betaFeedback

WorksMathematics

Some Expressions involving Bernoulli, Harmonic, and Stirling Numbers

Abstract

In this paper, we establish several identities involving Bernoulli numbers, harmonic numbers, and Stirling numbers. Our approach extends the method developed by Kargin and Mutluer.

Introduction

The Bernoulli numbers, BnB_n, arose in the study of sums of powers of positive integers. They are commonly defined via the generating function

zez−1=∑n=0∞Bnznn!.\frac{z}{e^z-1}=\sum_{n=0}^{\infty}B_n\frac{z^n}{n!}.

The first few Bernoulli numbers are B0=1B_0=1, B1=−1/2B_1=-1/2, B2=1/6B_2=1/6 and B4=−1/30B_4=-1/30. It is well-known that B2n+1=0B_{2n+1}=0 for all n≥1n\geq1.

In recent years, considerable effort has been devoted to establishing connections between Bernoulli numbers and harmonic numbers HnH_n, defined by

Hn:=∑k=1n1k,H_n:=\sum_{k=1}^{n}\frac{1}{k},

with the convention that H0=0H_0=0. To describe some of these identities, we introduce the Stirling numbers of the first and second kinds, defined, respectively, by

x(n)=x(x+1)⋯(x+n−1)=∑k=0n[nk]xk,x^{(n)}=x(x+1)\cdots(x+n-1)=\sum_{k=0}^{n}\begin{bmatrix}n\\k\end{bmatrix}x^k,

and

xn=∑k=0n(−1)n−k{nk}x(k).x^n=\sum_{k=0}^{n}(-1)^{n-k}\begin{Bmatrix}n\\k\end{Bmatrix}x^{(k)}.

Various properties of these number can be found in the book by Graham, Knuth, and Patashnik [2]. Several of these properties will be used throughout this paper.

A simple identity connecting Bernoulli numbers, harmonic numbers, and Stirling numbers was obtained by Kellner [4] using differentiation

∑k=1n(−1)k−1{nk}k!k+1Hk=n2Bn−1.\sum_{k=1}^{n}(-1)^{k-1}\begin{Bmatrix}n\\k\end{Bmatrix}\frac{k!}{k+1}H_k=\frac{n}{2}B_{n-1}.

More recently, Akkanat and Kargin [1] established two related identities

∑k=1n(−1)k−1{nk}(k−1)!Hk=Bn−1,\sum_{k=1}^{n}(-1)^{k-1}\left\{\begin{matrix} n \\ k \end{matrix}\right\}(k-1)!H_k=B_{n-1},
∑k=2n(−1)k−1{nk}(k−1)!HkHk−1=n+12Bn−2.\sum_{k=2}^{n}(-1)^{k-1}\left\{\begin{matrix} n \\ k \end{matrix}\right\}(k-1)!H_kH_{k-1}=\frac{n+1}{2}B_{n-2}.

Observe that (k−1)!=[k1](k-1)!=\left[\begin{matrix} k \\ 1 \end{matrix}\right] and (k−1)!Hk−1=[k2](k-1)!H_{k-1}=\left[\begin{matrix} k \\ 2 \end{matrix}\right]. These formulas suggest a more general formula involving [kj]\left[\begin{matrix} k \\ j \end{matrix}\right] for each jj. Indeed, Kargin and Mutluer [3] proved that

∑k=jn(−1)k−j{nk}[kj]Hk=[(nj)−1]Bn−jn−j.\sum_{k=j}^{n}(-1)^{k-j}\left\{\begin{matrix} n \\ k \end{matrix}\right\}\left[\begin{matrix} k \\ j \end{matrix}\right]H_k=\left[\binom{n}{j}-1\right]\frac{B_{n-j}}{n-j}.

In this paper, we build upon the method of Kargin and Mutluer to derive several further identities involving these numbers.

Main Result

We now state the main result, which will be used throughout this paper.

Theorem 1. Let f(x)f(x) be a real-valued function. Define

Sn,j(f):=∑k=jn(−1)k−j{nk}[kj]f(k).S_{n,j}(f):=\sum_{k=j}^{n}(-1)^{k-j}\left\{\begin{matrix} n \\ k \end{matrix}\right\}\left[\begin{matrix} k \\ j \end{matrix}\right]f(k).

Then, for 1≤j≤n1\le j\le n, we have

Sn+1,j(f)−Sn,j−1(f)=Sn+1,j(Δf)−Sn,j(xΔf),S_{n+1,j}(f)-S_{n,j-1}(f)=S_{n+1,j}(\Delta f)-S_{n,j}(x\Delta f),

where Δf(k)=f(k)−f(k−1)\Delta f(k)=f(k)-f(k-1) denotes the backward difference of ff.

Proof. We begin by considering the first term on the left-hand side

Sn+1,j(f)=∑k=jn+1(−1)k−j{n+1k}[kj]f(k).S_{n+1,j}(f)=\sum_{k=j}^{n+1}(-1)^{k-j}\left\{\begin{matrix} n+1 \\ k \end{matrix}\right\}\left[\begin{matrix} k \\ j \end{matrix}\right]f(k).

Applying the recurrence relation for Stirling numbers of the second kind [2 pg. 164]

{n+1k}=k{nk}+{nk−1},\left\{\begin{matrix} n+1 \\ k \end{matrix}\right\}=k\left\{\begin{matrix} n \\ k \end{matrix}\right\}+\left\{\begin{matrix} n \\ k-1 \end{matrix}\right\},

we see that

Sn+1,j(f)=∑k=jn+1(−1)k−j{nk}[kj]kf(k)+∑k=jn+1(−1)k−j{nk−1}[kj]f(k).S_{n+1,j}(f)=\sum_{k=j}^{n+1}(-1)^{k-j}\left\{\begin{matrix} n \\ k \end{matrix}\right\}\left[\begin{matrix} k \\ j \end{matrix}\right]kf(k)+\sum_{k=j}^{n+1}(-1)^{k-j}\left\{\begin{matrix} n \\ k-1 \end{matrix}\right\}\left[\begin{matrix} k \\ j \end{matrix}\right]f(k).

We also use the recurrence relation for Stirling numbers of the first kind [2 pg. 164]

[k+1j]=k[kj]+[kj−1],\begin{bmatrix} k+1 \\ j \end{bmatrix}=k\begin{bmatrix} k \\ j \end{bmatrix}+\begin{bmatrix} k \\ j-1 \end{bmatrix},

to obtain

Sn+1,j(f)=∑k=jn(−1)k−j{nk}[k+1j]f(k)−∑k=jn(−1)k−j{nk}[kj−1]f(k)−∑k=j−1n(−1)k−j{nk}[k+1j]f(k+1).(1)S_{n+1,j}(f)=\sum_{k=j}^{n}(-1)^{k-j}\begin{Bmatrix} n \\ k \end{Bmatrix}\begin{bmatrix} k+1 \\ j \end{bmatrix}f(k)-\sum_{k=j}^{n}(-1)^{k-j}\begin{Bmatrix} n \\ k \end{Bmatrix}\begin{bmatrix} k \\ j-1 \end{bmatrix}f(k)-\sum_{k=j-1}^{n}(-1)^{k-j}\begin{Bmatrix} n \\ k \end{Bmatrix}\begin{bmatrix} k+1 \\ j \end{bmatrix}f(k+1). \tag*{(1)}

Separating the term corresponding to k=j−1k=j-1 in the last sum gives

∑k=j−1n(−1)k−j{nk}[k+1j]f(k+1)=−{nj−1}f(j)+∑k=jn(−1)k−j{nk}[k+1j]f(k+1).\sum_{k=j-1}^{n}(-1)^{k-j}\begin{Bmatrix} n \\ k \end{Bmatrix}\begin{bmatrix} k+1 \\ j \end{bmatrix}f(k+1)=-\begin{Bmatrix} n \\ j-1 \end{Bmatrix}f(j)+\sum_{k=j}^{n}(-1)^{k-j}\begin{Bmatrix} n \\ k \end{Bmatrix}\begin{bmatrix} k+1 \\ j \end{bmatrix}f(k+1).

The first term on the right hand side can be combined with the second sum in Equation (1) to give Sn,j−1(f)S_{n,j-1}(f). Hence, Equation (1) simplifies to

Sn+1,j(f)=∑k=j−1n(−1)k−j{nk}[k+1j](f(k)−f(k+1))+Sn,j−1(f),S_{n+1,j}(f)=\sum_{k=j-1}^{n}(-1)^{k-j}\begin{Bmatrix} n \\ k \end{Bmatrix}\begin{bmatrix} k+1 \\ j \end{bmatrix}\left(f(k)-f(k+1)\right)+S_{n,j-1}(f),

or equivalently

Sn+1,j(f)=∑k=jn+1(−1)k−j{nk−1}[kj]Δf(k)+Sn,j−1(f).S_{n+1,j}(f)=\sum_{k=j}^{n+1}(-1)^{k-j}\begin{Bmatrix} n \\ k-1 \end{Bmatrix}\begin{bmatrix} k \\ j \end{bmatrix}\Delta f(k)+S_{n,j-1}(f).

Applying the recurrence relation for Stirling numbers of the second kind once more yields

Sn+1,j(f)−Sn,j−1(f)=∑k=jn+1(−1)k−j{n+1k}[kj]Δf(k)−∑k=jn(−1)k−j{nk}[kj]kΔf(k),S_{n+1,j}(f)-S_{n,j-1}(f)=\sum_{k=j}^{n+1}(-1)^{k-j}\begin{Bmatrix} n+1 \\ k \end{Bmatrix}\begin{bmatrix} k \\ j \end{bmatrix}\Delta f(k)-\sum_{k=j}^{n}(-1)^{k-j}\begin{Bmatrix} n \\ k \end{Bmatrix}\begin{bmatrix} k \\ j \end{bmatrix}k\Delta f(k),

which is precisely the required expression. □

Applications of Main Result

Example 1

When f(k)=Hkf(k)=H_k, we have Δf(k)=Hk−Hk−1=1/k\Delta f(k)=H_k-H_{k-1}=1/k. The right hand side of Theorem 1 becomes

∑k=jn+1(−1)k−j{n+1k}[kj]1k−∑k=jn(−1)k−j{nk}[kj].\sum_{k=j}^{n+1}(-1)^{k-j}\begin{Bmatrix} n+1 \\ k \end{Bmatrix}\begin{bmatrix} k \\ j \end{bmatrix}\frac{1}{k}-\sum_{k=j}^{n}(-1)^{k-j}\begin{Bmatrix} n \\ k \end{Bmatrix}\begin{bmatrix} k \\ j \end{bmatrix}.

Using the identities [2 Eq. (6.100)]

∑k=jn(−1)k−j(nk)[kj]1k=(n−1j)Bn−jn−j,\sum_{k=j}^{n}(-1)^{k-j}\binom{n}{k}\left[\begin{matrix}k\\j\end{matrix}\right]\frac{1}{k}=\binom{n-1}{j}\frac{B_{n-j}}{n-j},

and [2 pg. 164]

∑k=jn(−1)k−j(nk)[kj]=δn,j,\sum_{k=j}^{n}(-1)^{k-j}\binom{n}{k}\left[\begin{matrix}k\\j\end{matrix}\right]=\delta_{n,j},

where δm,n\delta_{m,n} denotes the Kronecker delta, we get for n>jn>j,

Sn+1,j(Hk)−Sn,j−1(Hk)=(nj)Bn+1−jn+1−j.S_{n+1,j}(H_k)-S_{n,j-1}(H_k)=\binom{n}{j}\frac{B_{n+1-j}}{n+1-j}.

We can iterate this identity by successively lowering the value of jj. Since Sn−j,0(Hk)=0S_{n-j,0}(H_k)=0, we obtain

Sn+1,j(Hk)=Bn+1−jn+1−j∑i=0j−1(n−ij−i)=[(n+1j)−1]Bn+1−jn+1−j.S_{n+1,j}(H_k)=\frac{B_{n+1-j}}{n+1-j}\sum_{i=0}^{j-1}\binom{n-i}{j-i}=\left[\binom{n+1}{j}-1\right]\frac{B_{n+1-j}}{n+1-j}.

Replacing n+1n+1 by nn, we arrive at

∑k=jn(−1)k−j(nk)[kj]Hk=[(nj)−1]Bn−jn−j,(2)\sum_{k=j}^{n}(-1)^{k-j}\binom{n}{k}\left[\begin{matrix}k\\j\end{matrix}\right]H_k=\left[\binom{n}{j}-1\right]\frac{B_{n-j}}{n-j}, \tag*{(2)}

which recovers the main theorem of Kargin and Mutlu­er [3].

Example 2

When f(k)=kHkf(k)=kH_k, we have Δf(k)=kHk−kHk−1+Hk−1=Hk−1+1=Hk−1/k+1\Delta f(k)=kH_k-kH_{k-1}+H_{k-1}=H_{k-1}+1=H_k-1/k+1. Therefore, Theorem 1 gives

Sn+1,j(kHk)−Sn,j−1(kHk)=∑k=jn+1(−1)k−j(n+1k)[kj](Hk−1k+1)−∑k=jn(−1)k−j(nk)[kj](kHk+k−1).S_{n+1,j}(kH_k)-S_{n,j-1}(kH_k)=\sum_{k=j}^{n+1}(-1)^{k-j}\binom{n+1}{k}\left[\begin{matrix}k\\j\end{matrix}\right]\left(H_k-\frac{1}{k}+1\right) -\sum_{k=j}^{n}(-1)^{k-j}\binom{n}{k}\left[\begin{matrix}k\\j\end{matrix}\right](kH_k+k-1).

As Sn,j(1)=δn,jS_{n,j}(1)=\delta_{n,j}, the constant terms in both sums vanishes for n>jn>j. Thus, the remaining terms become

Sn+1,j(kHk)+Sn,j(kHk)−Sn,j−1(kHk)=∑k=jn+1(−1)k−j(n+1k)[kj](Hk−1k)−∑k=jn(−1)k−j(nk)[kj]k.(3)S_{n+1,j}(kH_k)+S_{n,j}(kH_k)-S_{n,j-1}(kH_k)=\sum_{k=j}^{n+1}(-1)^{k-j}\binom{n+1}{k}\left[\begin{matrix}k\\j\end{matrix}\right]\left(H_k-\frac{1}{k}\right) -\sum_{k=j}^{n}(-1)^{k-j}\binom{n}{k}\left[\begin{matrix}k\\j\end{matrix}\right]k. \tag*{(3)}

The first sum can be evaluated using the identities established above. We therefore need only to look at the second sum. The following lemma will be useful.

Lemma 2. For 1≤j≤n1 \le j \le n, we have

∑k=jn(−1)k−j{nk}[kj]k=(−1)n−j(nj−1).\sum_{k=j}^{n}(-1)^{k-j}\left\{\begin{matrix}n\\k\end{matrix}\right\}\left[\begin{matrix}k\\j\end{matrix}\right]k=(-1)^{n-j}\binom{n}{j-1}.

Proof. The exponential generating function for Stirling numbers of the first kind [2 pg. 351] is

∑k=j∞(−1)k−j[kj]zkk!=(log⁡(1+z))jj!.\sum_{k=j}^{\infty}(-1)^{k-j}\left[\begin{matrix}k\\j\end{matrix}\right]\frac{z^k}{k!}=\frac{(\log(1+z))^j}{j!}.

Differentiating both sides with respect to zz and multiplying by zz, we get

∑k=j∞(−1)k−j[kj]kzkk!=z1+z(log⁡(1+z))j−1(j−1)!.\sum_{k=j}^{\infty}(-1)^{k-j}\left[\begin{matrix}k\\j\end{matrix}\right]\frac{kz^k}{k!}=\frac{z}{1+z}\frac{(\log(1+z))^{j-1}}{(j-1)!}.

Applying the Stirling transformation amounts to replacing zz with ez−1e^z-1 on the right hand side. Hence, we obtain

∑n=1∞(∑k=jn(−1)k−j{nk}[kj]k)znn!=ez−1ez(log⁡(ez))j−1(j−1)!=(1−e−z)zj−1(j−1)!.\sum_{n=1}^{\infty}\left(\sum_{k=j}^{n}(-1)^{k-j}\left\{\begin{matrix}n\\k\end{matrix}\right\}\left[\begin{matrix}k\\j\end{matrix}\right]k\right)\frac{z^n}{n!}=\frac{e^z-1}{e^z}\frac{(\log(e^z))^{j-1}}{(j-1)!}=(1-e^{-z})\frac{z^{j-1}}{(j-1)!}.

On the other hand, the Taylor expansion of e−ze^{-z} gives

(1−e−z)zj−1(j−1)!=∑k=1∞(−1)k−1zk+j−1k!(j−1)!=∑n=1∞(−1)n−j(nj−1)znn!.(1-e^{-z})\frac{z^{j-1}}{(j-1)!}=\sum_{k=1}^{\infty}(-1)^{k-1}\frac{z^{k+j-1}}{k!(j-1)!}=\sum_{n=1}^{\infty}(-1)^{n-j}\binom{n}{j-1}\frac{z^n}{n!}.

where, in the second equality, we set n=k+j−1n=k+j-1. Comparing coefficients of zn/n!z^n/n! yields

∑k=jn(−1)k−j{nk}[kj]k=(−1)n−j(nj−1),\sum_{k=j}^{n}(-1)^{k-j}\left\{\begin{matrix}n\\k\end{matrix}\right\}\left[\begin{matrix}k\\j\end{matrix}\right]k=(-1)^{n-j}\binom{n}{j-1},

as required. □

Returning to the Equation (3), we have

Sn+1,j(kHk)+Sn,j(kHk)−Sn,j−1(kHk)=[(nj−1)−1]Bn+1−jn+1−j−(−1)n−j(nj−1).(4)S_{n+1,j}(kH_k)+S_{n,j}(kH_k)-S_{n,j-1}(kH_k)=\left[\binom{n}{j-1}-1\right]\frac{B_{n+1-j}}{n+1-j}-(-1)^{n-j}\binom{n}{j-1}. \tag*{(4)}

Equation (4) provides a recurrence relation for evaluating Sn,j(kHk)S_{n,j}(kH_k). We first consider the case j=1j=1. Since Sn,0(kHk)=0S_{n,0}(kH_k)=0 for all n≥1n\geq1, Equation (4) reduces to

Sn+1,1(kHk)+Sn,1(kHk)=(−1)n.S_{n+1,1}(kH_k)+S_{n,1}(kH_k)=(-1)^n.

This leads to the following result.

Theorem 3. For n≥1n\geq1, it holds that

Sn,1(kHk)=∑k=1n(−1)k−1{nk}k!Hk=(−1)n−1n.S_{n,1}(kH_k)=\sum_{k=1}^{n}(-1)^{k-1}\left\{\begin{matrix}n\\k\end{matrix}\right\}k!H_k=(-1)^{n-1}n.

Proof. We proceed by induction on nn. For n=1n=1, we have S1,1(kHk)=1S_{1,1}(kH_k)=1, so the result holds in the base case. Suppose that Sm,1(kHk)=(−1)m−1mS_{m,1}(kH_k)=(-1)^{m-1}m for some m≥1m\ge1. Taking n=m+1n=m+1, we obtain

Sm+1,1(kHk)=(−1)m−Sm,1(kHk)=(−1)n−(−1)m−1m=(−1)m(m+1).S_{m+1,1}(kH_k)=(-1)^m-S_{m,1}(kH_k)=(-1)^n-(-1)^{m-1}m=(-1)^m(m+1).

This completes the induction step and the proof.

This identity also follows directly by applying the Stirling transformation to the well-known identity

n!Hn=∑k=1nk[nk].n!H_n=\sum_{k=1}^{n}k\begin{bmatrix}n\\k\end{bmatrix}.

Next, we consider the case j=2j=2. Using the fact that Sn,1(kHk)=(−1)n−1nS_{n,1}(kH_k)=(-1)^{n-1}n, Equation (4) becomes

Sn+1,2(kHk)+Sn,2(kHk)=(−1)n−1(2n)+Bn−1.S_{n+1,2}(kH_k)+S_{n,2}(kH_k)=(-1)^{n-1}(2n)+B_{n-1}.

This recurrence yields the following theorem.

Theorem 4. For n≥2n\ge2, it holds that

Sn,2(kHk)=∑k=2n(−1)k{nk}k!HkHk−1=(−1)nn(n−1)+∑k=0n−2(−1)n−kBk.S_{n,2}(kH_k)=\sum_{k=2}^{n}(-1)^k\begin{Bmatrix}n\\k\end{Bmatrix}k!H_kH_{k-1}=(-1)^n n(n-1)+\sum_{k=0}^{n-2}(-1)^{n-k}B_k.

Proof. The first equality follows from the fact that [k2]=(k−1)!Hk−1\begin{bmatrix}k\\2\end{bmatrix}=(k-1)!H_{k-1}. We prove the second equality by induction. For n=2n=2, we have S2,2(kHk)=2H2=3=2+B0S_{2,2}(kH_k)=2H_2=3=2+B_0, so the base case is done. Suppose that the equality holds for some m≥2m\ge2. Taking n=m+1n=m+1, we get

Sm+1,2(kHk)=(−1)m−1m(m−1)−∑k=0m−2(−1)m−kBk+(−1)m−1(2m)+Bm−1=(−1)mm(m+1)+∑k=0m−1(−1)m+1−kBk.\begin{aligned} S_{m+1,2}(kH_k)&=(-1)^{m-1}m(m-1)-\sum_{k=0}^{m-2}(-1)^{m-k}B_k+(-1)^{m-1}(2m)+B_{m-1}\\ &=(-1)^m m(m+1)+\sum_{k=0}^{m-1}(-1)^{m+1-k}B_k. \end{aligned}

Thus, the result holds for n=m+1n=m+1, completing the induction.

Notice that B2n+1=0B_{2n+1}=0 for all n≥1n\ge1, so we can further simplify the sum on the right hand side of Theorem 4. Suppose that n≥3n\ge3, then

∑k=0n−2(−1)kBk=1+∑k=0n−2Bk,\sum_{k=0}^{n-2}(-1)^kB_k=1+\sum_{k=0}^{n-2}B_k,

where the extra 1 comes from replacing −B1=1/2-B_1=1/2 in the sum with B1=−1/2B_1=-1/2. Therefore, by adding an extra k=1k=1 term in the sum, which does not change its value, we can rewrite Theorem 4 as

∑k=1n(−1)n−k{nk}k!HkHk−1=n(n−1)+1+∑k=0n−2Bk.\sum_{k=1}^{n}(-1)^{n-k}\begin{Bmatrix}n\\k\end{Bmatrix}k!H_kH_{k-1}=n(n-1)+1+\sum_{k=0}^{n-2}B_k.

Since Hk−1=Hk−1/kH_{k-1}=H_k-1/k, the left hand side becomes

∑k=1n(−1)n−k{nk}k!Hk2−∑k=1n(−1)n−k{nk}(k−1)!Hk=n(n−1)+1+∑k=0n−2Bk.\sum_{k=1}^{n}(-1)^{n-k}\left\{\begin{matrix}n\\k\end{matrix}\right\}k!H_k^2-\sum_{k=1}^{n}(-1)^{n-k}\left\{\begin{matrix}n\\k\end{matrix}\right\}(k-1)!H_k=n(n-1)+1+\sum_{k=0}^{n-2}B_k.

Using (2) with j=1j=1, the second sum on the left hand side equals to Bn−1B_{n-1}. Therefore, we arrive at the next corollary.

Corollary 5. For n≥3n\geq3, it holds that

∑k=1n(−1)n−k{nk}k!Hk2=n(n−1)−1+∑k=0n−1Bk.\sum_{k=1}^{n}(-1)^{n-k}\left\{\begin{matrix}n\\k\end{matrix}\right\}k!H_k^2=n(n-1)-1+\sum_{k=0}^{n-1}B_k.

References

  1. [1]P. Akkanat and L. Kargin, Harmonic geometric polynomials via geometric polynomials and their applications. Preprint, arXiv: 2512.07416, 2025.arxiv.org/abs/2512.07416
  2. [2]R. L. Graham, D. E. Knuth, and O. Patashnik, Concrete Mathematics, second edition. Addison-Wesley, Reading, 1994.
  3. [3]L. Kargin and M. Mutluer, New expression for the Bernoulli numbers and its applications. Preprint, arXiv: 2602.03440, 2026.
  4. [4]B. C. Kellner, Identities between polynomials related to Stirling and harmonic numbers, Integers 14 (2014), #A54.

Paper details

vlnc:W_tudm4tb7
Version
v1 · latest
Published
Fresh — added to the Hub within the last 30 days
Added
License
CC BY 4.0
Usage

views · 11

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Not recorded for this paper

View statement
Looking for related papers…
Contents