Integrals over an infinite range, defined as limits of ordinary integrals; their comparison tests; and substitution and integration by parts carried to infinity.
The integral test compares ∑ϕ(n) with ∫1ξϕ(x)dx as ξ→∞. That limit deserves a name of its own: it is the integral of ϕ over the infinite range [1,∞). §§184–186 define such infinite integrals, show that their theory runs parallel to that of series (convergence, comparison, the general principle), point out where the parallel breaks, and extend the rules of substitution and integration by parts to them. Infinite integrals are everywhere in later analysis, from ∫0∞e−x2dx to the Gamma function and Fourier transforms.
The idea
An ordinary integral ∫aXϕ(x)dx is (for positive ϕ) the area under the graph between a and X. Let the right-hand end run off to infinity. The region becomes infinitely long, but that does not make its area infinite: if the graph comes down to the axis fast enough, the areas ∫aXϕ approach a finite limit, and we call that limit ∫a∞ϕ. This is exactly how the sum of a series is the limit of its partial sums.
ExploreAn area that runs off to infinity
Try this.With x−s and s=2, move X to the right, then show x up to 100 and 1000. Now lower s to 1 and below. Then try cosx, xcosx and Hardy’s peaks.
∫1Xx−2dx
xφ(x)
XΦ(X)
the integrand; shaded: area from 1 to X (below the axis counts negative)
Φ(X)=∫1Xϕ
its limit
Φ(5.5)=0.81818. As X→∞ the integral converges: its value is 1. For s>1, Φ(X)=(1−X1−s)/(s−1)→1/(s−1).
What to notice
For s=2 the area creeps up to 1 and never passes it; for s≤1 it grows without bound, just like the partial sums of ∑n−s. ∫0Xcosxdx=sinX swings for ever between −1 and 1, and ∫0Xxcosxdx swings ever more widely: integrals can oscillate, as series can. The peaks never get lower, yet their total area is finite.
The last example is the surprise. For a convergent series the terms must tend to 0. For a convergent integral the integrand need not tend to 0, even when it is positive and continuous: it can have taller and taller (or, here, equally tall) spikes, provided they become thin fast enough.
Precise statements
DefinitionInfinite integral (of the first kind)Hardy: §184
Let ϕ be continuous for x≥a. If
X→∞lim∫aXϕ(x)dx=l
exists, we say that the infinite integral∫a∞ϕ(x)dxconverges and has the value l. If ∫aXϕ→∞ (or −∞) we say it diverges to ∞ (or −∞); otherwise it oscillates, finitely or infinitely, according as ∫aXϕ stays bounded or not.
Integrals ∫−∞a are defined in the same way, and ∫−∞∞ϕ is defined as ∫−∞0ϕ+∫0∞ϕwhen both converge (Examples LXXIII, 5–6). Hardy’s remarks on the definition (§184, (i)–(vii)), restated:
If Φ is any antiderivative of ϕ, then ∫a∞ϕ=Φ(∞)−Φ(a), where Φ(∞) means limx→∞Φ(x), whenever this limit exists.
If ϕ≥0, then ∫aXϕ increases with X, so the integral either converges or diverges to ∞.
General principle.∫a∞ϕ converges if and only if for every δ>0 there is an X(δ) with ∫x1x2ϕ<δ whenever x2>x1≥X(δ).
Integral test restated. If ϕ is positive and decreasing, then ∑ϕ(n) and ∫1∞ϕ(x)dx converge or diverge together.
If ∫a∞ϕ converges and b>a, then ∫b∞ϕ converges and ∫a∞ϕ=∫abϕ+∫b∞ϕ.
For positive integrands, the analogues of the theorems for positive series hold (§185).
TheoremComparison for positive integrandsHardy: §185
Let ϕ and ψ be continuous and positive for x≥a.
∫a∞ϕ converges if and only if there is a constant K with ∫aXϕ<K for all X>a.
If ∫a∞ϕ converges and ψ(x)≤Kϕ(x) for x≥a, then ∫a∞ψ converges and ∫a∞ψ≤K∫a∞ϕ.
In particular: if ϕ(x)≤Kx−s with s>1, the integral converges; if ϕ(x)≥Kx−s with K>0 and s≤1, it diverges; and if xsϕ(x)→l>0, it converges or diverges according as s>1 or s≤1 (here a>0).
There is no analogue of d’Alembert’s test, since an integrand has no “next term”. And the analogy with series breaks at one point:
WarningThe integrand need not tend to 0Hardy: §185, Fig. 47
There are positive continuous functions ϕ with ∫0∞ϕ convergent but ϕ(x)→0.
Hardy’s example: let ϕ be 0 except for a triangular peak of height 1 and base 2/(n+1)2 centred at each integer n≥1. The peak at n has area 1/(n+1)2, so ∫0Xϕ<∑n≥11/(n+1)2 for every X, and the integral converges, while ϕ(n)=1 for every n.
Finally, the rules of calculation carry over (§186).
TheoremSubstitution in an infinite integralHardy: §186 (1)
Let ∫a∞ϕ(x)dx converge. Let x=f(t), where f has a continuous derivative, f(b)=a, the substitution rule of §166 holds on every interval [b,τ], and f(t)→∞ as t→∞. Then
∫a∞ϕ(x)dx=∫b∞ϕ{f(t)}f′(t)dt,
the integral on the right being convergent.
TheoremIntegration by parts to infinityHardy: §186 (2)
If f and ϕ have continuous derivatives for x≥a, and two of the three limits below exist, then so does the third, and
The function Φ(X)=∫aXϕ plays the role of the partial sums. For positive ϕ it increases, and an increasing function of X tends to a limit exactly when it is bounded. Then compare areas.
Part 1: bounded increasing functions converge
For X2>X1, Φ(X2)−Φ(X1)=∫X1X2ϕ≥0, so Φ increases. An increasing function bounded above tends to a limit as X→∞ (the continuous-variable form of §69, proved in Ch. V); an unbounded one tends to ∞.
Part 2: compare the areas
If ψ≤Kϕ, then ∫aXψ≤K∫aXϕ≤K∫a∞ϕ for every X, by the monotonicity of the ordinary integral. So ∫aXψ is bounded, hence converges by Part 1, with the stated inequality in the limit.
Part 3: the yardstick x−s
By §181, ∫1∞x−sdx converges for s>1 and diverges for s≤1 (and likewise from any a>0). The first two claims are Part 2 and its divergence form. If xsϕ(x)→l>0, then 21lx−s<ϕ(x)<2lx−s for large x, and the first two claims apply from that point on (the finite piece ∫ax0ϕ does not affect convergence, by remark 5).
∎
ProofSubstitution in an infinite integral
Strategy
Apply the ordinary rule on a finite range, then let the upper limit go to infinity on both sides at once.
Finite ranges
For each τ>b, with ξ=f(τ), the rule of §166 gives ∫aξϕ(x)dx=∫bτϕ{f(t)}f′(t)dt.
Let τ→∞
As τ→∞, ξ=f(τ)→∞, so the left side tends to ∫a∞ϕ (a function with a limit at infinity, composed with a function tending to infinity, has the same limit). Hence the right side has the same limit, which is the statement.
Other endings
If instead ξ→∞ as τ→−∞, or as τ tends to a finite c, the same argument gives ∫a∞ϕ=−∫−∞bϕ{f(t)}f′(t)dt, or limτ→c∫bτϕ{f(t)}f′(t)dt. The second case leads to integrals of the second kind (next lesson).
∎
ProofIntegration by parts to infinity
Strategy
Write the finite-range formula with upper limit ξ and let ξ→∞.
The three quantities depending on ξ satisfy a linear relation with the constant f(a)ϕ(a). If two of them tend to limits as ξ→∞, the third, being a sum of the others and a constant, does too, and the relation holds in the limit.
∎
Worked examples and exercises
Exercise●○○Hardy: Examples LXXIII, 1 and 2
Let P and Q be polynomials of degrees r and s with positive leading coefficients, and a beyond the largest root of Q. Show that ∫a∞P(x)/Q(x)dx converges if s>r+1 and diverges otherwise. Which of these converge (a>0, c=0)?
xs−rP(x)/Q(x) tends to the ratio of the leading coefficients. Use part 3 of the comparison theorem.
Solution
P(x)/Q(x) is continuous and positive for large x (beyond the roots of P and Q, and the finite piece before that does not matter). xs−rP(x)/Q(x)→A/α>0, so the integral behaves like ∫∞x−(s−r)dx: convergent if s−r>1, divergent if s−r≤1.
So: ∫dx/x diverges (power 21); ∫dx/x4/3 converges; ∫dx/(c2+x2) converges (s−r=2); ∫xdx/(c2+x2) diverges (s−r=1; indeed it equals 21logc2+a2c2+X2→∞); ∫x2dx/(c2+x2) diverges (the integrand tends to 1).
Exercise●○○Hardy: Examples LXXIII, 7
Show that ∫−∞01+x2dx=∫0∞1+x2dx=21∫−∞∞1+x2dx=21π.
Solution
∫0Xdx/(1+x2)=arctanX→21π as X→∞, and ∫−X0dx/(1+x2)=−arctan(−X)=arctanX→21π. Both halves converge, so by definition ∫−∞∞=21π+21π=π.
Exercise●○○Hardy: Examples LXXIV, 1
Let s>1 and α>0. Show by the substitution x=tα that ∫1∞x−sdx=α∫1∞tα(1−s)−1dt, and check that both sides equal 1/(s−1).
Solution
x=tα maps [1,∞) onto [1,∞), increasingly, with dx=αtα−1dt, and x−s=t−αs. So the right-hand integrand is αt−αstα−1=αtα(1−s)−1, and the theorem applies because the left side converges (s>1). Directly: the left is 1/(s−1); on the right the exponent α(1−s)−1=−(1+α(s−1)) is less than −1, so α∫1∞t−(1+α(s−1))dt=α⋅α(s−1)1=s−11.
Exercise●●○Hardy: Examples LXXIV, 4
Show that ∫1∞(1+x)xdx=21π.
Hint
Put x=t2.
Solution
The integrand is positive and below x−3/2, so the integral converges. With x=t2 (t≥1, increasing, t→∞ as x→∞), dx=2tdt and x=t:
∫1∞(1+t2)t2tdt=2∫1∞1+t2dt=2(2π−4π)=2π.
Exercise●●○Hardy: Examples LXXV, 1
Show by integration by parts that ∫0∞(1+x)3xdx=21∫0∞(1+x)2dx=21.
Hint
Take f(x)=x and ϕ(x)=−21(1+x)−2, so that ϕ′(x)=(1+x)−3.
Solution
With f(x)=x, ϕ(x)=−21(1+x)−2: f(ξ)ϕ(ξ)=−2(1+ξ)2ξ→0, f(0)ϕ(0)=0, and ∫0∞f′ϕ=−21∫0∞(1+x)−2dx=−21 converges. Two limits exist, so the third does, and
For x>2a, the interval [x/2,x] has length x/2, and on it ϕ≥ϕ(x) because ϕ decreases. So 0≤21xϕ(x)≤∫x/2xϕ. By the general principle (remark 3), ∫x/2xϕ→0 as x→∞, since both limits pass X(δ). Hence xϕ(x)→0. Note that Hardy’s peaks do not contradict this: they are not decreasing.
Check yourself
Check yourselfTrue or false?
If ϕ is positive and continuous and ∫0∞ϕ converges, then ϕ(x)→0 as x→∞.
Why
False: Hardy’s peaks (Fig. 47) have height 1 at every integer, but total area ∑1/(n+1)2. For series the analogous statement is true; for integrals, a positive function can hide tall values on very short intervals. If ϕ is also decreasing, then ϕ→0 (and even xϕ(x)→0).
Check yourselfChoose one
Does ∫1∞x3+1xdx converge?
Why
x2⋅x3+1x→1, so by the comparison theorem the integral converges with ∫1∞x−2dx. We never need an antiderivative to decide convergence (here one exists, but it is messy).
Check yourselfEnter a number
What is ∫0∞1+x2dx? (Give a decimal.)
Why
arctanX→21π≈1.5708. An infinitely long region with a perfectly finite area, as Hardy’s remark (iv) insists.
Check yourselfChoose one
How does ∫0∞cosxdx behave?
Why
∫0Xcosxdx=sinX, which stays between −1 and 1 but has no limit: finite oscillation. Compare ∫0Xxcosxdx=XsinX+cosX−1, which is unbounded in both directions: infinite oscillation (Examples LXXIII, 3–4).
Check yourselfTrue or false?
∫−∞∞xdx converges to 0, since ∫−XXxdx=0 for every X.
Why
False. By definition ∫−∞∞ requires both∫−∞0 and ∫0∞ to converge, and ∫0Xxdx=21X2→∞. The symmetric limit lim∫−XX (today’s “principal value”) is a different and weaker notion.
Hardy’s language
The reader should not be puzzled by the use of the term infinite integral to denote something which has a definite value such as 2 or ½π.
— G. H. Hardy, §184
Summary
Summary
∫a∞ϕ=limX→∞∫aXϕ when the limit exists; otherwise the integral diverges or oscillates, just as a series does.
For positive integrands: converge iff bounded; comparison; ∫∞x−s converges iff s>1, and xsϕ(x)→l>0 decides by the same rule.
Unlike the terms of a series, a positive integrand of a convergent integral need not tend to 0 (Hardy’s peaks).
Substitution and integration by parts extend to infinite ranges by letting the upper limit tend to infinity in the finite formulas.