Summation and Integral

90 Views Asked by At

Suppose $f$ is continuous and there exist $a>1$ such that $\sup_{x \in \mathbb{R}}|x|^a|f(x)| < \infty$. Suppose there exist $\Omega>0$ such that $\hat{f}=0$ outside $[-\Omega, \Omega]$. Let $0<\lambda \leq \frac{2\pi}{\Omega}.$ Prove that $$\lambda \sum_{n \in \mathbb{Z}}f(n\lambda)=\int_{\mathbb{R}}f(x)dx.$$ Here $\hat{f}$ is the fourier transform of $f$.

I tried to used the poisson summation formula but i'm stuck and can't proceed to the desire conclusion! Any help or hints would be much appreciated!!!

1

There are 1 best solutions below

2
On

Yes, it's just Poisson summation. In fact it's a particularly simple application: The idea behind Poisson summation is clear, typically the hard part is getting from a formal Fourier series to convergence of the Fourier series. But here it turns out that we get the Fourier series of a trigonometric polynomial, so convergence is no problem.

First, as noted, different authors define the Fourier transform differently, with the $2\pi$s in different places. So when you state this sort of thing you should state explicitly what definition of $\hat f$ you have in mind, even though you're assuming the reader knows all about the Fourier transform; the result could be true with your version but false with the version the reader has in mind.

$\newcommand\ri{\frac1{\sqrt{2\pi}}\int}$ So, for the record, I'm going to be normalizing things as Rudin does, defining $$\hat f(\xi)=\ri f(t)e^{-it\xi}\,d\xi.$$

Now for your problem. Below we'll be dealing with the Fourier series of a $\lambda$-periodic function; this will be less confusing if we start by saying

Replacing $f$ by a suitable dilate, WLOG $$\lambda=2\pi.$$

So now your hypothesis becomes $\Omega\le1$. Which really doesn't seem right; since we're assuming that $\hat f$ vanishes outside $[-\Omega,\Omega]$ surely the condition should be that $\Omega$ is larger than something, not smaller.

So for now forget that inequality; assume $\hat f$ vanishes outside $[-\Omega,\Omega]$ and we'll see what $\Omega$ works. Define $$F(x)=\sum_{n\in\Bbb Z}f(x+2\pi n).$$Then $F$ has period $2\pi$. The bound on $f$ at infinity shows the series converges uniformly on $[-\pi,\pi]$, so $F$ is continuous and the Fourier coefficients of $F$ are related to the Fourier transform of $f$ by $$\hat F(k)=\ri\hat f(k)\quad(k\in\Bbb Z).$$

Assume now that $$\Omega\ge1.$$Then $$\hat F(k)=0,\quad(k\in\Bbb Z, k\ne0),$$so $F$ is actually a constant; since $F$ is continuous we have $$\sum f(2\pi n)=F(0)=\hat F(0)=\ri\hat f(0)=\frac1{2\pi}\int f.$$

Which is what you wanted in a special case, except for the extra $1/2\pi$; as noted above, I suspect this is just due to the Fourier transform being normalized differently.

Note That's a proof if $\lambda=2\pi$ and $\Omega\ge1$. You still have to work out what the constraint on $\Omega$ should be if $\lambda\ne2\pi$. (Apply the above to a suitable dilate of $f$. Or repeat the above, without assuming $\lambda=2\pi$, instead dealing with the Fourier coefficients of $F$ as a $\lambda$-periodic function.)