Improper integrals and geometric applications
Introduced| English | Meaning |
|---|---|
| improper integral/ɪmˈprɒpə ˈɪntɪɡrəl/ | An integral defined by a limit because an integration interval is unbounded or the integrand is unbounded at a boundary or interior point. |
| arc length/ɑːk leŋθ/ | The distance along a curved part of a circle or another curve. |
| principal value/ˈprɪnsɪpl ˈvæljuː/ | A value obtained by a prescribed symmetric limiting procedure, which need not equal an ordinary convergent improper integral. |
Finite volume, infinite surface
- Rotate $y=1/x$, $x\ge1$, about the $x$-axis. The resulting horn encloses a finite volume but has infinite lateral surface area. A physical coating has thickness; this mathematical example compares ideal geometric quantities.
- Improper integrals 反常积分 let us make that comparison using limits. We will also distinguish geometric area, volume, arc length 弧长 and surface area, and choose a formula from the geometry before integrating.
An infinite endpoint is a limit
- If $f$ is integrable on every finite interval $[a,b]$, define $\int_a^\infty f(x)\,dx$ as $\lim_{b\to\infty}\int_a^b f(x)\,dx$. It converges only when this limit exists and is finite. Unbounded cut-off values, or oscillation without a limit, mean divergence.
- For an interval extending to both infinities, choose a finite split point $c$ and require $\int_{-\infty}^c f$ and $\int_c^\infty f$ to converge separately. Taking only symmetric bounds $[-b,b]$ can hide divergent pieces; it is not the definition of the ordinary improper integral.
Singular endpoints need separate limits
- At a singular right endpoint $b$, use $\lim_{t\to b^-}\int_a^t f(x)\,dx$; at a singular left endpoint $a$, use $\lim_{t\to a^+}\int_t^b f(x)\,dx$. Keep the remaining endpoints regular, or split again. Never insert a singular endpoint into an antiderivative as an ordinary number.
- If $c$ is singular inside $(a,b)$, split at $c$. Both one-sided integrals must have finite limits before adding them. For several singularities or infinite endpoints, isolate each one and check every required limit.
Symmetric cancellation is a different question
- For $1/x$ on $[-1,1]$, the left cut-off integral is $\ln\alpha\to-\infty$ as $\alpha\to0^+$; the right one is $-\ln\beta\to+\infty$ as $\beta\to0^+$. Neither half converges, so their ordinary improper sum does not exist.
- With equal cut-offs $\alpha=\beta$, the combined value is zero: this is the Cauchy principal value 主值. But the combined expression is $\ln(\alpha/\beta)$; choosing $\alpha=2\beta$ gives $\ln2$. A special pairing cannot replace independent convergence.
Since $\lim_{\varepsilon \to 0}\left(\int_{-1}^{-\varepsilon} + \int_\varepsilon^1\right)\frac{dx}{x} = 0$, the improper integral $\int_{-1}^1 \frac{dx}{x}$ converges to 0.
Difficulty 1/5 · 100 coins + a small bonus on first completion today
Each half must converge on its own.
For $1/x$ on $[-1,1]$, compute the cut-off sum using independent positive cut-offs $\alpha$ and $\beta$. Compare $\alpha=\beta$ with $\alpha=2\beta$ and decide whether the ordinary improper integral exists.
Compare your argument with the model and checklist.
Derive the p-test at infinity
- For $p\ne1$, the finite cut-off integral $\int_1^b x^{-p}\,dx$ equals $(b^{1-p}-1)/(1-p)$. If $p>1$, the power of $b$ tends to zero, giving $1/(p-1)$. For example, $p=3$ gives $1/2$.
- If $p<1$, the same expression grows without bound. At $p=1$, the integral is $\ln b$, which also diverges. Thus $\int_1^\infty x^{-p}\,dx$ converges exactly when $p>1$; the borderline case needs the logarithm, not division by zero.
Evaluate $\int_1^\infty x^{-3}\,dx$.
Difficulty 2/5 · 140 coins + a small bonus on first completion today
$[-\tfrac{1}{2x^2}]_1^\infty = \tfrac12$.
Near zero, the threshold reverses
- For $p\ne1$, $\int_\varepsilon^1 x^{-p}\,dx$ equals $(1-\varepsilon^{1-p})/(1-p)$. As $\varepsilon\to0^+$, the limit is $1/(1-p)$ when $p<1$, and is infinite when $p>1$. At $p=1$, $-\ln\varepsilon\to\infty$.
- Hence $1/\sqrt{x}$ is integrable near zero but not at infinity; $1/x^2$ has the reverse behaviour. No real $p$ makes $\int_0^\infty x^{-p}\,dx$ converge: split at $1$ and the two necessary inequalities are incompatible.
There is a real $p$ for which $\int_0^\infty x^{-p}\,dx$ converges.
Difficulty 1/5 · 100 coins + a small bonus on first completion today
The piece near zero needs p < 1, while the tail at infinity needs p > 1. At p = 1, both pieces diverge logarithmically.
Comparison needs the right direction
- Suppose $0\le f\le g$ near the improper endpoint and both are integrable on the remaining finite pieces. A finite integral of $g$ forces a finite integral of $f$; an infinite integral of $f$ forces divergence of $g$. Being below a divergent function, or above a convergent one, proves neither conclusion.
- For $x\ge1$, $0<1/(x^2+\sin^2x)\le1/x^2$, so its integral converges. In the original four-choice problem, $1/\sqrt{x}$ and $1/x$ diverge at infinity, while $1/x^2$ diverges at zero. Comparison supplies convergence without needing a difficult antiderivative.
Which integral converges?
Difficulty 1/5 · 100 coins + a small bonus on first completion today
For x ≥ 1, 0 < 1/(x² + sin²x) ≤ 1/x², and ∫₁^∞ dx/x² converges, so the comparison test gives convergence. 1/√x and 1/x decay too slowly at infinity, and 1/x² blows up too fast at 0.
Compare asymptotic size and change variables
- For eventually positive integrands with no other singularities, if $f(x)/g(x)\to L$ with $0<L<\infty$, the two improper integrals have the same convergence behaviour. Eventually $f$ lies between positive constant multiples of $g$. A limit of zero or infinity does not by itself give this two-way conclusion.
- For $\int_e^\infty dx/[x(\ln x)^q]$, put $u=\ln x$. The transformed integral is $\int_1^\infty u^{-q}\,du$, so it converges exactly when $q>1$. The factor $1/x$ alone does not decide it; a logarithmic factor can change the borderline outcome.
Converge or diverge?
Apply the p-test at infinity and at zero.
Evaluate $\int_e^\infty dx/[x(\ln x)^2]$.
Difficulty 2/5 · 140 coins + a small bonus on first completion today
With u = ln x, this is ∫₁^∞ u⁻² du = 1.
For eventually positive $f$ and $g$, a limit $f/g\to0$ proves that their improper integrals have the same convergence behaviour.
Difficulty 1/5 · 100 coins + a small bonus on first completion today
Take f = 1/x² and g = 1/x on [1,∞). Their ratio tends to zero, but only the integral of f converges. A finite positive ratio gives the two-way limit comparison.
Determine all real $q$ for which $\int_0^{1/e}dx/[x|\ln x|^q]$ converges. Give the value when it converges, and explain why the p-test cannot be applied to $1/x$ while ignoring the logarithm.
Compare your argument with the model and checklist.
Signed convergence and absolute convergence differ
- Absolute convergence means $\int|f|$ is finite, and implies convergence of $\int f$. For $\int_1^b (\sin x)/x\,dx$, integration by parts gives the boundary term $[-\cos x/x]_1^b$ minus $\int_1^b (\cos x)/x^2\,dx$. The boundary term has a limit and the last integral converges absolutely, proving convergence.
- This example is not absolutely convergent. On each interval from $k\pi+\pi/6$ to $k\pi+5\pi/6$, $|\sin x|\ge1/2$. For integers $k\ge1$, that interval contributes at least $1/[3(k+1)]$ to $\int|\sin x|/x\,dx$. These lower bounds have a divergent harmonic sum. This is conditional convergence, unlike the nonconvergent $1/x$ principal-value example.
The integral $\int_1^\infty (\sin x)/x\,dx$ converges absolutely.
Difficulty 1/5 · 100 coins + a small bonus on first completion today
Integration by parts proves ordinary convergence. But on successive central pieces of sine half-waves, the absolute integral has harmonic lower bounds and diverges. Convergence is conditional.
Geometric area uses the upper curve
- The area between continuous curves is $\int_a^b |f(x)-g(x)|\,dx$. Find their crossings and split where their order changes. The signed integral $\int(f-g)$ can cancel regions; geometric area counts both positively.
- For $f(x)=x$ and $g(x)=x^2$ on $[-1,1]$, the crossings are $0$ and $1$. On $[-1,0]$, $x^2$ is above $x$ and gives area $5/6$; on $[0,1]$, $x$ is above $x^2$ and gives $1/6$. Total area is $1$, whereas $\int_{-1}^1(x-x^2)\,dx=-2/3$.
What is the geometric area between $y=x$ and $y=x^2$ on $[-1,1]$?
Difficulty 2/5 · 140 coins + a small bonus on first completion today
Split at zero, where the upper curve changes. The two positive areas are 5/6 and 1/6, totalling 1. The signed integral -2/3 is not this area.
A maximum chooses a different curve on each piece
- To integrate $\max(\sqrt{1-x^2},x+1)$ on $[-1,1]$, solve the crossing equation. Both sides are nonnegative there, so squaring is valid and gives $2x(x+1)=0$. The crossings are $-1$ and $0$; checking a point in each intervening interval fixes the order.
- The semicircle is higher on $[-1,0]$, contributing a quarter-circle area $\pi/4$. The line is higher on $[0,1]$, contributing $\int_0^1(x+1)\,dx=3/2$. Thus the integral of the maximum is $\pi/4+3/2$. It is not the area between the two curves.
Disks and washers use distances from the axis
- A slice perpendicular to the rotation axis is a disk or washer. With outer radius $R(x)\ge r(x)\ge0$, its area is $\pi(R^2-r^2)$, and volume is its integral with respect to $x$. For the region under $y=\sqrt{x}$ on $[0,4]$, rotation about the $x$-axis gives $V=\pi\int_0^4x\,dx=8\pi$.
- Rotate the region $x\le y\le2$, $0\le x\le1$, about the $x$-axis. The outer radius is $2$, inner radius $x$, and $V=\pi\int_0^1(4-x^2)\,dx=11\pi/3$. Subtract squared radii; squaring the thickness $2-x$ would describe a different cross-section.
Rotate $y = \sqrt x$, $0 \le x \le 4$, about the $x$-axis. The volume is $k\pi$. Find $k$.
Difficulty 2/5 · 140 coins + a small bonus on first completion today
$\pi\int_0^4 x\,dx = 8\pi$.
Rotate $x\le y\le2$, $0\le x\le1$, about the $x$-axis. The volume is $k\pi$. What is $k$?
Difficulty 1/5 · 100 coins + a small bonus on first completion today
Each washer has outer radius 2 and inner radius x. Integrate 4 - x² from 0 to 1, obtaining 11/3. Squaring 2 - x would incorrectly square the thickness.
Shells use parallel slices
- Rotating a vertical slice at $x\ge0$ about the $y$-axis gives a shell with circumference $2\pi x$, height $H(x)\ge0$ and thickness $dx$. Thus $V=2\pi\int xH(x)\,dx$, provided these slices do not count the same shell twice. A shifted axis requires its actual distance as the radius.
- For the region $0\le y\le x^2$, $0\le x\le1$, shells give $V=2\pi\int_0^1x^3\,dx=\pi/2$. Horizontal washers verify it: for $0\le y\le1$, the radii are $1$ and $\sqrt y$, giving $V=\pi\int_0^1(1-y)\,dy=\pi/2$.
For $0\le y\le x^2$, $0\le x\le1$, derive two volume integrals for rotation about the $y$-axis: vertical shells and horizontal washers. Explain the bounds and radii, then evaluate both.
Compare your argument with the model and checklist.
Arc length and surface area need a slope factor
- For a continuously differentiable graph $y=f(x)$ on a finite interval, put $s(x)=\sqrt{1+f'(x)^2}$. Arc length is $\int s(x)\,dx$. Rotating the graph about the $x$-axis gives lateral surface area $\int2\pi|f(x)|s(x)\,dx$; $|f|$ is the nonnegative radius. This excludes any end caps. For improper endpoints, apply these formulas on finite pieces and then take limits.
- For $f(x)=x$ on $[0,1]$, $s=\sqrt2$, so length is $\sqrt2$ and the cone's lateral area is $\pi\sqrt2$. The graph $f(x)=-x$ has the same length and swept surface; omitting the absolute value would incorrectly give a negative area.
For a nonnegative continuously differentiable graph $y=f(x)$, match each quantity to its integrand with respect to $x$. Volume means the region under the graph rotated about the $x$-axis; surface means its lateral surface.
Difficulty 2/5 · 140 coins + a small bonus on first completion today
The length factor is √(1 + f′²); circumference times that factor gives lateral surface area. Nonnegative f is the radius here. Disk volume uses πf².
Rotate the graph $y=-x$, $0\le x\le1$, about the $x$-axis. What is its lateral surface area?
Difficulty 1/5 · 100 coins + a small bonus on first completion today
The radius is |-x| = x and the length factor is √2. Thus the lateral area is 2π√2 ∫₀¹ x dx = π√2, with no end cap included.
The horn has two different limiting integrals
- For the horn $f(x)=1/x$ on $[1,b]$, volume is $V(b)=\pi(1-1/b)$ and tends to $\pi$. Its arc-length factor is $\sqrt{1+x^{-4}}$. Hence lateral surface area satisfies $S(b)\ge2\pi\int_1^b dx/x=2\pi\ln b$, which tends to infinity. The original figure shows this lower bound, not the exact surface area.
- The horn demonstrates that finite volume does not force finite lateral area. Its generating curve also has infinite length, since the length integrand is at least $1$ on an interval of unbounded width. Each geometric question needs its own integral and convergence argument.
Which integral conclusions are justified?
Check limiting and geometric assumptions.
For $p>0$, rotate $y=x^{-p}$, $x\ge1$, about the $x$-axis. Determine exactly when the volume and lateral surface area are finite. Is the generating curve ever of finite length?
Compare your argument with the model and checklist.
Recap
- improper integral: An integral defined by limits at an infinite endpoint or at a point where the integrand is singular.
- principal value: A value obtained by a prescribed symmetric limiting procedure, which need not equal an ordinary convergent improper integral.
- arc length: The distance measured along a curve.
- Check each improper limit separately, use comparison with its hypotheses, and choose geometric radii and slice directions before integrating.