Skip to content
Topos Uranos

Sequences, functions and uniform continuity

Summary

This lesson relates sequences to functions. It proves that the limit of a function, at an accumulation point of its domain or at infinity, and its continuity at a point are characterized, in both directions, by the sequences that tend to the point, and it derives from this criteria for proving that a limit does not exist. It then defines uniform continuity, in which one and the same response works for all the points of the domain: Lipschitz functions and functions with bounded derivative have it, and neither the square on the reals nor the reciprocal and the sine of the reciprocal on the open interval with endpoints 00 and 11 have it. It characterizes it by pairs of sequences and proves, with the Bolzano-Weierstrass theorem, the Heine-Cantor theorem: every function continuous on a closed and bounded interval is uniformly continuous, which is what the definite integral will need. Finally, it proves the Stolz-Cesàro lemma, a version of L'Hôpital's rule for sequences, and with it and the limits of Calculus I it orders by growth the logarithm, the powers, the exponentials, the factorial and nnn^n; as a corollary, the mean of the first terms of a convergent sequence tends to the same limit.

Learning objectives

  1. Prove the sequential characterization of the limit and of continuity, in both directions, and use it to prove that a limit does not exist.
  2. Define uniform continuity, distinguish it from continuity at each point, recognize it through the Lipschitz condition or through a bound on the derivative, and refute it with two sequences.
  3. Prove the Heine-Cantor theorem with the Bolzano-Weierstrass theorem and show that none of its hypotheses is superfluous.
  4. Prove the Stolz-Cesàro lemma and use it, with the limits of Calculus I, to compare the growth of sequences and to compute limits of means.

The limit of a function via sequences

The lesson on sequences and their limit passed from the limit of a function at infinity to that of the sequence of its values at the natural numbers, and showed with sin⁡(πx)\sin(\pi x) that the converse is false: the natural numbers are too few to decide the limit of the function. The lesson on monotonic sequences and the number e proved that a function continuous at a point preserves the limit of every sequence that tends to it. If instead of a single sequence one considers all those that tend to the point, the converse holds, and the limit and continuity are characterized by sequences; in some of the literature, the characterization is known as Heine's criterion.

TheoremSequential characterization of the limit

Let ff be a function defined on a set AA and LL a real number. First statement: if aa is an accumulation point of AA, then lim⁡x→af(x)=L\lim_{x \to a} f(x) = L if and only if f(xn)→Lf(x_n) \to L for every sequence (xn)(x_n) of elements of AA different from aa that tends to aa. Second statement: if AA is not bounded above, then lim⁡x→∞f(x)=L\lim_{x \to \infty} f(x) = L if and only if f(xn)→Lf(x_n) \to L for every sequence (xn)(x_n) of elements of AA that tends to infinity. In one direction, the response of the limit is used as a challenge for the sequence; in the other, by contraposition, the points that the negation of the limit provides for each response 1n\frac{1}{n}, or nn, form a sequence whose images do not tend to LL.

Proof

  1. n≥N→xn∈A∧0<∣xn−a∣<δ→∣f(xn)−L∣<εn \geq N \rightarrow x_n \in A \wedge 0 < \abs{x_n - a} < \delta \rightarrow \resaltar{\abs{f(x_n) - L} < \varepsilon}

    First statement, first direction. Let ε>0\varepsilon > 0, and let δ\delta be the response of the definition of the limit of a function at a point, from the lesson on the limit of functions. With the challenge δ\delta, the convergence xn→ax_n \to a gives a natural number NN, which depends on ε\varepsilon through δ\delta. For n≥Nn \geq N, the term xnx_n belongs to AA, is different from aa and lies at a distance from aa less than δ\delta.

  2. xn∈A,0<∣xn−a∣<1n,∣f(xn)−L∣≥ε\dato{xn}{x_n \in A, \qquad 0 < \abs{x_n - a} < \frac{1}{n}, \qquad \abs{f(x_n) - L} \geq \varepsilon}

    First statement, second direction, by contraposition. If LL is not the limit, the proposition on the negation of the definition of limit, from the same lesson, gives an ε>0\varepsilon > 0 such that, for every δ>0\delta > 0, some xx in AA satisfies 0<∣x−a∣<δ0 < \abs{x - a} < \delta and ∣f(x)−L∣≥ε\abs{f(x) - L} \geq \varepsilon. With δ=1n\delta = \frac{1}{n} one chooses one, xnx_n, for each natural number nn (one choice for each nn, as is customary in analysis).

  3. n≥N→∣xn−a∣<1n≤1N<η,∣f(xN)−L∣≥εn \geq N \rightarrow \abs{x_n - a} < \frac{1}{n} \leq \frac{1}{N} < \eta, \qquad \resaltar{\abs{f(x_N) - L} \geq \varepsilon}

    The terms of (xn)(x_n) belong to AA and are different from aa, and the sequence tends to aa: given η>0\eta > 0, the Archimedean property lemma, from the lesson on the decimal expansion of rational numbers, gives a natural number N>1ηN > \frac{1}{\eta}. But f(xn)f(x_n) does not tend to LL: for the challenge ε\varepsilon, no natural number NN works, because the term with index NN already lies at a distance from LL of at least ε\varepsilon. The condition on the sequences fails.

  4. n≥N→xn∈A∧xn>M→∣f(xn)−L∣<εn \geq N \rightarrow x_n \in A \wedge x_n > M \rightarrow \resaltar{\abs{f(x_n) - L} < \varepsilon}

    Second statement, first direction. Let ε>0\varepsilon > 0, and let MM be the real response of the definition of the limit as the variable tends to infinity, from the lesson on the limit at infinity. With the bound K=MK = M, the definition of the sequences that tend to infinity, from the lesson on sequences and their limit, gives a natural number NN.

  5. xn∈A,xn>n,∣f(xn)−L∣≥ε,xn→∞x_n \in A, \qquad x_n > n, \qquad \abs{f(x_n) - L} \geq \varepsilon, \qquad \resaltar{x_n \to \infty}

    Second statement, second direction. Negating the definition of the limit at infinity, quantifier by quantifier, one obtains an ε>0\varepsilon > 0 such that, for every real MM, some xx in AA satisfies x>Mx > M and ∣f(x)−L∣≥ε\abs{f(x) - L} \geq \varepsilon. With M=nM = n one chooses one, xnx_n, for each natural number nn (one choice for each nn, as is customary in analysis). Since xn>nx_n > n and n→∞n \to \infty, the proposition on comparison and reciprocals with infinite limits, from the lesson on sequences and their limit, gives xn→∞x_n \to \infty; and f(xn)f(x_n) does not tend to LL, for the same reason as in the first statement.

In the first statement, the terms must be different from aa, because the limit does not look at the value at aa: the function equal to 00 at every x≠0x \neq 0 and to 11 at 00 tends to 00 as xx tends to 00, but the constant sequence xn=0x_n = 0 gives f(xn)=1f(x_n) = 1. The condition does not hold vacuously: the construction of the second direction of the first statement, with any δ=1n\delta = \frac{1}{n}, shows that if aa is an accumulation point of AA, there are sequences of elements of AA different from aa that tend to aa. For continuity, the point belongs to the domain and is no longer excluded, as in the definition of Calculus I.

TheoremSequential characterization of continuity

Let ff be a function defined on a set AA and aa an element of AA. Then ff is continuous at aa if and only if f(xn)→f(a)f(x_n) \to f(a) for every sequence (xn)(x_n) of elements of AA that tends to aa. One direction is already proved; the other repeats the contraposition of the preceding theorem, without excluding the point.

Proof

  1. xn∈A,xn→a→f(xn)→f(a)x_n \in A, \qquad x_n \to a \rightarrow \resaltar{f(x_n) \to f(a)}

    If ff is continuous at aa, the proposition according to which a continuous function preserves the limit of a sequence, from the lesson on monotonic sequences, with B=AB = A and b=ab = a, gives the conclusion.

  2. xn∈A,∣xn−a∣<1n,∣f(xn)−f(a)∣≥εx_n \in A, \qquad \abs{x_n - a} < \frac{1}{n}, \qquad \resaltar{\abs{f(x_n) - f(a)} \geq \varepsilon}

    If ff is not continuous at aa, negating the definition of a function continuous at a point, from the lesson on limit and continuity, one obtains an ε>0\varepsilon > 0 such that, for every δ>0\delta > 0, some xx in AA satisfies ∣x−a∣<δ\abs{x - a} < \delta and ∣f(x)−f(a)∣≥ε\abs{f(x) - f(a)} \geq \varepsilon. With δ=1n\delta = \frac{1}{n} one chooses one, xnx_n, for each nn (one choice for each nn, as is customary in analysis); as in the first statement of the preceding theorem, xn→ax_n \to a, and f(xn)f(x_n) does not tend to f(a)f(a).

This fulfills what the lesson on monotonic sequences announced: if f(xn)→f(a)f(x_n) \to f(a) for every sequence of terms of the domain that tends to aa, then ff is continuous at aa. Its most frequent use is the following.

CorollaryCriteria for the nonexistence of the limit

Let ff be a function defined on a set AA and aa an accumulation point of AA. If there exist two sequences (xn)(x_n) and (yn)(y_n) of elements of AA different from aa that tend to aa, with f(xn)→L1f(x_n) \to L_1, f(yn)→L2f(y_n) \to L_2 and L1≠L2L_1 \neq L_2, or a single sequence (xn)(x_n) of such elements for which f(xn)f(x_n) does not converge, then ff has no limit as xx tends to aa. The same holds at infinity, if AA is not bounded above, with sequences of elements of AA that tend to infinity. If there were a limit, all the sequences of images would tend to it.

Proof

  1. f(xn)→L∧f(yn)→L→L1=L2f(x_n) \to L \wedge f(y_n) \to L \rightarrow \resaltar{L_1 = L_2}

    If lim⁡x→af(x)=L\lim_{x \to a} f(x) = L, the sequential characterization of the limit gives f(xn)→Lf(x_n) \to L and f(yn)→Lf(y_n) \to L, and by the uniqueness of the limit, from the theorem according to which the limit theorems hold for sequences, L1=L=L2L_1 = L = L_2, contrary to the hypothesis; in the second case, f(xn)f(x_n) would converge. At infinity one uses the second statement of the characterization.

For example, sin⁡1x\sin \frac{1}{x} has no limit at 00: the sequences xn=1π2+2nπx_n = \frac{1}{\frac{\pi}{2} + 2n\pi} and yn=13π2+2nπy_n = \frac{1}{\frac{3\pi}{2} + 2n\pi} tend to 00 without vanishing, and the function equals 11 at the terms of the first and −1-1 at those of the second. At infinity, cos⁡x\cos x has no limit, because it equals 11 at xn=2nπx_n = 2n\pi and −1-1 at yn=(2n+1)πy_n = (2n + 1)\pi; nor does sin⁡(πx)\sin(\pi x), because it equals 00 at yn=ny_n = n and 11 at xn=2n+12x_n = 2n + \frac{1}{2}, which is the argument of the lesson on sequences and their limit, written with two sequences.