Skip to content
Topos Uranos

Improper integrals: synthesis and challenges

Summary

This lesson closes the unit on improper integrals. It gathers into a single deductive chain what the four content lessons proved: the integral of a function on an unbounded interval, or of a function unbounded near an endpoint, is defined as the limit of Riemann integrals over closed intervals that approach the endpoint; the Cauchy criterion and, for nonnegative functions, the completeness axiom decide whether that limit exists without computing it, and from them follow comparison and absolute convergence; the second mean value theorem for integrals gives Dirichlet’s and Abel’s tests, which govern the integrals that converge by cancellation; uniform continuity on a rectangle allows an integral to be differentiated with respect to a parameter; and with these tools one computes the integral of sin⁡xx\frac{\sin x}{x} on the half-line, the Gaussian integral and Stirling’s formula. It explains how the unit rests on Calculus I and on the four preceding units, which of their promises it fulfills and what it leaves to the units that follow, in particular to numerical series. It ends with an exit assessment, parallel to the entry assessment, and with ten challenges, harder than the problems of the lessons, each of which combines results from several of them.

Learning objectives

  1. Set out in logical order the results of the unit, indicating on which results of Calculus I and of the preceding units each one depends and in which proofs the Cauchy criterion, the comparison test, the tail bound of Dirichlet’s test and the change of variable in an improper integral are reused.
  2. Distinguish absolute from conditional convergence and the convergence of an integral with a parameter from its uniform convergence, and recognize in each theorem the point at which the completeness axiom, monotonicity or uniform continuity intervenes.
  3. Relate the unit to Calculus I, to the units on sequences, on the definite integral and on its applications, and to the units on numerical series and on series of functions.
  4. Solve nonroutine problems that combine several lessons: deciding convergence with a well-chosen bound, integrating by parts in an improper integral, introducing a parameter and differentiating with respect to it, and passing to the limit in an integral that depends on a parameter.

What the unit achieved

The unit on the definite integral gave an integral to every function bounded on a closed and bounded interval, and only to those. This unit extended it with the passage to the limit that the unit on sequences had studied: if ff is integrable on each [a,b]\icc{a}{b} with a<b<βa < b < \beta, where β\beta is a real number or the symbol ∞\infty, its improper integral is the limit below, when it exists and is a real number:

∫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 this is so, the central question of the unit was not how to compute that integral, but when the limit exists, and the response repeated those of the unit on sequences: a Cauchy criterion, monotonicity with the completeness axiom, comparison and absolute convergence. Now, some integrals converge only by cancellation, like that of sin⁡xx\frac{\sin x}{x}, and for them finer tests were needed; and some converge without their antiderivative being expressible, so that their value required other routes: differentiation with respect to a parameter and the Wallis integrals. Hence the chain has five segments: the definition and its properties, the convergence tests, conditional convergence, integrals with a parameter and, finally, the bell curve, the gamma function and the factorial.

The definition and the convergence tests

The lesson on improper integrals and convergence tests defined the improper integral at each endpoint, with the notation b→βb \to \beta that abbreviates b→∞b \to \infty and b→β−b \to \beta^-, and that over an open interval by splitting it at a point, on which it proved the integral does not depend. One lemma gave a common form to the limit at the two kinds of endpoint, and another, the reflection, carried over to the left endpoint everything proved at the right one; consistency with the Riemann integral ensured that nothing changes when the function was already integrable on the closed interval, and linearity, additivity and monotonicity passed to the limit. The integrals of powers fixed the scale of comparison: ∫1∞dxxp\int_1^{\infty} \frac{\mathrm{d}x}{x^p} converges if and only if p>1p > 1, and ∫01dxxp\int_0^1 \frac{\mathrm{d}x}{x^p} if and only if p<1p < 1. The Cauchy criterion for improper integrals, proved with the one for sequences, decides convergence without knowing the limit; for nonnegative functions, the completeness axiom reduces convergence to the boundedness of the partial integrals, and from this follow the comparison test and the limit comparison test; and the Cauchy criterion shows that absolute convergence implies convergence. With all this, the lesson measured unbounded regions and curves, computed ∫0π2ln⁡sin⁡x dx=−π2ln⁡2\int_0^{\frac{\pi}{2}} \ln \sin x \, \mathrm{d}x = -\frac{\pi}{2} \ln 2 with the symmetries of the unit on the definite integral, once its convergence had been proved, and distinguished the improper integral from the principal value.

Conditional convergence

The lesson on conditional convergence of improper integrals studied the integrals that converge without converging absolutely. The second mean value theorem for integrals, without derivatives, and the Cauchy criterion gave Dirichlet’s test, for the product of a monotonic function tending to 00 and another whose integrals are bounded, with a tail bound, ∣∫uβfg∣≤K∣f(u)∣\abs{\int_u^{\beta} fg} \leq K \abs{f(u)}, that the lesson after it needed; and Abel’s test, for the product of a monotonic and bounded function and another with a convergent integral. With them, the integral of sin⁡xx\frac{\sin x}{x} on the half-line converges, while that of its absolute value diverges, because its integrals over the arcs of the graph exceed a multiple of the sums of the reciprocals of the natural numbers, which grow like a logarithm. A change of variable proved for improper integrals gave the convergence of the Fresnel integrals, whose integrand does not even tend to 00.

Integrals with a parameter

The lesson on differentiation under the integral sign defined the continuity of a function of two variables with squares instead of disks and proved with the Bolzano-Weierstrass theorem the Heine-Cantor theorem on a rectangle; with it, an integral with a parameter whose integrand is continuous on a rectangle is continuous, and the Leibniz rule for integrals with a parameter differentiates it by differentiating inside the integral if the derivative with respect to the parameter is also continuous. For improper integrals it defined uniform convergence, obtained it from an integrable common bound by the Weierstrass M-test and extended continuity and differentiation to them. It computed the Frullani integrals, ∫0∞f(ax)−f(bx)x dx=(f(0)−L)ln⁡ba\int_0^{\infty} \frac{f(ax) - f(bx)}{x} \, \mathrm{d}x = \left(f(0) - L\right) \ln \frac{b}{a}, with a change of variable on each closed interval and without differentiating under the sign, and the value ∫0∞sin⁡xx dx=π2\int_0^{\infty} \frac{\sin x}{x} \, \mathrm{d}x = \frac{\pi}{2}, by differentiating with respect to tt the integral of e−txsin⁡xxe^{-tx} \frac{\sin x}{x} for t>0t > 0 and passing to the limit as t→0+t \to 0^+ with the tail bound of Dirichlet’s test, because on t=0t = 0 no integrable common bound exists.

The bell curve, the gamma function and the factorial

The lesson on the Gaussian integral, the gamma function and Stirling’s formula enclosed the powers of the bell curve between those of 1−x21 - x^2 and of 11+x2\frac{1}{1 + x^2}, whose integrals are Wallis integrals, and obtained ∫−∞∞e−x2 dx=π\int_{-\infty}^{\infty} e^{-x^2} \, \mathrm{d}x = \sqrt{\pi} without double integrals. It defined the gamma function, proved that its integral converges exactly for positive arguments, that Γ(x+1)=x Γ(x)\Gamma(x + 1) = x \, \Gamma(x) and that Γ(n+1)=n!\Gamma(n + 1) = n!, and with the Gaussian integral obtained Γ(12)=π\Gamma\left(\frac{1}{2}\right) = \sqrt{\pi}. Finally, two bounds for the logarithm made the Stirling sequences monotonic, and their common limit was identified by the order of magnitude of the Wallis integrals: n!∼2πn(ne)nn! \sim \sqrt{2\pi n} \left(\frac{n}{e}\right)^n, with a relative error less than e14n−1e^{\frac{1}{4n}} - 1.

What is admitted, what is proved and what remains

The balance of the unit can be summarized as follows:

  • Admitted are the completeness axiom of Algebra and Geometry I, with the Archimedean property that is deduced from it there; the results of Algebra and Geometry on summations, roots and the trigonometric functions; those of Calculus I on limits at infinity and one-sided limits, equivalent infinitesimals, L’Hôpital’s rule, powers with real exponents, the exponential and the logarithm, and the mean value theorems; and those of the four preceding units on antiderivatives, sequences, the definite integral and its applications.
  • Proved are the definition of the improper integral and its independence of the splitting point, the convergence tests, differentiation under the integral sign and the three classical values. Several promises are thus fulfilled: that of the lesson on the Riemann integral, which announced the extension of the integral to unbounded intervals and functions; that of the lesson on the properties of the definite integral, which bounded an oscillating integral and announced its convergence on the half-line; that of the lesson on techniques for definite integrals, which left pending the integral of ln⁡sin⁡x\ln \sin x; those of the lessons on areas and on arc length, which left unbounded regions and curves; and those of the introduction and the synthesis of the unit on sequences, which announced Stirling’s formula.
  • Four instruments are reused instead of repeating arguments: the Cauchy criterion for improper integrals, which gives absolute convergence and Dirichlet’s and Abel’s tests; the comparison test, which decides almost all concrete convergences and underlies the Weierstrass M-test; the tail bound of Dirichlet’s test, which makes convergence uniform where comparison falls short; and the change of variable in an improper integral, which brings the Fresnel integrals, the Gaussian integral and the integral of the gamma function to a known form.
  • Left for later is what sums instead of integrating: the unit on numerical series repeats the passage to the limit with partial sums, and the integral test there compares a series with an improper integral. Also left, as facts from later courses on which nothing here depends, are the values of the Fresnel integrals, the characterization of the gamma function by the convexity of its logarithm and the Gaussian integral by double integrals.

The following figure gathers the five segments into a single chain. Each node of the column is a result, numbered according to the logical order and joined by an arc to the earlier links that its proof uses; around it float the supports, that is, the principles and tools to which the chain resorts again and again, each one joined by a solid line to the links that rest on it and by a dotted one to the link where it is established, except those that come from Calculus I, from Algebra and Geometry or from the preceding units, which are drawn as rings. Clicking a support highlights all the links that use it, and each link refers to the lesson in which it is proved.

Supports of the chain

  • ACompleteness axiom. Axiom of Algebra and Geometry I: every nonempty set of real numbers that is bounded above has a supremum. There the Archimedean property is deduced from it; here it gives the value of the integral of a nonnegative function as the supremum of its partial integrals and the limit of the Stirling sequences. Admitted without proof in this unit.
  • BCalculus I and Algebra and Geometry. Proved in Calculus I and in Algebra and Geometry: limits at infinity and one-sided limits, the algebra of limits and the squeeze theorem, powers with real exponents, the exponential and the logarithm, the differentiation rules, the mean value theorems and monotonicity from the sign of the derivative, the triangle inequality and the trigonometric formulas. Admitted without proof in this unit.
  • CSequences. From the unit on sequences: the limit theorems, the convergence of monotonic and bounded sequences, subsequences and the Bolzano-Weierstrass theorem, and the Cauchy criterion for sequences. Admitted without proof in this unit.
  • DDefinite integral. From the unit on the definite integral: the integrability of continuous and monotonic functions, linearity, additivity, monotonicity and the absolute value of the integral, the second mean value theorem for integrals, the fundamental theorem of calculus, Barrow’s rule, change of variable, integration by parts and the Wallis integrals. Admitted without proof in this unit.
  • ECauchy criterion. The integral ∫aβf\int_a^{\beta} f converges if and only if for every ε>0\varepsilon > 0 there exists cc with ∣∫uvf∣<ε\abs{\int_u^v f} < \varepsilon for all uu and vv in (c,β)\ioo{c}{\beta}: the way to establish convergence without knowing the value. Established in: 7. Cauchy criterion for improper integrals.
  • FComparison. If 0≤f≤g0 \leq f \leq g on [a,β)\ico{a}{\beta} and ∫aβg\int_a^{\beta} g converges, then ∫aβf\int_a^{\beta} f converges and does not exceed it: the usual way to decide a concrete convergence. Established in: 9. Comparison test.
  • GDirichlet tail. If ff is monotonic with limit 00 at β\beta and ∣∫stg∣≤K\abs{\int_s^t g} \leq K for all ss and tt, the integral of fgfg converges and its tail from uu is at most K∣f(u)∣K \abs{f(u)}: a bound that depends on nothing other than uu. Established in: 12. Dirichlet’s test.
  • HChange of variable. If ff is continuous, φ\varphi has a continuous derivative and φ(t)→β\varphi(t) \to \beta, the convergence of ∫aβf\int_a^{\beta} f gives that of ∫pτf(φ(t)) φ′(t) dt\int_p^{\tau} f(\varphi(t)) \, \varphi'(t) \, \mathrm{d}t, with the value ∫φ(p)βf\int_{\varphi(p)}^{\beta} f: the way to bring an integral to a known one. Established in: 16. Change of variable in an improper integral.
  1. Segment 1 The definition and its properties

    • Define the improper integral at each endpoint and on an open interval, and give the limit a common form.
    • Prove consistency with the Riemann integral, reflection and the inherited properties.
    • Decide for which exponents the integrals of powers converge.
    1. Common form of the limit at the endpoint. If GG is defined on [a,β)\ico{a}{\beta}, 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 cc in [a,β)\ico{a}{\beta} with ∣G(b)−L∣<ε\abs{G(b) - L} < \varepsilon for every bb in (c,β)\ioo{c}{\beta}; likewise for the limit ∞\infty.

      Rests on: B (Calculus I and Algebra and Geometry). Proved in: Improper integrals and convergence tests.

    2. Consistency with the Riemann integral. If β\beta is real and ff is integrable on [a,β]\icc{a}{\beta}, the improper integral on [a,β)\ico{a}{\beta} converges to the Riemann integral on [a,β]\icc{a}{\beta}; likewise at the left endpoint.

      Uses: 1. Common form of the limit at the endpoint. Rests on: D (Definite integral). Proved in: Improper integrals and convergence tests.

    3. Reflection of an improper integral. With g(x)=f(−x)g(x) = f(-x), the integral of ff on (α,b]\ioc{\alpha}{b} converges if and only if that of gg on [−b,−α)\ico{-b}{-\alpha} converges, with the same value: everything proved at the right endpoint holds at the left one.

      Rests on: D (Definite integral), B (Calculus I and Algebra and Geometry). Proved in: Improper integrals and convergence tests.

    4. Linearity, additivity and monotonicity of improper integrals. If ∫aβf\int_a^{\beta} f and ∫aβg\int_a^{\beta} g converge, so does ∫aβ(λf+μg)\int_a^{\beta} (\lambda f + \mu g), with the expected value; ∫aβf\int_a^{\beta} f converges if and only if ∫dβf\int_d^{\beta} f converges, and then ∫aβf=∫adf+∫dβf\int_a^{\beta} f = \int_a^d f + \int_d^{\beta} f; and if f≤gf \leq g, ∫aβf≤∫aβg\int_a^{\beta} f \leq \int_a^{\beta} g.

      Uses: 1. Common form of the limit at the endpoint. Rests on: D (Definite integral), B (Calculus I and Algebra and Geometry). Proved in: Improper integrals and convergence tests.

    5. The splitting point does not matter. If the integral over an open interval (α,β)\ioo{\alpha}{\beta} converges when split at a point cc, it converges when split at any other point dd, with the same sum.

      Uses: 4. Linearity, additivity and monotonicity of improper integrals, 3. Reflection of an improper integral. Rests on: D (Definite integral). Proved in: Improper integrals and convergence tests.

    6. The integrals of powers. The integral ∫1∞dxxp\int_1^{\infty} \frac{\mathrm{d}x}{x^p} converges if and only if p>1p > 1, with value 1p−1\frac{1}{p - 1}, and the integral ∫01dxxp\int_0^1 \frac{\mathrm{d}x}{x^p} if and only if p<1p < 1, with value 11−p\frac{1}{1 - p}; in the other cases they diverge to ∞\infty.

      Uses: 2. Consistency with the Riemann integral. Rests on: D (Definite integral), B (Calculus I and Algebra and Geometry). Proved in: Improper integrals and convergence tests.

  2. Segment 2 The convergence tests

    • Prove the Cauchy criterion for improper integrals with the one for sequences.
    • Reduce the convergence of an integral of a nonnegative function to boundedness, with the completeness axiom, and deduce the comparison tests.
    • Prove that absolute convergence implies convergence.
    1. Cauchy criterion for improper integrals. If ff is integrable on each [a,b]\icc{a}{b} with a<b<βa < b < \beta, the integral ∫aβf\int_a^{\beta} f converges if and only if for every ε>0\varepsilon > 0 there exists cc in [a,β)\ico{a}{\beta} with ∣∫uvf(x) dx∣<ε\abs{\int_u^v f(x) \, \mathrm{d}x} < \varepsilon for all uu and vv in (c,β)\ioo{c}{\beta}.

      Uses: 1. Common form of the limit at the endpoint. Rests on: C (Sequences), A (Completeness axiom), B (Calculus I and Algebra and Geometry). Proved in: Improper integrals and convergence tests.

    2. Integrals of nonnegative functions. If f≥0f \geq 0, the function F(b)=∫abfF(b) = \int_a^b f is increasing, and ∫aβf\int_a^{\beta} f converges if and only if FF is bounded, with value sup⁡{F(b):a≤b<β}\sup\left\{F(b) : a \leq b < \beta\right\}; otherwise it diverges to ∞\infty. One sequence bn→βb_n \to \beta with F(bn)F(b_n) convergent suffices.

      Uses: 1. Common form of the limit at the endpoint. Rests on: A (Completeness axiom), C (Sequences), D (Definite integral). Proved in: Improper integrals and convergence tests.

    3. Comparison test. If 0≤f(x)≤g(x)0 \leq f(x) \leq g(x) on [a,β)\ico{a}{\beta} and ∫aβg\int_a^{\beta} g converges, then ∫aβf\int_a^{\beta} f converges and ∫aβf≤∫aβg\int_a^{\beta} f \leq \int_a^{\beta} g; if ∫aβf\int_a^{\beta} f diverges, ∫aβg\int_a^{\beta} g diverges to ∞\infty.

      Uses: 8. Integrals of nonnegative functions. Rests on: D (Definite integral). Proved in: Improper integrals and convergence tests.

    4. Limit comparison test. If f≥0f \geq 0, g>0g > 0 and f(x)g(x)→L\frac{f(x)}{g(x)} \to L as x→βx \to \beta: with 0<L<∞0 < L < \infty, the two integrals converge or diverge together; with L=0L = 0, the convergence of that of gg gives that of ff; with L=∞L = \infty, the divergence of that of gg gives that of ff.

      Uses: 1. Common form of the limit at the endpoint, 4. Linearity, additivity and monotonicity of improper integrals. Rests on: F (Comparison). Proved in: Improper integrals and convergence tests.

    5. Absolute convergence implies convergence. If ∫aβ∣f∣\int_a^{\beta} \abs{f} converges, then ∫aβf\int_a^{\beta} f converges, and ∣∫aβf(x) dx∣≤∫aβ∣f(x)∣ dx\abs{\int_a^{\beta} f(x) \, \mathrm{d}x} \leq \int_a^{\beta} \abs{f(x)} \, \mathrm{d}x.

      Uses: 8. Integrals of nonnegative functions. Rests on: E (Cauchy criterion), D (Definite integral). Proved in: Improper integrals and convergence tests.

  3. Segment 3 Conditional convergence

    • Prove Dirichlet’s and Abel’s tests with the second mean value theorem for integrals.
    • Prove that the integral of sin⁡xx\frac{\sin x}{x} converges and that that of its absolute value diverges.
    • Prove the change of variable in an improper integral and the convergence of the Fresnel integrals.
    1. Dirichlet’s test. If ff is monotonic on [a,β)\ico{a}{\beta} with f(x)→0f(x) \to 0, and gg is integrable on each [a,b]\icc{a}{b} with ∣∫stg∣≤K\abs{\int_s^t g} \leq K for all ss and tt in [a,β)\ico{a}{\beta}, then ∫aβfg\int_a^{\beta} fg converges, and ∣∫uβf(x) g(x) dx∣≤K∣f(u)∣\abs{\int_u^{\beta} f(x) \, g(x) \, \mathrm{d}x} \leq K \abs{f(u)} for every uu in [a,β)\ico{a}{\beta}.

      Uses: 1. Common form of the limit at the endpoint, 4. Linearity, additivity and monotonicity of improper integrals. Rests on: E (Cauchy criterion), D (Definite integral). Proved in: Conditional convergence of improper integrals.

    2. Sine and cosine divided by a power. If aa, pp and λ\lambda are positive, ∫a∞sin⁡(λx)xp dx\int_a^{\infty} \frac{\sin(\lambda x)}{x^p} \, \mathrm{d}x and ∫a∞cos⁡(λx)xp dx\int_a^{\infty} \frac{\cos(\lambda x)}{x^p} \, \mathrm{d}x converge, and their tails from u≥au \geq a are at most 2λup\frac{2}{\lambda u^p}.

      Rests on: G (Dirichlet tail), B (Calculus I and Algebra and Geometry), D (Definite integral). Proved in: Conditional convergence of improper integrals.

    3. Abel’s test. If ff is monotonic and bounded on [a,β)\ico{a}{\beta}, gg is integrable on each [a,b]\icc{a}{b} and ∫aβg\int_a^{\beta} g converges, then ∫aβfg\int_a^{\beta} fg converges.

      Rests on: E (Cauchy criterion), D (Definite integral), B (Calculus I and Algebra and Geometry). Proved in: Conditional convergence of improper integrals.

    4. The integral of sin x / x converges conditionally. The integral ∫0∞sin⁡xx dx\int_0^{\infty} \frac{\sin x}{x} \, \mathrm{d}x converges, with tail from u>0u > 0 at most 2u\frac{2}{u}; on the other hand, ∫0nπ∣sin⁡x∣x dx≥2πln⁡(n+1)\int_0^{n\pi} \frac{\abs{\sin x}}{x} \, \mathrm{d}x \geq \frac{2}{\pi} \ln(n + 1), and the integral of the absolute value diverges.

      Uses: 13. Sine and cosine divided by a power, 4. Linearity, additivity and monotonicity of improper integrals, 8. Integrals of nonnegative functions. Rests on: D (Definite integral), B (Calculus I and Algebra and Geometry), A (Completeness axiom). Proved in: Conditional convergence of improper integrals.

    5. Change of variable in an improper integral. If ff is continuous on [a,β)\ico{a}{\beta}, φ\varphi is differentiable on [p,τ)\ico{p}{\tau} with φ′\varphi' continuous, takes values in [a,β)\ico{a}{\beta} and φ(t)→β\varphi(t) \to \beta as t→τt \to \tau, and if ∫aβf\int_a^{\beta} f converges, then ∫pτf(φ(t)) φ′(t) dt=∫φ(p)βf(x) dx\int_p^{\tau} f(\varphi(t)) \, \varphi'(t) \, \mathrm{d}t = \int_{\varphi(p)}^{\beta} f(x) \, \mathrm{d}x.

      Uses: 1. Common form of the limit at the endpoint. Rests on: D (Definite integral). Proved in: Conditional convergence of improper integrals.

    6. The Fresnel integrals. The integrals ∫0∞sin⁡(x2) dx\int_0^{\infty} \sin(x^2) \, \mathrm{d}x and ∫0∞cos⁡(x2) dx\int_0^{\infty} \cos(x^2) \, \mathrm{d}x converge conditionally: the change u=x2u = x^2 brings them to those of sin⁡u2u\frac{\sin u}{2\sqrt{u}} and of cos⁡u2u\frac{\cos u}{2\sqrt{u}}.

      Uses: 13. Sine and cosine divided by a power, 6. The integrals of powers, 4. Linearity, additivity and monotonicity of improper integrals. Rests on: H (Change of variable), F (Comparison), B (Calculus I and Algebra and Geometry). Proved in: Conditional convergence of improper integrals.

  4. Segment 4 Integrals with a parameter

    • Prove the Heine-Cantor theorem on a rectangle and, with it, the continuity of an integral with a parameter and the Leibniz rule for integrals with a parameter.
    • Define uniform convergence, obtain it from an integrable common bound and extend continuity and differentiation to improper integrals.
    • Compute the Frullani integrals and the Dirichlet integral.
    1. Heine-Cantor theorem on a rectangle. Every function continuous on a rectangle [a,b]×[p,q]\icc{a}{b} \times \icc{p}{q} is uniformly continuous on it: if it were not, the Bolzano-Weierstrass theorem, applied to each coordinate, would give a point where it is not continuous.

      Rests on: C (Sequences), B (Calculus I and Algebra and Geometry). Proved in: Differentiation under the integral sign.

    2. Continuity of an integral with a parameter. If ff is continuous on [a,b]×[p,q]\icc{a}{b} \times \icc{p}{q}, the function F(t)=∫abf(x,t) dxF(t) = \int_a^b f(x, t) \, \mathrm{d}x is uniformly continuous on [p,q]\icc{p}{q}.

      Uses: 18. Heine-Cantor theorem on a rectangle. Rests on: D (Definite integral). Proved in: Differentiation under the integral sign.

    3. Leibniz rule for integrals with a parameter. If ff and ∂f∂t\frac{\partial f}{\partial t} are continuous on [a,b]×[p,q]\icc{a}{b} \times \icc{p}{q}, the function F(t)=∫abf(x,t) dxF(t) = \int_a^b f(x, t) \, \mathrm{d}x is differentiable on [p,q]\icc{p}{q}, with continuous derivative F′(t)=∫ab∂f∂t(x,t) dxF'(t) = \int_a^b \frac{\partial f}{\partial t}(x, t) \, \mathrm{d}x.

      Uses: 18. Heine-Cantor theorem on a rectangle, 19. Continuity of an integral with a parameter. Rests on: B (Calculus I and Algebra and Geometry), D (Definite integral). Proved in: Differentiation under the integral sign.

    4. Weierstrass M-test for integrals with a parameter. If ∣f(x,t)∣≤g(x)\abs{f(x, t)} \leq g(x) for every xx in [a,β)\ico{a}{\beta} and every tt in JJ, with ∫aβg\int_a^{\beta} g convergent, the integral ∫aβf(x,t) dx\int_a^{\beta} f(x, t) \, \mathrm{d}x converges absolutely for each tt, uniformly on JJ, with tails at most ∫uβg\int_u^{\beta} g.

      Uses: 11. Absolute convergence implies convergence, 4. Linearity, additivity and monotonicity of improper integrals, 1. Common form of the limit at the endpoint. Rests on: F (Comparison), D (Definite integral). Proved in: Differentiation under the integral sign.

    5. Continuity of an improper integral with a parameter. If ff is continuous on [a,β)×J\ico{a}{\beta} \times J and the integral converges uniformly on each closed and bounded interval contained in JJ, the function F(t)=∫aβf(x,t) dxF(t) = \int_a^{\beta} f(x, t) \, \mathrm{d}x is continuous on JJ.

      Uses: 19. Continuity of an integral with a parameter, 4. Linearity, additivity and monotonicity of improper integrals. Proved in: Differentiation under the integral sign.

    6. Differentiation under the sign of an improper integral. If ff and ∂f∂t\frac{\partial f}{\partial t} are continuous on [a,β)×J\ico{a}{\beta} \times J, the integral of ff converges for each tt and that of ∂f∂t\frac{\partial f}{\partial t} converges uniformly on each closed and bounded interval of JJ, for instance with an integrable common bound, then F′(t)=∫aβ∂f∂t(x,t) dxF'(t) = \int_a^{\beta} \frac{\partial f}{\partial t}(x, t) \, \mathrm{d}x.

      Uses: 20. Leibniz rule for integrals with a parameter, 22. Continuity of an improper integral with a parameter, 21. Weierstrass M-test for integrals with a parameter. Rests on: D (Definite integral). Proved in: Differentiation under the integral sign.

    7. Frullani integrals. If ff is continuous on [0,∞)\ico{0}{\infty} and has a real limit LL at infinity, and aa and bb are positive, then ∫0∞f(ax)−f(bx)x dx=(f(0)−L)ln⁡ba\int_0^{\infty} \frac{f(ax) - f(bx)}{x} \, \mathrm{d}x = \left(f(0) - L\right) \ln \frac{b}{a}.

      Rests on: D (Definite integral), B (Calculus I and Algebra and Geometry). Proved in: Differentiation under the integral sign.

    8. The Dirichlet integral. For every t≥0t \geq 0, ∫0∞e−tx sin⁡xx dx=π2−arctan⁡t\int_0^{\infty} e^{-tx} \, \frac{\sin x}{x} \, \mathrm{d}x = \frac{\pi}{2} - \arctan t; in particular, ∫0∞sin⁡xx dx=π2\int_0^{\infty} \frac{\sin x}{x} \, \mathrm{d}x = \frac{\pi}{2}. Continuity at t=0t = 0 follows from the Dirichlet tail bound.

      Uses: 4. Linearity, additivity and monotonicity of improper integrals, 11. Absolute convergence implies convergence, 22. Continuity of an improper integral with a parameter, 23. Differentiation under the sign of an improper integral, 21. Weierstrass M-test for integrals with a parameter. Rests on: G (Dirichlet tail), F (Comparison), D (Definite integral), B (Calculus I and Algebra and Geometry). Proved in: Differentiation under the integral sign.

  5. Segment 5 The bell curve, the gamma function and the factorial

    • Compute the Gaussian integral with the Wallis integrals, without double integrals.
    • Define the gamma function and prove its convergence, its functional equation and its value at one half.
    • Prove Stirling’s formula with a monotonic sequence and the order of magnitude of the Wallis integrals.
    1. Increasing change of variable on an open interval. If ff is continuous on (α,β)\ioo{\alpha}{\beta} and φ\varphi, strictly increasing on (p,τ)\ioo{p}{\tau} with φ′\varphi' continuous, tends to α\alpha and to β\beta at the endpoints, the convergence of ∫αβf\int_{\alpha}^{\beta} f gives ∫pτf(φ(t)) φ′(t) dt=∫αβf(x) dx\int_p^{\tau} f(\varphi(t)) \, \varphi'(t) \, \mathrm{d}t = \int_{\alpha}^{\beta} f(x) \, \mathrm{d}x.

      Uses: 5. The splitting point does not matter, 3. Reflection of an improper integral, 4. Linearity, additivity and monotonicity of improper integrals. Rests on: H (Change of variable), D (Definite integral), B (Calculus I and Algebra and Geometry). Proved in: The Gaussian integral, the gamma function and Stirling’s formula.

    2. Bounds for the bell curve. For every real xx, 1−x2≤e−x2≤11+x21 - x^2 \leq e^{-x^2} \leq \frac{1}{1 + x^2}.

      Rests on: B (Calculus I and Algebra and Geometry). Proved in: The Gaussian integral, the gamma function and Stirling’s formula.

    3. Gaussian integral. The integral of e−x2e^{-x^2} on the line converges, and ∫−∞∞e−x2 dx=π\int_{-\infty}^{\infty} e^{-x^2} \, \mathrm{d}x = \sqrt{\pi}: the nnth powers of the bounds for the bell curve have as integrals multiples of Wallis integrals.

      Uses: 27. Bounds for the bell curve, 3. Reflection of an improper integral, 8. Integrals of nonnegative functions, 2. Consistency with the Riemann integral. Rests on: F (Comparison), H (Change of variable), D (Definite integral), C (Sequences), B (Calculus I and Algebra and Geometry). Proved in: The Gaussian integral, the gamma function and Stirling’s formula.

    4. Convergence of the integral of the gamma function. The integral ∫0∞tx−1e−t dt\int_0^{\infty} t^{x - 1} e^{-t} \, \mathrm{d}t converges if and only if x>0x > 0; its value defines Γ(x)\Gamma(x).

      Uses: 5. The splitting point does not matter, 6. The integrals of powers, 10. Limit comparison test, 3. Reflection of an improper integral. Rests on: F (Comparison), B (Calculus I and Algebra and Geometry). Proved in: The Gaussian integral, the gamma function and Stirling’s formula.

    5. Functional equation of the gamma function. For every x>0x > 0, Γ(x+1)=x Γ(x)\Gamma(x + 1) = x \, \Gamma(x); moreover, Γ(1)=1\Gamma(1) = 1 and Γ(n+1)=n!\Gamma(n + 1) = n! for every integer n≥0n \geq 0.

      Uses: 29. Convergence of the integral of the gamma function, 2. Consistency with the Riemann integral, 4. Linearity, additivity and monotonicity of improper integrals. Rests on: D (Definite integral), B (Calculus I and Algebra and Geometry). Proved in: The Gaussian integral, the gamma function and Stirling’s formula.

    6. The gamma function at one half. It holds that Γ(12)=π\Gamma\left(\frac{1}{2}\right) = \sqrt{\pi}, and therefore Γ(32)=π2\Gamma\left(\frac{3}{2}\right) = \frac{\sqrt{\pi}}{2}: the change t=u2t = u^2 turns the integral into twice that of the bell curve.

      Uses: 2. Consistency with the Riemann integral, 4. Linearity, additivity and monotonicity of improper integrals, 26. Increasing change of variable on an open interval, 29. Convergence of the integral of the gamma function, 28. Gaussian integral, 30. Functional equation of the gamma function. Rests on: B (Calculus I and Algebra and Geometry). Proved in: The Gaussian integral, the gamma function and Stirling’s formula.

    7. The Stirling sequence. The sequence an=n! ennnna_n = \frac{n! \, e^n}{n^n \sqrt{n}} is strictly decreasing, bn=ane−14nb_n = a_n e^{-\frac{1}{4n}} is strictly increasing, and the two converge to the same ρ≥e34\rho \geq e^{\frac{3}{4}}, with ρ<an<ρ e14n\rho < a_n < \rho \, e^{\frac{1}{4n}}: this follows from two bounds for the logarithm.

      Rests on: A (Completeness axiom), C (Sequences), D (Definite integral), B (Calculus I and Algebra and Geometry). Proved in: The Gaussian integral, the gamma function and Stirling’s formula.

    8. Stirling’s formula. It holds that n!∼2πn(ne)nn! \sim \sqrt{2\pi n} \left(\frac{n}{e}\right)^n, and 2πn(ne)n<n!<2πn(ne)ne14n\sqrt{2\pi n} \left(\frac{n}{e}\right)^n < n! < \sqrt{2\pi n} \left(\frac{n}{e}\right)^n e^{\frac{1}{4n}} for every natural number nn: the quotient an2a2n\frac{a_n^2}{a_{2n}} is written with W2nW_{2n}.

      Uses: 32. The Stirling sequence. Rests on: D (Definite integral), C (Sequences), B (Calculus I and Algebra and Geometry). Proved in: The Gaussian integral, the gamma function and Stirling’s formula.

Toward numerical series. The chain stops where the integral gives way to the sum. The unit that follows defines the sum of a series as the limit of its partial sums, the same passage to the limit as in this unit with a sum instead of an integral, and repeats its questions and its responses: the Cauchy criterion, the boundedness of the partial sums for series with nonnegative terms, with the completeness axiom, comparison and absolute convergence; and it studies alternating series, which converge by cancellation like the integrals of the second lesson. One difference appears from the outset: the general term of a convergent series tends to 00, whereas the integrand of a convergent integral need not tend to 00. The two units meet in the integral test, which compares a series with positive and decreasing terms with the improper integral of a decreasing function, and which will turn the integrals of powers of this unit into the scale of comparison for series.