Improper integrals: infinite limits, infinite integrands and comparison
Ten improper integrals: 1/x³ to infinity, the p-test at infinity and at zero, an integrand that blows up inside the interval, ln x on (0, 1], 1/(1 + x²) over the whole line, x e^(−x²), a comparison that proves convergence, a partial-fraction limit that gives ln 2, and the Cauchy distribution whose mean does not exist, with worked solutions and the mistakes of integrating straight through a singularity, canceling ∞ − ∞ and comparing in the wrong direction.
Before you start
Probability lives on improper integrals: a density has to integrate to 1 over an infinite range, and a mean or a variance exists only if another improper integral converges. The definition is always the same: replace the infinite limit, or the point where the integrand blows up, with a variable, integrate, and take a limit. When no antiderivative is available, a comparison with xp1 often settles convergence. The work goes wrong in a few places: a singularity inside the interval integrated straight through, two divergent pieces subtracted as if ,∞−∞=0, a symmetric limit taken as the value of a two-sided integral, a comparison made in the direction that proves nothing, and the p-test used with the wrong threshold near .0. These ten problems cover each of them.
Infinite limit..∫a∞f(x)dx=b→∞lim∫abf(x)dx. It converges if the limit is a finite number and diverges otherwise.
Infinite integrand. If f blows up at ,a, then .∫abf(x)dx=ε→0+lim∫a+εbf(x)dx. If it blows up at a point c inside ,[a,b], split at ;c; the integral converges only if both pieces do.
Both ends infinite.,∫−∞∞f=∫−∞0f+∫0∞f, with each piece a separate limit. Both must converge.
The p-test.∫1∞xpdx converges exactly when ,p>1, and ∫01xpdx converges exactly when .p<1. At p=1 both diverge.
Comparison. If 0≤f(x)≤g(x) for :x≥a: if ∫a∞g converges, so does ;∫a∞f; if ∫a∞f diverges, so does .∫a∞g. Being smaller than something divergent, or larger than something convergent, proves nothing.
Show that ∫1∞xpdx converges exactly when ,p>1, and find its value. Use the result on p=23 and .p=1.
··
Evaluate ,∫01xdx, then say for which p the integral ∫01xpdx converges.
··
Does ∫−12x2dx converge?
··
Evaluate .∫01lnxdx.
··
Evaluate .∫−∞∞1+x2dx.
··
Evaluate .∫0∞xe−x2dx.
···
Show that ∫1∞x3+1dx converges, and give an upper bound for it.
···
Evaluate .∫1∞x(x+1)dx.
···
Does ∫−∞∞1+x2xdx converge? What does the answer say about the mean of the Cauchy distribution, whose density is ?π(1+x2)1?
Answers
21
Converges to p−11 for ,p>1, diverges for ;p≤1;p=23 gives ,2,p=1 diverges
;2;∫01x−pdx converges to 1−p1 exactly when p<1
No: the integral diverges
−1
π
21
It converges, and its value is less than 2
ln2≈0.693
It diverges, so the Cauchy distribution has no mean
Worked solutions
Problem 1
Evaluate .∫1∞x3dx.
∫1bx−3dx=[−2x21]1b=21−2b21Replace ∞ with b and integrate.
2b21→0 as b→∞Take the limit.
21The integral converges: the area under x31 from 1 out to infinity is finite.
Problem 2
Show that ∫1∞xpdx converges exactly when ,p>1, and find its value. Use the result on p=23 and .p=1.
For :p=1:∫1bx−pdx=1−pb1−p−1The power rule.
If ,p>1,b1−p→0 and the integral is ;1−p−1=p−11; if ,p<1,b1−p→∞The exponent 1−p is negative in the first case and positive in the second.
For :p=1:∫1bxdx=lnb→∞The borderline case diverges, slowly.
Converges to p−11 for ,p>1, diverges for ;p≤1;p=23 gives ,2,p=1 divergesx1 and x1.011 look almost the same, but only the second has a finite area.
Problem 3
Evaluate ,∫01xdx, then say for which p the integral ∫01xpdx converges.
x1 blows up at ,x=0, so ∫01xdx=ε→0+lim[2x]ε1The problem is at the left end.
=limε→0+(2−2ε)=2Converges.
For :p=1:∫ε1x−pdx=1−p1−ε1−pThe same power rule as Problem 2.
ε1−p→0 when p<1 and →∞ when ;p>1;p=1 gives −lnε→∞Near 0 the roles of the exponents are reversed.
;2;∫01x−pdx converges to 1−p1 exactly when p<1Near 0 a small power blows up slowly enough, and near ∞ a large power decays fast enough. p=1 fails at both ends.
Problem 4
Does ∫−12x2dx converge?
x21 blows up at ,x=0, inside [−1,2]Check the interval for a zero in a denominator before integrating.
,∫−12=∫−10+∫02, and both pieces must convergeSplit at the singularity.
∫ε2x2dx=ε1−21→∞The right piece diverges: this is p=2≥1 near .0.
No: the integral divergesOne divergent piece is enough. The left piece diverges as well.
Problem 5
Evaluate .∫01lnxdx.
lnx→−∞ as ,x→0+, so take ε→0+lim∫ε1lnxdxAn infinite integrand at the left end.
∫lnxdx=xlnx−xBy parts with ,u=lnx,.dv=dx.
∫ε1lnxdx=(0−1)−(εlnε−ε)Evaluate at the limits.
εlnε=1/εlnε→0L'Hôpital: .−1/ε21/ε=−ε→0. The same limit justifies 0log0=0 in entropy.
−1Negative, because lnx<0 on :(0,1): the region lies below the axis.
Problem 6
Evaluate .∫−∞∞1+x2dx.
∫−∞∞=∫−∞0+∫0∞Both ends are infinite, so split and take two separate limits.
∫0b1+x2dx=arctanb→2πarctan is the antiderivative, and it levels off at .2π.
∫a01+x2dx=−arctana→2π as a→−∞The integrand is even, so the left half matches the right.
πBoth pieces converge. Dividing by π gives the Cauchy density ,π(1+x2)1, which integrates to 1 (see Problem 10).
Problem 7
Evaluate .∫0∞xe−x2dx.
Let ,u=x2,du=2xdxThe x in front is exactly what the substitution needs.
∫0bxe−x2dx=21∫0b2e−udu=21(1−e−b2)Change the limits with the variable: u runs from 0 to .b2.
e−b2→0Take the limit.
21The Gaussian e−x2 on its own has no elementary antiderivative; with the extra x it does.
Problem 8
Show that ∫1∞x3+1dx converges, and give an upper bound for it.
,x3+1>x3, so x3+1>x3/2 and x3+11<x3/21 for x≥1A bigger denominator makes a smaller fraction.
∫1∞x3/2dx=3/2−11=2The p-test with p=23>1 (Problem 2).
The integrand is positive and below a function whose integral convergesComparison in the direction that proves convergence.
It converges, and its value is less than 2Numerically it is about .1.895. There is no elementary antiderivative, so the comparison is the proof.
Problem 9
Evaluate .∫1∞x(x+1)dx.
x(x+1)1=x1−x+11Partial fractions, by the cover-up rule.
∫1b(x1−x+11)dx=[lnx+1x]1b=lnb+1b−ln21Combine the logs before taking any limit.
,b+1b→1, so lnb+1b→0The two logs grow together and their difference settles down.
ln2≈0.693The integrand behaves like x21 for large ,x, so convergence was expected; the split pieces x1 and x+11 each diverge on their own.
Problem 10
Does ∫−∞∞1+x2xdx converge? What does the answer say about the mean of the Cauchy distribution, whose density is ?π(1+x2)1?
Split at :0:∫0b1+x2xdx=21ln(1+b2)Substitute .u=1+x2.
21ln(1+b2)→∞ as b→∞The right half diverges, and by symmetry the left half goes to .−∞.
The integral divergesA two-sided integral needs both halves to converge on their own.
A symmetric cutoff gives ,∫−bb1+x2xdx=0, but ∫−b2b1+x2xdx=21ln1+b21+4b2→ln2Different ways of letting the limits go to infinity give different answers, which is why neither one is the value.
It diverges, so the Cauchy distribution has no meanThe mean would be ,π1∫x⋅1+x21dx, this integral divided by .π. Sample averages of Cauchy draws never settle down.
Where this goes wrong
1. Integrating straight through a singularity
The antiderivative exists on both sides of the problem point, so the formula F(b)−F(a) seems to apply.
,∫−12x2dx, with antiderivative −x1Right so far: Problem 4.
“Evaluate the antiderivative at the limits.”The habit that causes the mistake: the fundamental theorem used without checking that the integrand is continuous on .[−1,2].
[−x1]−12=−21−1=−23A positive integrand cannot have a negative integral. x21 blows up at ,0, inside the interval, and the integral diverges: ,∫ε2x2dx=ε1−21, which is 99.5 at ε=0.01 and keeps growing.
2. Canceling ∞ − ∞
After partial fractions the integral looks like two simpler ones.
x(x+1)1=x1−x+11Right so far: Problem 9, step 1.
“Integrate each term from 1 to ∞ separately.”The habit that causes the mistake: splitting an improper integral into pieces that are themselves improper.
,∫1∞xdx−∫1∞x+1dx=∞−∞, so the integral diverges∞−∞ has no value, so it decides nothing. Keep the pieces together up to a finite :b:.lnb+1b+ln2→ln2≈0.693. The integral converges.
3. Taking a symmetric limit as the value
For an odd integrand, cutting off at −b and b gives 0 every time, which looks like the answer.
,∫−∞∞1+x2xdx, and ∫−bb1+x2xdx=0 for every bRight so far: Problem 10, step 4.
“So the limit as b→∞ is .0.”The habit that causes the mistake: one particular way of going to infinity treated as the definition.
The integral is ,0, so the Cauchy distribution has mean 0The definition takes the two ends separately, and ∫0b already goes to .∞. Cutting off at −b and 2b gives ln2≈0.693 instead of .0. The integral diverges, and the Cauchy distribution has no mean at all, though its median is .0.
4. Comparing in the wrong direction
Comparison only works one way in each case, and the wrong way still looks like an argument.
1+x21≤x1 for x≥1Right so far: x≤1+x2 for every .x.
“∫1∞xdx diverges.”The habit that causes the mistake: knowing only that the bigger function's integral diverges.
So ∫1∞1+x2dx divergesBeing below something infinite says nothing: the smaller function's area could still be finite, and here it is, 2π−arctan1=4π≈0.785 (Problem 6). To prove divergence, show the integrand is above a divergent one; to prove convergence, below a convergent one, as 1+x21≤x21 does.
5. Using the p > 1 rule near zero
The p-test is usually learned on ,[1,∞), and the threshold gets carried over unchanged.
,∫01x2dx, which is improper at 0Right so far: the setting of Problem 3.
“,p=2>1, so it converges, to .p−11.”The habit that causes the mistake: the rule for [1,∞) used on .(0,1].
∫01x2dx=1Near 0 the test is reversed: it converges only for .p<1. Here ,∫ε1x2dx=ε1−1, which is 99 at ε=0.01 and 999 at :ε=0.001: it diverges. A larger power decays faster at infinity but blows up faster at .0.
Print this set: improper-integrals.pdf (problems, answers, and worked solutions on separate pages).