IOP PUBLISHING JOURNAL OF PHYSICS A: MATHEMATICAL AND THEORETICAL

J. Phys. A: Math. Theor. 42 (2009) 454014 (12pp) doi:10.1088/1751-8113/42/45/454014

The best constant of discrete Sobolev inequality

Atsushi Nagai1 , Yoshinori Kametaka2 and Kohtaro Watanabe3

1 College of Industrial Technology, Nihon University, 2-11-1 Shin-ei, Narashino 275-8576, Japan
2 Faculty of Engineering Science, Graduate School of Mathematical Sciences, Osaka University,
1-3 Machikaneyama, Toyonaka 560-8531, Japan
3 Department of Computer Science, National Defense University, Hashirimizu, Yokosuka

239-8686, Japan

E-mail: [email protected], [email protected] and [email protected]

Received 27 February 2009, in final form 27 June 2009

Published 27 October 2009
Online at stacks.iop.org/JPhysA/42/454014

Abstract
A discrete version of Sobolev inequalities in Hilbert spaces 2 and 2N , which
are equipped with an inner product defined by using 2Mth positive difference
operators, is presented. Their best constants are also found by means of the
theory of reproducing kernel and are given by a harmonic mean of the spectra
of the difference operator. Other expressions of the best constants are also
derived.

PACS numbers: 02.30Sa, 02.30Lt

Mathematics Subject Classification: 46E39, 41A44

1. Introduction

The best constant of the Sobolev inequality

uLq (RN )  C∇uLp (RN ) , u ∈ W 1,p (R N ),
was found by Talenti [6] in the case 1  p < N, q = Np/(N − p) and by Kametaka et al
[1, 3] in the case q = ∞, p = 2. In particular, we obtained the best constant of the Sobolev
inequalities which come from a physical background such as the string deflection problem [2]
and the ladder electric circuit [4].
In our previous work [5], we derived a discrete analogue of the Sobolev inequality
starting from the periodic boundary value problems for the 2Mth-order difference operator.
The best constant of the discrete Sobolev inequality is given by means of a discrete analogue
of Bernoulli polynomials and Riemann zeta functions. The purpose of this paper is to find
a discrete version of Sobolev inequalities and their best constants starting from 2Mth-order
difference equations equipped with M complex parameters z0 , . . . , zM−1 . We restrict ourselves
to a one-dimensional case N = 1.

1751-8113/09/454014+12$30.00 © 2009 IOP Publishing Ltd Printed in the UK 1

J. Phys. A: Math. Theor. 42 (2009) 454014 A Nagai et al

This paper is organized as follows. In section 2, we start with the simplest example
corresponding to the case M = 1, that is the second-order difference operator. We derive
the discrete analogue of the Sobolev inequality and find its best constant by using the
theory of reproducing kernels. In section 3, we generalize this result to the case M  2.
Section 4 is devoted to the derivation of the discrete Sobolev inequality corresponding to
periodic boundary problems for the same difference operator as in section 3. Finally in
section 5, we give concluding remarks.

2. Simple example

Let 2 be a set of all the infinite sequences u = t (. . . , u(i), . . .) (i ∈ Z) of complex numbers

∞ 
2 1/2
u(i) with finite 2 -norm u = i=−∞ |u(i)| < ∞. For any two elements u and v in
2 , we attach a usual unitary inner product:
∞

(u, v) = u(i)v(i).
i=−∞

With respect to this inner product, 2 is a Hilbert space.

We consider the shift operator L on 2 . L maps an element u = (. . . , u(i), . . .) to the
shifted element Lu = (. . . , u(i + 1), . . .), whose ith element is u(i + 1). If we introduce a
vector δ j (j ∈ Z) defined by

1 (i = 0),
δ j = (. . . , δ(i − j ), . . .), δ(i) =
0 (else),
the operator L is also given by the convolution by δ −1 , that is,
L = δ −1 ∗.

Convolution operator * is defined by (u ∗ v)(i) = ∞ j =−∞ u(i − j )v(j ).
Now we introduce a complex number z which satisfies |z| < 1. An operator (I − zL) is
a generalized difference operator of the first order. It is a bounded linear operator. Its inverse
operator is given by the Neumann series,
∞

(I − zL)−1 = z k Lk , (1)
k=0

which is also a bounded linear operator.

Now we introduce a special second-order difference operator given by
A = (I − zL)∗ (I − zL). (2)
The above operator A is bounded linear, self-adjoint and positive definite. A new inner product
in 2 ,
(u, v)A = (Au, v) = ((I − zL)u, (I − zL)v), (3)
gives an equivalent metric to the old one (·, ·). In fact, we have an inequality
(1 − |z|)u  uA  (1 + |z|)u (u ∈ 2 )
 
uA ≡ (u, u)A = (Au, u) = u − zLu.
Since L∗ = L−1 we have
A = −z̄L(L−1 − z−1 I )(L−1 − zI ) = −z1−1 L(L−1 − z0 I )(L−1 − z1 I ),

2
J. Phys. A: Math. Theor. 42 (2009) 454014 A Nagai et al

where z0 = z, z1 = z−1 . Given a characteristic polynomial p(λ) = (λ − z0 )(λ − z1 ), we have

an expansion formula by partial fractions

1
1 z 1 z
−1
p(λ) = ej (λ − zj )−1 , e0 = =− , e1 = = .
j =0
p (z0 ) 1 − |z|2 p (z1 ) 1 − |z|2

Using this formula, we have


1
A−1 = −z1 L−1 ej (L−1 − zj I )−1 .
j =0

It is easy to see that the following equalities hold:

∞
 ∞

(L−1 − z0 I )−1 = z0k Lk+1 = zk Lk+1 , (4)
k=0 k=0
∞
 ∞

(L−1 − z1 I )−1 = − z1−k−1 L−k = − zk+1 L−k , (5)
k=0 k=0

from which we finally have

 ∞ ∞

 
A −1
= −z1 e0 z0k Lk − e1 z1−k−1 L−k−1
k=0 k=0
∞ ∞

1  
−k−1
= k
z L + k
z k+1
L .
1 − |z|2 k=0 k=0

The above relation shows that the inverse operator A−1 is a convolution operator by the
sequence (. . . , A−1 (i), . . .) given by
 ∞ ∞

 
−1 −k−1
A (i) = −z1 e0 z0 δ(i + k) − e1
k
z1 δ(i − k − 1)
k=0 k=0
∞ ∞

1  
= k
z δ(i + k) + z k+1
δ(i − k − 1) . (6)
1 − |z|2 k=0 k=0

For the set of a Hilbert space 2 and an inner product (u, v)A , the kernel function
K(i, j ) = A−1 (i − j )
 ∞ ∞

 
−k−1
= −z1 e0 z0 δ(i − j + k) − e1
k
z1 δ(i − j − k − 1)
k=0 k=0
∞ ∞

1  
= z δ(i − j + k) +
k
z k+1
δ(i − j − k − 1) . (7)
1 − |z|2 k=0 k=0
is a reproducing kernel. In fact, we can show from (3) and (7) that for any j ∈ Z =
{0, ±1, ±2, . . .} fixed and for any u ∈ 2 , K(i, j ), as a function in i, belongs to 2 and the
reproducing relation,
(u, K(·, j ))A = u(j ) (8)
holds. From the theory of reproducing kernel, we have the following theorem.

3
J. Phys. A: Math. Theor. 42 (2009) 454014 A Nagai et al

Theorem 1. Let z be a complex number which satisfies |z| < 1. We can find a positive constant
C such that for any u ∈ 2 the discrete Sobolev inequality of the following form holds:
(sup |u(j )|)2  C(Au, u). (9)
j ∈Z

Among such constants C, the best (least) constant is given by

1
C(z) = A−1 (0) = . (10)
1 − |z|2
If we replace C by C(z) in the above inequality (9), then the equality holds for u(i) = K(i, j )
for any fixed j ∈ Z.

We call inequality (9) the discrete Sobolev inequality because (Au, u) = ∞ n=−∞ |u(j ) −
2
zu(j + 1)| is the square of the difference sequence for u.

Proof. [Proof of theorem 1.] Applying the Cauchy–Schwartz inequality to the reproducing
relation (8), we have
(sup |u(j )|)2 = (sup |(u, K(·, j ))A |)2
j ∈Z j ∈Z

 sup(K(·, j ), K(·, j ))A u2A = sup K(j, j )u2A = A−1 (0)uA .

j ∈Z j ∈Z

Therefore, the best constant of inequality (9) is given by

1
A−1 (0) =
1 − |z|2
where we have put i = 0 in (6). The equality in (9) holds for u(i) = K(i, j ) with j arbitrarily
fixed. 

3. Discrete Sobolev inequality of higher order

We introduce M distinct complex numbers z0 , . . . , zM−1 which satisfy |zj | < 1 (0  j 

M − 1) and a bounded linear operator on 2 defined by

M−1
A= A(zj ), (11)
j =0

where A(z) = (I − zL)∗ (I − zL) is a bounded linear, self-adjoint, positive-definite operator

which was treated in section 1. Operators A(zj ) are mutually commutative, so

M−1
A= (I − z̄j L−1 )(I − zj L)
j =0

is also self-adjoint and positive definite. Introducing new parameters zM , . . . , z2M−1 defined
by
zM+j = z̄j−1 (0  j  M − 1),
we can write
⎛ ⎞

M−1

2M−1
A = (−1)M ⎝ −1 ⎠ M
zM+j L (L−1 − zj I ).
j =0 j =0

4
J. Phys. A: Math. Theor. 42 (2009) 454014 A Nagai et al

We attach a 2Mth-order characteristic polynomial defined by p(λ) = (λ − z0 )(λ − z1 ) · · · (λ −

z2M−1 ). According to the theory of Lagrange interpolation polynomials we have the following
fundamental expansion by partial fractions:

2M−1
p(λ)−1 = ej (λ − zj )−1 , ej = 1/p (zj ) (0  j  2M − 1).
j =0

Using this formula, we have

M−1 
2M−1
A−1 = (−1)M zM+j L−M ej (L−1 − zj I )−1 .
j =0 j =0

From (4) and (5), A−1 is rewritten as follows:

⎡ ⎤

M−1  
M−1 ∞  
2M−1 ∞
A−1 = (−1)M zM+j ⎣ ej zjk Lk+1−M − ej zj−k−1 L−k−M ⎦ .
j =0 j =0 k=0 j =M k=0

A−1 is a convolution operator by a sequence

⎡

M−1  
M−1 ∞
A−1 (i) = (−1)M zM+j ⎣ ej zjk δ(i + k + 1 − M)
j =0 j =0 k=0
⎤

2M−1 ∞

− ej zj−k−1 δ(i − k − M)⎦ .
j =M k=0

Substitution of i = 0 in the above expression gives

M−1 
M−1
A−1 (0) = (−1)M zM+j ej zjM−1 . (12)
j =0 j =0

Owing to the property

⎛ ⎞
0
 −1 ⎜ ... ⎟
⎜
⎟
(ei ) = zji ⎜ ⎟ (0  i, j  2M − 1),
⎝0⎠
1
we have
⎛ ⎞
0

M−1
  −1 ⎜ .. ⎟
⎜
⎟
A−1 (0) = (−1)M zM+j z0M−1 , . . . , zM−1
M−1
, 0, . . . , 0 zji ⎜ . ⎟ ,
⎝0⎠
j =0
1
 
where z0M−1 , . . . , zM−1
M−1
, 0, . . . , 0 and (0, . . . , 0, 1) are 2M-dimensional vectors. In general,
for any n×n matrix A = (aij ) and n×1 matrices b = t (b0 , . . . , bn−1 ) and c = t (c0 , . . . , cn−1 )
the following formula holds:
 
 A c
t −1
bA c = − t   |A|.
b 0
5
J. Phys. A: Math. Theor. 42 (2009) 454014 A Nagai et al

Applying this formula to (13) we have

 
 0 
 
 ..

M−1  zji .  i
−1   z  ,
A (0) = (−1) M−1
zM+j  0 
 j
j =0  1 

 zM−1 · · · M−1
zM−1 0 ··· 0 0 
0
where the numerator is a (2M + 1) × (2M + 1) matrix and the denominator is a 2M × 2M
matrix. The Laplace expansion of the above numerator with respect to the last column gives
 
 
 

M−1
 zji   i
A−1 (0) = (−1)M zM+j  

z .
j (13)
j =0  
 zM−1 · · · zM−1 0 · · · 0 
0 M−1
We now arrive at the following conclusion.
Theorem 2. Let z0 , . . . , zM−1 be M complex numbers which satisfy |zj | < 1 (0  j  M −1).
zM , . . . , z2M−1 are given by zM+j = z̄j−1 (0  j  M − 1). We can find a positive constant C
such that for any u ∈ 2 the following discrete Sobolev inequality of the Mth order holds:
M−1 2


 
(sup |u(j )|)2  C  (I − zj L)u . (14)
j ∈Z  
j =0

Among such constants C, the best constant is given by

 
 

 
M−1
 zji   i
C(M; z0 , . . . , zM−1 ) = (−1)M zM+j  

z .
j (15)
j =0  
zM−1 · · · zM−1 0 ··· 0 
0 M−1
If we replace C by C(M; z0 , . . . , zM−1 ) in inequality (14), the equality holds for u(i) =
K(i, j ) = A−1 (i − j ) with j ∈ Z arbitrarily fixed.

As a special case, we have

1 1 − |z0 z1 |2
C(1; z0 ) = , C(2; z0 , z1 ) = .
1 − |z0 | 2 |1 − z0 z̄1 | (1 − |z0 |2 )(1 − |z1 |2 )
2

It is interesting to note that from the spectral decomposition

 2π √

M−1 √
1
A(i) = e −1iy |1 − zj e− −1y |2 dy,
2π 0 j =0

where

M−1 √
λ(y) = |1 − zj e− −1y 2
|
j =0

is the continuous spectrum of the self-adjoint operator A, we also have the spectral
decomposition
 2π √

M−1 √
−1 1 −1iy
A (i) = e |1 − zj e− −1y |−2 dy.
2π 0 j =0

6
J. Phys. A: Math. Theor. 42 (2009) 454014 A Nagai et al

Putting i = 0 we have


2π M−1 √
1
A−1 (0) = |1 − zj e− −1y −2
| dy. (16)
2π 0 j =0
We should remark, however, that it is not easy to perform integration if M  2.

4. Periodic discrete Sobolev inequality

In this section we add the restriction

∀
u(i + N ) = u(i) i∈Z (17)
for any fixed N = 2, 3, 4, . . .. We modify the definition of a Hilbert space  , an inner product 2

(u, v) and delta function vector δ in the previous section as follows:

2N = {u = t (u(0), . . . , u(N − 1))|u(i) ∈ C (0  i  N − 1)} = CN ,

N−1
(u, v) = u(i)v(i),
i=0
δ j = (. . . , δ(i − j ), . . .), δ(i) = 1 (i ≡ 0 mod N ), 0 (else).
The unitary shift operator L = δ −1 ∗ is the cyclic rotate left operator defined by
Lt (u(0), . . . , u(N − 1)) = t (u(1), . . . , u(N − 1), u(0)).
The convolution u ∗ v of u and v ∈ CN is obtained by

N−1
(u ∗ v)(i) = u(i − j )v(j )
j =0

extending the definition of u(i) ∈ CN outside 0  i  N − 1 by periodicity u(i + N ) =

u(i)(i ∈ Z). Note that we must add the new restriction
LN = I.
The formula of the inverse operator (1) is replaced by

N−1
(I − zL)−1 = (1 − zN )−1 z k Lk . (18)
k=0
We note that we have dealt with the same problem putting z = 1 in our previous work [5],
where we considered a generalized inverse of I − L.
We also treat the same form of the operator A (11) as in the previous section but the
formula of its inverse is replaced by
⎡

M−1  
M−1 
−1 N−1
A−1 = (−1)M zM+j ⎣ ej 1 − zjN zjk Lk−M0
j =0 j =0 k=0
⎤

2M−1
 −1 
N−1
− ej 1 − zj−N zj−k−1 L−k−1−M0 ⎦ , (19)
j =M k=0

where M0 = Mod(M − 1, N ) is the remainder of the natural number M − 1 divided by N and

takes values in {0, 1, 2, . . . , N − 1}. A−1 is also a convolution operator:

N−1 
N−1
−1 −1
(A u)(i) = A (i, j )u(j ) = A−1 (i − j )u(j ).
j =0 j =0

7
J. Phys. A: Math. Theor. 42 (2009) 454014 A Nagai et al

It should be noted that the relation A−1 (i + N ) = A−1 (i) holds for any i ∈ Z. Now we have
⎡

M−1  
M−1 
−1 N−1
A−1 (i) = (−1)M zM+j ⎣ ej 1 − zjN zjk δ(i + k − M0 )
j =0 j =0 k=0
⎤

2M−1
 −1

N−1
− ej 1 − zj−N zj−k−1 δ(i − k − 1 − M0 )⎦ . (20)
j =M k=0

Putting i = 0, we have
⎡

M−1 
M−1
 
−1 N−1
A−1 (0) = (−1)M zM+j ⎣ ej 1 − zjN zjk δ(k − M0 )
j =0 j =0 k=0
⎤

2M−1
 −1

N−1
− ej 1 − zj−N zj−k−1 δ(−k − 1 − M0 )⎦
j =M k=0

M−1 
2M−1
 −1
= (−1)M zM+j ej 1 − zjN zjM0 . (21)
j =0 j =0

Through the same argument as in the previous section, A−1 (0) is rewritten in the following
determinant expression:
 
 
 

M−1
 zji   i
−1
A (0) = (−1) M
zM+j  

z  ,
j (22)
j =0  
 ϕ(z ) 
j

where
zM0
ϕ(z) = .
1 − zN
We finally have the following theorem.

Theorem 3. Let N, M be two fixed numbers N = 2, 3, 4, . . . and M = 1, 2, 3, . . .. We choose

M distinct complex numbers z0 , . . . , zM−1 such that |zj | < 1 (0  j  M − 1) and define
zM+j by zM+j = z̄j−1 (0  j  M − 1). For any vector u ∈ CN there exists a positive constant
C such that the following discrete periodic Sobolev inequality holds:
 M−1 2


( max |u(j )|)2  C   (I − z j L)u .
 (23)
0j N−1
j =0

Among such constants C the best constant is given by

 
 
 

M−1
 zji   i
C(N, M; z0 , . . . , zM−1 ) = (−1)M zM+j  

z ,
j (24)
j =0  
 ϕ(zj ) 

where ϕ(z) = (1 − zN )−1 zMod(M−1,N) . If we replace C by C(N, M; z0 , . . . , zM−1 ) in the

above inequality (23), the equality holds for u(i) = K(i, j ) = A−1 (i − j ) for any fixed
j (0  j  N − 1).

8
J. Phys. A: Math. Theor. 42 (2009) 454014 A Nagai et al

Explicit forms of the best constants in the case M = 1, 2 are given in the following
theorem, which is proved in the appendix.

Theorem 4. For any fixed N = 2, 3, 4, . . ., the best constant C(N, 1; z0 ) and C(N, 2; z0 , z1 )
are given by

1 − |z0 |2N
C(N, 1; z0 ) =  2 > 0, (25)
(1 − |z0 |2 )1 − zN  0
N−1    j 2
z0i z1 + i+j =k+N z0i z1 
j
i+j =k
k=0 i,j 0
C(N, 2; z0 , z1 ) =
i,j >k
    > 0. (26)
1 − z N 2 1 − z N 2
0 1

It is interesting to note that A−1 (0) given by (22) has another expression, which follows
from the spectral decomposition or decomposition into Jordan canonical form. Using
√ 1
ω = exp( −1 2π/N), W = √ (ωij ), L̂ = (ωi δij ),
N
we have the decomposition

L = W L̂W ∗ .

The diagonal matrix L̂ is the Jordan canonical form of L; W is a unitary matrix, so we have
W ∗ = W −1 . Using this fact, we have the decomposition

A = W ÂW ∗ ,

where
M−1

M−1

 = (I − zk L̂)∗ (I − zk L̂) = |1 − zk ωi |2 δij

k=0 k=0
M−1
is the Jordan canonical form of A. λi = k=0 |1 − zk ωi |2 (0  i  N − 1) are eigenvalues
of A. From this fact, we have
M−1

−1 −1 ∗ −1 i −2
A = W Â W , Â = |1 − zk ω | δij ,
k=0

that is,

1  (i−j )l

N−1 M−1
A−1 (i − j ) = ω |1 − zk ωl |−2 .
N l=0 k=0

Finally, we have

1 

N−1 M−1
−1
A (0) = |1 − zk ωl |−2 .
N l=0 k=0

9
J. Phys. A: Math. Theor. 42 (2009) 454014 A Nagai et al

5. Concluding remarks

We have derived the discrete Sobolev inequality which estimates the supremum of the sequence
u(j ) by the norm of difference sequence uA , which is considered as a discrete analogue of
the Sobolev norm. It is expected that the obtained results have important relations with error
analysis in the field of numerical analysis.
In the continuous  limit, we essentially obtain the boundary value problem for 2Mth
differential operator M−1j =0 (−(d/dx) + qj ), where qj is constant. In the case qj = 0 (0 
2

j  M − 1), we have already obtained the result [3]. In the case M = 1 and q0 = 0, we have
a string deflection problem [2]. It is also an interesting problem to investigate in detail the
case M = 2, q0 , q1 = 0, which describes a beam deflection problem.

Acknowledgments

One of the authors (AN) is supported by J. S. P. S. Grant-in-Aid for Scientific Research (C)
No 20540138.

Appendix.

Proof of theorem 4. We here prove theorem 4 concerning the best constant of discrete Sobolev
inequality with periodic boundary condition. Since (25) is proved through simple calculations,
we prove only (26):
 
 1 1 1 1 

 z z3 
 0 z1 z2
 z2 2 2
z32 
 0 z1 z2
 z0 z1 z2 z3 
 1−zN 1−zN 1−zN 1−z N 
C(N, 2; z0 , z1 ) = z2 z3 0
 1 2
 3 . (A.1)
1 1 1 1 
 
 z0 z1 z2 z3 
 
z 2 z 2 z 2 z 2 
 0 1 2 3
z 3 z 3 z 3 z 3 
0 1 2 3
Noting that the denominator is calculated as

|z0 − z1 |2 |1 − z0 z¯1 |2 (1 − |z0 |2 )(1 − |z1 |2 )
(zi − zj ) = − ,
i<j
z¯0 3 z¯1 3
we have
 
 1/z0 1/z1 z̄0 z̄1 
 
 1 1 1 1 
2 
|z0 | |z1 |  z
2

 0 z1 1/z̄0 1/z̄1 
 1 1 
 1−zN 1
1−z1N
1
1−z̄0−N 1−z̄ −N 
C(N, 2; z0 , z1 ) = − 0 1
. (A.2)
|z0 − z1 |2 |1 − z0 z¯1 |2 (1 − |z0 |2 )(1 − |z1 |2 )
Expanding the numerator with respect to the last row, we have
   
 1/z0 1/z1 z̄0 z̄1   1 1 z̄02 z̄12 
  
 1 1 1 1    z z z̄0 z̄1 
  0 1 
−|z0 |2 |z1 |2  z 1/z̄1  = −  z02 z12 1 1 
 0 z1 1/z̄0 
 1   z0 −z̄0N+1 −z̄1N+1 
 1−zN 1−z
1
N
1
1−z̄−N
1
1−z̄−N
  N z1
N

0 1 0 1 1−z0 1−z1 1−z̄0N 1−z̄1N

10
J. Phys. A: Math. Theor. 42 (2009) 454014 A Nagai et al

z0  
= z̄0 + z̄02 z1 |z1 |2 + |z1 |2 z̄1 − z̄0 |z1 |4 − z̄02 z1 − z̄1
1 − z0 N

z1  
− z̄0 + |z0 |4 z̄1 + z0 z̄12 − |z0 |2 z0 z̄12 − |z0 |2 z̄0 − z̄1
1 − z1 N

z̄N+1  
− 0 N z1 + z0 |z1 |4 + z02 z̄1 − z02 z1 |z1 |2 − |z1 |2 z1 − z0
1 − z̄0
z̄N+1  
+ 1 N z1 + |z0 |2 z0 + |z0 |2 z̄0 z1 − |z0 |4 z1 − z̄0 z12 − z0
1 − z̄1
1  2  2
= 2 |z0 | (1 − |z0 |2N )(1 − |z1 |2 )|1 − z0 z̄1 |2 1 − z1N 
1 − z  |1 − z |2
N N
0 1
 2
+ |z1 | (1 − |z1 | )(1 − |z0 |2 )|1 − z0 z̄1 |2 1 − z0N 
2 2N
  
− z0 z̄1 (1 − (z0 z̄1 )N )(1 − |z0 |2 )(1 − |z1 |2 )(1 − z̄0 z1 ) 1 − z̄0N 1 − z1N
  
− z̄0 z1 (1 − (z̄0 z1 )N )(1 − |z0 |2 )(1 − |z1 |2 )(1 − z0 z̄1 ) 1 − z0N 1 − z̄1N

through straightforward calculations. Substitution of the above result into (A.2) gives

1
C(N, 2; z0 , z1 ) =  2  2
|z0 − z1 |2 (1
− |z0 |2 )(1
− |z1 − z0 z̄1 |2 1 − z0N  1 − z1N 
|2 )|1
  2
× |z0 |2 (1 − |z0 |2N )(1 − |z1 |2 )|1 − z0 z̄1 |2 1 − z1N 
 2
+ |z1 |2 (1 − |z1 |2N )(1 − |z0 |2 )|1 − z0 z̄1 |2 1 − z0N 
  
− z0 z̄1 (1 − (z0 z̄1 )N )(1 − |z0 |2 )(1 − |z1 |2 )(1 − z̄0 z1 ) 1 − z̄0N 1 − z1N
  
− z̄0 z1 (1 − (z̄0 z1 )N )(1 − |z0 |2 )(1 − |z1 |2 )(1 − z0 z̄1 ) 1 − z0N 1 − z̄1N
1
=  2  2
|z0 − z1 | 1 − zN  1 − zN 
2 
0 1

|z0 |2 (1 − |z0 |2N )  2 |z1 |2 (1 − |z1 |2N )  
× 1 − z1N  + 1 − z N 2
1 − |z0 | 2 1 − |z1 | 2 0

N 
1 − (z0 z̄1 )   1 − (z̄0 z1 )N   
− z0 z̄1 1 − z1N 1 − z̄0N − z̄0 z1 1 − z0N 1 − z̄1N
1 − z0 z̄1 1 − z̄0 z1
1 
N−1
 2(k+1)    
=  2  2 |z0 | 1 − zN 2 + |z1 |2(k+1) 1 − zN 2
|z0 − z1 |2 1 − zN  1 − zN 
1 0
0 1 k=0
     
− (z0 z̄1 )k+1 1 − z1N 1 − z̄0N − (z̄0 z1 )k+1 1 − z0N 1 − z̄1N
1 
N−1
   
=     zk+1 1 − zN − zk+1 1 − zN 2
|z0 − z1 |2 1 − z0N  1 − z1N  k=0
2 2 0 1 1 0

 2
 

N−1    
1  j j 
=     z i
z + z i
z  ,
1 − zN 2 1 − zN 2 k=0  i+j =k 0 1 0 1

0 1  i,j 0 i+j =k+N
i,j >k


which completes the proof.



11
J. Phys. A: Math. Theor. 42 (2009) 454014 A Nagai et al

References

[1] Kametaka Y, Watanabe K and Nagai A 2005 Proc. Japan. Acad. A 81 1

[2] Kametaka Y, Watanabe K, Nagai A, Yamagishi H and Takemura K 2007 Sci. Math. Japonicae Online e-2007
283
[3] Kametaka Y, Yamagishi H, Watanabe K, Nagai A and Takemura K 2007 Sci. Math. Japonicae Online e-2007 63
[4] Kametaka Y, Takemura K, Yamagishi H, Nagai A and Watanabe K 2007 Sci. Math. Japonicae Online e-2007
739
[5] Nagai A, Kametaka Y, Yamagishi H, Watanabe K and Takemura K 2008 Funkcialaj Ekvacioj 51 307
[6] Talenti G 1976 Ann. Mat. Pura. Appl. 110 353

12