Skip to content
Topos Uranos

Improper integrals and convergence tests

Summary

The Riemann integral requires a closed and bounded interval and a function bounded on it. This lesson extends it, by passing to the limit, to the two cases in which one of those requirements fails: an unbounded interval, such as [1,∞)\ico{1}{\infty}, and a function unbounded near an endpoint, such as 1x\frac{1}{\sqrt{x}} near 00. The improper integral is the limit of the integrals over the closed intervals that approach the endpoint; if it exists, the integral converges. It is proved that it inherits linearity, additivity and monotonicity, that the point at which an integral improper at both endpoints is split does not matter, and it is decided for which exponents the integrals of powers converge. As in the unit on sequences, the aim is to know whether the limit exists without computing it: the Cauchy criterion characterizes it; for nonnegative functions, it suffices that the integrals be bounded, and from this come the comparison test and the limit comparison test; and absolute convergence implies convergence. With these tools, unbounded regions and curves are measured, and the integral of ln⁡sin⁡x\ln \sin x on [0,π2]\icc{0}{\frac{\pi}{2}} that the unit on the definite integral left pending is computed.

Learning objectives

  1. Define the improper integral on an unbounded interval and that of a function unbounded near an endpoint, and prove that it inherits the linearity, additivity and monotonicity of the definite integral.
  2. Decide for which exponents the integrals of powers converge, and prove the Cauchy criterion for improper integrals.
  3. Prove and use the comparison test and the limit comparison test for nonnegative functions, and the convergence of absolutely convergent integrals.
  4. Compute improper integrals with the techniques of the definite integral, distinguish them from the principal value, and use them to measure unbounded regions and curves.

The integral as a limit

The lesson on the Riemann integral defined the integral only for functions bounded on a closed and bounded interval, and its solved problem “The Riemann sums of an unbounded function” showed that the sums of an unbounded function have no limit. That lesson announced that the integral would be extended through limits of integrals over intervals where the function is bounded. There are two situations: the interval is unbounded, such as [0,∞)\ico{0}{\infty}, or the function is unbounded near an endpoint, such as 1x\frac{1}{\sqrt{x}} on (0,1]\ioc{0}{1}. In both, the function is integrable on the closed intervals that approach the endpoint, and a single letter, β\beta, a real number or the symbol ∞\infty, makes it possible to treat them together.

DefinitionImproper integral

Let aa be a real number, β\beta a real number greater than aa or the symbol ∞\infty, and ff a function defined on [a,β)\ico{a}{\beta} and integrable on [a,b]\icc{a}{b} for every bb with a<b<βa < b < \beta. One writes b→βb \to \beta to mean b→∞b \to \infty if β=∞\beta = \infty, and b→β−b \to \beta^- if β\beta is real. If the following limit exists and is a real number, the improper integral of ff on [a,β)\ico{a}{\beta} is said to converge, and its value is that limit:

∫aβf(x) dx=lim⁡b→β∫abf(x) dx\int_a^{\beta} f(x) \, \mathrm{d}x = \lim_{b \to \beta} \int_a^b f(x) \, \mathrm{d}x

If the limit does not exist, the integral diverges; if it is ∞\infty or −∞-\infty, the integral is said to diverge to ∞\infty or to −∞-\infty, and one writes ∫aβf=∞\int_a^{\beta} f = \infty or ∫aβf=−∞\int_a^{\beta} f = -\infty. Likewise, if bb is real, α\alpha is a real number less than bb or the symbol −∞-\infty, and ff is defined on (α,b]\ioc{\alpha}{b} and integrable on [c,b]\icc{c}{b} for every cc with α<c<b\alpha < c < b, one writes c→αc \to \alpha to mean c→−∞c \to -\infty or c→α+c \to \alpha^+, and the improper integral of ff on (α,b]\ioc{\alpha}{b} is ∫αbf(x) dx=lim⁡c→α∫cbf(x) dx\int_{\alpha}^{b} f(x) \, \mathrm{d}x = \lim_{c \to \alpha} \int_c^b f(x) \, \mathrm{d}x, with the same terminology.

For example, by Barrow’s rule for continuous functions, from the lesson on the fundamental theorem of calculus, ∫0be−x dx=1−e−b\int_0^b e^{-x} \, \mathrm{d}x = 1 - e^{-b}, which tends to 11 as b→∞b \to \infty; and ∫c1dxx=2−2c\int_c^1 \frac{\mathrm{d}x}{\sqrt{x}} = 2 - 2\sqrt{c}, which tends to 22 as c→0+c \to 0^+. Both integrals converge: ∫0∞e−x dx=1\int_0^{\infty} e^{-x} \, \mathrm{d}x = 1 and ∫01dxx=2\int_0^1 \frac{\mathrm{d}x}{\sqrt{x}} = 2. On the other hand, ∫0bcos⁡x dx=sin⁡b\int_0^b \cos x \, \mathrm{d}x = \sin b equals 00 at b=nπb = n\pi and 11 at b=2nπ+π2b = 2n\pi + \frac{\pi}{2}, and by the corollary on criteria for the nonexistence of the limit, from the lesson on sequences, functions and uniform continuity, it has no limit at infinity: ∫0∞cos⁡x dx\int_0^{\infty} \cos x \, \mathrm{d}x diverges, and not to ∞\infty or to −∞-\infty.

Limits at ∞\infty and from the left at a point are defined in Calculus I with different responses, a bound MM and a δ\delta; the following lemma writes them with a single response, a point cc beyond which the condition holds, and so each proof in the lesson works for both cases.

LemmaCommon form of the limit at the endpoint

Let aa and β\beta be as in the definition of the improper integral, GG a function defined on [a,β)\ico{a}{\beta}, and LL a real number. Then lim⁡b→βG(b)=L\lim_{b \to \beta} G(b) = L if and only if for every ε>0\varepsilon > 0 there exists a cc in [a,β)\ico{a}{\beta} such that ∣G(b)−L∣<ε\abs{G(b) - L} < \varepsilon for every bb in (c,β)\ioo{c}{\beta}. Likewise, lim⁡b→βG(b)=∞\lim_{b \to \beta} G(b) = \infty if and only if for every real NN there exists a cc in [a,β)\ico{a}{\beta} such that G(b)>NG(b) > N for every bb in (c,β)\ioo{c}{\beta}. The bound MM of the definition at infinity and the endpoint β−δ\beta - \delta of the definition from the left play the role of cc.

Proof

  1. β=∞→(b>max⁡(a,M)→b>M),c=max⁡(a,M)\beta = \infty \rightarrow \left(b > \max(a, M) \rightarrow b > M\right), \qquad \resaltar{c = \max(a, M)}

    Let β=∞\beta = \infty. The domain [a,∞)\ico{a}{\infty} is not bounded above, and the definition of the limit as the variable tends to infinity, from the lesson on the limit at infinity, requires for each ε>0\varepsilon > 0 a real number MM such that ∣G(b)−L∣<ε\abs{G(b) - L} < \varepsilon for every bb in [a,∞)\ico{a}{\infty} with b>Mb > M. Given that MM, the point c=max⁡(a,M)c = \max(a, M) lies in [a,∞)\ico{a}{\infty}, and every b>cb > c satisfies b>Mb > M; conversely, given cc, M=cM = c works. For the limit ∞\infty only the condition changes, as in the variant of the definition of infinite limits, from the lesson on infinite limits.

  2. β∈R→(c<b<β→β−δ<b<β),c=max⁡(a,β−δ)\beta \in \mathbb{R} \rightarrow \left(c < b < \beta \rightarrow \beta - \delta < b < \beta\right), \qquad \resaltar{c = \max(a, \beta - \delta)}

    Let β\beta be real. Since a<βa < \beta, the point β\beta is an accumulation point of [a,β)\ico{a}{\beta}, and by the definition of one-sided limits, from the lesson on one-sided limits, the limit from the left requires for each ε>0\varepsilon > 0 a δ>0\delta > 0 such that ∣G(b)−L∣<ε\abs{G(b) - L} < \varepsilon for every bb in [a,β)\ico{a}{\beta} with β−δ<b\beta - \delta < b. Given that δ\delta, the point c=max⁡(a,β−δ)c = \max(a, \beta - \delta) lies in [a,β)\ico{a}{\beta}, and every bb in (c,β)\ioo{c}{\beta} satisfies β−δ<b<β\beta - \delta < b < \beta; conversely, given cc, δ=β−c\delta = \beta - c works, which is positive. For the limit ∞\infty, likewise.

The notation does not conflict with that of the definite integral: if β\beta is real and ff is integrable on all of [a,β]\icc{a}{\beta}, the improper integral converges to the same number. The improper integral says something new only when β=∞\beta = \infty or when ff is not integrable on [a,β]\icc{a}{\beta}, for example because it is unbounded near β\beta.

PropositionConsistency with the Riemann integral

Let a<βa < \beta be real numbers and ff a function integrable on [a,β]\icc{a}{\beta}. Then the improper integral of ff on [a,β)\ico{a}{\beta} converges, and its value is the integral of ff on [a,β]\icc{a}{\beta}. The same holds at the left endpoint: the improper integral on (a,β]\ioc{a}{\beta} converges to the same integral. The difference between the two integrals is the integral over a short interval, where ff is bounded.

Proof

  1. ∣∫aβf−∫abf∣=∣∫bβf∣≤B (β−b)\abs{\int_a^{\beta} f - \int_a^b f} = \abs{\int_b^{\beta} f} \leq \resaltar{B \, (\beta - b)}

    The function ff is integrable, and therefore bounded, because the definition of an integrable function applies only to bounded functions: ∣f(x)∣≤B\abs{f(x)} \leq B on [a,β]\icc{a}{\beta}. For a<b<βa < b < \beta, ff is integrable on [a,b]\icc{a}{b} by the corollary on integrability on the subintervals, from the lesson on the properties of the definite integral, and by additivity with respect to the interval and the corollary on bounds for the integral, from the same lesson:

  2. c<b<β→∣∫aβf−∫abf∣≤B εB+1<εc < b < \beta \rightarrow \abs{\int_a^{\beta} f - \int_a^b f} \leq \frac{B \, \varepsilon}{B + 1} < \resaltar{\varepsilon}

    Given ε>0\varepsilon > 0, let c=max⁡(a,β−εB+1)c = \max\left(a, \beta - \frac{\varepsilon}{B + 1}\right), which lies in [a,β)\ico{a}{\beta} and depends only on ε\varepsilon. For every bb in (c,β)\ioo{c}{\beta}, the bound is less than ε\varepsilon, and by the lemma on the common form of the limit at the endpoint, the improper integral converges to ∫aβf\int_a^{\beta} f. At the left endpoint, for a<u<βa < u < \beta, the difference between ∫aβf\int_a^{\beta} f and ∫uβf\int_u^{\beta} f is ∫auf\int_a^u f, bounded by B (u−a)B \, (u - a), less than ε\varepsilon if u<a+εB+1u < a + \frac{\varepsilon}{B + 1}: by the definition of the limit from the right, with δ=εB+1\delta = \frac{\varepsilon}{B + 1}, the improper integral on (a,β]\ioc{a}{\beta} converges to the same number.

Everything proved for the right endpoint holds for the left one, because the reflection x↦−xx \mapsto -x interchanges them.

LemmaReflection of an improper integral

Let bb be real, α\alpha a real number less than bb or the symbol −∞-\infty, and ff a function defined on (α,b]\ioc{\alpha}{b} and integrable on [c,b]\icc{c}{b} for every cc with α<c<b\alpha < c < b. Let g(x)=f(−x)g(x) = f(-x) on [−b,−α)\ico{-b}{-\alpha}, where −α-\alpha is ∞\infty if α=−∞\alpha = -\infty. Then gg is integrable on [−b,u]\icc{-b}{u} for every uu with −b<u<−α-b < u < -\alpha, and ∫αbf\int_{\alpha}^{b} f converges if and only if ∫−b−αg\int_{-b}^{-\alpha} g converges, with the same value; the same holds for divergence to ∞\infty and to −∞-\infty. Reflecting a closed interval does not change the integral, and reflecting the variable does not change the limit.

Proof

  1. ∫−b−cf(−x) dx=∫cbf(x) dx,α<c<b\dato{ref}{\int_{-b}^{-c} f(-x) \, \mathrm{d}x = \int_c^b f(x) \, \mathrm{d}x}, \qquad \alpha < c < b

    For α<c<b\alpha < c < b, the lemma on translation and reflection of the integral, from the lesson on techniques for definite integrals, with 00 in place of its constant, says that gg is integrable on [−b,−c]\icc{-b}{-c} and gives the equality. Each uu with −b<u<−α-b < u < -\alpha is −c-c for c=−uc = -u.

  2. lim⁡c→α∫cbf=L⇔lim⁡u→−α∫−bug=L\lim_{c \to \alpha} \int_c^b f = L \Leftrightarrow \resaltar{\lim_{u \to -\alpha} \int_{-b}^{u} g = L}

    Let H(c)=∫cbfH(c) = \int_c^b f. If α\alpha is real, the translation and reflection proposition, from the lesson on one-sided limits, says that H(c)H(c) has limit LL as c→α+c \to \alpha^+ if and only if H(−u)H(-u) has it as u→(−α)−u \to (-\alpha)^-. If α=−∞\alpha = -\infty, the condition c<Mc < M of the definition of the limit at −∞-\infty is the condition u>−Mu > -M for u=−cu = -c. The conditions for infinite limits carry over likewise.