AP Calculus BC

Integral Test and p-Series: Proving Convergence

If f is positive, continuous and decreasing for x >= 1 and a_n = f(n), then the series of a_n and the improper integral of f from 1 to infinity either both converge or both diverge. Applied to 1/n^p it gives the p-series rule: convergent when p > 1, divergent when p <= 1. The harmonic series, p = 1, diverges.

The test and its three conditions

The integral test trades a series for an improper integral that is often easier to evaluate. It is CED Topic 10.4, Integral Test for Convergence. Suppose an=f(n)a_n = f(n) for a function ff that meets three conditions on [1,∞)[1, \infty):

  • ff is positive.
  • ff is continuous.
  • ff is decreasing.

Then the series and the integral share the same fate.

∑n=1∞an  and  ∫1∞f(x) dxboth converge or both diverge\sum_{n=1}^{\infty} a_n \ \text{ and } \ \int_1^{\infty} f(x)\,dx \quad \text{both converge or both diverge}

The conditions only need to hold from some point on. If ff is positive, continuous and decreasing for x≥Nx \ge N, compare ∑n=N∞an\sum_{n=N}^{\infty} a_n with ∫N∞f(x) dx\int_N^{\infty} f(x)\,dx. The finitely many terms before NN change the sum but cannot change whether the series converges.

Running the test

  1. Replace nn with xx in the general term to get f(x)f(x).
  2. State that ff is positive, continuous and decreasing on the interval, and justify the decreasing claim. Showing f′(x)<0f'(x) < 0 is the standard way.
  3. Write the improper integral as a limit: ∫1∞f(x) dx=lim⁡b→∞∫1bf(x) dx\int_1^{\infty} f(x)\,dx = \lim_{b \to \infty} \int_1^b f(x)\,dx.
  4. Evaluate the integral and take the limit.
  5. Conclude: a finite limit means the series converges, and an infinite limit means the series diverges.

The test is a good choice when ff has an antiderivative you can find quickly, especially when the term pairs 1n\frac{1}{n} with ln⁡n\ln n or pairs nn with a function of n2n^2, since those set up a clean substitution.

p-series and the harmonic series

A p-series has the form ∑n=1∞1np\sum_{n=1}^{\infty} \frac{1}{n^p} for a constant pp. Its behavior is CED Topic 10.5, Harmonic Series and p-Series.

∑n=1∞1np  converges if p>1  and diverges if p≤1\sum_{n=1}^{\infty} \frac{1}{n^p} \ \text{ converges if } p > 1 \ \text{ and diverges if } p \le 1

The rule comes straight from the integral test. For p>1p > 1, ∫1∞x−p dx=lim⁡b→∞b1−p−11−p=1p−1\int_1^{\infty} x^{-p}\,dx = \lim_{b \to \infty} \frac{b^{1-p} - 1}{1 - p} = \frac{1}{p - 1}, which is finite. For p=1p = 1 the integral is lim⁡b→∞ln⁡b\lim_{b \to \infty} \ln b, which is infinite. For 0<p<10 < p < 1 the power b1−pb^{1-p} grows without bound. For p≤0p \le 0 the terms do not approach 00, so the series diverges by the nth term test.

The case p=1p = 1 is the harmonic series, ∑n=1∞1n=1+12+13+⋯\sum_{n=1}^{\infty} \frac{1}{n} = 1 + \frac{1}{2} + \frac{1}{3} + \cdots. Its terms approach 00 and it still diverges. It is the standard reminder that terms approaching 00 is required for convergence and is not enough to guarantee it.

Once the rule is known, use it

You do not need to rerun the integral test on a p-series. Identify pp, compare it with 11, and state the conclusion. A constant multiple does not change the verdict: ∑3n\sum \frac{3}{n} diverges because ∑1n\sum \frac{1}{n} does.

What the integral does not tell you

The integral test gives a verdict. It does not give the sum. The series ∑n=1∞1n2\sum_{n=1}^{\infty} \frac{1}{n^2} converges because ∫1∞1x2 dx=1\int_1^{\infty} \frac{1}{x^2}\,dx = 1 is finite, yet the sum of the series is π26≈1.645\frac{\pi^2}{6} \approx 1.645.

For a positive, continuous, decreasing ff whose integral converges, the integral does bracket the sum:

∫1∞f(x) dx ≤ ∑n=1∞an ≤ a1+∫1∞f(x) dx\int_1^{\infty} f(x)\,dx \ \le\ \sum_{n=1}^{\infty} a_n \ \le\ a_1 + \int_1^{\infty} f(x)\,dx

For 1n2\frac{1}{n^2} this says the sum lies between 11 and 22, which agrees with 1.6451.645.

Common mistakes

  • Skipping the conditions. The test applies only when ff is positive, continuous and decreasing, so state and check all three.
  • Reporting the value of the integral as the sum of the series.
  • Writing ∫1∞\int_1^{\infty} and substituting infinity directly. Use a limit as b→∞b \to \infty.
  • Deciding that a series converges because its terms approach 00. The harmonic series is the counterexample.
  • Misreading the exponent. 1nn\frac{1}{n\sqrt{n}} has p=32p = \frac{3}{2}, and 1n23\frac{1}{\sqrt[3]{n^2}} has p=23p = \frac{2}{3}.
  • Applying the p-series rule to 12n\frac{1}{2^n}. With nn in the exponent, the series is geometric.

Worked examples

Worked example

A convergent series by substitution

Determine whether ∑n=1∞ne−n2\sum_{n=1}^{\infty} n e^{-n^2} converges or diverges.

  1. Let f(x)=xe−x2f(x) = x e^{-x^2}. For x≥1x \ge 1 it is positive and continuous.
  2. Check decreasing: f′(x)=e−x2−2x2e−x2=e−x2(1−2x2)f'(x) = e^{-x^2} - 2x^2 e^{-x^2} = e^{-x^2}\left(1 - 2x^2\right), which is negative for x≥1x \ge 1. So ff is decreasing on [1,∞)[1, \infty) and the integral test applies.
  3. Integrate with u=−x2u = -x^2, du=−2x dxdu = -2x\,dx: ∫1bxe−x2 dx=−12e−x2∣1b=12(e−1−e−b2)\int_1^b x e^{-x^2}\,dx = -\frac{1}{2} e^{-x^2} \Big|_1^b = \frac{1}{2}\left(e^{-1} - e^{-b^2}\right).
  4. Take the limit: as b→∞b \to \infty, e−b2→0e^{-b^2} \to 0, so ∫1∞xe−x2 dx=12e\int_1^{\infty} x e^{-x^2}\,dx = \frac{1}{2e}.
  5. The integral is finite, so the series converges.

The series converges by the integral test, since ∫1∞xe−x2 dx=12e\int_1^{\infty} x e^{-x^2}\,dx = \frac{1}{2e} is finite. The sum of the series is not 12e\frac{1}{2e}.

Worked example

A divergent series with a logarithm

Determine whether ∑n=2∞1nln⁡n\sum_{n=2}^{\infty} \frac{1}{n \ln n} converges or diverges.

  1. Let f(x)=1xln⁡xf(x) = \frac{1}{x \ln x}. For x≥2x \ge 2 it is positive and continuous.
  2. Check decreasing: for x≥2x \ge 2 both xx and ln⁡x\ln x are positive and increasing, so the denominator xln⁡xx \ln x is increasing and ff is decreasing.
  3. Integrate with u=ln⁡xu = \ln x, du=1x dxdu = \frac{1}{x}\,dx: ∫2b1xln⁡x dx=ln⁡(ln⁡x)∣2b=ln⁡(ln⁡b)−ln⁡(ln⁡2)\int_2^b \frac{1}{x \ln x}\,dx = \ln(\ln x) \Big|_2^b = \ln(\ln b) - \ln(\ln 2).
  4. Take the limit: as b→∞b \to \infty, ln⁡(ln⁡b)→∞\ln(\ln b) \to \infty, so the improper integral diverges.
  5. The integral diverges, so the series diverges.

The series diverges by the integral test, even though its terms approach 00.

Worked example

Classifying p-series

Decide whether each series converges or diverges: (a) ∑n=1∞1n23\sum_{n=1}^{\infty} \frac{1}{\sqrt[3]{n^2}}, (b) ∑n=1∞1nn\sum_{n=1}^{\infty} \frac{1}{n\sqrt{n}}, (c) ∑n=1∞3n\sum_{n=1}^{\infty} \frac{3}{n}.

  1. (a) Rewrite the term as 1n2/3\frac{1}{n^{2/3}}. This is a p-series with p=23≤1p = \frac{2}{3} \le 1, so it diverges.
  2. (b) Rewrite the term as 1n3/2\frac{1}{n^{3/2}}, since n⋅n1/2=n3/2n \cdot n^{1/2} = n^{3/2}. This is a p-series with p=32>1p = \frac{3}{2} > 1, so it converges.
  3. (c) Factor out the constant: ∑3n=3∑1n\sum \frac{3}{n} = 3 \sum \frac{1}{n}. This is 33 times the harmonic series, p=1p = 1, so it diverges.

(a) diverges with p=23p = \frac{2}{3}; (b) converges with p=32p = \frac{3}{2}; (c) diverges as a multiple of the harmonic series.

Frequently asked questions

Does the harmonic series converge?

No. ∑n=1∞1n\sum_{n=1}^{\infty} \frac{1}{n} diverges. It is the p-series with p=1p = 1, and the integral test shows it: ∫1∞1x dx=lim⁡b→∞ln⁡b\int_1^{\infty} \frac{1}{x}\,dx = \lim_{b \to \infty} \ln b is infinite. The partial sums grow very slowly, roughly like ln⁡n\ln n, which is why the divergence is easy to doubt.

Is the value of the integral the sum of the series?

No. The integral test only says whether the series converges. For ∑1n2\sum \frac{1}{n^2} the integral equals 11 and the sum equals π26≈1.645\frac{\pi^2}{6} \approx 1.645. The integral gives a lower bound for the sum, and adding the first term gives an upper bound.

When is the integral test the right choice?

Use it when the general term turns into a function you can integrate easily and that is positive, continuous and decreasing. Terms such as 1nln⁡n\frac{1}{n \ln n} or ne−n2n e^{-n^2} are typical. For a plain power of nn, use the p-series rule directly. For terms with factorials, the ratio test fits better.