Calculus › Section I · Integration techniques and improper integrals › Topic I.8
Improper integrals
Some regions stretch to infinity or hide a vertical asymptote inside them. Improper integrals extend the definite integral to these cases, and the surprising news is that an infinitely long region can still have finite area.
When infinity sneaks in
The definite integral \(\int_a^b f(x)\,dx\) you have used so far assumes two things: the interval \([a, b]\) is finite, and \(f\) is continuous on all of it. Two situations break that contract. The interval can run forever, as in the area under a curve for all \(x \ge 1\). Or the function can blow up to infinity at a point inside the interval, like \(\frac{1}{x^2}\) at \(x = 0\). Both integrals are called improper, and both are handled the same way: replace the offending endpoint or point by a variable, do an ordinary integral, then take a limit.
Infinite intervals are limits
For an integral that runs to infinity, define
where \(t\) is an ordinary number. If the limit exists and is finite, the improper integral converges; if the limit does not exist, it diverges. For an integral over the whole line, \(\int_{-\infty}^{\infty} f(x)\,dx\), split it at any convenient point, say \(\int_{-\infty}^{c} f + \int_c^{\infty} f\), and require both pieces to converge. Note what the definition does not say: \(\infty\) is not a number, and you never plug it in. Everything is a limit of honest, finite integrals.
Worked example: a convergent tail and a divergent one
Compare \(\int_1^{\infty} \frac{dx}{x^2}\) with \(\int_1^{\infty} \frac{dx}{x}\). For the first, \(\lim_{t \to \infty}\left[-\frac{1}{x}\right]_1^t = \lim_{t \to \infty}\left(-\frac{1}{t} + 1\right) = 1\). The region under \(\frac{1}{x^2}\) is infinitely long yet its area is exactly 1. For the second, \(\lim_{t \to \infty}[\ln x]_1^t = \lim_{t \to \infty}\ln t\), which grows without bound, so the harmonic integral diverges. Two nearly identical curves, opposite verdicts: the exponent \(x^2\) in the denominator decays fast enough to make the area finite, while \(x\) does not.
The p-integral benchmark
The family \(\int_1^{\infty} \frac{dx}{x^p}\) settles the question for every power at once. The antiderivative of \(x^{-p}\) is \(\frac{x^{1-p}}{1-p}\) when \(p \ne 1\), and at \(t \to \infty\) that either dies out or explodes. The result:
and it diverges when \(p \le 1\). So \(\int_1^{\infty} \frac{dx}{x^{3/2}} = \frac{1}{1/2} = 2\), while \(\int_1^{\infty} \frac{dx}{\sqrt{x}}\) diverges because \(p = \frac{1}{2} \le 1\). This one family becomes the standard yardstick for every comparison you will make with improper integrals, and later with infinite series.
Blow-ups at a point: the second kind
An integral is also improper when \(f\) has a vertical asymptote at an endpoint or inside the interval. The fix is the same: approach the bad point by a limit. For example, \(\int_0^1 \frac{dx}{\sqrt{x}}\) has an asymptote at \(x = 0\), so define it as \(\lim_{t \to 0^+}\int_t^1 \frac{dx}{\sqrt{x}}\). Computing gives \(\lim_{t \to 0^+}[2\sqrt{x}]_t^1 = 2 - \lim_{t \to 0^+} 2\sqrt{t} = 2\). The integrand is unbounded, yet the area is finite, because the spike is narrow enough. Its neighbour \(\int_0^1 \frac{dx}{x}\) behaves differently: \(\lim_{t \to 0^+}[\ln x]_t^1 = -\lim_{t \to 0^+}\ln t\), which blows up, so it diverges.
Asymptotes hiding inside
The sneakiest case is an asymptote strictly inside the interval, where a naive application of the fundamental theorem gives a confidently wrong answer. Consider \(\int_{-1}^{1} \frac{dx}{x^2}\). Plugging in \(-\frac{1}{x}\) at the endpoints gives \(-1 - 1 = -2\), which is impossible for a positive function. The integrand blows up at \(x = 0\), so the integral must be split into \(\int_{-1}^{0} \frac{dx}{x^2}\) and \(\int_{0}^{1} \frac{dx}{x^2}\), and each is a limit approaching 0 from its own side. The right-hand piece is \(\lim_{t \to 0^+}\left[-\frac{1}{x}\right]_t^1 = -1 + \lim_{t \to 0^+}\frac{1}{t} = \infty\), so that piece diverges, and one divergent piece is enough to doom the whole integral. The correct verdict is divergence, never -2.
Exponential tails
Exponential decay is so fast that its improper integrals always converge, which is why they appear everywhere in applications. For \(k > 0\),
A probability model can exploit this: the exponential distribution with rate \(\lambda\) has density \(\lambda e^{-\lambda x}\) for \(x \ge 0\), and \(\int_0^{\infty} \lambda e^{-\lambda x}\,dx = \lambda \cdot \frac{1}{\lambda} = 1\). The total probability is exactly 1, as it must be, because the decay tail has finite area.
Worked example: what is a perpetuity worth?
A charitable trust pays out \(\$600\) per year, forever, and money earns interest at a continuous annual rate of 5%. What is that endless payment stream worth today? A payment of \(\$600\) arriving \(t\) years in the future has present value \(600e^{-0.05t}\) dollars, so the whole stream is worth \(\int_0^{\infty} 600e^{-0.05t}\,dt\). By the exponential rule with \(k = 0.05\), this equals \(\frac{600}{0.05} = \$12{,}000\). An infinite stream of payments is worth a finite amount today, and you can sanity-check the size: the trust earns \(5\%\) of \(\$12{,}000\), which is exactly \(\$600\) a year, so the capital never runs down. The two statements are two views of the same balance.
The comparison test for integrals
When you cannot find an antiderivative, comparison still answers the convergence question. If \(0 \le f(x) \le g(x)\) for all \(x \ge a\), then a smaller region sits inside a larger one: if \(\int_a^{\infty} g\,dx\) converges, so does \(\int_a^{\infty} f\,dx\). And if \(\int_a^{\infty} f\,dx\) diverges, so does the bigger one \(\int_a^{\infty} g\,dx\). For example, \(\frac{1}{x^2 + 1} \le \frac{1}{x^2}\) for \(x \ge 1\), and the larger integral converges, so \(\int_1^{\infty} \frac{dx}{x^2 + 1}\) converges too. In this case you can even confirm by integrating: \(\lim_{t \to \infty}[\arctan x]_1^t = \frac{\pi}{2} - \frac{\pi}{4} = \frac{\pi}{4}\), a finite number slightly smaller than the bounding value 1.
Common mistakes to avoid
The most dangerous slip is integrating straight across a vertical asymptote, as in \(\int_{-1}^{1} \frac{dx}{x^2}\): always scan the interval for places where the integrand is undefined before integrating. Second, treating \(\infty\) as a number you substitute into the antiderivative: the definition is a limit, so write \(\lim_{t \to \infty}\) and let \(t\) be finite. Third, forgetting that a split improper integral, like \(\int_{-\infty}^{\infty}\), needs every piece to converge; one divergent piece makes the whole thing divergent. And when using comparison, check the inequality direction: bounding a function above by a convergent one proves convergence, but bounding it above by a divergent one proves nothing.
Key takeaways
An improper integral is a limit of ordinary integrals, taken when the interval is infinite (first kind) or the integrand blows up (second kind). The benchmark family \(\int_1^{\infty} \frac{dx}{x^p}\) converges exactly when \(p > 1\), to \(\frac{1}{p - 1}\). Exponential tails always converge: \(\int_0^{\infty} e^{-kx}\,dx = \frac{1}{k}\) for \(k > 0\). An asymptote inside the interval forces a split, and every piece must converge. When exact integration is out of reach, the comparison test settles convergence by sandwiching against a known benchmark.
Flashcards
Flashcard
A limit of ordinary integrals
Test your understanding
Question 1 of 5
What is the value of the integral of dx/x³ from x = 1 to infinity?
Question 2 of 5
What is the value of the integral of e^(-2x) dx from x = 0 to infinity?
Question 3 of 5
What is the value of the integral of dx/√x from x = 0 to x = 1?
Question 4 of 5
Does the integral of dx/x² from x = -1 to x = 1 converge?
Question 5 of 5
The integral of dx/(x² + 1) from x = 1 to infinity converges. Why?
Standards: 6 (BC)
