Chapter VIII · The convergence of infinite series and infinite integrals

Integrals of unbounded functions, and a warning about substitution

§§187–189 40 min

Integrals whose integrand becomes infinite at a point, defined as limits; substitutions that turn one kind of infinite integral into another or into an ordinary integral; and why a substitution must be monotone.

Why this matters

The definite integral of Chapter VII needs two things: a bounded interval, and a continuous integrand. The last lesson dropped the first condition. §§187–188 drop the second: what does ∫01dx/x\int_0^1 dx/\sqrt x mean, when 1/x1/\sqrt x is unbounded near 00? The answer is again a limit of ordinary integrals, and again the yardstick is a power, now (x−a)−s(x - a)^{-s} with the dividing line at s=1s = 1, the other way round. These infinite integrals of the second kind appear constantly: the Beta and Gamma integrals, ∫01log⁡x dx\int_0^1 \log x\,dx, and every substitution that maps an infinite range onto a finite one. §189 ends with a warning about substitutions that is worth learning once and for all.

The idea

Near x=0x = 0 the graph of x−sx^{-s} shoots up to infinity. The region under it between 00 and 11 is infinitely tall, just as the region in the last lesson was infinitely long. Cut it off at x=εx = \varepsilon: the area from ε\varepsilon to 11 is an ordinary integral. Now let ε→0\varepsilon \to 0. If these areas approach a limit, that limit is what we mean by ∫01x−sdx\int_0^1 x^{-s}dx.

Explore An infinitely tall region

Try this. With s=0.5s = 0.5, slide ε\varepsilon towards 10−410^{-4} and watch the lower plot. Then raise ss through 11.

∫ε1x−0.5 dx\int_{\varepsilon}^{1} x^{-0.5}\,dx: the integrand is unbounded near 0
xφ(x)
ε (log scale)∫ from ε to 1
  • the integrand (cut off at height 8); shaded: from ε\varepsilon to 1
  • ∫ε1ϕ\int_\varepsilon^1 \phi as ε\varepsilon shrinks (to the left)
  • its limit
∫0.11ϕ=1.3675\int_{0.1}^{1} \phi = 1.3675. For s<1s < 1 this tends to 1/(1−s)=21/(1-s) = 2 as ε→0\varepsilon \to 0: the region is infinitely tall but has finite area, and the integral of the second kind converges.
What to notice

For s=0.5s = 0.5 the area approaches 22, although the shaded region becomes ever taller and thinner near 00. For s≥1s \ge 1 it grows without bound. Compare the last lesson: there ∫1∞x−s\int_1^\infty x^{-s} converged for s>1s > 1; here ∫01x−s\int_0^1 x^{-s} converges for s<1s < 1. The substitution x=1/yx = 1/y turns one region into the other.

Why does the dividing line flip? Under x=1/yx = 1/y, the interval ε≤x≤1\varepsilon \le x \le 1 becomes 1≤y≤1/ε1 \le y \le 1/\varepsilon, and ∫ε1x−sdx=∫11/εys−2dy\int_\varepsilon^1 x^{-s}dx = \int_1^{1/\varepsilon} y^{s-2}dy. The new integrand is y−(2−s)y^{-(2-s)}, which has finite area out to infinity exactly when 2−s>12 - s > 1, that is s<1s < 1.

Precise statements

Definition Infinite integral of the second kind Hardy: §187

Let ϕ\phi be continuous on a<x≤Aa < x \le A, and ϕ(x)→∞\phi(x) \to \infty (or −∞-\infty) as x→ax \to a from the right. If

lim⁡ε→+0∫a+εAϕ(x) dx=l\lim_{\varepsilon \to +0} \int_{a+\varepsilon}^{A} \phi(x)\,dx = l

exists, we say that ∫aAϕ(x) dx\int_a^A \phi(x)\,dx converges and has the value ll. Similarly at the upper end, with ∫aA−ε\int_a^{A-\varepsilon}. If ϕ\phi becomes infinite at an interior point ξ\xi, we define ∫aAϕ=∫aξϕ+∫ξAϕ\int_a^A\phi = \int_a^\xi\phi + \int_\xi^A\phi when both converge; and in the same way for finitely many such points.

The same definition serves for other discontinuities: for a jump, or for an oscillating function like sin⁡(1/x)\sin(1/x). If both a finite end and an infinite range are involved, as in ∫0∞xs−1/(1+x) dx\int_0^\infty x^{s-1}/(1+x)\,dx, split the range so that each piece has only one difficulty, and require each piece to converge. Most of Hardy’s remarks on infinite integrals of the first kind (last lesson) carry over word for word.

Theorem The standard integral of the second kind Hardy: §187; Examples LXXVI, 2

∫aAdx(x−a)s\displaystyle\int_a^A \frac{dx}{(x-a)^s} converges if s<1s < 1, with value (A−a)1−s1−s\dfrac{(A - a)^{1-s}}{1 - s}, and diverges if s≥1s \ge 1.

Theorem Comparison near a singular point Hardy: Examples LXXVI, 1 and 3

Let ϕ\phi be continuous on a<x≤Aa < x \le A.

  1. If ϕ≥0\phi \ge 0, then ∫aAϕ\int_a^A\phi converges if and only if ∫a+εAϕ<K\int_{a+\varepsilon}^A \phi < K for some constant KK and every ε>0\varepsilon > 0; otherwise it diverges to ∞\infty.
  2. If 0≤ϕ(x)<K(x−a)−s0 \le \phi(x) < K(x - a)^{-s} with s<1s < 1, the integral converges.
  3. If ϕ(x)>K(x−a)−s\phi(x) > K(x - a)^{-s} with K>0K > 0 and s≥1s \ge 1, the integral diverges.

Substitution can move a difficulty from one place to another, or remove it altogether (§188).

Theorem Substitution onto a finite range Hardy: §188

Suppose ∫a∞ϕ(x) dx\int_a^\infty \phi(x)\,dx converges, and x=f(t)x = f(t) maps b≤t<cb \le t < c increasingly onto a≤x<∞a \le x < \infty, with f′f' continuous. Then

∫a∞ϕ(x) dx=∫bcϕ{f(t)}f′(t) dt,\int_a^\infty \phi(x)\,dx = \int_b^c \phi\{f(t)\}f'(t)\,dt,

where the right-hand side is an infinite integral of the second kind if ϕ{f(t)}f′(t)\phi\{f(t)\}f'(t) is unbounded near cc, and an ordinary integral if it extends continuously to t=ct = c.

For example, with x=t/(1−t)x = t/(1-t),

∫0∞dx(1+x)m=∫01(1−t)m−2 dt(m>1),\int_0^\infty \frac{dx}{(1+x)^m} = \int_0^1 (1-t)^{m-2}\,dt \qquad (m > 1),

an integral of the second kind when 1<m<21 < m < 2 and an ordinary integral when m≥2m \ge 2. Both sides equal 1/(m−1)1/(m-1).

A substitution must run one way

The substitution rule ∫ϕ(x) dx=∫ϕ{f(t)}f′(t) dt\int \phi(x)\,dx = \int \phi\{f(t)\}f'(t)\,dt is proved for a function x=f(t)x = f(t) defined on the whole range. If instead we start from y=g(x)y = g(x) and solve for xx, there may be several solutions, and we must choose the right one on each part of the range. Hardy’s example is J=∫17(x2−6x+13) dx=48J = \int_1^7 (x^2 - 6x + 13)\,dx = 48 with y=x2−6x+13y = x^2 - 6x + 13.

Explore Which area did the substitution compute?

Try this. Choose each of the three ways of computing JJ in the variable yy, and look at which area is shaded.

J=∫17(x2−6x+13) dxJ = \int_1^7 (x^2 - 6x + 13)\,dx with the substitution y=x2−6x+13y = x^2 - 6x + 13
xy
  • y=x2−6x+13y = x^2 - 6x + 13
  • the levels y=8y = 8 (at x=1x = 1) and y=20y = 20 (at x=7x = 7)
  • the minimum y=4y = 4 at x=3x = 3, where dx/dydx/dy changes sign
Compute J in the variable y by
As xx runs from 1 to 3, yy falls from 8 to 4 on the branch x=3−y−4x = 3 - \sqrt{y-4}, with dx/dy<0dx/dy < 0; as xx runs from 3 to 7, yy rises from 4 to 20 on the other branch. So J=∫84y (−12y−4)dy+∫420y dy2y−4=323+1123=48J = \int_8^4 y\,\Big(-\frac{1}{2\sqrt{y-4}}\Big)dy + \int_4^{20} \frac{y\,dy}{2\sqrt{y-4}} = \frac{32}{3} + \frac{112}{3} = 48, the correct value.
What to notice

As xx runs from 11 to 77, yy does not simply run from 88 to 2020: it falls to 44 at x=3x = 3 and then rises. Using one branch for the whole range computes the area over a wrong interval (5≤x≤75 \le x \le 7, or even −1≤x≤1-1 \le x \le 1), which is why both signs give ±80/3\pm 80/3. Splitting at the turning point, with the right branch on each side, gives 4848.

Rule Split a substitution where it turns Hardy: §189

When transforming ∫abϕ(x) dx\int_a^b \phi(x)\,dx by a relation y=g(x)y = g(x), split [a,b][a, b] into intervals on which gg is monotone. On each, xx is a single-valued function of yy with dx/dydx/dy of one sign, and the substitution rule applies to that piece.

Proofs

Proof The standard integral of the second kind
Strategy

Compute the ordinary integral from a+εa + \varepsilon to AA and let ε→0\varepsilon \to 0. Everything depends on whether ε1−s\varepsilon^{1-s} tends to 00.

The integral from a+εa + \varepsilon

For s≠1s \ne 1, an antiderivative of (x−a)−s(x - a)^{-s} is (x−a)1−s/(1−s)(x-a)^{1-s}/(1-s), so

∫a+εAdx(x−a)s=(A−a)1−s−ε1−s1−s.\int_{a+\varepsilon}^A \frac{dx}{(x-a)^s} = \frac{(A-a)^{1-s} - \varepsilon^{1-s}}{1-s}.

s<1s < 1: convergence

Then 1−s>01 - s > 0 and ε1−s→0\varepsilon^{1-s} \to 0, so the integral tends to (A−a)1−s/(1−s)(A-a)^{1-s}/(1-s).

s>1s > 1: divergence

Then ε1−s=1/εs−1→∞\varepsilon^{1-s} = 1/\varepsilon^{s-1} \to \infty, and since 1−s<01 - s < 0 the whole expression tends to +∞+\infty.

s=1s = 1: divergence

The substitution x−a=1/yx - a = 1/y turns ∫a+εAdx/(x−a)\int_{a+\varepsilon}^A dx/(x-a) into ∫1/(A−a)1/εdy/y\int_{1/(A-a)}^{1/\varepsilon} dy/y, which tends to ∞\infty as ε→0\varepsilon \to 0 by the harmonic case of §181.

Proof Comparison near a singular point
Strategy

Exactly as for the first kind: for positive ϕ\phi the integral from a+εa + \varepsilon increases as ε\varepsilon decreases, so it either has a limit or tends to ∞\infty; then compare with (x−a)−s(x - a)^{-s}.

Monotonicity in ε\varepsilon

If ε′<ε\varepsilon' < \varepsilon then ∫a+ε′Aϕ−∫a+εAϕ=∫a+ε′a+εϕ≥0\int_{a+\varepsilon'}^A\phi - \int_{a+\varepsilon}^A\phi = \int_{a+\varepsilon'}^{a+\varepsilon}\phi \ge 0. So I(ε)=∫a+εAϕI(\varepsilon) = \int_{a+\varepsilon}^A\phi increases as ε→0\varepsilon \to 0, and tends to a limit if it is bounded, to ∞\infty if not. This is part 1.

Comparison

If 0≤ϕ<K(x−a)−s0 \le \phi < K(x-a)^{-s} with s<1s < 1, then I(ε)≤K∫a+εA(x−a)−sdx≤K(A−a)1−s/(1−s)I(\varepsilon) \le K\int_{a+\varepsilon}^A (x-a)^{-s}dx \le K(A-a)^{1-s}/(1-s), a bound independent of ε\varepsilon. If ϕ>K(x−a)−s\phi > K(x-a)^{-s} with s≥1s \ge 1, then I(ε)≥K∫a+εA(x−a)−sdx→∞I(\varepsilon) \ge K\int_{a+\varepsilon}^A (x-a)^{-s}dx \to \infty.

Proof Substitution onto a finite range
Strategy

Apply the ordinary rule on [b,τ][b, \tau] with τ<c\tau < c, and let τ→c\tau \to c.

Finite pieces

For b≤τ<cb \le \tau < c, with ξ=f(τ)\xi = f(\tau), §166 gives ∫aξϕ(x) dx=∫bτϕ{f(t)}f′(t) dt\int_a^\xi \phi(x)\,dx = \int_b^\tau \phi\{f(t)\}f'(t)\,dt.

Let τ→c\tau \to c

As τ→c\tau \to c, ξ=f(τ)→∞\xi = f(\tau) \to \infty, so the left side tends to ∫a∞ϕ\int_a^\infty\phi. Hence the right side tends to the same limit: by definition, the integral of the second kind ∫bc\int_b^c converges and has that value.

When the new integrand is continuous at cc

If ϕ{f(t)}f′(t)\phi\{f(t)\}f'(t) extends to a continuous function on [b,c][b, c], then ∫bτ\int_b^\tau is a continuous function of τ\tau (§165), so its limit as τ→c\tau \to c is the ordinary integral ∫bc\int_b^c. The substitution has turned an infinite integral into a finite one.

Proof Splitting a substitution at its turning point
Strategy

On an interval where gg is strictly monotone, it has a continuous inverse, and the ordinary substitution rule applies with that inverse. Then add the pieces.

The pieces of Hardy’s example

y=x2−6x+13=(x−3)2+4y = x^2 - 6x + 13 = (x-3)^2 + 4 decreases on [1,3][1, 3] (from 88 to 44) and increases on [3,7][3, 7] (from 44 to 2020). The inverses are x=3−y−4x = 3 - \sqrt{y - 4} and x=3+y−4x = 3 + \sqrt{y-4}, with dx/dy=∓12(y−4)−1/2dx/dy = \mp\tfrac12(y-4)^{-1/2}.

Substitute on each piece

J=∫84y(−12y−4)dy+∫420y2y−4 dy=∫48y dy2y−4+∫420y dy2y−4.J = \int_8^4 y\left(-\frac{1}{2\sqrt{y-4}}\right)dy + \int_4^{20} \frac{y}{2\sqrt{y-4}}\,dy = \int_4^8 \frac{y\,dy}{2\sqrt{y-4}} + \int_4^{20}\frac{y\,dy}{2\sqrt{y-4}}.

Both are integrals of the second kind at y=4y = 4 (convergent, since the integrand is like (y−4)−1/2(y-4)^{-1/2}).

Evaluate

An antiderivative of y2y−4\dfrac{y}{2\sqrt{y-4}} is 13(y−4)3/2+4(y−4)1/2\tfrac13(y-4)^{3/2} + 4(y-4)^{1/2}. It is 00 at y=4y = 4, 323\tfrac{32}{3} at y=8y = 8 and 1123\tfrac{112}{3} at y=20y = 20. So J=323+1123=48J = \tfrac{32}{3} + \tfrac{112}{3} = 48, as direct integration gives.

Worked examples and exercises

Exercise ●○○ Hardy: Examples LXXVI, 4

Which of these converge?

∫aAdx(x−a)(A−x),∫aAdx(A−x)x−a3,∫aAdx(A−x)A−x3.\int_a^A \frac{dx}{\sqrt{(x-a)(A-x)}}, \qquad \int_a^A \frac{dx}{(A-x)\sqrt[3]{x-a}}, \qquad \int_a^A \frac{dx}{(A - x)\sqrt[3]{A-x}}.
Hint
Look at each end separately and find the power of (x−a)(x - a) or (A−x)(A - x) that the integrand behaves like there.
Solution

First. Near aa the integrand is about (A−a)−1/2(x−a)−1/2(A-a)^{-1/2}(x-a)^{-1/2}; near AA it is about (A−a)−1/2(A−x)−1/2(A-a)^{-1/2}(A-x)^{-1/2}. Both exponents are 12<1\tfrac12 < 1: converges. (Its value is π\pi, Examples LXXVI, 22.)

Second. Near aa it behaves like (x−a)−1/3(x-a)^{-1/3}, fine; near AA like (A−x)−1(A-x)^{-1}: diverges.

Third. The integrand is (A−x)−4/3(A-x)^{-4/3}; the exponent exceeds 11: diverges.

Exercise ●●○ Hardy: Examples LXXVI, 7

Show that ∫0πdx(sin⁡x)s\displaystyle\int_0^\pi \frac{dx}{(\sin x)^s} converges if and only if s<1s < 1. (In particular ∫0πdx/sin⁡x\int_0^\pi dx/\sqrt{\sin x} converges: Examples LXXVI, 6.)

Hint
sin⁡x/x→1\sin x / x \to 1 as x→0x \to 0, and the integrand is symmetric about 12π\tfrac12\pi.

Solution

For s≤0s \le 0 the integrand is continuous and there is nothing to prove, so let s>0s > 0. The integrand is infinite at both ends, so split at 12π\tfrac12\pi; by symmetry (sin⁡(π−x)=sin⁡x\sin(\pi - x) = \sin x) it is enough to study ∫0π/2\int_0^{\pi/2}. On (0,12π](0, \tfrac12\pi] we have 2πx≤sin⁡x≤x\tfrac{2}{\pi}x \le \sin x \le x. So x−s≤(sin⁡x)−s≤(π2)sx−sx^{-s} \le (\sin x)^{-s} \le (\tfrac{\pi}{2})^s x^{-s}, and by the comparison theorem the integral converges exactly when ∫0x−sdx\int_0 x^{-s}dx does, that is, when s<1s < 1.

Exercise ●●○ Hardy: Examples LXXVI, 11

Show that ∫0∞xs−11+x dx\displaystyle\int_0^\infty \frac{x^{s-1}}{1+x}\,dx converges if and only if 0<s<10 < s < 1.

Hint
Split at x=1x = 1. Near 00 the integrand behaves like xs−1x^{s-1}; near ∞\infty like xs−2x^{s-2}.

Solution

By definition it is ∫01+∫1∞\int_0^1 + \int_1^\infty, and both must converge. On (0,1](0, 1], 12xs−1≤xs−11+x≤xs−1\tfrac12 x^{s-1} \le \dfrac{x^{s-1}}{1+x} \le x^{s-1}, so ∫01\int_0^1 converges iff ∫01x−(1−s)dx\int_0^1 x^{-(1-s)}dx does, that is iff 1−s<11 - s < 1, i.e. s>0s > 0. On [1,∞)[1, \infty), x2−s⋅xs−11+x=x1+x→1x^{2-s}\cdot\dfrac{x^{s-1}}{1+x} = \dfrac{x}{1+x} \to 1, so ∫1∞\int_1^\infty converges iff 2−s>12 - s > 1, i.e. s<1s < 1. Both together: 0<s<10 < s < 1. (Its value, π/sin⁡sπ\pi/\sin s\pi, comes much later.)

Exercise ●●○ Hardy: Examples LXXVI, 20

Show by the substitution x=t/(1−t)x = t/(1-t) that, for l,m>0l, m > 0,

∫0∞xl−1(1+x)l+m dx=∫01tl−1(1−t)m−1 dt.\int_0^\infty \frac{x^{l-1}}{(1+x)^{l+m}}\,dx = \int_0^1 t^{l-1}(1-t)^{m-1}\,dt .
Solution

x=t/(1−t)x = t/(1-t) maps 0≤t<10 \le t < 1 increasingly onto 0≤x<∞0 \le x < \infty, with 1+x=1/(1−t)1 + x = 1/(1-t) and dx=dt/(1−t)2dx = dt/(1-t)^2. Then

xl−1(1+x)l+m dx=tl−1(1−t)1−l (1−t)l+m dt(1−t)2=tl−1(1−t)m−1 dt.\frac{x^{l-1}}{(1+x)^{l+m}}\,dx = t^{l-1}(1-t)^{1-l}\,(1-t)^{l+m}\,\frac{dt}{(1-t)^2} = t^{l-1}(1-t)^{m-1}\,dt .

Both integrals converge: the left one behaves like xl−1x^{l-1} near 00 and like x−m−1x^{-m-1} at infinity; the right one like tl−1t^{l-1} near 00 and (1−t)m−1(1-t)^{m-1} near 11. Applying the substitution on [ε,τ][\varepsilon, \tau] and letting ε→0\varepsilon \to 0, τ→1\tau \to 1 gives the equality. The right-hand side is Euler’s Beta integral B(l,m)B(l, m).

Exercise ●●○ Hardy: §189, Example

Verify the rule of §189 on ∫01(4x2−x+116)dx\displaystyle\int_0^1 \left(4x^2 - x + \tfrac{1}{16}\right)dx with the substitution y=4x2−x+116y = 4x^2 - x + \tfrac1{16}.

Hint
y=4(x−18)2y = 4(x - \tfrac18)^2. Where is the turning point, and what are the two branches?

Solution

Directly: 43−12+116=4348\tfrac43 - \tfrac12 + \tfrac1{16} = \tfrac{43}{48}.

Now y=4(x−18)2y = 4(x - \tfrac18)^2, which falls from 116\tfrac1{16} to 00 on [0,18][0, \tfrac18] and rises from 00 to 4916\tfrac{49}{16} on [18,1][\tfrac18, 1]. The branches are x=18∓12yx = \tfrac18 \mp \tfrac12\sqrt y, with dx/dy=∓14ydx/dy = \mp\dfrac{1}{4\sqrt y}. Hence

∫01y dx=∫1/160y(−14y)dy+∫049/16y4y dy=14∫01/16y dy+14∫049/16y dy.\int_0^1 y\,dx = \int_{1/16}^0 y\left(-\frac{1}{4\sqrt y}\right)dy + \int_0^{49/16} \frac{y}{4\sqrt y}\,dy = \frac14\int_0^{1/16}\sqrt y\,dy + \frac14\int_0^{49/16}\sqrt y\,dy .

With ∫0cy dy=23c3/2\int_0^c \sqrt y\,dy = \tfrac23 c^{3/2}: 16(164+34364)=344384=4348\tfrac16\left(\tfrac1{64} + \tfrac{343}{64}\right) = \tfrac{344}{384} = \tfrac{43}{48}. A single branch over yy from 116\tfrac1{16} to 4916\tfrac{49}{16} would have given 16⋅34264=5764\tfrac16 \cdot \tfrac{342}{64} = \tfrac{57}{64}, which is wrong.

Check yourself

Check yourself Choose one

What is ∫−11dxx2\displaystyle\int_{-1}^{1}\frac{dx}{x^2}?

Why

The integrand is unbounded at the interior point 00, so the integral means ∫−10+∫01\int_{-1}^0 + \int_0^1, and ∫ε1x−2dx=1/ε−1→∞\int_\varepsilon^1 x^{-2}dx = 1/\varepsilon - 1 \to \infty. The answer −2-2 (negative, for a positive integrand!) comes from applying the fundamental theorem across a point where it does not hold. Symmetry would only matter if both halves converged.

Check yourself Enter a number

What is ∫01dxx\displaystyle\int_0^1 \frac{dx}{\sqrt x}?

Why

∫ε1x−1/2dx=2−2ε→2\int_\varepsilon^1 x^{-1/2}dx = 2 - 2\sqrt\varepsilon \to 2. This is the case s=12s = \tfrac12 of the standard integral: (A−a)1−s/(1−s)=1/12=2(A-a)^{1-s}/(1-s) = 1/\tfrac12 = 2.

Check yourself True or false?

∫01x−sdx\int_0^1 x^{-s}dx and ∫1∞x−sdx\int_1^\infty x^{-s}dx converge for the same values of ss.

Why

False: the first converges for s<1s < 1, the second for s>1s > 1, and neither for s=1s = 1. Near 00 a smaller power is milder; near ∞\infty a larger power decays faster. So ∫0∞x−sdx\int_0^\infty x^{-s}dx converges for no ss at all.

Check yourself Choose one

Hardy’s substitution y=x2−6x+13y = x^2 - 6x + 13 in ∫17y dx\int_1^7 y\,dx, using x=3+y−4x = 3 + \sqrt{y-4} for yy from 88 to 2020, gives 80/380/3 instead of 4848. What went wrong?

Why

The relation y=g(x)y = g(x) must be inverted separately on each interval where gg is monotone. The unboundedness at y=4y = 4 is harmless (it is a convergent integral of the second kind) and appears only in the correct, split computation.

Hardy’s language

Some care has occasionally to be exercised in applying the rule for transformation by substitution.

— G. H. Hardy, §189

Summary

Summary
  • If ϕ\phi is unbounded near aa, ∫aAϕ\int_a^A\phi means lim⁡ε→+0∫a+εAϕ\lim_{\varepsilon\to+0}\int_{a+\varepsilon}^A\phi; with several bad points, split the range and require every piece to converge.
  • ∫aA(x−a)−sdx\int_a^A (x-a)^{-s}dx converges iff s<1s < 1, the opposite of the condition at infinity; comparison with it decides most integrals.
  • A substitution can turn an infinite range into a finite one with an unbounded integrand (or a perfectly ordinary one), and back.
  • A substitution defined by y=g(x)y = g(x) must be split where gg turns, with the correct branch on each piece.