Eqs. (21) is not irrelevant since
[N , J± ] = ±J± ,
and, also in this case, the Hermiticity relation (J± )Ď = J∓ (and, thus, the unitarity of the group repre-
sentation) is imposed by the recurrence formulas.
The action of the operators J∓ on the vectors |n⟩ is from (21)
J− |n⟩ = n |n − 1⟩, J+ |n⟩ = (n + 1) |n + 1⟩ . (22)
Finally from (22) we obtain:
J+ J− − N 2 = −(1 − X 2 ) (1 − X 2 )D2x − 2XDx + N (N + 1) ≡ 0,
 

J− J+ − (N + 1)2 = −(1 − X 2 ) (1 − X 2 )D2x − 2XDx + N (N + 1) ≡ 0,

 

that, up to an irrelevant global factor (1 − X 2 ) that never vanishes, is the Legendre equation in oper-
atorial form.
As before, the algebraic approach starts from the commutator and the anticommutator of the op-
erators J±
[J+ , J− ] = −2 (N + 1/2) , {J+ , J− } = 2(N + 1/2)2 + 1/2.
Defining the operator J3 := N + 1/2, which has the same explicit form as in the Laguerre case, a
su(1, 1) algebra is again found:
[J3 , J± ] = ±J± , [J+ , J− ] = −2J3 .
Moreover, the Legendre functions support the same SU (1, 1)-UIR of Casimir C = −1/4 of the La-
guerre functions and the equation
(C + 1/4) Pn (x) = J32 − {J+ , J− }/2 + 1/4 Pn (x) = 0
 

together with Eqs. (21) gives the Legendre equation.

A coherent state generalization of {Pn (x)} can be also defined in this case similarly to (18)
∞
Pα (x) := eξ̂ J+ −ξ̂ J− P0 (x) = (1 − |α|2 )1/2
∗ 
α n Pn (x), (23)
n =0

where the complex numbers α and ξ̂ are the same of the Laguerre case (18).
 E. Celeghini, M.A. del Olmo / Annals of Physics 335 (2013) 78–85 85

5. Conclusions

This paper is the first of a program attempting to give a global description of orthogonal poly-
nomials where the different aspects – differential equations theory, Lie algebras, Hilbert spaces and
generalized Fourier series – are considered together (see Ref. [9] for the case of the associated Legen-
dre polynomials and the spherical harmonics). We discuss here orthogonal polynomials defined on
the line – where technical aspects are less relevant – and we give an algebraic description of them in a
global way. At the level of Hilbert spaces the Hermite polynomials allow to build a discrete basis from
the square integrable functions on the line while the Laguerre and Legendre polynomials are defined
on the half-line and the interval (−1, +1), respectively. From the point of view of Lie groups, the Her-
mite polynomials support the UIR of the Weyl–Heisenberg group associated to the Casimir C = 0
and the Laguerre and Legendre polynomials are related to the fundamental UIR (C = −1/4) of the
discrete series of SU (1, 1) that are all explicitly constructed at the Lie algebra level.
Summarizing, in all the three cases we introduce in the vector space the countable basis {|n⟩}n∈N ,
related to the uncountable coordinate basis {|x⟩}x∈E ⊂R by the appropriate orthogonal functions, on
which an infinite-dimensional IR of a non-compact Lie algebra (the Weyl–Heisenberg algebra h(1)
for Hermite and the su(1, 1) algebra for Laguerre and Legendre) is defined. As this basis {|n⟩} and
the associated orthogonal functions are in one-to-one correspondence, the algebraic structure can be
transferred from the L2 (E ) space straight to the space E.
As we are dealing with irreducible representations, all the operators acting on L2 (E ) can be gener-
ated in the UEA. In particular, the action of the elements of the group allows to construct all possible
alternative bases of the space L2 (E ).
A first result related to the introduction of the algebraic structure is that coherent states defined
on the vector space {|n⟩} by the Weyl–Heisenberg algebra can be transferred to the wave functions,
i.e. the Hermite polynomials and, analogously, coherent Laguerre or Legendre polynomials can be
defined à la Perelomov [7] starting from the properties of the appropriate vector space.
In conclusion, we believe that – combining properties of different origin like differential equations,
the factorization method, theory of Lie representations, Hilbert spaces and integrable functions – a
better description of the orthogonal polynomials has been obtained.

Acknowledgments

This work was partially supported by the Ministerio de Educación y Ciencia of Spain (Projects
FIS2009-09002 with EU-FEDER support), by the Junta de Castilla y León, by INFN-MICINN (Italy–Spain)
and the Universidad de Valladolid.

References

[1] S. Cambanis, Proc. Amer. Math. Soc. 29 (1971) 284.

[2] G. ’ t Hooft, S. Noobbenhuis, Special Functions and Polynomials, www.phys.uu.nl/~hooft101/GtH_lectures.html.
[3] P. Szekeres, A Course in Modern Mathematical Physics, Cambridge Univ. Press, Cambridge, 2006.
[4] L. Infeld, T.E. Hull, Rev. Modern Phys. 23 (1951) 21.
[5] W. Miller, Lie Theory and Special Functions, Academic Press, New York, 1968.
[6] E. Celeghini, M. Tarlini, Il Nuovo Cimento B61 (1981) 265.
[7] A. Perelomov, Generalized Coherent States and Applications, Springer, Berlin, 1986.
[8] F.W.J. Olver, D.W. Lozier, R.F. Boisvert, C.W. Clark, NIST Handbook of Mathematical Functions, Cambridge Univ. Press, New
York, 2010.
[9] E. Celeghini, M.A. del Olmo, Ann. Phys. 333 (2013) 90.