1 Definition and basic properties

Bernoulli numbers \(\{B_n\}_{n\ge 0}\) are a sequence of rational numbers defined implicitly by a generating function. They act as universal coefficients that appear in series expansions and in formulas converting sums over integers into analytic expressions.

1.1 Generating function

A standard definition uses the exponential generating function \[ \frac{x}{e^x-1}=\sum_{n=0}^{\infty} B_n\frac{x^n}{n!}. \] Because the right-hand side is a formal power series, this relation uniquely specifies each \(B_n\) as a rational number.

1.1.1 Exponential generating function

Expanding \(e^x=\sum_{m\ge 0}x^m/m!\) and then forming \(x/(e^x-1)\) yields the coefficients \(B_n\). In practice, one can compute \(B_n\) recursively by matching coefficients of equal powers of \(x\).

1.1.2 Relation to formal power series

The definition is valid both analytically for \(x\) small and algebraically in the ring of formal power series. In the formal viewpoint, questions of convergence are irrelevant; only coefficient identities matter.

1.2 Initial values and parity features

From the generating function, the first few terms can be extracted. Their explicit forms highlight two recurring phenomena: alternating signs (with a typical convention) and strong parity restrictions.

1.2.1 Values of \(B_0\) through \(B_4\)

Using the expansion of \(\frac{x}{e^x-1}\), one finds \[ B_0=1,\qquad B_1=-\frac12,\qquad B_2=\frac16,\qquad B_3=0,\qquad B_4=-\frac1{30}. \]

1.2.2 Vanishing of odd-indexed Bernoulli numbers (for \(n>1\))

For indices greater than \(1\), all odd-indexed Bernoulli numbers vanish: \[ B_{2m+1}=0\quad\text{for }m\ge 1. \] This follows from the symmetry properties of the generating function when expressed using the exponential series and the fact that the coefficient comparison eliminates odd powers beyond the linear term.

1.3 Recursive relations

Although the generating function is the cleanest definition, it is often computationally convenient to derive recurrences.

1.3.1 Standard recursion from the generating function

Write \[ \left(\sum_{n=0}^{\infty} B_n\frac{x^n}{n!}\right)\left(\sum_{m=0}^{\infty}\frac{x^m}{m!}\right)=\frac{x}{e^x-1}\cdot \frac{e^x-1}{1}=x. \] More directly, rearranging the identity \(\frac{x}{e^x-1}=\sum_{n\ge 0}B_n x^n/n!\) implies \[ \left(\sum_{n\ge 0} B_n\frac{x^n}{n!}\right)\left(\sum_{m\ge 0}\frac{x^m}{m!}-1\right)=x. \] Matching coefficients of \(x^k\) gives a recursion expressing \(B_k\) in terms of earlier \(B_j\). One common form is: \[ \sum_{j=0}^{k}\binom{k+1}{j}B_j=0\quad\text{for }k\ge 1, \] with \(B_0=1\) as the initial value. Solving for \(B_k\) yields \(B_k\) uniquely.

1.3.2 Consequences for computation

The recursion reduces the computation of \(B_n\) to arithmetic with rational numbers and already-known lower-index values. Together with the parity property, it implies that only even-index coefficients need to be computed for \(n>1\).

2 Characterizations and equivalent formulations

Bernoulli numbers admit multiple equivalent descriptions. These formulations connect them to summation identities, special-function expansions, and values of the Riemann zeta function.

2.1 Faulhaber’s theorem and power sums

Faulhaber’s theorem states that sums of integer powers can be expressed using polynomials in the upper limit.

2.1.1 Polynomial expressions for \(\sum_{m=1}^n m^k\)

For each nonnegative integer \(k\), there exists a polynomial \(P_k(n)\) of degree \(k+1\) such that \[ 1^k+2^k+\cdots+n^k=P_k(n). \] Bernoulli numbers provide a systematic way to write these polynomials and to read off their coefficients.

2.1.2 Degree and leading coefficient structure

A standard closed form is \[ \sum_{m=1}^{n} m^k=\frac{1}{k+1}\sum_{j=0}^{k}\binom{k+1}{j}B_j\,n^{k+1-j}, \] which makes the polynomial degree \(k+1\) explicit. The leading coefficient is \(1/(k+1)\), while lower-degree coefficients are rational combinations of Bernoulli numbers.

2.2 Via special function expansions

Bernoulli numbers appear as coefficients in expansions of quotients involving exponentials and related hyperbolic functions.

2.2.1 Expansion of \( \frac{x}{e^x-1} \)

The defining identity \[ \frac{x}{e^x-1}=\sum_{n=0}^{\infty} B_n\frac{x^n}{n!} \] is itself an example of a special-function expansion: the quotient of \(x\) by \(e^x-1\) has a Maclaurin series whose coefficients are Bernoulli numbers.

Through standard transformations, one can express related expansions in terms of Bernoulli numbers. For instance, using identities between \(\coth(x)\), \(\sinh(x)\), and \(\cosh(x)\), series for \(\coth(x)\) near \(x=0\) involve only even powers and therefore only even-index Bernoulli numbers. Such connections often simplify the study of Laurent or Taylor coefficients in problems involving hyperbolic functions.

2.3 In terms of zeta values

A celebrated relationship ties Bernoulli numbers to the Riemann zeta function at integer arguments.

2.3.1 Zeta at negative integers

For integers \(n\ge 1\), \[ \zeta(1-n)=-\frac{B_n}{n}. \] This formula converts Bernoulli numbers into analytic continuation values of \(\zeta(s)\) at nonpositive integers, making Bernoulli numbers central in discussions that extend beyond convergent series.

2.3.2 Zeta at nonnegative even integers (conceptual bridge)

While \(\zeta(2m)\) (for \(m\ge 1\)) is not directly expressed using Bernoulli numbers by the simplest linear formula above, Bernoulli numbers do enter the classic closed-form evaluation of \(\zeta(2m)\) in terms of \(\pi^{2m}\). Conceptually, this provides a bridge: Bernoulli numbers translate between discrete-power-sum information and analytic behavior of \(\zeta(s)\) at even positive integers.

3 Computation methods

Practical computation of Bernoulli numbers can be done in several ways, depending on whether one favors coefficient extraction, recursion, or tabulated conventions.

3.1 Direct use of the generating function

One can compute \(B_n\) by expanding \(\frac{x}{e^x-1}\) to the desired order and matching coefficients.

3.1.1 Coefficient extraction techniques

After expanding \(e^x-1\) into a power series, the quotient \(x/(e^x-1)\) can be computed using formal division or series inversion. The resulting coefficient of \(x^n\) multiplied by \(n!\) yields \(B_n\). This approach is straightforward for small \(n\) and can be automated.

3.1.2 Symbolic computation considerations

In exact arithmetic, series inversion requires careful handling of rational coefficients. Computer algebra systems typically manage these operations automatically, but the growth of intermediate numerators and denominators can still be significant for large \(n\).

3.2 Recurrence-based algorithms

Recurrence relations often outperform direct series expansion when many Bernoulli numbers are needed.

3.2.1 Using triangular recurrences

The coefficient-matching identity from the generating function leads to a triangular linear system: each new \(B_k\) depends only on earlier \(B_j\) with \(j<k\). Because the system is triangular, one solves sequentially without back-substitution.

3.2.2 Numerical stability and exact arithmetic

Because Bernoulli numbers grow rapidly in magnitude with \(k\), floating-point computation can suffer from cancellation. For high-precision requirements, exact rational arithmetic or modular methods are typically used to ensure correctness.

3.3 Tables and normalization conventions

Different sources may use slightly different sign conventions or indexing for \(B_1\), so careful normalization is required when comparing tables.

3.3.1 Sign conventions across literature

With the convention from \(\frac{x}{e^x-1}\), one has \(B_1=-\tfrac12\). Some authors instead define a variant with \(B_1=+\tfrac12\), which affects certain identities but leaves the even-index values unchanged.

3.3.2 Common indexing differences

While the index \(n\) itself is standard, formulas involving Bernoulli polynomials or zeta values can shift due to differing definitions of polynomial normalizations. Translation usually amounts to checking the generating function used in the source.

4 Analytic applications

Bernoulli numbers serve as coefficients controlling the accuracy of approximating sums by integrals and also govern large-order behavior of certain expansions.

4.1 Euler–Maclaurin summation formula

The Euler–Maclaurin formula expresses a sum of a smooth function in terms of an integral plus correction terms.

4.1.1 Where Bernoulli numbers enter

For a sufficiently smooth function \(f\), one version of the Euler–Maclaurin expansion includes terms of the form \[ \sum_{k=0}^{p} \frac{B_k}{k!}\, \bigl(f^{(k-1)}(b)-f^{(k-1)}(a)\bigr) \] (with details depending on which endpoint conventions are used). The key point is that the Bernoulli numbers multiply higher derivatives of \(f\), creating a systematic asymptotic series for the difference between discrete sums and continuous integrals.

4.1.2 Truncation error and asymptotic behavior

Euler–Maclaurin series are generally asymptotic: truncating after finitely many Bernoulli correction terms produces an error whose size depends on smoothness and on the scale of the derivatives. The magnitude and sign pattern of Bernoulli numbers influence how the remainder behaves as the truncation order increases.

4.2 Asymptotic growth and approximations

Bernoulli numbers grow quickly in absolute value, affecting asymptotic estimates.

4.2.1 Magnitude growth of \( |B_{2n}| \)

Although the exact values are rational, their denominators and numerators combine so that \(B_{2n}\) increases very rapidly with \(n\). This rapid growth is consistent with their role as coefficients in high-order correction terms.

4.2.2 Stirling-type intuition for large indices

One can develop an intuition—often via connections to zeta values and to \(\pi\)-dependent formulas—suggesting that the size of \(B_{2n}\) is comparable to factorial growth. This aligns with how higher derivatives in Euler–Maclaurin corrections typically bring in large combinatorial factors.

4.3 Series expansions involving Bernoulli numbers

Bernoulli numbers also appear when expanding rational functions of exponentials and related quotients.

4.3.1 Laurent/Taylor expansions of elementary quotients

For example, the quotient \(\frac{x}{e^x-1}\) gives a Taylor series at \(0\). Related quotients, such as those obtained by shifting or combining exponentials, yield series (sometimes Laurent in a transformed variable) whose coefficients can be expressed using Bernoulli numbers.

4.3.2 Coefficient comparisons

A common technique is to compute the expansion of a target function in two ways: first using a known functional identity and second using an assumed series form. Matching coefficients then produces explicit expressions in terms of \(B_n\).

5 Structural results

Beyond applications, Bernoulli numbers satisfy structural identities involving polynomial algebra, discrete calculus, and Bernoulli polynomials.

5.1 Convolution and polynomial identities

Multiplicative relationships of generating functions translate into convolution-type identities for Bernoulli numbers.

5.1.1 Identities derived from generating functions

When generating functions are multiplied, coefficients combine via binomial convolution. Since Bernoulli numbers are coefficients in \(\frac{x}{e^x-1}\), such multiplication yields identities among sums of products \(B_iB_j\) with combinatorial weights.

5.1.2 Symmetry and combinatorial interpretations

Many structural identities can be interpreted combinatorially through counting arguments encoded by power sums and finite differences. Even when a direct counting model is not explicit, the symmetry in these formulas typically reflects the symmetry of the underlying generating functions.

5.2 Connection to finite differences

Bernoulli numbers connect naturally to discrete difference operators, which translate polynomial behavior into shift relations.

5.2.1 Discrete calculus viewpoint

Finite differences of polynomials have degrees reduced predictably, and the coefficients of power sums align with these differences. In that setting, Bernoulli numbers act as bridge terms converting between discrete and continuous descriptions.

5.2.2 Relation to binomial sums

Because finite differences involve binomial coefficients, many identities featuring Bernoulli numbers can be rewritten as binomial-weighted sums. This often yields alternate proofs of the Faulhaber polynomial formula.

5.3 Bernoulli polynomials

Bernoulli polynomials generalize Bernoulli numbers by introducing a parameter \(x\).

5.3.1 Definition via generating function

Bernoulli polynomials \(B_n(x)\) are defined by \[ \frac{t e^{xt}}{e^t-1}=\sum_{n=0}^{\infty} B_n(x)\frac{t^n}{n!}. \] Setting \(x=0\) recovers the Bernoulli numbers.

5.3.2 Relationship \(B_n = B_n(0)\)

From the definition, \[ B_n=B_n(0). \] This connects the parameterized polynomials to the original sequence.

5.3.3 Difference properties and shift identities

Bernoulli polynomials satisfy a key difference relation: \[ B_n(x+1)-B_n(x)=n x^{n-1}. \] Such identities are central in discrete-to-continuous translations and provide another route to Faulhaber-type formulas.

6 Number-theoretic connections (non-controversial overview)

Bernoulli numbers possess arithmetic structure, particularly in the denominators and in congruence patterns studied in number theory.

6.1 Von Staudt–Clausen theorem (fractional parts)

A fundamental theorem describes which primes occur in the denominators of Bernoulli numbers.

6.1.1 Denominator structure of Bernoulli numbers

For \(n\ge 1\), the theorem implies that the fractional part of \(B_{2n}\) can be expressed as a sum of reciprocals of certain primes. Consequently, the denominator of \(B_{2n}\) is controlled by a specific finite set of primes.

6.1.2 Interpretation in terms of primes

The primes involved are selected by a condition depending on divisibility relations tied to the index. Thus the theorem links the “shape” of denominators of \(B_{2n}\) to prime arithmetic constraints.

6.2 Kummer-type congruences (high-level)

Bernoulli numbers also satisfy congruences modulo primes, especially when viewed through \(p\)-adic lenses.

6.2.1 Congruence patterns modulo primes

Kummer-type results describe how certain combinations of Bernoulli numbers behave modulo powers of primes. Such congruences often reflect deeper regularities in arithmetic sequences.

6.2.2 Implications for arithmetic properties

At a high level, these congruences constrain possible denominators and residues of Bernoulli numbers and help explain why they show up in arithmetic formulas beyond classical real-variable analysis.

7 Historical context and notation

Bernoulli numbers emerged from early attempts to systematize power sums, then gained broader importance as analysis and analytic continuation developed.

7.1 Bernoulli and early motivations

Jacob Bernoulli studied expressions for sums of consecutive integer powers and their relationships to polynomial expansions. His work helped formalize coefficients now recognized as Bernoulli numbers.

7.2 Development of standard modern conventions

As analytic methods matured, the generating function viewpoint became dominant. The choice of defining \(\frac{x}{e^x-1}\) fixes a convention for \(B_1\) and standardizes many identities used in calculus and number theory.

7.3 Notational variants and how to translate between them

Different authors may swap the sign of \(B_1\) or use shifted generating functions for related polynomials. Translation typically involves comparing the generating function at low degree and then adjusting the corresponding term(s) so that both conventions align.

8 Exercises and worked examples

The following examples illustrate practical computation and typical uses in power sums and summation approximations.

8.1 Computing small Bernoulli numbers by hand

Working to a low order requires expanding \(\frac{x}{e^x-1}\) or applying a simple recursion.

8.1.1 Example using coefficient extraction

Start with \[ \frac{x}{e^x-1}=\frac{x}{x+\frac{x^2}{2}+\frac{x^3}{6}+\frac{x^4}{24}+\cdots}. \] Factor out \(x\): \[ \frac{x}{e^x-1}=\frac{1}{1+\frac{x}{2}+\frac{x^2}{6}+\frac{x^3}{24}+\cdots}. \] Now invert the series \(1+u\) using \(1/(1+u)=1-u+u^2-u^3+\cdots\), keeping terms up to the desired order. The coefficients obtained match \[ 1-\frac{x}{2}+\frac{x^2}{12}-\frac{x^4}{720}+\cdots, \] so \(B_0=1\), \(B_1=-\frac12\), \(B_2=\frac16\), \(B_3=0\), \(B_4=-\frac1{30}\).

8.1.2 Example using a recurrence

Use the triangular identity \[ \sum_{j=0}^{k}\binom{k+1}{j}B_j=0 \quad (k\ge 1), \qquad B_0=1. \] For \(k=1\): \[ \binom{2}{0}B_0+\binom{2}{1}B_1=0 \Rightarrow 1\cdot 1+2B_1=0 \Rightarrow B_1=-\frac12. \] For \(k=2\): \[ \binom{3}{0}B_0+\binom{3}{1}B_1+\binom{3}{2}B_2=0 \Rightarrow 1+3\left(-\frac12\right)+3B_2=0 \Rightarrow B_2=\frac16. \] For \(k=3\), the relation yields \(B_3=0\), consistent with the parity property.

8.2 Using Bernoulli numbers in a power-sum formula

Take \(k=2\). The formula \[ \sum_{m=1}^{n} m^2=\frac{1}{3}\sum_{j=0}^{2}\binom{3}{j}B_j\,n^{3-j} \] gives \[ \sum_{m=1}^{n} m^2=\frac13\left(\binom{3}{0}B_0n^3+\binom{3}{1}B_1n^2+\binom{3}{2}B_2n\right). \] Substitute \(B_0=1\), \(B_1=-\frac12\), \(B_2=\frac16\): \[ \sum_{m=1}^{n} m^2=\frac13\left(n^3+3\left(-\frac12\right)n^2+3\left(\frac16\right)n\right) =\frac13\left(n^3-\frac{3}{2}n^2+\frac12 n\right) =\frac{n(n+1)(2n+1)}{6}. \]

8.3 Applying Euler–Maclaurin to a simple sum

Consider \(f(x)=x\) on \([1,n]\). The exact sum satisfies \[ \sum_{m=1}^{n} m=\int_{1}^{n} x\,dx+\frac{f(n)+f(1)}{2}. \] Euler–Maclaurin explains why this works: the integral plus the endpoint average corresponds to truncating the asymptotic correction series at the first nontrivial boundary term. In this case, higher Bernoulli corrections vanish because the required derivatives are constant or zero, illustrating how Bernoulli numbers control the refinement in more complex (nonlinear) settings.