This article was published in an Elsevier journal.

The attached copy

is furnished to the author for non-commercial research and
education use, including for instruction at the authors institution,
sharing with colleagues and providing to institution administration.
Other uses, including reproduction and distribution, or selling or
licensing copies, or posting to personal, institutional or third party
websites are prohibited.
In most cases authors are permitted to post their version of the
article (e.g. in Word or Tex form) to their personal website or
institutional repository. Authors requiring further information
regarding Elseviers archiving and manuscript policies are
encouraged to visit:
http://www.elsevier.com/copyright

Author's personal copy

J. Math. Anal. Appl. 340 (2008) 521535

www.elsevier.com/locate/jmaa

Strong asymptotics for Sobolev orthogonal polynomials

in the complex plane
G. Lpez Lagomasino a,,1,2 , A. Martnez-Finkelshtein b,3 , I. Prez Izquierdo c ,
H. Pijeira Cabrera a,4
a Universidad Carlos III de Madrid, Spain
b Universidad de Almera, Spain
c Universidad de La Habana, Cuba

Received 24 April 2007

Available online 7 August 2007
Submitted by Richard M. Aron

Abstract
We obtain the strong asymptotics for the sequence of monic polynomials minimizing the norm
 N
1/2

2
(k)


q
qS =
,
k

k=0

where  k , k = 0, . . . , N 1, are L2 norms with respect to measures supported on the same rectifiable Jordan closed curve or
arc , and  N is the L2 norm corresponding to a weight supported on , which satisfies the Szego condition, plus mass points
in the unbounded connected component of C \ .
2007 Published by Elsevier Inc.
Keywords: Sobolev orthogonal polynomials; Asymptotic behavior; Szegos condition

* Corresponding author.

E-mail addresses: lago@math.uc3m.es (G. Lpez Lagomasino), andrei@ual.es (A. Martnez-Finkelshtein), ignacio@matcom.uh.cu
(I. Prez Izquierdo), hpijeira@math.uc3m.es (H. Pijeira Cabrera).
1 Research partially supported by Ministerio de Ciencia y Tecnologa, grant MTM2006-13000-C03-02.
2 Research partially supported by Comunidad de Madrid-Universidad Carlos III de Madrid, grant CCG06-UC3M/EST-0690.
3 Research partially supported by Ministerio de Ciencia y Tenologa, project code MTM2005-09648-C02-01 and by Junta de Andaluca,
FQM-229, FQM-481 and P06-FQM-01738.
4 Research partially supported by a research grant from Centro de Investigacin Matemtica de Canarias (CIMAC), Spain.
0022-247X/$ see front matter 2007 Published by Elsevier Inc.
doi:10.1016/j.jmaa.2007.07.074

Author's personal copy

522

G. Lpez Lagomasino et al. / J. Math. Anal. Appl. 340 (2008) 521535

1. Introduction
Let be an arc or a closed rectifiable Jordan curve in the complex plane. For simplicity, we assume that the
parametrization of with respect to the arc length is a complex valued function f defined on an interval [a, b] R.
We assume that C 2+ ; that is, f has second derivative which satisfies the Lipschitz condition

 
f (x) f  (y)  C|x y| ,
for some constants C, > 0 and for all x, y [a, b].
By we denote the unbounded connected component of C \ . Throughout this paper we assume that
{z1 , . . . , zm } is a finite set of points, and {k }N
k=0 a set of N + 1 finite positive Borel measures supported on ,
where N is such that dN ( ) = N ( )|d |. On the vector space of polynomials P, we consider the inner products

(1)
p, qk = p( )q( ) dk ( ), k = 0, . . . , N 1,


p, qN =


p( )q( )N ( )|d | + p(Z)Aq(Z) ,

p(Z) = p(z1 ), . . . , p (d1 ) (z1 ), p(z2 ), . . . , p (d2 ) (z2 ), . . . , p(zm ), . . . , p (dm ) (zm ) ,

(2)

and
p, qS =

N



p (k) , q (k) k ,

(3)

k=0

p(Z) is the transposed conjugate vector of p(Z), and A is a hermitian

where p (k) denotes the kth derivative of p, 
positive definite matrix of order M = m + m
i=1 di . The norms corresponding to (1), (2), and (3) on P are denoted
 k ,  N , and  S , respectively. (3) is called a Sobolev inner product and (2) a discrete Sobolev inner product.
Notice that only derivatives of order  N are evaluated at the points zk , k = 1, . . . , m, in (3). The nth monic orthogonal
polynomial with respect to the inner product (3) is the unique polynomial Qn of degree n and leading coefficient equal
to 1, such that


n = Qn 2S = inf q2S : q(z) = zn + .
(4)
In the past two decades, the study of the algebraic and asymptotic properties of sequences of Sobolev orthogonal
polynomials has attracted much attention. Let us mention some results related with special cases of (3) dealing with
general classes of measures.
For N = 0, in [4] the authors study the limit of the sequence {Qn /Pn }, n N, where {Pn }, n N, is the sequence
of monic orthogonal polynomials with respect to 0 and 0 > 0 on its support consisting of a bounded interval of R.
In [1], the authors consider a similar problem for general measures 0 in the Szego class supported on an arc or a
closed rectifiable Jordan curve in C.
When N > 0 and A 0 (known as the continuous case) the strong asymptotic of Sobolev orthogonal polynomials
and their first derivative (N = 1) was studied in [5] assuming that 0 and 1 belong to the Szego class. A natural
extension when N > 1 was given in [6].
In this work, we extend the results on strong asymptotics contained in the papers mentioned above. To this aim,
we compare the norms and the monic orthogonal polynomials Qn with respect to the general inner product (3) with
the norms and the monic orthogonal polynomials with respect to (2).
Let be a finite positive Borel measure supported on and = ( )|d | + s its Lebesgue decomposition on
with respect to |d |. We say that satisfies the Szego condition on , and we write S( ), if




(5)
log ( )   ( )|d | > ,

Author's personal copy

G. Lpez Lagomasino et al. / J. Math. Anal. Appl. 340 (2008) 521535

523

where is the conformal mapping of onto the exterior of the unit circle such that
1
(z)
=
> 0.
z z
C( )

 () = lim

() = ,

C( ) is the logarithmic capacity of . In particular, (5) implies that > 0 almost everywhere on . If is absolutely
continuous with respect to |d | and verifies (5), we write S( ), and use in place of in the notation of norms
and polynomials.
Set
n = Ln 2N =

inf

Rn (z)=zn +

Rn 2N ,

(6)

where Ln (z) = zn + . Our first result establishes the asymptotic behavior of the Sobolev norms (4).
1
2+
Theorem 1. Let {k }N
k=0 be a set of N finite positive Borel measures supported on C , dN ( ) = N ( )|d |,
with N S( ), and {z1 , . . . , zm } a fixed set of points. Then

n
2N
n n nN
lim

= 1.

(7)

Consider the extremal problem

(N ) = min f 2H 2 (,

f H

N)

= F 2H 2 (, ) ,
N

(8)

 = {f H 2 (, N ): f () = 1, f (j ) (zi ) = 0, j = 0, . . . , di , i = 1, . . . , m} (see Section 2 for the definition

where H
2
 exists and is unique. Notice that
of H (, N )). Standard arguments show that a minimizing function F H
(F )(j ) (zi ) = 0,

j = 0, . . . , di , i = 1, . . . , m.

(9)

The function F allows to describe the asymptotic behavior of the N th derivative of the polynomials Qn .
By limn n = in , we denote local uniform convergence in the region of the sequence of functions
{n }nN as n (i.e. uniform convergence on each compact subset of ).
Theorem 2. Under the assumptions of Theorem 1, we have

nN 

)
N
Q(N
F (z) 1 + n (z) ,
n (z) = n C( )(z)

(10)

where limn n = 0 in . Equivalently,

(N )

Qn (z)
= 1 in .
lim
n nN LnN (z)

(11)
(N )

Notice that the asymptotic behavior of {Qn } only depends on the weight N and the interaction matrix A. Some of
the measures k , k = 0, . . . , N 1, may even be zero as long as ,S is an inner product in the space of polynomials.
1
o condition on , the asymptotic behavior of
If we assume additionally that the measures {k }N
k=0 satisfy the Szeg
the polynomials Qn and all its derivatives can be described as follows:
Theorem 3. If j S( ), j = 0, . . . , N , and N is absolutely continuous, then for k Z+
k
Q(k)
n (z) = n

C( )nN (z)nk 
F
(z)
1
+

(z)
,
n
(  (z))N k

(12)

where limn n = 0 in . Equivalently, for k Z+

(k)

1
Qn (z)
=
k
n n Lnk (z)
([C( )] (z))N k
lim

in .

(13)

Author's personal copy

524

G. Lpez Lagomasino et al. / J. Math. Anal. Appl. 340 (2008) 521535

From (9), (12), and Hurwitz theorem, the asymptotic behavior of the zeros of the polynomials Qn is derived. Let
G = C \ ; therefore, G is an open connected set if is a closed curve, and it is empty if is an arc.
Corollary 1. Under the assumptions of Theorem 3, for n sufficiently large, each point zi , i = 1, . . . , m, attracts exactly
di + 1 zeros of Qn . The rest of the zeros accumulate on G.
The paper is organized as follows. The next section, includes the definitions and auxiliary results needed in the
sequel. Section 3 is dedicated to the proof of Theorem 1. Finally, Section 4 contains the proof of Theorems 2 and 3.
2. Definitions and auxiliary results
In order to give a unified treatment to the cases of an arc and a closed curve, we will consider a rectifiable Jordan
arc as a two sided cut of the complex plane. If is an arc and g is an analytic function in with boundary values
almost everywhere on (with respect to the Lebesgue measure) when we approach the arc from each side, we denote
by g+ ( ) and g ( ) the two boundary values of g at a given point .
If g+ and g are integrable on with respect to the weight , we define



g( )( )|d | = g+ ( )( )|d | + g ( )( )|d |.

When is a closed curve and g is analytic on with boundary values g( ) for almost all , we define


g( )( )|d | = g( )( )|d |.

Assume that is a finite positive Borel measure supported on C 2+ and d = ( )|d |, S( ). Let Pn (; z)
denote the nth monic orthogonal polynomial with respect to , and



2


p( )2 ( )|d |.
n () = Pn (; ) ( )|d | =
infn
(14)
p(z)=z +

= 1 ({w

C: |w| = r}), r > 1. Since C 2+ , by Carathodorys theorem (see [7, Chapter 2]), and
Set r
1 have one-to-one continuous extensions to the boundary of their respective domains of definition (considering in
the case that is an arc that it has two sides) and |( )| = 1 (| ( )| = 1 if is an arc) for all .
Since S( ), there exists a unique function D(; ) analytic on satisfying:
(1) for all z , D(; z) = 0,
(2) D(; ) > 0,
(3) for almost all , D(; ) has non-tangential limit D(; ) (D+ (; ), D (; ) if is an arc) and |D(; )| =
( ) (|D (; )| = ( )).
If g(z; ) = log|(z)| denotes the Green function of with singularity at infinity, and / denotes the exterior
normal derivative to , then



g( ; )
1
log ( )
D(; ) = exp
|d | .
(15)
2

An analytic function f on is said to belong to E 1 (), if




sup f (z)|dz| < .
r>1

By H 2 (, ) we denote the space of analytic functions f on such that f 2 D(, ) E 1 ().

Author's personal copy

G. Lpez Lagomasino et al. / J. Math. Anal. Appl. 340 (2008) 521535

525

Each f H 2 (, ) has a non-tangential limit f ( ) (f+ ( ), f ( ) if is an arc) at almost all . Moreover,

f ( ) L2 (), (f+ ( ), f ( ) L2 ()). The inner product

f, gH 2 (,) = f ( )g( )( )|d |,
(16)

makes H 2 (, ) a Hilbert space. By the Riesz representation theorem it follows that there exists (see [8, 7]) a
function K(t, z), called Szego reproducing kernel, such that for any f H 2 (, )


f (z) = f, K(., z) H 2 (,) , z .
(17)
Let  H 2 (,) be the norm induced by (16). The extremal property (14) of the polynomials Pn (; ) motivates the
following extremal problem:


(18)
() = min f 2H 2 (,) : f H 2 (, ), f () = 1 .
It has a unique minimizing function, which we denote by F , satisfying additionally (see [8, 6 and 7])
D(; )
, z ,
D(; z)
() = 2D(; )C( ),

(19)

F(z) = ()K(z, ),

(20)

F 2 (z) =  (z)

z .

The function F and the extremal constant () are intimately connected with the asymptotic properties of the orthogonal polynomials Pn (; ). In fact,
n ()
= (),
n C( )2n

n


Pn (; z) = C( )(z) F(z) 1 + n (z) ,
lim

(21)
(22)

where limn n = 0 in .
In [1], the following extremal problem is considered
() = min f 2H 2 (,) ,

f H

(23)

 = {f H 2 (, ): f () = 1, f (j ) (zi ) = 0, j = 0, . . . , di , i = 1, . . . , m}. Let F be the extremal function

where H
for the problem (23). There exists a close connection between the extremal constants and functions of problems (18)
and (23) (see [2,3]). In fact,
F (z) = B(z)F(z),
m

2(d +1)

() = () (zi ) i ,

(24)
(25)

i=1

where


m 

(z) (zi ) |(zi )|2 di +1
B(z) =
(z)(zi ) 1 (zi )
i=1

has the following properties:

(1) B H 2 (, ), B() = 1,
di , i = 1, . . . , m,
(2) B (j ) (zi ) =

0, j = 0, . . .d,+1
di +1 , if is an arc).
i
|(z
)|
, (|B ( )| = m
(3) |B( )| = m
i
i=1
i=1 |(zi )|
The authors of [1] prove that
m
2(di +1)
n () 
(zi )
lim
=
,
n n
i=1

(26)

Author's personal copy

526

G. Lpez Lagomasino et al. / J. Math. Anal. Appl. 340 (2008) 521535

and the relation of the extremal function F and constant () with the asymptotic behavior of the orthogonal
polynomials Ln with respect to (3) with N = 0 is
n
lim
= (),
(27)
n C( )2n

n 

(28)
Ln (z) = C( )(z) F (z) 1 + n (z) ,
where limn n = 0 in .
Notice that the asymptotic formulas (27) and (28) coincide in form with those given in (21) and (22) for n () and
Pn (, ), respectively, only the extremal constant and function change.
The following result shows that for functions in H 2 (, ), estimates of in the L2 norm imply estimates on
K in the uniform norm (see [8, Corollary 7.4]).
Lemma 1. Let K be a compact set and S( ). Then, there exists a constant A(K), which only depends on K,
such that for all f H 2 (, )
2

maxf (z)  A(K)f 2H 2 (,) .
zK

3. Asymptotics of the Sobolev norms

We begin establishing two auxiliary results needed for the proof of (7) in Theorem 1.
Let P be the family of monic polynomials in P. Fix a point 0 and consider the mapping : P P given
by
z
R(z) = z +
n

(R)(z) := (n + 1)

R( ) d,
0

that is, (R) is the monic primitive of the monic polynomial R that vanishes at 0 . Now, define 0 = I (the identity
operator on P ), k = k1 , k N. If R(z) = zn + , then deg(k (R)) = n + k and for each j {0, . . . , k}
(n + k)!
(j )
kj (R)(z).
k (R)(z) =
(29)
(n + k j )!
Lemma 2. Let be a finite positive Borel measure supported on , d( ) = ( )|d |, a monic polynomial whose
zeros lie in C \ , deg = M, W = ||2 and Pn = Pn (W; ) the nth monic orthogonal polynomial with respect to
the measure W|d |. If 1/ L1 ( ), then for each k N, the sequence of polynomials
n! k (PnM )(z)

, n N, n > M,
(30)
n,k (z) :=
(n + k)!
nM (W)
is uniformly bounded on and tends to zero for all z as n .
Proof. Let l be the length of and = (t), t [0, l], (0) = (l) = 0 , 0 , be the parametrization of with
respect to the arc length. If z ,
1 1 (PnM )(z)

=
n,1 (z) :=
n+1
nM (W)

s

(z)

( )pnM ( ) d =





(t) pnM (t)  (t) dt,

where (z) = [0, s] denotes the arc along from 0 to z, following the orientation given by the parametrization,
and pn = pn (W; ) is the nth orthonormal polynomial with respect to the measure W|d |.
If 1(z) is the characteristic function of (z), then



W ((t))((t))


n,1 (z) =
pnM (t) (t) dt = f (z; )pnM ( )W ( )( )|d |
1(z) (t)
((t))((t))
[0,l]

= f (z; ), pnM

L2 (W) ,

(31)

Author's personal copy

G. Lpez Lagomasino et al. / J. Math. Anal. Appl. 340 (2008) 521535

527

where
f (z; ) =

1(z) ( )  (t)
,
( ) ( )

Thus, n,1 (z) is the conjugate of the (n M)th Fourier coefficient of f (z; ) with respect to the orthonormal system
{pn } in L2 (W). Since 1/ L1 ( ), then for all z ,
 

2



1


f (z; )2 2
 
f
(z;

)
=
W
(
)(
)|d
|

(32)
  1 .
L (W)
L ( )

Using (31), (32), and the Bessel inequality, we obtain

 1/2


 
1

n,1 (z)  f (z; ) 2

  1 , z , n > M,
L (W)
L ( )
lim n,1 (z) = 0,

z ,

which establishes Lemma 2 for k = 1. For the case of k  2 one can proceed by induction making use of the identity
z
n,k+1 (z) =

n, k N, n > M, z C,

n,k (x) dx,

0

and the Lebesgue dominated convergence theorem.

For what follows, we fix the notation
(z) =

m


di +1

(z zi )

M = deg = m +

i=1

m


di ,

W = ||2 .

(33)

i=1

Lemma 3. Assume that the assumptions of Theorem 1 are satisfied and let Pn = Pn (WN ; ) be the nth monic orthogonal polynomials with respect to the measure WN |d |, with W given in (33). Then
n
 1,
(34)
lim sup 2N
n n nN M (WN )
where n = Qn 2S and n (WN ) = Pn 2L2 (W

Proof. Let us assume initially that

n = Qn 2S 

N
1


(n N)!
n!

2

L1 ( ). By the extremal property of Qn , we have

N
1

k=0

Hence






 N (PnN M ) (k) 2 +  N (PnN M ) (N ) 2

k=0

1
N

|d |)

n!
(n k)!

2



N k (PnN M )2 +
k

n!
(n N)!

2
PnN M 2N .


N
1

(n N )! 2 N k (PnN M )2k
n

+ 1,
nN M (WN )
(n k)!
nN M (WN )
k=0

and
n
n2N nN M (WN )

N
1

k=0

nN,N k 2k + 1.

Author's personal copy

528

G. Lpez Lagomasino et al. / J. Math. Anal. Appl. 340 (2008) 521535

From Lemma 2 and the Lebesgue dominated convergence theorem, it follows that
lim nN,N k 2k = 0,

k = 0, . . . , N 1.

Consequently, we have (34) when 1/N L1 ( ).

Now, assume that the weight N S( ) is arbitrary. Take a constant > 0 (to be fixed later) and define N ( ) =
n = P
n (W N ; ) be the nth monic orthogonal polynomial with respect to the measure d (
)=
N ( ) + . Let P
Then
W ( )N ( )|d |, and let n,k be the sequence defined in (30) corresponding to the measure .
 N 1


nN M (W N )
n
2
.
(35)

 nN,N k k + 1
2N
nN M (WN )
n nN M (WN )
k=0

But 1/N

L1 ( );

lim sup
n

therefore, from Lemma 2 and (35), we have that

n

n2N nN M (WN )

nN M (W N )
.
n nN M (WN )

 lim

(36)

Since N S( ), then WN , W N S( ). Therefore, from (21), (18) and (15), we obtain





 g( ; )
nN M (W N ) (W N )
1
lim
log(N + ) log N ( )
=
= exp
|d | .
n nN M (WN )
(WN )
2

(37)

It remains to use the continuity of the extremal constant (), S( ), in the metric

g( ; )
1
|d |, , S( ).
|log log |( )
dist(, ) =

In fact, by the Lebesgue dominated convergence theorem

dist(N , N ) 0,

0.

Then, for an arbitrary fixed > 0, we can take > 0 in (37) so that
nN M (W N )
 1 + .
n nN M (WN )
lim

(38)

Now, (34) follows from (36), (38), and the arbitrariness of > 0.

Proof of Theorem 1. By the extremal property (6), for all n Z+ , n > N , we have
2

N
1

 (N ) 2
 (k) 2  (N ) 2
n!






Qn k + Qn N  Qn N 
nN .
n =
(n N)!
k=0

Therefore,
lim inf
n

n
 1.
n2N nN

(39)

On the other hand,

n
nN M (WN ) nN (N )
n
=
.
nN
n2N nN
n2N nN M (WN ) nN (N )
According to (21) and (19), we have
m



1
nN M (WN ) D(WN ; )
(zi )2(di +1) ,
=
=
n
nN (N )
D(N ; ) C( )2M

lim

i=1

since due to (15),

m



D(WN ; )
2M
(zi )2(di +1) .
= C( )
D(N ; )
i=1

Author's personal copy

G. Lpez Lagomasino et al. / J. Math. Anal. Appl. 340 (2008) 521535

Using (26) and (34), we conclude that

n
lim sup 2N
 1.
n
nN
n

529

(40)
2

Now, (7) follows from (39) and (40).

4. Asymptotics of Sobolev polynomials

The following two lemmas pave the way for the proof of Theorem 2.
Lemma 4. Assume that the assumptions of Theorem 1 are satisfied and let
(N )

nN (z) = nN

Qn
(z),
nN

where nN =

(n N )!
, n N, n > N.
n!C( )nN

(41)

Then {nN } is uniformly bounded on compact subsets of and

(j )

lim nN (zi ) = 0,

j = 0, . . . , di , i = 1, . . . , m.

(42)

Proof. From the definition of n , we have

2
n
nN

N
1


 (k) 2 

2
Q  + nN Q(N ) 2 2
nN
n k
n
L (

N)

2
)
(N )

+ nN
Q(N
n (Z)AQn (Z) .

(43)

k=0

Using (7) and (27), it follows that

lim 2 n
n nN

= (N ).

(44)

On the other hand, (43) and (44) imply

lim sup nN 2H 2 (,

N)

 (N ).

Therefore, the sequence {nN } is bounded in H 2 (, N ) and Cauchys integral formula renders that {nN } is
uniformly bounded on compact subsets of .
For A hermitian and positive definite,
x Ax
= 1 > 0,
xCM , x =0 x x
min

where 1 > 0 is the smallest eigenvalue of A. In particular, for every x CM ,

xx 

x Ax
.
1

(45)

2
Qn (Z)AQn (Z) }nN is bounded; that is, there exists a
From (43) and (44) we have that the sequence {nN
constant C > 0 such that

 

)
(N )

(46)
nN Q(N
n (Z) A nN Qn (Z)  C, n N.
(N )

Hence, by (45), sequence


)

M
nN Q(N
n (Z) nN C ,

(N )

(47)

is uniformly bounded with respect to the Euclidean norm. Since in a finite-dimensional space all norms are equivalent,
we conclude that sequence (47) is bounded with respect to the uniform norm; that is,


(N +j )
nN Qn
(zi ) nN , j = 0, . . . , di , i = 1, . . . , m,
are uniformly bounded.

Author's personal copy

530

G. Lpez Lagomasino et al. / J. Math. Anal. Appl. 340 (2008) 521535

We have


+j )
nN Q(N
(zi ) =
n



(N +j )
nN j 

(n N )! 
Q
n

(zi )
(zi ),

nN
j
j
n! C( )
[C( )]

and
nN j
(n N)! 
(zi )
= .
n
n!
Therefore,
lim

(N +j )

lim

Qn

[C( )]nN j

(zi ) = 0,

j = 0, . . . , di , i = 1, . . . , m,

(48)

and convergence takes place with geometric rate. Let us see that

(j )
(N )
Qn
lim
(zi ) = 0, j = 0, . . . , di , i = 1, . . . , m,
n [C( )]nN

(49)

also converges with geometric rate.

According to the Leibnitz formula,


(j )

(N )

Qn

(zi ) =

[C( )]nN

j  

j
k=0

+k)
Q(N
(zi )
n

(j k)
(50)

(zi ).

[C( )]nN

The derivative of order l, l N, of the function 1/[C( )]nN , is of the form

l


(1)

t=1

t


(n N + r 1)

r=1

Ct
[C( )]nN +t

(51)

(z),

where Ct (z), t = 1, . . . , l, do not depend on n, but only on C( ) and on some of the values  (z), . . . , (l) (z).
From (50) and (51), we have


(j )

(N )

Qn

[C( )]

nN

(zi ) =

j  
j k

j
k=0

(1)

t=1

t


(n N + r 1)

r=1

(N +k)

Qn

nN k

[C( )]

Now, using (52) and (48), we obtain (49). Finally, (42) follows from (41) and (49).

(zi )

Ct
[C( )]k+t

(zi ).

(52)

Associated with the extremal function F and the conformal mapping , we consider the sequence of functions {Hn }, defined almost everywhere on by

F ( ) n ( ),
if is a closed curve,

Hn ( ) =
n
n

F+ ( )+ ( ) + F ( ) ( ), if is an arc.
Lemma 5. Under the assumptions of Theorem 1, we have


(N )


Q
n

HnN N
= 0.
lim 

 2
nN
n
n C( )
L (N )
Proof. Obviously,


(N ) 2
H
nN nN Qn
L2 (

N)

 2
 2
= HnN
L (

N)

(53)



) 2
+ nN Q(N
n
L2 (

In connection with the first term of (54), if is a closed curve, then

N)

)
2 Re HnN
, nN Q(N
n L2 ( ) .
N

(54)

Author's personal copy

G. Lpez Lagomasino et al. / J. Math. Anal. Appl. 340 (2008) 521535


H

2
 2

nN L (N

=
)


 2 
F ( ) ( )2(nN ) N ( )|d | =

531

 2
F ( ) N ( )|d |

= F 2H 2 (, )
N

= (N ),

(55)

since F is the solution of the extremal problem (23) and |( )| = 1 for . If is an arc, we have
 2

nN
nN
nN
nN 
H

+ F
, F+ +
+ F
nN L2 (N ) = F+ +
L2 (N )


2(nN )
2(nN )
 2 
 2 
= F+ ( ) + ( )
N ( )|d | + F ( )  ( )
N ( )|d |

nN
nN 
, F
+ 2 Re F+ +
L2 (N )


 2
nN
nN


F ( ) N ( )|d | + 2 Re +
=
( )
( )F+ ( )F ( )N ( )|d |

= F 2H 2 (,

N)

(nN )

nN
+
( )

+ 2 Re

( )F+ ( )F ( )N ( )|d |,

(56)

and the second term tends to zero when n (see [8, Lemma 12.1]). Therefore, if is an arc or a closed curve,
 2
 2
lim HnN
= (N ).
(57)
L ( )
n

Let us consider the second term in the right-hand side of (54). We have
 (N ) 2


) 2
Q  2
 Q(N
 Qn 2S = n .
n
n
L ( )
N
N

Hence,
 (N ) 2
2
Q  2
nN
n
L (N )

(n N )!
n!

From (27), (7) and (58),



) 2
lim supnN Q(N
n
L2 (

N)

2

n
nN

nN
.
C( )2(nN )

(58)

 (N ).

(59)

Consider the third term in (54). If is a closed curve,

 (N )


(N ) 
Qn
(N )

nN
Qn , HnN L2 ( ) = Qn ( )F ( )
( )N ( )|d | =
, F
N
nN

If is an arc,

(N ) 
Qn , HnN L2 (


=
)

Q(N
n ( )HnN ( )N ( )|d |

 (N )
nN
nN
)

( ) + Q(N
( ) N ( )|d |
Qn ( )F+ ( )+
n ( )F ( )

 

.
H 2 (,N )

(N )

Qn ( )
nN
+
( )

F+ ( ) +

(N )

Qn ( )
nN

( )

F ( )

N ( )|d |

 (N )
(N )
Qn ( )
Qn
F ( )N ( )|d | =
, F
nN

( )
nN

,
H 2 (,N )

where we used again that || = 1 (| | = 1) on . Therefore, if is an arc or a closed curve, from the definition of
nN , (24), (25), (20), and the properties of B(z), it follows that

Author's personal copy

532

G. Lpez Lagomasino et al. / J. Math. Anal. Appl. 340 (2008) 521535

nN Q(N
n , HnN


L2 (N )


= nN , F H 2 (,N ) = (N ) nN , BK(, ) H 2 (, )
N

nN
|B|2 , K(, )
= (N )
B
H 2 (,N )

nN
, K(, )
= (N )
.
B
H 2 (,N )

(60)

/ H 2 (, N ), since the points zi , i = 1, . . . , m, are poles of this

Notice that (nN /B)() = 1, and nN /B
function. For all n N,
m di
anN
nN (z)  
=
+ gn (z),
B(z)
(z zi )j
(i,j )

gn H 2 (, N ), gn () = 1.

(61)

i=1 j =0

Substituting (61) in (60) and using property (17) of the reproducing kernel, we obtain



di
m 



1
(i,j
)
)

nN Q(N
anN
, K(, )
.
n , HnN L2 (N ) = (N ) 1 +
(z zi )j
H 2 (,N )

(62)

i=1 j =0

Now, suppose that

lim inf Re nN Q(N

n , HnN L2 (

N)

Let N be such that



lim Re nN Q(N
n , HnN L2 (

N)

< (N ).

(63)

= lim inf Re nN Q(N

n , HnN L2 ( ) .
n

(64)

Since {nN } is uniformly bounded on compact subsets of , there exists  and  , holomorphic in , such
that for j Z+ ,
(j )

(j )

lim nN = 

in where n  .

(65)

From (42) and (65), we have that

(j )

 (zi ) = 0,

j = 0, . . . , di , i = 1, . . . , m.

(66)

Using the definition of B(z), (65) and (66), we have that  /B H 2 (, N ) and for n  ,
lim


nN
=
B
B

in \ {z1 , . . . , zm } where n  .

(67)

On account of (61) and (67), we conclude that anN 0, j = 0, . . . , di , i = 1, . . . , m, for n  , and from (62), we
obtain


lim Re nN Q(N
(68)
n , HnN L2 ( ) = (N ).
(i,j )

Since  , (64) and (68) contradict the assumption (63). Therefore,



lim inf Re HnN

, nN Q(N
n L2 ( )  (N ).
n

(69)

Taking upper limit on both sides of (54), (53) follows from (57), (59) and (69).

Proof of Theorem 2. If is a closed curve,




)
) 



 (n N )!
Q(N
(n N )! Q(N
n
n



F
H
=

.
nN

n!
[C( )]nN H 2 (,N ) 
n!
C( )nN L2 (N )
Lemma 1 and (53) imply
(N )

Qn
(z) = F (z) in ,
N
n n [C( )]nN
which is (10). Due to (28), (10) is equivalent to (11).
lim

Author's personal copy

G. Lpez Lagomasino et al. / J. Math. Anal. Appl. 340 (2008) 521535

533

If is an arc, using (17), we have


(N )
(N )
(n N )!
Qn
Qn
(n N)!
(z)
=
, K(, z)
n!
[C( )]nN
n!
[C( )]nN
H 2 (,N )

(N )
Qn
(n N )!
=
( )K(, z)N ( )|d |
n!
[C( )]nN

(N )

(n N )! Qn ( ) nN

( )K(, z)N ( )|d |

n!
C( )nN

(n N )! Qn ( )  nN
nN
+ ( )K+ (, z) +
( )K (, z) N ( )|d |
nN
n!
C( )


nN
nN
= HnN
, +
K+ (, z) +
K (, z) L2 ( )
N

(N )
(n N )! Qn
nN

HnN

, nN K+ (, z) +
K (, z)
.
n!
C( )nN +
L2 (N )
(N )

Using the CauchySchwartz inequality and (53), it is easy to see that the second term in the last equality tends to zero
as n . Additionally,


nN
nN
HnN , +
K+ (, z) +
K (, z) L2 ( )
N


= F , K(., z) H 2 (, )
N



(nN )
(nN )
nN
nN
F+ ( )+
+
( )
( )K (, z) + F ( )+
( )
( )K+ (, z) N ( )|d |

= F (z) +

(nN )

nN
F+ ( )+
( )

(nN )

( )K (, z) + F ( )+


nN
( )
( )K+ (, z) N ( )|d |.

In the last equality, the second term tends to zero since it is the Fourier coefficient of an integrable function. Therefore,
we also have (10) when is an arc and we conclude. 2
In proving Theorem 3, we use the following auxiliary result.
Lemma 6. Under the assumptions of Theorem 3, we have for k = 0, . . . , i 1 and i = 1, . . . , N ,
(k)

Qn
(z) = 0 in .
i
n n [C( )]nk
lim

Proof. Let k ( )|d |, k = 0, . . . , N , be the absolutely continuous component of dk ( ). We have

2 
2
N
1
(k)
(N )





Qn
Qn
n




=
+ N



N
nN
nN
2N
2(nN
)
n C( )
n
C(
)
n C( )
k
N
k=0

N
1

k=0

2


Q(k)
n


 nN C( )nN  2

Since (see (44))

lim

n
= (N ),
n2N C( )2(nN )

and, according to (53) and (57),

L (k )

2

)


Q(N
n


+ N
 2
nN
n C( )

L (N )

Author's personal copy

534

G. Lpez Lagomasino et al. / J. Math. Anal. Appl. 340 (2008) 521535


2
(N )


Q
n

lim 

 2
N
nN
n n C( )

L (N )

 2
 2
= lim HnN
L (

N)

it follows that

2
(k)


Qn


lim  N
= 0,
n n [C( )]nk  2
L (k )

= F 2H 2 (,

N)

= (N ),

k = 0, . . . , N 1.

Since k S( ), k = 0, . . . , N , using Lemma 1, we obtain for k = 0, . . . , N 1,

(k)

Qn
(z) = 0 in .
n nN [C( )]nk
Therefore, the lemma is true when i = N .
By the Weierstrass theorem, for k = 0, . . . , N 1,


(k)
Qn
lim
(z) = 0 in .
n nN [C( )]nk
lim

(70)

(71)

It is easy to verify that



Q(k)
Qn(k+1)
(n k)  (z)
1
Q(k)
n
n
(z) =
(z) N
(z).
n(z)
C( )(z) nN [C( )]n(k+1)
n [C( )]nk
nN 1 [C( )]nk

Therefore, using (70) and (71),

(k)

Qn (z)
= 0 in
N
1
n n
[C( )]nk (z)
for k = 0, . . . , N 2. Repeating the arguments, we conclude the proof.
lim

Proof of Theorem 3. For a given k Z+ , (12) and (13) are equivalent due to (28). On the other hand, using the
Weierstrass theorem it is easy to prove that if (12) is true for a certain k then it holds for all i  k. For k = N , (12)
coincides with (10). Therefore, (12) takes place for all k  N . To conclude, we will assume that (12) is satisfied for
some k, 1  k  N , and prove that it also holds for k 1.
We have


(k1)
(k)
(k1)
Qn
Qn
1
Qn
(n + 1 k)  (z)
(z) =
(z) k
(z).
n(z)
C( )(z) nk [C( )]nk
nk1 [C( )]n(k1)
n [C( )]n(k1)
From Lemma 6 and the Weierstrass theorem, the second term on the right-hand side of this equality tends to zero
uniformly on compact subsets of . By assumption,
Qn (z)
1
F (z)
1
=
in .
n C( )(z) nk [C( )]nk (z)
C( )(z) ([C( )] (z))N k
Taking limit on both sides of the equality above, (12) immediately follows for k 1 and we are done.
(k)

lim

Remark. The following questions arise naturally. First of all, can the assumption that N is absolutely continuous be
suppressed? Moreover, consider a Sobolev inner product of the form
N 

=
p (k) ( )q (k) ( ) dk ( ) + p(Z)Aq(Z) ,
p, q
S
k=0

where


p(Z) = p(z1 ), . . . , p (d1 ) (z1 ), p(z2 ), . . . , p (d2 ) (z2 ), . . . , p(zm ), . . . , p (dm ) (zm ) ,


zi , i = 1, . . . , m, and A is a hermitian semi positive definite matrix of order M = m + m
i=1 di . Under what
assumptions can we obtain theorems similar to those above for the corresponding sequence of Sobolev orthogonal
polynomials? How does the appearance of derivatives of order < N in the discrete part affect the results?

Author's personal copy

G. Lpez Lagomasino et al. / J. Math. Anal. Appl. 340 (2008) 521535

535

References
[1] A. Branquinho, A. Foulqui, F. Marcelln, Asymptotic behavior of Sobolev-type orthogonal polynomials on a rectifiable Jordan curve or arc,
Constr. Approx. 18 (2002) 161182.
[2] V.A. Kaliaguine, R. Benzine, Sur la formule asymptotique des polynmes orthogonaux associs une mesure concentre sur un contour plus
une partie discrte finie, Bull. Soc. Math. Belg. Ser. B 41 (1989) 2946.
[3] V.A. Kaliaguine, A note on the orthogonal polynomials on a complex arc: The case of a measure with a discrete part, J. Approx. Theory 80
(1995) 138145.
[4] G. Lpez Lagomasino, F. Marcelln, W. Van Assche, Relative asymptotics for polynomials orthogonal with respect to a discrete Sobolev inner
product, Constr. Approx. 11 (1995) 107137.
[5] A. Martnez-Finkelshtein, BernsteinSzegos theorem for Sobolev orthogonal polynomials, Constr. Approx. 16 (2000) 7384.
[6] A. Martnez-Finkelshtein, H. Pijeira, Strong asymptotics for Sobolev orthogonal polynomials, J. Anal. Math. 78 (1999) 143156.
[7] C. Pommerenke, Boundary Behaviour of Conformal Maps, Springer, Berlin, 1992.
[8] H. Widom, Extremal polynomials associated with a system of curves in the complex plane, Adv. Math. 3 (1969) 127232.