A Sharp Version of Mahlers Inequality for Products of Polynomials

Andr as Kro o

and Igor E. Pritsker

Abstract
In this note we give some sharp estimates for norms of polynomials via the products of norms of their
linear terms. Dierent convex norms on the unit disc are considered.
1 Introduction
Let p
1
(z), . . . , p
m
(z) be complex polynomials such that their product p :=
m

j=1
p
j
is of degree
n. Then, by a well-known inequality of Mahler [6],
p
1

. . . p
m

2
n
p

, n IN, (1.1)
where f

:= max
|z|=1
|f(z)| denotes the uniform norm on the unit circle. (A weaker
version of (1.1) appeared earlier in Gelfond [4].) Choosing p(z) = z
n
+ 1, m = n, and
p
j
(z) (1 j n) to be the linear factors of z
n
+ 1, one can easily see that the constant in
(1.1) cannot be, in general, smaller than 2
n1
, i.e., 2
n
in (1.1) is sharp up to the factor 2.
Based on this observation, it was conjectured by Sarantopoulos [7] that the constant 2
n
in
(1.1) can be replaced by 2
n1
. We shall verify this conjecture in the present note. In fact,
this will be accomplished in the context of generalized polynomials, and other norms on the
unit circle will be discussed as well. It should be noted that, for m = o(

n), the constant in

(1.1) was substantially improved by Boyd [3], see also Borwein [1] and Borwein-Erdelyi [2]
for some recent developments in this area.

1991 Mathematics Subject Classication: 30C10, 11C08.

Written during the authors visit at Kent State University, Department of Mathematics and Computer Science, Kent, OH
44242-0001, U.S.A.
1
A SHARP VERSION OF MAHLERS INEQUALITY 2
2 Results
For = (
1
, . . . ,
n
) IR
n
+
, denote

n
() := min
|a
j
|=1
_
_
_
_
_
_
n

j=1
|z a
j
|

j
_
_
_
_
_
_

and d
n
() :=
n

j=1

j
. (2.2)
Now we can state the next
Theorem 2.1 For any IR
n
+
and any set {a
j
I C, 1 j n}, we have
n

j=1
|z a
j
|

2
dn()

n
()
_
_
_
_
_
_
n

j=1
|z a
j
|

j
_
_
_
_
_
_

. (2.3)
Moreover, the equality in (2.3) holds if and only if {a
j
, 1 j n} is a solution of the
minimization problem (2.2).
Functions p(z) =
n

j=1
|za
j
|

j
are usually called generalized polynomials of degree
n

j=1

j
=
d
n
() ( IR
n
+
) (see [2]). Using Theorem 2.1, we can easily derive the following
Corollary 2.2 Let p
1
(z), . . . , p
m
(z) be generalized complex polynomials such that p(z) :=
m

j=1
p
j
(z) =
n

j=1
|z a
j
|

j
, where = (
1
, . . . ,
n
) IR
n
+
and {a
j
I C, 1 j n} are
arbitrary. Then
p
1

. . . ||p
m

2
dn()

n
()
p

, (2.4)
with equality being attained only if p is a solution of the minimization problem (2.2).
It is well-known that when

= (1, . . . , 1), the solution of the minimization problem (2.2)

is given by
n
(

) = 2, with z
n
+1 being the unique (up to a rotation) extremal polynomial
for (2.2). Hence we obtain an improvement of Mahlers inequality (1.1) from (2.4).
Corollary 2.3 Let p
1
, . . . , p
m
be complex polynomials such that their product p =
m

j=1
p
j
is
of degree n. Then
p
1

. . . p
m

2
n1
p

, (2.5)
and equality in (2.5) is attained if and only if p(z) = z
n
+ , with || = 1, and m = n.
A SHARP VERSION OF MAHLERS INEQUALITY 3
Inequalities (2.3)-(2.5) provide sharp estimates for the products of generalized polynomials
in the uniform norm. Next, we present the L
2
-version of Theorem 2.1 for ordinary complex
polynomials. Consider the L
q
-norm on the unit circle given by
f
q
=
_
1
2
_
2
0
|f(e
i
)|
q
d
_
1/q
, 0 < q < .
Theorem 2.4 For any set {a
j
I C, 1 j n},
2
(1n)/2
n

j=1
z a
j

2

_
_
_
_
_
_
n

j=1
(z a
j
)
_
_
_
_
_
_
2
2
n/2
_
2n
n
_
1/2
n

j=1
z a
j

2
. (2.6)
Moreover, the estimates (2.6) are sharp; the lower bound is attained for z
n
+ 1, the upper
bound is attained for (z + 1)
n
, and these extremal polynomials are unique up to a rotation.
Note that inequality (2.3) can be written in an equivalent form
_
_
_
_
_
_
n

j=1
|z a
j
|

j
_
_
_
_
_
_

n
()2
dn()
n

j=1
(1 +|a
j
|)

j
. (2.7)
Let us also mention an interesting explicit form of (2.6): for p
n
(z) =
n

j=1
(za
j
) =
n

k=0
c
k
z
k
,
where c
n
= 1, we have
2
n+1
n

k=1
(1 +|a
k
|
2
)
n

k=0
|c
k
|
2
2
n
_
2n
n
_
n

k=1
(1 +|a
k
|
2
), (2.8)
and both of these bounds are sharp. Although the coecients c
k
of a polynomial p
n
can
be expressed explicitly via its zeros, the estimates (2.8) do not seem to follow directly from
these expressions.
Now we shall address the question of extending (2.7) for the L
q
-norms.
Let
n,q
() be dened as follows:

n,q
() := min
|a
j
|=1
_
_
_
_
_
_
n

j=1
|z a
j
|

j
_
_
_
_
_
_
q
, = (
1
, . . . ,
n
) IR
n
+
. (2.9)
Theorem 2.5 Let = (
1
, . . . ,
n
) IR
n
+
and 0 < q be such that q
j
1, 1 j n.
Then, for any set {a
j
I C, 1 j n}, we have
A SHARP VERSION OF MAHLERS INEQUALITY 4
_
_
_
_
_
_
n

j=1
|z a
j
|

j
_
_
_
_
_
_
q

n,q
() 2
dn()
n

j=1
(1 +|a
j
|)

j
. (2.10)
The equality in (2.10) is attained for solutions of the minimization problem (2.9). If, in
addition, 1 < q
j
< , 1 j n, then this equality holds only for solutions of (2.9).
Note that the estimate (2.10) is more general than (2.3), but on the other hand the
additional claim of Theorem 2.1, that equality in (2.3) can be attained only when the roots
of generalized polynomials are on the unit circle, does not follow from Theorem 2.5. Thus,
Theorems 2.1 and 2.5 complement each other. In addition, their proofs are based on dierent
methods. Theorem 2.1 (as well as Theorem 2.4) will follow by variational arguments, while
the proof of Theorem 2.5 will be based on applications of symmetry and convexity.
The above results indicate that the sharp constants, appearing in inequalities for the
norms of products of polynomials, depend on the solution of the extremal problems (2.2)
and (2.9). For ordinary polynomials, the explicit solution of such extremal problem can be
given for a wide class of norms. Let

n
:= min
|a
j
|=1
_
_
_
_
_
_
n

j=1
(z a
j
)
_
_
_
_
_
_
, (2.11)
where is an arbitrary convex norm in the space of polynomials on the unit circle. We
shall say that is rotation invariant if for any xed
0
IR and any polynomial p
p(e
i(+
0
)
) = p(e
i
) and p = |p|.
Our next result extends Theorem 2.5 for rotation invariant norms in the case of ordinary
polynomials.
Theorem 2.6 For any rotation invariant convex norm on the unit circle and any set
{a
j
I C, 1 j n}, we have
n
= z
n
+ 1 and
_
_
_
_
_
_
n

j=1
(z a
j
)
_
_
_
_
_
_
2
n
z
n
+ 1
n

j=1
(1 +|a
j
|). (2.12)
The equality in (2.12) is attained for solutions of the minimization problem (2.11). If, in
addition, is strictly convex, then this equality holds only for solutions of (2.11).
A SHARP VERSION OF MAHLERS INEQUALITY 5
3 Proofs
3.1 Proof of Theorem 2.1.
First, we note that (2.3) is trivial for n = 1 and assume that n 2. Let us consider the
following minimization problem:

n
() := inf
a
j
I C
_
_
_
_
_
_
n

j=1
|z a
j
|

j
_
_
_
_
_
_

j=1
(1 +|a
j
|)

j
, IR
n
+
. (3.13)
Since the functional minimized in (3.13) is invariant with respect to the transformation
a
j
1/a
j
, it is clear that the inf in (3.13) is attained for |a
j
| 1, 1 j n. We are
going to verify now that this inf can be attained only when |a
j
| = 1, 1 j n. This will
immediately imply the statement of Theorem 2.1. We shall show that the inf in (3.13) can
be attained only for a
j
s on the unit circle, by using variational arguments based on the
following well-known formula for the directional derivative of the L

-norm (see [5]).

lim
t0
+
f + tg

t
= max
zE(f)
Re g sgn f, (3.14)
where E(f) = {|z| = 1 : |f(z)| = f

}, sgn f = f/|f|.
Since
_
_
_
_
_
_
n

j=1
|z a
j
|

j
_
_
_
_
_
_

(1+|a
1
|)

1
_
_
_
_
_
_
n

j=2
|z a
j
|

j
_
_
_
_
_
_

, it easily follows that we do not lose

generality by assuming that the inf in (3.13) is attained for 0 < |a
j
| 1, 1 j n. Set
p(z) :=
n

j=1
|z a
j
|

j
, p
t
(z) :=
n

j=1
|z a
j
+ tw
j
|

j
S :=
n

j=1
(1 +|a
j
|)

j
, S
t
:=
n

j=1
(1 +|a
j
tw
j
|)

j
where w
j
I C, 1 j n, and t > 0 are arbitrary. Evidently,
1
t
(Sp
t

S
t
p

) 0. (3.15)
A SHARP VERSION OF MAHLERS INEQUALITY 6
Furthermore, it can be easily shown that
|z a
j
+ tw
j
|

j
= |z a
j
|

j
+ t
j
|z a
j
|

j
2
Re w
j
(z a
j
) + O(t
2
), 1 j n,
where O(t
2
) above is uniform in z from compact subsets of I C\{a
j
}
n
j=1
. Thus,
p
t
(z) = p(z) + tp(z)
n

j=1

j
Re
_
w
j
z a
j
_
+ O(t
2
). (3.16)
Similarly, for a
j
= 0
(1 +|a
j
tw
j
|)

j
= (1 +|a
j
|)

j
(1 +|a
j
|)

j
1
t Re w
j
sgn a
j
+ O(t
2
)
and therefore
S
t
= S tS
n

j=1

j
1 +|a
j
|
Re w
j
sgn a
j
+ O(t
2
). (3.17)
Using (3.14)-(3.17) we obtain
0 lim
t0
+
1
t
(p
t

) +p

j=1

j
1 +|a
j
|
Re w
j
sgn a
j
= p

max
zE(p)
Re
n

j=1

j
w
j
z a
j
+p

j=1

j
1 +|a
j
|
Re w
j
sgn a
j
.
This means that for every w
j
I C, 1 j n, there exists a z E(p) so that
n

j=1

j
Re
_
w
j
_
1
z a
j
+
sgn a
j
1 +|a
j
|
__
0. (3.18)
It can be easily seen that
1
z a
j
+
sgn a
j
1 +|a
j
|
=
(1 |a
j
|)(|a
j
| + Re a
j
z) + i (1 +|a
j
|)Im a
j
z
a
j
(1 +|a
j
|)|z a
j
|
2
This and (3.18) yield that for every w
j
I C, 1 j n, there exists a point z E(p) so
that
n

j=1

j
Re
_
w
j
|z a
j
|
2
((1 |a
j
|)(|a
j
| + Re a
j
z) + i (1 +|a
j
|)Im a
j
z)
_
0. (3.19)
A SHARP VERSION OF MAHLERS INEQUALITY 7
Assume now that |a
j
| = 1 for 1 j k (1 k n) and |a
j
| = 1 if k < j n. Setting
in (3.19) w
j
= 1/(|a
j
| 1) for 1 j k, and w
j
= 0 for k < j n, we obtain that for some
z
0
E(p)
k

j=1

j
|z
0
a
j
|
2
(|a
j
| + Re a
j
z
0
) 0. (3.20)
On the other hand |z
0
| = 1 and, therefore, each term in the sum (3.20) is nonnegative, i.e.,
z
0
= a
j
/|a
j
|, 1 j k. Thus, we may assume that z
0
= 1 and a
j
= x
j
> 0 for
1 j k. Using, in addition, that
Re
1 + z
z x
j
=
(1 x
j
)(1 + Re z)
|z x
j
|
2
,
we obtain from (3.18) that for every b
j
IR, 1 j k, and w
j
I C, k < j n, there exists
a point z E(p) so that
k

j=1

j
b
j
1 + Re z
|z x
j
|
2
+
n

j=k+1

j
Re
_
w
j
a
j
+ z
a
j
z
_
0.
Choose b
j
= N, 1 j k. Then, it follows that for some z
N
E(p)
n

j=k+1

j
Re
_
w
j
a
j
+ z
N
a
j
z
N
_
N
k

j=1

j
1 + Re z
N
|z
N
x
j
|
2
. (3.21)
But 1 +Re z
N
0 and z
N
z

E(p) as N (for a proper subsequence). Thus we

obtain from (3.21) that z

= 1 and for every w

j
I C, k + 1 j n
n

j=k+1

j
Re
_
w
j
a
j
1
a
j
+ 1
_
0.
Thus a
j
= 1, k + 1 j n, i.e., p(z) = |z 1|

j=1
|z x
j
|

j
with some > 0 and x
j
>
0, 1 j k. Hence
n
(), dened by (3.13), equals 1, which is an evident contradiction if
n 2. This completes the proof of Theorem 2.1. 2
3.2 Proof of Theorem 2.4.
Set
A SHARP VERSION OF MAHLERS INEQUALITY 8
M
n
:= sup
a
j
I C
_
_
_
_
_
_
n

j=1
(z a
j
)
_
_
_
_
_
_
2
n

j=1
z a
j

2
; m
n
:= inf
a
j
I C
_
_
_
_
_
_
n

j=1
(z a
j
)
_
_
_
_
_
_
2
n

j=1
z a
j

2
.
Again, it is evident that sup and inf in the above expressions are attained.
Proposition 3.1 If the set of points {a
j
, 1 j n} I C is extremal for M
n
or m
n
and
|a
1
| = 1, then for p
n
(z) =
n

j=1
(z a
j
)
p
n

2
=
_
_
_
_
p
n
z a
1
_
_
_
_
2
z a
1

2
. (3.22)
Proof. (3.22) is trivial for a
1
= 0, so we may assume that a
1
= 0. Consider the functional
(t) =
_
2
0

p
n
(z)
ta
1
p
n
(z)
z a
1

2
d
1 + (1 + t)
2
|a
1
|
2
, z = e
i
.
The extremality of the set {a
j
, 1 j n} yields that

(0) = 0. Therefore, by dieren-

tiating (t), we obtain
(1 +|a
1
|
2
)
_
2
0
|p
n
(z)|
2
Re
a
1
z a
1
d +|a
1
|
2
_
2
0
|p
n
(z)|
2
d
=
_
2
0
|p
n
(z)|
2
_
Re
a
1
(1 +|a
1
|
2
)
z a
1
+|a
1
|
2
_
d = 0.
(3.23)
Moreover, for |z| = 1
Re
a
1
(1 +|a
1
|
2
)
z a
1
+|a
1
|
2
= Re
a
1
+ z|a
1
|
2
z a
1
=
(1 |a
1
|
2
) Re za
1
|z a
1
|
2
.
Since |a
1
| = 1, we obtain by substituting the above expression into (3.23) that
_
2
0

p
n
(z)
z a
1

2
Re za
1
d = 0, z = e
i
.
Finally, using this relation implies
A SHARP VERSION OF MAHLERS INEQUALITY 9
p
n

2
2
=
1
2
_
2
0
|p
n
(z)|
2
d =
1
2
_
2
0

p
n
(z)
z a
1

2
(z a
1
)(z a
1
) d
=
1
2
_
2
0

p
n
(z)
z a
1

2
(1 +|a
1
|
2
2Re za
1
)d =
=
1
2
_
2
0

p
n
(z)
z a
1

2
(1 +|a
1
|
2
) d = z a
1

2
2
_
_
_
_
p
n
z a
1
_
_
_
_
2
2
. 2
Corollary 3.2 If the set of points {a
j
, 1 j n} I C is extremal for M
n
(or m
n
) and
|a
1
| = 1, then M
n
= M
n1
(respectively, m
n
= m
n1
).
Proof. Clearly, M
n
M
n1
and m
n
m
n1
for every n 2. On the other hand, if |a
1
| = 1
then (3.22) yields that M
n
M
n1
and m
n
m
n1
. Thus M
n
= M
n1
and m
n
= m
n1
.
2
The next two statements are straightforward.
Proposition 3.3 For every n IN
M
n
2
n/2
(z + 1)
n

2
=

2
n
n

k=0
_
n
k
_
2

1/2
=
_
2
n
_
2n
n
__
1/2
,
m
n
2
n/2
z
n
+ 1
2
= 2
(1n)/2
.
Proposition 3.4 If the set of points {a
j
, 1 j n} {|z| = 1} is extremal for M
n
or
m
n
, then
M
n
= 2
n/2
_
_
_
_
_
_
n

j=1
(z a
j
)
_
_
_
_
_
_
2

2
n
n

k=0
_
n
k
_
2

1/2
=
_
2
n
_
2n
n
__
1/2
.
Respectively,
m
n
= 2
n/2
_
_
_
_
_
_
n

j=1
(z a
j
)
_
_
_
_
_
_
2
2
n/2
z
n
+ 1
2
= 2
(1n)/2
.
Let us verify now that
M
n
=
_
2
n
_
2n
n
__
1/2
and m
n
= 2
(1n)/2
.
A SHARP VERSION OF MAHLERS INEQUALITY 10
Assume that M
n
= . . . = M
m
> M
m1
, where 2 m n. Then, by Corollary 3.2,
|a
j
| = 1, 1 j m, where a
j
s are extremal for M
m
. Thus, by Propositions 3.3 and 3.4
2
n
_
2n
n
_
M
2
n
= M
2
m
2
m
_
2m
m
_
.
It is easy to show that this inequality is possible only if m = n, i.e., M
n
=
_
2
n
_
2n
n
__
1/2
.
Similarly, if m
n
= . . . = m

< m
1
, we obtain by Propositions 3.3 and 3.4
2
(1n)/2
m
n
= m

2
(1)/2
,
yielding that n = and m
n
= 2
(1n)/2
. The above arguments show that M
n
> M
n1
and
m
n
< m
n1
for every n 2. Thus, by Corollary 3.2, the extremal sets for M
n
and m
n
must belong to the unit circle. Therefore, if p
n
(z) =
n

j=1
(z a
j
) is extremal for m
n
, then
we must have |a
j
| = 1, 1 j n, and p
n

2
=

2. It is well-known that only p

n
(z) =
z
n
+ ( I C, || = 1) can satisfy the above properties. Similarly, if g
n
(z) =
n

j=1
(z b
j
) is
extremal for M
n
, then |b
j
| = 1, 1 j n, and g
n

2
=
_
_
2n
n
_
. Again, it is easy to see that
g
n
(z) = (z + )
n
( I C, || = 1).
The proof of Theorem 2.4 is now complete. 2
3.3 Proof of Theorem 2.5.
Consider the functional
(a
1
, . . . , a
n
) :=
_
_
_
_
_
_
n

j=1
|z a
j
|

j
_
_
_
_
_
_
q
n

j=1
(1 +|a
j
|)

j
and the corresponding extremal problem

n
() := inf
a
j
I C
(a
1
, . . . , a
n
). (3.24)
Evidently, it suces to show that (3.24) possesses a solution {a
j
, 1 j n} I C such that
|a
j
| = 1, 1 j n. First, observe that for every a
j
= 0, 1 j n,
A SHARP VERSION OF MAHLERS INEQUALITY 11
|z a
j
|

j
(1 +|a
j
|)

j
=

z
1
a
j

j
_
1 +
1
|a
j
|
_

j
,
and therefore
(a
1
, . . . , a
j
, . . . , a
n
) =
_
a
1
, . . . ,
1
a
j
, . . . , a
n
_
. (3.25)
(Note that this relation easily implies the existence of a solution of (3.24).)
Set
p(z) :=
n

j=1
|z a
j
|

j
, p
k
(z) := p(z)/|z a
k
|

k
,
S :=
n

j=1
(1 +|a
j
|)

j
, S
k
:= S/(1 +|a
k
|)

k
(1 k n).
Then, using that
j
q 1 (1 j n), we obtain for any a
k
= 0 and 0 t 1 :
_
_
_
_

z
_
ta
k
+ (1 t)
1
a
k
_

k
p
k
_
_
_
_
1/
k
q
=
_
_
_
_

z
_
ta
k
+ (1 t)
1
a
k
_

p
1/
k
k
_
_
_
_

k
q
t
_
_
_|z a
k
| p
1/
k
k
_
_
_

k
q
+ (1 t)
_
_
_
_

z
1
a
k

p
1/
k
k
_
_
_
_

k
q
= tp
1/
k
q
+ (1 t)
_
_
_
_

z
1
a
k

k
p
k
_
_
_
_
1/
k
q
= S
1/
k
k
_
t (1 +|a
k
|) ((a
1
, . . . , a
n
))
1/
k
+ (1 t)
_
1 +
1
|a
k
|
_
_

_
a
1
, . . . ,
1
a
k
, . . . , a
n
__
1/
k
_
.
(3.26)
Using (3.25) and the relation
t(1 +|a
k
|) + (1 t)
_
1 +
1
|a
k
|
_
= 1 +

ta
k
+
1 t
a
k

,
we obtain from (3.26)

_
a
1
, . . . , ta
k
+
1 t
a
k
, . . . , a
n
_
(a
1
, . . . , a
n
), 1 k n, (3.27)
A SHARP VERSION OF MAHLERS INEQUALITY 12
if a
k
= 0 and 0 t 1. Furthermore, the inequality p
q
(1 + |a
k
|)

k
p
k

q
yields that
(3.24) possesses a solution {a
j
, 1 j n} such that a
j
= 0, 1 j n. Moreover, (3.27)
implies that this solution can be chosen so that |a
j
| = 1, 1 j n.
Note that if 1 <
j
q < , then

j
q
is strictly convex, 1 j n. Therefore, it follows
from (3.26) that the equality in (3.27) can hold only for |a
k
| = 1, 1 k n, in this case.
2
Remark 3.5 It can be seen from the above proof that Theorem 2.5 could be generalized by re-
placing
n

j=1
(1+|a
j
|)

j
by a function g(|a
1
|, . . . , |a
n
|) such that g
j
(t) := g
1/
j
(|a
1
|, . . . , |a
j1
|, t,
|a
j+1
|, . . . , |a
n
|) is concave and satises the symmetry property tg
j
_
1
t
_
= g
j
(t) (1 j
n), t > 0.
3.4 Proof of Theorem 2.6.
We need to verify rst that
n
= z
n
+ 1. The rest of the proof will then follow by the
same arguments as in the proof of Theorem 2.5. Consider the following best approximation
problem

n
:= min
c
j
I C
_
_
_
_
_
_
z
n
+ 1
n1

j=1
c
j
z
j
_
_
_
_
_
_
. (3.28)
Clearly,
n

n
. Assume rst that is strictly convex. Then (3.28) possesses a unique
solution q(z) :=
n1

j=1
c

j
z
j
. This and the rotation invariance of the norm yield that q(e
2i/n
z) =
q(z) for every |z| = 1, i.e., q 0. This veries that
n
=
n
= z
n
+ 1 when is strictly
convex. In the general case, set
p

= p + p
2
( > 0).
Evidently,

is rotation invariant and strictly convex, i.e., the above argument is

applicable to

. Letting 0, we obtain that

n
=
n
= z
n
+ 1. 2
Acknowledgement. The authors are grateful to Prof. Y. Sarantopoulos for many helpful
discussions concerning the content of this paper.
A SHARP VERSION OF MAHLERS INEQUALITY 13
References
[1] P. B. Borwein, Exact inequalities for the norms of factors of polynomials, Can. J.
Math. 46 (1994) 687698.
[2] P. B. Borwein and T. Erd elyi, Polynomials and Polynomial Inequalities (Springer-
Verlag, New York, 1995).
[3] D. W. Boyd, Sharp inequalities for the product of polynomials, Bull. London Math.
Soc. 26 (1994) 449454.
[4] A. O. Gelfond, Transcendental and Algebraic Numbers (Dover, New York, 1960).
[5] G. Godini, Best approximation in certain classes of normed linear spaces, J. Approx.
Theory 39 (1983) 157171.
[6] K. Mahler, An application of Jensens formula to polynomials, Mathematica 7 (1960)
98100.
[7] Y. Sarantopoulos, Personal communication (1996).
Andr as Kro o Igor E. Pritsker
Mathematical Institute of the Institute for Computational Mathematics
Hungarian Academy of Sciences Department of Mathematics and Computer Science
Budapest, Realtanoda u. 13-15 Kent State University
H-1053, HUNGARY Kent, OH 44242-0001, USA