Dirichlet's Test

A wobbling series gets tamed by a sinking weight in this scene. Two ingredients meet and produce convergence.

The partial sums of the sine terms wobble around but stay trapped inside a fixed fence. Multiply them by weights that sink to 0. The wobble gets pressed flatter and flatter until the series converges. The test's exact conditions are in the formulas below.

For example, a ⚽ ball bounces around inside a fence and never escapes. Push it more and more gently, and its wandering dies down until it settles.

∑k=1Nsin⁡(kx)=sin⁡Nx2 sin⁡(N+1)x2sin⁡x2,∣∑k=1Nsin⁡(kx)∣≤1∣sin⁡(x/2)∣\sum_{k=1}^{N}\sin(kx) = \frac{\sin\frac{Nx}{2}\,\sin\frac{(N+1)x}{2}}{\sin\frac{x}{2}},\quad \left|\sum_{k=1}^{N}\sin(kx)\right| \le \frac{1}{|\sin(x/2)|}
∑n=1∞sin⁡(nx)n=π−x2,x∈(0,2π)\sum_{n=1}^{\infty}\frac{\sin(nx)}{n} = \frac{\pi-x}{2},\quad x \in (0, 2\pi)

Part of these learning paths

Examples & Exercises

Show that ∑n=1∞sin⁡(nx)n\sum_{n=1}^{\infty} \frac{\sin(nx)}{n} converges for every xx that is not a multiple of 2π2\pi.

∑k=1Nsin⁡(kx)=sin⁡Nx2 sin⁡(N+1)x2sin⁡x2\sum_{k=1}^{N} \sin(kx) = \frac{\sin\frac{Nx}{2}\,\sin\frac{(N+1)x}{2}}{\sin\frac{x}{2}}
∣∑k=1Nsin⁡(kx)∣≤1∣sin⁡(x/2)∣\left|\sum_{k=1}^{N} \sin(kx)\right| \le \frac{1}{|\sin(x/2)|}
x=1x=1: the partial sums ∣∑k≤Nsin⁡(kx)∣\left\lvert\sum_{k \le N}\sin(kx)\right\rvert stay below 1/sin⁡(0.5)≈2.08581/\sin(0.5) \approx 2.0858.
Show proof
  1. If xx is not a multiple of 2π2\pi then sin⁡(x/2)≠0\sin(x/2) \neq 0. The product-to-sum identity 2sin⁡x2sin⁡(kx)=cos⁡(k−12)x−cos⁡(k+12)x2\sin\frac{x}{2}\sin(kx) = \cos\left(k-\tfrac{1}{2}\right)x - \cos\left(k+\tfrac{1}{2}\right)x, summed over k=1,…,Nk = 1, \dots, N, telescopes to ∑k=1Nsin⁡(kx)=cos⁡x2−cos⁡(N+12)x2sin⁡x2=sin⁡Nx2sin⁡(N+1)x2sin⁡x2\sum_{k=1}^{N} \sin(kx) = \frac{\cos\frac{x}{2} - \cos\left(N+\tfrac{1}{2}\right)x}{2\sin\frac{x}{2}} = \frac{\sin\frac{Nx}{2}\sin\frac{(N+1)x}{2}}{\sin\frac{x}{2}}.

  2. Each factor in the numerator has absolute value at most 11, so the partial sums are bounded uniformly in NN by ∣∑k=1Nsin⁡(kx)∣≤1∣sin⁡(x/2)∣\left\lvert\sum_{k=1}^{N} \sin(kx)\right\rvert \le \frac{1}{|\sin(x/2)|}. At x=1x = 1 this is 1/sin⁡(0.5)≈2.08581/\sin(0.5) \approx 2.0858 — matching the caption's figure.

  3. bn=1/nb_n = 1/n decreases to 00.

    Since the partial sums of an=sin⁡(nx)a_n = \sin(nx) are bounded, Dirichlet's test (summation by parts) gives convergence of ∑n=1∞anbn=∑n=1∞sin⁡(nx)n\sum_{n=1}^{\infty} a_n b_n = \sum_{n=1}^{\infty} \frac{\sin(nx)}{n}.

    (In fact this series equals (π−x)/2(\pi - x)/2 for x∈(0,2π)x \in (0, 2\pi).)

Show that the Leibniz alternating series test is a corollary of Dirichlet's test.

∣∑n=1N(−1)n∣≤1\left|\sum_{n=1}^{N} (-1)^n\right| \le 1
The alternating partial sums ∣∑n≤N(−1)n∣\left\lvert\sum_{n \le N}(-1)^n\right\rvert never exceed 11 — the Leibniz test is a corollary of Dirichlet's test.
Show proof
  1. Set an=(−1)na_n = (-1)^n. Its partial sums are 00 when NN is even and −1-1 when NN is odd, so ∣∑n=1Nan∣≤1\left\lvert\sum_{n=1}^{N} a_n\right\rvert \le 1 for every NN — the boundedness condition of Dirichlet's test holds.

  2. The Leibniz hypothesis is bn↓0b_n \downarrow 0. Applying Dirichlet's test to the pair (an,bn)(a_n, b_n) yields convergence of ∑n=1∞(−1)nbn\sum_{n=1}^{\infty} (-1)^n b_n at once — the alternating series test is Dirichlet's test in the case an=(−1)na_n = (-1)^n.

Show that ∑n=1∞(−1)nnn2+1\sum_{n=1}^{\infty} (-1)^n \frac{n}{n^2+1} converges.

bn=nn2+1,bn>bn+1  ⟺  n2+n−1>0b_n = \frac{n}{n^2+1}, \qquad b_n > b_{n+1} \iff n^2+n-1 > 0
b1=0.5>b2=0.4>⋯b_1 = 0.5 > b_2 = 0.4 > \cdots: bn=nn2+1b_n = \frac{n}{n^2+1} decreases monotonically for n≥1n \ge 1 and tends to 00.
Show proof
  1. The partial sums of an=(−1)na_n = (-1)^n have absolute value at most 11 for every NN (as in the previous problem). For bn=nn2+1b_n = \frac{n}{n^2+1} we have 0<bn<1n0 < b_n < \frac{1}{n}, so bn→0b_n \to 0.

  2. Monotonicity of bnb_n: bn>bn+1b_n > b_{n+1} is, after cross-multiplication, n((n+1)2+1)>(n+1)(n2+1)n\left((n+1)^2+1\right) > (n+1)(n^2+1), i.e. n3+2n2+2n>n3+n2+n+1n^3+2n^2+2n > n^3+n^2+n+1, i.e. n2+n−1>0n^2+n-1 > 0.

    At n=1n = 1 we have 12+1−1=1>01^2+1-1 = 1 > 0, so this holds for all n≥1n \ge 1.

    Hence bn↓0b_n \downarrow 0.

  3. b1=12=0.5>b2=25=0.4b_1 = \frac{1}{2} = 0.5 > b_2 = \frac{2}{5} = 0.4 — matching the caption's figures. By Dirichlet's test, ∑n=1∞(−1)nnn2+1\sum_{n=1}^{\infty} (-1)^n \frac{n}{n^2+1} converges.

7. Series