6.15 Sums6.17 Physical Applications

§6.16 Mathematical Applications

Contents

§6.16(i) The Gibbs Phenomenon

Consider the Fourier series

6.16.1 \mathop{\sin\/}\nolimits x+\tfrac{1}{3}\mathop{\sin\/}\nolimits\!\left(3x\right)+\tfrac{1}{5}\mathop{\sin\/}\nolimits\!\left(5x\right)+\dots=\begin{cases}\frac{1}{4}\pi,&0<x<\pi,\\
0,&x=0,\\
-\frac{1}{4}\pi,&-\pi<x<0.\end{cases}

The nth partial sum is given by

6.16.2 S_{n}(x)=\sum _{{k=0}}^{{n-1}}\frac{\mathop{\sin\/}\nolimits\!\left((2k+1)x\right)}{2k+1}=\frac{1}{2}\int _{0}^{x}\frac{\mathop{\sin\/}\nolimits\!\left(2nt\right)}{\mathop{\sin\/}\nolimits t}dt=\tfrac{1}{2}\mathop{\mathrm{Si}\/}\nolimits\!\left(2nx\right)+R_{n}(x),

where

6.16.3 R_{n}(x)=\frac{1}{2}\int _{0}^{x}\left(\frac{1}{\mathop{\sin\/}\nolimits t}-\frac{1}{t}\right)\mathop{\sin\/}\nolimits\!\left(2nt\right)dt.

By integration by parts

6.16.4 R_{n}(x)=\mathop{O\/}\nolimits\!\left(n^{{-1}}\right), n\to\infty,

uniformly for x\in[-\pi,\pi]. Hence, if x is fixed and n\to\infty, then S_{n}(x)\to\frac{1}{4}\pi, 0, or -\frac{1}{4}\pi according as 0<x<\pi, x=0, or -\pi<x<0; compare (6.2.14).

These limits are not approached uniformly, however. The first maximum of \frac{1}{2}\mathop{\mathrm{Si}\/}\nolimits\!\left(x\right) for positive x occurs at x=\pi and equals (1.1789\dots)\times\frac{1}{4}\pi; compare Figure 6.3.2. Hence if x=\pi/(2n) and n\to\infty, then the limiting value of S_{n}(x) overshoots \frac{1}{4}\pi by approximately 18%. Similarly if x=\pi/n, then the limiting value of S_{n}(x) undershoots \frac{1}{4}\pi by approximately 10%, and so on. Compare Figure 6.16.1.

This nonuniformity of convergence is an illustration of the Gibbs phenomenon. It occurs with Fourier-series expansions of all piecewise continuous functions. See Carslaw (1930) for additional graphs and information.

See accompanying text
Figure 6.16.1: Graph of S_{n}(x), n=250, -0.1\leq x\leq 0.1, illustrating the Gibbs phenomenon. Magnify

§6.16(ii) Number-Theoretic Significance of \mathop{\mathrm{li}\/}\nolimits\!\left(x\right)

If we assume Riemann’s hypothesis that all nonreal zeros of \mathop{\zeta\/}\nolimits\!\left(s\right) have real part of \tfrac{1}{2}25.10(i)), then

6.16.5 \mathop{\mathrm{li}\/}\nolimits\!\left(x\right)-\pi(x)=\mathop{O\/}\nolimits\!\left(\sqrt{x}\mathop{\ln\/}\nolimits x\right), x\to\infty,

where \pi(x) is the number of primes less than or equal to x. Compare §27.12 and Figure 6.16.2. See also Bays and Hudson (2000).

See accompanying text
Figure 6.16.2: The logarithmic integral \mathop{\mathrm{li}\/}\nolimits\!\left(x\right), together with vertical bars indicating the value of \pi(x) for x=10,20,\dots,1000. Magnify