6
$\begingroup$

I can't understand a lemma in "An introduction to harmonic analysis" by Yitzhak Katznelson which is stated as follows:

Corollary. Let $\mu\in M(\mathbb T)$. Then $$\sum\limits_{\tau\in\mathbb T}|\mu(\{\tau\})|^2=\lim\limits_{N\rightarrow\infty }\frac{1}{2N+1}\sum\limits_{-N}^N|\hat\mu(n)|^2.$$

It looks bizzare to me since the left hand sum up all variable $\tau$, while the other side not. What I only know is that $\mu(\{\tau\})=\frac{1}{2N+1}\sum\limits_{-N}^N|\hat\mu(n)|,$ but it seems impossible to derive the corollary from this relation.


(note added by YC: the result is stated without proof at the end of Chapter I Section 7.11 in the 1976 Dover edition)

$\endgroup$
4
  • 1
    $\begingroup$ it may help if you give the source more precisely (which corollary of which book? Katznelson's?) $\endgroup$ – Carlo Beenakker Feb 28 at 10:48
  • $\begingroup$ Notice that both sides of the stated equation are computing something like an $L^2$-norm. This is not surprising, because it's presumably some form of the Plancherel theorem. Though I do not quite understand it as stated. However, your "hoped for" formula looks very different: a point evaluation one side, and something like an $L^2$-norm on the other. That looks very unlikely to be true. $\endgroup$ – Matthew Daws Feb 28 at 11:44
  • 1
    $\begingroup$ I'm really sorry, It's a typo, the correct is: $\mu(\{\tau\})=\lim\limits_{N\rightarrow\infty}\sum\limits_{-N}^N\hat \mu(n)e^{in\tau}$. $\endgroup$ – Christoff_ferland Feb 28 at 12:09
  • $\begingroup$ A finite measure has at-most countably many atoms, that should make the definition of the LHS clear as it is the sum of at-most countably many non-zero values... $\endgroup$ – Asaf Feb 28 at 20:54
7
$\begingroup$

As far as I understand, $\mu$ is a Borel probability measure on the unit circle $\mathbb{T}$. Then $$ \frac{1}{2N+1}\sum\limits_{n=-N}^N|\hat\mu(n)|^2=\int\int\frac1{2N+1}\sum_{n=-N}^N (x/y)^nd\mu(x)d\mu(y). $$ The integrand $$ \frac1{2N+1}\sum_{n=-N}^N (x/y)^n $$ has absolute value at most 1 and converges pointwise to 1 for $x=y$ and to $0$ for $x\ne y$ (the sum is a geometric progression; you may sum it up to see that it is bounded when $x\ne y$). Thus, by the Dominated Convergence Theorem, the limit is the $\mu\times \mu$-measure of the diagonal $\{x=y\}$, which is exactly the LHS.

$\endgroup$
3
  • $\begingroup$ Why is the first equality right? I'm not quite sure of this, please let me know. $\endgroup$ – Christoff_ferland Feb 28 at 12:44
  • $\begingroup$ well, $\mu(n)=\int y^{-n} d\mu(y)$, $\bar{\mu}(n)=\int x^{n}d\mu(x)$, multiply to get $|\mu(n)|^2=\int \int (x/y)^{n} d\mu(x)d\mu(y)$ $\endgroup$ – Fedor Petrov Feb 28 at 13:00
  • $\begingroup$ I got your idea. Thanks a lot. $\endgroup$ – Christoff_ferland Feb 28 at 13:22

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.