Simultaneous Preservation of Orthogonality of Polynomials by Linear Operators Arising from Dilation of Orthogonal Polynomial Systems
1Frank Filbir,2Roland Girgensohn,2Anu Saxena,3Ajit Iqbal Singh4 and Ryszard Szwarc5
For an orthogonal polynomial system p pnn2IN0and a sequence d dnn2IN0 of nonzero numbers, let Sp;dbe the linear operator de®ned on the linear space of all polynomials via Sp;dpn dnpnfor all n 2 IN0.We investigate conditions on p and d under which Sp;dcan simultaneously preserve the orthogonality of dierent polynomial systems. As an application, we get that for p Ln, a generalized Laguerre polynomial system, no d can simultaneously preserve the orthogonality of two additional Laguerre systems, Ltn 1 and Ltn 2, where t1; t26 0 and t16 t2. On the other hand, for p Tn, the Chebyshev polynomial system and d 1n, Sp;d simultaneously preserves the orthogonality of uncountably many kernel polynomial systems associated with p. We study many other ex- amples of this type.
KEY WORDS: Orthogonal polynomials; dilation map; kernel polynomials;
Jacobi polynomials; Laguerre polynomials.
1. INTRODUCTION
We start with some notation. Let IN be the set of natural numbers and IN0 IN [ f0g. For a set X; ] X will denote the cardinality of X. Let P1be the linear space of all polynomials and let P be the set of sequences p pnn2IN0 in P1 with p0 1 and deg pn n for all n 2 IN0. We shall
0Dedicated to Professor P. L. Butzer on the occasion of his 70th birthday.
0GSFÐNational Research Center for Environment and Health, Institute of Biomathematics and Biometry, IngolstaÈdter Landstrae 1, D-85764 Neuherberg, Germany.
0Department of Mathematics, Jesus and Mary College, Chanakyapuri, New Delhi 110021, India.
0Department of Mathematics, University of Delhi South Campus, Benito Juarez Road, New Delhi 110021, India.
0Institute of Mathematics, Wrocaw University, pl. Grunwaldzki 2/4, 50-384 Wrocaw, Poland.
177
1521-1398/00/0400-0177$18.00/0 ß 2000 Plenum Publishing Corporation
often write pnn2IN0 simply as pn or even p. S will usually stand for a degree preserving linear operator on P1 to itself with S 1 1. Ortho- gonality will always be with respect to some quasi-de®nite moment func- tional, whereas positive orthogonality will be with respect to some positive de®nite moment functional and in the context of real polynomials only (see [5], Theorem I.3.3). The word orthonormality will be used only in the context of positive orthogonality, where, in addition, the positive moment functional L is normalized so that L 1 1. For any such pair p; L and any polynomial q, L jqj21=2will be denoted by kqkL or kqkp; the sux L or p will be suppressed if no confusion arises. Orthogonality with respect to a symmetric moment functional (see [5], De®nition I.4.1) will be termed symmetric orthogonal. We refer to [2], [5], and [13] for the basics on orthogonal polynomials.
Operators preserving orthogonality of polynomials have been studied by various authors, among them [1], [8], and [10]. We start by quoting two simple recent results from [8] which, to a certain extent, provided the motivation for the present paper. For 2 IR n IN, let Ln be the generalized Laguerre polynomial system, given by
Ln x Xn
k 0
1k n n k
xk
k!; x 2 IR; n 2 IN0; where
t 0 !
1; t
k !
t t 1 t k 1
k!
t k 1k
1k ; t 2 IR; k 2 IN:
From Ref. [8]: Corollary 1. If S preserves the orthogonality of Ln for 0 1; 0; 0 1; 0 2 (with some 0 and, moreover, k S L1 k2 1 for at least two of f0; 0 1; 0 2g, then for some a;
b 2 IR; Sp x p ax b for x 2 IR and p 2 P1.
From Ref. [8]: Theorem 3. For any 2 IR, the linear operator S de®ned on P1via SLn x Ln x ; x 2 IR; n 2 IN, is independent of . Further, S simultaneously preserves the orthogonality of Ln for all . Let Pm; Po; Pmo; and Po be the subsets of P consisting of monic, orthogonal, monic orthogonal, and positive orthogonal systems, resp- ectively. Let D be the set of nonconstant sequences d dnn2IN0of non-zero numbers with d0 1.
In Section 2 we shall de®ne a dilation-type linear operator Sp;don P1to itself associated to a p in Poand d in D. In fact, Sp;dis the linear operator on P1 to itself, given by Sp;dpn dnpn for all n 2 IN0. We ®rst observe that if d 2 D is the sequence of eigenvalues of a linear dierential operator F such that the corresponding sequence of eigenfunctions consists of an orthogonal polynomial system p, then Sp;d coincides with F. (For these topics, one can consult [7] and the related expository articles [4] and [11], or the books [3], Section 3.5, and [5], Section V.2 and Section V.3.) This is illustrated by the following simple examples.
(a) Let ; > 1 and p Pn ;, the Jacobi polynomial system. Let dn n n 1 1 for n 2 IN0. Then Sp;d is the dier- ential operator F determined by
Ff x 1 x2 f00 x 2 xf0 x f x:
(b) For p Hn, the Hermite polynomial system, and d given by dn 2n 1 for n 2 IN0, Sp;d coincides with the dierential operator F de®ned by
Ff x f00 x 2xf0 x f x:
(c) Let 2 IR n IN and p Ln, the generalized Laguerre system.
If dn 2n 1 for n 2 IN0, then Sp;d coincides with the dierential operator F determined by
Ff x 2xf00 x 2 1 xf0 x f x:
For a ®xed p; d 2 Po D we shall look for those subsets S of P, which are mapped into Po by Sp;d, i.e., where orthogonality is simulta- neously induced by Sp;d. Also, for a class S contained in Poand a p in Po, we shall study the question of existence of a d in D such that Sp;d preserves the orthogonality of S.
To motivate, we may note the following result, which follows readily from the connection coecients of the Jacobi system (cf. [2], 7.32±7.34).
Theorem 1.1. Let ; > 1 and p be the Jacobi polynomial system Pn ;. Let
dn 1n 1n
1n 1n Qn
j 1 j
Qn
j 1 j
and d0n 1=dn for n 2 IN0.
(a) Then, for t > 0 and n 2 IN0,
Sp;dPn ;t dnPn t; and
Sp;d0Pn t; d0nPn ;t:
Therefore Sp;d simultaneously preserves the orthogonality of fPn ;t; t > 0g and Sp;d0 that of fPn t;; t > 0 g.
(b) For , we have dn 1n; n 2 IN0, and d d0. In this case, Sp;dPn ; 1nPn ; for all > 1, and thus Sp;d S P ;
n ;d
for > 1.
This is true for any sequence d satisfying d2n 1 and d2n1 d1 for all n 2 IN0.
Further, for such d's, it is enough to consider the case p P 1=2; 1=2
n Tn instead of the whole class fPn ; : > 1g of symmetric Jacobi polynomials.
In Section 3, we shall de®ne and study the notion of a q qn in P to be analytic with respect to a p pn in P. Let Papbe the class of those q's in P, which are analytic with respect to p. It turns out that for a p in Pmoand q in Pap; q 2 Po if, and only if, q is a kernel polynomial K p; y associated with p where y 62 Zp, Zpbeing the set of zeros of pn's. An equivalent form for a p in Pmois obtained using the general method developed in Section 2. This enables us to have nice transparent conditions on d for Sp;d to simulta- neously preserve the orthogonality of various subsets of Pap\ Po. For some special d's we can obtain many dierent subsets of Pap, whose orthogonality is simultaneously preserved by Sp;d.
In the last section, we apply the results of the second and third sections to the class of generalized Laguerre polynomial systems and obtain a striking contrast to the situation for the Chebyshev polynomials!
2. DILATION-TYPE OPERATORS AND ORTHOGONALITY 2.1. Discussion and De®nition
Let p pn be an orthogonal polynomial system. Then the set P of all polynomial systems is in one±one correspondence with the set A of (in®nite) matrices A aj;kj;k2IN0 with aj;k 0 for k > j; a0;0 1 and aj;j6 0 for
all j via q A p, i.e., for all n 2 IN0; qnPn
k 0 an;kpk.
(a) Clearly, in the case p has real coecients, q has real coecients if, and only if, A is real.
(b) In the case p 2 Pm, i.e., p is monic and A 2 A, q A p 2 Pm if, and only if, aj;j 1 for all j. Such matrices will be called monic.
The set of monic matrices will be denoted by Am.
(c) For a d dn 2 D let Sp;d be the linear operator on P1 to itself given by Sp;dpn dnpnfor all n 2 IN0. The map induced by Sp;don P will be called the p; d-dilation and denoted by Sp;d only.
(d) Clearly for an A 2 A and q A p, Sp;dq corresponds to the matrix Ad aj;kdk A D; D being the diagonal matrix given by the diagonal d. In the case p is monic, the associated monic polynomial system Q has the matrix aj;k dk= aj;j dj
(e) For a ®xed p, the sequence d and the operator S. p;ddetermine each other. However, for a ®xed d; Sp;d Sq;d can hold without p q, as is seen from Theorem 1.1. In fact, for q A p and d; d02 D, Sp;d Sq;d0 if, and only if, aj;k d0j dk 0 for all j; k 2 IN0. As a consequence, if d has all terms distinct and p; q are monic, then p q()Sp;d Sq;d. On the other hand, if d has at least two terms coincident then fq monic : Sp;d Sq;dg is in®nite.
(f) Monic orthogonal polynomial systems pn can be characterized via their recurrence relation
xpn x pn1 x bnpn x cnpn 1 x; x 2 IR; n 2 IN0
with recurrence sequences b bn, c cn and cn6 0 for all n 2 IN0 (see again [5]). For notational convenience, we set p 1 q 1 0; aj;k 0 for j < 0 or k < 0, d 1 0, and we take empty sums to be zero and empty products to be 1.
(g) I. M. Sheer [12] de®ned polynomial systems of type zero as follows. The system q qn is of type zero if there are numbers n
with 16 0 such that 1q0n 2qn00 nq nn qn 1 for all n 2 IN. Among other things, he connected systems of type zero to orthogonal systems by dilations. He proved ([12], Theorem 4.1):
Assume that a polynomial system q dnpn is given with d 2 D.
Then q is of type zero if, and only if, there exist numbers a1; . . . ; a4 with
bn a1 a2n and cn n a3 a4n for all n 2 IN0:
Theorem 2.2. Let p pn be a monic orthogonal polynomial system with recurrence sequences b bn and c cn. Let A aj;k be a monic matrix and q qn Ap.
(a) The system q is orthogonal if, and only if, (i) a2;0 a2;1a1;06 c1 b0 b1a1;0 a1;02; (ii) an1;n 1 an1;nan;n 1
6cn bn 1 bnan;n 1 an;n 12
an;n 2; n 2;
(iii) an1;0 an1;nan;0 an1;n 1 an1;nan;n 1an 1;0
b0 bn an;n 1an;0 c1an;1
cn bn 1 bnan;n 1 an;n 12 an;n 2an 1;0; n 2;
(iv) aandn1;j an1;nan;j an1;n 1 an1;nan;n 1an 1;j
an;j 1 bj bn an;n 1an;j cj1an;j1
cn bn 1 bnan;n 1 an;n 12 an;n 2an 1;j;
n 3 ; 1 j n 2:
(b) Suppose p is positive orthogonal and A is real. Then q is positive orthogonal if, and only if, all the conditions in (a) above are satis®ed with 6 in (i) and (ii) strengthened to <.
Proof. (a) By Favard's Theorem ([5], Theorem I.4.4), combined with the fundamental recurrence relation ([5], Theorem I.4.1), q is orthogonal if, and only if, there exist sequences nn2IN0; nn2IN0; 0 arbitrary (and ineective) such that for each n 2 IN; n6 0
xqn x qn1 x nqn x nqn 1 x; x 2 IR; n 2 IN0: Using the recurrence relation for the pn's, this happens if, and only if, for all n 2 IN0,
Xn
j 0
an;j pj1 bjpj cjpj 1
Xn1
j 0
an1;jpj nXn
j 0
an;jpj n Xn 1
j 0
an 1;jpj:
Since the pj's are linearly independent, we can equate their coecients on both sides, and simple computations then give that the conditions stated in
the theorem are necessary and sucient. Along the way we also have that 0 b0 a1;0;
1 b1 a1;0 a2;1; 1 c1 b 0 b1 a1;0
a1;0 a2;0 a2;1a1;0;
and for n 2 ;
n bn an;n 1 an1;n; n cn b n 1 bn an;n 1
an;n 1
an;n 2 an1;n 1 an1;nan;n 1:
(b) We shall again use the fundamental recurrence relation and Favard's Theorem. We have that for each n, bn is real and cn1> 0.
Further, q is positive orthogonal if, and only if, for each n, n is real and n1> 0. This gives us that 6 in (i) and (ii) of (a) have to be replaced by
<. &
Remark 2.3.
(a) The above theorem tells us how to construct all in®nite matrices which preserve orthogonality or positive orthogonality.
Take any sequence s snn2IN0 of numbers and another sequence t tnn2IN0 of numbers satisfying
t06 c1 s0 b0 b1 s0 s1;
tn tn 16 cn1 sn bn bn1 sn sn1; n 1:
Set an1;n sn and an2;n tn for all n. De®ne, for n 2; 0 j n 2, the numbers an1;j recursively as in (a) (iii) and (iv). Finally, take aj;j 1 and aj;k 0 for k > j, and obtain As;t aj;k. Then q As;tp is orthogonal. Also, every monic matrix A having this preservation property has this form for a unique pair s; t.
(b) In case p is positive orthogonal, we have to restrict s and t to be real sequences and take
t0< c1 s0 b0 b1 s0 s1;
tn tn 1< cn1 sn bn bn1 sn sn1; n 1:
(c) On the other hand, if p is symmetric orthogonal, then bn 0 for all n. Thus q is symmetric orthogonal if, and only if, n 0 is satis®ed together with (a) of Theorem 2.2, i.e., the sequence s sn is constant. This simpli®es the condition on t to tn tn 16 cn1 for all n 2 IN, and it simpli®es (a) as follows:
(i) a2;06 c1,
(ii) an1;n 16 an;n 2 cn, n 2,
(iii) an1;0 c1an;1 an1;n 1 an;n 2 cn an 1;0, n 2, and(iv) an1;j an;j 1 cj1an;j1 an1;n 1 an;n 2 cn an 1;j,
n 3, 1 j n 2.
As before, for positive orthogonality, in (i) and (ii), 6 has to be replaced by <.
(d) Given a monic matrix A, the above theorem also gives us conditions on the sequences b and c for the polynomial system q Ap to be orthogonal.
Theorem 2.4. Let A be a monic in®nite matrix, d a sequence of nonzero numbers with d0 1 and p pn a monic orthogonal system with recurrence sequences b and c. Let q A p and r Sp;dq.
(ad) The system r is orthogonal if, and only if, (i) 1
d2 a2;0 a2;1a1;0 6 c1 1
d1 b0 b1a1;0 1 d1a1;0
2
;
(ii) dn 1
dn1 an1;n 1 an1;nan;n 1 6 cndn 1
dn bn 1 bnan;n 1 dn 1 dn an;n 1
2
dn 2
dn an;n 2; n 2, (iii) 1
dn1 an1;0 an1;nan;0 an1;n 1 an1;nan;n 1an 1;0
1 dn
"
b0 bn dn 1 dn an;n 1
!
an;0 d1c1an;1 dn
dn 1 cndn 1
dn bn 1 bnan;n 1 dn 1
dn an;n 1
2
dn 2
dn an;n 2
! an 1;0
#
; n 2;
(iv) 1
dn1 an1;j an1;nan;j an1;n 1 an1;nan;n 1an 1;j
1 dn
"
dj 1
dj an;j 1 bj bn dn 1
dn an;n 1
an;j
dj1
dj cj1an;j1 dn
dn 1cn bn 1 bnan;n 1 dn 1
dn an;n 12
dn 2 dn 1an;n 2
an 1;j
#
; n 3 ; 1 j n 2:
(bd) Suppose p is positive orthogonal. Then q is positive orthogonal if, and only if, dk=djaj;k
is a real matrix and all the conditions in (ad) above are satis®ed with 6 in (i) and (ii) strengthened to <.
Proof. (ad) We ®rst note that r is orthogonal if, and only if, the cor- responding monic polynomial system R Rn 1=d nrn is orthogonal.
The matrix associated with R is eA eaj;k, given by eaj;k dk=djaj;k for j; k 2 IN0. Thus, by Theorem 2.2 above, R is orthogonal if, and only if, eA satis®es the Condition 2.2(a) where a is replaced by ea everywhere. We al- ready bracketed the expressions in 2.2(a) in such a way that (a) transforms to (ad) immediately.
(bd) By 2.1(a), eA has to be real in the ®rst place. The rest follows from (ad) in the same manner as (b) does from (a) in Theorem 2.2. &
Remark 2.5.
(a) The above theorem is a two-edged sword, although not equally sharp! The sharper one says that if a d is given, then we can construct all those A's for which r is orthogonal (respectively, positive orthogonal in case p is so). All we have to do is to choose any sequence sn of numbers and then take another sequence tn satisfying
1
d2t06 c1 1
d1s0 b0 b1 1 d1s0 d1
d2s1
;
and, for n 1 , dn
dn2tn dn 1
dn1tn 16 cn1
dn
dn1sn bn bn1 dn
dn1sn dn1 dn2sn1
:
Then we can set an1;n sn and an2;n tn for all n 2 IN. Next we can de®ne, for n 2 and 0 j n 1 , the numbers an1;j recursively using (iii) and (iv) of (ad). Finally, take aj;j 1 for all j and aj;k 0 for all k > j and obtain As;t aj;k. Then for q As;tp, the system Sp;dq is orthogonal.
In addition, every monic matrix A having this preservation property has this form for a unique s; t.
In case p is positive orthogonal, we have to restrict s sn so that dn= dn1snis real for all n and t tn so that dn= dn2tnis real for all n, and strengthen the inequalities to < in the above construction.
(b) We now come to the question of determining the sequence d when a monic matrix A is given. For j; n 2 IN, we put
un an1;n 1 an1;nan;n 1 an;n 1 an1;n 1
1 an1;n
; vn;j an1;j an1;nan;j;
and for n 2, 0 j n 2, we put
wn;j vn;j unan 1;j an 1;j 1 0
an;j an;n 1 1
an1;j an1;n 1 an1;n
: For each n 2, one of the following two conditions is satis®ed:
(n1) for some jn; 0 jn n 2; wn;jn 6 0; wn;j 0; 0 j < jn. (n2) wn;j 0; 0 jn n 2. This holds if, and only if, for
0 j n; an1;j an1;nan;j unan 1;j if, and only if, qn1 pn1 an1;nqn unqn 1.
Let
T1 n : n 2; n1 holdsf g;
and
T2 n : n 2; n2 holdsf g [ f0; 1g:
The problem reduces to ®nding a suitable function on the set T2to the set of nonzero numbers such that, setting dn1 n for n 2 T2 and then de-
®ning the remaining dn's recursively using (ad)(iii) if jn 1 0 and (ad)(iv) if jn 1> 0, the condition (ad) [respectively, (bd)] is satis®ed.
Theorem 2.6. Let p; A; d; q and r be as in Theorem 2.4 above.
(a) We may combine the conditions (a) of Theorem 2.2 and (ad) of Theorem 2.4 to obtain necessary and sucient conditions on A and d for both q and r to be orthogonal.
(b) For positive orthogonality, we have to combine (b) of Theorem 2.2 with (bd) of Theorem 2.4 which, in turn, put together are equivalent to:
(i) Both A and d are real,
(ii) a2;0 a2;1a1;0< c1 b0 b1a1;0 a1;02; and1
d2 a2;0 a2;1a1;0 < c1 1
d1 b0 b1a1;0 1 d1a1;0
2
;
(iii) an1;n 1 an1;nan;n 1
< cn bn 1 bnan;n 1 an;n 12 an;n 2
anddn 1
dn1 an1;n 1 an1;nan;n 1
< cndn 1
dn bn 1 bnan;n 1 dn 1 dn an;n 1
2
dn 2
dn an;n 2; n 2 ;
(iv) an1;j an1;nan;j an1;n 1 an1;nan;n 1an 1;j
an;j 1 bj bn an;n 1an;j cj1an;j1
cn bn 1 bnan;n 1 an;n 12 an;n 2
h i
an 1;j; and
1 dn1dj 1
dndj
an;j 1
1 dn1
dn
bj bn 1 dn1dn 1
d2n
an;n 1
an;j
1 dn1dj1 dndj
cj1an;j1
"
1 dn1
dn 1
cn 1 dn1
dn
bn 1 bnan;n 1
1 dn1dn 2
dndn 1
an;n 2 1 dn1dn 1
d2n
an;n 12
#
an 1;j 0;
n 2 ; 0 j n 2:
(c) For a set A of monic matrices, we can have a set of conditions on d in order that the members of fSp;dAp : A 2 Ag are all orthogonal or positive orthogonal systems.
In this general form these conditions look formidable. We shall discuss a few special cases here and in the next two sections.
Corollary 2.7. Let d 1n , p a monic orthogonal system with recurrence sequences b and c, and A a monic in®nite matrix. Let q A p and r Sp;dq.
(a) (i) In case bj bnan;j 0 for 0 j < n, q is orthogonal if, and only if, r is orthogonal.
(ii) Suppose b is a constant sequence. Then q 2 Poif, and only if, Sp;dq 2 Po. Thus Sp;dPo Po. In particular, it is so if p is a symmetric orthogonal system. In this case, we also have that q is symmetric if, and only if, r is.
(iii) Suppose b is eventually constant, i.e. , for a least number j12 IN0; bn bj for all n j1. Then Sp;dPo \ Po is in®nite.
(b) Suppose p is positive orthogonal. Then (i), (ii) and (iii) above hold with orthogonal replaced by positive orthogonal and Po by Po.
Proof. (a) (i) In this case Condition (a) of Theorem 2.2 and Condition (ad) of Theorem 2.4 become equivalent.
(ii) Because b is a constant sequence, every monic matrix A satis®es the requirement in (i). For symmetric orthogonal polynomial systems, we have bn 0 for all n.
(iii) Let Aj1 f A 2 Am : an;j 0 for 0 j n ; j < j1 g. Then every A in Aj1 satis®es the requirement in (i) and, therefore, q A p is orthogonal if, and only if, Sp;d q is orthogonal if, and only if, A satis®es (a) of Theorem 2.2. There are in®nitely many matrices in Aj1 satisfying
this condition, as can be easily seen, because aj01;j0 can be chosen arbi- trarily.
(b) In the above proof of Part (a), we only have to use (b) of Theorem 2.2 and (bd) of Theorem 2.4 instead of (a) and (ad). &
Remark 2.8.
(a) Examples of p satisfying condition (iii) of Corollary 2.7 above are provided by Modi®ed Lommel polynomials, Tricomi±Carlitz poly- nomials and polynomials related to Bernoulli numbers (cf. [5], VI.6, VI.7 and VI.8).
(b) For any d1 with 0 6 d216 1, de®ne the sequence d dn dn1.
Let p be a symmetric orthogonal system and A; q; r; as in the above corollary. Then both q and r are symmetric orthogonal if, and only if, aj;k 0 for 0 k j 3 ; j 3 ; ak2;k a2;06 c1 and ak1;k a1;0; k 2 IN0. For the positive orthogonal case, the extra conditions d1; a1;0; a2;02 IR and a2;0< minf c1; d21c1g are needed.
3. KERNEL POLYNOMIALS AND DILATIONS 3.1. Discussion and De®nition
Let p pn be a monic orthogonal polynomial system with the asso- ciated recurrence sequences b bn and c cn. Let Zpbe the set of zeros of pn; n 2 IN0, and Ip the open interval inf Zp; sup Zp. For a number y not in Zp, let py pyn be the corresponding monic kernel polynomial system. We may refer to [5], Section 7, for this and related results to be used in this section.
(i)
pyn cn1 pn y
Xn
k 0
pk y
ck1 pk; where ckj Yk
lj
cl :
Thus, the associated monic matrix Ay ayn;k satis®es ayn;k cnk1pk y
pn y for 0 k < n : (ii) Also, py pyn is orthogonal.
(iii) Further, if p is positive orthogonal, then pyis positive orthogonal if, and only if, y 2 IR n Ip. In this case, we denote by ep epn the corresponding orthonormal system. Then epn 1=
cn1
p pn for n 2 IN, and
thus, for n 2 IN ,
Xn
j 0
epj y epj cn1 1pn y pyn:
This motivates the de®nition and the name of the main object of study in this section.
(iv) A polynomial system q qn will be called analytic withrespect to a polynomial system p pn, in short, p-analytic, if there is a sequence h hn of nonzero numbers with h0 1 such that qnPn
k 0 hkpk; n 2 IN0. The sequence hn will be called the coecient sequence for q. If p is monic, then the monic system Q corresponding to q is given by the matrix A an;k (with respect to p) that satis®es an;k hk=hn; 0 k n. We shall say that q is monic analytic withrespect to p, in short, p-monic analytic.
Any monic matrix A an;k having an;kof the form hk=hn; 0 k n for a sequence hn 2 D gives rise to a q that is p-monic analytic. We note that A and h determine each other uniquely.
(v) If q is p-analytic or p-monic analytic, then for d; d02 D , we have Sp;d6 Sq;d0.
Theorem 3.2. Let p be a monic orthogonal system with recurrence sequences b and c, and let q be monic analytic with respect to p with monic analytic sequence h.
(a) The following conditions are equivalent:
(i) q is orthogonal,
(ii) b0 c1h1hn1 hn bn1hn1 cn2hn2; n 2 IN0, (iii) hn 1=cn1 pn b0 c1h1 ; n 2 IN ;
(iv) q py with y b0 c1h1:
(b) Suppose p is positive orthogonal. Then q is positive ortho±
gonal if, and only if, q py for some y 2 IR n Ip. Further, y and hn are related via
y b0 c1h1 and hn 1
cn1 pn y ; n 2 lN:
Proof. (a) We ®rst note that the monic matrix A for q satis®es an;kak;l an;l, 0 l k n. In this case (a)(iii) and (a)(iv) of Theorem 2.2 are equivalent.
Let q be orthogonal. Then (ii) follows from (a)(iii) of Theorem 2.2.
Moreover, (ii) gives (a)(i) and (a)(ii) of Theorem 2.2 simply because cn6 0 6 hn for all n. This shows the equivalence of (i) and (ii).
Let, for n 2 IN; h0n cn1hn and h00 1. We put y b0 c1h1. Then for n 2 IN0; y h0n h0n1 bnhn0 cnh0n 1. This gives that h0n pn y for n 2 IN0, which readily implies (iii).
(iii) ) (iv) follows from 3.1(i), and (iv) ) (i) is immediate from 3.1(ii).
(b) We only have to combine part (a) with 3.1(iii). &
Theorem 3.3. Let p be an orthogonal polynomial system with recurrence relation
xpn x npn1 x npn x npn 1 x ; x 2 lR; n 2 IN0; p0 x 1; p 1 x 0 :
Let wnQn 1
j 0j=j1 for n 2 IN0 and w0 1. Let q be p-analytic with coecient sequence h hn.
(a) The system q is orthogonal if, and only if, for some number y 62 Zp, we have hn wnpn y , n 2 IN0.
(b) Suppose p is a positive orthogonal polynomial system.
(i) The system q is orthogonal if, and only if, for some y 2 IR n Zp,
hn pn y
kpnk2p ; n 2 IN0:
In particular, if p is an orthonormal polynomial system, then the condition reduces to hn pn y , n 2 IN0.
(ii) The system q is positive orthogonal if, and only if, y in (i) above satis®es y inf Zp or y sup Zp. Moreover, if p is monic then, in the former case, 1nhn> 0 for all n, while in the latter case, hn> 0 for all n .
Proof. (a) Let b and c be the recurrence sequences for the monic polynomial system P corresponding to p. Then q is P-analytic as well with the coecient sequence hn Qn 1
j 0j 1. We note that n n=n1 and
n cn n1=n 1n, where n is the leading coecient of the polynomial pn. From this we can derive wn 1=cn12n. All that we have to do now is to apply the above theorem to P.
(b) In this case, wnabove has the value kpnkp2for each n. Finally, pn x
retains the same sign in 1; inf Zp and the same sign in sup Zp; 1. &
Theorem 3.4. Let p pn be an orthogonal polynomial system with
; ; ; w as in Theorem 3.3 above and d 2 D.
(a) Sp;dPap Pap. (b) Sp;d Pap\ Po
\ Po6 ; if, and only if, there exist distinct numbers y1 and y2, both not in Zp, satisfying dnpn y1 pn y2 for all n.
(c) ] Sp;d Pap\ Po
\ Po 2 if, and only if, there exist distinct pairs y1; y2 and y3; y4 with yi62 Zp; y16 y2, y16 y3, y26 y4, y36 y4, such that
dnpn y2
pn y1pn y4
pn y3 for all n:
(d) If p is positive orthogonal, then we can replace Poby Po, provided we make the yj's satisfy yj2 IR n Ip instead of just yj62 Zp.
Proof. For an h 2 D and qn Pn
j 0 hjpj for n 2 IN0, we have Sp;dqnPn
j 0 djhjpj; n 2 IN0. This gives (a).
Further, by Theorem 3.3, both q and Sp;dq can be in Poif, and only if, there exist distinct numbers y1 and y2 not in Zp satisfying
hn wnpn y1; hndn wnpn y2 for all n; i:e:;
hn wnpn y1; dnpn y1 pn y2 for all n : Thus we have (b).
(c) and (d) follow immediately from (b) in view of 3.1. &
Corollary 3.5. Let p pn be a symmetric orthogonal polynomial system. Then for any two distinct nonzero numbers y1 and y2, both not in Zp, and dn pn y2=pn y1; n 2 IN0, we have ]Sp;d Pap\ Po \ Po 2 .
Proof. We may take y3 y1and y4 y2. &
Remark 3.6.
(a) If we choose y2 y1 in the above corollary, then we always get the sequence d 1n and, thus, Sp;d Pap\ Po
Pap\ Po. This follows from Corollary 2.7, as well!
(b) We may use Mehler's formula ([13], Ex. 23, p. 377 or [2], 2.44 on p. 16) to deduce that in the case of Hermite polynomials there are no other solutions for the condition in Theorem 2.4(iii) except those given by the proof in Corollary 3.5 or (a) above. The same is true for Chebyshev polynomials of the ®rst, as well as the second, kind, as can be easily checked
by just considering the case n 1 and 2. Thus, for these polynomials the following statements hold.
(i) For dn 1n, we have Sp;d Pap\ Po
Pap\ Po:
(ii) For dn pn y2=pn y1 with y1; y2 two distinct numbers, both not in Zp, and y2 6 y1, we have
Sp;d Pap\ Po
\ Po f py2; p y2 g:
(iii) For all other d's, we have Sp;d Pap\ Po
\ Po ; .
(c) We can use [13], Theorem 5.1, or [2], 2.43, to show that there are no solutions to condition (iii) of Theorem 3.4 when we take p to be any Laguerre polynomial system and thus we have that for distinct y1; y2 not in Zp and dn pn y2=pn y1 for all n, Sp;d Pap\ Po
\ Po fpy2g.
(d) As already noted in Theorem 3.3(b)(i), the coecient sequence takes a simple form for an orthonormal polynomial system p. In fact, a direct proof of this part of Theorem 3.3 can be given using the orthogonality measure .
Theorem 3.7.
(a) Let p be an orthogonal polynomial system with recurrence sequences a; b; c, i.e.,
xpn x anpn1 x bnpn x cnpn 1 x; x 2 lR; n 2 IN0: Let g be a polynomial system such that p is g-analytic with constant one as the coecient sequence. Then g is orthogonal if, and only if,
an bn cn a1 b1 c16 a0 b0; n 2:
(b) Let g; p; q be a triple of polynomial systems such that p is g- analytic and q is p-analytic with constant one as the coecient sequence in each case. Then g; p; q are all orthogonal if, and only if, p is orthogonal and the recurrence sequences a; b; c of p satisfy
(i) an bn cn a1 b1 c16 a0 b0, n 2; and (ii) an bn1 cn2 b0 c1, n 2 IN0. (c) The conditions in (b) are equivalent to
(i)0 a1 b1 c1 6 a0 b0,
(ii)0 an 1 bn b0 cn1 c1 0, n 1, and (iii)0 an1 an 1 a1 bn1 bn b1 b0 0, n 1:
Proof. (a) Suppose that g is orthogonal. Let ; ; be its recurrence sequences. Then n; n16 0, n 2 IN0. Also,
Xn
j 0
jgj1 x jgj x jgj 1 x
Xn
j 0
xgj x x pn x
anXn1
j 0
gj x bnXn
j 0
gj x
cnXn 1
j 0
gj x; x 2 lR; n 2 lN0:
This implies Xn
j 0
jgj1 jgj jgj 1
an
Xn1
j 0
gj bn
Xn
j 0
gj cn
Xn1
j 0
gj; n 2 lN0:
Equating coecients of gj on both sides gives 0 a0; 0 a0 b0;
1 a1; 0 1 a1 b1; 0 1 a1 b1 c1; n an; n 1 n an bn;
j 1 j j1 an bn cn and
0 1 an bn cn
for n 2 and 1 j n 1. This implies an bn cn a1 b1 c1 for n 2 and, because 16 0, it also implies a1 b1 c16 a0 b0.
We also note that n an; n 2 lN0;
0 a0 b0; n an an 1 bn; n 2 lN;
1 a1 b1 c1 a0 b0; n cn 1; n 2:
Now, on the other hand, if a; b; c satisfy
an bn cn a1 b1 c16 a0 b0; n 2;
then ; ; , de®ned by the expressions given above, work as recurrence sequences for g and this shows orthogonality of g.
(b) As a byproduct of the computations in the proof of (a), we note that if ; ; are sequences satisfying
n; n16 0; n n1 n2 0 1; n 2 lN0; then the sequences a; b; c, de®ned by
an n; n 2 lN0;
b0 0 0; bn n n n 1; n 2 lN;
cn n1; n 2 IN;
satisfy an bn cn a1 b1 c16 a0 b0, n 2. Thus, (b) follows from (a) and its proof both applied to the pairs g; p and p; q.
(c) is trivial. &
De®nition 3.8. Let p pn be an orthonormal polynomial system with the orthogonality measure . Let y 2 IR n Ip. Let q qn Pn
k 0pk y
pk and Q be the corresponding orthonormal system, i.e., Q Qn qn=kqnkq, and let r rn Pn
k 0 Qk yQk. Then p; q; r
will be called an analytic triple through y.
Theorem 3.9. Let p ; q ; r be an analytic triple through y.
(a) The following are equivalent:
(i) There exists a d 2 D such that Sp;d simultaneously preserves the positive orthogonality of f q; r g.
(ii) There exist y0; y002 IR n Ip; y 6 y0, satisfying p1 y0
pn y0 pn1 y0
"
Xn
j 0
pj y00 pj y0
#
p1 y
pn y pn1 y
"
Xn
j 0
pj y2
#
for all n 2 lN0:
(iii) There exist y0; y002 IR n Ip; y 6 y0, such that either y0 y}and p1 y0p0n1 y0
pn1 y0
p1 yp0n1 y
pn1 y; n 2 IN;
or y06 y00 and p1 y0 y00 y0
pn1 y00 pn1 y0 1
p1 y p0n1 y
pn1 y ; n 2 IN:
The point y0(respectively y00) chosen in (ii) can be taken to be the same as in (iii), and vice versa, and in each case d p n y0=pn y.
(b) For given y0; y002 IR n Ip; y 6 y0, Part (iii)() above is equivalent to each of the following, where nis the leading coecient of pn, as before.
(i) y06 y00 y0 1
2
p1 y p2 y0
p1 y0 p2 yp02 y p02 y0
2 IR n Ip and
p1 y0 y00 y0
pn1 y00 pn1 y0 1
p1 yp0n1 y0
pn1 y ; n 2:
(ii) y06 y0 1
2
p1 y p2 y0
p1 y0 p2 yp02 y p02 y0
2 IR n Ip
and p1 y0 pn1 y0
Xn1
j 1
p jn1 j!
1
2
p1 y p2 y0
p1 y0 p2 yp02 y p02 y0
j 1
" #
p1 yp0n1 y
pn1 y ; n 2: