TAIWANESE JOURNAL OF MATHEMATICS Vol. 4, No. 4, pp.

569-598, December 2000

SOME FAMILIES OF RAPIDLY CONVERGENT SERIES REPRESENTATIONS FOR THE ZETA FUNCTIONS H. M. Srivastava

Abstract. Many interesting families of rapidly convergent series representations for the Riemann Zeta function (2n + 1) (n N) were considered recently by various authors. In this survey-cum-expository paper, the author presents a systematic (and historical) investigation of these series representations. Relevant connections of the results presented here with several other known series representations for (2n + 1) (n N) are also pointed out. In one of many computationally useful special cases presented here, it is observed that (3) can be represented by means of a series which converges much faster than that in Eulers celebrated formula as well as the series used recently by Apry in his proof of the e irrationality of (3). Symbolic and numerical computations using Mathematica (Version 4.0) for Linux show, among other things, that only 50 terms of this series are capable of producing an accuracy of seven decimal places.

1. INTRODUCTION AND HISTORICAL BACKGROUND Let S denote the set of all nontrivial integer kth powers, that is, (1.1) S : = nk : n, k N \ {1} (N := {1, 2, 3, . . . }) = {4, 8, 9, 16, 25, 27, 32, 36, . . . }.

0 Received June 13, 2000. Communicated by P. Y. Wu. 2000 Mathematics Subject Classication: Primary 11M06, 11M35, 33B15; Secondary 11B68, 33E20, 40A30. Key words and phrases: Riemann (and Hurwitz) Zeta function, Bernoulli number, Goldbachs theorem, harmonic number, Euler polynomial, Eulers formula, Bernoulli polynomial, lHpitals rule, Wiltons formula. o

569

570

H. M. Srivastava

An over two centuries old theorem of Christian Goldbach (1690 - 1764), which was stated in a letter dated 1729 from Goldbach to Daniel Bernoulli (1700 1782), was revived, not too long ago, as the following problem [19]: (1.2)
S

( 1)1 = 1,

the sum being extended over all members of the set S. In terms of the Riemann Zeta function (s), Goldbachs theorem (1.2) can easily be restated as

(1.3)
k=2

{(k) 1} = 1

or, equivalently, as

(1.4)
k=2

F((k)) = 1,

where, for convenience, F(x) := x [x] denotes the fractional part of x R. In fact, it is fairly easy to show also that

(1.5) (1.6)

k=2

1 (1)k F((k)) = , 2 3 F((2k)) = , 4

k=1

and

(1.7)
k=1

1 F((2k + 1)) = . 4

Here, as usual, the Riemann Zeta function (s) and the (Hurwitzs) generalized Zeta function (s, a) are dened (for R(s) > 1) by 1 1 1 = (R(s) > 1) s n=1 ns 1 2 n=1 (2n 1)s (1.8) (s) := (1)n1 1 (R(s) > 0; s = 1) 1 21s n=1 ns and

(1.9)

(s, a) :=
n=0

1 (n + a)s

R(s) > 1; a Z := {0, 1, 2, . . . } , / 0

Series Representations for the Zeta Functions

571

and (for R(s) 1; s = 1) by their meromorphic continuations (see, for details, Titchmarsh [28]), so that (obviously) (1.10) (s, 1) = (s) = (2s 1)1 s, 1 2 and (s, 2) = (s) 1.

Another remarkable result involving Riemanns -function is the following series representation for (3): (1.11) (3) = 4 2 7
k=0

(2k) , (2k + 1)(2k + 2)22k

which was contained in a 1772 paper, entitled Exercitationes Analyticae, by Leonhard Euler (1707 - 1783) (cf., e.g., Ayoub [2, pp. 1084 - 1085]). In fact, this result of Euler was rediscovered (among others) by Ramaswami [18] (see also Srivastava [20, p. 7, Equation (2.23)]) and (more recently) by Ewell [8]. And, as pointed out by (for example) Chen and Srivastava [4, pp. 180 - 181], another series representation: (1.12) 5 (3) = 2 (1)k1 , 2k k=1 k 3 k

which played a key rle in Aprys celebrated proof [1] of the irrationality of o e (3), was proven independently by (among others) Hjortnaes [13], Gosper [11], and Apry [1]. e Clearly, Eulers series in (1.11) converges faster than the dening series for (3), but obviously not as fast as the series in (1.12). Such Zeta values as (3), (5), et cetera are known to arise naturally in a wide variety of applications (see, for example, Tricomi [29], Witten [32], and Nash and OConnor [16, 17]). On the other hand, in the case of even integer arguments, we have the computationally useful relationship: (1.13) (2n) = (1)n1 (2)2n B2n 2 (2n)!

(n N0 := N {0})

with the well-tabulated Bernoulli numbers dened by the generating function: (1.14) z = z 1 e Bn
n=0

zn n!

(|z| < 2),

as well as the familiar recursion formula: (1.15) (2n) = 1 n+ 2

(2k) (2n 2k)

(n N \ {1}).

572

H. M. Srivastava

Thus there is a need for expressing (2n + 1) as a rapidly converging series for all n N. With this objective in view, we propose to develop here a rather systematic investigation of the various families of rapidly convergent series representations for the Riemann (2n + 1) (n N). We also consider relevant connections of the results presented here with many other known series representations for (2n + 1) (n N). In one of many computationally useful special cases considered here, it is observed that (3) can be represented by means of a series which converges much more rapidly than that in Eulers celebrated formula (1.11) as well as the series (1.12) used recently by Apry [1] e in his proof of the irrationality of (3). Symbolic and numerical computations using Mathematica (Version 4.0) for Linux show, among other things, that only 50 terms of this series are capable of producing an accuracy of seven decimal places. 2. THE FIRST SET OF SERIES REPRESENTATIONS The various series identities considered in the preceding section, including (for example) Goldbachs theorem (1.2), are known to be derivable also from the following simple consequence of the binomial theorem and the denition (1.9):

(2.1)
k=0

(s)k (s + k, a)tk = (s, a t) (|t| < |a|), k!

which, for a = 1 and t = 1/m, readily yields the series identity: (s)2k (s + 2k) (2k)! m2k k=0 (2s 1) (s) 2s1 m2 = j 1 (ms 1) (s) ms s, 2 m j=2

(2.2)

(m = 2) (m N \ {1, 2}),

()n := ( + n)/() being the Pochhammer symbol (or the shifted factorial, since (1)n = n!). In terms of the familiar harmonic numbers
n

(2.3)

Hn :=
j=1

1 j

(n N),

Series Representations for the Zeta Functions

573

the following series representations for (2n + 1) were proven recently by appealing appropriately to the series identity (2.2) in its special cases when m = 2, 3, 4, and 6 (see Srivastava [24]): (2n + 1) = (1)n1
n1

(2)2n H2n log 22n+1 1 (2n)!

(2.4)

+
k=1

(1)k (2k + 1) (2n 2k)! 2k (2k 1)! (2k) (2n + 2k)! 22k (n N);

+2
k=1

(2n + 1) = (1)n1
n1

2(2)2n H2n log 32n+1 1 (2n)!

2 3

(2.5)

+
k=1

(1)k (2k + 1) (2n 2k)! 2 2k

+2
k=1

(2k 1)! (2k) (2n + 2k)! 32k

(n N);

(2n + 1) = (1)n1
n1

H2n log 2(2)2n 4n+1 + 22n 1 2 (2n)!

1 2

(2.6)

+
k=1

(1)k (2k + 1) (2n 2k)! 1 2k

+2
k=1

(2k 1)! (2k) (2n + 2k)! 42k

(n N);

(2n + 1)

= (1)n1
n1

2(2)2n 32n (22n + 1) + 22n 1

H2n log (2n)!

1 3

(2.7)

+
k=1

(1)k (2k + 1) (2n 2k)! 1 2k

+2
k=1

(2k 1)! (2k) (2n + 2k)! 62k

(n N).

574

H. M. Srivastava

Here (and elsewhere in this work) an empty sum is to be interpreted (as usual) to be nil. We choose to recall the proof of (2.4) detailed by Srivastava [24]. Each of the other results (2.5), (2.6), and (2.7) can be proven mutatis mutandis. The following properties of the Riemann -function will also be required in these derivations: (2.8) 1 (0) = ; 2 (2n) = 0 (n N); 1 (0) = log(2), 2

and (in general) (2n) = lim (2.9) = (2n + )

(1)n (2n)! (2n + 1) (n N), 2(2)2n

where use is made of the familiar functional equation: (2.10) 2s (1 s) (1 s) sin 1 s 2 = 1s (s).

Furthermore, by lHpitals rule, it is easily seen that o (2.11) and (2.12) lim (s + 2k) s + 2n = (1)nk (2n 2k)! (2n 2k + 1) 2(2)2(nk) lim sin 1 s 2 s + 2n = (1)n 2 (n N)

s2n

s2n

(k = 1, , n 1; n N \ {1}). First of all, upon separating the rst n + 1 terms of the series occurring on the left-hand side of the case m = 2 of the general result (2.2), if we transpose the terms for k = 0 and k = n to the right-hand side, we readily obtain the identity:
n1

(2.13)

k=1

(s)2k 2(nk) 2 (s + 2k) + (2k)!

k=1

(s)2n+2k (s + 2n + 2k) (2n + 2k)! 22k (n N),

= 22n (2s 2)(s) 2s+2n1

(s)2n (s + 2n) (2n)!

Series Representations for the Zeta Functions

575

it being understood, as mentioned before, that an empty sum is to be interpreted as nil. Now we apply the functional equation (2.10) in the rst term on the righthand side of (2.13) and divide both sides by s + 2n. We thus nd that
n1 k=1

(s)2k 2(nk) 2 (2k)!

(s + 2k) s + 2n

+
k=1

(s)2n (s + 2n + 1)2k1 (s + 2n + 2k) (2n + 2k)! 22k sin 1 s 2 s + 2n

(2.14)

= 2s+2n (2s 2) s1 (1 s)(1 s)

2s+2n1 + (s)2n (s + 2n) (2n)! (s = 2n; n N). s + 2n Since (n)k = (1)k so that, obviously, (2.15) (n)n = (1)n n! (n N), n! (n k)! (k = 0, 1, , n; n N),

it is easily seen by logarithmic dierentiation that d {(s)n } = (s)n ds

(2.16)

1 s+j

(n N),

so that (2.17) d {(s)2n } ds = (2n)! H2n

(n N),

where Hn denotes the harmonic numbers dened by (2.3). We observe also that the limit formula (2.12) is needed in the rst sum on the left-hand side of (2.14) only when this sum is nonzero (that is, only when n N \ {1}).

576

H. M. Srivastava

Furthermore, by lHpitals rule once again, we have o 2s+2n1 + (s)2n (s + 2n) (2n)! lim s2n s + 2n (2.18) d (s + 2n) (s)2n = 2s+2n1 log 2 + {(s)2n } + (s + 2n) ds (2n)! (2n)! 1 = (H2n log ) (n N). 2

s=2n

Finally, letting s 2n in (2.14), and making use of the limit relationships (2.12) and (2.18), we obtain the rst series representation for (2n+1) asserted by (2.4). The series representation (2.4) is markedly dierent from each of the series representations for (2n + 1), which were given earlier by Zhang and Williams [33, p. 1590, Equation (3.13)] and (subsequently) by Cvijovi and Klinowski c [5, p. 1265, Theorem A]. Since (2k) 1 as k , the general term in the series representation (2.4) has the order estimate: O 22k k 2n1 (k ; n N),

whereas the general term in each of these earlier series representations has the order estimate: O 22k k 2n (k ; n N). By suitably combining (2.4) and (2.6), it is fairly straightforward to obtain the series representation: (2n + 1) = (1)n1
n1

log 2 2(2)2n 2n 1)(22n+1 1) (2n)! (2

(2.19)

+
k=1

(1)k (22k 1) (2k + 1) (2n 2k)! 2k (2k 1)!(22k 1) (2k) (2n + 2k)! 24k (n N).

2
k=1

Now, in terms of the Bernoulli numbers Bn and the Euler polynomials En (x) dened by the generating functions (1.14) and (2.20) 2exz = ez + 1

En (x)
n=0

zn n!

(|z| < ),

Series Representations for the Zeta Functions

577

respectively, it is known that (cf., e.g., Magnus et al. [15, p. 29]) (2.21) En (0) = (1)n En (1) = 2(1 2n+1 ) Bn+1 n+1 (n N),

which, together with the identity (1.13), implies that (2.22) E2n1 (0) = 4(1)n (2n 1)!(22n 1) (2n) (2)2n (n N).

By appealing to the relationship (2.22), the series representation (2.19) can immediately be put in the form: (2n + 1) = (1)n1
n1

log 2 2(2)2n 2n 1)(22n+1 1) (2n)! (2

(2.23)

+
k=1

(1)k (22k 1) (2k + 1) (2n 2k)! 2k (1)k1 (2n + 2k)! 2

1 2

k=1

E2k1 (0)

(n N),

which is a slightly modied (and corrected) version of a result proven in a signicantly dierent way by Tsumura [30, p. 383, Theorem B]. Another interesting combination of our series representations (2.4) and (2.6) leads us to the following variant of Tsumuras result (2.19) or (2.23): (2n + 1) = (1)n1
n1

H2n log 2n 2n+1 1 2 (2n)!

1 4

(2.24)

+
k=1

(1)k (22k+1 1) (2k + 1) (2n 2k)! 2k (2k 1)!(22k1 1) (2k) (2n + 2k)! 24k (n N),

4
k=1

which is essentially the same as the determinantal expression for (2n + 1) derived recently by Ewell [9, p. 1010, Corollary 3] by employing an entirely dierent technique from ours. Other similar combinations of the series representations (2.4) to (2.7) would

578

H. M. Srivastava

yield the following interesting companions of Ewells result (2.24): (2n + 1) = (1)n1
n1

H2n log 2(2)2n (22n+1 1)(32n + 1) (2n)!

1 6

(2.25)

+
k=1

(1)k (22k+1 1) (2k + 1) 2k 2 (2n 2k)!

4
k=1

(2k 1) (2k) (2n + 2k)! 62k

1)!(22k1

(n N),

(2n + 1) (2.26)

= (1)n1
n1

2H2n log 2(2)2n 2n + 1)(32n+1 1) (2 (2n)!

2 27

+
k=1

(1)k (32k+1 1) (2k + 1) (2n 2k)! 2k (2k 1)!(32k1 1) (2k) (2n + 2k)! 62k (n N),

6
k=1

(2n + 1) = (1)n1
n1

H2n log 2(2)2n 2n+2 22n+3 + 1 3 (2n)!

8 27

(2.27)

+
k=1

(1)k (32k+1 22k+1 ) (2k + 1) (2n 2k)! (2)2k (2k 1)!(32k1 22k1 ) (2k) (2n + 2k)! 62k (n N),

12
k=1

(2n + 1) = (1)n1
n1

H2n log 2(2)2n 4n+3 + 22n+2 32n+2 1 2 (2n)!

27 128

(2.28)

+
k=1

(1)k (42k+1 32k+1 ) (2k + 1) (2n 2k)! (2)2k (2k 1)!(42k1 32k1 ) (2k) (2n + 2k)! 122k (n N),

24
k=1

Series Representations for the Zeta Functions

579

and (2n + 1) = (1)n1

H2n log 2(2)2n 2n+1 (22n + 1) 24n+2 + 22n 1 3 (2n)!

4 27

(2.29)

+
k=1

(1)k (32k+1 22k+1 ) (2k + 1) (2n 2k)! 2k (2k 1)!(32k1 22k1 ) (2k) (2n + 2k)! 122k (n N).

12
k=1

Next we turn to the following obvious consequence of the series identity (2.1):

(2.30)

1 = [(s, a t) (s, a + t)] (|t| < |a|). 2

k=0

(s)2k+1 (s + 2k + 1, a) t2k+1 (2k + 1)!

By setting t = 1/m and dierentiating both sides with respect to s, we nd from (2.30) that

(s)2k+1 (s + 2k + 1, a) + (s + 2k + 1, a) (2k + 1)!m2k s, a 1 m s, a + 1 m

2k j=0

1 s+j

(2.31)

k=0

m 2 s

(m N \ {1}),

where we have made use of the derivative formula (2.16). In particular, when m = 2, (2.31) immediately yields

(2.32)

k=0

(s)2k+1 (s + 2k + 1, a) + (s + 2k + 1, a) (2k + 1)!22k 1 2

2k j=0

1 s+j

= a

log a

1 2

By letting s 2n1 (n N) in the further special of this last identity (2.32) when a = 1, Wilton [31, p. 92] obtained the following series representation for

580

H. M. Srivastava

(2n + 1) (see also Hansen [12, p. 357, Entry (54.6.9)]): (2n + 1) = (1)n1 2n
n1

H2n+1 log (2n + 1)!

(2.33)

+
k=1

(1)k (2k + 1) (2n 2k + 1)! 2k (2k 1)! (2k) (2n + 2k + 1)! 22k (n N),

+2
k=1

which, in view of the identity: (2.34) (2k)! (2k 1)! (2k 1)! = 2n (2n + 2k)! (2n + 2k 1)! (2n + 2k)! (n N),

would combine with the result (2.4) to yield the series representation: (2n + 1) = (1)n (2.35) +
k=0

(2)2n n(22n+1 1)

n1 k=1

(1)k1 k (2k + 1) (2n 2k)! 2k (n N).

(2k)! (2k) (2n + 2k)! 22k

The series representation (2.35) is precisely the aforementioned main result of Cvijovi and Klinowski [5, p. 1265, Theorem A]. In fact, in view of c the derivative formula (2.19), the series representation (2.35) is essentially the same as a result given earlier by Zhang and Williams [33, p. 1590, Equation (3.13)] (see also Zhang and Williams [33, p. 1591, Equation (3.16)], where an obviously more complicated (asymptotic) version of (2.35) was proven similarly). Observing also that (2.36) (2k 1)! (2k 1)! (2k)! = (2n + 1) (2n + 2k + 1)! (2n + 2k)! (2n + 2k + 1)! (n, k N),

we obtain yet another series representation for (2n + 1) by applying (2.4) and (2.33): (2n + 1) = (1)n (2.37) +
k=0

2(2)2n (2n 1)22n + 1

n1 k=1

(1)k1 k (2k + 1) (2n 2k + 1)! 2k (n N),

(2k)! (2k) (2n + 2k + 1)! 22k

Series Representations for the Zeta Functions

581

which provides a signicantly simpler (and much more rapidly convergent) version of the other main result of Cvijovi and Klinowski [5, p. 1265, Theorem c B]: (2.38) (2n + 1) = (1)n 2(2)2n (2n)!

n,k
k=0

(2k) 22k

(n N),

where the coecients n,k are given explicitly as a nite sum of Bernoulli numbers [5, p. 1265, Theorem B(i)] (see, for details, Srivastava [24, pp. 393 394]):
2n

(2.39)

n,k :=
j=0

B2nj 2n j (j + 2k + 1)(j + 1)2j

(n N; k N0 ).

3. ANOTHER FAMILY OF SERIES REPRESENTATIONS Starting once again from the identity (2.1) with (of course) a = 1, t = 1/m, and s replaced by s + 1, and applying (2.2), we nd yet another class of series identities including, for example,

(3.1)
k=1

(s + 1)2k (s + 2k) = (2s 2)(s) (2k)! 22k

and
k=1

(s + 1)2k (s + 2k) (2k)! m2k

(3.2)

1 1 m(ms 3)(s) + (ms+1 1)(s + 1) 2 s + 1, 2m m

m s,

j m

+ s + 1,

j m

(m N \ {1, 2}).

It is the series identity (3.1) which was rst applied by Zhang and Williams [33] (and, subsequently, by Cvijovi and Klinowski [5]) in order to prove two c (only seemingly dierent) versions of the series representation (2.35). Indeed, by appealing to (3.2) with m = 4, we can derive the following much more rapidly convergent series representation for (2n + 1) (see Srivastava [23, p. 9,

582 Equation (41)]): (2n + 1) = (1)n (3.3)

H. M. Srivastava

2(2)2n 4n1 1 B2n log 2 n(24n+1 + 22n 1) (2n)! 1 2n, 1 4 (n N),

22n1 1 42n1 (1 2n) 2(2n 1)! (2n 1)!

+
k=1

(1)k1 k (2k + 1) + (2n 2k)! 1 2k

k=0

(2k)! (2k) (2n + 2k)! 42k

where (and in what follows) a prime denotes the derivative of (s) or (s, a) with respect to s. In view of the identities (2.34) and (2.36), the results (2.6) and (3.3) would lead us eventually to the following additional series representations for (2n+1) (see Srivastava [23, p. 10, Equations (42) and (43)]): (2n + 1) = (1)n1 + (3.4) 2
2n

H2n+1 log 1 2 (2n + 1)!

2(4n 1) 22n+1 1 B2n+2 log 2 (2n 1) (2n + 2)! (2n + 1)! 24n+3 (2n + 1)!

2n 1,

1 4

n1

+
k=1

(1)k (2k + 1) (2n 2k + 1)! 1 2k

+2
k=1

(2k 1)! (2k) (2n + 2k + 1)! 42k

(n N);

(2n + 1) = (1)n + (3.5) +

4(2)2n 22n+1 1 (2n 1) n 42n+1 22n + 1 2 (2n)! 2n 1, 1 4 (2n + 1)(4n 1) B2n+2 log 2 (2n + 2)!

42n+1 (2n)!
n1

(1)k1 k (2k + 1) (2n 2k + 1)! 1 2k

+
k=0

(2k) (2k)! (2n + 2k + 1)! 42k

(n N).

1 Explicit expressions for the derivatives (2n 1) and 2n 1, 4 , occurring in the series representations (3.3), (3.4), and (3.5), can be found and substituted into these results in order to represent (2n + 1) in terms of Bernoulli numbers and polynomials and various rapidly convergent series of -functions (see, for details, Srivastava [23, Section 3]).

Series Representations for the Zeta Functions

583

Of the four seemingly analogous results (2.6), (3.3), (3.4), and (3.5), the innite series in (3.4) would obviously converge most rapidly, with its general term having the order estimate: O k 2n2 42k (k ; n N).

We now turn to the work of Srivastava and Tsumura [27], who derived the following three new members of the class of the series representations (2.6) and (3.4): (3.6) (2n + 1) = (1)n1 2 3
2n

H2n+1 log 2 (32n+2 1) 3 + B2n+2 (2n + 1)! 2 3(2n + 2)!

(1)n1 1 + 2n + 2, 2n+1 3 3(2)

+
k=1

(1)k (2k + 1) (2n 2k + 1)! 2 2k

+2
k=1

(2k 1)! (2k) (2n + 2k + 1)! 32k 2

(n N),

(2n + 1) = (1)n1

H2n+1 log( 1 ) 22n (22n+2 1) 2 + B2n+2 (2n + 1)! (2n + 2)!

(3.7)

(1)n1 1 + 2n + 2, 2n+1 2(2) 4

+
k=1

(1)k (2k + 1) (2n 2k + 1)! ( 1 )2k 2

+2
k=1

(2k 1)! (2k) (2n + 2k + 1)! 42k

(n N),

and (2n + 1) = (1)n1 3

H2n+1 log( 1 ) 3 (2n + 1)!

(3.8)

22n (32n+2 1) (1)n1 + B2n+2 + 3(2n + 2)! 2 3(2)2n+1 1 1 + 2n + 2, 2n + 2, 3 6

+
k=1

(1)k (2k + 1) (2n 2k + 1)! ( 1 )2k 3 (2k 1)! (2k) (2n + 2k + 1)! 62k (n N).

+2
k=1

584

H. M. Srivastava

Indeed the general terms of the innite series occurring in these three members [(3.6), (3.7), and (3.8)] have the order estimates: (3.9) O k 2n2 m2k (k ; n N; m = 3, 4, 6),

which exhibit the fact that each of the three series representations (3.6), (3.7), and (3.8) converges more rapidly than Wiltons result (2.33) and two of them (cf. Equations (3.7) and (3.8)) at least as rapidly as Srivastavas result (3.4). 4. FURTHER SERIES REPRESENTATIONS In their aforecited work on the Ray-Singer torsion and topological eld theories, Nash and OConnor ([16] and [17]) obtained a number of remarkable integral expressions for (3), including (for example) the following result [17, p. 1489 et seq.]: (4.1) (3) = 2 2 8 log 2 7 7
/2 0

z 2 cot z dz.

Since [7, p. 51, Equation 1.20(3)]

(4.2)

z cot z = 2
k=0

(2k)

2k

(|z| < ),

the result (4.1) is obviously equivalent to the series representation (cf. Dbrowski a [6, p. 202]; see also Chen and Srivastava [4, p. 191, Equation (3.19)]): (4.3) (3) = 2 2 7

log 2 +
k=0

(2k) (k + 1)22k

Moreover, by integrating by parts, it is easily seen that

(4.4)
0

z 2 cot z dz = 2
0

/2

z log sin z dz,

so that the result (4.1) is equivalent also to the integral representation: (4.5) (3) = 2 2 16 log 2 + 7 7
/2

z log sin z dz,

which was proven in the aforementioned 1772 paper by Euler (cf., e.g., Ayoub [2, p. 1084]).

Series Representations for the Zeta Functions

585

Next, since (4.6) i cot iz = coth z = 2 +1 e2z 1 (i := 1),

by replacing z in the known expansion (4.2) by 1 iz, it is easily seen that (cf., 2 e.g., Koblitz [14, p. 25]; see also Erdlyi et al. [7, p. 51, Equation 1.20(1)]) e (4.7) z z + = z 1 e 2
k=0

(1)k+1 (2k) 2k z 22k1

(|z| < 2).

By setting z = it in (4.7), multiplying both sides by tm1 (m N), and then integrating the resulting equation from t = 0 to t = (0 < < 2), Srivastava [25] derived the following series representations for (2n + 1) (see also Srivastava et al. [26]): (2n + 1) = (1)n1 (4.8) (2)2n (2n)!(22n+1 1)
n1

log 2 +
j=1

(1)j

2n (2j)!(22j 1) (2j + 1) 2j (2)2j (n N)

+
k=0

(2k) (k + n)22k

and (2n + 1) = (1)n1 (4.9) (2)2n (2n + 1)!(22n 1)

log 2 +
j=1

(1)j

2n + 1 (2j)!(22j 1) (2j + 1) 2j (2)2j (n N).

+
k=0

(2k) k + n + 1 22k 2

For n = 1, (4.9) immediately reduces to the following series representation for (3): (4.10) (3) = 2 2 9

log 2 + 2
k=0

(2k) (2k + 3)22k

which was proven independently by (among others) Glasser [10, p. 446, Equation (12)], Zhang and Williams [33, p. 1585, Equation (2.13)], and Dbrowski a

586

H. M. Srivastava

[6, p. 206] (see also Chen and Srivastava [4, p. 183, Equation (2.15)]). And a special case of (4.8) when n = 1 yields (cf. Dbrowski [6, p. 202]; see also a Chen and Srivastava [4, p. 191, Equation (3.19)]) (4.11) 2 2 (3) = 7

log 2 +
k=0

(2k) (k + 1)22k

In view of the familiar sum: (4.12)

1 (2k) = log 2, (2k + 1)22k 2

Eulers formula (1.11) is indeed a simple consequence of (4.11). We remark in passing that an integral representation for (2n + 1), which is easily seen to be equivalent to the series representation (4.8), was given by Dbrowski [6, p. 203, Equation (16)], who [6, p. 206] mentioned the existence a of (but did not fully state) the series representation (4.9) as well. The series representation (4.8) is derived also in a forthcoming paper by Borwein et al. (cf. [3, Equation (57)]). By suitably combining the series occurring in (4.3), (4.10), and (4.12), it is not dicult to derive several other series representations for (3), which are analogous to Eulers formula (1.11). More generally, since k 2 + k + (2k + 2n 1)(2k + 2n)(2k + 2n + 1) (4.13) A B C = + + , 2k + 2n 1 2k + 2n 2k + 2n + 1 where, for convenience, (4.14) (4.15) and (4.16) C = Cn (, , ) := 1 1 n2 + ( )n + ( 2 + 4) , 2 4 A = An (, , ) := 1 1 n2 ( + )n + ( + 2 + 4) , 2 4 B = Bn (, , ) := (n2 n + ),

by applying (4.8), (4.9), and another result (proven by Srivastava [25, p. 341, Equation (3.17)]):
n

(1)j1 (4.17)
j=1

2n + 1 (2j)!(22j 1) (2j + 1) 2j (2)2j (2k) k + n + 1 22k 2 (n N0 ),

= log 2 +
k=0

Series Representations for the Zeta Functions

587

with n replaced by n 1, Srivastava [25] derived the following unication of a large number of known (or new) series representations for (2n + 1) (n N), including (for example) Eulers formula (1.11):

(2n + 1) =

(1)n1 (2)2n (2n)!{(22n+1 1)B + (2n + 1)(22n 1)C} 1 log 2 + 4

(1)j
j=1

2n 1 2j 2 1 2

(4.18)

2j(2j 1)A + [(4n 1) 2]nj + n n + (2j 2)!(22j 1) (2j + 1) (2)2j (k 2 + k + )(2k) + (2k + 2n 1)(k + n)(2k + 2n + 1)22k
k=0

(n N; , , C),

where A, B, and C are given by (4.14), (4.15), and (4.16), respectively. Numerous other interesting series representations for (2n + 1), which are analogous to (4.8) and (4.9), were also given by Srivastava et al. [26]. For the sake of completeness, we choose to recall their results as follows:

(4.19) (2n + 1) = (1)n1 (2)2n log 3 + 4 (2n + 1)!(32n 1)

(2k) (2k + 2n + 1)32k (2j + 1)

+(2n + 1)! (2n + 1)! 3

(1)j (2n 2j + 1)! (1)j (2n 2j + 2)! 1)(2j)

32j

1 (2)2j

2(2j,

j=1 (32j

(2)2j1

(n N),

588

H. M. Srivastava

(2n + 1) = (1)n1 +(2n)! (4.20) (2n)! 3

(2)2n log 3 + 2 (2n)!(32n+1 1) (1)j (2n 2j)! 32j 1 (2)2j

k=0

(2k) (k + n)32k

n1 j=1 n j=1

(2j + 1)

(1)j (2n 2j + 1)!

2(2j, 1 ) (32j 1)(2j) 3 (n N), (2)2j1 (2n + 1) = (1)n1 (2)2n (2n + 1)!(22n 1)

log 2 + 4
k=0 n1

(2k) (2k + 2n + 1)42k (1)j (2n 2j + 1)! (1)j (2n 2j + 2)! 1)(2j) (n N), 22j 1 (2)2j (2j + 1)

(4.21)

+(2n + 1)!
j=1 n+1

(2n + 1)! (2j,

j=1 22j1 (22j

(2)2j1

(2n + 1) = (1)n1

(2)2n (2n)!(24n+1 + 22n 1)

log 2 + 2
k=0 n1

(2k) (k + n)42k 22j 1 (2)2j (2j + 1)

(4.22)

+(2n)!
j=1 n

(1)j (2n 2j)!

(2n)!
j=1

(1)j (2n 2j + 1)!

(2j, 1 ) 22j1 (22j 1)(2j) 4 (n N), (2)2j1

Series Representations for the Zeta Functions

589

(2n + 1) = (1)n1

(2)2n (22n 1)(32n 1)

4 (2n + 1)!

(2k) (2k + 2n + 1)62k (22j 1)(32j 1) (2)2j (2j + 1)

n1

(4.23)

+
j=1

(1)j (2n 2j + 1)!

1 + 3 and

(1)j (2n 2j + 2)!

(2j, 1 ) + (2j, 1 ) 22j1 (32j 1)(2j) 3 6 (n N), (2)2j1

(2n + 1) = (1)n1
n1

(2)2n 2 2n + 32n + 62n 1 (2n)! 2

k=0

(2k) (k + n)62k

(4.24)
j=1

(1)j (2n 2j)!

(22j 1)(32j 1) (2)2j

(2j + 1)

1 3

(1)j (2n 2j + 1)! (n N).

(2j, 1 ) + (2j, 1 ) 22j1 (32j 1)(2j) 3 6 (2)2j1

It is not dicult to derive further series representations for (2n + 1) (n N) by appropriately combining two or more of the results (4.8), (4.9), (4.17), and (4.19) to (4.24). Thus we can arrive at several general results analogous (for example) to (4.18). 5. SOME INTERESTING DEDUCTIONS For = 0, the series representation (4.18) simplies to the form:

590

H. M. Srivastava

(2n + 1) = (1)n1 (2)2n 1 1 (2n)!{(22n+1 1)(n ) (22n 1)(n + 2 )[(n + 2 ) ]} n1 2n 1 1 2nj (1)j j(2j 1) n 2 2j 2
j=1

(5.1)

(2j 2)!(22j 1) (2j + 1) (2)2j

+
k=0

(k + )(2k) (2k + 2n 1)(k + n)(2k + 2n + 1)22k

(n N; , C).

Furthermore, by setting = = 0 and = 1 in (4.18) or (alternatively) by setting =0 and = 1

in (5.1), we immediately obtain the series representation: (2n + 1) = (1)n1 (2)2n (2n)!{22n (2n 3) 2n + 1} n1 2n 1 (2j)!(22j 1) (1)j (2j + 1) (2)2j 2j 2
j=1

(5.2)

+2
k=0

(2k) (2k + 2n 1)(k + n)(2k + 2n + 1)22k

(n N),

which, in the special case when n = 1, was given by Chen and Srivastava [4, p. 189, Equation (2.45)]. Of the three representations (4.18), (5.1), and (5.2) for (2n + 1) (n N), the innite series in (5.2) converges most rapidly. For various other suitable special values of the parameters , , and , we can easily deduce from (4.18) and (5.1) several known (or new) series representations for (2n + 1) (n N). For example, if we set =2 and = 2n + 1

Series Representations for the Zeta Functions

591

in the series representation (5.1), we shall obtain (2n + 1) (5.3) = (1)n1 (2)2n (2n)!(22n+1 1) 1)
j=1 n1

(1)j

2n 1 2j 1

(2j)!(22j

(2)2j

(2j + 1) (n N),

k=0

(2k) (2k + 2n 1)(k + n)22k

which, in the special case when n = 1, immediately yields Eulers formula (1.11). The following additional series representations for (2n+1) (n N), which are analogous to (5.3), can also be deduced similarly from (5.1): (2n + 1) = (1)n1 (5.4) 2n
j=1

(2)2n (2n)!{(2n 1)22n 2n} (1)j 2n 1 2j 2 (2j)!(22j 1) (2j + 1) (2)2j (n N)

n1

k=0

(2k) (k + n)(2k + 2n + 1)22k

and (2n + 1) = (1)n1 (5.5)

(2)2n (2n + 1)!(22n 1) 4nj 2j + 1 2j 1 2n 1 2j 2

(1)j

(2j)!(22j 1) (2j + 1) (2)2j

4
k=0

(2k) (2k + 2n 1)(2k + 2n + 1)22k

(n N).

The special case of each of the last two series representations (5.4) and (5.5) when n = 1 was given by Zhang and Williams [33, p. 1586]. Next, with a view to further improving the rate of convergence in the

592

H. M. Srivastava

reasonably rapidly convergent series representation (5.2), we observe that 1 (2k + 2n 1)(2k + 2n)(2k + 2n + 1)(2k + 2n + 2) = 1 6 1 1 2k + 2n 1 2k + 2n + 2 1 1 . 2 (2k + 2n)(2k + 2n + 1)

(5.6)

Thus, by applying the series representations (4.17) with n replaced by n 1, (4.8) with n replaced by n + 1, and (5.4), we obtain (2n + 3) = 2 2 {22n+2 + n(2n 3)(22n 1) 1} (2n + 1) (n + 1)(2n + 1)(22n+3 1) +(1)n1 (5.7)
j=1 n1

(2)2n+2 (2n + 2)!(22n+3 1) 2n 1 2n + 2 2j 2j (2j)!(22j 1) (2j + 1) (2)2j

(1)j 2n 1 2j 2

+6n +12

k=0

(2k) (2k + 2n 1)(2k + 2n)(2k + 2n + 1)(2k + 2n + 2)22k

(n N), where the series converges faster than that in (5.2). In its special case when n = 1, (5.7) readily yields the following improved version of the series representation derivable from (5.2) for n = 2 (cf. [33, p. 1590, Equation (3.14)]): (5.8) 8 4 4 2 (3) + (5) = 31 31
k=0

(2k) , (2k + 1)(2k + 2)(2k + 3)(2k + 4)22k

in which (3) can be replaced by its known value 4 2 (2) given by (2.9) for n = 1. Yet another rapidly convergent series representation for (2n + 3) (n N), analogous to (5.7), can be derived by means of the identity: 1 (2k + 2n)(2k + 2n + 1)(2k + 2n + 2)(2k + 2n + 3) = 1 6 1 1 2k + 2n 2k + 2n + 3 1 1 , 2 (2k + 2n + 1)(2k + 2n + 2)

(5.9)

Series Representations for the Zeta Functions

593

together with our series representations (4.8), (4.9) with n replaced by n + 1, and (5.3) with n replaced by n + 1. We thus obtain the series representation: (2n + 3) =
1 2 2 { 3 (2n + 1)(2n2 4n + 3)(22n 1) 22n+1 + 1} (2n + 1) (n + 1)(2n + 1){(2n 3)22n+2 2n} n1 2n (2)2n+2 (1)j +(1)n1 2n+2 2n} (2n + 2)!{(2n 3)2 2j j=1

(5.10)

2n + 3 2n + 1 +3 2j 2j 1

(2j)!(22j

1)

(2)2j

(2j + 1)

+12
k=0

(2k) (2k + 2n)(2k + 2n + 1)(2k + 2n + 2)(2k + 2n + 3)22k

(n N), which, in the special case when n = 1, yields (5.11) (5) = 2 2 4 4 (3) 27 9
k=0

(2k) , (2k + 2)(2k + 3)(2k + 4)(2k + 5)22k

where the series obviously converges faster than that derivable from (5.2) for n = 2. Lastly, by applying the identity: 1 2k(2k + 2n 1)(2k + 2n)(2k + 2n + 1) 1 1 1 1 = 2n(2n 1)(2n + 1) 2k 2(2n 1) 2k + 2n 1 1 1 1 1 + 2n 2k + 2n 2(2n + 1) 2k + 2n + 1

(5.12)

in conjunction with the series representations (4.17) with n replaced by n 1, (4.8), (4.9), and the known result (cf., e.g., [12, p. 356, Entry (54.5.3)]):

(5.13)
k=1

(2k) 2k t = log[t csc(t)] k

with t =

1 2,

we arrive at the following series representation for (2n + 1)

594 (n N) : (2n + 1) = (1)n1 (5.14)

H. M. Srivastava

(2)2n 2 (2n 1)!{n 1 (n 2)22n }

12n2 1 log 2 (4n2 1)2 2n n(4n2 1)

(1)j

2n 2 (2j 1)!(22j 1) (2j + 1) 2j 3 (2)2j (n N),

+
k=1

(2k) k(k + n)(2k + 2n 1)(2k + 2n + 1)22k

1 where we have also applied the fact that (0) = 2 . For n = 1, (5.14) reduces immediately to Wiltons formula (cf. Wilton [31, p. 92] and Hansen [12, p. 357, Entry (54.5.9)]; see also Chen and Srivastava [4, p. 181, Equation (2.1)]):

(5.15)

(3) =

2 2

11 1 log + 18 3

k=1

(2k) k(k + 1)(2k + 1)(2k + 3)22k

Furthermore, in its special case when n = 2, (5.14) would yield the following interesting companion of the series representations (5.8) and (5.11): (5.16) 2 4 (5) = 45 47 log 30 60
k=1

(2k) k(k + 2)(2k + 3)(2k + 5)22k

which does not contain a term involving (3) on the right-hand side. By eliminating (2n+3) between the results (5.7) and (5.10), we can obtain a series representation for (2n + 1) (n N), which would converge as rapidly as the series in (5.14). We thus nd that (cf. Srivastava [25, pp. 548-549, Equation (3.47)])

Series Representations for the Zeta Functions

595

(5.17) (2n + 1)

n1 (2)2n = (1)n1 (1)j (2n)!n

{(2n 3)22n+2 2n} +6n +3 +12

2n 1 2n + 2 2j 2j 2n 2n + 3 2j 2j

2n 1 2j 2 2n + 1 2j 1

(22n+3 1)

(2j)!(22j 1) (2j + 1) (2)2j (n k + n ) (2k) (2k + 2n 1)(2k + 2n)(2k + 2n + 1)(2k + 2n + 2)(2k + 2n + 3)22k

(n N), where, for convenience, n := (22n+3 1) 1 (2n + 1)(2n2 4n + 3)(22n 1) 22n+1 + 1 3

(5.18)

{(2n 3)22n+2 2n}{22n+2 + n(2n 3)(22n 1) 1}, n := 2 (2n 5)22n+2 2n + 1 ,

(5.19) and (5.20)

n := (4n2 4n 7)22n+2 (2n + 1)2 .

In its special case when n = 1, (5.17) yields the following (rather curious) series representation: (5.21) 6 2 (3) = 23
k=0

(98k + 121)(2k) , (2k + 1)(2k + 2)(2k + 3)(2k + 4)(2k + 5)22k

where the series obviously converges much more rapidly than that in each of the celebrated results (1.11) and (1.12).

596

H. M. Srivastava

6. SYMBOLIC AND NUMERICAL COMPUTATIONS In this concluding section, we choose to summarize below the results of our symbolic and numerical computations with the series in (5.21) using Mathematica (Version 4.0) for Linux: In[1] : = (98k + 121)Zeta [2k]/ (2k + 1)(2k + 2)(2k + 3) (2k + 4)(2k + 5)2q (2k) Out[1] = (121 + 98k) Zeta [2k] 22k (1 + 2k)(2 + 2k)(3 + 2k)(4 + 2k)(5 + 2k)

In[2] : = Sum[%, {k, 1, Innity}] // Simplify Out[2] = 121 23 Zeta[3] 240 6Pi2 In[3] : = N[%]

Out[3] = 0.0372903 In[4] : = Sum [N[%1] // Evaluate, {k, 1, 50}] Out[4] = 0.0372903 In[5] : = N Sum [%1 // Evaluate, {k, 1, Innity}] Out[5] = 0.0372903 Since (0) = 1 , Out[2] evidently validates the series representation (5.21) 2 symbolically. Furthermore, our numerical computations in Out[3], Out[4], and Out[5], together, exhibit the fact that only 50 terms (k = 1 to k = 50) of the series in (5.21) can produce an accuracy of seven decimal places. ACKNOWLEDGMENTS The present investigation was supported, in part, by the Natural Sciences and Engineering Research Council of Canada under Grant OGP0007353. Indeed it is a great pleasure for me to express my sincere thanks to Professor Pei Yuan Wu (Editor-in-Chief, Taiwanese Journal of Mathematics) for his kind invitation to publish this survey-cum-expository work in his Journal. REFERENCES
1. R. Apry, Irrationalit de (2) et (3), in: Journes Arithmtiques de Luminy e e e e (Colloq. Internat. CNRS, Centre Univ. Luminy, Luminy, 1978), Astrisque e 61, Soc. Math. France, Paris, 1979, pp. 11-13.

Series Representations for the Zeta Functions

597

2. R. Ayoub, Euler and the Zeta function, Amer. Math. Monthly 81 (1974), 1067-1086. 3. J. M. Borwein, D. M. Bradley and R. E. Crandall, Computational strategies for the Riemann Zeta function, J. Comput. Appl. Math., to appear. 4. M.-P. Chen and H. M. Srivastava, Some families of series representations for the Riemann (3), Results Math. 33 (1998), 179-197. 5. D. Cvijovi and J. Klinowski, New rapidly convergent series representations for c (2n + 1), Proc. Amer. Math. Soc. 125 (1997), 1263-1271. 6. A. Dbrowski, A note on the values of the Riemann Zeta function at positive a odd integers, Nieuw Arch. Wisk. (4) 14 (1996), 199-207. 7. A. Erdlyi, W. Magnus, F. Oberhettinger and F. G. Tricomi, Higher Transcene dental Functions, Vol. I, McGraw-Hill, New York, 1953. 8. J. A. Ewell, A new series representation for (3), Amer. Math. Monthly 97 (1990), 219-220. 9. J. A. Ewell, On the Zeta function values (2k +1), k = 1, 2, ..., Rocky Mountain J. Math. 25 (1995), 1003-1012. 10. M. L. Glasser, Some integrals of the arctangent function, Math. Comput. 22 (1968), 445-447. 11. R. W. Gosper, Jr., A calculus of series rearrangements, in: Algorithms and Complexity: New Directions and Recent Results, J. F. Traub, ed., Academic Press, New York, 1976, pp. 121-151. 12. E. R. Hansen, A Table of Series and Products, Prentice-Hall, Englewood Clis, New Jersey, 1975. 13. M. M. Hjortnaes, Overfring av rekken k=1 (1/k 3 ) til et bestemt integral, in: Proceedings of the Twelfth Scandinavian Mathematical Congress (Lund; August 10-15, 1953), Scandinavian Mathematical Society, Lund, 1954, pp. 211-213. 14. N. Koblitz, p-Adic Numbers, p-Adic Analysis, and Zeta-Functions, Graduate Texts in Mathematics, Vol. 58, Springer-Verlag, New York, 1977. 15. W. Magnus, F. Oberhettinger and R. P. Soni, Formulas and Theorems for the Special Functions of Mathematical Physics, 3rd enlarged ed., Die Grundlehren der Mathematischen Wissenschaften in Einzeldarstellungen mit besonderer Bercksichtingung u der Anwendungsgebiete, Bd. 52, Springer-Verlag, New York, 1966. 16. C. Nash and D. OConnor, Ray-Singer torsion, topological eld theories and the Riemann Zeta function at s = 3, in: Low-Dimensional Topology and Quantum Field Theory (Proceedings of a NATO Advanced Research Workshop held at the Isaac Newton Institute at Cambridge, U.K.; September 6-12, 1992), H. Osborn, ed., Plenum Press, New York and London, 1993, pp. 279-288. 17. C. Nash and D. OConnor, Determinants of Laplacians, the Ray-Singer torsion on lens spaces and the Riemann Zeta function, J. Math. Phys. 36 (1995), 1462-1505.

598

H. M. Srivastava

18. V. Ramaswami, Notes on Riemanns -function, J. London Math. Soc. 9 (1934), 165-169. 19. J. D. Shallit and K. Zikan, A theorem of Goldbach, Amer. Math. Monthly 93 (1986), 402-403. 20. H. M. Srivastava, A unied presentation of certain classes of series of the Riemann Zeta function, Riv. Mat. Univ. Parma (4) 14 (1988), 1-23. 21. H. M. Srivastava, Sums of certain series of the Riemann Zeta function, J. Math. Anal. Appl. 134 (1988), 129-140. 22. H. M. Srivastava, Certain families of rapidly convergent series representations for (2n + 1), Math. Sci. Res. Hot-Line 1 (6) (1997), 1-6. 23. H. M. Srivastava, Further series representations for (2n + 1), Appl. Math. Comput. 97 (1998), 1-15. 24. H. M. Srivastava, Some rapidly converging series for (2n + 1), Proc. Amer. Math. Soc. 127 (1999), 385-396. 25. H. M. Srivastava, Some simple algorithms for the evaluations and representations of the Riemann Zeta function at positive integer arguments, J. Math. Anal. Appl. 246 (2000), 331-351. 26. H. M. Srivastava, M. L. Glasser and V. S. Adamchik, Some denite integrals associated with the Riemann Zeta function, Z. Anal. Anwendungen 19 (2000), 831-846. 27. H. M. Srivastava and H. Tsumura, A certain class of rapidly convergent series representations for (2n + 1), J. Comput. Appl. Math. 118 (2000), 323-335. 28. E. C. Titchmarsh, The Theory of the Riemann Zeta-Function, Oxford University (Clarendon) Press, Oxford and London, 1951; 2nd ed. (revised by D.R. Heath-Brown), 1986. 29. F. G. Tricomi, Sulla somma delle inverse delle terze e quinte potenze dei numeri naturali, Atti Accad. Naz. Lincei Rend. Cl. Sci. Fis. Mat. Natur. (8) 47 (1969), 16-18. 30. H. Tsumura, On evaluation of the Dirichlet series at positive integers by qcalculation, J. Number Theory 48 (1994), 383-391. 31. J. R. Wilton, A proof of Burnsides formula for log (x + 1) and certain allied properties of Riemanns -function, Messenger Math. 52 (1922-1923), 90-93. 32. E. Witten, On quantum gauge theories in two dimensions, Phys. 141 (1991), 153-209. Comm. Math.

33. N.-Y. Zhang and K. S. Williams, Some series representations of (2n+1), Rocky Mountain J. Math. 23 (1993), 1581-1592.

Series Representations for the Zeta Functions

599

Department of Mathematics and Statistics University of Victoria Victoria, British Columbia V8W 3P4 Canada E-mail: [email protected]