Skip to content
Topos Uranos

Sequences and series of functions

Summary

The preceding units studied sequences and series of numbers. This lesson studies sequences whose terms are functions defined on one and the same set, and series whose partial sums are functions. At each point, the values form a sequence of numbers, and if all those sequences converge, their limits define a function: this is pointwise convergence. But the pointwise limit of continuous functions may be discontinuous, like that of the powers xnx^n on [0,1]\icc{0}{1}, and the integral of the limit may fail to be the limit of the integrals. Uniform convergence, in which one and the same index serves for all points, just as in uniform continuity one and the same δ\delta serves for all of them, preserves continuity, allows passing to the limit under the integral on a closed and bounded interval and, if the derivatives satisfy it, allows differentiating the limit. The uniform Cauchy criterion decides uniform convergence without knowing the limit, and the Weierstrass M-test yields it for series whose terms are bounded by those of a convergent numerical series. Each theorem comes with a counterexample showing that its hypotheses are not superfluous.

Learning objectives

  1. Distinguish pointwise convergence from uniform convergence by the order of the quantifiers, and decide which one a sequence of functions has by means of the supremum of its distance to the limit.
  2. Prove the uniform Cauchy criterion and the Weierstrass M-test, and use them to show that a series of functions converges uniformly.
  3. Prove that the uniform limit of continuous functions is continuous, that one can integrate term by term on a closed and bounded interval, and under which conditions one can differentiate term by term.
  4. Construct counterexamples showing that none of those hypotheses can be omitted.

Pointwise convergence and uniform convergence

The lesson on sequences and their limit defined the limit of a sequence of numbers with an ε\varepsilon and a natural number NN, quantified in that order, and the lesson on sequences, functions and uniform continuity showed that, for continuity, moving the “for every point” behind the “there exists δ\delta” changes the concept. With a sequence of functions the same happens with the “there exists NN”. In what follows, AA is a nonempty set of real numbers and, for each natural number nn, fnf_n is a function defined on AA; one says that (fn)(f_n) is a sequence of functions on AA.

DefinitionPointwise convergence

Let (fn)(f_n) be a sequence of functions on AA and ff a function defined on AA. One says that (fn)(f_n) converges pointwise to ff on AA if, for each xx in AA, the sequence of numbers (fn(x))(f_n(x)) converges to f(x)f(x):

(∀x∈A)(∀ε>0)(∃N∈N)(∀n∈N)(n≥N→∣fn(x)−f(x)∣<ε)(\forall x \in A)(\forall \varepsilon > 0)(\exists N \in \mathbb{N})(\forall n \in \mathbb{N})\left(n \geq N \rightarrow \abs{f_n(x) - f(x)} < \varepsilon\right)

One then writes fn→ff_n \to f on AA, and ff is called the pointwise limit of the sequence.

By the uniqueness of the limit, from the theorem according to which the limit theorems hold for sequences, in the lesson on sequences and their limit, the pointwise limit, if it exists, is unique. In the definition, the natural number NN is chosen after xx and ε\varepsilon, and may depend on both: near some points, the sequence may take longer and longer to approach its limit.

DefinitionUniform convergence

Let (fn)(f_n) be a sequence of functions on AA and ff a function defined on AA. One says that (fn)(f_n) converges uniformly to ff on AA if for every ε>0\varepsilon > 0 there exists a natural number NN such that ∣fn(x)−f(x)∣<ε\abs{f_n(x) - f(x)} < \varepsilon for every natural number n≥Nn \geq N and every xx in AA:

(∀ε>0)(∃N∈N)(∀n∈N)(∀x∈A)(n≥N→∣fn(x)−f(x)∣<ε)(\forall \varepsilon > 0)(\exists N \in \mathbb{N})(\forall n \in \mathbb{N})(\forall x \in A)\left(n \geq N \rightarrow \abs{f_n(x) - f(x)} < \varepsilon\right)

One then writes “fn→ff_n \to f uniformly on AA”. If BB is a nonempty subset of AA, one says that (fn)(f_n) converges uniformly to ff on BB if the condition holds for all xx in BB.

Here NN depends only on ε\varepsilon and on the sequence, not on the point: the “for every xx” has moved behind the “there exists NN”, just as in the definition of a uniformly continuous function the “for all xx and yy” comes behind the “there exists δ\delta”. Uniform convergence implies pointwise convergence, with the same NN for each xx; and if (fn)(f_n) converges uniformly on AA, it also does so on every nonempty subset of AA, with the same NN. If AA has finitely many points, the two kinds of convergence coincide: it suffices to take the largest of the natural numbers NN of the individual points. In the graph, uniform convergence says that, from the index NN on, the graph of each fnf_n stays inside the strip between the graphs of f−εf - \varepsilon and of f+εf + \varepsilon, on all of AA.

PropositionUniform convergence via the supremum

Let (fn)(f_n) be a sequence of functions on AA and ff a function defined on AA. Then fn→ff_n \to f uniformly on AA if and only if there exists a natural number n0n_0 such that, for every n≥n0n \geq n_0, the set of the numbers ∣fn(x)−f(x)∣\abs{f_n(x) - f(x)}, with xx in AA, is bounded above, and their suprema tend to 00:

dn=sup⁡x∈A∣fn(x)−f(x)∣→0d_n = \sup_{x \in A} \abs{f_n(x) - f(x)} \to 0

The number dnd_n is the least bound of the vertical distance between the graphs of fnf_n and of ff; uniform convergence requires that this bound, which does not depend on the point, tend to 00.

Proof

  1. n≥n0→0≤dn≤1n \geq n_0 \rightarrow \resaltar{0 \leq d_n \leq 1}

    First direction. With ε=1\varepsilon = 1, the definition gives a natural number n0n_0 such that, for every n≥n0n \geq n_0, the number 11 is an upper bound of the numbers ∣fn(x)−f(x)∣\abs{f_n(x) - f(x)}. Since AA is not empty, the set of those numbers is not empty, and by the completeness axiom, from the lesson on number systems, it has a supremum dnd_n, which is not negative.

  2. n≥max⁡(N,n0)→∣dn−0∣=dn≤ε2<εn \geq \max(N, n_0) \rightarrow \abs{d_n - 0} = d_n \leq \frac{\varepsilon}{2} < \resaltar{\varepsilon}

    Given ε>0\varepsilon > 0, the definition, with ε2\frac{\varepsilon}{2} in place of ε\varepsilon, gives a natural number NN; for n≥Nn \geq N, the number ε2\frac{\varepsilon}{2} is an upper bound of the numbers ∣fn(x)−f(x)∣\abs{f_n(x) - f(x)}, and it is not smaller than their supremum. With the larger of NN and n0n_0, the sequence (dn)(d_n), defined from n0n_0 on, satisfies the definition of the limit 00.

  3. n≥N∧x∈A→∣fn(x)−f(x)∣≤dn<εn \geq N \wedge x \in A \rightarrow \abs{f_n(x) - f(x)} \leq d_n < \resaltar{\varepsilon}

    Second direction. If dn→0d_n \to 0, given ε>0\varepsilon > 0 there is a natural number N≥n0N \geq n_0, which depends only on ε\varepsilon, such that dn<εd_n < \varepsilon for every n≥Nn \geq N. Since dnd_n is an upper bound of the numbers ∣fn(x)−f(x)∣\abs{f_n(x) - f(x)}, the definition of uniform convergence holds with that NN.

To refute uniform convergence it suffices to negate the definition: find a c>0c > 0 such that, for every natural number NN, there are an n≥Nn \geq N and a point xx of AA with ∣fn(x)−f(x)∣≥c\abs{f_n(x) - f(x)} \geq c. The simplest example is that of the powers.

PropositionThe powers on the unit interval

Let fn(x)=xnf_n(x) = x^n on [0,1]\icc{0}{1}. Then (fn)(f_n) converges pointwise on [0,1]\icc{0}{1} to the function ff that equals 00 on [0,1)\ico{0}{1} and 11 at x=1x = 1. The convergence is not uniform on [0,1)\ico{0}{1} nor on [0,1]\icc{0}{1}, and it is uniform on [0,r]\icc{0}{r} for every rr with 0≤r<10 \leq r < 1. Away from 11, the powers are flattened against the axis at the speed of rnr^n; near 11, each one climbs up to 11, and there always remains a point where it is farther than 12\frac{1}{2} from the limit. The figure shows xnx^n for n=1,2,4,8,16,32n = 1, 2, 4, 8, 16, 32 and the limit, which jumps at 11; the strip of radius ε\varepsilon around the limit on [0,r]\icc{0}{r}, which contains the graphs from some index on; and, on each graph, the point of [0,1)\ico{0}{1} where it equals 34\frac{3}{4}, which approaches 11 without leaving [0,1)\ico{0}{1}.

Proof

  1. 0≤x<1→xn→0,1n=1→10 \leq x < 1 \rightarrow x^n \to 0, \qquad \resaltar{1^n = 1 \to 1}

    By the theorem on the geometric sequence, from the lesson on sequences and their limit, if 0≤x<10 \leq x < 1, then xn→0x^n \to 0; and 1n=11^n = 1 for every nn.

  2. n≥N∧0≤x≤r→∣xn−0∣≤rn<εn \geq N \wedge 0 \leq x \leq r \rightarrow \abs{x^n - 0} \leq r^n < \resaltar{\varepsilon}

    Let 0≤r<10 \leq r < 1. If 0≤x≤r0 \leq x \leq r, multiplying nn times inequalities between nonnegative numbers gives 0≤xn≤rn0 \leq x^n \leq r^n: the number rnr^n bounds the distance to the limit on all of [0,r]\icc{0}{r} and does not depend on xx. Since rn→0r^n \to 0, given ε>0\varepsilon > 0 there is a natural number NN, which depends only on ε\varepsilon and on rr, such that rn<εr^n < \varepsilon for n≥Nn \geq N, and it serves for all xx in [0,r]\icc{0}{r}.

  3. u∈[0,1)∧∣fn(u)−f(u)∣=un>12u \in \ico{0}{1} \wedge \abs{f_n(u) - f(u)} = u^n > \resaltar{\frac{1}{2}}

    Let nn be a natural number. By the corollary on known continuous functions, from the lesson on limit and continuity, the polynomial xnx^n is continuous at 11, where it equals 11; with ε=12\varepsilon = \frac{1}{2} there is a δ>0\delta > 0 such that xn>12x^n > \frac{1}{2} for every xx in [0,1]\icc{0}{1} with ∣x−1∣<δ\abs{x - 1} < \delta. The point uu, the larger of 00 and 1−δ21 - \frac{\delta}{2}, lies in [0,1)\ico{0}{1}, because δ>0\delta > 0, and satisfies ∣u−1∣<δ\abs{u - 1} < \delta. With c=12c = \frac{1}{2}, no NN satisfies the definition of uniform convergence on [0,1)\ico{0}{1}, nor on [0,1]\icc{0}{1}, which contains it.

The limit is discontinuous at 11, although each xnx^n is continuous: the theorem on continuity, further below, will show that this suffices to know that the convergence is not uniform on [0,1]\icc{0}{1}. On [0,1)\ico{0}{1}, by contrast, the limit is the zero function, which is continuous, and the convergence is not uniform either; continuity of the limit does not suffice for uniform convergence, as the solved problem “A bump approaching the origin” also shows.