Calculus › Section J · Infinite series › Topic J.3

The integral test and p-series

Most series have no closed-form sum at all, so the real question is a yes-or-no one: does the running total settle? The integral test answers it by turning series into an area you already know how to measure.

When you cannot find the sum

Geometric and telescoping series are the lucky ones: a subtraction or a collapse hands you the exact sum. Most series are not lucky. Nobody can write down an elementary formula for the sum of \(\sum \frac{1}{n^2}\) by algebra alone, and yet it is vital to know that this total is finite, because a convergent series can be approximated by its partial sums to any accuracy you like. Convergence tests trade the hard question (what is the sum?) for a tractable one (does the sum exist?). The first and most intuitive test borrows the improper integrals you already command.

The integral test

Suppose \(f\) is positive, continuous and decreasing for \(x \ge 1\), and the series terms are \(a_n = f(n)\). Then

\sum_{n=1}^{\infty} a_n \text{ and } \int_1^{\infty} f(x)\,dx

either both converge or both diverge. All three hypotheses matter: positivity keeps the areas meaningful, and decreasing is what makes the geometry work. The test never tells you the sum, only that the series and the integral share the same verdict. Any starting point works the same way, so a series beginning at \(n = 3\) can be tested with \(\int_3^{\infty}\) once \(f\) meets the conditions there.

Why it works: rectangles under a curve

Draw \(f\) decreasing and erect a rectangle of height \(f(n)\) over each interval from \(n\) to \(n + 1\). Because \(f\) is decreasing, a rectangle built from the left endpoint of its interval sits above the curve, while one built from the right endpoint sits below it. Totalling rectangles over the intervals from 1 to \(N\) gives the squeeze

\int_1^{N+1} f(x)\,dx \;\le\; S_N \;\le\; f(1) + \int_1^{N} f(x)\,dx

Both bounds share the limiting behaviour of the integral: if the area is finite, the partial sums are trapped between two closing walls and must converge; if the area is infinite, the left wall pushes them past every bound.

p-series

Apply the test to the most important family in the subject: \(f(x) = \frac{1}{x^p}\), whose integral you already evaluated among improper integrals. The integral \(\int_1^{\infty} \frac{dx}{x^p}\) converges exactly when \(p > 1\), so the p-series \(\sum \frac{1}{n^p}\) does the same:

\sum_{n=1}^{\infty} \frac{1}{n^p} \text{ converges exactly when } p > 1

Thus \(\sum \frac{1}{n^2}\) converges (its sum is \(\frac{\pi^2}{6}\), a celebrated result of Euler), while \(\sum \frac{1}{\sqrt{n}}\) diverges, since \(p = \frac{1}{2} \le 1\). Memorise this family: it is the benchmark against which the comparison tests of the next section measure everything else.

Worked example: a logarithmic series

Does \(\sum_{n=2}^{\infty} \frac{1}{n \ln n}\) converge? The function \(f(x) = \frac{1}{x \ln x}\) is positive, continuous and decreasing for \(x \ge 2\), so the integral test applies. Compute with the substitution \(u = \ln x\), \(du = \frac{dx}{x}\):

\int_2^{\infty} \frac{dx}{x \ln x} = \int_{\ln 2}^{\infty} \frac{du}{u} = \lim_{t \to \infty}[\ln u]_{\ln 2}^{t} = \infty

The integral diverges, so the series diverges too. Notice the surprise: the terms shrink noticeably faster than \(\frac{1}{n}\), but multiplying by \(\ln n\) is not enough to buy convergence. The same substitution shows \(\sum \frac{1}{n(\ln n)^2}\) converges, since \(\int \frac{du}{u^2} = \frac{1}{u}\) closes up: the borderline series need the extra logarithmic power.

Remainders: how good is a partial sum?

Convergence alone does not produce numbers; partial sums do. The integral test supplies a bonus in the form of a remainder bound: for a decreasing positive \(f\), the tail left over after \(N\) terms obeys

0 \le R_N = S - S_N \le \int_N^{\infty} f(x)\,dx

Geometrically, the tail terms are rectangles sitting under the leftover area to the right of \(N\). This is what makes convergent series computationally usable: you can name in advance how many terms suffice for a given accuracy.

Worked example: bounding a tail

How many terms of \(\sum \frac{1}{n^3}\) are enough to be within 0.01 of the true sum? The tail after \(N\) terms is at most \(\int_N^{\infty} \frac{dx}{x^3} = \left[-\frac{1}{2x^2}\right]_N^{\infty} = \frac{1}{2N^2}\). You need \(\frac{1}{2N^2} \le 0.01\), that is \(N^2 \ge 50\), and \(N = 8\) already gives \(64 \ge 50\). Eight terms suffice. As a sanity check, \(S_8 = 1 + \frac{1}{8} + \frac{1}{27} + \cdots + \frac{1}{512} \approx 1.195\), and the true sum of this p-series is known to be approximately 1.202: the difference is about 0.007, inside the promised 0.01. The bound is honest, and slightly generous, as bounds should be.

Common mistakes to avoid

The gravest error is applying the integral test without checking that \(f\) is decreasing; for a function that rises and falls, the rectangle argument collapses and the verdict can be wrong. Second, do not confuse the p-series \(\sum \frac{1}{n^p}\) with the p-integral itself: the test compares them, it does not say they are equal, and the test never produces the value of the sum. Third, watch the starting index: \(\sum_{n=2}^{\infty} \frac{1}{n \ln n}\) is undefined at \(n = 1\), and the matching integral must start where the hypotheses hold. Fourth, convergence of the integral does not hand you a closed form; it only certifies that partial sums will approach something.

Key takeaways

For a positive, continuous, decreasing \(f\), the series \(\sum f(n)\) and the integral \(\int_1^{\infty} f\,dx\) live or die together, because the partial sums are sandwiched between left- and right-endpoint rectangles. The p-series \(\sum \frac{1}{n^p}\) converges exactly when \(p > 1\) and is the standard benchmark from here on. Logarithmic borderline series, like \(\sum \frac{1}{n \ln n}\), are decided by the same test after a substitution. The remainder bound \(R_N \le \int_N^{\infty} f\,dx\) converts convergence into a practical accuracy guarantee for partial sums.

Flashcards

Flashcard

Integral test

For f positive, continuous, decreasing: Σ f(n) and ∫ f dx converge or diverge together

Decided by rectangles under the curve.

Test your understanding

Question 1 of 5

For which values of p does the series 1/nᵖ from n = 1 to infinity converge?

Question 2 of 5

What does the integral test say about the series 1/(n ln n) from n = 2 to infinity?

Question 3 of 5

Which condition is NOT required by the integral test?

Question 4 of 5

Using the remainder bound, how many terms of the series 1/n³ guarantee the partial sum is within 0.01 of the true sum?

Question 5 of 5

Does the series 1/√n from n = 1 to infinity converge?

Standards: 10 (BC)

← Previous topicNext topic →