✧*。٩(ˊᗜˋ*)و✧*。 白麓的 web-log

The Herglotz Trick

An idea that can be used only once is a trick, many times, a method.

-Problems and Theorems in Analysis

This morning when I was having my routine coffee, I came across this interesting video, "how to build your own periodic function". The rational is fairly simple, to ensure that

f(x)=f(x+T)

where T=1 for simplicity, a possible setting up would be let

f(x)=k=+g(x+k)

Now this is a infinite serie, if it converges absolutely, we can rearrange the order of the summation:

f(x)=g(x)+k=1(g(x+k)+g(xk)).

However to ensure the convergence, we at least require that limkg(x+k)=0. Then comes the Herglotz trick:

Set g(x)=1x:

f(x)=1x+k=1(1x+k+1xk)=1x+k=12xx2k2.

Show that f(x)=πcotπx

  1. f(x)+f(x)=0,
  2. f(x) is not defined on integers, but continuous otherwise,
  3. f(x) satisfies the functional equation:
f(x2)+f(x+12)=2f(x).

Let h(x)=f(x)πcotπx, due to (1), it is reasonable to let h(n)=0,n. Thus we have a continuous periodic function in , and we only need to focus on one of the periods (that is a closed interval). Let x0 be the point where h(x)=m is maximised in [0,1], by (3)

h(x02)+h(x0+12)=2m

Thus, h(x)m for x[0,1], but notice that h(0)=0, we have

f(x)πcotπx0,

or

f(x)=πcotπx.

Thus a periodic function is created.

But, what if we make other decisions of g(x)? Notice that

n=1(x+n)2=ddxn=1x+n=f(x)=π2sin2πx.

Taking different xk yields various derivatives of πcotπx. Also the above formula is particularly interesting as if x=0 we will have

limx01x2+2n=11(x+n)2=limx0π2π2x213!π4x4.

Exchange the limits, we have

n=11n2=limx01x216π2x41x2=π26.