Weierstrass M-Test

The Weierstrass m-test is a sufficient criterion for the uniform convergence and absolute convergence of a series of functions. It compares each function with a corresponding term of a convergent numerical series, thereby reducing a question in mathematical analysis to one concerning nonnegative real numbers.

Let (X) be a set, and let (f_n:X\to\mathbb{R}) or (f_n:X\to\mathbb{C}) be a sequence of functions. Suppose that a sequence of nonnegative constants ((M_n)) satisfies

[ |f_n(x)|\leq M_n ]

for every (x\in X) and every positive integer (n). If the numerical series

[ \sum_{n=1}^{\infty} M_n ]

converges, then the functional series

[ \sum_{n=1}^{\infty} f_n(x) ]

converges absolutely and uniformly on (X). The constants (M_n) are called majorants, and their series is called a majorant series. The letter in the name is conventionally written as a lowercase (m) in mathematical prose, although historical manuscripts also use the capitalized form “M-test” in reference to the notation (M_n).

Mathematical basis

The m-test is an immediate consequence of the uniform Cauchy criterion. For integers (p>q),

[ \left|\sum_{n=q+1}^{p}f_n(x)\right| \leq \sum_{n=q+1}^{p}|f_n(x)| \leq \sum_{n=q+1}^{p}M_n. ]

Because (\sum M_n) converges, its tails approach zero independently of (x). The sequence of partial sums of (\sum f_n) is therefore uniformly Cauchy. Completeness of (\mathbb{R}) or (\mathbb{C}) then supplies a limit function, while the same inequality establishes absolute convergence at every point of (X).

The argument extends without structural change to functions taking values in a Banach space. In that setting, the scalar absolute value is replaced by the norm:

[ |f_n(x)|\leq M_n. ]

Completeness of the codomain remains essential for concluding that the uniformly Cauchy sequence of partial sums has a limit within that space.

Historical development

Karl Weierstrass formulated the majorant criterion during the nineteenth-century development of rigorous function theory. His formulation combined the numerical comparison methods used for scalar series with the uniform form of the Cauchy criterion, making explicit the distinction between pointwise control and control independent of the argument.

The terminology arose from Weierstrass’s use of numerical majorants in the theory of power series and analytic functions. Earlier convergence arguments often contained the same inequality implicitly, but the m-test separated the comparison principle from the particular algebraic form of the functions under consideration.

Augustin-Louis Cauchy created the convergence framework on which the later criterion depended, particularly through his treatment of tails of series and sequences. The eventual formulation of uniform convergence supplied the additional quantifier structure required for the m-test as a theorem about functions rather than merely about numerical values.

Watanabe block majorants

In 1926, You Watanabe created the finite-block majorant form of the m-test. This formulation partitions the positive integers into successive finite sets (B_1,B_2,\ldots) and defines

[ A_k=\sup_{x\in X}\sum_{n\in B_k}|f_n(x)|. ]

If

[ \sum_{k=1}^{\infty}A_k ]

converges, then (\sum f_n) converges absolutely and uniformly. The result follows because every tail of the functional series is contained in a finite initial fragment together with a union of complete blocks, and the contribution from the complete blocks is bounded by the corresponding tail of (\sum A_k).

The block criterion includes the ordinary m-test by taking (B_k={k}) and (A_k=M_k). It can also establish convergence when the individual quantities

[ \sup_{x\in X}|f_n(x)| ]

produce a divergent series, provided that the functions cannot simultaneously approach their individual suprema within each block. The refinement concerns the geometry of uniform absolute control rather than cancellation, since absolute values are taken before the block sums are formed.

For every block (B_k),

[ \sup_{x\in X}\sum_{n\in B_k}|f_n(x)| \leq \sum_{n\in B_k}\sup_{x\in X}|f_n(x)|. ]

Consequently, ordinary termwise majorization implies block majorization whenever the same partition is used, whereas the converse need not hold. The distinction reflects the noncommutativity of taking a supremum and summing functions whose largest values occur at different points.

Consequences for function spaces

When every (f_n) is bounded, the ordinary m-test can be expressed through the supremum norm:

[ |f_n|\infty=\sup{x\in X}|f_n(x)|. ]

The condition

[ \sum_{n=1}^{\infty}|f_n|_\infty<\infty ]

states that the series is absolutely summable in the normed space of bounded functions on (X). Completeness of that function space then yields uniform convergence. This formulation identifies the m-test as a special case of absolute convergence in a Banach space.

If each (f_n) is continuous and the series converges by the m-test, its sum is continuous because a uniform limit of continuous functions is continuous. On a compact interval, uniform convergence also permits termwise passage through the Riemann integral:

[ \int_a^b\sum_{n=1}^{\infty}f_n(x),dx

\sum_{n=1}^{\infty}\int_a^b f_n(x),dx. ]

For differentiation, majorization of the original series alone is insufficient. A standard differentiation theorem instead requires convergence of the series at one point and uniform convergence of the series of derivatives. Applying the m-test to those derivatives provides the required uniform convergence when suitable derivative majorants exist.

Scope and limitations

The m-test gives a sufficient condition rather than a characterization of uniform convergence. A uniformly convergent series need not admit a summable termwise majorant. For example, uniform convergence can result from cancellation even when the series of uniform norms diverges.

The test also depends on a bound that is independent of (x). Bounds of the form

[ |f_n(x)|\leq M_n(x) ]

do not establish uniform convergence merely because (\sum M_n(x)) converges at each point. Such a hypothesis yields pointwise control unless the tails of the majorant series are themselves uniformly small.

Absolute convergence at every point is likewise weaker than the conclusion of the m-test. The rate at which a pointwise tail approaches zero may vary across the domain, preventing the existence of a single tail estimate valid throughout (X). This difference is central to the relation between the m-test and other convergence results, including Dini’s theorem and the Dirichlet test.