Faulhaber’s formula, Bernoulli numbers and the equation $f(x)+x^{k}=f(x+1)$

Abstract

In modern usage the Bernoulli numbers and polynomials follow Euler’s approach and are defined using generating functions. We consider the functional equation $f(x)+x^{k}=f(x+1)$ and show that a solution can be derived from Faulhaber’s formula for the sum of powers. We show how these solutions provide a characterization of Bernoulli numbers and related results.

1 Introduction

We consider the functional equation described by

$$f(x)+x^{k}=f(x+1)$$ (1)

A purpose of this brief note is to show that Faulhaber’s formula for the sums of powers (which initiated the study of Bernoulli numbers) can be used to derive a solution to this equation and leads to a characterization of Bernoulli numbers, Bernoulli polynomials and related identities.

2 Faulhaber’s formula

Faulhaber’s formula (also known as Bernoulli’s formula) equates the sum of powers of consecutive integers to a polynomial involving Bernoulli numbers. It was discovered by various people including Takanaga, Faulhaber, and Bernoulli. In particular, for $n$ and $p$ nonnegative integers,

$$F(n,p)=\sum_{i=1}^{n}i^{p}=\frac{1}{p+1}\sum_{j=0}^{p}\binom{p+1}{j}B^{+}_{j}n^{p+1-j}$$ (2)

where $B^{+}_{j}$ are the Bernoulli numbers of the second kind. There are in general two conventions when defining Bernoulli numbers, with the only difference being the sign of $B_{1}$. In particular, $B^{+}_{1}=\frac{1}{2}$, whereas the Bernoulli numbers of the first kind $B^{-}_{j}$ are identical to $B^{+}_{j}$ except for $B^{-}_{1}=-\frac{1}{2}$. $B^{-}_{j}$ has been used in modern textbooks although recently there are discussions on whether $B^{+}_{j}$, which was used in the past, should be favored. Computer algebra software such as Mathematica currently use $B^{-}_{j}$ to denote Bernoulli numbers. However, the bernoulli function in Sympy which until very recently also returns $B^{-}_{j}$ has changed from returning $B^{-}_{j}$ to $B^{+}_{j}$ starting from version 1.12. Note that the number of terms in the left hand side is $n$ while it is $p+1$ on the right hand side, meaning that the computation of $F(n,p)$ is more efficient using the right hand side when $n\gg p$.

The Bernoulli numbers are defined in modern usage via the generating functions [1]:

$$\sum_{j=0}^{\infty}\frac{B^{-}_{j}x^{j}}{j!}=\frac{x}{e^{x}-1}$$ (3)

$$\sum_{j=0}^{\infty}\frac{B^{+}_{j}x^{j}}{j!}=\frac{x}{1-e^{-x}}$$ (4)

They also satisfy the following recursive definitions [1]:

Lemma 1.

$$\sum_{k=0}^{m}\binom{m+1}{k}B^{-}_{k}=\delta_{m,0}$$ (5)

$$\sum_{k=0}^{m}\binom{m+1}{k}B^{+}_{k}=m+1$$ (6)

Where $\delta_{m,0}$ is the Kronecker delta function defined as $1$ if $m=0$ and $0$ otherwise.

2.1 A variant of Faulhaber’s formula

Eq. (2) expresses the sum of powers $F(n,p)$ using Bernoulli numbers of the second kind $B^{+}_{j}$. The following formula gives a variant of Faulhaber’s formula using Bernoulli numbers of the first kind $B^{-}_{j}$ [2].

Theorem 1.

If $p>0$, then

$$F(n,p)=\sum_{i=1}^{n}i^{p}=\frac{1}{p+1}\sum_{j=0}^{p}\binom{p+1}{j}B^{-}_{j}(n+1)^{p+1-j}$$ (7)

Proof.

By Eq. (2),

$$F(n+1,p)=F(n,p)+(n+1)^{p}=\frac{1}{p+1}\sum_{j=0}^{p}\binom{p+1}{j}B^{+}_{j}(n+1)^{p+1-j}$$

For $j=1$, the corresponding term is $\frac{1}{2}(n+1)^{p}$ whereas $\frac{1}{p+1}\binom{p+1}{1}B^{-}_{1}(n+1)^{p}=\frac{1}{2}(n+1)^{p}$. This means that $F(n+1,p)=(n+1)^{p}+\frac{1}{p+1}\sum_{j=0}^{p}\binom{p+1}{j}B^{-}_{j}(n+1)^{p+1-j}$ and the conclusion follows. ∎

Note that the formula in Eq. (7) is almost identical to the formula in Eq. (2) except for using $B^{-}_{j}$ rather than $B^{+}_{j}$ and shifting the base of the powers from $n$ to $n+1$.

3 A solution to the equation $f(x)+x^{k}=f(x+1)$

Define $f_{k}(x)=\frac{1}{k+1}\sum_{j=0}^{k}\binom{k+1}{j}B^{-}_{j}x^{k+1-j}$, then the above discussion shows that $f_{k}(x+1)=f_{k}(x)+x^{k}$. Note that $(k+1)f_{k}(x)=B_{k+1}(x)$, the $k+1$-th Bernoulli polynomial. We show that the converse is also true among polynomials in the following sense.

Suppose that $p$ is a minimal degree polynomial that satisfied Eq. (1). Clearly $p$ must have degree larger than or equal to $k-1$. If $p$ has degree $k$, then the coefficient of $x^{k}$ does not match in Eq. (1). Therefore, $p$ must be of degree $k+1$ or larger. Consider the case where $p$ has degree $k+1$ written as $p(x)=\sum_{j=0}^{k+1}a_{i}x^{j}$. Then Eq. (1) can be written as $\sum_{j=0}^{k+1}a_{j}x^{j}+x^{k}=\sum_{j=0}^{k+1}a_{j}(x+1)^{j}$. Matching the coefficients results in the following $k+2$ equations $a_{j}=\sum_{l=0}^{k+1}a_{l}\binom{l}{j}=\sum_{l=j}^{k+1}a_{l}\binom{l}{j}$ for $j\neq k$ and $a_{k}+1=\sum_{l=0}^{k+1}a_{l}\binom{l}{k}$. This last equation can be simplified as: $a_{k}+1=a_{k}+a_{k+1}(k+1)$, i.e. $a_{k+1}=\frac{1}{k+1}$. The equation for $j=k+1$ is the identity $a_{k+1}=a_{k+1}$. The remaining $k$ equations $a_{j}=\sum_{l=j}^{k+1}a_{l}\binom{l}{j}$ for $j<k$ uniquely determines the other coefficients $a_{j}$. In particular, $a_{j}=a_{j}+a_{j+1}\binom{j+1}{j}+\sum_{l=j+2}^{k+1}a_{l}\binom{l}{j}$, i.e. for $j\leq k$,

$$a_{j}=-\frac{\sum_{l=j+1}^{k+1}a_{l}\binom{l}{j-1}}{\binom{j}{j-1}}=-\frac{\sum_{l=j+1}^{k+1}a_{l}\binom{l}{j-1}}{j}$$

This implies that $a_{k}=-a_{k+1}\binom{k+1}{k-1}/k=-\frac{1}{2}$.

Let $b_{k+1-j}=(k+1)a_{j}/\binom{k+1}{j}$ or equivalently $b_{j}=(k+1)a_{k+1-j}/\binom{k+1}{j}$. This implies that $b_{0}=1$ and $b_{1}=-\frac{1}{2}$. After straightforward algebraic manipulations we get the following identity regarding $b_{j}$’s’: $b_{j}=-\frac{\sum_{w=0}^{j-1}b_{w}\binom{j+1}{w}}{j+1}$ for $0<j\leq k+1$ which is exactly the recursive definition of the Benoulli number $B^{-}_{j}$ in Lemma 1. Thus we have shown that

Proposition 1.

The first $k+2$ Bernoulli numbers $B^{-}_{j}$ are defined by the coefficients $a_{j}$ of the minimal degree polynomial that satisfies the functional equation $f(x)+x^{k}=f(x+1)$ via the equation $B^{-}_{j}=(k+1)a_{k+1-j}/\binom{k+1}{j}$.

Define $g(x)=\frac{1}{k+1}\sum_{j=0}^{k}\binom{k+1}{j}B^{+}_{j}x^{k+1-j}$, a $(k+1)^{th}$-degree polynomial with rational coefficients. Then Eqns 2-7 show that $g(x)+(x+1)^{k}=g(x+1)$. An analogous derivation shows that

Proposition 2.

The first $k+2$ Bernoulli numbers $B^{+}_{j}$ are defined by the coefficients $a_{j}$ of the minimal degree polynomial that satisfies the functional equation $f(x)+(x+1)^{k}=f(x+1)$ via the equation $B^{+}_{j}=(k+1)a_{k+1-j}/\binom{k+1}{j}$.

Since the Bernoulli polynomials $B_{k}(x)$ form an Appell sequence [4], i.e., it satisfies the differential equation $\frac{d}{dx}B_{k}(x)=kB_{k-1}(x)$, it is clear that $f_{k}$ is an Appell sequence as well. This implies that it satisfies Appell’s identity: $f_{k}(x+y)=\sum_{i=0}^{k}\binom{k}{i}f_{i}(x)y^{k-i}$, i.e. $f_{k}(x)+x^{k}=f_{k}(x+1)=\sum_{i=0}^{k}\binom{k}{i}f_{i}(x)$. This simplifies to the equation $\sum_{i=0}^{k-1}\binom{k}{i}f_{i}(x)=x^{k}$.