Higher-order recurrences for Bernoulli numbers (Q1025890): Difference between revisions

A classical formula by Euler for Bernoulli numbers \(B_n\) reads, in symbolic notation, as \((B_0 + B_0)^n = -nB_{n-1} - (n-1)B_n\) \(\,(n \geq 1)\). The authors extend this to \((B_{k_1} + \cdots + B_{k_m})^n\) for \(m \geq 2,\,n \geq m-1\), where \(k_1,\dots, k_m\) are arbitrary nonnegative integers. They in fact show that this expression equals the sum of the numbers \(C_\nu^K(n)B_{n+\nu}\), where \(\nu\) runs from \(-m+1\) to \(s_m = k_1 + \cdots + k_m\) and where \(C_\nu^K(x)\) are polynomials of degree \(\leq m-1\) having rational coefficients and depending only on \(K = (k_1,\dots,k_m)\). These polynomials are recursively computable. The authors describe the computation of \(C_\nu^K(n)\) first in general and then more detailed for \(m=3\). Explicit formulas are written down in the case \(k_1 = k_2 = k_3\). Some further information about \(C_{-m+1}^K(n)\) and \(C_{s_m}^K(n)\) is also given. This work continues the authors' previous work [J. Number Theory 124, No. 1, 105--122 (2007; Zbl 1137.11014)] concerning the case \(m=2\).