Kyushu J. Math. 65 (2011), 39–53 doi:10.2206/kyushujm.65.39
THE MODULAR RELATION AND THE DIGAMMA FUNCTION
K. CHAKRABORTY, S. KANEMITSU and X.-H. WANG (Received 5 March 2010)
Dedicated to Professor Ken’ichi Sato on his sixtieth birthday with compliments and friendship
Abstract. In this paper we shall locate a class of fundamental identities for the gamma function and trigonometric functions in the chart of functional equations for the zeta- functions as a manifestation of the underlying modular relation. We use the beta-transform but not the inverse Heaviside integral. Instead we appeal to the reciprocal relation for the Euler digamma function which gives rise to the partial fraction expansion for the cotangent function. Through this we may incorporate basic results from the theory of the digamma (and gamma) function, thereby dispensing also with the beta-transform. Section 4 could serve as a foundation of the theory of the gamma function through the digamma function.
1. Introduction and the digamma function
We recall that Euler’s solution of the Basel problem, the evaluation of the value of the infinite ∞ 2 series n=1(1/n ), depends on the product representation for the sine function (cf. (20) below). Since the value is ζ(2) = π2/6, there must be a relation between the sine function and the Riemann zeta-function, which has been established as a circle of equivalent identities in [13, Ch. 5]. In this paper we shall give another foundation of this circle of identities by making use of the Euler digamma function, thereby suggesting a new foundation of the theory of the gamma function through the digamma function. We use the following notation throughout this paper (cf. e.g. [5]): (a) s = σ + it is the complex variable; z is another complex variable, used interchangeably with s; N 1 (b) ψ(s) =− lim − log(N + s) N→∞ n + s n=0 is the digamma function (cf. (40)); s (c) log (s) = ψ(u) du 1 is the log-gamma function (in both the digamma and log-gamma functions, non- positive integer values of s are excluded);
2000 Mathematics Subject Classification: Primary 11M35; Secondary 11M06, 33B15. Keywords: Euler digamma function; gamma function; Hurwitz zeta-function.
c 2011 Faculty of Mathematics, Kyushu University 40 K. Chakraborty et al
(d) γ =−ψ(1) is the Euler constant (cf. (44)); ∞ 1 (e) ζ(s, a)= (n + a)s n=0 is the Hurwitz zeta-function, where σ>1, a ∈ C,anda = non-positive integer; ∞ 1 (f) ζ(s)= ζ(s, 1) = ,σ>1, ns n=1 is the Riemann zeta-function; (g) for x ≥ 0, a ∈ C, a = non-positive integer, and u ∈ C, u Lu(x, a) = (n + a) 0≤n≤x is the partial sum of the Hurwitz zeta-function ζ(−u, a); ¯ (h) Bk(t) = Bk(t −[t]) is the kth periodic Bernoulli polynomial; k k − (i) B (t) = B tk r k r r r=0
is the kth Bernoulli polynomial (in both of which Bk is the kth Bernoulli number, and [t] is the integral part of t); and m,n aj (j) Gp,q x bj is the Meijer G-function defined by (cf. [5, Ch. VI, p. 206] and [25]) m,n a1,a2,...,ap Gp,q x b1,b2,...,bq m − n − + 1 j= (bj s) i= (1 ai s) = 1 1 xs ds, (1) πi q − + p − 2 (γ ) j=m+1 (1 bj s) i=n+1 (ai s) where (γ ) is the vertical Bromwich path extending from γ − i∞ to γ + i∞ so that all poles of (bj − s), j = 1, 2,...,m, are to the right, and all poles of (1 − ai − s), i = 1, 2,...,n, are to the left of (γ ). The last separation of poles condition is essential for an integral to be the G-function. The notation is slightly different in main references [5], [19]and[25]. In [5]and[25], the notation is the same except for the sign of the abscissa of the integration path being opposite. In [13, Ch. 5] we have presented the theory of the gamma function through the Hurwitz zeta-function and located various fundamental identities. Here we shall appeal rather to the Euler digamma function, which is the Laurent constant of the Hurwitz zeta-function around s = 1, and locate the same class of identities, the highlight being the claim that the partial fraction expansion for the cotangent function is equivalent to the functional equation for the Modular relation and digamma function 41
Riemann zeta-function. The partial fraction expansion for the cotangent function ∞ ∞ 1 2z 1 1 1 1 1 cot πz= − = + − (2) πz π n2 − z2 πz π n + z n − z n=1 n=1 is usually proved as an application of the residue calculus (cf. e.g. [22, p. 113]) and regarded as independent of the theory of zeta-functions. However, our recent theory of general modular = relations (cf. [8, 9, 14]), which gives identities equivalent to the functional equation ϕ(s) − = ∞ s = ∞ s R(s)ψ(r s),whereϕ(s) n=1(αn/λn), ψ(s) n=1(βn/μn) and R(s) is a general gamma factor, reveals that it is a specification of the following general modular relation: ∞ ∞ αn 2,0 − βn 1,1 μn 1 − a G zλn = − G λs 0,2 s, a r s 1,1 z r − s n=1 n n=1 μn L s−w + Res((a − s + w)χ(w)z ,w= sk). (3) k=1 Indeed, the left-hand side of (3) reduces to the cotangent function in view of the inverse Heaviside integral − 2,0 2 = z+1/2 G0,2 x 1 2x Kz−1/2(2x) (4) z, 2 √ −z and the reduction formula K1/2 = (π/2z)e . On the other hand, the right-hand side of (3) reduces to the partial fraction expansion of the coth in view of the beta-transform (in previous literature it used to be called the Mellin– Barnes integral, but this refers to a much wider class of functions, cf. [19]) (z + s)(−s) + −z = 1 s = 1 1,1 −1 1 (1 x) x ds G1,1 x , (5) 2πi (c) (z) (z) z for x>0and−ξ