Improper Integrals: Decide Convergence Before You Integrate

An improper integral is an ordinary definite integral with one extra piece of information attached: the interval is unbounded, or the integrand is unbounded somewhere on it. The extra work is not a new integration technique. It is a limit placed around an integral you already know how to evaluate.

Two unrelated reasons an integral is improper

The first reason is an infinite interval. The definition replaces the infinite limit by a variable and then lets that variable run:

$\int_{a}^{\infty} f(x)\,dx=\lim_{t\to\infty}\int_{a}^{t} f(x)\,dx,\qquad \int_{-\infty}^{b} f(x)\,dx=\lim_{t\to-\infty}\int_{t}^{b} f(x)\,dx.$

For an integral running from $-\infty$ to $+\infty$ you split it at any interior point and require both pieces to converge. If either piece diverges, the whole integral diverges — there is no cancellation between the two sides.

The second reason is a vertical asymptote. If the integrand is unbounded at an endpoint, the integral is the one-sided limit taken from inside the interval. If it is unbounded at an interior point $c$, split there and again require both pieces to converge. A discontinuity that stays finite, such as a jump, does not by itself make an integral improper.

Convergence is a statement about a limit, not about size

Writing the integral as a limit makes the vocabulary precise. If the limit is a finite number, the integral converges to that number. If the limit is infinite or fails to exist, the integral diverges. That wording carries no claim about how large the integrand gets: integrands that blow up can still produce finite areas, and bounded integrands on an infinite interval can still produce infinite ones.

The p-rules point in opposite directions

For the family $f(x)=x^{-p}$ there are two rules, and confusing them is the most common source of wrong answers:

• On an infinite interval, $\int_{1}^{\infty}x^{-p}dx$ converges exactly when $p>1$.
• At a finite endpoint, $\int_{0}^{1}x^{-p}dx$ converges exactly when $p<1$.

The same integrand family, opposite conclusions, because the two integrals describe different ends of the graph. Moving outward, $x^{-p}$ decays toward zero, and only sufficiently fast decay is integrable. Moving toward the origin, $x^{-p}$ grows, and only sufficiently slow growth is integrable.

Worked examples with new integrals

The logarithmic case sits exactly on the borderline of the first rule. Consider $\int_{2}^{\infty}\dfrac{dx}{x\ln x}$. Substituting $u=\ln x$, so that $du=\dfrac{dx}{x}$, turns the integral into $\int_{\ln 2}^{\infty}\dfrac{du}{u}$, which is a logarithm again: $\ln u$ evaluated at the upper limit grows without bound. The integral diverges, even though $x\ln x$ grows faster than $x$ alone.

A blow-up at an endpoint is often harmless. Consider $\int_{0}^{8}\dfrac{dx}{x^{2/3}}$. Here $p=\frac{2}{3}<1$, and the antiderivative gives

$\lim_{t\to 0^{+}}\left[3x^{1/3}\right]_{t}^{8}=6-0=6,$

so the integral converges even though the integrand is unbounded at $0$. The area near the origin is squeezed into a narrow spike; an unbounded height does not force an unbounded area.

Comparing instead of integrating

When the antiderivative is hard, comparison often settles the question. If $0\le f(x)\le g(x)$ on the interval and $\int g$ converges, then $\int f$ converges; if $f(x)\ge g(x)\ge 0$ and $\int g$ diverges, then $\int f$ diverges. For example, $0\le\dfrac{1+\sin x}{x^{2}}\le\dfrac{2}{x^{2}}$ for $x\ge 1$, and $\int_{1}^{\infty}\dfrac{2}{x^{2}}dx$ converges by the $p$-rule with $p=2>1$, so $\int_{1}^{\infty}\dfrac{1+\sin x}{x^{2}}dx$ converges as well. Comparison never produces the value of an integral — only the verdict.

What a verdict does not say

Two habits prevent most errors. First, do not treat a divergent integral as zero, and do not use symmetry as a convergence test: an odd integrand over a symmetric infinite interval gives two divergent halves, not a cancelling value of zero. Second, do not apply a convergence rule without checking which type of improper integral you have, since the endpoint and infinite-interval versions of the $p$-rule have opposite conditions.