nonclassical jacobi polynomials and sobolev orthogonality

31
Results. Math. 61 (2012), 283–313 c 2011 Springer Basel AG 1422-6383/12/030283-31 published online March 15, 2011 DOI 10.1007/s00025-011-0102-4 Results in Mathematics Nonclassical Jacobi Polynomials and Sobolev Orthogonality Andrea Bruder and L. L. Littlejohn Abstract. In this paper, we consider the second-order Jacobi differential expression α,β [y](x)= 1 (1 x) a (1 + x) 1 ((1 x) α+1 y (x)) (x (1, 1)); here, the Jacobi parameters are α> 1 and β = 1. This is a non- classical setting since the classical setting for this expression is generally considered when α, β > 1. In the classical setting, it is well-known that the Jacobi polynomials {P (α,β) n } n=0 are (orthogonal) eigenfunctions of a self-adjoint operator T α,β , generated by the Jacobi differential expres- sion, in the Hilbert space L 2 ((1, 1); (1 x) α (1 + x) β ). When α> 1 and β = 1, the Jacobi polynomial of degree 0 does not belong to the Hilbert space L 2 ((1, 1); (1 x) α (1 + x) 1 ). However, in this paper, we show that the full sequence of Jacobi polynomials {P (α,1) n } n=0 forms a complete orthogonal set in a Hilbert–Sobolev space Wα, generated by the inner product φ (f,g) := f (1) g(1) + 1 1 f (t) g (t)(1 t) α+1 dt. We also construct a self-adjoint operator Tα, generated by α,1[·] in Wα, that has the Jacobi polynomials {P (α,1) n } n=0 as eigenfunctions. Mathematics Subject Classification (2010). Primary 33C45, 34B30, 47B25; Secondary 34B20, 47B65.

Upload: l-l

Post on 23-Aug-2016

212 views

Category:

Documents


0 download

TRANSCRIPT

Results. Math. 61 (2012), 283–313c© 2011 Springer Basel AG1422-6383/12/030283-31published online March 15, 2011DOI 10.1007/s00025-011-0102-4 Results in Mathematics

Nonclassical Jacobi Polynomialsand Sobolev Orthogonality

Andrea Bruder and L. L. Littlejohn

Abstract. In this paper, we consider the second-order Jacobi differentialexpression

�α,β [y](x) =−1

(1 − x)a(1 + x)−1((1 − x)α+1 y′(x))′ (x ∈ (−1, 1));

here, the Jacobi parameters are α > −1 and β = −1. This is a non-classical setting since the classical setting for this expression is generallyconsidered when α, β > −1. In the classical setting, it is well-known that

the Jacobi polynomials {P(α,β)n }∞

n=0 are (orthogonal) eigenfunctions ofa self-adjoint operator Tα,β , generated by the Jacobi differential expres-sion, in the Hilbert space L2((−1, 1); (1 − x)α(1 + x)β). When α > −1and β = −1, the Jacobi polynomial of degree 0 does not belong to theHilbert space L2((−1, 1); (1 − x)α(1 + x)−1). However, in this paper, we

show that the full sequence of Jacobi polynomials {P(α,−1)n }∞

n=0 forms acomplete orthogonal set in a Hilbert–Sobolev space Wα, generated by theinner product

φ (f, g) := f(−1)g(−1) +

1∫

−1

f ′(t)g′(t)(1 − t)α+1dt.

We also construct a self-adjoint operator Tα, generated by �α,−1[·] in Wα,

that has the Jacobi polynomials {P(α,−1)n }∞

n=0 as eigenfunctions.

Mathematics Subject Classification (2010). Primary 33C45, 34B30, 47B25;Secondary 34B20, 47B65.

284 A. Bruder and L. L. Littlejohn Results. Math.

1. Introduction

For α, β > −1, the special functions properties as well as the spectral proper-ties of the classical Jacobi differential expression

lα,β [y](x) :=1

ωα,β(x)

×[((−(1 − x)α+1(1 + x)β+1

)y′(x)

)′+ k(1 − x)α(1 + x)βy(x)

]

= −(1 − x2)y′′(x) + (α− β + (α+ β + 2)x)y′(x) + ky(x)

where k ≥ 0 is fixed, x ∈ (−1, 1) and ωα,β(x) = (1−x)α(1+x)β are well known.In this case, the nth degree Jacobi polynomial y = Pα,β

n (x) is a solution of theequation

lα,β [y](x) = (n(n+ α+ β + 1) + k) y(x) (n ∈ N0);

details of the properties of these polynomials can be found in [7,16]. The right-definite spectral analysis has been studied in [1,9]. Through theGlazman–Krein–Naimark (GKN) theory it has been known that there exists aself-adjoint operator A(α,β) generated from the Jacobi differential expressionin the Hilbert space L2((−1, 1); (1−x)α(1+x)β) having the Jacobi polynomialsas a complete set of eigenfunctions.

For α, β ≥ −1, let

L2α,β(−1, 1) := L2((−1, 1); (1 − x)α(1 + x)β) (1.1)

be the weighted Hilbert space with usual inner product

(f, g)α,β =

1∫

−1

f(x)g(x)(1 − x)α(1 + x)βdx (1.2)

and related norm ‖·‖α,β . We will study the nonclassical Jacobi case whenα > −1 and β = −1. One main difference to the classical case is that theconstant Jacobi polynomial Pα,−1

0 (x) does not belong to the Hilbert spaceL2

α,−1(−1, 1). However, it is true that the Jacobi polynomials {P (α,−1)n }∞

n=1

still form a complete orthogonal set in L2α,−1(−1, 1). Moreover, by the GKN

theory, there exists a self-adjoint differential operator A(α,−1), generated bylα,−1[·], that is positively bounded below in L2

α,−1(−1, 1), that has the set

{P (α,−1)n }∞

n=1as eigenfunctions. Furthermore, in [12], Kwon and Littlejohnshowed that the full sequence of Jacobi polynomials {P (α,−1)

n }∞n=0 are orthog-

onal in a Hilbert–Sobolev function space Wα with inner product

φ (f, g) := f(−1)g(−1) +

1∫

−1

f ′(x)g′(x)(1 − x)α+1dx. (1.3)

Vol. 61 (2012) Nonclassical Jacobi Polynomials and Sobolev Orthogonality 285

In this paper we prove that the entire sequence of Jacobi polynomials{P (α,−1)

n }∞n=0 are, in fact, complete in Wα. More importantly, we construct

a self-adjoint, positively bounded below operator Tα, generated from lα,−1[·],in Wα having the entire set of Jacobi polynomials {P (α,−1)

n }∞n=0 as a complete

set of eigenfunctions. The general left-definite theory, that was recently devel-oped by Littlejohn and Wellman [13] is, surprisingly, of paramount importancein the construction of this self-adjoint operator.

We note that, for m ∈ N, the Jacobi polynomials {P (α,−m)n }∞

n=0 areorthogonal with respect to inner products of the form (1.3) but whose inte-grand involves the mth derivative of the functions. In this respect, we referthe reader to [2–4,11,12] where general results on the Sobolev orthogonalityof the Jacobi or Gegenbauer polynomials when one or both parameters α andβ are negative integers.

2. Preliminaries: Properties of the Jacobi Polynomials

For α, β > −1, the Jacobi polynomials {P (α,β)n (x)}∞

n=0 are defined by

P (α,β)n (x) := kα,β

n

n∑j=0

(1 + α)n(1 + α+ β)n+j

j!(n− j)!(1 + α)j(1 + α+ β)n

(1 − x

2

)j

, (2.1)

where

kα,βn :=

(n!)1/2(1 + α+ β + 2n)1/2(Γ(α+ β + n+ 1))1/2

2(α+β+1)/2(Γ(α+ n+ 1))1/2(Γ(β + n+ 1))1/2;

see [7,15,16] for detailed discussions of these polynomials. For each n ∈ N0,

y = P(α,β)n (x) is a solution of the Jacobi differential equation

lα,β [y](x) = (n(n+ α+ β + 1) + k) y(x).

The Jacobi polynomials can always be normalized, in any number of ways,and we will assume that the polynomials are normalized in various spacesthroughout this paper. They form a complete orthogonal set in the Hilbertspace L2

α,β(−1, 1). In fact, they satisfy the orthogonality relation

1∫

−1

P (α,β)n (x)P (α,β)

r (x)(1 − x)α(1 + x)βdx

=2α+β+1Γ(n+ α+ 1)Γ(n+ β + 1)

(2n+ α+ β + 1)Γ(n+ α+ β + 1)n!δn,r

for α, β > −1. The derivatives of the Jacobi polynomials satisfy the identity

dj

dxjP (α,β)

n (x) = a(α,β)(n, j)P (α+j,β+j)n−j (x) (n, j ∈ N0), (2.2)

286 A. Bruder and L. L. Littlejohn Results. Math.

where

a(α,β)(n, j) =(n!)1/2 (Γ(α+ β + n+ 1 + j))1/2

((n− j)!)1/2 (Γ(α+ β + n+ 1))1/2(j = 0, 1, . . . , n),

and a(α,β)(n, j) = 0 if j > n. Furthermore, for n, r, j ∈ N0, we have theorthogonality relation

1∫

−1

dj(P

(α,β)n (x)

)

dxj

dj(P

(α,β)r (x)

)

dxjwα+j,β+j(x)dx (2.3)

=n!Γ(α+ β + n+ 1 + j)

(n− j)!Γ(α+ β + n+ 1)δn,r. (2.4)

When α > −1 and β = −1, the natural setting for the Jacobi polynomi-als {P (α,−1)

n } is still the Hilbert space L2α,−1(−1, 1). However, because of the

singularity in the weight function, P (α,−1)0 /∈ L2

α,−1(−1, 1). Moreover, with thedefinition of the Jacobi polynomials as given in, say [7],

P (α,−1)n (x) :=

n∑j=0

(n+ α

j

)(n− 1n− j

)(x− 1

2

)j (x+ 1

2

)n−j

,

notice that the first Jacobi polynomial P (α,−1)1 (x) is degenerate. However, any

multiple of the first degree polynomial y = x+1 will be a solution of the Jacobidifferential equation

lα,−1[y](x) = (n(n+ α) + k)y(x).

Therefore, by renormalizing the sequence of Jacobi polynomials as follows:

P(α,−1)1 (x) =

√(α+ 1)(α+ 2)

2α+2(x+ 1) ,

and, for n ≥ 2,

P (α,−1)n (x) =

√n (2n+ α)2α (n+ α)

n∑j=0

(n+ α

n− j

)(n− 1j

)(x− 1

2

)j (x+ 1

2

)n−j

,

the sequence {P (α,−1)n }∞

n=1 is an orthonormal set in L2α,−1(−1, 1). We remark

that the full set of Jacobi polynomials {P (α,−1)n }∞

n=0 cannot be orthogonal onthe real line with respect to any bilinear form of type∫

R

fgdμ,

where μ is a (possibly signed) measure. This is an application of Favard’sclassic theorem in orthogonal polynomials. However,

Vol. 61 (2012) Nonclassical Jacobi Polynomials and Sobolev Orthogonality 287

Lemma 1. The sequence {P (α,−1)n (x)}∞

n=1 of Jacobi polynomials of degree n ≥ 1forms a complete orthogonal set in the Hilbert space L2

α,−1(−1, 1). Equivalently,the set of all polynomials p ∈ P[−1, 1] of degree ≥ 1 satisfying p(−1) = 0 isdense in the space L2

α,−1(−1, 1).

Proof. We have

1∫

−1

|f(x)|2 (1 − x)α(1 + x)−1dx =

1∫

−1

∣∣(1 + x)−1f(x)∣∣2 (1 − x)α(1 + x)dx,

that is to say,

f ∈ L2α,−1(−1, 1) ⇐⇒ (1 + x)−1f ∈ L2

α,1(−1, 1),

and in this case,

‖f‖L2α,−1

=∥∥(1 + x)−1f

∥∥L2

α,1.

Let f ∈ L2α,−1(−1, 1) and let ε > 0. Then

(1 + x)−1f ∈ L2α,1(−1, 1);

since the Jacobi polynomials {P (α,1)n (x)}∞

n=0 forms a complete orthogonal setin L2

α,1(−1, 1), there exists q ∈ P[−1, 1] such that∥∥(1 + x)−1f − q

∥∥L2

α,1< ε.

Let p(x) := (1 + x)q(x), so deg(p) ≥ 1, p(−1) = 0, and q(x) = (1 + x)−1p(x).Then

ε >∥∥(1 + x)−1f − (1 + x)−1p

∥∥L2

α,1=∥∥(1 + x)−1(f − p)

∥∥L2

α,1= ‖f − p‖L2

α,−1.

3. An Operator Inequality

The following result, due to Chisholm, Everitt, and Littlejohn [6], is impor-tant in establishing our main results. The authors in [6] proved a more generalresult for conjugate indices p and q; for the purposes of this paper, we onlystate the result when p = q = 2.

Theorem 1. Suppose I = (a, b) is an open interval of the real line, where−∞ ≤ a < b ≤ ∞. Suppose w is a positive Lebesgue measurable function on(a, b) and ϕ,ψ are functions satisfying the three conditions:

(i) ϕ ∈ L2loc((a, b);w) and ψ ∈ L2

loc((a, b);w);(ii) for some c ∈ (a, b), suppose ϕ ∈ L2((a, c];w) and ψ ∈ L2([c, b);w);

288 A. Bruder and L. L. Littlejohn Results. Math.

(iii) for all [α, β] ⊂ (a, b),α∫

a

|ϕ(t)|2 w(t)dt > 0 and

b∫

β

|ψ(t)|2 w(t)dt > 0.

Define the linear operators A and B on L2((a, b);w) and L2((a, b);w), respec-tively, by

(Ag)(x) := ϕ(x)

b∫

x

ψ(t)g(t)w(t)dx (x ∈ (a, b) and g ∈ L2((a, b);w))

(Bg)(x) := ψ(x)

x∫

a

ϕ(t)g(t)w(t)dx (x ∈ (a, b) and g ∈ L2((a, b);w));

then

A : L2((a, b);w) → L2loc((a, b) : w)

B : L2((a, b);w) → L2loc((a, b);w).

Define K(·) : (a, b) → (0,∞) by

K(x) :=

⎧⎨⎩

x∫

a

|ϕ(t)|2 w(t)dx

⎫⎬⎭

1/2⎧⎨⎩

b∫

x

|ψ(t)|2 w(t)dt

⎫⎬⎭

1/2

(x ∈ (a, b))

and the number K ∈ (0,∞]

K := sup{K(x) | x ∈ (a, b)}.Then a necessary and sufficient condition that A and B are bounded linearoperators on L2((a, b);w) is that the number K is finite, i.e.

K ∈ (0,∞).

4. Right-Definite Analysis of the Jacobi Expression Whenα > −1 and β = −1

Recall that, for α > −1 and β = −1, the second-order Jacobi differentialexpression is defined to be

α,−1[y](x) := (1 − x2)y′′(x) − (α+ 1)(x+ 1)y′(x) + ky(x)

=1

wα,−1(x)[((1 − x)α+1y′(x))′ + k(1 − x)α(1 + x)−1y], (4.1)

where wα,−1(x) := (1−x)α(1+x)−1, x ∈ (−1, 1) and k is a fixed, non-negativeconstant.

Both points x = ±1 are regular singular endpoints of lα,−1[·] in the senseof Frobenius. In fact, in the case α ≥ 1 and β = −1, both endpoints x = ±1

Vol. 61 (2012) Nonclassical Jacobi Polynomials and Sobolev Orthogonality 289

are limit-point in L2α,−1(−1, 1), defined in (1.1), and thus the right-definite

GKN self-adjoint operator is unique and there are no boundary conditionsprescribed in the domain of A(α,−1). For −1 < α < 1, the endpoint x = −1 isin the limit-point condition, whereas x = +1 is in the limit-circle case. There-fore, when −1 < α < 1, one boundary condition is necessary to define theself-adjoint operator A(α,−1) having the Jacobi polynomials {P (α,−1)

n }∞n=1 as

eigenfunctions.For α > −1, the maximal domain Δα,−1 of α,−1[·] in the Hilbert space

L2α,−1(−1, 1) is defined to be

Δα,−1 :={f : (−1, 1) → C |f, f ′ ∈ ACloc(−1, 1); f, α,−1[f ] ∈ L2

α,−1(−1, 1)}

={f : (−1, 1) → C |f, f ′ ∈ ACloc(−1, 1);

(1 − x)α/2

(1 + x)1/2f ∈ L2(−1, 1);

(1 − x)−α/2(1 + x)1/2((1 − x)α+1f ′(x)

)′ ∈ L2(−1, 1)}.

The maximal operator T (α,−1)max associated with lα,−1[·] is given by

T (α,−1)max (f) := lα,−1[f ]

D(T (α,−1)max ) := Δα,−1.

The minimal operator is defined as T (α,−1)min := (T (α,−1)

max )∗, the Hilbert spaceadjoint of T (α,−1)

max . The operator T (α,−1)min is closed, symmetric and satisfies

(T (α,−1)min )∗ = T (α,−1)

max .

The deficiency index d(T (α,−1)min ) of T (α,−1)

min is

d(T (α,−1)min ) =

{(0, 0) if β = −1, α ≥ 1(1, 1) if β = −1, −1 < α < 1.

This can be seen from the limit-point/limit-circle classification of the singularendpoints x = ±1:

(i) x = ±1 are limit-point if β = −1, α ≥ 1 and(ii) x = −1 is limit-point, x = 1 is limit-circle if β = −1,−1 < α < 1.

Consequently, by von Neumann’s theory of self-adjoint extensions of symmetricoperators ([8], chapter XII), T (α,−1)

min has self-adjoint extensions in L2α,−1(−1, 1)

for α > −1 and β = −1.We define the operator A(α,−1) : D(A(α,−1)) ⊂ L2

α,−1(−1, 1) → L2α,−1

(−1, 1) as follows:{

(A(α,−1)f)(x) := lα,−1[f ](x) (a.e. x ∈ (−1, 1))f ∈ D(A(α,−1)),

(4.2)

290 A. Bruder and L. L. Littlejohn Results. Math.

where D(A(α,−1)) is defined by

D(A(α,−1)) :=

{{f ∈ Δα,−1 | limx→1−(1 − x)α+1f ′(x) = 0} if − 1 < α < 1

Δα,−1 if α ≥ 1.

(4.3)

From the GKN theory, A(α,−1) is self-adjoint in L2α,−1(−1, 1). More impor-

tantly, from our point of view, the Jacobi polynomials {P (α,−1)n }∞

n=1 of degree≥ 1 form a complete set of eigenfunctions of A(α,−1); see Lemma 1. Further-more, the spectrum of A(α,−1) is discrete and given by

σ(A(α,−1)) = {n(n+ α) | n ∈ N}.In order to develop left-definite properties of A(α,−1) which, in turn, are

important to construct the self-adjoint operator Tα in the Sobolev space Wα,we turn our attention to developing properties of functions in the spaces Δα,−1

and D(A(α,−1)).Observe that if f, g ∈ Δα,−1, then α,−1[f ]g ∈ L2

α,−1(−1, 1). Furthermore,for f, g ∈ Δα,−1, Green’s formula gives us

x∫

0

(1 − t)α+1f ′(t)g′(t)dt = (1 − x)α+1f ′(x)g(x) − f ′(0)g(0)

−x∫

0

α,−1[f ](t)g(t)(1 − t)α(1 + t)−1dt

+k

x∫

0

f(t)g(t)(1 − t)α(1 + t)−1dt (−1 < x < 1).

(4.4)

Lemma 2. For α > −1 and f ∈ Δα,−1, we have f ′ ∈ L2(−1, 0). In particular,we can define f at x = −1 such that f ∈ AC[−1, 0]. Moreover, it is necessarythat f(−1) = 0.

Proof. Let f ∈ Δα,−1. Note that

f ′(x) = −(1 − x)−α−1

0∫

x

(1 − t)−α/2(1 + t)1/2

(1 − t)−α/2(1 + t)1/2((1 − t)α+1f ′(t))′dt

+(1 − x)−α−1f ′(0) (x ∈ (−1, 0]).

From the definition of Δα,−1, we see that (1−t)−α/2(1+t)1/2((1−t)α+1f ′(t))′ ∈L2(−1, 1); consequently, we apply the inequality in Theorem 1 on the interval(−1, 0] with

ϕ(x) = (1 − x)−α−1, ψ(x) =1

(1 − x)−α/2(1 + x)1/2(x ∈ (−1, 0]).

Vol. 61 (2012) Nonclassical Jacobi Polynomials and Sobolev Orthogonality 291

Since the function 1 − t is bounded on (−1, 0], there exists a constant M suchthat

x∫

−1

ϕ2(t)dt

0∫

x

ψ2(t)dt =

x∫

−1

(1 − t)−2α−2dt

0∫

x

(1 − t)α(1 + t)−1dt

≤ M

x∫

−1

dt

0∫

x

(1 + t)−1dt = −M(x+ 1) ln(1 + x);

since (x + 1) ln(1 + x) is bounded on (−1, 0], the inequality from Theorem 1implies that f ′ ∈ L2(−1, 0) as claimed. From the identity

f(x) = f(0) −0∫

x

f ′(t)dt (−1 < x ≤ 0),

define

f(−1) := f(0) −0∫

−1

f ′(t)dt.

With f defined this way, we see that f ∈ AC[−1, 0].We now show that f(−1) =0. We suppose that f is real-valued and, by way of contradiction and withoutloss of generality, we assume that f(−1) > 0. Then there exists x∗ ∈ (−1, 0]such that f(t) ≥ c for all t ∈ (−1, x∗], where c is some positive number. Since(1 − x)α is bounded below on (−1, 0] by a constant K > 0, we see that

−∞ >

1∫

−1

|f(t)|2 (1 − t)α(1 + t)−1dt

≥x∗∫

−1

|f(t)|2 (1 − t)α(1 + t)−1dt ≥ Kc2x∗∫

−1

(1 + t)−1dt = ∞,

a contradiction. Hence f(−1) = 0. �

Lemma 3. For α > −1 and f ∈ Δα,−1, we have (1 − x)(α+1)/2f ′ ∈ L2(−1, 0).

Proof. Suppose, to the contrary, that

0∫

−1

|f ′(t)|2 (1 − t)α+1dt = ∞.

292 A. Bruder and L. L. Littlejohn Results. Math.

We assume, without loss of generality, that f is real-valued. From (4.4), we seethat

0∫

x

(1 − t)α+1(f ′(t))2dt = f ′(0)f(0) − (1 − x)α+1f ′(x)f(x)

−0∫

x

α,−1[f ](t)f(t)(1 − t)α(1 + t)−1dt

+k

x∫

0

(f(t))2(1 − t)α(1 + t)−1dt (−1 < x ≤ 0);

(4.5)

consequently, it must be the case that

limx→−1+

(1 − x)α+1f ′(x)f(x) = −∞,

and hence

limx→−1+

f ′(x)f(x) = −∞.

It follows that there exists x∗ ∈ (−1, 0], f ′(x)f(x) ≤ −1 for x ∈ (−1, x∗].Integrating, we see that

(f(x∗))2

2− (f(x))2

2=

x∗∫

x

f ′(t)f(t)dt ≤ −1

x∗∫

x

dt = x− x∗

so that

(f(x))2 ≥ −2x+ (f(x∗))2 + 2x∗ ≥ −2(x− x∗) (x ∈ (−1, x∗]). (4.6)

With

Mα = mint∈(−1,x∗]

(1 − t)α,

after integrating the inequality in (4.6), we see that

x∗∫

x

(f(t))2(1 − t)α(1 + t)−1dt ≥ −2

x∗∫

x

(t− x∗)(1 − t)α(1 + t)−1dt

≥ −2Mα

x∗∫

x

(t− x∗)(1 + t)−1dt

→ ∞ as x → −1+.

Vol. 61 (2012) Nonclassical Jacobi Polynomials and Sobolev Orthogonality 293

However, this contradicts the fact that f ∈ Δα,−1, and, in particular, that

1∫

−1

|f(t)|2 (1 − t)α(1 + t)−1dt < ∞.

Lemma 4. Let f, g ∈ Δα,−1. Then

limx→−1+

(1 − x)α+1f ′(x)g(x) = 0.

Proof. Let f, g ∈ Δα,−1; assume that both f and g are real-valued. We notethat, from Lemma 2, that f(−1) = g(−1) = 0. From (4.5) and Lemma 3, wesee that

limx→−1+

(1 − x)α+1f ′(x)g(x)

exists and is finite. Suppose, then, that

limx→−1+

(1 − x)α+1f ′(x)g(x) = c > 0.

Then we can assume there exists x∗ ∈ (−1, 0] such that

f ′(x) > 0, g(x) > 0, and (1 − x)α+1f ′(x) ≥ c

2g(x)(x ∈ (−1, x∗]).

Multiply this inequality by |g′(x)| to get

(1 − x)α+1f ′(x) |g′(x)| ≥ c

2|g′(x)|g(x)

(x ∈ (−1, x∗]).

Now integrate to obtain

∞ >

1∫

−1

(1 − t)α+1f ′(t) |g′(t)| dt (using Lemma 3)

≥x∗∫

x

(1 − t)α+1f ′(t) |g′(t)| dt ≥ c

2

x∗∫

x

|g′(t)|g(t)

dt ≥ c

2

∣∣∣∣∣∣x∗∫

x

g′(t)g(t)

dt

∣∣∣∣∣∣=c

2|ln g(x∗) − ln g(x)| → ∞ as x → −1+ since g(−1) = 0.

This contradiction gives us the required result. �

We summarize the results from Lemmas 2, 3, and 4.

Theorem 2. Assume α > −1 and β = −1. Let f, g ∈ Δα,−1. Then(a) f ′ ∈ L2(−1, 0);(b) f ∈ AC[−1, 0];

294 A. Bruder and L. L. Littlejohn Results. Math.

(c) The Jacobi differential expression (4.1) is strong limit point at x = −1;that is to say,

limx→−1+

(1 − x)α+1f ′(x)g(x) = 0;

(d) The Jacobi differential expression (4.1) is Dirichlet at x = −1; that is tosay,

0∫

−1

(1 − x)α+1 |f ′(x)|2 dx < ∞.

As a consequence of this theorem, note that for f, g ∈ Δα,−1, and for−1 < x ≤ 0,

0∫

−1

(1 − t)α+1f ′(t)g′(t)dt = f ′(0)g(0) −0∫

x

α,−1[f ](t)g(t)(1 − t)α(1 + t)−1dt.

(4.7)

We now turn our attention to establishing that the Jacobi differentialexpression (4.1) is both strong-limit point and Dirichlet at x = 1 on D(A(α,−1)),defined in (4.3).

Lemma 5. Let α > −1. Then, for all f ∈ D(A(α,−1)),

limx→1−

(1 − x)α+1f ′(x) = 0.

Proof. From the definition of D(A(α,−1)) in ( 4.3), this result is automaticallytrue for −1 < α < 1. So we assume α ≥ 1. Let f ∈ D(A(α,−1)). Since theJacobi expression is limit point at x = 1 in this case, we know that

limx→1−

(1 − x)α+1[f ′(x)g(x) − f(x)g′(x)] = 0 (g ∈ Δα,−1).

Construct a real-valued g ∈ C2[−1, 1] such that

g(x) ={

0 for x near − 11 for x near + 1;

for example, we could take

g(x) =

⎧⎨⎩

0 if − 1 < x ≤ 0−16x3 + 12x2 if 0 < x < 1/21 if 1/2 ≤ x ≤ 1.

Clearly g ∈ Δα,−1; moreover, with this choice of g, we have

0 = limx→1−

(1 − x)α+1[f ′(x)g(x) − f(x)g′(x)

= limx→1−

(1 − x)α+1f ′(x).

Vol. 61 (2012) Nonclassical Jacobi Polynomials and Sobolev Orthogonality 295

The proof of the following lemma is similar to the proof given in Lemma 4;we omit the details. This result shows that the Jacobi expression is Dirichletat x = 1 on D(A(α,−1)).

Lemma 6. Suppose α > −1. Then, for any f ∈ D(A(α,−1)), we have

(1 − x)(α+1)/2f ′ ∈ L2(0, 1).

We are now in position to prove that the Jacobi expression (4.1) is stronglimit point at x = 1 on D(A(α,−1)). The proof is similar to the one of Lemma 4but different enough to warrant a demonstration.

Lemma 7. Suppose α > −1. Then for f, g ∈ D(A(α,−1)), we have

limx→1−

(1 − x)α+1f ′(x)g(x) = 0.

Proof. Let f, g ∈ D(A(α,−1)). From (4.4), the definition of D(A(α,−1)), andLemma 6, we see that

limx→1−

(1 − x)α+1f ′(x)g(x)

exists and is finite. Without loss of generality, assume that both f and g arereal-valued on (−1, 1); furthermore, by way of contradiction, we assume that

limx→1−

(1 − x)α+1f ′(x)g(x) = c > 0.

Again, we assume that there exists x∗ ∈ [0, 1) such that

(1 − x)α+1f ′(x) > 0, g(x) > 0, (1 − x)α+1f ′(x)g(x) ≥ c/2 (x ∈ [x∗, 1)).

Hence it follows that

∣∣∣((1 − x)α+1f ′(x))′∣∣∣ g(x) ≥ c

2

∣∣∣((1 − x)α+1f ′(x))′∣∣∣

(1 − x)α+1f ′(x)(x ∈ [x∗, 1)).

Integrate over [x∗, x] to get

∞ >

1∫

−1

α,−1[f ](t)g(t)(1 − t)α(1 + t)−1dt− k

1∫

−1

(f(t))2(1 − t)α(1 + t)−1dt

=

1∫

−1

∣∣∣((1 − t)α+1f ′(t))′∣∣∣ g(t)dt ≥

x∗∫

x

∣∣∣((1 − t)α+1f ′(t))′∣∣∣ g(t)dt

≥ c

2

x∗∫

x

∣∣∣((1 − t)α+1f ′(t))′∣∣∣

(1 − t)α+1f ′(t)dt ≥ c

2

∣∣∣∣∣∣x∗∫

x

((1 − t)α+1f ′(t)

)′

(1 − t)α+1f ′(t)dt

∣∣∣∣∣∣=c

2

∣∣ln((1 − x∗)α+1f ′(x∗)) − ln((1 − x)α+1f ′(x))∣∣ → ∞ as x → 1−

by Lemma 5. This contradiction gives us the required result. �

296 A. Bruder and L. L. Littlejohn Results. Math.

Summarizing Lemmas 6 and 7, we have the following theorem.

Theorem 3. For α > −1 and β = −1, the Jacobi differential expression α,−1[·],given in (4.1), is strong limit point and Dirichlet at x = 1 on D(A(α,−1)). Thatis to say,(a) limx→1−(1 − x)α+1f ′(x)g(x) = 0 for all f, g ∈ D(A(α,−1)), and(b)

∫ 1

0(1 − x)α+1 |f ′(x)|2 dx < ∞ for all f ∈ D(A(α,−1)).

Combining Theorems 2 and 3, we see that Dirichlet’s formula, for f, g ∈D(A(α,−1)), over the interval (−1, 1) reads:

(A(α,−1)f, g)α,−1 =

1∫

−1

[(1 − t)α+1f ′(t)g′(t) + kf(t)g(t)(1 − t)α(1 + t)−1]dt.

(4.8)

In particular, we see that

(A(α,−1)f, f)α,−1 ≥ k(f, f)α,−1 (f ∈ D(A(α,−1))); (4.9)

that is to say, A(α,−1) is bounded below by kI in L2α,−1(−1, 1). The conse-

quence of the inequality in (4.9) is that the left-definite theory, developed byLittlejohn and Wellman, can be applied to A(α,−1). We now briefly review thistheory.

5. General Left-Definite Theory

In [13], Littlejohn and Wellman developed a general abstract left-definite the-ory for a self-adjoint operator A that is bounded below in a Hilbert space(H, (·, ·)).

Let V be a vector space over C with inner product (·, ·) such that H :=(V, (·, ·)) is a Hilbert space. Let r > 0 and suppose that Vr is a vector sub-space of V with inner product (·, ·)r; denote this inner product space by Wr :=(Vr, (·, ·)r). Let A : D(A) ⊂ H → H be a self-adjoint operator that is boundedbelow by rI for some r > 0, that is to say

(Ax, x) ≥ r(x, x) (x ∈ D(A)).

Then, for any s > 0, the operator As is self-adjoint and bounded below in Hby rsI.

Definition 1. Let s > 0 and let Vs be a vector subspace of the Hilbert spaceH = (V, (·, ·)) with inner product (·, ·)s. We say that Ws = (Vs, (·, ·)s is an sth

left-definite space associated with the pair (H,A) if

(i) Ws is a Hilbert space(ii) D(As) is a vector subspace of Vs

(iii) D(As) is dense in Ws

(iv) (x, x)s ≥ rs(x, x) for all x ∈ Vs

Vol. 61 (2012) Nonclassical Jacobi Polynomials and Sobolev Orthogonality 297

(v) (x, y)s = (Asx, y) for all x ∈ D(As), y ∈ Vs.

Littlejohn and Wellman in [13, Theorem 3.1] prove the following result.

Theorem 4. Let A : D(A) ⊂ H → H be a self-adjoint operator that is boundedbelow by rI for some r > 0. Let s > 0 and define Ws := (Vs, (·, ·)s) by

Vs = D(As/2)

and

(x, y)s = (As/2x,As/2y) (x, y ∈ Vs).

Then Ws is the unique left-definite space associated with the pair (H,A).

Definition 2. For s > 0, let Ws := (Vs, (·, ·)s) be the sth left-definite spaceassociated with (H,A). If there exists a self-adjoint operator Bs : D(Bs) ⊂Ws → Ws satisfying

Bsf = Af (f ∈ D(Bs) ⊂ D(A)),

we call such an operator an sth left-definite operator associated with the pair(H,A).

In [13, Theorem 3.2], the authors prove the following existence/unique-ness result.

Theorem 5. Let A be a self-adjoint operator in a Hilbert space H that isbounded below by rI for some r > 0. For any s > 0, let Ws := (Vs, (·, ·)s)denote the sth left-definite space associated with (H,A). Then there exists aunique left-definite operator Bs in Ws associated with (H,A). Furthermore,

D(Bs) = Vs+2 ⊂ D(A).

We refer the reader to other results established in [13]. We note if, inaddition to the hypotheses assumed in this section, that A is bounded thenthe left-definite theory is trivial; specifically, V = Vs and A = As for alls > 0. Only in the unbounded case will there be a non-trivial left-definite the-ory associated with (H,A). Moreover, we note that the authors in [13] proveσ(A) = σ(As) for all s > 0; in fact, the point spectrum σp and continuousspectrum σc match up for A and each left-definite operator As.

6. Left-Definite Theory of the Nonclassical Jacobi DifferentialEquation

Let k > 0. For each n ∈ N, define

V (α,−1)n := {f : (−1, 1) → C|f ∈ AC

(n−1)loc ; f (j) ∈ L2

α+j,j−1(−1, 1), j = 0, . . . , n}

298 A. Bruder and L. L. Littlejohn Results. Math.

and let (·, ·)(α,−1)n and ‖·‖(α,−1)

n denote, respectively, the Sobolev inner product

(f, g)(α,−1)n :=

n∑j=0

c(α,−1)j (n, k)

1∫

−1

f (j)(t)g(j)(t)(1 − t)α+j(1 + t)j−1dt

(f, g ∈ V(α,−1)n ), and norm ‖f‖(α,−1)

n :=((f, f)(α,−1)

n

)1/2

, where the numbers

c(α,−1)j (n, k) are defined as in [9] by

c(α,−1)0 (n, k) :=

{0 if k = 0kn if k > 0 ,

and, for j ∈ {1, 2, . . . , n},

c(α,−1)j (n, k) :=

{P (α,−1)S

(j)n if k = 0∑n−j

s=0

(ns

)P (α,−1)S

(j)n−sk

s if k > 0,

where the Jacobi–Stirling number P (α,−1)S(j)n of order (n, j) is defined by

P (α,−1)S(j)n :=

j∑r=0

(−1)r+j Γ(α+ r)Γ(α+ 2r + 1)[r(r + α)]n

r!(j − r)!Γ(α+ 2r)Γ(α+ j + r + 1),

(n, j ∈ N; j ≤ n). We extend this definition to include

P (α,−1)S(0)0 := 1, P (α,−1)S(j)

n := 0 if j ∈ N and 0 ≤ n ≤ j − 1,

P (α,−1)S(0)n := 0 for n ∈ N.

In [9], the authors prove that P (α,−1)S(j)n > 0 for n, j ∈ N, j ≤ n. Let

W (α,−1)n :=

(V (α,−1)

n , (·, ·)(α,−1)n

).

We will show that the vector space W (α,−1)n is the nth left-definite space

associated with the pair (L2α,−1(−1, 1), A(α,−1)), where A(α,−1) is the self-

adjoint Jacobi operator defined in (4.2) and (4.3). Using the results from [9]mutatis mutandis, Theorem 6 follows; for a proof see [5].

Theorem 6. Let k > 0. For each n ∈ N,W(α,−1)n is a Hilbert space.

Theorem 7. The Jacobi polynomials {P (α,−1)m }∞

m=1 form a complete orthogo-nal set in each W

(α,−1)n . Equivalently, the set of polynomials P is dense in

W(α,−1)n .

Proof. Fix n ∈ N , and let f ∈ W(α,−1)n , so

f (n) ∈ L2α+n,n−1(−1, 1).

Vol. 61 (2012) Nonclassical Jacobi Polynomials and Sobolev Orthogonality 299

Since {P (α+n,n−1)m }∞

m=0 is a complete orthonormal set in L2α+n,n−1(−1, 1), we

knowr∑

m=0

c(α,−1)m,n P (α+n,n−1)

m → f (n) as r → ∞ in L2α+n,n−1(−1, 1) (6.1)

where c(α,−1)m,n are the Fourier coefficients given by

c(α,−1)m,n =

1∫

−1

f (n)(t)P (α+n,n−1)m (t)(1 − t)α+n(1 + t)n−1dt

for m ∈ N0. For r ≥ n define the polynomials

pr(t) :=r∑

m=max{2,n}

c(α,−1)m−n,n ((m− n)!)1/2 (Γ(α+m))1/2

(m!)1/2 (Γ(α+m+ n)!)1/2P (α,−1)

m (t).

From

dj

dtjP (α,−1)

m (t) =(m!)1/2 (Γ(α+m+ j))1/2

((m− j)!)1/2 (Γ(α+m))1/2P

(α+j,j−1)m−j (t),

we see that, for j = 0, 1, . . . , n,

p(j)r (t) =

r∑m=max{2,n}

c(α,−1)m−n,n ((m− n)!)1/2 (Γ(α+m+ j)!)1/2

(Γ(α+m+ n)!)1/2 ((m− j)!)1/2P

(α+j,j−1)m−j (t).

In particular, by (6.1),

p(n)r (t) =

r∑m=max{2,n}

c(α,−1)m−n,nP

(α+n,n−1)m−n =

r−max{2,n}∑l=0

c(α,−1)l,n P

(α+n,n−1)l

=s∑

m=0

c(α,−1)m,n P (α+n,n−1)

m → f (n)

as r → ∞ in L2α+n,n−1(−1, 1). Furthermore, by Riesz-Fischer, there exists a

subsequence {p(n)rj } of {p(n)

r } such that

p(n)rj

→ f (n) for a.e. t ∈ (−1, 1).

By Dirichlet’s test, the sequence{c(α,−1)m−n,n ((m− n)!)1/2 (Γ(α+m+ j)!)1/2

(Γ(α+m+ n)!)1/2 ((m− j)!)1/2

}∈ 2,

so there exists a gj ∈ L2α+j,j−1(−1, 1) such that

prj→ gj in L2

α+j,j−1(−1, 1). (6.2)

300 A. Bruder and L. L. Littlejohn Results. Math.

For a.e. a, t ∈ (−1, 1),t∫

a

p(n)rj

(u)du →t∫

a

f (n)(u)du.

Integrate both sides and obtain

p(n−1)rj

(t) → f (n−1)(t) + c1 for a.e. t ∈ (−1, 1) (6.3)

for some constant c1. Passing through the subsequence implies

gn−1(t) = f (n−1)(t) + c1 for a.e. t ∈ (−1, 1).

From (6.3), we see thatt∫

a

p(n−1)rj

(u)du →t∫

a

f (n−1)(u)du+ c1

t∫

a

du,

that is,

p(n−2)rj

(t) → f (n−2)(t) + c1t+ c2 for a.e. t ∈ (−1, 1)

or

gn−2(t) = f (n−2)(t) + c1t+ c2 for a.e. t ∈ (−1, 1).

Continue this process to see that for j ∈ {0, 1, . . . , n− 1} ,gj(t) = f (j)(t) + qn−j+1 for a.e. t ∈ (−1, 1),

where qn−j−1 is a polynomial of degree ≤ n−j−1 and where q′n−j−1 = qn−j−2.

Hence, with (6.2),

p(j)r → f (j) + qn−j−1 in L2

α+j,j−1(−1, 1). (6.4)

For r ≥ n, define πr(t) := pr(t) − qn−1(t). Note that, with (6.4),

π(j)r (t) = p(j)

r (t) − q(j)n−1(t) = p(j)

r (t) − qn−j−1(t) → f (j)(t).

Hence, as r → ∞,

(‖f − πr‖(α,−1)

n,k

)2

=n∑

j=0

c(α,−1)j (n, k)

×−1∫

−1

∣∣∣f (j)(t) − π(j)r (t)

∣∣∣2 (1 − t)α+j(1 + t)j−1dt → 0.

Lemma 8. For p, q ∈ P, the vector space of all polynomials p : [−1, 1] → C,and for n ≥ 1,

(p, q)(α,−1)n =

((A(α,−1)

)n

p, q)

α,−1.

Vol. 61 (2012) Nonclassical Jacobi Polynomials and Sobolev Orthogonality 301

Proof. First we note that this may be restated as

(lnα,−1[p], q

)α,−1

=

1∫

−1

lnα,−1[p](x)q(x)wα,−1(x)dx

=n∑

j=0

c(α,−1)j (n, k)p(j)(x)q(j)(x)(1 − x)j+α(1 + x)j−1dx.

(6.5)

Since the Jacobi polynomials form a basis for P, it suffices to prove (6.5) forp = P

(α,−1)m and q = P

(α,−1)r for arbitrary m, r ∈ N0. From

lnα,−1[P(α,−1)m ](x) = (m(m− 1) + k)nP (α,−1)

m (x) (m ∈ N0)

and (P (α,−1)

r , P (α,−1)m

)α,−1

= δr,m (r,m ∈ N0) ,

the left-hand side of (6.5) becomes

(lnα,−1[P

(α,−1)m ], P (α,−1)

r

)α,−1

=

1∫

−1

lnα,−1[P(α,−1)m ](x)P (α,−1)

r (x)wα,−1(x)dx

= (m(m− 1) + k)nδr,m. (6.6)

Upon using (2.3) for α > −1, β = −1 and the recurrence relation for thec(α,−1)j (n, k), that is,

(m(m+ α) + k)n =n∑

j=0

c(α,−1)j (n, k)

m!(m+ α+ j − 1)!(m− j)!(m+ α− 1)!

the right-hand side of (6.5) becomesn∑

j=0

c(α,−1)j (n, k) (6.7)

×1∫

−1

(P (α,−1)

m (x))(j)

(x)(P

(α,−1)r (x)

)(j)

(x)(1 − x)j−1(1 + x)j−1dx

=n∑

j=0

c(α,−1)j (n, k)

m!(m+ α+ j − 1)!(m− j)!(m+ α− 1)!

δr,m (6.8)

= (m(m+ α) + k)nδr,m.

Comparing (6.6) and (6.8) completes the proof of the lemma. �

302 A. Bruder and L. L. Littlejohn Results. Math.

Theorem 8. For k > 0, let A(α,−1) be the Jacobi self-adjoint operator inL2

α,−1(−1, 1), defined in (4.2) and (4.3), having the sequence of Jacobi polyno-

mials {P (α,−1)m }∞

m=1 as eigenfunctions. For each n ∈ N, let

V (α,−1)n =

{f : (−1, 1) → C | f∈AC(n−1)

loc ; f (j) ∈ L2α+j,j−1(−1, 1), j=0, . . . , n

}

and, for f, g ∈ V(α,−1)n ,

(f, g)(α,−1)n =

n∑j=0

c(α,−1)j (n, k)

1∫

−1

f (j)(t)g(j)(t)(1 − t)α+j(1 + t)j−1dt.

Then W(α,−1)n =

(V

(α,−1)n , (·, ·)(α,−1)

n

)is the nth left-definite space associ-

ated with the pair(L2

α,−1(−1, 1), A(α,−1)). Moreover, the Jacobi polynomials

{P (α,−1)m }∞

m=1 form a complete orthogonal set in each W (α,−1)n , and they satisfy

the orthogonality relation(P (α,−1)

m , P(α,−1)l

)n

= (m(m− 1) + k)nδm,l.

Furthermore, define

B(α,−1)n := D

(B(α,−1)

n

)⊂ W (α,−1)

n → W (α,−1)n

by

B(α,−1)n f := lα,−1 [f ]

(f ∈ D

(B(α,−1)

n

):= V

(α,−1)n+2

).

Then B(α,−1)n is the nth left-definite operator associated with (L2

α,−1(−1, 1),

A(α,−1)). Lastly, the spectrum of B(α,−1)n is given by

σ(B(α,−1)

n

)= {m(m− 1) + k | m ∈ N0} = σ{A(α,−1)},

with the Jacobi polynomials {P (α,−1)m }∞

m=1 forming a complete set of eigen-functions of each B(α,−1)

n .

Proof. Let n ∈ N. We need to show that W (α,−1)n satisfies the five properties

in Definition 1 in Sect. 5. (i) W (α,−1)n is a Hilbert space (see Theorem 6 and

Ref. [5]). (ii) We need to show

D((A(α,−1)

)n)⊂ W (α,−1)

n ⊂ L2α,−1(−1, 1).

Let f ∈ D((A(α,−1)

)n). Since the Jacobi polynomials {P (α,−1)

m }∞m=1 form a

complete orthonormal set in L2α,−1(−1, 1), we see that

pj → f in L2α,−1(−1, 1) as j → ∞ (6.9)

Vol. 61 (2012) Nonclassical Jacobi Polynomials and Sobolev Orthogonality 303

where

pj(t) :=j∑

m=0

c(α,−1)m P (α,−1)

m (t) (t ∈ (−1, 1)) ,

and, for m ∈ N0,

c(α,−1)m :=

(f, P (α,−1)

m

)α,−1

=

1∫

−1

f(t)P (α,−1)m (t) (1 − t)α (1 + t)−1dt.

Since(A(α,−1)

)nf ∈ L2

α,−1(−1, 1), we see that as j → ∞,

j∑m=0

c (α,−1)m P (α,−1)

m →(A(α,−1)

)n

f in L2α,−1(−1, 1)

where

c(α,−1)m :=

((A(α,−1)

)n

f, P (α,−1)m

)−1,−1

=(f,(A(α,−1)

)n

P (α,−1)m

)α,−1

= (m(m+ α) + k)n(f, P (α,−1)

m

)α,−1

= (m(m+ α) + k)nc(α,−1)m ;

consequently, (A(α,−1)

)n

pj →(A(α,−1)

)n

f

in L2α,−1(−1, 1) as j → ∞. Moreover, by Lemma 8,

(‖pj − pr‖(α,−1)

n

)2

=((A(α,−1)

)n

[pj − pr] , pj − pr

)α,−1

→ 0 as j, r → ∞so {pj}∞

j=0 is Cauchy inW (α,−1)n . SinceW (α,−1)

n is a Hilbert space (Theorem 6),there exists

g ∈ W (α,−1)n ⊂ L2

α,−1(−1, 1)

such that

pj → g in W (α,−1)n as j → ∞.

Furthermore, since

(f, f)(α,−1)n =

n∑j=0

c(α,−1)j (n, k)

∥∥∥f (j)∥∥∥2

j+α,j−1

≥ c(α,−1)0 (n, k)

∥∥∥f (j)∥∥∥2

α,−1= kn (f, f)α,−1 ,

we see that

‖pj − g‖α,−1 ≤ k−n/2 ‖pj − g‖(α,−1)n ,

304 A. Bruder and L. L. Littlejohn Results. Math.

and hence,

pj → g in L2α,−1(−1, 1). (6.10)

Comparing (6.9) and (6.10), f = g ∈ W(α,−1)n .

(iii) We need to show: D((A(α,−1))n) is dense in W (α,−1)n . Since the set of poly-

nomials is contained in D((A(α,−1))n) and is dense in W (α,−1)n (by Theorem 7),

D((A(α,−1))n) is dense in W(α,−1)n . Furthermore, from Theorem 7, the Jacobi

polynomials {P (α,−1)m }∞

m=1 form a complete orthonormal set in W (α,−1)n .

(iv) We need to show that (f, f)(α,−1)n ≥ kn(f, f)α,−1 for f ∈ V

(α,−1)n . This is

clear from the definition of (·, ·)(α,−1)n .

(v) We need to show: (f, g)(α,−1)n = ((A(α,−1))nf, g)α,−1 for f ∈ D((A(α,−1))n)

and g ∈ V(α,−1)n . This is true for any f, g ∈ P by Lemma 8. Let f ∈

D((A(α,−1))n) ⊂ W(α,−1)n , g ∈ W

(α,−1)n . Since the set of polynomials P is

dense in both W(α,−1)n and L2

α,−1(−1, 1), and since convergence in the space

W(α,−1)n implies convergence in L2

α,−1(−1, 1) (by (iv)), there exist sequences{pj}∞

j=0 and {qj}∞j=0 such that

pj → f in W (α,−1)n as j → ∞(

A(α,−1))n

pj →(A(α,−1)

)n

f

in L2α,−1(−1, 1) as j → ∞ and

qj → g

in W (α,−1)n and L2

α,−1(−1, 1) as j → ∞. Hence, by Lemma 8,((A(α,−1)

)n

f, g)

−1,−1= lim

j→∞

((A(α,−1)

)n

pj , qj

)α,−1

= limj→∞

(pj , qj)n

= (f, f)(α,−1)n .

The results listed in the theorem on B(α,−1)n and the spectrum of B(α,−1)

n followimmediately from the general left-definite theory. �

7. The Sobolev Orthogonality of the Jacobi Polynomials

If we renormalize the Jacobi polynomials as follows:

P(α,−1)0 (x) := 1, P

(α,−1)1 (x) :=

(α+ 22α+2

)1/2

(x+ 1),

and, for n ≥ 2,

Vol. 61 (2012) Nonclassical Jacobi Polynomials and Sobolev Orthogonality 305

P (α,−1)n (x) :=

(2n+ α)12

2n+α/2 (n+ α)

n∑j=0

(n+ α

n− j

)(n− 1j

)(x− 1

2

)j (x+ 1

2

)n−j

,

we obtain the following theorem; a proof can be found in [12]. A key result inestablishing this result is in the following important identity

P (α,−1)n (x) =

(n+ α)!(n− 1)!2n!(n+ α− 1)!

(x+ 1)P (α,1)n−1 (x) (n ≥ 2)

whose proof can easily be checked.

Theorem 9. The Jacobi polynomials {P (α,−1)n (x)}∞

n=0 are orthonormal withrespect to the Sobolev inner product

φ (f, g) := f(−1)g(−1) +

1∫

−1

(1 − x)α+1f ′(x)g′(x)dx,

that is,

φ(P (α,−1)

n , P (α,−1)m

)= δnm (n,m ∈ N0).

8. Jacobi Polynomials and a Self-Adjoint Operator in a SobolevSpace

Definition 3. Define

Wα :={f : [−1, 1) → C | f ∈ AC [−1, 1) ; f ′ ∈ L2

α+1,0(−1, 1)}

φ(f, g) := f(−1)g(−1) +

1∫

−1

f ′(x)g′(x)(1 − x)α+1dx (f, g ∈ Wα).

Write ‖f‖2φ = φ(f, f) for f ∈ Wα.

Theorem 10. (Wα, φ(·, ·)) is a Hilbert space.

Proof. Let {fn} ⊂ W1 be a Cauchy sequence. Hence

‖fn − fm‖2φ = |fn(−1) − fm(−1)|2 +

1∫

−1

|f ′n(x) − f ′

m(x)|2 (1 − x)α+1dx

→ 0 as n,m → ∞.

In particular, since1∫

−1

|f ′n(x) − f ′

m(x)|2 (1 − x)α+1dx ≤ ‖fn − fm‖2φ ,

we see that {f ′n} is Cauchy in L2((−1, 1); (1 − x)α+1).

306 A. Bruder and L. L. Littlejohn Results. Math.

Since L2((−1, 1); (1 − x)α+1) is complete, there exists g ∈ L2α+1,0(−1, 1)

such that

f ′n → g as n → ∞ in L2

α+1,0(−1, 1). (8.1)

Also, since

|fn(−1) − fm(−1)|2 ≤ ‖fn − fm‖2φ

we see that the sequence {fn(−1)} is Cauchy in C and, hence, there existsA ∈ C such that

fn(−1) → A. (8.2)

Furthermore, since fn ∈ AC[−1, 1)(n ∈ N), we see that

1∫

−1

f ′n(t)(1 − t)α+1dt →

1∫

−1

g(t)(1 − t)α+1dt,

Since g ∈ AC[−1, 1), we may define f : [−1, 1) → C by

f(x) = A+

x∫

−1

g(t)dt.

It is clear that f ∈ AC[−1, 1) and f ′(x) = g(x) ∈ L2α+1,0(−1, 1) for a.e.

x ∈ [−1, 1), so f ∈ Wα. Furthermore, f(−1) = A. Now

‖fn − f‖2φ = |fn(−1) − f(−1)|2 +

1∫

−1

|f ′n(t) − f ′(t)|2 (1 − t)α+1dt

= |fn(−1) −A|2 +

1∫

−1

|f ′n(t) − g(t)|2 (1 − t)α+1dt → 0

as n → ∞ by (8.1) and (8.2). Thus, (Wα, φ(·, ·)) is complete. �

Theorem 11. Let Wα and φ(·, ·) be as before, and

Wα,1 := {f ∈ Wα | f(−1) = 0}Wα,2 := {f ∈ Wα | f ′(x) = 0} .

Then Wα,1 and Wα,2 are closed, orthogonal subspaces of Wα and Wα = Wα,1⊕Wα,2.

Proof. Since Wα,2 is one-dimensional, it is a closed subspace of Wα. Theorthogonal complement of Wα,2 is given by

W⊥α,2 := {f ∈ Wα | φ(f, g) = 0 (g ∈ Wα,2)}.

Vol. 61 (2012) Nonclassical Jacobi Polynomials and Sobolev Orthogonality 307

To see that Wα,1 ⊂ W⊥α,2, let f ∈ Wα,1, g ∈ Wα,2 and consider

φ(f, g) = f(−1)g(−1) +

1∫

−1

f ′(x)g′(x)(1 − x)α+1dx = 0.

The first summand vanishes because f ∈ Wα,1, and the integral is 0 becauseg ∈ Wα,2. Now let f ∈ Wα. We need to find f1 ∈ Wα,1 and f2 ∈ Wα,2 such thatf = f1 + f2. Let f1(x) = f(x) − f(−1) and f2(x) = f(−1); clearly, fi ∈ Wα,i

for i = 1, 2. �

The next result shows that, surprisingly, the space Wα,1 is precisely thefirst left-definite vector space V (α,−1)

1 . It is this theorem that allows us to con-struct a self-adjoint operator Tα in Wα having the entire sequence of Jacobipolynomials {P (α,−1)

n }∞n=0 as eigenfunctions. For n ≥ 2,

P (α,−1)n (x) =

(n+ α)!(n− 1)!2n!(n+ α− 1)!

(x+ 1)P (α,1)n−1 (x).

Theorem 12. Wα,1 = V(α,−1)1 .

Proof. Notice that functions in Wα,1 are defined on [−1, 1) while functions inV

(α,−1)1 are defined on (−1, 1) so the connection between these two spaces is

not immediately obvious.

V(α,−1)1 ⊆ Wα,1: Let f ∈ V

(α,−1)1 . Since V (α,−1)

1 ⊂ Δα,−1 we see, by Lemma 2,that f ∈ AC[−1, 1) and f(−1) = 0 so f ∈ Wα,1.

Wα,1 ⊆ V(α,−1)1 : Let f ∈ Wα,1. From the definition of both spaces, it clearly

suffices to show that f ∈ L2α,−1(−1, 1). For −1 < x < 0,

(1 − x)α/2(1 + x)−1/2

x∫

−1

f ′(t)dt = (1 − x)α/2(1 + x)−1/2f(x)

since f(−1) = 0. We use Theorem 1 on (−1, 0) with ψ(x) = (1 − x)α/2(1 +x)−1/2 and ϕ(x) = 1. Clearly, ψ and ϕ satisfy the conditions of Theorem 1.Moreover

x∫

−1

dt

0∫

x

(1 − t)α(1 + t)−1dt ≤ c

x∫

−1

dt

0∫

x

dt

1 + t= −c(x+ 1) ln(1 + x),

and, since this is a bounded function on (−1, 0), we see that f ∈ L2α,−1(−1, 0).

A similar argument shows that f ∈ L2α,−1(0, 1) and this completes the proof.

Theorem 13. The inner products φ(·, ·) and (·, ·)(α,−1)1 are equivalent on Wα,1 =

V(α,−1)1 .

308 A. Bruder and L. L. Littlejohn Results. Math.

Proof. First of all, (Wα,1, φ(·, ·)) is a Hilbert space, and, by definition,(V (α,−1)

1 , (·, ·)(α,−1)1 ) is a Hilbert space. Let f ∈ Wα,1 = V

(α,−1)1 . Then

‖f‖2φ =

1∫

−1

|f ′|2 (1 − x)α+1dx

≤1∫

−1

[|f ′|2 (1 − x)α+1 + k(1 − x)α(1 + x)−1 |f |2

]dx = (‖f‖1)

2.

By the Open Mapping Theorem, these inner products are equivalent. �

We now construct a self-adjoint operator Tα,1 in the space Wα,1, gener-ated by the Jacobi expression lα,−1, having the sequence of Jacobi polynomials{P (α,−1)

n }∞n=1 as eigenfunctions. Recall that by Theorem 12, V (α,−1)

1 = Wα,1.We also know that the operator

B(α,−1)1 : D(B(α,−1)

1 ) := V(α,−1)3 ⊂ V

(α,−1)1 → V

(α,−1)1

namely, the first left-definite operator associated with (L2α,−1(−1, 1), A(α,−1)),

is self-adjoint and given specifically by

B(α,−1)1 [f ](x) = lα,−1[f ](x) (a.e. x ∈ (−1, 1))

f ∈ D(B(α,−1)1 ) = V

(α,−1)3 ,

where

f ∈ D(B

(α,−1)1

)= V

(α,−1)3 = {f : (−1, 1) → C | f, f ′, f ′′ ∈ ACloc(−1, 1);

(1 − x)(α+3)/2(1 + x)f ′′′, (1 − x)(α+2)/2(1 + x)1/2f ′′, (1 − x)(α+1)/2f ′,

(1 − x)α/2(1 + x)−1/2f ∈ L2(−1, 1)}.More specifically, B(α,−1)

1 is self-adjoint with respect to the first left-definiteinner product (·, ·)(α,−1)

1 which is equivalent to the inner product φ(·, ·). Weshall prove that the operator Tα,1 : D(Tα,1) ⊂ Wα,1 → Wα,1 given by

Tα,1f = B(α,−1)1 f = lα,−1[f ]

f ∈ D(Tα,1) := V(α,−1)3

is self-adjoint in (Wα,1, φ(·, ·)).Lemma 9. Tα,1 in (Wα,1, φ(·, ·)) is densely defined.

Proof. The Jacobi polynomials {P (α,−1)n }∞

n=1 are eigenfunctions of Tα,1 andthey are complete in V (α,−1)

3 . �

Theorem 14. Tα,1 is symmetric in (Wα,1, φ(·, ·)).

Vol. 61 (2012) Nonclassical Jacobi Polynomials and Sobolev Orthogonality 309

Proof. From the previous lemma, it suffices to show that Tα,1 is Hermitian. Letf, g ∈ D(Tα,1) = V

(α,−1)3 . Since V (α,−1)

3 ⊂ V(α,−1)1 and Tα,1f, Tα,1g ∈ V

(α,−1)1 ,

we know that

f(−1) = g(−1) = 0 = Tα,1f(−1) = Tα,1g(−1).

Integration by parts shows that φ(Tα,1f, g) = φ(f, Tα,1g). �

Theorem 15. The operator Tα,1 has the following properties:

(i) Tα,1 is self-adjoint in (Wα,1, φ(·, ·)).(ii) σ(Tα,1) = {n(n+ α) + k | n ≥ 2}.(iii) {P (α,−1)

n }n≥1 is a complete orthonormal set of eigenfunctions of Tα,1 inthe space (Wα,1, φ(·, ·)).

(iv) Tα,1 is bounded below by kI in (Wα,1, φ(·, ·)).

Proof. For (iii): We know that {P (α,−1)n }n≥0 is a complete orthonormal set

in (Wα, φ(·, ·)) and we know that Wα = Wα,1 ⊕Wα,2. Also, we have Wα,2 =span{P (α,−1)

0 } and so Wα,1 = W⊥α,2 = span{P (α,−1)

n }n≥1. We next prove that

Tα,1 is closed in (Wα,1, φ(·, ·)). Take a sequence {fn} ⊆ D(Tα,1) = V(α,−1)3

such that

fn → f in (Wα,1, φ(·, ·))T1fn → g in (Wα,1, φ(·, ·)) .

We show that f ∈ D(Tα,1) and Tα,1f = g. We know that B(α,−1)1 is self-adjoint

and hence closed in (Wα,1, (·, ·)(α,−1)1 ), and we know, since φ(·, ·) and (·, ·)(α,−1)

1

are equivalent, there exist constants c1 and c2 such that

c1 ‖f‖φ ≤ ‖f‖1 ≤ c2 ‖f‖φ (f ∈ Wα,1 = V1).

Hence,

‖fn − f‖1 ≤ c2 ‖fn − f‖φ → 0;

that is,

fn → f in(Wα,1, (·, ·)(α,−1)

1

)

and

‖Tα,1fn − g‖1 ≤ c2 ‖Tα,1fn − g‖φ → 0;

hence

Tα,1fn → g in(Wα,1, (·, ·)(α,−1)

1

)

310 A. Bruder and L. L. Littlejohn Results. Math.

and since Tα,1 is closed in (Wα,1, (·, ·)(α,−1)1 ), we see that f ∈ D(Tα,1) and

Tα,1f = g. Also, we know that, for n ≥ 2,(Tα,1P

(α,−1)n

)(x) = lα,−1[P (α,−1)

n ](x)

= (n(n+ α) + k)P (α,−1)n (x).

This implies

{n(n+ α) + k | n ≥ 2} ⊆ σ(Tα,1).

Since {P (α,−1)n }n≥1 is complete and λn := n(n+ α) + k → ∞, we know that

σ(Tα,1) = {n(n+ α) + k | n ≥ 2}which proves (ii) and (iii). To summarize: Tα,1 is a closed, symmetric oper-ator with a complete set of eigenfunctions. From [14], Tα,1 is self-adjoint.This proves (i). To prove (iv), let f ∈ D(Tα,1). Then, since Tα,1 : V (α,−1)

3 ⊂V

(α,−1)1 → V

(α,−1)1 ,

φ (Tα,1f, f) = (Tα,1f) (−1)f(−1) +

1∫

−1

(Tα,1f)′ (x)f′(x)(1 − x)α+1dx

=

1∫

−1

(Tα,1f)′ (x)f′(x)(1 − x)α+1dx

=

1∫

−1

⎡⎢⎣∣∣∣((1 − x)α+1f ′(x)

)′∣∣∣2(1 − x)α(1 + x)−1

+ k |f ′(x)|2 (1 − x)α+1

⎤⎥⎦dx

≥ k

1∫

−1

|f ′(x)|2 (1 − x)α+1dx = kφ (f, f) .

Next, we define the operator Tα,2 : D(Tα,2) ⊂ Wα,2 → Wα,2 by

(Tα,2f)(x) = lα,−1[f ](x)D(Tα,2) := Wα,2.

It is straightforward to check that Tα,2 is symmetric in Wα,2 and since thedomain of Tα,2 is the entire space, it follows that Tα,2 is self-adjoint.

We now construct the self-adjoint operator Tα in (Wα, φ(·, ·)) that isgenerated by the Jacobi differential expression lα,−1[.], having the entire setof Jacobi polynomials {P (α,−1)

n }n≥0 as eigenfunctions and having spectrumσ(Tα) = {n(n+ α) + k | n ∈ N0}. For f ∈ Wα, write

f = f1 + f2

Vol. 61 (2012) Nonclassical Jacobi Polynomials and Sobolev Orthogonality 311

where fi ∈ Wα,i, (i = 1, 2). Define Tα : D(Tα) ⊂ Wα → Wα by

Tαf = Tα,1f1 + Tα,2f2 = lα,−1[f1] + lα,−1[f2] = lα,−1[f ],D(Tα) = D(Tα,1) ⊕ D(Tα,2).

A proof that operators of this form are self-adjoint can be found in [10, Theo-rem 11.1]. Furthermore, since we know explicitly the domains of Tα,1 and Tα,2,we can specifically determine the domain D(Tα) of Tα.

Theorem 16. Tα is self-adjoint in (Wα, φ(·, ·)) and

D(Tα) = {f : [−1, 1) → C | f ∈ AC[−1, 1); f ′, f ′′ ∈ ACloc(−1, 1);

(1 − x)(α+3)/2(1 + x)f ′′′, (1 − x)(α+2)/2(1 + x)1/2f ′′,

(1 − x)(α+1)/2f ′ ∈ L2(−1, 1)}= {f : [−1, 1) → C | f ∈ AC[−1, 1); f ′, f ′′ ∈ ACloc(−1, 1);

(1 − x)(α+3)/2(1 + x)f ′′′ ∈ L2(−1, 1)}.Furthermore, σ(Tα) = {n(n+α) + k | n ∈ N0} and Tα is bounded below by kIin (Wα, φ(·, ·)).For the following theorem let us recall the definitions of the first and thirdleft-definite spaces:

V(α,−1)1 = {f : (−1, 1) → C | f ∈ ACloc(−1, 1);

(1 − x)α/2(1 + x)−1/2f, (1 − x)(α+1)/2f ′ ∈ L2(−1, 1)}= {f : [−1, 1) → C | f ∈ AC [−1, 1) ;

(1 − x)(α+1)/2f ′ ∈ L2(−1, 1); f(−1) = 0}= Wα,1

V(α,−1)3 = D(Tα,1) = {f : (−1, 1) → C | f, f ′, f ′′ ∈ ACloc(−1, 1);

(1 − x)(α+3)/2(1 + x)f ′′′, (1 − x)(α+2)/2(1 + x)1/2f ′′,

(1 − x)(α+1)/2f ′, (1 − x)α/2(1 + x)−1/2f ∈ L2(−1, 1)}= {f ∈ V1 | f ′, f ′′ ∈ ACloc(−1, 1); (1 − x)(α+3)/2(1 + x)f ′′′

(1 − x)(α+2)/2(1 + x)1/2f ′′ ∈ L2(−1, 1)}= {f : [−1, 1) → C | f ∈ AC[−1, 1); f ′, f ′′ ∈ ACloc(−1, 1);

f(−1) = 0; (1 − x)(α+3)/2(1 + x)f ′′′, (1 − x)(α+2)/2(1 + x)1/2f ′′,

(1 − x)(α+1)/2f ′ ∈ L2(−1, 1)}.Theorem 17. Let

D := {f : [−1, 1) → C | f ∈ AC[−1, 1); f ′, f ′′ ∈ ACloc(−1, 1);

(1 − x)(α+3)/2(1 + x)f ′′′, (1 − x)(α+2)/2(1 + x)1/2f ′′,

(1 − x)(α+1)/2f ′ ∈ L2(−1, 1)}.Then D(Tα) = D.

312 A. Bruder and L. L. Littlejohn Results. Math.

Proof. We first show D(Tα) ⊆ D. Let f ∈ D(Tα) = D(Tα,1) ⊕ D(Tα,2). Write

f = f1 + f2

where f1 ∈ D(Tα,1) = V(α,−1)3 ⊆ D, f2 ∈ D(Tα,2) ⊆ D. Then f ∈ D. To show

that D ⊆ D(Tα), let f ∈ D. Write

f(x) = f1(x) + f2(x)

where

f1(x) := f(x) + f(−1)f2(x) := −f(−1).

Then f1 ∈ D and f1(−1) = 0; that is, f1 ∈ V(α,−1)3 = D(Tα,1). Also, f ′′

2 (x) = 0so f2 ∈ D(Tα,2). Together, f ∈ D(Tα). �

References

[1] Akhiezer, N.I., Glazman, I.M.: Theory of Linear Operators in Hilbert Space.Dover Publications, New York (1993)

[2] Alfaro, M., Alvarez de Morales, M., Rezola, M.L.: Orthogonality of the Jacobipolynomials with negative integer parameters. J. Comput. Appl. Math. 145, 379–386 (2002)

[3] Alfaro, M., Rezola, M.L., Perez, T.E., Pinar, M.A.: Sobolev orthogonal polyno-mials: the discrete-continuous case. Methods Appl. Anal. 6, 593–616 (1999)

[4] Alvarezde Morales, M., Perez, T.E., Pinar, M.A.: Sobolev orthogonality for the

Gegenbauer polynomials {C(−N+1/2)n }n≥0. J. Comput. Appl. Math. 100, 111–120

(1998)

[5] Bruder, A.: Applied Left-Definite Theory: Jacobi Polynomials, Their SobolevOrthogonality, and Self-Adjoint Operators. Ph.D. Thesis. Baylor University,Waco, TX (2009)

[6] Chisholm, R.S., Everitt, W.N., Littlejohn, L.L.: An integral operator inequalitywith applications. J. Inequal. Appl. 3, 245–266 (1999)

[7] Chihara, T.: An Introduction to Orthogonal Polynomials. Gordon and Breach,New York (1978)

[8] Dunford, N., Schwartz, J.T.: Linear Operators, Part II. Wiley Interscience,New York (1963)

[9] Everitt, W.N., Kwon, K.H., Littlejohn, L.L., Wellman, R., Yoon, G.J.: Jacobi–Stirling numbers, Jacobi polynomials, and the left-definite analysis of the clas-sical Jacobi differential expression. J. Comput. Appl. Math. 208, 29–56 (2007)

[10] Everitt, W.N., Littlejohn, L.L., Wellman, R.: The Sobolev orthogonality andspectral analysis of the Laguerre polynomials {L−k

n } for positive integers k.J. Comput. Appl. Math. 171, 199–234 (2004)

[11] Kwon, K.H., Lee, D.W., Littlejohn, L.L.: Sobolev orthogonal polynomials andsecond order differential equations II. Bull. Korean Math. Soc. 33(1), 135–170(1996)

Vol. 61 (2012) Nonclassical Jacobi Polynomials and Sobolev Orthogonality 313

[12] Kwon, K.H., Littlejohn, L.L.: Sobolev orthogonal polynomials and second-orderdifferential equations. Rocky Mountain J. Math. 28(2), 547–594 (1998)

[13] Littlejohn, L.L., Wellman, R.: A general left-definite theory for certainself-adjoint operators with applications to differential equations. J. Differ.Equ. 181, 280–339 (2002)

[14] Naimark, M.A.: Linear Differential Operators II. Frederick Ungar PublishingCo., New York (1968)

[15] Rainville, E.D.: Special Functions. Chelsea Publishing Co., New York (1960)

[16] Szego, G.: Orthogonal polynomials, vol. 23. American Mathematical SocietyColloquium Publications, Providence, Rhode Island (1978)

Andrea BruderDepartment of Mathematics and Computer ScienceTutt Science CenterColorado College14 E. Cache la Poudre StreetColorado Springs, CO 80903USAe-mail: [email protected]

L. L. LittlejohnDepartment of MathematicsBaylor UniversityOne Bear Place #97328Waco, TX 76798-7328USA

Received: November 15, 2010.

Revised: January 18, 2011.

Accepted: January 25, 2011.