Alternative Proofs of a Formula for Bernoulli Numbers in Terms of Stirling Numbers

DOI 10.1515/anly-2014-0003 Ë Analysis 2014; 34 (3):311–317

Erratum

Feng Qi and Bai-Ni Guo Alternative proofs of a formula for Bernoulli numbers in terms of Stirling numbers

This is a corrected version of [Analysis 34 (2014), 187–193]

Abstract: In the paper, the authors provide four alternative proofs of an explicit formula for computing Bernoulli numbers in terms of Stirling numbers of the second kind.

Keywords: Alternative proof, explicit formula, Bernoulli number, Stirling number of the second kind, Faá di Bruno formula, Bell polynomial

MSC 2010: Primary: 11B68, secondary: 11B73

ËË Feng Qi, Bai-Ni Guo: School of Mathematics and Informatics, Henan Polytechnic University, Jiaozuo City, Henan Province, 454010, P.R. China, e-mail: [email protected], [email protected]

1 Introduction

It is well known that Bernoulli numbers B for k ≥ 0 may be generated by k x ∞ xk x ∞ x2k = H B = 1 − + H B , |x| < 2ð. (1.1) ex − 1 k k! 2 2k (2k)! k=0 k=1 In combinatorics, Stirling numbers of the second kind S(n, k) for n ≥ k ≥ 0 may be computed by 1 k k S(n, k) = H(−1)k−~ ~n k! ~ ~=0 and may be generated by

(ex − 1)k ∞ xn = H S(n, k) , k ∈ {0} ∪ ℕ. k! n! n=k In [5, p. 536] and [6, p. 560], the following simple formula for computing Bernoulli numbers B in terms of Stirling numbers of the second kind S(n, k) was n incidentally obtained. 312 Ë F. Qi and B.-N. Guo

Theorem 1.1. For n ∈ {0} ∪ ℕ, we have

n k! B = H(−1)k S(n, k). (1.2) n k + 1 k=0 The aim of this paper is to provide four alternative proofs for the explicit formula (1.2).

2 Four alternative proofs of the formula (1.2)

Considering S(0, 0) = 1, it is clear that the formula (1.2) is valid for n = 0. Further considering S(n, 0) = 0 for n ≥ 1, it is sucient to show

n k! B = H(−1)k S(n, k), n ≥ 1. n k + 1 k=1 First proof. It is listed in [1, p. 230, 5.1.32] that

∞ b e−au − e−bu ln = X du. (2.1) a u 0 Taking a = 1 and b = 1 + x in (2.1) yields

∞ ∞ 1 ln(1 + x) 1 − e−xu = X e−u du = XX txu−1 dte−u du. (2.2) x xu 0 0 1/e Replacing x by ex − 1 in (2.2) results in

∞ 1 x = XX tue −u−1 dte−u du. (2.3) ex − 1 x 0 1/e In combinatorics, Bell polynomials of the second kind (also called partial Bell polynomials) B (x , x , . . . , x ) are dened by n,k 1 2 n−k+1

n! n−k+1 x ~ B (x , x , . . . , x ) = H I i n,k 1 2 n−k+1 ∏n−k+1 ~ ! i! i 1≤i≤n,~ ∈ℕ i=1 i i=1 ∑ i~ =n i ∑n ~ =k i=1 i n i for n ≥ k ≥ 1, see [4, p. 134, Theorem A].i= They1 satisfy

B abx , ab2x , . . . , abn−k+1x = akbnB (x , x , . . . , x ) (2.4) n,k 1 2 n−k+1 n,k 1 n n−k+1 An explicit formula for Bernoulli numbers Ë 313 and n−k+1 B 1, 1, . . . , 1 = S(n, k), (2.5) n,k see [4, p. 135], where a and b are any complex numbers. The well-known Faà di Bruno formula may be described in terms of Bell polynomials of the second kind B (x , x , . . . , x ) by n,k 1 2 n−k+1 dn n f ∘ g(x) = H f(k)(g(x))B gὔ(x), gὔὔ(x), . . . , g(n−k+1)(x), (2.6) dxn n,k k=1 see [4, p. 139, Theorem C]. Applying in (2.6) the function f(y) = ty and y = g(x) = uex − u − 1 gives

dntue n n−k+1 x = H(ln t)ktue B uex, uex, . . . , uex . (2.7) dxn x n,k k=1 Making use of the formulas (2.4) and (2.5) in (2.7) reveals

dntue n x = tue H S(n, k)uk(ln t)kekx. (2.8) dxn x k=1 Dierentiating n times on both sides of (2.3) and considering (2.8), we obtain

∞ 1 dn x n = H S(n, k)ekx X ukX(ln t)ktue −u−1 dte−u du. (2.9) dxn ex − 1 x k=1 0 1/e On the other hand, dierentiating n times on both sides of (1.1) gives

dn x ∞ xk−n = H B . (2.10) dxn ex − 1 k (k − n)! k=n Equating (2.9) and (2.10) and taking the limit x → 0, we deduce

∞ 1 n (ln t)k B = H S(n, k) X ukX dte−u du n t k=1 0 1/e ∞ n (−1)k = H S(n, k) X uke−u du k + 1 k=1 0 n (−1)kk! = H S(n, k). k + 1 k=1 The rst proof of Theorem 1.1 is complete. 314 Ë F. Qi and B.-N. Guo

Second proof. In the book [2, p. 386] and in the papers [3, p. 615] and [11, p. 885], it was given that 1 ln b − ln a 1 = X dt, b − a (1 − t)a + tb 0 where a, b > 0 and a ̸= b. Replacing a by 1 and b by ex yields

1 x 1 = X dt. ex − 1 1 + (ex − 1)t 0 Applying the functions f(y) = 1 and y = g(x) = 1 + (ex − 1)t in the formula (2.6) y and simplifying by (2.4) and (2.5) give

1 dn x dn 1 = X dt dxn ex − 1 dxn 1 + (ex − 1)t 0 1 n k! n−k+1 = X H(−1)k B tex, tex, . . . , tex dt [1 + (ex − 1)t]k+1 n,k 0 k=1 1 n tk n−k+1 = H(−1)kk! X B ex, ex, . . . , ex dt [1 + (ex − 1)t]k+1 n,k k=1 0 1 n n−k+1 → H(−1)kk! X tkB 1, 1, . . . , 1 dt, x → 0 n,k k=1 0 1 n = H(−1)kk!S(n, k) X tk dt

k=1 0 n k! = H(−1)k S(n, k). k + 1 k=1 On the other hand, taking the limit x → 0 in (2.10) leads to

dn x ∞ xk−n = H B → B , x → 0. dxn ex − 1 k (k − n)! n k=n The second proof of Theorem 1.1 is thus complete. An explicit formula for Bernoulli numbers Ë 315

Third proof. Let CT[f(x)] be the coecient of x0 in the power series expansion of f(x). Then

n k! n n! (ex − 1)k H(−1)k S(n, k) = H(−1)kCT k + 1 xn k + 1 k=1 k=1 1 ∞ (ex − 1)k = n!CT H(−1)k xn k + 1 k=1 1 ln[1 + (ex − 1)] − (ex − 1) = n!CT xn ex − 1 1 x = n!CT xn ex − 1 = B . n Thus, the formula (1.2) follows.

Fourth proof. It is clear that the equation (1.1) may be rewritten as

ln[1 + (ex − 1)] ∞ xk = H B . (2.11) ex − 1 k k! k=0 Dierentiating n times on both sides of (2.11) and taking the limit x → 0 reveal ∞ xk−n B = lim H B n k x→0 (k − n)! k=n dn ln[1 + (ex − 1)] = lim x→0 dxn ex − 1 n ln(1 + u) (k) n−k+1 = lim H B ex, ex, . . . , ex , u = ex − 1 n,k x→0 u k=1 n ∞ u~−1 (k) n−k+1 = lim H H(−1)~−1 B ex, ex, . . . , ex n,k x→0 ~ k=1 ~=1 n ∞ (~ − 1)! n−k+1 = lim H H (−1)~−1 u~−k−1B ex, ex, . . . , ex n,k x→0 (~ − k − 1)!~ k=1 ~=k+1 n ∞ (~ − 1)! n−k+1 = H lim H (−1)~−1 u~−k−1 lim B ex, ex, . . . , ex n,k u→0 (~ − k − 1)!~ x→0 k=1 ~=k+1 n k! n−k+1 = H(−1)k B 1, 1, . . . , 1 k + 1 n,k k=1 n k! = H(−1)k S(n, k). k + 1 k=1 The fourth proof of Theorem 1.1 is thus complete. 316 Ë F. Qi and B.-N. Guo

Remark 2.1. In [9, p. 1128, Corollary], among other things, it was found that

1 1 k−1 A B = − − 2k H 2(k−i) 2k 2 2k + 1 2(k − i) + 1 i=1 for k ∈ ℕ, where A is dened by m

n k+1 H mk = H A nm. m m=1 m=0 It was listed in [6, p. 559] and recovered in [8, Theorem 2.1] that

1 (k) k+1 1 m = (−1)k H (m − 1)!S(k + 1, m) (2.12) ex − 1 ex − 1 m=1 for k ∈ {0} ∪ ℕ. In [8, Theorem 3.1], by the identity (2.12), it was obtained that

2k−1 S(2k + 1, m + 1)S(2k, 2k − m) B = 1 + H 2k 2k m=1 m 2k 2k S(2k, m)S(2k + 1, 2k − m + 1) − H , k ∈ ℕ. 2k + 1 2k m=1 m−1 In [12, Theorem 1.4], among other things, it was presented that

(−1)k−1k k−1 k−i−1 2k B = H H (−1)i+~ (k − i − ~)2k−1, k ∈ ℕ. 2k 22(k−1)(22k − 1) ~ i=0 ~=0 Remark 2.2. The identities in (2.12) have been generalized and applied in [7, 13].

Acknowledgement: We thank Professor Doron Zeilberger for drawing our atten- tion to the books [5, 6] and sketching the third proof in an e-mail on October 10, 2013. Thanks to his advice, we could nd that the formula (1.2) originated from [10] and was listed as an incidental consequence of an answer to an exercise in [5, p. 536] and [6, p. 560].

References

[1] M. Abramowitz and I. A. Stegun (eds.), Handbook of Mathematical Functions with Formu- las, Graphs, and Mathematical Tables, 9th printing with corrections, Applied Mathemat- ics Series 55, National Bureau of Standards, Washington, 1970. [2] P. S. Bullen, Handbook of Means and Their Inequalities, Kluwer Academic Publishers, Dordrecht, 2003. An explicit formula for Bernoulli numbers Ë 317

[3] B. Carlson, The logarithmic mean, Am. Math. Mon. 79 (1972), 615–618. [4] L. Comtet, Advanced Combinatorics: The Art of Finite and Innite Expansions, D. Reidel, Dordrecht, 1974. [5] R. L. Graham, D. E. Knuth and O. Patashnik, Concrete Mathematics. A Foundation for Computer Science, Addison-Wesley, Reading, 1989. [6] R. L. Graham, D. E. Knuth and O. Patashnik, Concrete Mathematics. A Foundation for Computer Science, 2nd ed., Addison-Wesley, Amsterdam, 1994. [7] B.-N. Guo and F. Qi, Explicit formulae for computing Euler polynomials in terms of Stirling numbers of the second kind, J. Comput. Appl. Math. 272 (2014), 251–257. [8] B.-N. Guo and F. Qi, Some identities and an explicit formula for Bernoulli and Stirling numbers, J. Comput. Appl. Math. 255 (2014), 568–579. n k Z. Anal. Anwend. 18 [9] S.-L. Guo and F. Qi, Recursion formulae for ∑m=1 m , (1999), no. 4, 1123–1130. [10] B. F. Logan, Polynomials related to the Stirling numbers, AT&T Bell Laboratories Internal Technical Memorandum, August 10, 1987. [11] E. Neuman, The weighted logarithmic mean, J. Math. Anal. Appl. 188 (1994), no. 3, 885–900. [12] F. Qi, Explicit formulas for derivatives of tangent and cotangent and for Bernoulli and other numbers, preprint (2012), http://arxiv.org/abs/1202.1205. [13] A.-M. Xu and Z.-D. Cen, Some identities involving exponential functions and Stirling numbers and applications, J. Comput. Appl. Math. 260 (2014), 201–207.

Received December 10, 2013; accepted January 14, 2014.