Power Series, Bieberbach Conjecture and

the de Branges and Weinstein Functions

Wolfram Koepf
Department of Mathematics and Computer Science
University of Kassel
Heinrich-Plett-Str. 40
D-34132 Kassel
Germany
[email protected]

ABSTRACT General Terms

It is well-known that de Branges’ original proof of the Bieber- Algorithms
bach and Milin conjectures

on the coefficients an of univa-
lent functions f (z) = ∞ n
k=1 n z of the unit disk as well as
a
Weinstein’s later proof deal with the same special function
Keywords
system that de Branges had introduced in his work. Bieberbach conjecture, Askey-Gasper inequality, de Branges
These hypergeometric polynomials had been already stud- functions, Weinstein functions, FPS algorithm, algorithmic
ied by Askey and Gasper who had realized their positiveness. computation of formal power series, Sturm sequences.
This fact was the essential tool in de Branges’ proof.
In this article, we show that many identities, e.g. the rep-
resentation of their generating function w.r.t. n, for these
1. INTRODUCTION
polynomials,
which are intimately related to the Koebe func- Let S denote the family of analytic and univalent functions
tion K(z) = ∞ n f (z) = z + a2 z 2 + . . . of the unit disk D. S is compact with
k=1 nz and therefore to univalent functions,
can be automatically detected from power series computa- respect to the topology of locally uniform convergence so
tions by a method developed by the author and accessible that kn := max |an (f )| exists. In 1916 Bieberbach [2] proved
f ∈S
in several computer algebra systems. that k2 = 2, with equality if and only if f is a rotation of
In other words, in this paper a new and interesting ap- the Koebe function
plication of the FPS (Formal Power Series) algorithm is  2 
given. As working engine we use a Maple implementation z 1 1+z ∞
K(z) := = − 1 = nz n , (1)
by Dominik Gruntz and the author. In particular, the hy- (1 − z) 2 4 1−z n=0
pergeometric representation of both the de Branges and the
Weinstein functions are determined by successive power se- and in a footnote he mentioned “Vielleicht ist überhaupt
ries computations from their generating functions. kn = n.”.1 This statement is known as the Bieberbach
The new idea behind this algorithm is the observation that conjecture.
hypergeometric function coefficients of double series can be In 1923 Löwner [14] proved the Bieberbach conjecture for
automatically detected by an iteration of the FPS procedure. n = 3. His method was to embed a univalent function f (z)
In a final section we show how algebraic computation en- into a Löwner chain, i.e. a family {f (z, t) | t
0 } of univa-
ables the fast verification of Askey-Gasper’s positivity re- lent functions of the form
sults for specific (not too large) n using Sturm sequences or 
∞
similar approaches. f (z, t) = et z+ an (t)z n , (z ∈ D, t
0, an (t) ∈ C (n
2))
n=2

Categories and Subject Descriptors which start with f

G.4 [Mathematical Software]: Algorithm design and anal-
ysis; G.2.1 [Discrete Mathematics]: Combinatorics—Gen- f (z, 0) = f (z) ,
erating functions, Recurrences and difference equations and for which the relation
 ˙ 
f (z, t)
Re p(z, t) = Re >0 (z ∈ D) (2)
zf  (z, t)
Permission to make digital or hard copies of all or part of this work for
personal or classroom use is granted without fee provided that copies are is satisfied. Here  and ˙ denote the partial derivatives with
not made or distributed for profit or commercial advantage and that copies respect to z and t, respectively. Equation (2) is referred to as
bear this notice and the full citation on the first page. To copy otherwise, to
republish, to post on servers or to redistribute to lists, requires prior specific
the Löwner differential equation, and geometrically it states
permission and/or a fee. that the image domains of f (D, t) expand as t increases.
ISSAC’03, August 3–6, 2003, Philadelphia, Pennsylvania, USA. 1
Copyright 2003 ACM 1-58113-641-2/03/0008 ...$5.00. Perhaps kn = n is generally valid.

169
 The history of the Bieberbach conjecture showed that it As working engine we use a Maple implementation by Do-
was easier to obtain results about the logarithmic coeffi- minik Gruntz and the author. In particular, the hyperge-
cients of a univalent function f , i.e. the coefficients dn of ometric representations of the de Branges and Weinstein
the expansion functions are determined by successive power series compu-
tations from their generating functions.
f (z) ∞
ϕ(z) = ln =: dn z n
z n=1 2. AUTOMATIC COMPUTATION OF POWER
rather than for the coefficients an of f itself. So Lebedev and SERIES COEFFICIENTS
Milin [13] in the mid sixties of the last century developed In this section, we review the FPS algorithm enabling the
methods to exponentiate such information. They proved automatic computation of power series as designed in [8].
that if for f ∈ S the Milin conjecture Given an expression f (x) in the variable x, one would like
n   to find the Taylor series
2 4
(n + 1 − k) k|dk | − 0 
∞
k f (x) = Ak xk ,
k=1
k=0
on its logarithmic coefficients is satisfied for some n ∈ N,
then the Bieberbach conjecture for the index n + 1 follows. i.e., a formula for the coefficient Ak . For example, if f (x) =
In 1984 de Branges [3] verified the Milin, and therefore exp(x), then
the Bieberbach conjecture, and in 1991, Weinstein [16] gave 
∞
1 k
a different proof. Both proofs use special function systems, f (x) = x ,
k!
and independently, Todorov [15] and Wilf [17] discovered k=0
that these essentially are the same, see also [11], [12]. 1
hence Ak = k! .
In [5], Zeilberger showed how parts of Weinstein’s proof If the result is simple enough, the FPS (formal power se-
version can be principally computerized. The main argu- ries) procedure of the Maple package FPS.mpl ([8], [6], [7])
ment is the following: The coefficients Ck,n (x) of the func- computes this series, even if it is a Laurent series (includ-
tion ing negative powers) or Puiseux series (including rational
1 powers).
  =
The main idea behind this procedure is
1 − z 2x2 + (1 − z 2 )(w + 1/w) + z 2
1. to compute a holonomic differential equation for f (x),
∞ n i.e., a homogeneous linear differential equation with
(n − k)!
(1 − x2 )k Ck,n (x)(wk + w−k )z n polynomial coefficients,
n=0
(n + k)!
k=0
2. to convert the differential equation to a holonomic re-
are polynomials Ck,n (x) ∈ Q[x] with rational coefficients. currence equation for Ak ,
To prove that these form the squares of another system of
3. and to solve the recurrence equation for Ak .
polynomials Dk,n ∈ Q[x],
The above procedure is successful at least is f (x) is hy-
Dk,n (x)2 = Ck,n (x) , pergeometric. A series
Zeilberger suggests to calculate the first polynomials Dk,n (x) 
∞

for 0 ≤ k ≤ n ≤ 20, then to “guess” a holonomic recurrence ak

equation w.r.t. n k=0

is called hypergeometric, if the series coefficients ak have

σ2 Dk,n+2 (x) + σ1 Dk,n+1 (x) + σ0 Dk,n (x) = 0
rational term ratio
with polynomials σj ∈ Q[k, n] of fixed degrees w.r.t. k and ak+1
∈ C(k) .
n, and to use linear algebra to find σ0 , σ1 , σ2 . In a final ak
part, the initial guess can be a posteriori verified by the The function
WZ method [18] finishing this part of Weinstein’s proof.2  
Note, however, that this approach is much different from the a1 , a2 , . . . , ap
p Fq x :=
approach of the given paper which mechanizes other parts b1 , b2 , . . . , b q
of the given proofs.
In this article, we show that many identities, e.g. the rep- 
∞ 
∞
(a1 )k · (a2 )k · · · (ap )k xk
resentation of their generating function, for the de Branges Ak xk =
(b1 )k · (b2 )k · · · (bq )k k!
and Weinstein polynomials, which are intimately related to k=0 k=0
the Koebe function and therefore to univalent function the- is called the generalized hypergeometric series, since its term
ory, can be automatically detected from power series compu- ratio
tations by a method developed by the author and accessible
Ak+1 xk+1 (k + a1 ) · · · (k + ap ) x
in several computer algebra systems. =
Ak xk (k + b1 ) · · · (k + bq ) (k + 1)
2
Although the ideas of the article [5] are very interesting, the
is a general rational function, in factorized form. Here (a)k =
presentation of some details is not completely correct. Hence
we refer the reader to my review 0894.30013 in Zentralblatt a(a + 1) · · · (a + k − 1) denotes the Pochhammer symbol or
für Mathematik which contains a corrected version which shifted factorial. The summand ak of the generalized hyper-
was approved by Zeilberger. geometric series is called a hypergeometric term.

170
3. THE LÖWNER CHAIN OF THE KOEBE Note that the above series result can be used to find a hy-
FUNCTION pergeometric representation for the Taylor coefficients An (y)
of the function
In this section, we consider the Löwner chain

∞
−1 −t w(z, y) = An (y) z n .
W (z, t) := K e K(z) (z ∈ D, t
0) (3)
n=0

of bounded univalent functions in the unit disk D which is Writing

defined in terms of the Koebe function (1). Since K maps

∞
(n)
the unit disk onto the entire plane slit along the negative x- An (y) = aj yj ,
axis in the interval (−∞, 1/4], W (D, t) is the unit disk with j=0
a radial slit increasing with t.
A computation shows that one arrives at the double sum

∞ 
∞
4e−t z w(z, y) = aj
(n)
yj zn . (4)
W (z, t) = 2
1−z+ 1 − 2(1 − 2e−t )z + z2 n=0 j=0

By another application of the FPS algorithm, the coefficients

(see [11], after Eq. (47)). (n)
aj of this double series can be determined, which therefore
As a first application of the FPS algorithm, we compute yields a representation for An (y). We compute
the Taylor series of w(z, y) = W (z, − ln y), considered as > ss:=standardsum(FPS(s,z,i));
function of the variable y := e−t , ∞ 
 
∞
2 (−1)j (2 j)! (2 j + 1) y j y (2 j + 2)i z (j+i+1)
4yz ss := ( )
w(z, y) = 2 , (j!)2 (j + 1) (j + 2) i!
1−z+ 1 − 2(1 − 2y)z + z 2 j=0 i=0
By (4), the exponent of z is n, hence we substitute i by
which turns out to be hypergeometric: n−j−1
> read "FPS.mpl";
> summand:=subs(i=n-j-1,op([1,1],ss));

Package Formal Power Series, Maple V-8 2 (−1)j (2 j)! (2 j + 1) y j y (2 j + 2)n−j−1 z n

summand :=
Copyright 1995, Dominik Gruntz, University of Basel (j!)2 (j + 1) (j + 2) (n − j − 1)!
Copyright 2002, Detlef Müller & Wolfram Koepf, to get the hypergeometric representation for An (y)
University of Kassel > sumtools[sumtohyper](summand,j);
> assume(z<1,z>0); interface(showassumed=0); y n z n hypergeom([n + 1, 1 − n], [3], y)
> w:=4*y*z/(1-z+sqrt(1-2*(1-2*y)*z+z^2))^2: Hence our computations have generated the result
> s:=standardsum(FPS(w,y,j)); ∞
2 (−1)j (2 j)! (2 j + 1) (2 j + 2)n−j−1 j+1
  An (y) = y
j 1 j (j+1) (j+1) (j!)2 (j + 1) (j + 2) (n − j − 1)!
∞
2 (−1) ( ) (1 + 2 j)! z y 
j=0
 (z − 1)2   
s :=   n + 1, 1 − n
(z − 1) (j!) (j + 1) (j + 2)
2 2
j=0 = n y 2 F1 y
3
Note that standardsum puts the sum signs in front of the
expression. in agreement with [11], Eq. (19).
Essentially by computing and simplifying the coefficient The above method can be summarized in
term ratio, from the above result one gets the hypergeomet-
Theorem 1. Assume w(z, y), given as function in z and
ric representation for w(z, y) ([9], [10], Chapter 2)3 (n)
y, can be written as double sum (4) with coefficients aj
> factor(sumtools[Sumtohyper](op(1,s),j)); that are hypergeometric terms w.r.t. both summation vari-
−1 4zy ables j and n. Then a double application of the FPS algo-
(z − 1)2 Hypergeom([ ], [], − ) (n)
1 2 (z − 1)2 rithm, applied to w(z, y), computes aj , and hence gives the
−
2 zy hypergeometric representation
To obtain this representation, as an intermediate result

∞
(n)
FPS uses the differential equation An (y) = aj yj
> DE:=HolonomicDE(w,W(y)); j=0

2 d2 2
DE := y (1 − 2 z + 4 z y + z ) ( 2 W(y))− for the coefficient functions An (y) that are generated by
dy w(z, y), i.e.

d 
∞
(−1 + z)2 W(y) + (6 z y + 1 + z 2 − 2 z) y ( W(y)) = 0 w(z, y) = An (y) z n .
dy n=0
for w(z, y). We would like to mention how efficiently the above result
3 was obtained compared to the lengthy and complicated de-
Sumtohyper is the inert form of sumtohyper which outputs
the inert form Hypergeom for the hypergeometric series to duction given in [11]. This statement is valid for all results
prevent evaluation. of this paper.

171
4. THE DE BRANGES AND WEINSTEIN which starts for j = 0, we put the factor y k+1 in front of
FUNCTIONS FPS and compute
In [3] de Branges showed that the Milin conjecture is valid
if for all n
2 the de Branges functions τkn : R
0 → R (k = > s:=standardsum(y^(k+1)*
1, . . . , n) defined by the system of differential equations > FPS(1/y*(w/y)^(k+1)/(1-w^2),y,j));
j (j+1+k) 1
τ̇kn (t) τ̇ n (t) 
k
∞ y y (−1) z (1 + 2 k)2j ( )(k+j) y (j−1)
n
τk+1 (t) − τkn (t) = + k+1 (k = 1, . . . , n) (z − 1)2
k k+1 s :=
j=0
(z − 1)2 (1 + 2 k)j j!
with the initial values
Note that the differential equation w.r.t. y
τkn (0) = n + 1 − k
have the properties > DE1:=HolonomicDE(1/y*w^(k+1)/(1-w^2),L(y));
lim τkn (t) =0, (5) DE1 := (2 z y − k2 + 2 z k2 − z 2 k2 ) L(y)
t→∞

and +y (10 z y + 1 + z 2 − 2 z) ( dy
d
L(y))
2
τ̇kn (t)  0 (t ∈ R
0 ) . (6) +y 2 (1 − 2 z + 4 z y + z 2 ) ( dy
d
2 L(y)) = 0

for the generating function Lk (z, t) leads to the above hy-

Whereas the relation (5) is easily checked using standard pergeometric result.
methods for ordinary differential equations, (6) is a deep An iterated power series computation w.r.t. z gives
result.
L. de Branges gave an explicit representation of the func-
tion system τkn (t) [3] with which the proof of the de Branges ss:=FPS(s,z,i);
>
theorem was completed as soon as de Branges realized that 
(6) was a theorem previously proved by Askey and Gasper 
∞ ∞
[1]. ss :=  (
On the other hand, Weinstein [16] uses the Löwner chain j=0 i=0
(3), and shows the validity of Milin’s conjecture if for all (i+1+k+j)
n
2 the Weinstein functions Λn 2 (−1) z j
Γ(2 j + 2 k) y (k+j) (2 j + 2 + 2 k)i j
k : R
0 → R (k = 1, . . . , n)
defined by j! (j + 2 k) Γ(j + 2 k) i!

t
e W (z, t) k+1 
∞ 2 (−1)j z (i+1+k+j) y (k+j) k (2 j + 2 + 2 k)i Γ(2 j + 2 k) 
=: Λn
k (t)z
n+1
= Lk (z, t) (7) + )
1 − W 2 (z, t) Γ(j + 2 k) (j + 2 k) j! i!
n=k
with summand
satisfy the relations

k (t)
0
Λn (t ∈ R
0 , k, n ∈ N) . (8)
> summand:=op([1,1],ss):
Weinstein did not identify the functions Λn
k (t),
but was able
to prove (8) without an explicit representation.
Independently, both Todorov [15] and Wilf [17] proved— > summand:=simplify(subs(i=n-j-k,summand));
using the explicit representation of the de Branges func- (−1)j z (n+1) y (k+j) Γ(j + 2 + k + n)
tions to be developed in Section 6—that for all n ∈ N, summand :=
(2j +1+2k)Γ(j +1+2k)Γ(j +1)Γ(n−j −k+1)
k = 1, . . . , n, one has the identity This finally yields the hypergeometric representation for Λn
k
τ̇kn (t) = −kΛn
k (t) , (9)
i.e. the de Branges and the Weinstein functions essentially > result1:=convert(
> sumtools[sumtohyper](summand,j),binomial);
are the same, and the main inequalities (6) and (8) are
identical. Note that in [11] an algebraic proof of (9) was result1 := z (n+1) y k binomial(k + n + 1, 1 + 2 k)
given which does not use the explicit representation of the
1 3
de Branges functions. hypergeom([ + k, k − n, 2 + k + n], [1 + 2 k, + k], y)
2 2
showing that
5. CLOSED FORM REPRESENTATION OF
THE WEINSTEIN FUNCTIONS    1

2
+k, k−n, 2+k+n
k+n+1
3 F2
−kt −t 
In this section, we use Theorem 1 to generate the hyper- Λn
k (z, t) =e 3 e ,
geometric representation for the Weinstein functions Λn 1 + 2k 1 + 2 k, + k
k (t) 2
directly from their defining generating function (7). (10)
To generate a power series w.r.t. y for the generating func- see e.g. [11], Section 5. The plot
tion of the Weinstein functions
1 w(z, y)k+1  ∞
· = αj y j > plot(simplify(
y 1 − w(z, y)2 j=0
> subs(y=(1-x)/2,z=1,n=30,k=6,result1)),x=-1..1);

172
 for w(z, y)k is used.
2.5 At this point we would like to mention that this differen-
tial equation can also be automatically determined applying
a version of Zeilberger’s algorithm (see e.g. [10], Chapter
2 7) to the summand of the hypergeometric representation of
w(z, y):
1.5 > read "hsum6.mpl";
Package ”Hypergeometric Summation”, Maple V-8
1 Copyright 1998-2002, Wolfram Koepf, University of Kassel
> DE2:=sumdiffeq(op(1,s),j,B(y));
0.5 2
DE2 := −(1 − 2 z + 4 z y + z 2 ) y 2 ( dy
d
2 B(y))

−y (z 2 + 1 + 6 z y − 2 z) ( dy
d
B(y)) + k2 (−1 + z)2 B(y) = 0
–1 –0.8 –0.6 –0.4 –0.2 0 0.2 0.4 0.6 0.8 1 By (11), a second application of the FPS procedure leads
x
to a hypergeometric representation of τkn :
of Λ30
6 with y =
1−x
2
for x = −1, . . . , 1 shows the nonneg- > ss:=standardsum(FPS(s,z,i));
ativity of this polynomial. Moreover the highly oscillatory

∞
character of the Weinstein functions, resp. the derivatives of ss :=
the de Branges functions, can be seen. j=0
∞ 
 2 (−1)j y k y j k Γ(2 j + 2 k) (2 j + 2 + 2 k)i z (j+i+1+k)
6. GENERATING FUNCTION OF THE DE ( )
Γ(j + 2 k) (j + 2 k) j! i!
BRANGES FUNCTIONS i=0

In [11], a very simple generation function for the de Branges

> summand:=op([1,1],ss):
function was given: > summand:=subs(i=n-j-k,summand):

∞ > result:=sumtools[sumtohyper](summand,j):
Bk (z, t) = τkn (t) z n+1 = K(z) W (z, t)k (11) > result:=convert(result,binomial);
n=k
 
result := y k z (n+1) binomial(1 + k + n, 1 + 2 k)
−kt k, k + 1/2
= K(z) k+1
e 2 F1 −4K(z)e−t
(12) 1 3
2k + 1 hypergeom([ + k, n + k + 2, k, −n + k], [k + 1, 1 + 2 k, + k], y)
2 2
  Therefore, we have finally computed that
∞
2k 2j − 1  
= (−1)j+k K(z)j+1 e−jt . (13) 1 + k + n
j+k j−k n
τk (t) = e−kt
·
j=k
1+2k
Equation (11) shows the intimate relation of the de Branges  1 
functions with the Koebe function: Their generating func- + k, n + k + 2, k, −n + k −t
4 F3  e .
2
tion is basically a power of the Löwner chain of the Koebe 3
k + 1, 1 + 2 k, + k
function. 2
By Theorem 1 we can obtain these representations auto- By taking derivative w.r.t t, this representation of τkn as
matically. Note that these results include a hypergeometric 4 F3 -hypergeometric function is easily seen to be equivalent
representation of the k-th power of w(z, y), extending the to (10). Hence our computations have proved identity (9).
result for k = 1 given in Section 3. The plot
We compute for K(z) w(z, y)k > plot(simplify(
> s:=standardsum(y^k*FPS(z/(1-z)^2*(w/y)^k,y,j)); > subs(y=(1-x)/2,z=1,n=30,k=6,result)),x=-1..1);

j (j+1+k) 1
k
∞ y (−1) z
 ( )(k+j) (2 k)2 j y j
(−1 + z)2 25
s :=
j=0
(−1 + z)2 (1 + 2 k)j j!
and therefore get (13), which by the calculation 20
> sumtools[Sumtohyper](op(1,s),j);
1 1 4zy 15
y k z (k+1) ( )k Hypergeom([k, + k], [1 + 2 k], − )
(−1 + z)2 2 (−1 + z)2
(−1 + z)2 10
gives the hypergeometric representation (12) of K(z) w(z, y)k .
As intermediate result, the differential equation
5
> DE:=HolonomicDE(z/(1-z)^2*w^k,B(y));
2
DE := (1 − 2 z + 4 z y + z 2 ) y 2 ( dy
d 2 2
2 B(y)) − k (−1 + z) B(y)
–1 –0.8 –0.6 –0.4 –0.2 0 0.2 0.4 0.6 0.8 1
+ y (6 z y + 1 − 2 z + z 2 ) ( dy
d
B(y)) = 0 x

173
shows the nonnegativity of τ630 . Morover it indicates that Using Mathematica’s CountRoots command which is based
τkn is much less oscillatory than its derivative. on a more efficient approach than Sturm sequences [4], the
author could show relations (6) for n = 100, k = 1, . . . , 100,
e.g., hence Bieberbach’s conjecture for n = 101, within 110
7. FURTHER RESULTS seconds (Intel Pentium III, 1066 MHz under Windows 2000).
In this section we briefly mention some more examples for
the use of Theorem 1 which, although not directly connected
to the proof of the Bieberbach conjecture, still deal with the
9. CONCLUSION
Löwner chain w(z, y) of the Koebe function. These examples In this article, we show how the algorithmic computation
emphasize the rich structure of this function. of power series can be applied to generating functions that
For this purpose we present some generating functions appear naturally in the study of the Koebe function, to ob-
with hypergeometric representations, and leave the respec- tain hypergeometric representations for the de Branges and
tive second application of the FPS algorithm to the reader: Weinstein functions. Moreover, we have considered how al-
m
1+w(z,y) gebraic computation enables the fast verification of the pos-
• generating function 1−w(z,y) :
itivity results in the proofs of de Branges and Weinstein for
> standardsum(FPS(((1+w)/(1-w))^m,y,j)); fixed n and k using Sturm sequences or similar approaches.
j 1 m
∞ z (
 )j (−1)j 4j (− )j y j 10. REMARKS
(−1 + z)2 2
j! The Maple packages FPS.mpl and hsum6.mpl can be down-
j=0
1−w(z,y)
loaded from the author’s web site http:/www.mathematik.
• generating function w(z, y)k 1+w(z,y)
: uni-kassel.de/~koepf/Publikationen, where also a cor-
> standardsum(y^k*FPS((w/y)^k*(1-w)/(1+w),y,j)); responding Maple Help Database is available.
j (j+k) 1
k
∞ y (−1) z
 ( )(j+k) (1 + 2 k)2 j y j 11. REFERENCES
(−1 + z)2
(1 + 2 k)j j! [1] Askey, R. and Gasper, G.: Positive Jacobi polynomial
j=0 sums II. Amer. J. Math. 98, 1976, 709–737.
w(z,y)k
• generating function 1−w(z,y) m: [2] Bieberbach, L.: Über die Koeffizienten derjenigen
Potenzreihen, welche eine schlichte Abbildung des
> standardsum(y^k*FPS((w/y)^k/(1-w)^m,y,j));
Einheitskreises vermitteln. Semesterberichte Preuss.
y k
(−1) j (j+k)
z (
1
) (j+k)
(2 k − m) y j Akad. Wiss. 38, 1916, 940–955.
∞
(−1 + z)2
2j
[3] De Branges, L.: A proof of the Bieberbach conjecture.
j=0
(−m + 2 k + 1)j j! Acta Math. 154, 1985, 137–152.
w(z,y)k 1
[4] Collins, G. E. and Krandick, W.: An efficient
• generating function 1+w(z,y) 1−w(z,y) m: algorithm for infallible polynomial complex roots
> standardsum(y^k*FPS((w/y)^k/(1-w)^m/(1+w),y,j)); isolation. In: Wang, Paul S. (Ed.): Proceedings of
1 ISSAC’92, 1992, 189–194.
j (j+k)

k
∞ y (−1) z ( 2
)(j+k) (2 k − m)2 j y j [5] Ekhad, S. B. and Zeilberger, D.: A high school
(−1 + z)
algebra, “formal calculus”, proof of the Bieberbach
j=0
(2 k − m)j j! conjecture [after L. Weinstein]. Jerusalem
Combinatorics’93, Contemp. Math. 178, Amer. Math.
8. POSITIVITY AND STURM SEQUENCES Soc., Providence, 1994, 113–115. Review in
When de Branges had found his function system τk (t), he n Zentralblatt für Mathematik 0894.30013,
was able to check the Bieberbach conjecture by hand com- [6] Gruntz, D. and Koepf, W.: Maple package on formal
putations for n  6. For this purpose he proved the non- power series. Maple Technical Newsletter 2 (2), 1995,
positivity of τ̇kn (t) for 1  n  5, 1  k  n, t
0. 22–28.
He asked his colleague Walter Gautschi from Purdue Uni- [7] Gruntz, D., Koepf, W. and Müller, D.: FPS.mpl for
versity to verify (6) numerically, which was done for n  30, Maple 8, http:/www.mathematik.uni-kassel.de/
and de Branges became confident of the validity of the gen- ~koepf/Publikationen, 2002.
eral statement (6). [8] Koepf, W.: Power series in computer algebra. Journal
Note, however, although Gautschi used Sturm sequences, of Symbolic Computation 13, 1992, 581–603.
because of the oscillatory nature of Λn k his numerical com- [9] Koepf, W.: Algorithms for m-fold hypergeometric
putations had to be very careful to obtain correct results. summation. Journal of Symbolic Computation 20,
Nowadays, we can apply Sturm sequences and count the 1995, 399–417.
roots easily by rational arithmetic to obtain correct count- [10] Koepf, W.: Hypergeometric Summation. Vieweg,
ings since the input polynomials are elements of Q[y]. Braunschweig/Wiesbaden, 1998.
For example, the computation [11] Koepf, W. and Schmersau, D.: On the de Branges
> sturm(simplify( Theorem. Complex Variables 31, 1996, 213–230.
> subs(y=(1-x)/2,z=1,n=30,k=6,result1)),x,-1,1); [12] Koepf, W. and Schmersau, D.: Weinstein’s functions
1 and the Askey-Gasper identity. Integral Transforms
proves that Λ30 6 (x) has exactly one zero in the semi-open and Special Functions 5, 1997, 227–246.
interval (−1, 1] (namely at the point x = 1), and since its [13] Lebedev, N. A. and Milin, I. M.: An inequality.
value at x = −1 is positive, Λ30 6 (x) is nonnegative in [−1, 1]. Vestnik Leningrad Univ. 20, 1965, 157–158 (Russian).

174
[14] Löwner, K.: Untersuchungen über schlichte konforme [17] Wilf, H.: A footnote on two proofs of the
Abbildungen des Einheitskreises I. Math. Ann. 89, Bieberbach-de Branges Theorem. Bull. London Math.
1923, 103–121. Soc. 26, 1994, 61–63.
[15] Todorov, P.: A simple proof of the Bieberbach [18] Wilf, H. and Zeilberger, D.: An algorithmic proof
conjecture. Bull. Cl. Sci., VI. Sér., Acad. R. Belg. 3 theory for hypergeometric (ordinary and “q”)
(12), 1992, 335–346. multisum/integral identities. Invent. Math. 103, 1992,
[16] Weinstein, L.: International Mathematics Research 575–633.
Notices 5, 1991, 61–64.

175