Izvestiya: Mathematics 66:3 489542
2002

c RAS(DoM) and LMS

Izvestiya RAN : Ser. Mat. 66:3 49102 DOI 10.1070/IM2002v066n03ABEH000387

Irrationality of values of the Riemann zeta function

W. Zudilin

Abstract. The paper deals with a generalization of Rivoals construction, which

enables one to construct linear approximating forms in 1 and the values of the
zeta function (s) only at odd points. We prove theorems on the irrationality of the
number (s) for some odd integers s in a given segment of the set of positive integers.
Using certain rened arithmetical estimates, we strengthen Rivoals original results
on the linear independence of the (s).

Introduction
The study of the arithmetical nature of the values of the Riemann zeta function


1
(s) = s
n=1
n

at integers s > 1 is one of the most attractive topics of the modern number theory.
In spite of the deceptive simplicity of this problem, the results obtained in this area
in the course of more than two centuries are far from being exhaustive. Eulers
formula
(2i)s Bs
(s) = , s = 2, 4, 6, . . .,
2s!
in which the values of the zeta function at even integers are expressed in terms
of 3.1415926 and the Bernoulli numbers Bs Q, which are dened by the
generating function

t t
ts
=1 + Bs , Bs = 0 for odd s
3,
e 1
t 2 s=2 s!

undoubtedly marked the rst progress in this area. In 1882 Lindemann proved
that is a transcendental number, which implies that (s) is transcendental if s is
even.
The problem of the irrationality of the values of the zeta function at odd integers
seemed inaccessible until 1978, when Apery [1] produced a sequence of rational
approximations proving that (3) is irrational.

This research was supported in part by INTAS and the Russian Foundation for Basic Research
(grant no. IR-97-1904).
AMS 2000 Mathematics Subject Classication. 11J81, 11M06.
490 W. Zudilin

Ap
erys Theorem. The number (3) is irrational.
The history of this discovery and a rigorous mathematical justication of Aperys
observations can be found in [2]. The phenomenon of Aperys sequence was repeat-
edly interpreted from the point of view of various analytical methods of number
theory (see [3][8]). New approaches made it possible to strengthen Aperys result
quantitatively, that is, to obtain a good measure of the irrationality of (3) ([9]
and [10] represent the latest stages of the competition in this area).
Unfortunately, natural generalizations of Aperys construction involve linear
forms containing values of the zeta function at both odd and even integers (the
interest in which faded after the EulerLindemann result). This circumstance
makes it impossible to obtain results on the irrationality of (s) for odd s
5
by this method. Interesting attempts to approach this problem can be found in the
preprints [11] and [12].
Finally, in 2000 Rivoal [13] equipped an auxiliary rational function with a sym-
metry and constructed linear forms containing the values of the zeta function only
at odd integers s > 1.
Rivoals Theorem. The sequence (3), (5), (7), . . . contains innitely many
irrational numbers. More precisely, the following estimate holds for the dimen-
sion (a) of the spaces generated over Q by 1, (3), (5), . . ., (a 2), (a) with an
odd integer a:
log a  
(a)
1 + o(1) , a .
1 + log 2
In this paper we generalize Rivoals construction [13] and prove the following
theorems.
Theorem 0.1. Each of the sets

{(5), (7), (9), (11), (13), (15), (17), (19), (21)},

(0.1)
{(7), (9), . . . , (35), (37)}, {(9), (11), . . . , (51), (53)}

contains at least one irrational number.1

Theorem 0.2. For every odd integer b
1 at least one of the numbers

(b + 2), (b + 4), . . . , (8b 3), (8b 1)

is irrational.
Theorem 0.3. There are odd integers a1  145 and a2  1971 such that 1, (3),
(a1 ), (a2 ) are linearly independent over Q.
Theorem 0.3 strengthens the corresponding theorem in [15], where it was estab-
lished that 1, (3), (a) with some odd integer a  169 are linearly independent.
1 When the work on this paper was nished, I was informed that Rivoal [14] had obtained the
assertion of Theorem 0.1 for the rst set in (0.1) independently, using another generalization of
the construction in [13].
Added in proof. I have recently obtained some renements of Theorem 0.1. In particular, I
have shown [28] that at least one of the four numbers (5), (7), (9), (11) is irrational.
 Irrationality of values of the zeta function 491

Theorem 0.4. The following absolute estimate holds for every odd integer a
3:

2 log a
(a) > 0.395 log a > . (0.2)
3 1 + log 2

This paper is organized as follows. In 1 we describe an analytical construction

of linear forms in the values of the zeta function at odd integers. In 24 we
study the asymptotics of linear forms, their coecients and denominators. Finally,
in 5 we give complete proofs of the results. The main ingredient in the proofs of
Theorems 0.3 and 0.4 (as well as in [13]) is the following special case of Nesterenkos
theorem [16].
Criterion for linear independence. Assume that for a given set of real numbers
0 , 1 , . . . , m , m
1, one can nd a sequence of linear forms

In = A0,n 0 + A1,n 1 + + Am,n m , n = 1, 2, . . . ,

with integer coecients and numbers > 0 and > 0 such that
 
log |In | = n + o(n), log max |Aj,n|  n + o(n)
0
j
m

as n . Then

dimQ (Q0 + Q1 + + Qm )
1 + .

Let us note that the justication of Theorems 0.10.4 is based on the saddle-point
method and follows the scheme of proof of Aperys theorem in [8]. It should be
noted that some arithmetical results for the values of polylogarithms were obtained
in [17], where the saddle-point method was used. The dissertation [17] served a
link between our empirical observations and the rigorous justication in 2. In 4
we improve the arithmetical estimates (for the denominators of numerical linear
forms) following [18], [9], [10], which enables us to rene the lower estimate for (a)
in Theorems 0.3 and 0.4 for small values of a. Finally, in 3 we obtain an upper
estimate for the coecients of linear forms and asymptotics of their growth.
The main results of this paper were announced in [19].
I am grateful to Professor Yu. V. Nesterenko for his constant interest and valuable
advice, which have enabled me to improve this paper.

1. An analytical construction
In this section we describe an analytical construction that enables us to obtain
good linear forms with rational coecients in the values of the zeta function at
odd integers.
Let us x positive integer parameters a, b, r such that 2rb  a and a + b is even.
For every positive integer n we consider the rational function
 b
(t (n + 1)) (t (n + 2rn))
R(t) = Rn (t) :=  a (2n)!a2rb . (1.1)
t(t 1) (t n)
492 W. Zudilin

Here and below (t l) means that the product (sum or set) contains the factors
(summands or elements) t l and t + l. We assign to the function dened by (1.1)
the sum


1 db1 R(t)
I = In := . (1.2)
t=n+1
(b 1)! dtb1

It is easy to see that the sum in (1.2) is taken only over integers t > n + 2rn.
The series on the right-hand side of (1.2) converges absolutely, since the degrees of
the numerator and the denominator of the rational function (1.1) are equal to 4rbn
and a(2n + 1)
4rbn + a
4rbn + 2, respectively, whence
   
1 db1 R(t) 1
R(t) = O 2 and = O , t . (1.3)
t dtb1 t2

Applying Laplaces method to the integral representation of the sum in formula (1.2),
we can calculate the quantity
log |In |
= lim > (1.4)
n n
(the details can be found in 2 below), following [8] and [17].
Lemma 1.1. For every n = 1, 2, . . . the number dened by formula (1.2) is a linear
form in 1 and the values of the zeta function at odd integers s, b < s < a + b:


I= As (s) A0 . (1.5)
s is odd
b<s<a+b

Proof. Decomposing the function (1.1) into a sum of partial fractions

n 


Ak,a Ak,2 Ak,1
R(t) = ++ + , (1.6)
(t + k)a (t + k)2 t + k
k=n

we note that


n
Ak,1 = Res R(t) = 0 (1.7)
t=
k=n

by (1.3). Since the function (1.1) is even (odd), that is,

R(t) = (1)a R(t), (1.8)

and the decomposition (1.6) is unique, we have

Ak,j = (1)aj Ak,j , k = 0, 1, . . . , n, j = 1, 2, . . . , a.

Hence,


n

n
Ak,j = (1)aj Ak,j = 0 if a j is odd, j = 2, . . . , a. (1.9)
k=n k=n
 Irrationality of values of the zeta function 493

Substituting (1.6) into (1.2), we obtain that

n  
b1 a+b2 Ak,a
(1) I=
t=n+1 k=n
b1 (t + k)a+b1
    
a+b3 Ak,a1 b Ak,2 Ak,1
+ ++ + .
b1 (t + k)a+b2 b 1 (t + k)b+1 (t + k)b

Simple transformations yield the formula

I = Aa+b1 (a + b 1) + Aa+b2 (a + b 2) + + Ab (b) A0 , (1.10)

where
 

n
b1 j 1

Aj = (1) Ak,jb+1, j = b, b + 1, . . . , a + b 1, (1.11)
b1
k=n

a + b 2 Ak,a
n k+n 
b

Ak,2 Ak,1

A0 = (1)b1 ++ + b .
b1 la+b1 b 1 lb+1 l
k=n l=1
(1.12)

By (1.7), we have Ab = 0. Hence, the expression on the right-hand side of (1.10)

is well dened even in the case when b = 1. Since a+b is even, formula (1.9) implies
that Aj = 0 if j
b is even. Hence, formula (1.10) means that I is a linear form
in 1 and the values of the zeta function at odd integers s, b + 1  s  a + b 1, as
was to be shown.
The following lemma enables us to calculate the denominators of linear
forms (1.2).
Lemma 1.2 (cf. [20], [13]). Assume that for some polynomial P (t) of degree  n
the rational function

P (t)
R(t) = (1.13)
(t + s)(t + s + 1) (t + s + n)

(this representation does not have to be irreducible) satises the conditions

 
R(t)(t + k) t=k Z, k = s, s + 1, . . . , s + n. (1.14)

Then 
Dnj dj  
R(t)(t + k)  Z, k = s, s + 1, . . . , s + n,
j! dtj t=k

for all non-negative integers j, where Dn is the least common multiple of 1, 2, . . . , n.

Proof. Decomposing the rational function (1.13) into a sum of partial fractions, we
obtain that

Bl
s+n
R(t) = ,
t+l
l=s
494 W. Zudilin

where  
Bl = R(t)(t + l) t=l Z, l = s, s + 1, . . . , s + n.
Therefore,


s+n
t+k

s+n
kl

R(t)(t + k) = Bk + Bl = Bk + Bl 1 + ,
t+l t+l
l=s l=s
l=k l=k

whence

Bl (k l) 
Bl
1 dj   s+n s+n
R(t)(t + k)  = (1)j  = .
j! dtj  (t + l)j+1  (k l)j
t=k l=s t=k l=s
l=k l=k

This completes the proof of the assertion.

Remark 1.1. Lemma 1.2 was stated by Nesterenko [21]. Nikishin [20] applied it to
the polynomial
P (t) = (t + p + 1) (t + p + n)
with p s Z and p + n < s or p
s + n. Rivoal observed [13] that (1.14) holds
for P (t) = n! . In each of these cases Bk , k = s, s + 1, . . . , s + n, are binomial
coecients, and (1.14) is easily veried.
It is easy to see that the rational function (1.1) can be written as a product of
the functions
(t (m + 1)) (t (m + 2n))
F (t) = Fn,m (t) :=  2 , n  m, (1.15)
t(t 1) (t n)
(2n)!
H(t) = Hn (t) := (1.16)
t(t 1) (t n)
(see formula (1.22) below).
Lemma 1.3. The following inclusions hold for functions (1.15) and (1.16):

j
D2n dj  
2 
F (t)(t + k)  Z, k = 0, 1, . . . , n, j = 0, 1, 2, . . .,
j! dtj t=k (1.17)

j
D2n d j  
j
H(t)(t + k)  Z, k = 0, 1, . . . , n, j = 0, 1, 2, . . .,
j! dt t=k (1.18)

where D2n is the least common multiple of 1, 2, . . . , 2n.

Proof. The inclusions (1.18) follow from Lemma 1.2 applied to the rational func-
tion (1.16). We prove (1.17) using Lemma 1.2 for the rational functions
(t (m + 1)) (t (m + 2n))
R (t) = ,
t(t 1) (t n)
(1.19)
(t + (m + 1)) (t + (m + 2n))
R+ (t) =
t(t 1) (t n)
 Irrationality of values of the zeta function 495

and the Leibniz rule for the jth derivative of the product R(t)R+ (t) = F (t). The
assumptions (1.14) hold for each rational function in (1.16) and (1.19), since
 
  (2n)! 2n
H(t)(t + k) t=k = (1)n+k = (1)n+k Z,
(n + k)! (n k)! n+k
   
  m + 2n + k 2n
R(t)(t + k) t=k = (1)n+k Z,
2n n+k
   
  m + 2n k 2n
R+ (t)(t + k) t=k = (1)n+k Z
2n n+k
for all k = 0, 1, . . . , n. This completes the proof of the lemma.
In what follows the denominator den(I) of the linear form I = A0 +
A1 1 + + Am m is dened to be the least rational (not necessarily integer)
D > 0 such that every DA0 , DA1 , . . . , DAm is an integer.
Lemma 1.4. The following inclusions hold for the coecients of the linear
form (1.5):
a+b1s
D2n As Z, (1.20)
a+b1
where s is zero or an odd integer, b < s < a + b. Therefore, den(In ) divides D2n
and
a+b1
log den(In ) log D2n
lim  lim = 2(a + b 1). (1.21)
n n n n
Proof. To determine the coecients in (1.6), we use the formula

1 daj  
a 
Ak,j = R(t)(t + k)  , k = 0, 1, . . . , n, j = 1, 2, . . . , a,
(a j)! dtaj t=k
and the representation
R(t) = (Hn (t))a2rb (Fn,n(t))b (Fn,3n (t))b (F(2r1)n,n(t))b (1.22)
of the rational function (1.1). The inclusions
aj
D2n Ak,j Z, k = 0, 1, . . . , n, j = 1, 2, . . ., a, (1.23)
follow from Lemma 1.3 and the Leibniz rule for the dierentiation of a product.
Inclusions (1.20) follow from (1.11), (1.12) and (1.23). The limiting relation (1.21)
follows from the formula
log Dn
lim =1 (1.24)
n n
(the prime number theorem). This completes the proof of the lemma.
So we have constructed a sequence of linear forms In , n = 1, 2, . . . . By (1.4),
a+b1
it contains an innite subsequence of non-zero forms. The forms D2n In ,
n = 1, 2, . . . , have integer coecients. Hence, for
+ 2(a + b 1) < 0 (1.25)
at least one of the numbers (s), where s is an odd integer and b < s < a + b, is
irrational. Moreover, upper estimates for the coecients of the linear forms (1.10)
in the case when (1.25) holds enable us to obtain a lower estimate for the number
of irrational numbers in the given set of values of the zeta function. This is why our
next step will be to calculate (1.4) and to nd the asymptotics of the coecients
of the linear forms (1.10) as n .
496 W. Zudilin

2. An asymptotic estimate for linear forms

We shall need some supplementary properties of the function cot z.
Lemma 2.1. The maximum value of the real-valued non-negative function h(y) =
| cot(x + iy)|, y R, depends only on x (mod Z) R.
Proof. If x = k for some integer k, then z = x is a (unique) pole of the func-
tion cot z on the line Im z = x (see (2.6) below). Therefore, in this case the
absolute maximum of h(y), which is equal to innity, is attained at y = 0. In what
follows we assume that x / Z, whence cos 2x < 1. We have
   
 cos(x + iy) 2  (ey + ey ) cos x + i(ey ey ) sin x 2
h(y)2 =   = 
sin(x + iy)   (ey ey ) cos x + i(ey + ey ) sin x 
e2y + e2y + 2 cos 2x 4 cos 2x
= 2y = 1 + 2y
e + e2y 2 cos 2x e + e2y 2 cos 2x
4| cos 2x|
1+ . (2.1)
2 2 cos 2x
Here we have used the inequality Y + 1/Y
2 for Y = e2y > 0, which becomes an
equality only in the case when Y = 1. The estimate on the right-hand side of (2.1)
depends only on x (mod Z), which completes the proof of the lemma.
We dene dierential iterations of the cotangent:

(1)b1 db1 cot z

cotb z = , b = 1, 2, . . . . (2.2)
(b 1)! dz b1

Lemma 2.2. For every b = 1, 2, . . .

(a) the function cotb z is a polynomial in cot z with rational coecients:

cotb z = Ub (cot z), Ub (y) = (1)b Ub (y), deg Ub = b;

(b) the function sinb z cotb z is a polynomial in cos z with rational coecients:

sinb z cotb z = Vb (cos z), Vb (y) = (1)b Vb (y), deg Vb = max{1, b 2}.
(2.3)
Proof. We proceed by induction. For b = 1 we have cot1 z = cot z and
sin z cot1 z = cos z, whence U1 (y) = V1 (y) = y.
(a) According to the dierential equation

dy
= (y2 + 1), y = cot z,
dz
and formula (2.2), the following recurrence relation holds for the polynomials Ub (y),
b = 1, 2, . . . :
1
Ub+1 (y) = (y2 + 1)Ub (y), b = 2, 3, . . . .
b
The induction step from b to b + 1 shows that Ub+1 (y) = (1)b+1 Ub+1 (y) and the
degree of Ub+1 is one greater than the degree of Ub .
 Irrationality of values of the zeta function 497

(b) Assume that (2.3) holds for some integer b
1. Then

d d
sin z Vb (cos z) = sin z (sinb z cotb z)
dz dz
d
= b cos z sinb z cotb z + sinb+1 z cotb z
dz
= b cos z Vb (cos z) b sinb+1 z cotb+1 z,

whence
1 d
sinb+1 z cotb+1 z = cos z Vb (cos z) sin z Vb (cos z)
b dz
1
= cos z Vb (cos z) + sin2 z Vb (cos z)
b
= Vb+1 (cos z),

where
1
Vb+1 (y) = yVb (y) + (1 y2 )Vb (y). (2.4)
b
Formula (2.4) and the induction hypothesis imply that Vb+1 (y) is a polynomial
whose degree cannot exceed the degree of Vb (y) more than by one. Besides,
Ub+1 (y) = (1)b+1 Ub+1 (y). Since V2 (y) = 1 (by (2.4)), we have deg Vb+1  b 1.
The induction step from b to b+1 and relation (2.4) show that the coecient of yb1
in the polynomial Vb+1 (y) is non-zero and equals 2b1 /b! . Hence, (2.3) holds for
all b = 1, 2, . . . , which completes the proof of the lemma.
Lemma 2.3. For any b = 1, 2, . . . and any integer k the following representation
holds in the neighbourhood of t = k:
1
b cotb t = + O(1), (2.5)
(t k)b
where the function in O(1) is analytic in the neighbourhood of t = k. The function
cotb t is analytic in the neighbourhood of t C \ Z
Proof. The second assertion of the lemma follows from formula (2.2) dening the
function cotb z and the decomposition of the cotangent into a sum of partial frac-
tions:
 
1
1 1
cot t = + + (2.6)
t m=1 t m t + m
(see, for example, [22], Ch. X, 3, formula (3)).
In the neighbourhood of t = k Z we have


1 (2i)s (t k)s
cot t = cot (t k) = 1+ Bs , (2.7)
tk s=2
s!

where Bs , s = 2, 3, . . . , are the Bernoulli numbers (see, for example, [22], Ch. X,
3, formula (2)). Dierentiating identity (2.7) b 1 times, we obtain representa-
tion (2.5) and the rst assertion of the lemma, which completes the proof.
498 W. Zudilin

Lemma 2.4. The following integral representation holds for the sum in (1.2):
M +i
b1 i
I= cotb t R(t) dt, (2.8)
2 M i

where M R is an arbitrary constant in the interval n < M < (2r + 1)n.

Proof. Consider the integrand in (2.8) on the contour of the rectangle with vertices
M iN , N + 1/2 iN (see Fig. 1), where the integer N is suciently large,
N > (2r + 1)n. Since the function R(t) is analytic inside this rectangle and on its
boundary, Cauchys theorem and Lemma 2.3 imply that the integral
 M iN N+1/2iN N+1/2+iN M +iN 
b
+ + + cotb t R(t) dt (2.9)
2i M +iN M iN N+1/2iN N+1/2+iN

is equal to the sum of the residues of the integrand at t Z, M < t  N . Using (2.5)
and the expansion

R(b1)(k)  
R(t) = R(k) + R (k) (t k) + + (t k)b1 + O (t k)b
(b 1)!

in the neighbourhood of t = k Z, M < k  N , we obtain that the integral (2.9)

is equal to

 
R(b1)(k)

N
R(b1)(k)
Res b cotb t R(t) = = . (2.10)
t=k (b 1)! (b 1)!
M <k
N M <k
N k=n+1

Figure 1. The neighbourhood of the point x

 Irrationality of values of the zeta function 499

Let us note that on the sides [N + 1/2 iN, N + 1/2 + iN ], [M iN, N + 1/2 iN ]
and [N + 1/2 + iN, M + iN ] of the rectangle we have R(t) = O(N 2 ) by (1.3), and
the function cotb t is bounded. The latter assertion follows from assertion (a) of
Lemma 2.2 and the boundedness of cot t on the sides under consideration: for the
segment [N + 1/2 iN, N + 1/2 + iN ] we use Lemma 2.1, while for the other two
segments we use the trivial estimate
 
 1 + e2ix e2|y|  1 + e2|y|

| cot (x + iy)| =   , x R,
1 e2ix e2|y|  1 e2|y|
for y = N . Therefore,
 N+1/2iN N+1/2+iN M +iN 
+ + cotb t R(t) dt = O(N 1 ).
M iN N+1/2iN N+1/2+iN

Passing to the limit in (2.9) as N , we obtain the right-hand side of (2.8). Pass-
ing to the limit in (2.10), we obtain the desired sum (1.2), which completes the proof
of the lemma.
Lemma 2.5. The following relation holds for (1.2) as n :

(1)bn (2 n )a2rb (2)b  
I =I
a1
1 + O(n1 ) ,
n
where
+i
1
I = In := sinb n cotb n enf() g( ) d, (2.11)
2i i

f( ) = b( + 2r + 1) log( + 2r + 1) + b( + 2r + 1) log( + 2r + 1)
+ (a + b)( 1) log( 1) (a + b)( + 1) log( + 1) + (a 2rb)2 log 2,
(2.12)
( + 2r + 1)b/2 ( + 2r + 1)b/2
g( ) = , (2.13)
( + 1)(a+b)/2 ( 1)(a+b)/2
and R is an arbitrary constant in the interval 1 < < 2r + 1.
Remark 2.1. To dene the functions f( ) and g( ) unambiguously, we consider
them in the -plane with cuts along the rays (, 1] and [2r + 1, +), xing the
branches of logarithms that take real values on the interval (1, 2r + 1) of the real
axis. This choice of branches is motivated by the following proof.
Proof. Using the formula (z + 1) = z(z), we express the rational function (1.1)
in terms of the gamma function:
 b  a
(t + (2r + 1)n + 1) (t)(1 t)
R(t) = (1)an (2n)!a2rb , (2.14)
(t + n + 1) (t + n + 1)
where, as before, stands for two factors in the product. We transform the inte-
grand in (2.8) using formula (2.14) for R(t) and the identity

(t)(1 t) = = (1)n (t + n + 1)(t n).
sin t
500 W. Zudilin

We have

(1)bn sinb t (t + (2r + 1)n + 1)b (t n)a+b

R(t) = (2n)!a2rb . (2.15)
b (t + n + 1)a+b

Substituting this expression into (2.8), we obtain the formula

M +i
(1)bn i
I= sinb t cotb t
2 M i
(t + (2r + 1)n + 1)b (t n)a+b (2n)!a2rb
dt
(t + n + 1)a+b

(1)bn (2n)a2rb i M +i b (t + (2r + 1)n)b (t + (2r + 1)n)b
= sin t cotb t
2 M i (t + n)a+b
(t + (2r + 1)n)b (t n)a+b (2n)a2rb
dt, (2.16)
(t + n)a+b

where M R is an arbitrary constant in the interval n < M < (2r + 1)n. We put
M = n for R in the interval 1 < < 2r + 1.
The asymptotics of the gamma function
 
1  
log (z) = z log z z + log 2 + O |z|1 (2.17)
2

(see, for example, [23], 3.10) implies that the following formulae hold on the
contour of integration in (2.16):
 
1
log (t + (2r + 1)n) = t + (2r + 1)n log(t + (2r + 1)n)
2

(t + (2r + 1)n) + log 2 + O(n1 ),
 
1
log (t + n) = t + n log(t + n) (t + n) + log 2 + O(n1 ),
2
 
1
log (t n) = t n log(t n) (t n) + log 2 + O(n1 ),
2
 
1
log (2n) = 2n log(2n) 2n + log 2 + O(n1 ).
2

The change t = n in the integrals in (2.16) yields

(1)bn (2 n )a2rb (2)b1 i +i b
I= sin n cotb n enf()
na1 i

( + 2r + 1)b/2 ( + 2r + 1)b/2  
1 + O(n1 ) d,
( + 1)(a+b)/2 ( 1)(a+b)/2

where f( ) is the function dened in (2.12) and = M/n R is an arbitrary

constant in the interval 1 < < 2r + 1. This completes the proof of the lemma.
 Irrationality of values of the zeta function 501

By assertion (b) of Lemma 2.2 the integral in (2.11) can be written as

+i
1

b
I = ck en(f()ki) g( ) d, ck = ck , (2.18)
2i i
k=b

where the sum is taken over those k that have even parity with b, and cb = cb = 0
if b > 1.
Lemma 2.6. For , R, 1 < < 2r + 1, and n a positive integer we put
+i
1
Jn, = en(f()i) g( ) d. (2.19)
2i i
Then
Jn, = Jn,,
where the bar stands for complex conjugation.
Proof. Applying Schwarz reection principle to the functions f( ), g( ) analytic
in C \ ((, 1] [2r + 1, +)) and taking real values on the interval (1, 2r + 1)
(see, for example, [24], Ch. III, 7, Proposition 7.1) and making the change 
in the integral in (2.19), we obtain that
+i +i
1 1
Jn, = en(f( )i ) g( =
) d en(f( )i ) g(
) d
2i i 2i i
+i
1
= en( f()+i ) g( ) d = Jn,,
2i i
as was to be shown.
Corollary 2.1. The integral in (2.11) can be written as


b
I = 2 ck Re Jn,k , (2.20)
k=0
kb (mod 2)

where ck are some (rational ) constants, c1 = 1 for b = 1 and cb = 0, cb2 = 0

for b > 1.
Proof. The desired representation (2.20) follows immediately from assertion (b) of
Lemma 2.2, formula (2.18), and Lemma 2.6.
Now let us calculate the asymptotics of the integrals on the right-hand side
of (2.18) using the saddle-point method. We have to determine the saddle points of
the integrands, that is, the zeros of the derivatives of the functions f( ) ki ,
k = 0, 1, . . . , b, k b (mod 2). It is easy to verify that the roots of the derivative

d
(f( ) ki ) = b log( + 2r + 1) b log( + 2r + 1)
d
+ (a + b) log( 1) (a + b) log( + 1) ki,
k = 0, 1, . . . , b, k b (mod 2),
502 W. Zudilin

are simultaneously the roots of the polynomial

( + 2r + 1)b ( 1)a+b ( 2r 1)b ( + 1)a+b , (2.21)

whose quantity is equal to the degree a + 2b 1 of this polynomial.

Lemma 2.7. Let a, b, r be positive integers, let a + b be even, a
3rb, and let f( )
be the function dened by (2.12), where the corresponding branches of the logarithms
in the cut plane C\((, 1][2r+1, +)) take real values on the interval (1, 2r+1).
Then the equation
f  ( ) i = 0, R, (2.22)
has
(a) a pair of real solutions 0 i0 with 0 symmetric with respect to the imag-
inary axis for = 0, where 0 (1, 2r + 1), and + ( ) in i0 corresponds to the
upper (lower) bank of the cut (, 1],
(b) a pair of real solutions 1 i0 and 1 i0 symmetric with respect to
the imaginary axis for = b, where 1 (2r + 1, +), and + ( ) in i0
coincides with the sign of and corresponds to the upper (lower) banks of the cuts
(, 1], [2r + 1, +),
(c) a real solution i0 for = (a + b), where + ( ) in i0 coincides with the
sign of and corresponds to the upper (lower) bank of the cut (, 1],
(d) a solution on the imaginary axis for the real such that b < || < a + b,
(e) a pair of complex solutions symmetric with respect to the imaginary axis for
the real such that 0 < || < b.
All solutions of equation (2.22) corresponding to positive are contained in the
half-plane Im > 0. All solutions corresponding to negative are contained in
the half-plane Im < 0. All solutions of equation (2.22) appear in the list (a)(e).
Proof. Since

f  ( ) = b log( + 2r + 1) b log( + 2r + 1)
+ (a + b) log( 1) (a + b) log( + 1) (2.23)

and log z = log |z| + i arg z, < arg z < , for the main branch of the logarithm
with the cut (, 0] in the z-plane, we have

| + 2r + 1|b | 1|a+b
Re f  ( ) = log , (2.24)
| + 2r + 1|b | + 1|a+b
Im f  ( ) = b arg( + 2r + 1) b arg( + 2r + 1)
+ (a + b) arg( 1) (a + b) arg( + 1), (2.25)

and the arguments arg( ) take the value zero on (1, 2r + 1). Formula (2.25) has
the following geometrical interpretation in the notation of Fig. 2:

Im f  ( ) = b( + + ) + (a + b)(+ ) = b( ) + (a + b), (2.26)

 Irrationality of values of the zeta function 503

whence Im f  ( ) > 0 if Im > 0 and Im f  ( ) < 0 if Im < 0. By (2.24) and (2.26),

the set of solutions of equation (2.22) for every is symmetric with respect
to the imaginary axis, and the symmetry  maps it onto the set of solu-
tions of equation (2.22) with instead of . In what follows we deal with > 0
and the upper half-plane. The results thus obtained can b transferred to the case
when < 0 by the symmetry with respect to the real axis.

Figure 2

By (2.23), any solution of equation (2.22) is a root of the polynomial

( + 2r + 1)b ( 1)a+b ei ( + 2r + 1)b ( + 1)a+b , (2.27)

whose degree is equal to a + 2b 1 if is an integer that has even parity with b

(otherwise, its degree is equal to a + 2b). If b (mod 2), then this polynomial
coincides with (2.21).
(a), (c). If = 0 (or, which is the same, = a + b), then the polynomial (2.27)
is an odd function. Therefore, 0 is one of its roots. The corresponding solution
+i0 of equation (2.22) corresponds to = a + b, since = = in (2.26)
for = 0. Calculating the values of the polynomial at 1, 2r + 1 and using the
fact that this polynomial is odd, we obtain that it has a pair of real roots 0 ,
where 0 (1, 2r + 1). We shall show later that 0 is the unique real root in
(1, 2r + 1), but until then we assume that 0 is the real root nearest to 2r + 1 of
the polynomial (2.27) with = 0 in the above interval. By (2.26), the solutions
0 + i0 and 0 correspond to = 0 in equation (2.22).
(b) Calculating the values of (2.27) with = b at 2r + 1, +, we likewise obtain
that this polynomial has a pair of real roots 1 . We assume for the moment that
among the roots in the interval (2r + 1, +) the root 1 (2r + 1, +) is the
nearest to 2r + 1. By (2.26), the solutions 1 + i0 and 1 + i0 correspond to = b
in equation (2.22).
In the case when a
3rb there is a better localization of the root 1 , namely,
1 (2r + 1, 4r + 1). Indeed, the value of the polynomial (2.27) at 4r + 1 is equal
to
 b  a+b 
1 1
b
(6r + 2) (4r)a+b
(2r) (4r + 2)
b a+b b
= (2r) (4r) a+b
3+ 1+ < 0,
r 2r
504 W. Zudilin

since
   
1 3 1
(3r + 1) log 1 + = 4 log > 2 log 2 = log 3 + if r = 1,
2r 2 r
   
1 3r + 1 7 1
(3r + 1) log 1 + >
> 2 log 2 > log 3 + if r
2,
2r 2r + 1 5 r

where we have used the elementary inequality

 
1 1 1
< log 1 + < , (2.28)
n+1 n n

which holds for all positive integers n, whence

     
1 1 1
(a + b) log 1 +
(3r + 1)b log 1 + > b log 3 + .
2r 2r r

(d) By (2.24), we have Re f  ( ) = 0 on the imaginary axis. As ranges over

the ray iy, y > 0, the angles , decrease continuously from to 0. Therefore, the
function Im f  ( ) = b( ) + (a + b) takes all intermediate values in the interval
(b, (a + b)). In particular, for every real , b < < a + b, we obtain at least one
solution of equation (2.22) that lies on the imaginary axis.
(e) We shall now localize the complex solutions of (2.22) corresponding to the
real , 0 < || < b.
Consider the semicircle of radius 2 with centre 2r +1, where < r, in the upper
half-plane. For = x + iy on this semicircle we have

| + 2r + 1|2 = 42 , | + 2r + 1|2 = 42 + 4(2r + 1)x,

| 1|2 = 42 + 4rx 4r(r + 1), | + 1|2 = 42 + 4(r + 1)x 4r(r + 1).
(2.29)
By (2.24),

(2 + (2r + 1)x)b (2 + rx r(r + 1))a+b

2 Re f  ( ) = log . (2.30)
2b (2 + (r + 1)x r(r + 1))a+b

We denote the function in (2.30) by f(x), 2r + 1 2 < x < 2r + 1 + 2. It is a

monotonically increasing function of x, since

b(2r + 1) (a + b)r (a + b)(r + 1)

f (x) = + 2 2
+ (2r + 1)x + rx r(r + 1) + (r + 1)x r(r + 1)
2
b(2r + 1) (a + b)(r(r + 1) 2 )
= 2 + 2
+ (2r + 1)x ( + rx r(r + 1))(2 + (r + 1)x r(r + 1))
4b(2r + 1) 16(a + b)(r(r + 1) 2 )
= + > 0.
| + 2r + 1| 2 | 1|2 | + 1|2
 Irrationality of values of the zeta function 505

Therefore,

f(2r + 1 2) < f(x) < f(2r + 1 + 2), x (2r + 1 2, 2r + 1 + 2), (2.31)

and
Re f  (2r + 1 2) < Re f  ( ) < Re f  (2r + 1 + 2 + i0) (2.32)

on the semicircle under consideration. Consider = (1 2r 1)/2, where

1 (2r + 1, +) is the real root of the polynomial (2.27) with = b.
As mentioned above, 2r + 1 < 1 < 4r + 1 if a
3rb, that is, the assumption
< r holds. By (2.32), we have

Re f  ( ) < Re f  (1 + i0) = 0 (2.33)

for the points of the semicircle. Besides,

lim Re f  ( ) = +. (2.34)
2r+1

Consider the rays starting from the point 2r + 1 and contained in the upper half-
plane. By (2.33) and (2.34), each of these contains a point belonging to the domain
bounded by the above semicircle and the real axis and such that Re f  ( ) = 0.
Hence, this domain contains a part of the continuously dierentiable curve

| + 2r + 1|b | 1|a+b
Re f  ( ) = log = 0. (2.35)
| + 2r + 1|b | + 1|a+b

(See Fig. 3, where this part of the curve is situated, according to (2.31), between
the two semicircles corresponding to = (1 2r 1)/2 and = (2r + 1 0 )/2.)
By (2.26), the values of Im f  ( ) on this curve as it approaches the real axis are equal
to 0 (at 0 ) and b (at 1 ). Therefore, Im f  ( ) takes on the curve all intermediate
values between 0 and b, that is, the values , where 0 < < b. So we have
produced solutions of equation (2.22) for these . Every point obtained from these
by reection in the imaginary axis also is a solution.

Figure 3
506 W. Zudilin

We have produced all the solutions of equation (2.22) listed in (a)(e). To

complete the proof, we have to show that there are no others. Assume the contrary,
that is, assume that for some 0 R there is a solution of equation (2.22) that is not
listed in (a)(e). As mentioned above, this solution is a root of the polynomial (2.27)
with = 0 . We claim that the number of roots of (2.27) listed in (a)(e) is equal
to its degree, which will contradict our assumption.
If 0 is not an integer, then we have a purely imaginary solutions (see (d))
corresponding to 0 (mod 2), b < || < a + b, and b pairs of complex solutions
(see (e)) corresponding to 0 (mod 2), || < b. Therefore, the total number of
roots of (2.27) with = 0 listed in (d) and (e) is equal to the degree a + 2b of the
polynomial.
If 0 is an integer (even or odd), then a solution of equation (2.22) with = 0
is a root of the polynomial

( + 2r + 1)2b ( 1)2(a+b) ( 2r 1)2b ( + 1)2(a+b). (2.36)

For this polynomial we have found two real roots 0 in (a), two real roots 1
in (b), one real root 0 in (c), 2(a 1) purely imaginary roots in (d) and 4(b 1)
complex roots in (e). The total number of roots of the polynomial (2.36) listed
in (a)(e) is equal to its degree 2a + 4b 1.
Hence, any solution of equation (2.22) is contained in the list (a)(e). This
contradiction completes the proof of the lemma.

Remark 2.2. We replaced the original assumption a
2rb by the stronger assump-
tion a
3rb in Lemma 2.7 only in order to localize the root 1 . Therefore,
Lemma 2.7 remains valid in the case when a
2rb if

1 < 4r + 1. (2.37)

Corollary 2.2. The set of C for which (2.35) holds is the union of the imagi-
nary axis and a pair of closed curves symmetric with respect to the imaginary axis
(see Fig. 4). These curves divide the complex plane into four parts, in each of which
the sign of Re f  ( ) is constant.

Figure 4
 Irrationality of values of the zeta function 507

Corollary 2.3. Any semicircle of radius , 2r + 1 0 < 2 < 1 2r 1, with

centre 2r + 1 lying in the upper half-plane, intersects the curve (2.35) (see Fig. 3)
at precisely one point.

Proof. This assertion follows from inequalities (2.32), since the function Re f  ( ) is
monotonic on any of these semicircles.

Lemma 2.8. Let a, b, r, n be positive integers, a
2rb, let f( ), g( ) be as dened

in (2.12), (2.13), and let , R, 1 < < 2r + 1, ||  b. Then the contour of
integration Re = in the integral

+i
en(f()i) g( ) d (2.38)
i

can be replaced by any other contour L joining innitely remote points in the
domains Re
, Im < 0 and Re
, Im > 0 and intersecting the real
axis at precisely one point = . In particular, L can go along the upper and (or)
lower bank of the cut [2r + 1, +).

Proof. For any real N > we join the points of the original contour Re = and
those of the new contour L by two arcs of radius N with centre at the origin. We
denote the corresponding points of intersection with the contours by M and L
(see Fig. 5).

Figure 5
508 W. Zudilin

There are no singular points of the integrand in (2.38) inside the curvilinear
triangles L+ M+ and M L . The lemma will be proved if we can show that

en(f()i)
g( ) d 0, en(f()i) g( ) d 0 (2.39)
L+ M + M L

as N , where L+ M+ and M L are arcs of the circle of radius N with centre

at the origin.
On the arcs L+ M+ and M L of the circle = N eit we have the inequalities
0  t < /2 and /2 < t  0, respectively (the value t = 0 is taken on the upper
and lower bank of the cut [2r + 1, +)). Using Taylors formula, we obtain the
relations
 
eit
log( + 1) = log(N eit + 1) = log(N eit ) + log 1 +
N
it
 
e 1
= log N + it + +O , (2.40)
N N2
 
eit 1
log( 1) = log N + it +O , (2.41)
N N2
 
it it (2r + 1)eit
log( + 2r + 1) = log(N e + 2r + 1) = log(N e ) + log 1 +
N
it
 
(2r + 1)e 1
= log N + it + +O , (2.42)
N N2
log( + 2r + 1) = i + log(N eit (2r + 1))
 
(2r + 1)eit 1
= i + log N + it +O , (2.43)
N N2

where (+) in i in (2.43) corresponds to the arc L+ M+ ( M L ) according

to the choice of branches of the logarithm. The constant in O(1/N 2 ) in formu-
lae (2.40)(2.43) is absolute, since |eit | = 1 for real t. Substituting these expan-
sions into (2.12), we obtain the formula

f( ) = f(N eit )
 
= (log N + it) b( + 2r + 1) + b( + 2r + 1) + (a + b)( 1) (a + b)( + 1)
eit 
+ b(2r + 1)( + 2r + 1) b(2r + 1)( + 2r + 1)
N 
(a + b)( 1) (a + b)( + 1)
 
| |
i b( + 2r + 1) + (a 2rb)2 log 2 + O
N2
   
Ne 1
= 2(a 2rb) log + it bi(N e 2r 1) + O
it
,
2 N
 Irrationality of values of the zeta function 509

where the sign of b coincides with the sign of t (and sin t). Therefore,
 
Ne 1 1
Re(f( ) i ) = 2(a 2rb) log (b )N | sin t| + O 
2 N n

on the arcs L+ M+ and M L as N . (We have used the inequalities a
2rb

and ||  b.) Hence, the estimate

|en(f()i) | = en Re(f()i)  e (2.44)

holds for all suciently large N . The following trivial estimate holds for the func-
tion (2.13) on the arcs L+ M+ and M L :
     
N b/2 N b/2 1 1
|g( )| = O =O =O as N , (2.45)
N (a+b)/2 N (a+b)/2 Na N2

since a
2rb
2. Since the length of each of the arcs L+ M+ and M L does not
exceed N/2, estimates (2.44) and (2.45) imply that every integral in (2.39) is of
order O(1/N ). Hence, the limiting relations (2.39) hold, which completes the proof
of the lemma.
Lemma 2.9. Let a, b, r be positive integers, a
2rb, let f( ), g( ) be the functions
dened in (2.12) and (2.13), and let R, ||  b. Assume, moreover, that (2.37)
and the assumption

0 + 20 1
1 (2.46)

hold for the real roots 0 (1, 2r + 1) and 1 (2r + 1, +) of (2.36).

Then the asymptotic behaviour of the integral (2.38) with = 0 as n
is determined by the single saddle point 0 (the solution of equation (2.22)) in the
domain Im > 0. More precisely, the following asymptotic formula holds:
0 +i
1
Jn, = en(f()i) g( ) d
2i 0 i



1/2
|f  (0 )|1/2 en Re f0 (0 ) |g(0 )| e 2 arg f
i
(0 )+i arg g(0 )+in Im f0 (0 )
= (2)
 
n1/2 1 + O(n1 ) as n , (2.47)

where

f0 ( ) = f( ) f  ( )
= b(2r + 1) log( + 2r + 1) + b(2r + 1) log( + 2r + 1)
(a + b) log( + 1) (a + b) log( 1) + (a 2rb)2 log 2. (2.48)

Proof. We shall prove the lemma only for
0. The case when < 0 can be
reduced to this case by reection in the real axis.
510 W. Zudilin

According to Lemma 2.8, we replace the contour of integration in (2.38) with

= 0 by a contour consisting of rectilinear rays and segments. Let us note at once
that the derivative of Re f( ) on every such rectilinear part = 1 + ei t, t R, is
given by the formula
 
d df( ) d df( )
Re f( ) = Re = Re = Re f  ( ) cos Im f  ( ) sin .
dt dt dt d
(2.49)
If = 0, then 0 is the unique maximum point of the function Re f( ) on the
line = 0 + it, < t < +, since

d
Re f( ) = Im f  ( )
dt
and Im f  ( ) takes negative values in the domain Im < 0 and positive ones in
the domain Im > 0. So, in the case when = 0, we use the original contour
Re = 0 .
If = b, then we replace the ray going from 0 to 0 + i in the contour of
integration in (2.38) with = 0 by a ray going from 0 to + along the upper
bank of the cut [2r + 1, +). The function

Re(f( ) i ) (2.50)

increases on the set = 0 + it, < t  0, since

d
Re(f( ) i ) = Im f  ( ) +
> 0. (2.51)
dt
On the set = t, 0  t < +, the function Re(f( ) i ) = Re f( ) has a
unique maximum point at t = 1 (by Lemma 2.7 and Corollary 2.2).

Figure 6

It remains to consider the case when (0, b). We replace

the ray going
from 0 to 0 + i by the segment = 0 + ei t, 0  t  20 1, passing
through the saddle point 0 (the unique solution of equation (2.22) in the domain
 Irrationality of values of the zeta function 511

Re > 0) and the ray = ei 20 1 + t, 0  t < +. By Corollary 2.3, the ray

= 0 +ei t, t > 0, intersects the curve Re f  ( ) = 0 at a single point (namely, 0 ),
and 0 < < /2. By (2.46), 0 is an interior point of the above segment (see
Fig. 6). On the
ray = 0 + it, t  0, the function (2.50) increases by (2.51). On
the ray = ei 20 1 + t, t
0 , it decreases, since

d
Re(f( ) i ) = Re(f  ( ) i) = Re f  ( ) < 0
dt

by Corollary 2.2. Hence, it is sucient to show that 0 is the unique maximum

of the function (2.50) on the segment = 0 + ei t, 0  t  t1 , where
point

t1 = 20 1. Assume that t0 corresponds to the point 0 = 0 + ei t0 on the

segment under consideration. By Corollary 2.3,

Re f  (0 + ei t) < 0 for 0 < t < t0 ,Re f  (0 + ei t) > 0 for t0 < t  t1 .

(2.52)
We claim that the function Im f  ( ) increases monotonically on the segment
under consideration. Let us use the geometric interpretation (2.26) of the function
Im f  ( ) once again. As t
0 increases on the ray = 0 + ei t, the angle
decreases monotonically from to 0, whereas the angle rst increases from 0 to
some 0 and then decreases from 0 to 0. We claim that the maximal value 0 is
attained on the ray at t = t1 , which will imply that the function

Im f  ( ) = b( ) + (a + b)

is monotonically increasing on the segment = 0 + ei t, 0  t  t1 .

Figure 7

The points = x + iy of the upper half-plane at which the real segment [1, 1]
subtends less than a given angle belong to an arc of the circle of radius 1/ sin
with centre i cot (this assertion for an acute angle is depicted in Fig. 7):

x2 + y2 2y cot = 1,
512 W. Zudilin

whence
x2 + y 2 1
= cot1 .
2y
On the given ray, x = 0 + t cos and y = t sin if t
0, whence

d (2x cos + 2y sin ) 2y 2 sin (x2 + y2 1)

=
dt (x2 + y2 1)2 + (2y)2
t (2 1)
2
= 2 sin 2 .
(x + y2 1)2 + 4y2

Therefore, the maximal value of is attained at t = 20 1.

Since Im f  ( ) is monotonically increasing on the segment = 0 + ei t, 0 
t  t1 , we have

Im f  (0 + ei t) < for 0  t < t0 , Im f  (0 + ei t) >

for t0 < t  t1 .
(2.53)
Combining (2.52), (2.53) and (2.49), we obtain that the function

d
Re(f( ) i ) = Re f  ( ) cos (Im f  ( ) ) sin
dt
on the segment = 0 + ei t, 0  t  t1 , changes sign from plus to minus only
when passes through 0 = 0 + ei t0 . Hence, 0 is indeed the unique maximum
point on the segment and on the whole contour of integration consisting of this
segment and two rays.
To prove the asymptotic formula (2.47), we write down the contribution of the
saddle point 0 :


 
(2)1/2 e 2 2 arg f (0 ) |f  (0 )|1/2 en(f(0 )i0 ) g(0 ) n1/2 1 + O(n1 ) (2.54)
i i

(see, for example, [23], 5.7, formula (5.7.2)). We have

b b a+b a+b
f  ( ) = + +
+ 2r + 1 + 2r + 1 1 + 1
(a 2rb) 2 (2r + 1)(a 2rb + 2ra)
=2 = 0
( 2 1)( 2 (2r + 1)2 )

at those for which Re f  ( ) = 0 (in particular, 0 ). Using the equality

f(0 ) i0 = f(0 ) f  (0 )0 = f0 (0 )

and separating the modulus and argument in (2.54), we obtain (2.47), which com-
pletes the proof of the lemma.
Corollary 2.4. Let the assumptions of Lemma 2.9 hold. Assume, moreover, that
1
arg f  (0 ) + arg g(0 )  (mod Z) or Im f0 (0 )  0 (mod Z). (2.55)
2 2
 Irrationality of values of the zeta function 513

Then the following limiting relation holds for the integral

0 +i
1 1
Re Jn, = (Jn, + Jn,) = enf() cos(n )g( ) d :
2 2i 0 i

log | Re Jn,|
lim = Re f0 (0 )
n n
22(a2rb)|0 + 2r + 1|b(2r+1)| 0 + 2r + 1|b(2r+1)
= log .
|0 + 1|a+b |0 1|a+b
(2.56)
Moreover, in the case when
1
arg f  (0 ) + arg g(0 ) 0 (mod Z) and Im f0 (0 ) 0 (mod Z), (2.57)
2
the integral Jn, has the real asymptotics (2.47), which implies that the upper limit
in (2.56) becomes a limit.
Proof. Assumption (2.55) implies that the real part of the coecient of n1/2 in
the asymptotic formula (2.47) is non-zero for an innite sequence of numbers n. It
is on this sequence that the limiting relation (2.56) is attained. If (2.57) holds, then
the principal term of the asymptotics (2.47) is a real number, and (2.56) with lim
instead of lim follows immediately from (2.47).
We now state our denitive results concerning
log |In | log |In |
= lim = lim (2.58)
n n n n
(see Lemma 2.5). If b = 1 or b = 2, then the results take a simple form, and the
upper limits in (2.58) can be replaced by limits. Our assertions in these cases will
be stated as separate propositions.
Proposition 2.1. Let b = 1, let r be a positive integer, let a > 2r be an odd
integer, and let 1 (2r + 1, +) be a real root of the polynomial (2.21). Then
log |In | 22(a2r)(1 + 2r + 1)2r+1 (1 2r 1)2r+1
= lim = log . (2.59)
n n (1 + 1)a+1 (1 1)a+1

Proof. In the case when b = 1, = 1 is the unique maximum point of the function
Re f( ) on the contour of integration for calculating the asymptotics of the inte-
gral Jn,1 in the proof of Lemma 2.9 (the ray = 0 + it, t  0, and the ray = t,
t
0 , that goes along the upper bank of the cut [2r + 1, +)). This implies, in
particular, that f  (1 ) < 0, whence
1
arg f  (1 ) (mod Z).
2 2
In the case when b = 1 we have for the function (2.13) that

arg g(1 ) = and Im f0 (1 ) = (2r + 1).
2
514 W. Zudilin

Using Corollary 2.4 with = 1 and 0 = 1 , we obtain that (2.57) holds and

log | Re Jn,1 |
lim = Re f0 (1 ).
n n

The formula In = Re Jn,1 (see Corollary 2.1) and Lemma 2.5 imply that (2.59)
holds.

Proposition 2.2. Let b = 2, let r be a positive integer, assume that a
4r is an

even integer, and let 0 (1, 2r + 1) be a real root of the polynomial (2.21). Then

log |In | 22(a4r)(0 + 2r + 1)2(2r+1) (0 + 2r + 1)2(2r+1)

= lim = log . (2.60)
n n (0 + 1)a+2 (0 1)a+2

Proof. If b = 2, then In = Jn,0 , and = 0 is the unique maximum on the

contour of integration = 0 + it, t R, for calculating the asymptotics of
the integral Jn,0 in the proof of Lemma 2.9. Therefore, f  (0 ) > 0, whence

1
arg f  (0 ) 0 (mod Z).
2

It is obvious that
arg g(1 ) = 0, Im f0 (1 ) = 0.

Corollary 2.4 with = 0 and 0 = 0 implies that (2.57) holds and

log |Jn,0|
lim = Re f0 (0 ) = f0 (0 ).
n n

Combining this with Lemma 2.5, we obtain (2.60).

Lemma 2.10. Assume that the real root 1 (2r+1, +) of the polynomial (2.21)
satises the inequality
 
br(r + 1) r(r + 1)
1  2r + 1 + min , . (2.61)
2(a + b) 3(2r + 1)

Then Re f0 ( ) regarded as a function of Re increases in the domain Re > 0,

Im
0 on curve (2.35).

Proof. As was shown in the proof of Lemma 2.7 (see also Corollary 2.3), on the
smooth curve (2.35) in the domain Re > 0, Im
0 the quantity = | (2r +1)|/2,
= x + iy, can be represented as an implicit function of x, 0  x  1 , which is
continuously dierentiable and increases. Using formulae (2.29), (2.30) and dier-
entiating with respect to x, we obtain the following formula on the curve dened
 Irrationality of values of the zeta function 515

by formula (2.35):

 
b b a+b a+b
2 +
2 + (2r + 1)x 2 2 + rx r(r + 1) 2 + (r + 1)x r(r + 1)
 
b(2r + 1) (a + b)r (a + b)(r + 1)
+ + = 0.
2 + (2r + 1)x 2 + rx r(r + 1) 2 + (r + 1)x r(r + 1)

Simple transformations yield that

 
 b(2r + 1) a+b b(2r + 1)2 (a + b)(r(r + 1) 2 )
2 x = + .
| (2r + 1)| 2 | 1|2 | (2r + 1)|2 | 1|2
(2.62)
Therefore, the function Re f0 ( ) can be regarded on the curve (2.35) as a function
of x, 0  x  1 . By (2.29), we have

2b(2r+1) (2 + (2r + 1)x)b(2r+1)

f0 (x) := 2 Re f0 ( ) = log .
(2 + rx r(r + 1))a+b (2 + (r + 1)x r(r + 1))a+b

Combining this with (2.29) and (2.62), we obtain that


b(2r + 1) b(2r + 1)
f0 (x) = 2 + 2
2 + (2r + 1)x

a+b a+b
2 2
+ rx r(r + 1) + (r + 1)x r(r + 1)
 
b(2r + 1)2 (a + b)r (a + b)(r + 1)
+
2 + (2r + 1)x 2 + rx r(r + 1) 2 + (r + 1)x r(r + 1)
 
b(2r + 1) a+b
= 32 (22 + (2r + 1)x)
| (2r + 1)|2 | 1|2
 
b(2r + 1)2 (a + b)(r(r + 1) 2 )
+ 16(2r + 1) +
| (2r + 1)|2 | 1|2
64 (a + b)r(r + 1) 32(a + b)r(r + 1)x
+
| 1|2 | 1|2
 
32(2 + (2r + 1)x) b(2r + 1)2 (a + b)(r(r + 1) 2 )
= +
x | (2r + 1)|2 | 1|2
32(a + b)r(r + 1)
+ (2 x). (2.63)
| 1|2
516 W. Zudilin

The function = (x) increases on the curve (2.35). Hence, 
0. Continuing

the chain in (2.63), we obtain that

32(2 + (2r + 1)x) b(2r + 1)2 (a + b)(r(r + 1) 2 )

f0 (x)
+ 32(2r + 1)
x | (2r + 1)|2 | 1|2
32(a + b)r(r + 1)x

| 1|2
2b(2r + 1) 2(a + b)((2r + 1)2 + r(r + 1)x r(r + 1)(2r + 1))
= .
x (2 + rx r(r + 1))(2 + (r + 1)x r(r + 1))
Put  
br(r + 1) r(r + 1)
= min , . (2.64)
2(a + b) 3(2r + 1)
We have
1 2r 1. The localization of the curve (2.35) (see Corollary 2.3)
implies that

2r + 1  x  2r + 1 + , 0< ,
2
whence
2b(2r + 1) 2(a + b)((2r + 1)/4 + r(r + 1))
f0 (x) >
2r + 1 + r(r + 1)(r )(r + 1 )
 
b (a + b)(2r + 1 + )
> 2(2r + 1)
2r + 1 + 4r(r + 1)(r )(r + 1 )
4br(r + 1)(r )(r + 1 ) (a + b)(2r + 1 + )2
= 2(2r + 1) .
4r(r + 1)(r )(r + 1 )(2r + 1 + )
(2.65)

Combining the inequalities

br(r + 1)
br(r + 1)
(a + b) if  ,
2 2(a + b)
r(r + 1)
8(r )(r + 1 ) > (2r + 1 + )2 if  ,
3(2r + 1)

with (2.64) and (2.65), we obtain that f0 (x) > 0. Hence, Re f0 ( ), regarded as a
function of x = Re , increases on the curve (2.35), as was to be shown.
Proposition 2.3. Let a, b, r be positive integers, assume that a + b is even, b
3,
a
2rb, and assume that (2.61) holds for the real root 1 (2r + 1, +) of the
polynomial (2.21). Let

22(a2rb)|0 + 2r + 1|b(2r+1)| 0 + 2r + 1|b(2r+1)

:= log ,
|0 + 1|a+b |0 1|a+b
where 0 is a complex root of the polynomial (2.21) in the domain Re > 0, Im > 0
with maximal real part Re 0 . Assume, moreover, that (2.55) holds. Then
log |In |
lim = . (2.66)
n n
 Irrationality of values of the zeta function 517

Proof. The inequality (2.61) implies that 1 (2r + 1) < r/3. Hence, (2.37)
and (2.46) hold. By Lemma 2.10, the maximal value of the function Re f0 ( )
on the roots of polynomial (2.21) is attained at = 0 , for which = k = b 2. By
Corollary 2.4, the asymptotics of the integral (2.20) is determined by the contribu-
tion of Re Jn,b2, since the contributions of the other quantities are exponentially
smaller in comparison with Re Jn,b2 , and the coecient cb2 in (2.20) is non-zero
by assertion (b) of Lemma 2.2. Therefore,

log |In | log | Re Jn,b2 |

lim = lim = .
n n n n

Using Lemma 2.5, we obtain the desired relation (2.66).

3. Estimates for the coecients of linear forms

Our rst result will be an upper estimate for the coecients of linear forms.

Proposition 3.1. Let a, b, r be positive integers, assume that a+b is even, a
2rb,
and let the linear forms (1.5) be dened by (1.2) and (1.1). Then the following
estimate holds for the coecients As , s = 0 or s = b + 1, . . . , a + b 1:

s |
log |A
lim  2b(2r + 1) log(2r + 1) + 2(a 2rb) log 2,
n n (3.1)
s = 0 or s = b + 1, . . . , a + b 1 and is odd.

Moreover, estimate (3.1) is exact in the case when a is even:

log |A
s |
lim = 2b(2r + 1) log(2r + 1) + 2(a 2rb) log 2,
n n (3.2)
s = b + 1, . . . , a + b 1 and is odd.

Proof. Consider decomposition (1.6), where the coecients can be calculated using
the formulae

1 daj  
a 
Ak,j = R(t)(t + k)  , k = 0, 1, . . . , n, j = 1, 2, . . . , a.
(a j)! dtaj t=k

First of all we claim that

((2r + 1)n)!2b (2n)!a2rb

max |Ak,a| = |A0,a | = . (3.3)
k=0,1,...,n n!2(a+b)
518 W. Zudilin

Since |Ak,a| = |Ak,a| for k = 0, 1, . . . , n, it is sucient to show that |Ak,a| decreases

as k increases from 0 to n. This can be veried directly:

|Ak,a| ((2r + 1)n + k)b (n k + 1)a+b

=
|Ak1,a| ((2r + 1)n k + 1) b (n + k)a+b
 b  a+b
((2r + 1)n + k) (n k + 1) nk+1
=
((2r + 1)n k + 1) (n + k) n+k
 b  a+b
(2r + 1)n2 k(k 1) + n(2rk + 2r + 1) n (k 1)
=
(2r + 1)n2 k(k 1) + n(2rk + 1) n+k
< 1, k = 1, . . . , n.

We now x an integer k, |k|  n, and denote the logarithmic derivative of

R(t)(t + k)a by gk (t):

(2r+1)n
1

n
1
gk (t) = b (a + b) .
t+l t+l
l=(2r+1)n l=n
l=k l=k

Then for j = 0, 1, 2, . . . the modulus of the quantity

 (n+1)
1 dj gk (t) 
(1)j

(2r+1)n
(1)j
=b + b
j! dtj t=k (l k)j+1 (l k)j+1
l=(2r+1)n l=n+1

n
(1)j
a
(l k)j+1
l=n
l=k

is bounded above by 2(2rb + a)n (we estimate every summand on the right-hand
side by 1). Using Leibniz rule for the dierentiation of a product, we obtain that

1 dj1  
a 
Ak,aj = g k (t) R(t)(t + k) 
j! dtj1 t=k

1
j1
1 d j1m
gk (t) 
= Ak,am , j = 1, . . . , a 1,
j m=0 (j 1 m)! dtj1m t=k
(3.4)

whence

1

j1
|Ak,aj |  2(2rb + a)n |Ak,am |
j m=0
 2(2rb + a)n max |Ak,am |, j = 1, . . . , a 1. (3.5)
m=0,1,...,j1

We obtain the following estimate from (3.5) by induction:

 j1
|Ak,aj |  2(2rb + a)n |Ak,a|, j = 1, . . . , a.
 Irrationality of values of the zeta function 519

By (3.3), we have
 a1 ((2r + 1)n)!2b (2n)!a2rb
|Ak,aj |  2(2rb + a)n ,
n!2(a+b)
k = 0, 1, . . . , n, j = 1, . . . , a.

Combining this with (1.11) and (1.12), we obtain that

  2b a2rb
s |  2a+b2 (2n + 1)2 2(2rb + a)n a1 ((2r + 1)n)! (2n)!
|A ,
n!2(a+b) (3.6)
s = 0 or s = b + 1, . . . , a + b 1.

We obtain the limiting relations (3.1) from (3.6) using Stirlings formula (2.17) for
(z) = (z 1)! with positive integers z .
If a is an even integer and j is an even integer such that 0  j < a, then
formula (3.4) with k = 0 implies that |A0,aj |
|A0,a|. If a is even and s is odd,
b < s < a + b, then the summands on the right-hand side of (1.11) have the same
sign, whence
|A
s |
|A0,a|, s = b + 1, . . . , a + b 1 and is odd.
So we have obtained both upper and lower estimates for the asymptotics of coef-
cients of linear forms (1.5) for even a using Stirlings formula. This proves the
limiting relations (3.2) and completes the proof of the proposition.
It turns out that the estimate (3.1) holds for odd values of a as well. The results
obtained in 2 enable us to calculate the asymptotics of the coecients (1.11) for
odd a. This will occupy the rest of this section.
For positive integers m the following expansion holds in the neighbourhood of
the point t = 0:
 2m

sin t (m)
= dl t2l , (3.7)

l=m
(m) (m)
where dl , l = m, m + 1, . . . , are real numbers, dm = 1.
Lemma 3.1. Assume that a is an odd integer. Then the following recurrence
relations hold for the coecients of the linear form (1.5):
 2m
(2m + b 1)! (1)b iM + sin t
A2m+b = R(t) dt
(2m)! (b 1)! i iM


(a1)/2
(2l)! (2m + b 1)! (m) a1
d A2l+b , m = 1, 2, . . . , ,
(2m)! (2l + b 1)! l 2
l=m+1 (3.8)
where M > 0 is an arbitrary real constant.
Proof. If m is a positive integer, then (3.7) implies that the following expansion
holds in the neighbourhood of t = k Z:
 2m  2m

sin t sin (t + k) (m)
= = dl (t + k)2l .

l=m
520 W. Zudilin

Formula (1.6) implies that the following formulae hold in the neighbourhood of
t = k for the function dened by (1.1):

Ak,a Ak,a1 Ak,1

R(t) = a
+ a1
++ + O(1),
(t + k) (t + k) t+k

where k = 0, 1, . . . , n, and R(t) = O(1) in the neighbourhood of t = k Z,

|k| > n. Therefore,

 2m  0 if |k| > n,
sin t
Res R(t) =
(m)
(a1)/2
t=k
dl Ak,2l+1 if |k|  n,
l=m

for m < a/2 and any integer k. If the closed contour L goes around the points
0, 1, . . . , n anticlockwise, then

  2m

(a1)/2
1 sin t (2l)! (b 1)! (m)
R(t) dt = (1)b1 d A2l+b
2i L (2l + b 1)! l
l=m

(2l)! (b 1)! (m)
(a1)/2 
b1 (2m)! (b 1)!
= (1) A2m+b + d A2l+b ,
(2m + b 1)! (2l + b 1)! l
l=m+1 (3.9)

(m)
which follows from (1.11) and the relation dm = 1. We deduce from (3.9) the
following recurrence formulae:
  2m
(2m + b 1)! (1)b sin t
A2m+b = R(t) dt
(2m)! (b 1)! 2i L


(a1)/2
(2l)! (2m + b 1)! (m) a1
d A2l+b , m = 1, 2, . . . , .
(2m)! (2l + b 1)! l 2
l=m+1 (3.10)

Figure 8
 Irrationality of values of the zeta function 521

Now let the contour of integration L be the rectangle with vertices N iM ,

where M > 0 is a xed real constant and N > n is suciently large (see Fig. 8).
As N , the following estimates hold on the vertical sides of the rectangle:
 
 sin t  eM
  R(t) = O(N 2 ).
  ,

Therefore, we have
  2m
1 sin t
R(t) dt
2i L
 iM N iM +N  2m
1 sin t
= + R(t) dt + O(N 1 ) as N
2i iM +N iM N
(3.11)

for m = 1, 2, . . ., (a1)/2, where the constant in O(N 1 ) depends only on M . Since

the integrand in (3.11) is an odd function (see (1.8)) and the contour L is symmetric
with respect to 0, it is sucient to integrate over half the contour and double the
result. Passing to the limit as N , we obtain the recurrence formulae (3.8)
from (3.10), which completes the proof of the lemma.
It follows from Lemma 3.1 that the asymptotic behaviour of the coecients of
the linear form (1.5) is closely connected with the asymptotics of the integrals
iM +  2m
1 sin t
Kn,m = Rn (t) dt
i iM
 2m+b
(1)bn iM + sin t
=
i iM
(t + (2r + 1)n + 1)b (t n)a+b (2n)!a2rb a1
dt, m = 1, 2, . . . , ,
(t + n + 1)a+b 2
(3.12)

as n , where we have used formula (2.15). As in 2, we assume that the cuts

(, n] and [(2r + 1)n, +) are made in the t-plane.
Lemma 3.2 (compare [23], 6.5). For any y0 > 0 the asymptotic equality
 
1    
log (z) = z log z z + log 2 + O |z|1 + O e2 Im z (3.13)
2

holds in the domain Im z
y0 .

Proof. The asymptotic relation (2.17) holds in the domain | arg z| < , but the
identity
2i
(z)(z) = = iz (3.14)
z sin z ze (1 e2iz )
522 W. Zudilin

enables us to write down asymptotics in the quadrant Im z > 0, Re z < 0 as well.

Making the cut (, 0] in the z-plane and xing principal values of the argument,
we deduce from (2.17) and (3.14) the following relation in this quadrant:

log (z) = log(2i) log z + iz log(1 e2iz ) log (z)

   
1
= log(2) + i z + log z log(1 e2iz )
2
  
1  
z (log z i) + z + log 2 + O |z|1
2
 
1  
= z log z z + log 2 + O |z|1 + log(1 e2iz ). (3.15)
2

Observing that |e2iz |  u0 := e2y0 < 1 in the domain Im z
y0 and combin-

ing (3.15) with the relation | ln(1 u)|  C|u|, which holds for |u|  u0 with some
constant C depending only on u0 , we complete the proof of formula (3.13).
Lemma 3.3. Let be a positive real constant. Then the relation
bn a2rb b  
Kn,m = K n,m (1) (2 n ) 2
1 + O(n1 ) + O(e2n )
2m
n a1

holds for the integrals (3.12) as n , where

n,m = 1
i+
a1
K sin2m+b n enf() g( ) d, m = 1, 2, . . . , , (3.16)
i i 2

and the functions f( ), g( ) are dened by (2.12) and (2.13) (see also Remark 2.1).
Proof. The proof repeats the proof of Lemma 2.5 except that we replace the asymp-
totics (2.17) by (3.13) (when integrating over the contour Im t = M = n, > 0).
Lemma 3.4. Assume that (2.61) holds for the real root 1 (2r + 1, +) of the
polynomial (2.21), and assume that
 
log(r2 + r)
|1 | < min 3, (3.17)

for the purely imaginary root 1 (0, +i) of minimal absolute value. Then the
following relation holds for integrals (3.16) as n :

|g(1 )|  
|K
n,(a1)/2| = en Re f0 (1 ) 1 + O(n1 ) ,
|f  (1 )|1/2 2a+b1/2 1/2 n1/2
(3.18)
n,(a1)/2 | O(e2n ), a1
|K
n,m| = |K m = 1, 2, . . . , 1,
2
where the constant is dened by the equality i = 1 .
Proof. By Lemma 2.7, = a + b 1 in equation (2.22) corresponds to the root
i = 1 of the polynomial (2.21).
 Irrationality of values of the zeta function 523

Using the equality

 2m+b
ein ein
sin2m+b n =
2i
(1)(a+b)/2 i(a+b1)n

a+b1
= m a+b1
e + hm,l iei(a+b2l1)n ,
2 i
l=1
a1
m = 1, 2, . . ., ,
2
where  a1
1 if m = ,
m = 2
0 otherwise
and hm,l , m = 1, . . . , (a 1)/2, l = 1, . . . , a + b 1 are real coecients, we write
the integrals (3.16) as

1

(a+b)/2 a+b1
n,m = m (1)
K Jn,a+b1 + hm,l Jn,a+b2l1 ,
2a+b1 (3.19)
l=1
a1
m = 1, 2, . . . , ,
2
where
i+
Jn,k = en(f()ki) g( ) d, k = 0, 1, . . . , (a + b 1). (3.20)
i

First we shall study the asymptotics of the integral Jn,a+b1 . The path of inte-
gration passes through the saddle point 1 = i. We claim that x = 0 is a maximum
point of the function

f(x) = Re(f( ) (a + b 1)i )=x+i . (3.21)

We have

f (x) = Re(f  ( ) (a + b 1)i)=x+i = Re f  (x + i).

Hence, the sole candidates for maximum points of (3.21), besides x = 0, are
x = x0 , where x0 + i is the point of intersection of the ray = x + i, x > 0,
with the curve Re f  ( ) = 0 in the domain Re > 0 (see Fig. 4) at which Re f  ( )
changes the sign from + to (if there is such a point). On the other hand,

f( ) (a + b 1)i = f0 ( ) + f  ( ) (a + b 1)i

for the function f0 ( ) dened in (2.48). By Lemma 2.10 and the inequality
Im f  (x0 + i) > 0, we have

f(x0 ) = f(x0 ) = Re f0 (x0 + i) Im f  (x0 + i) + (a + b 1)

< Re f0 (1 ) + (a + b 1). (3.22)
524 W. Zudilin

The inequality
r(r + 1)
2r + 1 < 1  2r + 1 + 2, := , (3.23)
6(2r + 1)
which follows from (2.61), enables us to obtain an upper estimate for Re f0 (1 ).
We have
Re f0 (1 ) < b(2r +1) log((2r +1+))(a+b) log((r +1)r) < (a2rb) log(r 2 +r),
since (2r + 1 + ) < r(r + 1) by (3.23). Continuing estimate (3.22) and using
inequality (3.17), we obtain the inequalities
f(x0 ) < (a 2rb) log(r2 + r) + (a + b)
 (a 2rb) log(r2 + r) + (a + b) log(r2 + r)
= b(2r + 1) log(r2 + r). (3.24)
To obtain a lower estimate for f(0), we use the inequalities

|i (2r + 1)| > 2r + 1, |i 1|  |i 3 1| = 2.
We have
f(0) = Re f0 (i) > 2(a 2rb) log 2 + 2b(2r + 1) log(2r + 1) 2(a + b) log 2
 
1
= 2b(2r + 1) log r + > b(2r + 1) log(r2 + r). (3.25)
2
Comparing (3.24) with (3.25), we obtain that f(0) > f(x 0 ). Hence, the saddle
point = i is a maximum point of the function Re(f( ) (a + b 1)i ) on the
contour Im = . Therefore, Jn,a+b1 is equal to the contribution of the saddle
point i = 1 :
(2)1/2 g(1 ) n Re(f(1 )(a+b1)i1 )  
Jn,a+b1 =  e 1 + O(n1 )
|f (1 )|1/2 n1/2
(2)1/2 g(1 ) n Re f0 (1 )  
=  e 1 + O(n1 ) , (3.26)
|f (1 )| n
1/2 1/2

and a similar estimate holds for the integral of the modulus of the integrand:
i+
(2)1/2 |g(1 )| n Re f0 (1 )  
|en(f()(a+b1)i) g( )| d =  e 1 + O(n1 ) .
i |f (1 )| n 1/2 1/2

(3.27)
For the integrals (3.20) with k < a + b 1 we use the relation
|en(f()ki) g( )| = |en(f()(a+b1)i) g( )| e(a+b1k)n
on the contour Im = . Combining this with (3.27) and (3.26), we obtain the
inequality
|Jn,k |  Ce(a+b1k)n|Jn,a+b1 |, k = 0, 1, . . . , (a + b 1), (3.28)
with some constant C > 0 that does not depend on n or k. (Inequality (3.28)
is obtained by integration along a nite segment, say [i 1, i + 1], since the
contribution of the integral over the remaining innite part is exponentially small
compared with the contribution of i.) Substituting estimates (3.26) and (3.28)
into (3.19), we obtain (3.18), which completes the proof of the lemma.
 Irrationality of values of the zeta function 525

Proposition 3.2. Let a, b, r be positive integers, let a, b be odd integers, a > 2rb,
assume that the real root 1 (2r + 1, +) of the polynomial (2.21) is such
that (2.61) holds, and assume that (3.17) holds for the purely imaginary root
1 (0, +i) of minimal absolute value. Then the following asymptotic formula
holds for the coecients As of the linear form (1.5):

log |As | 22(a2rb)|1 + 2r + 1|b(2r+1)| 1 + 2r + 1|b(2r+1)

lim  log , (3.29)
n n |1 + 1|a+b |1 1|a+b

where s = 0 or s = b + 1, . . . , a + b 1 and is odd. In the case when s = a + b 1

the upper limit becomes a limit and the inequality becomes an equality.

Proof. The limiting relation

log |A
a+b1 |
lim = Re f0 (1 )
n n

follows from formula (3.8) with m = (a 1)/2 by Lemmas 3.3 and 3.4. For the
other odd integers s, b < s < a+b1, estimates (3.29) follow from Lemmas 3.1, 3.3
and 3.4. In the case when s = 0 formula (1.5) implies that


|A
0 |  |I| + |As |(s),
s is odd
b<s<a+b

whence
0 |
log |A
lim  max{Re f0 (0 ), Re f0 (1 )}, (3.30)
n n
where the root 0 of polynomial (2.21) is dened in Proposition 2.3. Lemma 2.10
implies that Re f0 (0 ) < Re f0 (1 ). The inequality Re f0 (1 ) < Re f0 (1 ) (even a
stronger one) was proved in Lemma 3.4. Therefore, the maximum on the right-hand
side of (3.30) is equal to Re f0 (1 ), which completes the proof of the proposition.

Remark 3.1. Estimate (3.29) is but a slight improvement of (3.1), and will not be
used in the proofs of Theorems 0.3 and 0.4. However, it is quite natural (and could
have been foreseen) that the asymptotics of linear forms and their coecients is
determined by the values of the same function Re f0 ( ) at dierent roots of the
same polynomial (2.21). (In the case when a is an even integer the asymptotics
of the coecients of the linear form (1.5) is determined by the root 0 = 0 of
polynomial (2.21); see Proposition 3.1.)

4. Rened estimates for the denominators of linear forms

The asymptotics of the denominators of linear forms (1.2) (as n ) obtained
in Lemma 1.4, although somewhat coarse, is sucient for the proof of Rivoals theo-
rem. We shall rene our results on denominators using the following generalization
of Lemma 1.2.
526 W. Zudilin

Lemma 4.1. Assume that for some polynomial P (t), deg P (t) < m(n + 1), the
rational function

P (t)
R(t) =  m
(t + s)(t + s + 1) (t + s + n)

(which may be reducible) satises the conditions


Dnj dj  
m 
R(t)(t + k)  Z,
j! dtj t=k (4.1)
k = s, s + 1, . . . , s + n, j = 0, 1, . . . , m 1,

where Dn is the least common multiple of the numbers 1, 2, . . . , n. Then


Dnj dj  
m 
R(t)(t + k)  Z, k = s, s + 1, . . . , s + n, (4.2)
j! dtj t=k

for all non-negative integers j.

Proof. For the integers j = 0, 1, . . . , m 1, the inclusions (4.2) follow directly
from (4.1). In what follows we consider only the integers j
m.
The rational function R(t) can be decomposed into a sum of partial fractions as
follows:


s+n
Bl,0 Bl,1 Bl,m2 Bl,m1

R(t) = + + + + , (4.3)
(t + l)m (t + l)m1 (t + l)2 t+l
l=s

where

1 dj  
m 
Bk,j = R(t)(t + k)  , k = s, s + 1, . . . , s + n, j = 0, 1, . . . , m 1.
j! dtj t=k

Since dierentiation is a linear operation and formulae (4.3) and (4.1) hold, it is
sucient to show that
 
Dnjq dj 1 
(t + k)m  Z,
j! dt j (t + l)mq  (4.4)
t=k
q = 0, 1, . . ., m 1, k, l = 0, 1, . . . , n.

Since
 
((t + l) (l k))m

m
m 1 p m (t + l)
mp
(l k)p
(t + k) = = (1)
(t + l)mq (t + l)mq p=0
p (t + l)mq

m  
m
= (1)p (l k)p (t + l)qp ,
p=0
p
q = 0, 1, . . . , m 1, k, l = 0, 1, . . . , n,
 Irrationality of values of the zeta function 527

we have for the integers j
m that

 
1 dj 1 
(t + k)m 
j! dtj (t + l)mq 
t=k

m  
m (q p)(q p 1) (q p j + 1)
= (1)p
p=0
p j!


(l k)p (t + l)qpj 
t=k

m  
1 p+j m (p q)j
= (1) (4.5)
(l k)jq p=0
p j!

if l = k and
 
1 dj m 1
(t + k) =0
j! dtj (t + l)mq
if l = k. Since
 
Dnjq m (p q)j
Z and Z,
(l k)jq p j!
q = 0, 1, . . . , m 1, p = 0, 1, . . . , m, k, l = 0, 1, . . . , n, k = l,

the inclusions (4.4) follow from (4.5), which completes the proof of the lemma.
The next lemma strengthens the inclusions (1.17). It will be used in our expo-
sition of the general case.
Lemma 4.2. We denote by E = En,m the set of primes dividing each of the numbers

  (m + 2n k)!
Bk,0 = F (t)(t + k)2 t=k = , k = 0, 1, . . . , n,
(m k)! (n k)!2

where the function F (t) is dened in (1.15). Let


= n,m = p. (4.6)
pE
m+3n<p
2n

Then 
j
D2n dj  
2 
1 F (t)(t + k)  Z, k = 0, 1, . . . , n, (4.7)
j! dtj t=k

for all non-negative integers j.

Proof. The fact that 1 Bk,0 Z, k = 0, 1, . . . , n (inclusions (4.7) with j = 0)
follows from the denition of E and . Let us verify inclusions (4.7) with j = 1.
528 W. Zudilin

We have 
d 
2 
Bk,1 = F (t)(t + k) 
dt t=k
 m+2n   



n 
1 1 1 
= Bk,0 + 2 
t+l tl t+l 
l=m+1 l=n t=k
l=k
 


m+2n
1 1


n
1
= Bk,0 2 , k = 0, 1, . . . , n.
lk l+k lk
l=m+1 l=n (4.8)
l=k
By Lemma 1.3, we have D2n Bk,1 Z, k = 0, 1, . . . , n. Therefore,
ordp (1 D2n Bk,1 ) = ordp (D2n Bk,1 )
0 (4.9)
if p is coprime with . Now assume that a prime p divides , whence
ordp = 1. (4.10)

For p > m + 3n we have ordp (lk)  1, where (lk) is any denominator in (4.8).
Therefore,
 m+2n   

1 1
n
1
ordp 2
1, k = 0, 1, . . . , n. (4.11)
lk l+k lk
l=m+1 l=n
l=k

We have
ordp D2n
1 (4.12)
for p  2n. Finally,
ordp Bk,0
1, k = 0, 1, . . . , n, (4.13)
if p E.
Combining estimates (4.10)(4.13) and taking (4.8) into account in the case
when p divides , we obtain that
ordp (1 D2n Bk,1 )
0. (4.14)
1
By (4.9) and (4.14), we have D2n Bk,1 Z, k = 0, 1, . . . , n, which proves
that (4.7) holds for j = 1. In the case when j
2 it remains to use Lemma 4.1
with m = 2.
Note that under the assumptions of Lemma 4.2 we have
     
m + 2n k m k n k
ordp Bk,0 = 2 , k = 0, 1, . . . , n, (4.15)
p p p p p p

for any prime p > m + 3n, since
     
q q q
ordp q! = + 2 + 3 + , q = 1, 2, . . . , (4.16)
p p p
where   stands for the integer part of a number. Formula (4.15) enables us to
calculate the asymptotics of the factor (4.6) as n, m , since
 
{p E : p > m + 3n } = p > m + 3n : min {ordp Bk,0 }
1 . (4.17)
k=0,1,...,n

Consider, for example, the simplest version of (1.15) with m = n.

 Irrationality of values of the zeta function 529

Lemma 4.3.
     
log n,n 1 1 5
1 = lim = + (1) 1 0.4820, (4.18)
n n 2 3 6
where (x) =  (x)/(x) is the logarithmic derivative of the gamma function.

Proof. In the case when m = n and p > 2 n, formula (4.15) implies that
 
n k
min {ordp Bk,0 } = min 1 , , (4.19)
k=0,1,...,n |k/p|
|n/p| p p
where
1 (x, y) = 3x + y + 3x y 3x + y 3x y. (4.20)
It is obvious that the function (4.20) is periodic (with period 1) in each argument:
1 (x, y) = 1 ({x}, {y}), {x} = x x.

Therefore, it is sucient to calculate the values of this function inside the unit
square. By the denition of the integer part of a number, the summands on the
right-hand side of (4.20) change their values only when the point passes through
the lines 3x y = const Z and x y = const Z (see Fig. 9). This enables us to
obtain the values of 1 (x, y) shown in Fig. 10.

Figure 9. The values of the functions
3x + y,
3x y and

3(
x + y +
x y) inside the unit square

Figure 10. The values of the function 1 (x, y) inside the unit square
530 W. Zudilin

Consider the perpendiculars to the x-axis passing through the points of inter-
section of the lines shown in Fig. 10. We have
   
1 if {x} E = 1 , 1 5 , 1 ,
1
min 1 (x, y) = min 1 ({x}, y) = 3 2 6 (4.21)
|y|
x 0
y
1
0 otherwise.
Formulae (4.17), (4.19) and (4.21) imply that
   
n
{p En,n : p > 2 n } = p > 2 n : E1 ,
p
whence
   
n,n = p= p p. (4.22)
p:{n/p}E p:{n/p}E 2n<p
3n
1 1
2 n<p
2n p>2 n

We have the following formula for the denominator of the fraction on the right-hand
side of (4.22):
 D3n
p= . (4.23)
D2n
2n<p
3n

We can nd asymptotics of the numerator using following lemma (cf. [25], Theo-
rem 4.3 and 6, and [26], Lemma 3.2).
Lemma 4.4. The limiting relation
1

lim log p = (v) (u)
n n
p> Cn
{n/p}[u,v)

holds for any C > 1 and any interval [u, v) (0, 1).
Using Lemma 4.4 and relations (4.22), (4.23) and (1.24), we obtain the desired
asymptotics (4.18), which completes the proof of Lemma 4.3.
Remark 4.1. Simple as it is, the calculation of (4.21) can be made four times simpler.
It is easy to verify that function (4.20) is not only periodic, but invariant under the
transformations
 
1 1
1 : (x, y)  x + , y + , 2 : (x, y)  (x, 1 y) (4.24)
2 2
(in the last map we use the shift by 1 and the fact that 1 (x, y) is odd with respect
to y). Therefore, it is sucient to nd the values of the function (4.20) on the
square 0  x, y < 12 and then use the transformations 1 , 1 2 and 2 to extend
this function to the unit square and so by periodicity to the whole of R2 .
The next object of our study is the rational function
(t (n + 1)) (t (n + 2rn))
G(t) = Gn (t) :=  2r . (4.25)
t(t 1) (t n)
Lemma 1.3 and the Leibniz rule for dierentiating a product imply that

j
D2n dj  
2r 
G(t)(t + k)  Z, k = 0, 1, . . . , n, j = 0, 1, 2, . . . . (4.26)
j! dtj t=k
 Irrationality of values of the zeta function 531

(r)
Lemma 4.5. For every integer r
1 there is a sequence of integers n = n
1,
n = 1, 2, . . . , such that

j
D2n dj  
2r 
1
n G(t)(t + k)  Z, k = 0, 1, . . . , n, (4.27)
j! dtj t=k

for all non-negative integers j, and

(r)
r     
r
log n l l 1
r = lim = 2 + (4r + 1) 4r + 4r,
n n r r + 1/2 l
l=1 l=1
(4.28)
where 0.57712 is Eulers constant.
Proof. For n  2r we put n = 1. Then the inclusions (4.27) follow from (4.26).
This nite part of the sequence has no inuence on the limit (4.28). In what follows
we assume that n > 2r.
For every prime p we consider the quantity
 
((2r + 1)n + k)! ((2r + 1)n k)!
p = min ordp . (4.29)
k=0,1,...,n (n + k)!2r+1 (n k)!2r+1
Put 
n = p p . (4.30)

p: (2r+2)n<p
2n

We x an integer k in the interval |k|  n and consider the function

Gk (t) := G(t)(t + k)2r . (4.31)

We claim that

j
D2n dj Gk (t) 
1 Z, j = 0, 1, 2, . . . . (4.32)
n
j! dtj t=k

By Lemma 4.1, it is sucient to prove the inclusions (4.32) for j  2r.

If p is a prime that does not divide n , then (4.26) implies that
    j  
j j 
1 D2n d Gk (t)  D2n dj Gk (t) 
ordp n = ordp
0. (4.33)
j! dtj t=k j! dtj t=k

Now assume that p is a prime dividing n . We shall prove by induction on

j = 0, 1, . . . , 2r that
  
j j 
1 D2n d Gk (t) 
ordp n
0. (4.34)
j! dtj t=k

For j = 0 relation (4.34) follows from the denition of n , since

 ((2r + 1)n + k)! ((2r + 1)n k)!
Gk (t)t=k = .
(n + k)!2r+1 (n k)!2r+1
532 W. Zudilin

We claim that (4.34) holds for j + 1 if it holds for all preceding j. We shall use the
following notation for the logarithmic derivative of the function (4.31):

Gk (t)

(2r+1)n
1

n
1
gk (t) = = (2r + 1) .
Gk (t) t+l t+l
l=(2r+1)n l=n
l=k l=k

We have

1 dj+1 Gk (t) 1 dj  
= gk (t)Gk (t)
(j + 1)! dtj+1 (j + 1)! dtj
1

j
1 djm gk (t) 1 dm Gk (t)
= . (4.35)
j + 1 m=0 (j m)! dtjm m! dtm

We also have
1
ordp = 0, (4.36)
j +1

since p > (2r + 2)n > 2r + 1
j + 1. Further, we have

  
1 djm gk (t) 
ordp
(j m)! dtjm t=k
 (2r+1)n 

(1)jm

n
(1)jm
= ordp (2r + 1)
(l k)jm+1 (l k)jm+1
l=(2r+1)n l=n
l=k l=k

(j m + 1) (4.37)

for m < j, since p > (2r + 2)n and |l k|  (2r + 2)n for all denominators
in (4.37). The inequality

jm+1
ordp D2n
j m+1 (4.38)

holds, since p  2n. Finally,

  
m
D2n dm Gk (t) 
ordp 1
0 (4.39)
n
m! dtm t=k

by the induction hypothesis. Substituting t = k into (4.35) and using esti-

mates (4.36)(4.39), we obtain that relation (4.34) holds for j + 1. This completes
the justication of the induction step.
Combining estimates (4.33) for p  n and (4.34) for p | n , we obtain that the
inclusions (4.32) and (4.27) hold for j = 0, 1, . . . , 2r. Therefore, they hold for all
non-negative integers j. We claim that the limiting relation (4.28) holds for the n ,
n = 1, 2, . . . .
 Irrationality of values of the zeta function 533

By (4.29) and (4.16),

 
n k
p =
min r , (4.40)
|k/p|
|n/p| p p

for every integer n > 2r and prime p > (2r + 2)n, where the function
r (x, y) = (2r +1)x+y+(2r +1)x y (2r +1)x+y(2r +1)xy (4.41)
is periodic (with period 1) in each argument. We claim that
min r (x, y) = if {x} E , = 0, 1, . . . , 2r 1, (4.42)
yR
where    
l l+1 1 l 1 l+1
E2l = , + , + , l = 0, 1, . . ., r 1,
2r 2r + 1 2 2r 2 2r + 1
    (4.43)
l l 1 l 1 l
E2l1 = , + , + , l = 1, 2, . . ., r.
2r + 1 2r 2 2r + 1 2 2r
Taking into account that (4.41) is an odd function, we can easily show that this
function is invariant under the transformations (4.24). Hence, relations (4.42) will
be proved if we prove them in the domain 0  x, y < 12 . In this domain
 
min r (x, y) = min r (x, y) = min (2r + 1)x + y + (2r + 1)x y ,
0
y<1/2 0
y
x 0
y
x

since

0 if 0  y  x < 12 ,
(2r + 1)x + y (2r + 1)x y =
2r + 1 if 0  x < y < 12 .
To complete the proof of relations (4.42), it remains to calculate the values of
the function (2r + 1)x + y + (2r + 1)x y in the domain 0  y  x < 12 (see
Fig. 11) using arguments similar to those used in the proof of Lemma 4.3.

Figure 11. The values of the function r (x, y) in the domain 0  y < x < 1
2
534 W. Zudilin

By (4.30), (4.40) and (4.42), (4.43), we have


2r1 
n = p
=1
p:{n/p}E
(2r+2)n<p
2n
2r1 r1 
    
r 
= p
p
2l
p 2l1
.

=1 p:{n/p}E l=1 2r+1 2r l=1 2r 2r+1
l+1 n<p
l n l n<p
l n
p> (2r+2)n
To obtain the limiting relation (4.28), we use Lemma 4.4, the identity
 
1
(x) + x + = 2 log 2 + 2(2x)
2
(see formula (4.47) below with r = 2) and the relation
1

lim log p = , where < ,
n n
n<p
n

which follows from the prime number theorem (the asymptotic law of distribution
of primes). We have

  l+1 
r1  
l

1 l+1
 
1 l

r = 2l + + +
2r + 1 2r 2 2r + 1 2 2r
l=1
 
2r 2r + 1

l l+1

r         
l l 1 l 1 l
+ (2l 1) + + +
2r 2r + 1 2 2r 2 2r + 1
l=1
 
2r + 1 2r

l l


r1     
r     
l+1 l l l
=2 2l +2 (2l 1)
r + 1/2 r r r + 1/2
l=1 l=1

r1  
r  
2r 2r + 1 2r + 1 2r
2l (2l 1)
l l+1 l l
l=1 l=1

 
 
r
l
r
l  
=2 (2l 1 2l) + 4r(1) + 2 2(l 1) (2l 1)
r r + 1/2
l=1 l=1

r
1 
4rl 2(l 1)(2r + 1) + (2l 1) + 4r
l
l=1

r     

r
l l 1
= 2 + (4r + 1) + 4r(1) + 4r.
r r + 1/2 l
l=1 l=1

Since (1) = (see, for example, [27], 1.1, formula (8)), we obtain the desired
result, which completes the proof of the lemma.
 Irrationality of values of the zeta function 535

Lemma 4.6. The following estimates hold for the quantity dened by (4.28):

1
log(r + 1) 4 < r 4r + (4r + 1) < log(r + 1) + + 2. (4.44)
r
In particular,
r = 4r(1 ) + O(log r) as r . (4.45)

Proof. Since (x) is monotonically increasing on (0, 1], we have

r+1  

r  

r  
l l l
(1) + < < . (4.46)
r+1 r + 1/2 r
l=1 l=1 l=1

We can calculate the sums in the upper and lower estimates using the formula

r  
l1
x+ = r log r + r(rx) (4.47)
r
l=1

(see, for example, [27], 1.1, formula (6)).

Inequality (2.28) implies that the sequence

1 1 1
1+ + + + log r
2 3 r
decreases and the sequence

1 1 1
1+ + + + log(r + 1)
2 3 r

increases. Therefore, their common limit lies between the elements of these
sequences, that is,


r
1
log r < < log(r + 1). (4.48)
l
l=1

Substituting estimates (4.46)(4.48) into (4.28), we obtain that

r+1 r+1
4r log log(r + 1) < r 4r + (4r + 1) < (2r + 1) log + log(r + 1).
r r

Using inequality (2.28) once again, we complete the proof of estimates (4.44). The
limiting relation (4.45) follows immediately from (4.44). The lemma is proved.
Proposition 4.1. The denominator den(In ) of the linear form (1.5) divides
b a+b1 (r)
n D2n , where the sequence of integers n = n , n = 1, 2, . . . , is dened
in Lemma 4.5. Hence,

log den(In ) a+b1

log D2n b log n
lim  lim = 2(a + b 1) br .
n n n n
536 W. Zudilin

Proof. We prove this proposition by replacing (1.22) in the proof of Lemma 1.4 by
the representation
R(t) = Hn (t)a2rb Gn (t)b ,
applying Lemmas 1.3 and 4.5 to the functions (1.16) and (4.25), and using the
Leibniz rule for the dierentiation of a product.
To conclude this section we give some particular values of (4.28) (for r = 1
see (4.18)):

2 2.01561, 3 3.64442, 10 15.38202, 50 82.98948,

100 1.67541 100, 1000 1.68956 1000, r 4(1 )r 1.69114 r.

5. Proof of the results on linear independence

The following theorem is based on the results obtained in 2 and 4.
Theorem 5.1. Let the assumptions of Proposition 2.3 hold, let b be an odd integer,
and let
+ 2(a + b 1) br < 0,
where
r   

 

r
l l 1
r = 2 + (4r + 1) 4r + 4r.
r r + 1/2 l
l=1 l=1

Then at least one of the numbers

(b + 2), (b + 4), . . . , (a + b 1) (5.1)

is irrational.
Proof. By Lemma 1.1, for any integer n
1 the quantity dened by (1.2) is a
linear form in 1 and the numbers (5.1). Assuming that every number in (5.1)
is rational and denoting their least common denominator by D, we obtain that
Jn = D|In | den(In ) is a positive integer for innitely many n
1. This contradicts
the limiting relation
log Jn
lim  + 2(a + b 1) br < 0,
n n
which follows from Propositions 2.3 and 4.1.
Proof of Theorem 0.1. We use Theorem 5.1 with suitable a, b and r = 1 for each
set of numbers. Note that 1 < 3.05 in each of the cases considered below. Hence,
assumption (2.61) holds.
We prove that at least one of the elements of the rst set in (0.1) is irra-
tional by putting a = 19 and b = 3 in Theorem 5.1. In this case 1 3.04028,
0 2.98027 + 0.02985i, and Im f0 (0 ) + 3 0.09308  0 (mod Z). Combining
the estimate
+ 2(a + b 1) br 0.72567 < 0
with Theorem 5.1, we complete the proof of the assertion for the rst set in (0.1).
 Irrationality of values of the zeta function 537

For the second we put a = 33 and b = 5. Then

1 3.03309, 0 3.00783 + 0.03046i, Im f0 (0 ) + 9 0.14978,

+ 2(a + b 1) br 0.76662 < 0.

Finally, putting a = 47 and b = 7 for the third, we obtain

1 3.03043, 0 3.01730 + 0.02406i, Im f0 (0 ) + 15 0.16232,

+ 2(a + b 1) br 0.82928 < 0.

The theorem is proved.

Proof of Theorem 0.2. We put a = 7b and r = 1 for every odd integer b
3
and denote by b the root of the corresponding polynomial (2.21) in the domain
Re > 0, Im > 0 with the maximal possible Re b . Note that the root of poly-
nomial (2.21) in the interval (3, +) coincides with the root of the polynomial

( + 3)( 1)8 ( 3)( + 1)8 ,

which is equal to 1 3.02472. Since (2.61) holds, Lemma 2.10 implies that
Re f0 (b )  Re f0 (1 ), whence

22(a2)(1 + 3)3 (1 3)3

 b0 := b log 15.56497 b.
(1 + 1)8 (1 1)8

Therefore,

+ 2(a + b 1) br < b0 + 16b b1 0.04701 b < 0.

We shall be able to use Theorem 5.1 if we can show that (2.55) holds. It is easy to
verify that

g0 (b) = Im f0 (b ) + 3(b 2)
   
= 3b arg(b 3) + arg(b + 3) 8b arg(b 1) + arg(b + 1) + 3(b 2)
 
= 2b 3 arg(b 3) + 8 arg(b 1) 16 arg(b + 1) ,

regarded as a function of b
3, increases, whence g0 (b)
g0 (3) 0.05935 > 0.

The relations

b2
arg(b + 3) , |b + 3| |1 3| as b , b is odd,
b

enable us to calculate the limit

lim g0 (b) < .

b
b is odd
538 W. Zudilin

Therefore, Im f0 (b )  0 (mod Z). An application of Theorem 5.1 completes the

proof.
Proof of Theorem 0.3. By Lemma 1.1 and Proposition 4.1 with b = 1, the In (a, r) =
a (r)
In D2n /n dened by (1.2) are linear forms in 1, (3), (5), . . . , (a) with integer
coecients. By Propositions 2.1, 3.1, and 4.1 the , in the criterion for linear
independence (see the introduction) are dened by the formulae

(a, r) = (a, r) 2a + r ,
(5.2)
(a, r) = 2(2r + 1) log(2r + 1) + 2(a 2r) log 2 + 2a r ,

where = (a, r) is dened by (2.59). For a = 145, r = 10 and a = 1971, r = 65

we have

(145, 10)
1 (145, 10) 21 0.38013 104 , 1+ 2.000397,
(145, 10)
(1971, 65)
1 (1971, 65) 131 0.22019 1010, 1+ 3.000103,
(1971, 65)

where 1 (a, r) is the real root of polynomial (2.21) with b = 1 in the interval
(2r + 1, +). The criterion for linear independence implies that

(145)
3, (1971)
4, (5.3)

since the dimension of space is an integer. By Aperys theorem, (3) = 2. The

assertion of the theorem follows from the estimates obtained.
Lemma 5.1. Assume that the positive integers a, r are such that

a > 2(r + 2) log(r + 1) + 1, (5.4)

and let 1 = 1 (a, r) be a real root of the polynomial

h( ) = ( + 2r + 1)( 1)a+1 ( 2r 1)( + 1)a+1 (5.5)

in the interval (2r + 1, +). Then

2r + 1
0 < 1 (2r + 1) < 2, where := < 1. (5.6)
(r + 1)2

Proof. Direct calculations show that

h(2r + 1) = 2a+2 (r + 1)ra+1 > 0, (5.7)

 
h(2r + 1 + 2) = 2a+2 (2r + 1 + )(r + )a+1 (r + 1 + )a+1 . (5.8)

Inequality (5.4) implies that

(r + 1 + )a1 > (r + )a+1 , (5.9)

 Irrationality of values of the zeta function 539

since

(a 1) log(r + 1 + ) (a + 1) log(r + )
 
1
= (a 1) log 1 + 2 log(r + )
r+
 
1
> (a 1) log 1 + 2 log(r + 1)
r+1
a1 a 1 2(r + 2) log(r + 1)
> 2 log(r + 1) = > 0.
r+2 r+2

Besides,

(2r + 1)(r + 1 + )2
(r + 1 + )2 =
(r + 1)2
(2r + 1)(r + 1 + )2 (3r2 + 4r + 1 + 2r + )
> 2
= 2r + 1 + .
(r + 1) (r + 1)2
(5.10)

Substituting inequalities (5.9) and (5.10) into (5.8), we obtain that h(2r+1+2) < 0.
Combining this inequality with (5.7), we obtain that the interval (2r +1, 2r +1+2)
contains at least one root of the polynomial (5.5). Now (5.6) follows from the fact
that 1 is the unique real root in the interval (2r + 1, +) (Lemma 2.7). The
lemma is proved.
Corollary 5.1. Let the assumptions of Lemma 5.1 hold with r > 6. Then the
following estimate holds for = (a, r) in (2.59):

(a, r) < 2(2r + 1) log(2r + 1) 2(a + 1)(1 + log r).

Proof. Since h(1 ) = 0, we have

22(a2r)(1 + 2r + 1)2(2r+1)(1 1)2(a+1)r

= log .
(1 + 1)2(a+1)(r+1)

Combining this with (5.6), we obtain that

(a 2r) log 2
2    
2r 2
= (2r + 1) log 1 + (a + 1)r log 1 + (a 2r) log(1 + 1)
1 + 1 1 1
 
r (a + 1)r
< (2r + 1) log 1 + (a 2r) log(2r + 2)
r+1 r+1
 
r
= (2r + 1) log(2r + 1) (a + 1) + log(r + 1) (a 2r) log 2
r+1
< (2r + 1) log(2r + 1) (a + 1)(1 + log r) (a 2r) log 2.
540 W. Zudilin

Here we have used the inequality

     
2 1 1 1
log 1 + > log 1 + > 2 log 1 + >
1 1 r+ 2r + 1 r+1
(see also (2.28) with n = 2r + 1), which holds for r > 6 since
 2
1 1 r 4r 3r + 1
1+ 1+ =
r+ 2r + 1 (r + )(2r + 1)2
r 5r 7r 2
3 2
= >0 if r > 6.
(r + )(r + 1)2 (2r + 1)2
The corollary is proved.
Proof of Theorem 0.4. Since (a) is an integer and
0.395 log 3 0.43395, 0.395 log 145 1.96581, 0.395 log 1971 2.99659,
Aperys theorem and estimates (5.3) imply that inequality (0.2) holds for all odd
integers a < e4/0.395, that is, a < 24999. In fact we shall show that the following
estimate holds for odd integers a
20737 = 124 + 1, which is stronger than (0.2):
log a
(a) > , (5.11)
log 12
or, which is the same,
(12m + 1) > m, m = 4, 5, 6, . . . . (5.12)
For every a = 12m + 1 we put
 2   m 
log 12 a 12 + 1
r= = , m = 4, 5, 6, . . .,
3 log2 a 3m2
and obtain a lower estimate for (a) using the criterion for linear independence,
formulae (5.2), and the inequalities in Lemma 4.6 and Corollary 5.1:
(a, r)
(a)
1 +
(a, r)
2(a + 1)(1 + log r) + 2(a 2r) log 2
> .
2(2r + 1) log(2r + 1) + 2(a 2r) log 2 + 2a 4r(1 ) + log(r + 1) + 4 +
(5.13)

Now for m = 4, 5, 6, 7 estimate (5.12) follows immediately from (5.13):

 
m a
m a = 12 + 1 r= (a)
3m2
4 20737 432 > 4.00882
5 248833 3317 > 5.15339
6 2985985 27648 > 6.35168
7 35831809 243753 > 7.58967
 Irrationality of values of the zeta function 541

For the remaining values of m we combine (5.13) with the trivial estimates

r 1 2r + 1 2 1 2
0<  ,  +  2 ,
a 3m2 a 3m2 a m log 12
12m
log r
log = m log 12 2 log 2 2 log m,
4m2
log(2r + 1)  log(a 1) = m log 12.

We have

m log 12 2 log m + 1 log 2 1/(3m2 ) m log 12 2 log m

(a) = (12m+1) > > .
1 + log 2 + 2/m 1 + log 2 + 2/m

Now (5.12) with m
8 follows from the fact that the function
 
2
(m) = (m log 12 2 log m) m 1 + log 2 + = m(log 6 1) 2 log m 2
m

increases monotonically and takes a positive value, (8) 0.17519, at m = 8. This

completes the proof of the theorem.
In fact we have proved estimate (5.11) for odd integers a
3 with the exception
of nitely many values in the range 123 + 1 = 1729  a  1969.

Bibliography
[1] R. Apery, Irrationalite de (2) et (3), Ast
erisque 61 (1979), 1113.
[2] A. van der Poorten, A proof that Euler missed... Aperys proof of the irrationality of (3).
An informal report, Math. Intelligencer 1:4 (1978/79), 195203.
[3] F. Beukers, A note on the irrationality of (2) and (3), Bull. London Math. Soc. 11
(1979), 268272.
[4] L.A. Gutnik, The irrationality of certain quantities involving (3), Uspekhi Mat. Nauk
34:3 (1979), 190; English transl., Russian Math. Surveys 34:3 (1979), 200; Acta Arith.
42:3 (1983), 255264.
[5] F. Beukers, Pade approximations in number theory, Lecture Notes in Math., vol. 888,
Springer-Verlag, Berlin 1981, pp. 9099.
[6] F. Beukers, Irrationality proofs using modular forms, Ast erisque 147148 (1987),
271283.
[7] V. N. Sorokin, HermitePade approximations of Nikishin systems and the irrationality
of (3), Uspekhi Mat. Nauk 49:2 (1994), 167168; English transl., Russian Math. Surveys
49:2 (1994), 176177.
[8] Yu. V. Nesterenko, A few remarks on (3), Mat. Zametki 59 (1996), 865880; English
transl., Math. Notes 59 (1996), 625636.
[9] M. Hata, A new irrationality measure for (3), Acta Arith. 92:1 (2000), 4757.
[10] G. Rhin and C. Viola, The group structure for (3), Acta Arith. 97:3 (2001), 269293.
[11] D. V. Vasilyev, On small linear forms for the values of the Riemann zeta-function at odd
points, Preprint no.1 (558), Inst. Math., Nat. Acad. Sci. Belarus, Minsk 2001.
[12] K. Ball, Diophantine approximation of hypergeometric numbers, Manuscript, 2000.
[13] T. Rivoal, La fonction zeta de Riemann prend une innite de valeurs irrationnelles aux
entiers impairs, C. R. Acad. Sci. Paris S er. I Math. 331:4 (2000), 267270; http://
arXiv.org/abs/math/0008051.
[14] T. Rivoal, Propriet
es diophantiennes des valeurs de la fonction z eta de Riemann aux entiers
impairs, Th`ese de Doctorat, Univ. de Caen, Caen 2001.
542 W. Zudilin

[15] K. Ball and T. Rivoal, Irrationalit

e dune innit
e de valeurs de la fonction zeta aux entiers
impairs, Invent. Math. 146:1 (2001), 193207.
[16] Yu. V. Nesterenko, On the linear independence of numbers, Vestnik Moskov. Univ. Ser. 1
Mat. Mekh. 1985, no. 1, 4654; English transl., Moscow Univ. Math. Bull. 40:1 (1985),
6974.
[17] T. G. Hessami Pilehrood, Arithmetical properties of values of hypergeometric functions,
Diss. Cand. Phys.-Math. Sci., Mosk. Gos. Univ., Moscow 1999; Linear independence of
vectors with polylogarithmic coordinates, Vestnik Moskov. Univ. Ser. 1 Mat. Mekh. 1999,
no. 6, 5456; English transl., Moscow Univ. Math. Bull. 54:6 (1999), 4042
[18] E. A. Rukhadze, A lower estimate for rational approximations of ln 2, Vestnik Moskov.
Univ. Ser. 1 Mat. Mekh. 1987, no. 6, 2529; English transl., Moscow Univ. Math. Bull.
42:6 (1987), 3035.
[19] V. V. Zudilin, On the irrationality of the values of the zeta function at odd integer
points, Uspekhi Mat. Nauk 56:2 (2001), 215216; English transl., Russian Math. Surveys
56 (2001), 423424.
[20] E. M. Nikishin, On the irrationality of the values of the functions F (x, s), Mat. Sbornik
109 (1979), 410417; English transl., Math. USSR-Sb. 37 (1980), 381388.
[21] Yu. V. Nesterenko, Diophantine approximations to Riemanns zeta function, Report on
the Workshop on Number Theory (4 October, 2000), Mosk. Gos. Univ., Moscow 2000.
[22] S. Lang, Introduction to modular forms, Grundlehren Math. Wiss., vol. 222, Springer-Verlag,
Berlin 1976; Russian transl., Mir, Moscow 1979.
[23] N. G. de Bruijn, Asymptotic methods in analysis, North-Holland, Amsterdam 1958; Russian
transl., Inostr. Lit., Moscow 1961.
[24] M. Yoshida, Hypergeometric function, my love, Aspects of Math., vol. E 32, Vieweg,
Wiesbaden 1997.
[25] G. V. Chudnovsky, On the method of ThueSiegel, Ann. of Math. (2) 117 (1983),
325382.
[26] M. Hata, Legendre type polynomials and irrationality measures, J. Reine Angew. Math.
407:1 (1990), 99125.
[27] Yu. L. Luke, Mathematical functions and their approximations, Academic Press, New York
1975; Russian transl., Mir, Moscow 1980.
[28] V. V. Zudilin, One of the numbers (5), (7), (9), (11) is irrational, Uspekhi Mat. Nauk
56:4 (2001), 149150; English transl., Russian Math. Surveys 56 (2001), 774776.

Moscow State University,

Moscow 117234, Russia
E-mail address: wadim@ips.ras.ru
Received 24/APR/01
Translated by V. M. MILLIONSHCHIKOV

Typeset by AMS-TEX