MathOverflow is a question and answer site for professional mathematicians. Join them; it only takes a minute:

Sign up
Here's how it works:
  1. Anybody can ask a question
  2. Anybody can answer
  3. The best answers are voted up and rise to the top

Let $c_0$ be the Banach space of doubly infinite sequences $$\lbrace a_n: -\infty\lt n\lt \infty, \lim_{|n|\to \infty} a_n=0 \rbrace.$$ Let $T$ be the space of $2\pi$ periodic functions integrable on $[0,2\pi]$.

Let $$S=\lbrace \lbrace a_n\rbrace \in c_0: a_n=\hat{f}(n) \forall n \mbox{ for some function } f\in T\rbrace,$$ where $\hat{f}(n)$ denotes the $n$-th Fourier coefficient of $f$, i.e. $$\hat{f}(n)=\frac{1}{2\pi}\int_{-\pi}^\pi f(x)e^{-inx}\,dx.$$ When I was a graduate student, I was told that no known characterizations of $S$ were known. Is this still true?

share|cite|improve this question
Well, what does "characterisation" count as? S is just the Fourier algebra of $\mathbb Z$, which is the convolution of two $\ell^2(\mathbb Z)$ functions. But that follows rather immediately from it also being the Fourier transform of $L^1(\mathbb T)$. – Matthew Daws Nov 19 '10 at 13:39
The Fourier transform takes $L^2(\mathbb T) \rightarrow \ell_2(\mathbb Z)$ unitary (if you normalise the measure on $[0,2\pi]$ correctly). Every $L^1(\mathbb T)$ function is the pointwise product of two $L^2(\mathbb T)$ functions, and the fourier transform converts pointwise product to convolution, so that's the proof! (Maybe a few $\epsilons$s and $\delta$s are needed to make this 100% rigorous). – Matthew Daws Nov 19 '10 at 14:22
The product of two (periodic) $L^2$ functions is an $L^1$ function, and conversely every $L^1$ function is the square of a (complex) $L^2$ function. And multiplying functions corresponds to convolving Fourier coefficients. Is that it?. – Tom Goodwillie Nov 19 '10 at 14:27
The statement that the Fourier transform of $L^1(T)$ cannot be characterized as a subspace of $c_0(Z)$ has always struck me as something of a folk principle rather than a precise theorem. I think there is a theorem in this direction which says that there exists a sequence $(a_n)$ which is the Fourier series of an integrable function, but where $(|a_n|)$ is not the Fourier series of any integrable function -- however I'd have to check this as I've forgotten the example. – Yemon Choi Nov 19 '10 at 15:16
Provided $f,g$ are in $L^2$, not in $L^1$, that first expression is correct. You can even see it as a consequence of Parseval's lemma, if you group the integrand as $f(x)$ and $g(x)e^{-inx}$ and use that phase rotation in physical space is the same as translation in frequency space... – Willie Wong Nov 19 '10 at 16:40
up vote 1 down vote accepted

This is to summarize what were discussed in the comments, so the title will not be listed as unanswered.

The linear subspace $S$ of $c_0(\mathbb{Z})$ is equal to the convolution product of two copies of $\ell^2(\mathbb{Z})$.

More precisely, $\lbrace a_n \rbrace$ is in $S$ if and only if there exist two sequences $\lbrace b_n \rbrace$ and $\lbrace c_n \rbrace$ in $\ell^2(\mathbb{Z})$ such that $$a_n=\sum_{k=-\infty}^\infty b_k c_{n-k} $$ for all $n$.

This follows since every function in $L^1[0,2\pi]$ is a product of two functions in $L^2[0,2\pi]$, and that for any functions $f,g$ in $L^2[0,2\pi]$ one has, by Parseval identity, $$\frac{1}{2\pi}\int_{-\pi}^\pi f(x)g(x)e^{-inx}dx=\frac{1}{2\pi}\int_{-\pi}^\pi f(x)\overline{h(x)}e^{-inx}dx=\sum_{k=-\infty}^\infty \hat{f}(k) \hat{g}(n-k)$$ where $h(x)=\overline{g(x)}$.

(One also uses that the mapping that maps each $f$ in $L^2[0,2\pi]$ to its Fourier coefficient sequence in $\ell^2(\mathbb{Z})$ is a surjective isomorphic isometry.)

share|cite|improve this answer

Your Answer


By posting your answer, you agree to the privacy policy and terms of service.

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