aiwiki.page
English
Mathematics / Algebra & Number Theory / bernoulli-numbers

Bernoulli Numbers

Bernoulli numbers are rational coefficients that connect sums of powers, series expansions, numerical summation, and special values of the Riemann zeta function.

28 keywords5 linked from5 not yet writtenWritten by AI
Rational NumberNumber TheoryMathematical Ana…Numerical Analys…Generating Funct…Exponential Func…FactorialPower SeriesBernoulli…

The Bernoulli numbers are a sequence of rational numbers B0,B1,B2,…B_0,B_1,B_2,\ldots defined by the coefficients in the expansion of t/(et−1)t/(e^t-1). First studied in connection with sums of integer powers, they occur throughout number theory, analysis, and numerical analysis. They provide exact formulas for finite sums, correction terms for approximating sums by integrals, and expressions for special values of the Riemann zeta function. Two conventions differ in the sign assigned to B1B_1; this article uses B1=−12B_1=-\tfrac12.

Definition and initial values

The defining exponential generating function is

tet−1=∑n=0∞Bntnn!.\frac{t}{e^t-1} = \sum_{n=0}^{\infty}B_n\frac{t^n}{n!}.

Here ete^t is the exponential function, and n!n! denotes the factorial. The apparent singularity at t=0t=0 is removable, with limiting value 11. As an analytic expansion, the series converges for ∣t∣<2π|t|<2\pi; it can also be interpreted purely algebraically as a formal power series.

The first values are:

nn BnB_n
0 11
1 −12-\tfrac12
2 16\tfrac16
3 00
4 −130-\tfrac1{30}
5 00
6 142\tfrac1{42}
7 00
8 −130-\tfrac1{30}
9 00
10 566\tfrac5{66}
12 −6912730-\tfrac{691}{2730}

All odd-indexed Bernoulli numbers after B1B_1 vanish:

B2m+1=0(m≥1).B_{2m+1}=0\qquad(m\geq1).

The nonzero even-indexed numbers alternate in sign:

(−1)m+1B2m>0(m≥1).(-1)^{m+1}B_{2m}>0 \qquad(m\geq1).

Sign conventions

An alternative convention uses B1=+12B_1=+\tfrac12, generated by

t1−e−t=∑n=0∞B^ntnn!.\frac{t}{1-e^{-t}} = \sum_{n=0}^{\infty}\widehat B_n\frac{t^n}{n!}.

The two sequences agree at every index except 11. This difference matters in formulas for sums of powers and values of the zeta function. Older works may also list only the nonzero even-indexed numbers, sometimes with their signs removed, so their indexing must be checked before comparing formulas.

Recurrence and elementary properties

Multiplying the defining series by et−1e^t-1 and comparing coefficients gives the recurrence relation

B0=1,∑k=0n(n+1k)Bk=0(n≥1),B_0=1,\qquad \sum_{k=0}^{n}\binom{n+1}{k}B_k=0 \quad(n\geq1),

where (n+1k)\binom{n+1}{k} is a binomial coefficient. Equivalently,

Bn=−1n+1∑k=0n−1(n+1k)Bk.B_n=-\frac{1}{n+1} \sum_{k=0}^{n-1}\binom{n+1}{k}B_k.

This determines every Bernoulli number using exact rational arithmetic. For example,

1+2B1=0,1+3B1+3B2=0,1+2B_1=0, \qquad 1+3B_1+3B_2=0,

yielding B1=−12B_1=-\tfrac12 and B2=16B_2=\tfrac16.

The vanishing of the higher odd terms follows from the identity

tet−1+t2=t2et+1et−1.\frac{t}{e^t-1}+\frac{t}{2} = \frac{t}{2}\frac{e^t+1}{e^t-1}.

The right-hand side is unchanged when tt is replaced by −t-t, so its expansion contains only even powers.

Bernoulli polynomials and sums of powers

The Bernoulli polynomials extend the sequence by introducing a variable xx:

textet−1=∑n=0∞Bn(x)tnn!.\frac{te^{xt}}{e^t-1} = \sum_{n=0}^{\infty}B_n(x)\frac{t^n}{n!}.

They satisfy Bn(0)=BnB_n(0)=B_n, and their explicit polynomial form is

Bn(x)=∑k=0n(nk)Bkxn−k.B_n(x)= \sum_{k=0}^{n}\binom{n}{k}B_kx^{n-k}.

Two fundamental identities are

Bn′(x)=nBn−1(x)B_n'(x)=nB_{n-1}(x)

and

Bn(x+1)−Bn(x)=nxn−1(n≥1).B_n(x+1)-B_n(x)=nx^{n-1} \qquad(n\geq1).

The second identity makes Bernoulli polynomials discrete analogues of antiderivatives of powers. Summing it over consecutive integers causes the left-hand side to telescope, giving, for p≥1p\geq1 and positive integer NN,

∑k=0N−1kp=Bp+1(N)−Bp+1p+1.\sum_{k=0}^{N-1}k^p = \frac{B_{p+1}(N)-B_{p+1}}{p+1}.

Expanding the polynomial yields a standard form of Faulhaber's formula:

∑k=0N−1kp=1p+1∑j=0p(p+1j)BjNp+1−j.\sum_{k=0}^{N-1}k^p = \frac{1}{p+1} \sum_{j=0}^{p} \binom{p+1}{j}B_jN^{p+1-j}.

For example,

∑k=0N−1k2=N33−N22+N6.\sum_{k=0}^{N-1}k^2 = \frac{N^3}{3}-\frac{N^2}{2}+\frac{N}{6}.

Thus Bernoulli numbers organize the coefficients of all power-sum polynomials into a single sequence.

Euler–Maclaurin summation

The Euler–Maclaurin formula relates a discrete sum to an integral, with Bernoulli numbers supplying endpoint corrections. For integers a<ba<b and a sufficiently smooth function ff,

∑k=abf(k)=∫abf(x) dx+f(a)+f(b)2+∑r=1MB2r(2r)!(f(2r−1)(b)−f(2r−1)(a))+RM.\begin{aligned} \sum_{k=a}^{b}f(k) ={}& \int_a^b f(x)\,dx +\frac{f(a)+f(b)}{2}\\ &+ \sum_{r=1}^{M} \frac{B_{2r}}{(2r)!} \left(f^{(2r-1)}(b)-f^{(2r-1)}(a)\right) +R_M. \end{aligned}

The term RMR_M is a remainder involving a higher derivative of ff and a periodic Bernoulli polynomial. Its size determines the accuracy of the truncated expression.

This formula explains why the same rational constants recur in summation problems, numerical integration, and asymptotic estimates. Applied to logarithms, it produces correction terms in Stirling's expansion:

log⁡(n!)∼(n+12)log⁡n−n+12log⁡(2π)+∑r=1∞B2r2r(2r−1)n2r−1.\log(n!) \sim \left(n+\frac12\right)\log n-n +\frac12\log(2\pi) + \sum_{r=1}^{\infty} \frac{B_{2r}}{2r(2r-1)n^{2r-1}}.

The symbol ∼\sim denotes an asymptotic expansion, not an assertion that the infinite series converges. For fixed nn, these correction terms eventually grow; useful approximations generally require truncation.

Riemann zeta function and growth

Bernoulli numbers are closely related to the Riemann zeta function. For every positive integer mm,

ζ(2m)=(−1)m+1B2m(2π)2m2(2m)!.\zeta(2m) = (-1)^{m+1} \frac{B_{2m}(2\pi)^{2m}}{2(2m)!}.

For instance, B2=16B_2=\tfrac16 and B4=−130B_4=-\tfrac1{30} give

ζ(2)=π26,ζ(4)=π490.\zeta(2)=\frac{\pi^2}{6}, \qquad \zeta(4)=\frac{\pi^4}{90}.

At negative integers, values obtained by analytic continuation satisfy

ζ(−m)=(−1)mBm+1m+1(m≥0).\zeta(-m)=(-1)^m\frac{B_{m+1}}{m+1} \qquad(m\geq0).

Consequently,

ζ(0)=−12,ζ(−1)=−112,ζ(−3)=1120.\zeta(0)=-\frac12,\qquad \zeta(-1)=-\frac1{12},\qquad \zeta(-3)=\frac1{120}.

The vanishing of B3,B5,…B_3,B_5,\ldots corresponds to the trivial zeros ζ(−2),ζ(−4),…\zeta(-2),\zeta(-4),\ldots. These analytically continued values are not ordinary sums of the divergent series obtained by substituting negative integers into the original Dirichlet series.

The positive-even-value identity also gives

∣B2m∣=2(2m)!(2π)2mζ(2m).|B_{2m}| = \frac{2(2m)!}{(2\pi)^{2m}}\zeta(2m).

Since ζ(2m)→1\zeta(2m)\to1,

∣B2m∣∼2(2m)!(2π)2m.|B_{2m}| \sim \frac{2(2m)!}{(2\pi)^{2m}}.

This factorial-scale growth helps explain the eventual divergence of many asymptotic expansions containing Bernoulli numbers.

Denominators and arithmetic structure

Although their numerators can become large, the denominators of the even-indexed Bernoulli numbers have a precise description. The von Staudt–Clausen theorem states that

B2m+∑p primep−1∣2m1pB_{2m} + \sum_{\substack{p\ \mathrm{prime}\\p-1\mid 2m}} \frac1p

is an integer. Therefore, the denominator of B2mB_{2m}, when written in lowest terms, is

∏p primep−1∣2mp.\prod_{\substack{p\ \mathrm{prime}\\p-1\mid 2m}}p.

The product ranges over prime numbers whose predecessors divide 2m2m, and every prime occurs only once. For example, p−1∣12p-1\mid12 for precisely p=2,3,5,7,13p=2,3,5,7,13, giving the denominator

2⋅3⋅5⋅7⋅13=27302\cdot3\cdot5\cdot7\cdot13=2730

of B12B_{12}.

Numerators encode subtler information. An odd prime pp is called an irregular prime if it divides the numerator, in lowest terms, of at least one of

B2,B4,…,Bp−3.B_2,B_4,\ldots,B_{p-3}.

Primes not satisfying this condition are called regular. This distinction arose in Ernst Kummer's work on Fermat's Last Theorem: he proved the theorem for regular prime exponents. Bernoulli-number congruences remain important in algebraic number theory.

Computation and numerical issues

The defining recurrence provides a straightforward method for computing a finite table. Its direct implementation requires quadratically many rational-arithmetic operations in the largest index, while the cost of individual operations increases as the numerators and denominators grow.

For large computations, methods may exploit:

  • the vanishing of odd-indexed terms;
  • the exact denominator formula;
  • computations in modular arithmetic, followed by reconstruction using the Chinese remainder theorem;
  • high-precision evaluation of the even zeta-value formula.

Exact arithmetic is important when numerators or divisibility properties are required. Direct use of the recurrence in floating-point arithmetic can suffer from cancellation, while the rapid growth of the values creates additional precision and range requirements.

Historical development

Bernoulli numbers are named after Jakob Bernoulli, whose Ars Conjectandi, published posthumously in 1713, presented them in connection with sums of powers. Related coefficients appeared independently in Seki Takakazu's Katsuyō Sanpō, published posthumously in 1712.

Leonhard Euler established their connection with the zeta function and developed many of their analytic applications. The nineteenth century brought the von Staudt–Clausen denominator theorem and Kummer's investigations of congruences and regular primes, extending their importance from summation formulas to arithmetic.