On a finite moment perturbation of

linear functionals and the inverse

Szegő transformation
Edinson Fuentes & Luis E. Garza
Universidad Nacional de Colombia, efuentes@unal.edu.co
Universidad de Colima, luis garza1@ucol.mx

Abstract
Given a sequence of moments {cn }n∈Z associated with an Hermitian
linear functional L defined in the space of Laurent polynomials, we study a
new functional LΩ which is a perturbation of L in such a way that a finite
number of moments are perturbed. Necessary and sufficient conditions
are given for the regularity of LΩ , and a connection formula between the
corresponding families of orthogonal polynomials is obtained. On the
other hand, assuming LΩ is positive definite, the perturbation is analyzed
through the inverse Szegő transformation.

1 Preliminaries and introduction

Consider a linear functional L defined in the linear space of Laurent polynomials
Λ = span{z n }n∈Z such that L is Hermitian, i.e.,
cn = hL, z n i = hL, z −n i = c−n , n ∈ Z.
Then, a bilinear functional can be defined in the linear space P = span{z n }n≥0
of polynomials with complex coefficients by
hp(z), q(z)iL = hL, p(z)q̄(z −1 )i, p, q ∈ P.
The sequence of complex numbers {cn }n∈Z is called the sequence of moments
associated with L. On the other hand, the Gram matrix associated with the
canonical basis {z n }n≥0 of P is
 
c0 c1 ··· cn ···
 c−1 c0 · · · cn−1 · · · 
 .. .. ..
 
.. 
T=  . . . . ,
 (1)
 c−n c−n+1 · · · c 0 · · · 
 
.. .. .. ..
. . . .

which is known in the literature as Toeplitz matrix [7]. A sequence of monic

polynomials {Φn }n≥0 , with deg (Φn ) = n, is said to be orthogonal with respect
to L if the condition
hΦn , Φm iL = δm,n Sn ,

1
where Sn 6= 0, holds for every n, m ≥ 0. Notice that the sequence {Φn }n≥0 can
be obtained by using the Gram-Schmidt orthogonalization process with respect
to the basis {z n }n≥0 . The necessary and sufficient conditions for the existence of
such a sequence can be expressed in terms of the Toeplitz matrix T: {Φn }n≥0
satisfies the orthogonality condition if and only if Tn , the (n + 1) × (n + 1)
principal leading submatrix of T, is non-singular for every n ≥ 0. In such a case,
L is said to be quasi-definite (or regular). On the other hand, if det Tn > 0
for every n ≥ 0, then L is said to be positive definite and it has the integral
representation Z
hL, p(z)i = p(z)dσ(z), p ∈ P, (2)
T
where σ is a nontrivial positive Borel measure supported on the unit circle
T = {z : |z| = 1}. In such a case, there exists a (unique) family of polynomials
{ϕn }n≥0 , with deg ϕn = n and positive leading coefficient, such that
Z
ϕn (z)ϕm (z)dσ(z) = δm,n . (3)
T

{ϕn }n≥0 is said to be the sequence of orthonormal polynomials with respect to

σ. If we denote by κn the leading coefficient of ϕn (z), then we have Φn (z) =
ϕn (z)/κn . These polynomials satisfy the following forward and backward re-
currence relations (see [7], [10], [11])

Φn+1 (z) = zΦn (z) + Φn+1 (0)Φ∗n (z), n ≥ 0, (4)

 
2
Φn+1 (z) = 1 − |Φn+1 (0)| zΦn (z) + Φn+1 (0)Φ∗n+1 (z), n ≥ 0, (5)

where Φ∗n (z) = z n Φ̄n (z −1 ) is the so-called reversed polynomial and the complex
numbers {Φn (0)}n≥1 are known as Verblunsky (Schur, reflection) parameters.
It is important to notice that in the positive definite case we get |Φn (0)| < 1,
n ≥ 1, and
Sn = hΦn , Φn iL > 0, n ≥ 0.
Moreover, we have
det Tn
Sn = , n ≥ 1, S0 = c0 , T−1 ≡ 0. (6)
det Tn−1

The n-th kernel polynomial Kn (z, y) associated with {Φ}n≥0 is defined by

n
X Φj (y)Φj (z) Φ∗n+1 (y)Φ∗n+1 (z) − Φn+1 (y)Φn+1 (z)
Kn (z, y) = = , (7)
j=0
Sj Sn+1 (1 − yz)

and the right hand side is known in the literature as Christoffel-Darboux formula
and it holds if ȳz 6= 1. It satisfies the so called reproducing property
Z
Kn (z, y)p(z)dσ(z) = p(y), (8)
T

(i,j)
for every polynomial p of degree at most n. Kn (z, y) will denote the i−th
and j−th partial derivative of Kn (z, y) with respect to z and y, respectively.
Notice that we have Φ∗n (z) = Sn Kn (z, 0), n ≥ 0.

2
 Furthermore, in terms of the moments, an analytic function can be defined
by
∞
X
F (z) = c0 + 2 c−k z k . (9)
k=1

If L is a positive definite functional, then (9) is analytic in D and its real part
is positive in D. In such a case, (9) is called a Carathéodory function, and can
be represented by the Riesz-Herglotz transform
Z iθ
e +z
F (z) = iθ
dσ(θ),
T e −z

where σ is the positive measure associated with L. By extension, for a quasi-

definite linear functional, (9) will denote its corresponding Carathodory func-
tion.
On the other hand, given a positive, nontrivial Borel measure µ supported
in [−1, 1], we can define a positive, nontrivial Borel measure σ supported in
[−π, π] in such a way that if dµ(x) = ω(x)dx, then
1
dσ(θ) = ω(cos θ)| sin θ|dθ. (10)
2
There exists a relation between the corresponding families of orthogonal poly-
nomials (see [6]).
On the other hand, since the moments {cn }n≥0 are real (see [6]), F (z), the
Carathodory function associated with σ, has real coefficients. Therefore, we
have
ReF (eiθ ) = ReF (ei(2π−θ) ),
and then dσ(θ) + dσ(2π − θ) = 0. Thus, there exists a simple relation between
the Stieltjes
P∞ function (the real line analog of the Carathdory functions, given by
S(x) = n=0 µn x−(n+1) , where {µn }n≥0 are the moments associated with the
measure on the real line) and the Carathodory function associated with µ and
σ, respectively, given by (see [9])
1
1 − z2 1 − z2
Z
dµ(t)
F (z) = = S(x), (11)
2z −1 x−t 2z
−1 √
where x = z+z2 , z = x + x2 − 1. In the literature, this relation is known
as the Szegő transformation. Conversely, if σ is a positive, nontrivial Borel
measure with support in the unit circle such that its moments are real, then
there exists a positive, nontrivial Borel measure µ, supported in [−1, 1], such
that (10) holds. This is called the inverse Szegő transformation.
Given a measure σ supported on the unit circle, the perturbations
(1) dσ̃C = |z − ξ|2 dσ, |z| = 1, ξ ∈ C,
−1
(2) dσ̃U = dσ + mc δ(z − ξ) + m̄c δ(z − ξ ), ξ ∈ C − {0}, mc ∈ C,
dσ −1
(3) dσ̃G = |z−ξ|2 + mc δ(z − ξ) + m̄c δ(z − ξ ), ξ ∈ C − {0}, mc ∈ C, |ξ| =
6 1,

3
are called Christoffel, Uvarov, and Geronimus transformations, respectively.
They are the unit circle analogue of the Christoffel, Uvarov and Geronimus
transformations on the real line (see [12]). In general, a linear spectral trans-
formation of a Stieltjes function is another Stieltjes function S̃(x) that has the
form
A(x)S(x) + B(z)
S̃(x) = ,
D(z)
where A, B and D are polynomials in x. The three transformations defined
above are important due to the fact that any linear spectral transformation
of a given Stieltjes function (i.e. for any polynomials A, B and D) can be
obtained as a combination of Christoffel and Geronimus transformations (see
[12]). A similar result holds for linear spectral transformations of Carathéodory
functions, which are defined in a similar way (see [4]).
In [2], the authors studied the perturbation associated with the linear func-
tional Z
dz
hp(z), q(z)iL̃ := hp(z), q(z)iL + m p(z)q(z) , (12)
T 2πiz
where m ∈ R, p, q ∈ P, and L is (at least) a quasi-definite Hermitian linear
functional defined in the linear space of Laurent polynomials. Notice that all
moments associated with L̃ are equal to the moments associated with L, except
for the first moment, which is c̃0 = c0 + m. The corresponding Toeplitz matrix
T̃ is the result of adding m to the main diagonal of T. Later on, the linear
functional
hp(z), q(z)iLj = hp(z), q(z)iL + mhz j p(z), q(z)iLθ + m̄hp(z), z j q(z)iLθ , (13)
where j ∈ N is fixed and h·, ·iLθ is the bilinear functional associated with the
normalized Lebesgue measure on the unit circle was studied in [3]. It is easily
seen that the moments associated with Lj are equal to those of L, except for
the moments of order j and −j, which are perturbed by adding m and m̄, re-
spectively. In other words, the corresponding Toeplitz matriz is perturbed on
the j-th and −j-th subdiagonals. In both cases, the authors obtained the reg-
ularity conditions for such a linear functional and deduced connection formulas
between the corresponding orthogonal sequences.
Assuming that both L and Lj are positive definite, the perturbation (13)
can be expressed in terms of the corresponding measures as
dθ dθ
dσ̃ = dσ + mz j + m̄z −j . (14)
2π 2π
On the other hand, the connection between the measure (14) and its corre-
sponding measure supported in [−1, 1] via the inverse Szegő transformation was
analyzed in [5], and it is deduced that the perturbed moments on the real line
depend on the Chebyshev polynomials of the first kind.

In this contribution, we will extend those results to the case where a pertur-
bation of a finite number of moments is introduced in (13). In Section 2, neces-
sary and sufficient conditions for the regularity of the perturbed functional are
obtained, as well as a connection formula that relates the corresponding families
of monic orthogonal polynomials. For the positive definite case, the study of
the perturbation through the inverse Szegő transformation will be analyzed in
Section 3. An illustrative example will be presented in Section 4.

4
2 A perturbation on a finite number of moments
associated with a linear functional L
Let L be a quasi-definite linear functional on the linear space of Laurent poly-
nomials, and let {cn }n∈Z be its associated sequence of moments.
Definition 2.1. Let Ω be a finite set of non negative integers. The linear
functional LΩ is defined such that the associated bilinear functional satisfies
X
hp(z), q(z)iLΩ = hp(z), q(z)iL + (mr hz r p(z), q(z)iLθ + m̄r hp(z), z r q(z)iLθ ) ,
r∈Ω
(15)
where mr ∈ C, p, q ∈ P, and h·, ·iLθ is the bilinear functional associated with
the normalized Lebesgue measure in the unit circle.
Notice that, from (15), one easily sees that


 cn , if n∈/ Ω,



c̃n = hLΩ , z n i = hz n , 1iLΩ = c−n + m−n , if n ∈ Ω and n ∈ Z− , (16)



/ Z− .

cn + m̄n , if n ∈ Ω and n ∈


In other words, LΩ represents an additive perturbation of the moments cr and

c−r of L, with r ∈ Ω. The rest of the moments remain unchanged. This is, the
Toeplitz matrix associated with LΩ is
0 · · · m̄r 0 · · ·
 
 .. 
 .
 0 · · · m̄r · · · 
T
e =T+
X .. .. .. ..  ,
 mr
 . . . . 

r∈Ω  0 mr · · · 0 ···  

.. .. .. .. ..
. . . . .
| {z }
mr is on the r−th subdiagonal

and therefore LΩ is also Hermitian. Moreover, if L is a positive definite func-

tional, then the above perturbation can be expressed in terms of the correspond-
ing measures as
X 
r dθ −r dθ
dσ̃Ω = dσ + mr z + m̄r z
2π 2π
r∈Ω
(17)
X dθ
= dσ + 2 Re(mr z r ) .
2π
r∈Ω

On the other hand, if FΩ (z) is the Carathéodory function associated with LΩ ,

then X
FΩ (z) = F (z) + 2 mr z r , (18)
r∈Ω

which is a linear spectral transformation of F (z).

The following notation will be used hereinafter:

5
 • A(s1 ,s2 ;l1 ,l2 ;r) will denote a (s2 − s1 + 1) × (l2 − l1 + 1) matrix whose entries
are a(s,l)r , where s1 ≤ s ≤ s2 and l1 ≤ l ≤ l2 ,. For instance,
 
a(2,4)6 a(2,5)6
A(2,3;4,5;6) = .
a(3,4)6 a(3,5)6

(0) (n−1) (0) (n−1)

• Ψn (0) = [ψn (0), · · · , ψn (0)]T and Φn (0) = [Φn (0), · · · , Φn (0)]T .

• In will denote the n × n identity matrix.

(0,−2)
• Derivatives of negative order are defined as zero. For instance, Kn (z, y) ≡
0.
Necessary and sufficient conditions for the regularity of LΩ , as well as the rela-
tion between the corresponding sequences of orthogonal polynomials, are given
in the next result.

Proposition 2.2. Let L be a quasi-definite linear functional and let {Φn }n≥0 be
its associated monic orthogonal polynomials sequence (MOPS). The following
statements are equivalent 1
1. LΩ is a quasi-definite linear functional.

2. The matrix In + r∈Ω Lrn is nonsingular, and

P

(n−r)
X X Φn (0)
S̃n = Sn + (Qn Wn )T Yrn + m̄r 6= 0, n ≥ 0, (19)
(n − r)!
r∈Ω r∈Ω

with  
(r)
mr Φn0!r!(0)
..
 
 

 . 

(2r−1)
 Φn (0) 

 m r (r−1)!(2r−1)!


 
 
(2r) (0)
mr Φr!(2r)!(0)
+ m̄r Φnr!0!(0)
 n

 
 
..
Yrn = 
 
. , (20)
 
 (n) (n−2r) 
 m Φ n (0) Φn (0)
 r (n−r)!(n)! + m̄r (n−r)!(n−2r)!


 
 
 (n−2r+1)

Φn (0)
m̄r (n−r+1)!(n−2r+1)!
 
 
..
 
 

 . 

(n−r−1)
Φn (0)
m̄r (n−1)!(n−r−1)!
X
Qn = (In + Lrn )−1 ,
r∈Ω

1 Thisresult generalizes the case when Ω has a single element k 6= 0 (see [3]) and the case
Ω = {0} (see [2]).

6
 X
Wn = Φn (0) − m̄r n!C(0,n−1;n,n;r) ,
r∈Ω
 (0,r)

Kn−1 (z,0)
 mr 0!r! 
 .. 

 . 

(0,2r−1)
 Kn−1 (z,0) 

 mr (r−1)!(2r−1)! 

 
 (0,2r) (0,0)

 Kn−1 (z,0) Kn−1 (z,0) 

 m r r!(2r)! + m̄ r r!0!


..
Krn−1 (z, 0) = 
 
. , (21)
 
(0,n−1) (0,n−2r−1)
 m Kn−1 (z,0) + m̄ Kn−1 (z,0)
 

 r (n−r−1)!(n−1)! r (n−r−1)!(n−2r−1)! 
 
 
 (0,n−2r) 
Kn−1 (z,0)

 m̄r (n−r)!(n−2r)! 

..
 
 
 . 
 (0,n−r−1) 
Kn−1 (z,0)
m̄r (n−1)!(n−r−1)! .
and
 
mr A(0,r−1;0,r−1;r) B(0,r−1;r,n−r−1;r) m̄r C(0,r−1;n−r,n−1;r)
Lrn =  mr A(r,n−r−1;0,r−1;r) B(r,n−r−1;r,n−r−1;r) m̄r C(r,n−r−1;n−r,n−1;r)  ,
mr A(n−r,n−1;0,r−1;r) B(n−r,n−1;r,n−r−1;r) m̄r C(n−r,n−1;n−r,n−1;r)

where the entries of the matrices A, B and C are given by

(s,l+r)
Kn−1 (0, 0)
a(s,l)r = ,
l!(l + r)!
(s,l+r) (s,l−r)
Kn−1 (0, 0) K (0, 0)
b(s,l)r = mr + m̄r n−1 ,
l!(l + r)! l!(l − r)!
(s,l−r)
Kn−1 (0, 0)
c(s,l)r = .
l!(l − r)!

Furthermore, if {ψn }n≥0 denotes the MOPS associated with LΩ , then

(0,n−r)
X X Kn−1 (z, 0)
ψn (z) = Φn (z) − (Qn Wn )T Krn−1 (z, 0) − m̄r , (22)
(n − r)!
r∈Ω r∈Ω

for every n ≥ 1.
Remark 2.3. Notice that Qn and Lrn are n × n matrices, whereas Yrn , Wn and
Krn−1 (z, 0) are n-th dimensional column vectors.
Proof. Assume LΩ is a quasi-definite linear functional, and denote by {ψn }n≥0
its associated MOPS. Let us write
n−1
X
ψn (z) = Φn (z) + λn,k Φk (z), (23)
k=0

7
where, for 0 ≤ k ≤ n − 1,
hψn (z), Φk (z)iL
λn,k =
Sn
dy dy
y r ψn (y)Φk (y) 2πiy y −r ψn (y)Φk (y) 2πiy
P R P R
hψn (z), Φk (z)iLΩ − r∈Ω mr T
− r∈Ω m̄r T
=
Sk
X mr Z dy X m̄r
Z
dy
=− y r ψn (y)Φk (y) − y −r ψn (y)Φk (y) ,
Sk T 2πiy Sk T 2πiy
r∈Ω r∈Ω

and notice that hψn (z), Φk (z)iL 6= 0 (in general), and hψn (z), Φk (z)iLΩ = 0 for
n > k.
Substituting in (23) and using (7), we get
Z n−1
!
X X Φk (z)Φk (y) dy
r
ψn (z) = Φn (z) − mr y ψn (y)
T Sk 2πiy
r∈Ω k=0
Z n−1
!
X X Φk (z)Φk (y) dy
− m̄r y −r ψn (y)
T Sk 2πiy
r∈Ω k=0
 Z Z 
X
r dy −r dy
= Φn (z) − mr y ψn (y)Kn−1 (z, y) − m̄r y ψn (y)Kn−1 (z, y) .
T 2πiy T 2πiy
r∈Ω

From the power series expansion of ψn (y) and Kn−1 (z, y), we have
n (l)
X ψn (0)
y r ψn (y) = y r yl
l!
l=0
n+r
X ψn(l−r) (0)
l
= y,
(l − r)!
l=r

and for |y| = 1,

n−1 (0,l)
X Kn−1 (z, 0) 1
Kn−1 (z, y) = ,
l! yl
l=0
dy
y r−t 2πiy
R
and since T
= 1 if r = t and zero otherwise, we arrive at
Z n+r
X ψn(l−r) (0) n−1 (0,l)
X Kn−1 (z, 0) 1 dy
Z
dy
y r ψn (y)Kn−1 (z, y) = yl
T 2πiy T (l − r)! l! y l 2πiy
l=r l=0
n−r−1 (l) (0,l+r)
X ψn (0) Kn−1 (z, 0)
= .
(l)! (l + r)!
l=0

Similarly,
n−r (0,l)
X ψn(l+r) (0) Kn−1 (z, 0)
Z
dy
y −r ψn (y)Kn−1 (z, y) =
T 2πiy (l + r)! l!
l=0
n (l) (0,l−r)
X ψn (0) Kn−1 (z, 0)
= .
(l)! (l − r)!
l=r

8
As a consequence, we get
n−r−1 (0,l+r) n (0,l−r)
!
(l) X ψn(l) (0) Kn−1 (z, 0)
X X ψn (0) Kn−1 (z, 0)
ψn (z) = Φn (z)− mr − m̄r ,
(l)! (l + r)! (l)! (l − r)!
r∈Ω l=0 l=r
(24)
which after a reorganization of the terms becomes
r−1 (l) (0,l+r) n (0,l−r)
!
(l)
X X ψn (0) Kn−1 (z, 0) X ψn (0) Kn−1 (z, 0)
ψn (z) = Φn (z) − mr + m̄r
l! (l + r)! (l)! (l − r)!
r∈Ω l=0 l=n−r
n−r−1 (0,l+r) (0,l−r)
!!
(l)
X X ψn (0) Kn−1 (z, 0) Kn−1 (z, 0)
− mr + m̄r .
l! (l + r)! (l − r)!
r∈Ω l=r
(25)
(l)
In order to find the constant values ψn (0), we take s derivatives, 0 ≤ s ≤ n,
with respect to the variable z and evaluate at z = 0 to obtain the (n+1)×(n+1)
linear system
r−1 (l) (s,l+r) n (s,l−r)
!
(l)
X X ψn (0) Kn−1 (0, 0) X ψn (0) Kn−1 (0, 0)
ψn(s) (0) = Φ(s)
n (0) − mr + m̄r
l! (l + r)! (l)! (l − r)!
r∈Ω l=0 l=n−r
n−r−1 (s,l+r) (s,l−r)
!!
(l)
X X ψn (0) Kn−1 (0, 0) Kn−1 (0, 0)
− mr + m̄r .
l! (l + r)! (l − r)!
r∈Ω l=r

If we denote
(s,l+r)
Kn−1 (0, 0)
a(s,l)r = ,
l!(l + r)!
(s,l+r) (s,l−r)
Kn−1 (0, 0) K (0, 0)
b(s,l)r = mr + m̄r n−1 ,
l!(l + r)! l!(l − r)!
(s,l−r)
Kn−1 (0, 0)
c(s,l)r = ,
l!(l − r)!
then the linear system becomes
r−1 n−r−1 n
!
X X X X
ψn(s) (0) = Φ(s)
n (0)− mr a(s,l)r ψn(l) (0) + b(s,l)r ψn(l) (0) + m̄r c(s,l)r ψn(l) (0) .
r∈Ω l=0 l=r l=n−r

Notice that the last equation (i.e., when s = n) gives no information, since
(n) (n)
a(n,l)r = b(n,l)r = c(n,l)r = 0 and ψn (0) = Φn (0) = n!. As a consequence, the
(n + 1) × (n + 1) linear system can be reduced to an n × n linear system that
can be expressed in matrix form as
X X
(In + Lrn )Ψn (0) = Φn (0) − m̄r n!C(0,n−1;n,n;r)
r∈Ω r∈Ω (26)
= Wn .

9
Since LΩ is quasi-definite, it has a unique MOPS and thereforePthe linear
system has a unique solution. As a consequence, the matrix In + r∈Ω Lrn is
nonsingular and X
Ψn (0) = (In + Lrn )−1 Wn . (27)
r∈Ω
This is, we have
!T (0,n−r)
X X X Kn−1 (z, 0)
ψn (z) = Φn (z)− WTn (In + Lrn )−1 Krn−1 (z, 0)− m̄r ,
(n − r)!
r∈Ω r∈Ω r∈Ω
(28)
which is (22). On the other hand, for n ≥ 0,
S̃n = hψn (z), ψn (z)iLΩ = hψn (z), Φn (z)iLΩ
X
= hψn (z), Φn (z)iL + (mr hz r ψn (z), Φn (z)iLθ + m̄r hψn (z), z r Φn (z)iLθ )
r∈Ω
 
Z X n (l) n (l)
X ψn (0) l+r X Φn (0) −l dz 
= Sn + mr z z
T l=0 l! l! 2πiz
r∈Ω l=0
 
Z X n (l) n (l)
X ψn (0) l−r X Φn (0) −l dz 
+ m̄r z z
T l=0 l! l! 2πiz
r∈Ω l=0
  
r−1 (l) (l+r) n−r (l) (l+r) (l−r)
X X ψn (0) Φn (0) X ψn (0)  Φn (0) Φn (0) 
= Sn + mr + mr + m̄r
l! (l + r)! l! (l + r)! (l − r)!
r∈Ω l=0 l=r
 
n−1
X ψn(l) (0) Φn(l−r) (0) (n−r)
X X Φn (0)
+ m̄r + m̄r .
l! (l − r)! (n − r)!
r∈Ω l=n−r+1 r∈Ω

Using (20) and (27), we get

!T
(n−r)
X X X Φn (0)
S̃n = Sn + WTn (In + Lrn )−1 Yrn + m̄r .
(n − r)!
r∈Ω r∈Ω r∈Ω
P r
Conversely, assume In + is nonsingular for every n ≥ 1 and define
r∈Ω Ln
{ψn }n≥0 as in (22). For 0 ≤ k ≤ n − 1, we have
*n−r−1 (0,l+r)
+ n−r−1
* (0,l+r)
+
X ψn(l) (0) Kn−1 (z, 0) X (l)
ψn (0) Kn−1 (z, 0)
, Φk (z) = , Φk (z)
l! (l + r)! l! (l + r)!
l=0 L l=0 L
n−r−1
* n−1
+
(l) (l+r)
X ψn (0) X Φt (z)Φt (0)
= , Φk (z)
l!(l + r)! t=0 St
l=0 L
n−r−1 (l) (l+r)
X ψn (0) Φk (0)
= ,
l! (l + r)!
l=0

and, similarly,
* n (0,l−r)
+ n (l−r)
X ψn(l) (0) Kn−1 (z, 0) X (l)
ψn (0) Φ (0)k
, Φk (z) = .
l! (l − r)! l! (l − r)!
l=j L l=r

10
In addition,
Z
dz
hz r ψn (z), Φk (z)iLθ = z r ψn (z)Φk (z)
T 2πiz
n
! k 
Z (l)
X ψn (0)z l X Φ(l) −l
n (0)z  dz
= zr 
T l! l! 2πiz
l=0 l=0

n−r−1 (l) (l+r)

X ψn (0) Φk (0)
= ,
l! (l + r)!
l=0

and also
n (l) (l−r)
X ψn (0) Φ (0)
hψn (z), z r Φk (z)iLθ = k
.
l! (l − r)!
l=r
Thus, for 0 ≤ k ≤ n−1, and taking into account (24) and the previous equations,
we have
X
hψn (z), Φk (z)iLΩ = hψn (z), Φk (z)iL + (mr hz r ψn (z), Φk (z)iLθ + m̄r hψn (z), z r Φk (z)iLθ )
r∈Ω
*n−r−1 (0,l+r)
+
X X ψn(l) (0) Kn−1 (z, 0)
= hΦn (z), Φn (z)iL − mr , Φk (z)
(l)! (l + r)!
r∈Ω l=0 L
* n (0,l−r)
+
(l)
X X ψn (0) Kn−1 (z, 0)
− m̄r , Φk (z)
(l − r)! (l)!
r∈Ω l=r L
   
n−r−1 (l)
X ψn (0) Φ (l+r) n (l) (l−r)
X
k (0)  X X ψn (0) Φk (0) 
+ mr  + m̄r 
l! (l + r)! l! (l − r)!
r∈Ω l=0 r∈Ω l=r

= 0.
On the other hand,
Sen = hψn (z), Φn (z)iLΩ
X
= hψn (z), Φn (z)iL + (mr hz r ψn (z), Φn (z)iLθ + m̄r hψn (z), z r Φn (z)iLθ )
r∈Ω
 
Z X n (l) n (l)
X ψn (0) l+r X Φn (0) −l dz 
= Sn + mr z z
T l=0 l! l! 2πiz
r∈Ω l=0
 
Z X n (l) n (l)
X ψn (0) l−r X Φn (0) −l dz 
+ m̄r z z
T l! l! 2πiz
r∈Ω l=0 l=0

(n−r)
X X Φn (0)
= Sn + ΨTn (0)Yrn (0) + m̄r ,
(n − r)!
r∈Ω r∈Ω

which is different from 0 by assumption. Therefore, LΩ is quasi-definite.

Notice that if rm = min{r : r ∈ Ω}, then from Proposition 2.2 we conclude
(0,n−r )
that if n < rm then Krn−1
m
is the zero vector, Kn−1 m (z, 0) = 0 and according
to (22) we have ψn (z) = Φn (z). This means that the only affected polynomials
are those with degree n ≥ rm .

11
3 Finite moments perturbation through the in-
verse Szegő transformation
Let σ be a positive measure supported on the unit circle such that its corre-
sponding moments {cn }n∈Z are real. Assume also that the perturbed measure
σ̃Ω , defined by (17), is also positive and that mr with r ∈ Ω is real, so that the
moments associated with σ̃Ω are also real. Our goal in this Section is to deter-
mine the relation between the positive Borel measures µ and µ̃Ω , supported in
[−1, 1], which are associated with σ and σ̃Ω , respectively, via the inverse Szegő
transformation. This relation will be stated in terms of the corresponding mea-
sure and their sequences of moments.

Proposition 3.1. Let σ be a positive nontrivial Borel measure with real mo-
ments supported in the unit circle, and let µ be its corresponding measure sup-
ported in [−1, 1], obtained through the inverse Szegő transformation. Let {cn }n∈Z
and {µn }n≥0 be their corresponding sequences of moments. Assume that σ̃Ω ,
defined by (17) with r ∈ Ω and mr ∈ R, is positive. Then, the measure µ̃Ω ,
obtained by applying the inverse Szegő transformation to σ̃Ω , is given by
2X dx
dµ̃Ω = dµ + mr Tr (x) √ , (29)
π 1 − x2
r∈Ω

where Tr (x) := cos(rθ) is the r-th degree Chebyshev polynomial of the first kind.
Its corresponding sequence of moments is

 µn , if 0 ≤ n < rm ,
µ̃n = (30)
µn + π2 r∈Ω mr B(n, r), if
P
rm ≤ n,


with
πr
P[r/2]  (−1)k (r−k−1)!(2)r−2k 
• B(n, r) = 2 k=0 k!(r−2k)! , if r + n − 2k = 0,

• B(n, r) = 0, if r + n is odd,
 
[r/2] (−1) (r−k−1)!(2)r−2k
k Q(r+n−2k)/2 r+n−2k−(2i−1)
• B(n, r) = πr
P
2 k=0 k!(r−2k)! i=1 r+n−2k−2(i−1) , if
r + n is even.
Proof. Notice that, setting z = eiθ , x = cos θ, and taking into account that
the inverse Szegő transformation applied to the normalized Lebesgue measure
dθ/2π yields the Chebyshev measure of the first kind π√dx1−x2
, the measure µ̃Ω
obtained by applying the inverse Szegő transformation to σ̃Ω is given by
X p p  dx
dµ̃Ω = dµ + mr (x + i 1 − x2 )r + mr (x + i 1 − x2 )−r √
r∈Ω
π 1 − x2
X dx
= dµ + (mr (cos rθ + i sin rθ) + mr (cos rθ − i sin rθ)) √
r∈Ω
π 1 − x2
2X Tr (x)dx
= dµ + mr √ .
π 1 − x2
r∈Ω

12
Notice that a measure that changes its sign in the interval [−1, 1] is added to
dµ. Then, the moments associated with µ̃Ω are given by
Z 1 Z 1
2X Tr (x)dx
µ̃n = xn dµ̃Ω (x) = µn + mr xn √ .
−1 π −1 1 − x2 r∈Ω

As a consequence, by the orthogonality of Tr (x), we obtain for the n−th mo-

ments with n ∈/Ω
if 0 ≤ n < rm ,

 µn ,
µ̃n = R1 (31)
µn + π2 r∈Ω mr −1 xn T√r1−x
(x)dx
P
, if r ≤ n.

2 m

Furthermore (see [8]), we have

[r/2]
r X (−1)k (r − k − 1)!(2x)r−2k
Tr (x) = , r = 1, 2, 3, ...,
2 k!(r − 2k)!
k=0

where [r/2] = r/2 if r is even and [r/2] = (r − 1)/2 if r is odd. Therefore,

1 1 [r/2]
Z
dx r
Z X (−1)k (r − k − 1)!(2x)r−2k dx
xn Tr (x) √ = xn √
−1 1−x 2 2 −1 k!(r − 2k)! 1 − x2
k=0
[r/2]
r X (−1)k (r − k − 1)!(2)r−2k 1 r+n−2k dx
Z
= x √ ,
2 k!(r − 2k)! −1 1 − x2
k=0

and, since


 π, if k = 0,


Z 1 
dx 
0, if k is odd,
xk √ =
−1 1 − x2 

 Q 
 k/2 k−(2i−1)
i=1 k−2(i−1) π, if k is even,



(31) becomes (30).

From the previous proposition we can conclude that a perturbation of the
moments cr and c−r with r ∈ Ω, associated with a measure σ supported in
the unit circle, results in a perturbation, defined by (30), of the moments µn ,
n ≥ rm , associated with a measure µ supported in [−1, 1], when both measures
are related through the inverse Szegő transformation.

4 Example
Let L be the Christoffel transformation of the normalized Lebesgue measure
with parameter ξ = 1, i.e., hp(z), q(z)iL = h(z − 1)p(z), (z − 1)q(z)iLθ , and let
Ω = {1, 2}. Then,
hp(z), q(z)iL{1,2} = h(z − 1)p(z), (z − 1)q(z)iLθ + m1 hzp(z), q(z)iLθ + m̄1 hp(z), zq(z)iLθ
+ m2 hz 2 p(z), q(z)iLθ + m̄2 hp(z), z 2 q(z)iLθ ,
(32)

13
i.e., the moments of order 1 and 2 are perturbed. Since the sequence {z n }n≥0
is orthogonal with respect to Lθ , the MOPS associated with h(z − 1)p(z), (z −
1)q(z)iLθ is given by (see [1])
 
n
1  n+1 1 X j
Φn (z) = z − z , n ≥ 1,
z−1 n + 1 j=0

or, equivalently,
n
Φn (z) = z n +
Φn−1 (z), n ≥ 1,
n+1
and its corresponding reversed polynomial is
 
n
1  1 X
Φ∗n (z) = 1− z j+1  , n ≥ 1.
1−z n + 1 j=0

Furthermore, if 0 ≤ s ≤ n, we have

(s + 1)!
Φ(s)
n (0) = ,
(n + 1)

s!(n + 1 − s)
Φn∗(s) (0) = ,
(n + 1)
n+2
Sn = ,
n+1
and if 0 ≤ t, s ≤ n − 1,
n−1
(0,t)
X (t + 1)!
Kn−1 (z, 0) = Φp (z),
p=t
p+2

n−1
(s,t)
X (t + 1)!(s + 1)!
Kn−1 (0, 0) = .
(p + 1)(p + 2)
p=max{s,t}

As a consequence, we have
n−1
(s + 1)!(l + r + 1) X 1
a(s,l)r = ,
l! (p + 1)(p + 2)
p=max{s,l+r}

n−1
(s + 1)!(l − r + 1) X 1
c(s,l)r = ,
l! (p + 1)(p + 2)
p=max{s,l−r}
 
n−1 n−1
(s + 1)!  X l+r+1 X l−r+1 
b(s,l)r = mr + m̄r .
l! (p + 1)(p + 2) (p + 1)(p + 2)
p=max{s,l+r} p=max{s,l−r}

We now proceed to obtain the MOPS associated with L{1,2} , denoted by

(s+1)! 2(s+1)!
{ψn }n≥0 . Notice that we have c(s,n)1 = n!(n+1) , c(s,n)2 = n!(n+1) if 0 ≤ s ≤ n−2

14
 n−1
and c(n−1,n)2 = n(n+1) , and thus
 (0) 
Φn (0)
   
c(0,n)1 c(0,n)2
(1)
Φn (0)  c(1,n)1   c(1,n)2
  
     
 .. 
 − m̄1 n!  .. ..
Wn =   − m̄2 n! 
  
 .  
 .   . 

 (n−2)   c(n−2,n)   c(n−2,n)
 Φn (0)  1 2

(n−1)
Φn (0) c(n−1,n)1 c(n−1,n)2
   
1! 1!
 2!   2! 
1 − m̄1 
..
 2m̄2  .

=

.

−
 .
. .

n+1  n+1
  
 
 (n − 1)!   (n − 1)! 
n! n! n−1
2n

On the other hand, for n ≥ 2 we have

 Pn−1 Φp (z) 
2m1 p=1 p+2
 
 Pn−1 Φp (z) Pn−1 Φp (z) 
3m1 p=2 p+2 + m̄1 p=0 p+2
 
 
..
 
K1n−1 (z, 0) =  ,
 
 . 
 nm1 Pn−1 Φp (z) m1 Pn−1 Φp (z)
 (n−2)! p=n−1 p+2 + (n−3)! p=n−3 p+2


 
 
m̄1 Pn−1 Φp (z)
(n−2)! p=n−2 p+2
 
m1 a(0,0)1 b(0,1)1 ··· b(0,n−2)1 m̄1 c(0,n−1)1

 m1 a(1,0)1 b(1,1)1 ··· b(1,n−2)1 m̄1 c(1,n−1)1 

L1n = .. .. .. .. ..
,
 
 . . . . . 
 m1 a(n−2,0)
1
b(n−2,1)1 ··· b(n−2,n−2)1 m̄1 c(n−2,n−1)1 
m1 a(n−1,0)1 b(n−1,1)1 ··· b(n−1,n−2)1 m̄1 c(n−1,n−1)1
and for n ≥ 3,
Pn−1 Φp (z)
 
3m2 p=2 p+2
 
 
Pn−1 Φp (z)
4m2
 

 p=3 p+2 

 
Pn−1 Φp (z) Pn−1 Φp (z)
 
5m2 m̄2

 2 p=4 p+2 + 2 p=0 p+2


2
 .. 
Kn−1 (z, 0) = 
 .
,

 nm2 Pn−1 Φp (z) m̄2 Pn−1 Φp (z) 

 (n−3)! p=n−1 p+2 + (n−3)(n−5)! p=n−5 p+2


 
 
 m̄2 Pn−1 Φp (z) 
 (n−2)(n−4)! p=n−4 p+2 
 
 
m̄2 Pn−1 Φp (z)
(n−1)(n−3)! p=n−3 p+2

15
  
m2 a(0,0)2 m2 a(0,1)2 b(0,2)2 ··· b(0,n−3)2 m̄2 c(0,n−2)2 m̄2 c(0,n−1)2

 m2 a(1,0)2 m2 a(1,1)2 b(1,2)2 ··· b(1,n−3)2 m̄2 c(1,n−2)2 m̄2 c(1,n−1)2 


 m2 a(2,0)2 m2 a(2,1)2 b(2,2)2 ··· b(2,n−3)2 m̄2 c(2,n−2)2 m̄2 c(2,n−1)2 

2
Ln =  .. .. .. .. .. .. ..
,
 
 . . . . . . . 
 m2 a(n−3,0)
 2
m2 a(n−3,1)2 b(n−3,2)2 ··· b(n−3,n−3)2 m̄2 c(n−3,n−2)2 m̄2 c(n−3,n−1)2 

 m2 a(n−2,0)
2
m2 a(n−2,1)2 b(n−2,2)2 ··· b(n−2,n−3)2 m̄2 c(n−2,n−2)2 m̄2 c(n−2,n−1)2 
m2 a(n−1,0)2 m2 a(n−1,1)2 b(n−1,2)2 ··· b(n−1,n−3)2 m̄2 c(n−1,n−2)2 m̄2 c(n−1,n−1)2
For illustrative purposes, we compute the first polynomials of the sequence:

• Degree one:
(0,0)
m̄1 K0 (z, 0)
ψ1 (z) = Φ1 (z) −
0!
(1 − m̄1 )
=z+ Φ0 (z),
2
since K10 (z, 0) = 0, K20 (z, 0) = 0 and Φ1 (z) = z + 21 Φ0 (z).
• Degree two:

ψ2 (z) = Φ2 (z)

m1 2m̄1
−1 T
 T  T ! +1
1 − m̄1 1! 2m̄2 1! 3 3
− −
  
3 2! 3 1/2 
2m1 m̄1

3 3 +1
2m1
 
3 Φ1 (z)  
 2m̄1 Φ0 (z) Φ1 (z)
 − Φ1 (z) − m̄2 +

3 2 3
 
Φ0 (z) Φ1 (z)
m̄1 2 + 3
2m1
  
3 Φ1 (z)
m̄1
− 2m

 T +1 1
1 − m̄1  1! 3 3
= z2 + 2Φ1 (z) − A 
 
3 2!  
Φ0 (z) Φ1 (z)
 
− 2m̄
3
1 m1
3 +1 m̄1 + 2 3
2m1
  
3 Φ1 (z)
m̄1
− 2m
 
 T +1 1
 Φ0 (z) Φ1 (z) 2 1! 3 3
− m̄2  + − A    ,
 
2 3 3 1/2  
Φ0 (z) Φ1 (z)
− 2m̄
3
1 m1
3 +1 m̄1 2 + 3

since K21 (z, 0) = 0, Φ2 (z) = z 2 + 32 Φ1 (z) and A = 1

det(I2 +L12 +L22 )
=
9
|m1 +3|2 −4|m1 |2 .

16
 • In general, the n-th degree polynomial is
  T  T 
1! 1!
2! 2!
      " #T
   2
 1 − m̄1  .
 2m̄ 2

.  X
r −1
ψn (z) = Φn (z) −   n+1 
 ..  −
  .. 
  (In + Ln )
n+1

   
r=1
(n − 1)! (n − 1)!
     
 
n! n!(n − 1)/2n
2
!  
X
r nm̄1 Φn−2 (z) Φn−1 (z)
× Kn−1 (z, 0) − Φn−1 (z) − m̄2 (n − 1) +
r=1
n+1 n n+1
  T 
1!
2!
   " #T !
 2 2 
n 1 − m̄1   
.
 X
r −1
X
r

=z + nΦn−1 (z) − 
 ..  (In +

Ln ) Kn−1 (z, 0) 
n+1    
r=1 r=1

  (n − 1)!  

n!
 
Φn−1 (z) Φn−2 (z)
− m̄2 (n − 1) +
n+1 n
 T
1!
 2!  " 2
#T 2
!
2m̄2  .
 X X
+ ..  (In + Lrn )−1 Krn−1 (z, 0) ,
 
n+1


 (n − 1)!  r=1 r=1

n! n−1
2n

n
since Φn (z) = z n + n+1 Φn−1 (z).

On the other hand, assuming that the linear functional (32) is positive definite,
the associated measure is
dθ dθ dθ
dσ = |z − 1|2 + 2Re(m1 z) + 2Re(m2 z 2 ) ,
2π 2π 2π
and the corresponding moments are given by


 2, if n = 0,
 −1 + m̄1 , if n = 1,



−1 + M1 , if n = −1,

c̃n =

 m̄2 , if n = 2,
M2 , if n = −2,




0, in other case.


Thus, the perturbed Toeplitz matrix is

 
2 −1 + m̄1 m̄2 0 0 ···
 −1 + m1 2 −1 + m̄1 m̄2 0 ··· 
 
 m 2 −1 + m1 2 −1 + m̄1 m̄2 ··· 
T
e = ,


 0 m 2 −1 + m1 2 −1 + m̄1 ··· 
 0 0 m 2 −1 + m1 2 ··· 
.. .. .. .. ..
 
..
. . . . . .

17
i.e.,
q the first and second subdiagonals are perturbed. Furthermore, since dµ =
2 1−x
π 1+x dx is the measure obtained applying the inverse Szegő transformation
dθ
to dσ = |z − 1|2 2π , then according to (29) the measure supported in [−1, 1] is
r
2 1−x 2 1 2 1
dµ̃ = dx + m1 T1 (x) √ dx + m2 T2 (x) √ dx. (33)
π 1+x π 1−x 2 π 1 − x2
Then, according to (30), the perturbed moments associated with the measure
(33) are


 µn , if n = 0,




 Q n+1 n+1−(2i−1)
µ̃n = µn + 2m1 i=1 2
n+1−2(i−1) , if n is odd,



n+2
  n

 µn + 2m2 2 Q 2 n+2−(2i−1) − Q 2 n−(2i−1) ,

if n is even,

i=1 n+2−2(i−1) i=1 n−2(i−1)

q
2 1−x
where {µn }n≥0 are the moments associated with the measure dµ = π 1+x ,
and

 2, if n = 0,




 Q n+1 n+1−(2i−1)
µn = −2 i=1 2
n+1−2(i−1) , if n is odd,




 Q n2 n−(2i−1)
2 i=1 n−2(i−1) , if n is even.


References
[1] Bueno M.I. and Marcellán F., ”Polynomial perturbations of bilinear func-
tionals and Hessenberg matrices”, Linear Algebra Appl. 414 (2006), no. 1,
64-83.
[2] Castillo K., Garza L.E. and Marcelln F., ”Linear spectral transforma-
tions, Hessenberg matrices and orthogonal polynomials”, Rend. Circ. Mat.
Palermo. 82 (2010), no. 2, 3-26.

[3] Castillo K., Garza L.E. and Marcelln F., ”Perturbations on the subdiag-
onals of Toeplitz matrices”, Linear Algebra Appl. 434 (2011), no. 6, 1563-
1579.
[4] Castillo K. and Marcellán F., ”Generators of rational spectral transforma-
tions for nontrivial C-functions”, Math. Comp. 82 (2013), 1057-1068.

[5] Fuentes E. and Garza L.E., ”Analysis of perturbations of moments associ-

ated with orthogonality linear functionals through the Szegő transforma-
tion”, Rev. Integr. Temas Mat. 33 (2015), no. 1, 61-82.
[6] Garza L., Hernández J. and Marcellán F., ”Spectral transformations of
measures supported on the unit circle and the Szegő transformation”, Nu-
mer. Algorithms. 49 (2008), no. 1-4, 169-185.

18
 [7] Grenander U. and Szegő G., Toeplitz Forms and their Applications, second
ed., University of California Press, Berkeley 1958, Chelsea, New York, 1984.

[8] Kreyszig E., Introductory functional analysis with applications, John Wiley
& Sons Inc., New York, 1989.
[9] Peherstorfer F., ”A special class of polynomials orthogonal on the circle
including the associated polynomials”, Constr. Approx. 12 (1996), no 2,
161-185.

[10] Simon B., ”Orthogonal Polynomials on the unit circle, Parts 1 and 2”,
Amer. Math. Soc. Coll. Publi. Series. vol. 54, 2005.
[11] Szegő G., ”Orthogonal Polynomials”, Amer. Math. Soc. Colloq. Publ. Ser.
vol. 23, fourth ed., 1975.

[12] Zhedanov A., ”Rational spectral transformations and orthogonal polyno-

mials”, J. Comput. Appl. Math. 85 (1997), no 1, 67-86.

19