4.2 Improper Integrals
Integrals over unbounded intervals and of unbounded functions: the $p$-tests, comparison and limit comparison, absolute versus conditional convergence, the Dirichlet and Abel tests, $\int_0^\infty\frac{\sin x}{x}\,dx=\frac\pi2$, the Gamma and Beta functions and the Cauchy principal value.
The Riemann integral of §4.1 lives on a bounded interval and only accepts bounded functions. Yet the integrals that matter most in applications break both rules: the normal density $\tfrac1{\sqrt{2\pi}}e^{-x^2/2}$ is integrated over the whole line, the Gamma function $\int_0^\infty x^{s-1}e^{-x}\,dx$ has an infinite range and, for $s\lt1$, an infinite integrand at $0$; the escape velocity of a rocket is an integral of $1/r^2$ out to infinity; the electric potential of a point charge has a singularity. An improper integral is defined as a limit of proper ones, and the whole subject is about deciding when that limit exists.
The decision is usually made by comparison — near a bad point, how fast does the integrand blow up or decay, compared with a power $x^{-p}$? — and, for oscillating integrands such as $\frac{\sin x}{x}$, by cancellation, which needs the subtler Dirichlet and Abel tests. CSIR NET asks for exactly these decisions: Part B questions of the form "for which $p$ does this integral converge?" or "evaluate using Gamma/Beta", and Part C questions that mix absolutely convergent, conditionally convergent and divergent integrals among the options, or test whether convergence of $\int f$ forces $f(x)\to0$.
Improper Integrals of the First and Second Kind
We say $f$ is locally integrable on an interval $J$ if $f$ is Riemann integrable on every closed bounded subinterval of $J$ (in particular $f$ is bounded on each of them).
Let $f$ be locally integrable on $[a,\infty)$. If the limit $$\int_a^\infty f(x)\,dx=\lim_{b\to\infty}\int_a^bf(x)\,dx$$ exists in $\mathbb{R}$, the improper integral converges; otherwise it diverges (to $\pm\infty$, or by oscillation). $\int_{-\infty}^bf$ is defined in the same way. For $f$ locally integrable on $\mathbb{R}$, $\int_{-\infty}^{\infty}f$ converges if both $\int_{-\infty}^cf$ and $\int_c^\infty f$ converge for some (hence every) $c$, and then $\int_{-\infty}^\infty f=\int_{-\infty}^cf+\int_c^\infty f$.
The two limits must exist independently: $\int_{-\infty}^\infty x\,dx$ diverges, although $\int_{-R}^Rx\,dx=0$ for every $R$ (that symmetric limit is the principal value, Definition 4.10).
Let $f$ be locally integrable on $(a,b]$ (typically unbounded near $a$). If $$\int_a^bf(x)\,dx=\lim_{c\to a^+}\int_c^bf(x)\,dx$$ exists in $\mathbb{R}$, the integral converges. Similarly at the right end point $b$. If $f$ has finitely many "bad points" (singularities, or $\pm\infty$), split the interval at points between them; the integral converges if and only if every piece converges, and its value is the sum.
If $f$ happens to be Riemann integrable on $[a,b]$, the improper integral converges to the ordinary one, because $c\mapsto\int_c^bf$ is continuous (Theorem 4.8). So the notation is unambiguous.
- $\displaystyle\int_1^\infty\frac{dx}{x^p}$ converges if and only if $p\gt1$, and then equals $\dfrac1{p-1}$.
- $\displaystyle\int_0^1\frac{dx}{x^p}$ converges if and only if $p\lt1$, and then equals $\dfrac1{1-p}$.
- $\displaystyle\int_2^\infty\frac{dx}{x(\log x)^q}$ converges if and only if $q\gt1$.
- $\displaystyle\int_0^\infty e^{-ax}\,dx$ converges if and only if $a\gt0$, and then equals $\dfrac1a$.
Proof. All four integrands are continuous with explicit antiderivatives. (1) For $p\ne1$, $\int_1^bx^{-p}dx=\frac{1-b^{1-p}}{p-1}$, which tends to $\frac1{p-1}$ if $p\gt1$ (since $b^{1-p}\to0$) and to $+\infty$ if $p\lt1$; for $p=1$ it is $\log b\to\infty$. (2) For $p\ne1$, $\int_c^1x^{-p}dx=\frac{1-c^{1-p}}{1-p}$, which tends to $\frac1{1-p}$ as $c\to0^+$ if $p\lt1$, and to $+\infty$ if $p\gt1$; for $p=1$ it is $-\log c\to\infty$. (3) Substitute $u=\log x$: $\int_2^b\frac{dx}{x(\log x)^q}=\int_{\log2}^{\log b}\frac{du}{u^q}$, and apply (1) (the lower limit $\log2\gt0$ does not matter). (4) $\int_0^be^{-ax}dx=\frac{1-e^{-ab}}a$ for $a\ne0$, and $b$ for $a=0$. ∎
Notice the two thresholds point in opposite directions: at infinity we need fast decay ($p\gt1$), near $0$ we need a mild singularity ($p\lt1$). Consequently $\int_0^\infty x^{-p}\,dx$ diverges for every $p$. The borderline $p=1$ diverges at both ends, but only logarithmically — and part (3) shows that a power of a logarithm can tip the balance.
Proof. Let $F(b)=\int_a^bf$, so $\int_x^yf=F(y)-F(x)$. If $F(b)\to I$, choose $A$ with $|F(b)-I|\lt\varepsilon/2$ for $b\ge A$; then $|F(y)-F(x)|\lt\varepsilon$. Conversely, if the condition holds, then for every sequence $b_n\to\infty$ the sequence $F(b_n)$ is Cauchy, hence converges (completeness of $\mathbb{R}$, §2.1); interleaving two such sequences shows that all the limits coincide, and the sequential characterisation of limits gives $\lim_{b\to\infty}F(b)=I$. ∎
Comparison Tests for Non-negative Integrands
For a non-negative integrand the partial integrals $F(b)=\int_a^bf$ increase with $b$, so by the monotone convergence principle (§2.1) they either converge or tend to $+\infty$. Everything rests on this one observation.
- If $f\ge0$, then $\int_a^\infty f$ converges if and only if $F(b)=\int_a^bf$ is bounded above.
- If $0\le f(x)\le g(x)$ for all $x\ge a$ (or for all large $x$), then: $\int_a^\infty g$ converges $\Rightarrow\int_a^\infty f$ converges; and $\int_a^\infty f$ diverges $\Rightarrow\int_a^\infty g$ diverges.
- (Integral test.) If $f\ge0$ is decreasing on $[1,\infty)$, then $\int_1^\infty f$ and $\sum_{n\ge1}f(n)$ converge or diverge together, and $\sum_{n=2}^Nf(n)\le\int_1^Nf\le\sum_{n=1}^{N-1}f(n)$.
Proof. (1) $F$ is increasing, so $\lim_{b\to\infty}F(b)=\sup_bF(b)$, which is finite iff $F$ is bounded (§1.2). (2) If the inequality holds for $x\ge A$, then $\int_A^bf\le\int_A^bg\le\int_A^\infty g$, so the partial integrals of $f$ beyond $A$ are bounded; apply (1) (the piece over $[a,A]$ is a proper integral). The second implication is the contrapositive. (3) For $k\le x\le k+1$, $f(k+1)\le f(x)\le f(k)$; integrate over $[k,k+1]$ and sum from $k=1$ to $N-1$. Both partial sums and partial integrals are increasing, and each is bounded iff the other is. ∎
- If $0\lt L\lt\infty$, then $\int_a^\infty f$ and $\int_a^\infty g$ converge or diverge together.
- If $L=0$ and $\int_a^\infty g$ converges, then $\int_a^\infty f$ converges.
- If $L=\infty$ and $\int_a^\infty g$ diverges, then $\int_a^\infty f$ diverges.
Proof. (1) There is $A$ with $\frac L2\le\frac{f(x)}{g(x)}\le2L$ for $x\ge A$, i.e. $\frac L2g\le f\le2Lg$ there; apply Theorem 4.13(2) in both directions. (2) Eventually $f\le g$. (3) Eventually $f\ge g$. ∎
- Locate the bad points: $\pm\infty$, and every point where the integrand is unbounded. Split the integral so that each piece has exactly one bad point, and test each piece separately — all must converge.
- Find the leading behaviour at each bad point: $f(x)\sim Cx^{-p}$ as $x\to\infty$, or $f(x)\sim C|x-c|^{-p}$ as $x\to c$ (use Taylor expansions: $\sin x\sim x$, $1-\cos x\sim\tfrac{x^2}2$, $e^x-1\sim x$, $\log(1+x)\sim x$, $x-\sin x\sim\tfrac{x^3}6$).
- Apply the $p$-test by limit comparison: at $\infty$ converge iff $p\gt1$; at a finite point converge iff $p\lt1$.
- Logs and exponentials: $e^{-x}$ beats every power ($x^ke^{-x}\le x^{-2}$ for large $x$); every power beats every power of $\log$ ($(\log x)^k\le x^\delta$ for large $x$, any $\delta\gt0$). So $\int_2^\infty\frac{dx}{x^p(\log x)^q}$ is decided by $p$ alone unless $p=1$, where $q$ decides (Theorem 4.11(3)).
- Sign changes? Comparison tests need $f\ge0$ (at least eventually). For oscillating integrands go to the next subsection.
Absolute and Conditional Convergence; the Dirichlet and Abel Tests
Proof. $|f|$ is locally integrable (Theorem 4.6). For $y\gt x$, $\left|\int_x^yf\right|\le\int_x^y|f|$, and the right side is small for large $x$ by the Cauchy criterion (Theorem 4.12) applied to $|f|$; so the Cauchy criterion holds for $f$. The inequality follows by letting $b\to\infty$ in $\left|\int_a^bf\right|\le\int_a^b|f|$. (Alternatively: $0\le f+|f|\le2|f|$, so $\int(f+|f|)$ converges by comparison, and $f=(f+|f|)-|f|$.) ∎
For example $\int_1^\infty\frac{\sin x}{x^2}dx$ converges absolutely, because $\left|\frac{\sin x}{x^2}\right|\le\frac1{x^2}$. But the most important oscillating integral, $\int_0^\infty\frac{\sin x}{x}dx$, does not converge absolutely — its convergence comes from the cancellation between positive and negative lobes. The next test captures that cancellation; it is the integral analogue of the Dirichlet test for series (§2.3).
Proof. Let $F(x)=\int_a^xf$, so $F'=f$ (Theorem 4.8) and $|F|\le K$. Integration by parts (Theorem 4.10) on $[a,b]$ gives $$\int_a^bfg=F(b)g(b)-\int_a^bF(x)g'(x)\,dx.$$ As $b\to\infty$, $|F(b)g(b)|\le K|g(b)|\to0$. For the last integral, $g'$ has constant sign (as $g$ is monotone), so $$\int_a^b|F\,g'|\le K\int_a^b|g'|=K\,|g(b)-g(a)|\le K|g(a)|,$$ since $g(b)$ moves monotonically from $g(a)$ towards $0$. So $\int_a^\infty Fg'$ converges absolutely (Theorem 4.13(1)), hence converges (Theorem 4.15), and $\int_a^bfg$ has a limit. ∎
The smoothness of $g$ is a convenience: the test remains true for any monotone $g\to0$ and locally integrable $f$ with bounded partial integrals, by Bonnet's second mean value theorem: for monotone $g\ge0$ decreasing, $\int_x^yfg=g(x^+)\int_x^\xi f$ for some $\xi\in[x,y]$. The monotonicity of $g$ cannot be dropped (pitfall 8 below).
Proof. A bounded monotone function has a limit $\ell=\lim_{x\to\infty}g(x)$ (§2.1, applied along sequences). Write $fg=f\,(g-\ell)+\ell f$. The function $g-\ell$ is monotone, $C^1$, and tends to $0$; the partial integrals of $f$ converge, so they are bounded. Dirichlet's test gives convergence of $\int f(g-\ell)$, and $\int\ell f$ converges by hypothesis. ∎
Typical uses: $\int_1^\infty\frac{\sin x}{x}\tan^{-1}x\,dx$ (Abel, $g=\tan^{-1}$), $\int_1^\infty\cos x\cdot\frac{\log x}{x}\,dx$ (Dirichlet, $g=\frac{\log x}x$ decreasing for $x\ge e$).
Proof. Convergence and non-absolute convergence are Example 4 with $p=1$. For the value, use the Dirichlet kernel: from $2\sin\frac t2\cos kt=\sin(k+\frac12)t-\sin(k-\frac12)t$ and telescoping, $$D_n(t)=\frac12+\sum_{k=1}^n\cos kt=\frac{\sin\big((n+\frac12)t\big)}{2\sin\frac t2}\qquad(0\lt t\le\pi),$$ and integrating the cosine sum term by term over $[0,\pi]$ gives $\int_0^\pi D_n(t)\,dt=\frac\pi2$ for every $n$. Now let $$\varphi(t)=\frac1t-\frac1{2\sin\frac t2}\quad(0\lt t\le\pi),\qquad\varphi(0)=0.$$ Expanding $\sin\frac t2=\frac t2-\frac{t^3}{48}+\cdots$ shows $\varphi(t)=-\frac{t}{24}+O(t^3)$, so $\varphi$ is continuous on $[0,\pi]$. By the Riemann–Lebesgue lemma (§4.1, Exercise IV.1; the same proof works for $\sin\big((n+\frac12)t\big)=\sin nt\cos\frac t2+\cos nt\sin\frac t2$), $$\int_0^\pi\varphi(t)\sin\big((n+\tfrac12)t\big)\,dt\to0.$$ Hence, since $\frac1t=\frac1{2\sin(t/2)}+\varphi(t)$, $$\int_0^\pi\frac{\sin((n+\frac12)t)}{t}dt=\frac\pi2+\int_0^\pi\varphi(t)\sin\big((n+\tfrac12)t\big)\,dt\to\frac\pi2.$$ Substituting $u=(n+\frac12)t$, the left side is $\int_0^{(n+\frac12)\pi}\frac{\sin u}u\,du$, which tends to $\int_0^\infty\frac{\sin u}u\,du$ since that integral converges. So the value is $\frac\pi2$. ∎
By substitution, $\int_0^\infty\frac{\sin ax}{x}dx=\frac\pi2\operatorname{sign}a$ for every real $a$ — a function of $a$ with a jump at $0$, a first warning that limits and improper integrals need care (Chapter 5, §4.5).
The Gamma and Beta Functions; Principal Values
- The integral for $\Gamma(s)$ converges if and only if $s\gt0$; that for $B(p,q)$ if and only if $p\gt0$ and $q\gt0$.
- $\Gamma(s+1)=s\,\Gamma(s)$ for $s\gt0$; $\Gamma(1)=1$, so $\Gamma(n+1)=n!$ for $n=0,1,2,\dots$
- $B(p,q)=B(q,p)$, and $$B(p,q)=2\int_0^{\pi/2}\sin^{2p-1}\theta\,\cos^{2q-1}\theta\,d\theta,$$ $$B(p,q)=\int_0^\infty\frac{t^{p-1}}{(1+t)^{p+q}}\,dt.$$
- $B(p,q)=\dfrac{\Gamma(p)\Gamma(q)}{\Gamma(p+q)}$.
- $\Gamma(\tfrac12)=\sqrt\pi$, hence $\int_{-\infty}^\infty e^{-x^2}dx=\sqrt\pi$, and $\int_0^\infty x^{m}e^{-x^n}dx=\frac1n\Gamma\big(\frac{m+1}n\big)$ for $n\gt0$, $m\gt-1$.
- (Reflection formula, statement.) $\Gamma(p)\Gamma(1-p)=\dfrac{\pi}{\sin p\pi}$ for $0\lt p\lt1$; equivalently $\int_0^\infty\frac{t^{p-1}}{1+t}dt=\frac\pi{\sin p\pi}$.
Proof. (1) The integrand of $\Gamma$ is positive. Near $0$, $x^{s-1}e^{-x}\big/x^{s-1}\to1$, and $\int_0^1x^{s-1}dx$ converges iff $1-s\lt1$, i.e. $s\gt0$ (Theorem 4.14 and 4.11). Near $\infty$, $x^{s-1}e^{-x}\le x^{-2}$ for large $x$ because $x^{s+1}e^{-x}\to0$; so $\int_1^\infty$ converges for every $s$. For $B$, the integrand behaves like $x^{p-1}$ at $0$ and like $(1-x)^{q-1}$ at $1$.
(2) On $[c,R]$, integrate by parts with $u=x^s$, $dv=e^{-x}dx$: $\int_c^Rx^se^{-x}dx=\big[-x^se^{-x}\big]_c^R+s\int_c^Rx^{s-1}e^{-x}dx$. Let $c\to0^+$, $R\to\infty$; the boundary terms vanish because $s\gt0$. And $\Gamma(1)=\int_0^\infty e^{-x}dx=1$; induct.
(3) Symmetry: $x\mapsto1-x$. The trigonometric form: $x=\sin^2\theta$, $dx=2\sin\theta\cos\theta\,d\theta$. The last form: $x=\frac t{1+t}$, $1-x=\frac1{1+t}$, $dx=\frac{dt}{(1+t)^2}$. (Substitution in improper integrals is justified on compact subintervals and then by passing to the limit.)
(4) (Careful sketch.) Substituting $x=u^2$ gives $\Gamma(p)=2\int_0^\infty u^{2p-1}e^{-u^2}du$. Hence $$\Gamma(p)\Gamma(q)=4\iint_{u,v\gt0}u^{2p-1}v^{2q-1}e^{-(u^2+v^2)}\,du\,dv.$$ The integrand is non-negative and continuous, so the iterated improper integrals may be combined into one over the quarter plane and computed in polar coordinates $u=r\cos\theta$, $v=r\sin\theta$ (Tonelli's theorem for non-negative functions, §4.5; or exhaust the quarter plane by quarter discs and squares, which squeeze the same integral). This gives $$\Gamma(p)\Gamma(q)=\Big(2\int_0^\infty r^{2(p+q)-1}e^{-r^2}dr\Big)\Big(2\int_0^{\pi/2}\cos^{2p-1}\theta\sin^{2q-1}\theta\,d\theta\Big)=\Gamma(p+q)\,B(p,q).$$ (5) By (3), $B(\frac12,\frac12)=2\int_0^{\pi/2}d\theta=\pi$, and by (4) it equals $\Gamma(\frac12)^2/\Gamma(1)$; as $\Gamma(\frac12)\gt0$, $\Gamma(\frac12)=\sqrt\pi$. Then $\int_0^\infty e^{-x^2}dx=\frac12\Gamma(\frac12)$ (substitute $t=x^2$), and the general formula follows from $t=x^n$. (6) is proved with complex analysis (residues, Unit 2) or with infinite products; we use it as a fact. ∎
If $f$ is locally integrable on $\mathbb{R}$, its principal value is $$\mathrm{PV}\!\int_{-\infty}^\infty f=\lim_{R\to\infty}\int_{-R}^Rf,$$ when this limit exists. If $f$ is locally integrable on $[a,b]$ except near one interior point $c$, $$\mathrm{PV}\!\int_a^bf=\lim_{\varepsilon\to0^+}\Big(\int_a^{c-\varepsilon}f+\int_{c+\varepsilon}^bf\Big).$$
If the improper integral converges, the principal value exists and equals it; the converse fails. Principal values arise naturally in the Hilbert transform, in dispersion relations in physics, and in contour integration (Unit 2).
- Symmetric limits are not enough. $\int_{-\infty}^\infty x\,dx$ and $\int_{-\infty}^\infty\frac{x}{1+x^2}dx$ diverge although their principal values are $0$.
- Hidden singularities. $\int_{-1}^1\frac{dx}{x^2}$ is not $\big[-\frac1x\big]_{-1}^1=-2$: it diverges. Always look for points where the integrand is unbounded before using an antiderivative.
- $\int_a^\infty f$ convergent does not force $f\to0$, nor even $f$ bounded: spikes of area $\frac1{n^2}$; $\sin(x^2)$; $x\sin(x^4)$. It does if $f$ is uniformly continuous, or monotone (Example 5).
- Convergent is not absolutely convergent. $\int_0^\infty\frac{\sin x}x\,dx=\frac\pi2$, while $\int_0^\infty\frac{|\sin x|}x\,dx=\infty$. In Lebesgue's theory (§4.5) $\frac{\sin x}{x}$ is not integrable on $(0,\infty)$ — the improper Riemann integral exists only as a limit.
- No implication between $\int f$ and $\int f^2$. On $[1,\infty)$, $f=\frac1x$: $\int f^2$ converges, $\int f$ diverges. On $(0,1]$, $f=\frac1{\sqrt x}$: $\int f$ converges, $\int f^2$ diverges. And $f=\frac{\sin x}{\sqrt x}$ on $[1,\infty)$: $\int f$ converges, $\int f^2=\int\frac{\sin^2x}x$ diverges.
- Limit comparison needs a sign condition. $g=\frac{\sin x}{\sqrt x}$ and $f=\frac{\sin x}{\sqrt x}+\frac{\sin^2x}{x}$ on $[1,\infty)$ have $\frac fg=1+\frac{\sin x}{\sqrt x}\to1$ (where $\sin x\ne0$), yet $\int g$ converges and $\int f$ diverges.
- Integral and series can disagree without monotonicity. $f(x)=\sin^2(\pi x)$ has $\sum f(n)=0$ but $\int_1^\infty f=\infty$; narrow spikes of height $1$ at the integers give $\int f\lt\infty$ but $\sum f(n)=\infty$. The integral test needs $f$ decreasing.
- Dirichlet's test needs monotone $g$. With $f=\sin x$ and $g=\frac{\sin x}{x}\to0$ (not monotone), $\int_1^\infty fg=\int_1^\infty\frac{\sin^2x}x\,dx$ diverges.
- Every piece must converge. $\int_0^\infty\frac{dx}{x^p}$ diverges for every $p$: the two ends need $p\lt1$ and $p\gt1$.
- Domains of $\Gamma$ and $B$. $\int_0^\infty x^{s-1}e^{-x}dx$ diverges for $s\le0$ (at $0$); $B(p,q)$ needs $p,q\gt0$. Formulas like $\Gamma(s)=\Gamma(s+1)/s$ extend $\Gamma$ to negative non-integers, but not as a convergent integral.
Exercises
- 1 Evaluate $\int_0^\infty xe^{-x^2}dx$.
- 2 Evaluate $\int_1^\infty\frac{\log x}{x^2}dx$.
- 3 Evaluate $\int_0^1x\log x\,dx$.
- 4 Evaluate $\int_0^\infty e^{-x}\sin x\,dx$.
- 5 Evaluate $\int_{-\infty}^\infty\frac{dx}{x^2+2x+5}$.
- 6 Evaluate $\int_0^3\frac{dx}{(x-1)^{2/3}}$.
- 7 Show that $\int_0^\infty\cos x\,dx$ diverges, and that $\int_{-\infty}^\infty x^3dx$ diverges although $\lim_{R\to\infty}\int_{-R}^Rx^3dx=0$.
- 8 Using $\int_0^1\log x\,dx=-1$, prove $\displaystyle\lim_{n\to\infty}\frac{(n!)^{1/n}}{n}=\frac1e$. (Take care: $\log x$ is not Riemann integrable on $[0,1]$.)
- 1 $\displaystyle\int_1^\infty\frac{dx}{x\sqrt{1+x^2}}$
- 2 $\displaystyle\int_0^1\frac{dx}{\sqrt{x+x^3}}$
- 3 $\displaystyle\int_0^\infty\frac{dx}{\sqrt x\,(1+x)}$ — and evaluate it.
- 4 $\displaystyle\int_0^1\frac{dx}{e^x-1}$
- 5 $\displaystyle\int_1^\infty\Big(1-\cos\frac1x\Big)dx$
- 6 $\displaystyle\int_2^\infty\frac{dx}{(\log x)^{10}}$
- 7 $\displaystyle\int_0^1\frac{\log x}{\sqrt x}dx$ — and evaluate it.
- 8 For which $\alpha\gt0$ does $\displaystyle\int_0^{\pi/2}\frac{dx}{(\sin x)^\alpha}$ converge?
- 9 For $0\lt a\lt b$, show that $\displaystyle\int_0^\infty\frac{dx}{x^a+x^b}$ converges if and only if $a\lt1\lt b$.
- 1 Show that $\int_1^\infty\frac{\cos x}{x}dx$ converges conditionally.
- 2 Show that $\int_1^\infty\frac{\sin x}{x^2}dx$ converges absolutely.
- 3 Show that $\int_0^\infty\sin(x^2)\,dx$ converges but $\int_0^\infty|\sin(x^2)|\,dx$ diverges.
- 4 Decide the convergence of $\int_3^\infty\frac{\log x}{x}\sin x\,dx$.
- 5 Show that $\int_1^\infty\frac{\sin x}{x}\tan^{-1}x\,dx$ converges.
- 6 Show that $\int_1^\infty x\sin x\,dx$ diverges by the Cauchy criterion.
- 7 Using Theorem 4.18, prove $\int_0^\infty\frac{\sin^2x}{x^2}dx=\frac\pi2$.
- 1 Compute $\Gamma(\tfrac72)$.
- 2 Evaluate $\int_0^\infty x^6e^{-2x}dx$.
- 3 Evaluate $\int_0^1x^4\sqrt{1-x}\,dx$ as a Beta value and as a fraction.
- 4 Evaluate $\int_0^{\pi/2}\sin^4\theta\cos^2\theta\,d\theta$.
- 5 Express $\int_0^1\frac{dx}{\sqrt{1-x^4}}$ in terms of the Beta function.
- 6 Show that $\int_0^\infty e^{-x^3}dx=\Gamma(\tfrac43)$.
- 7 Find $\mathrm{PV}\int_{-\infty}^\infty\frac{1+x}{1+x^2}dx$. Does the improper integral converge?
- 8 Prove that $\displaystyle\lim_{n\to\infty}\int_0^\infty\frac{\sin x}{x}e^{-x/n}\,dx=\frac\pi2$. (Hint: bound the tails $\int_A^\infty$ uniformly in $n$ with the second mean value theorem mentioned after Theorem 4.16.)
- 9 Prove the integral test with error bound: for $f\ge0$ decreasing on $[1,\infty)$ with $\int_1^\infty f\lt\infty$, show $0\le\sum_{n=N+1}^\infty f(n)\le\int_N^\infty f$.
E1. CSIR Part B GATE MA The integral $\displaystyle\int_0^\infty\frac{x^{p-1}}{1+x^2}\,dx$ converges if and only if (A) $0\lt p\lt1$; (B) $0\lt p\lt2$; (C) $p\gt0$; (D) $1\lt p\lt2$.
E2. CSIR Part B The value of $\displaystyle\int_0^1x^2\Big(\log\frac1x\Big)^3dx$ is (A) $\dfrac2{27}$; (B) $\dfrac19$; (C) $6$; (D) $\dfrac29$.
E3. CSIR Part B Which of the following improper integrals converges? (A) $\displaystyle\int_1^\infty\frac{dx}{x\log(1+x)}$; (B) $\displaystyle\int_0^1\frac{dx}{(x-\sin x)^{1/3}}$; (C) $\displaystyle\int_0^\infty\frac{dx}{\sqrt{x+x^4}}$; (D) $\displaystyle\int_0^{\pi/2}\frac{dx}{\sqrt{1-\cos x}}$.
E4. CSIR Part C Which of the following integrals converge? (One or more options may be correct.) (A) $\displaystyle\int_1^\infty\frac{\sin x}{x^{1/3}}dx$; (B) $\displaystyle\int_1^\infty\frac{|\sin x|}{x^{1/3}}dx$; (C) $\displaystyle\int_0^1\frac1x\sin\frac1x\,dx$; (D) $\displaystyle\int_1^\infty\frac{\sin^2x}{x}dx$.
E5. CSIR Part C Let $f:[0,\infty)\to[0,\infty)$ be continuous with $\int_0^\infty f(x)\,dx$ convergent. Which of the following must be true? (A) $\lim_{x\to\infty}f(x)=0$; (B) $f$ is bounded; (C) $\liminf_{x\to\infty}f(x)=0$; (D) $\int_0^\infty f(x)e^{-x}\,dx$ converges.
E6. CSIR Part C Which of the following are true? (A) $\Gamma(\tfrac12)=\sqrt\pi$; (B) $\displaystyle\int_0^\infty e^{-x^4}dx=\Gamma(\tfrac54)$; (C) $B(\tfrac12,\tfrac12)=\dfrac\pi2$; (D) $\displaystyle\int_0^{\pi/2}\sqrt{\tan\theta}\,d\theta=\dfrac{\pi}{\sqrt2}$.
Suggested plan — 3 lectures × 60 min.
- Definitions and the $p$-tests. Definitions 4.6–4.7, Example 1 (spend time on (d) — ask the class to find the error), Theorem 4.11 with Interactive 1 (move $p$ slowly through $1$ in each mode), the Cauchy criterion.
- Comparison. Theorems 4.13–4.14, the power-counting strategy, Examples 2–3; have students do Exercise II in pairs and report the leading behaviour at each bad point before deciding. Finish with Example 3(d), the symmetry trick for $\int\log\sin x$.
- Oscillation and special functions. Absolute vs conditional convergence, Dirichlet and Abel (Theorems 4.15–4.17), Example 4 with Interactive 2 (try $p=0.5$, $1$, $1.5$), the value $\frac\pi2$ (sketch Theorem 4.18), Example 5 (prove/disprove — do it as a class debate), then Gamma, Beta and principal values (Theorem 4.19, Examples 6–7) and Exam-corner E2, E6.
Discussion questions. (1) For series, $\sum a_n$ convergent implies $a_n\to0$; why does the analogous statement fail for integrals, and which extra hypothesis rescues it? (2) Why must $\int_{-\infty}^\infty f$ be defined with two independent limits? Give an example where a lopsided limit gives a different answer. (3) Is $\frac{\sin x}x$ "integrable" on $(0,\infty)$? In which sense — and what will Lebesgue's theory say?
Misconceptions to watch for. Applying an antiderivative across a singularity; using limit comparison for sign-changing integrands; "the integrand tends to $0$, so the integral converges" (and its converse); confusing the thresholds $p\gt1$ at $\infty$ and $p\lt1$ at $0$; using $\Gamma(s)$ for $s\le0$; treating a principal value as the value of a divergent integral.
Suggested homework. I(2, 5, 6, 8), II(3, 4, 6, 7, 9), III(1, 3, 4, 7), IV(1, 3, 4, 6, 7); Exam corner E1, E3, E4, E5.
Research question. How do you compute an oscillatory improper integral such as $\int_0^\infty\frac{\sin x}{x}\,dx$ numerically when its integrand decays so slowly that the naive partial integrals $\int_0^b$ crawl towards the answer?
Why it matters. Slowly decaying oscillatory integrals are everywhere in applied science: Fourier transforms in signal processing and MRI image reconstruction, the diffraction integrals of optics (Fresnel's $\int\sin x^2$), and option-pricing formulas in finance that invert characteristic functions. Engineers accelerate them by integrating lobe by lobe and summing the resulting alternating series cleverly.
What to do.
- Compute the lobe integrals $a_k=\int_{k\pi}^{(k+1)\pi}\frac{\sin x}{x}dx$ for $k=0,\dots,40$ (Simpson's rule with $200$ points per lobe is plenty). Check that they alternate in sign and decrease in modulus, and explain why with the Leibniz test (§2.3).
- Compare the error of the partial sums $\sum_{k\lt n}a_k$ with $\frac\pi2$ and with the bound $|a_n|\approx\frac2{n\pi}$.
- Average consecutive partial sums (one step of Euler's transformation) and repeat the averaging a few times. How many lobes do you need for $6$ correct digits, with and without averaging?
- Apply the same method to the Fresnel integral $\int_0^\infty\sin(x^2)dx$ (lobes between $\sqrt{k\pi}$ and $\sqrt{(k+1)\pi}$) and compare with the exact value $\frac12\sqrt{\frac\pi2}$.
Tools: Python (NumPy, SciPy for reference values) or a spreadsheet; Interactive 2 for the picture.
Hand in: A table of errors against the number of lobes for the raw and the averaged sums, and a short explanation of why averaging helps for alternating series.
Going further. Read about the Shanks transformation and Wynn's $\varepsilon$-algorithm, and about Longman's method for Fourier-type integrals; try $\int_0^\infty\frac{\cos x}{1+x}dx$.
🧪 Self-check quiz
Q1. $\displaystyle\int_1^\infty\frac{dx}{x^p}$ converges exactly when
Q2. The integral $\displaystyle\int_0^\infty\frac{\sin x}{x}\,dx$
Q3. $\Gamma(5)$ equals
Q4. About $\displaystyle\int_{-1}^1\frac{dx}{x}$ we can say:
Q5. $B(2,3)=\int_0^1x(1-x)^2\,dx$ equals