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

The Fourier Transform $\mathcal{F}:L^1(\mathbb{R})\to C_0(\mathbb{R})$ is an injective, bounded linear map that isn't onto. It is known (if I remember correctly) that the range isn't closed, but is dense in $C_0$. Everything I have read/heard says that the range is "difficult to describe".

But, since $\mathcal{F}$ is injective and continuous, the image of $L^1$ must be Borel inside of $C_0$. Is anything else known about its descriptive complexity? If not, might this be an example of a natural set of high Borel rank?

This question may be relevant. Thanks for any insight/references.

share|cite|improve this question
Isn't this space related to the Wiener Algebra? This is the space of $2\pi$-periodic functions with absolute summable Fourier coefficients. I vaguely remember that I heard somebody calling the image of $L^1$ under Fourier transform also "Wiener Algebra"... – Dirk Apr 9 '13 at 8:46
Dirk: yes. Also called the Fourier algebra – Yemon Choi Apr 9 '13 at 18:28
Ah, I see. Interesting that the Wikipedia pages of the Wiener algebra and the Fourier algebra do not mention each other... – Dirk Apr 10 '13 at 6:37
up vote 5 down vote accepted

Here is an attempt, somewhat rough around the edges but I think it works:

Claim: The range of the Fourier transform $\mathcal{F}:L^1(\mathbb R)\to C_0(\mathbb R)$ is a Borel set in $C_0(\mathbb R)$ of the form

\begin{equation} \bigcap_{k=1}^\infty \bigcup_{N=1}^\infty \bigcap_{m,n\geq N} E_{m,n,k} \end{equation}

where each $E_{m,n,k}$ is an $F_\sigma$.

Proof: Consider the cutoff functions $\{e^{-a\pi|t|}\}$ and fix a sequence $a_n\to 0$. It is a fact that a function $g\in C_0(\mathbb R)$ is the Fourier transform of some $f\in L^1$ if and only if the sequence \begin{equation} T_n(g)(x) := \int_{\mathbb R} e^{-a_n\pi |t|} g(t)e^{2\pi itx} dt \end{equation} is Cauchy in $L^1$, in which case if we put $f=\lim T_ng$ then $g =\widehat{f}$. Let $R$ denote the range of the Fourier transform in $C_0$. If we define

\begin{equation} E_{n,m,k} = \lbrace g\in C_0: T_ng, T_mg\in L^1 ,{||T_n(g)-T_m(g)||}_1 \leq\frac{1}{k} \rbrace \end{equation}

then $R$ has the claimed form.

To prove that $E_{m,n,k}$ is an $F_\sigma$, put $T_n(g)-T_m(g):=T_{mn}(g)$, explicitly

\begin{equation} T_{mn}g(x):= T_ng(x)-T_mg(x) =\int_{\mathbb R} h_{mn}(x,t)g(t) dt \end{equation}

where we define

\begin{equation} h_{mn}(x,t) = (e^{-a_m\pi|t|} - e^{-a_n\pi|t|} )e^{2\pi itx} \end{equation}

Note that by monotone convergence, $g\in E_{m,n,k}$ if and only if two conditions are satisfied: first, for fixed $n$, we need $T_ng\in L^1$. By monotone convergence this is equivalent to: There exists an integer $N$ such that for all integers $d\geq 1$,

\begin{equation} || {\bf 1}_{[-d,d]}(x)T_ng(x)||_1 \leq N. \end{equation}

By dominated convergence the set of all such $g$ obeying this for fixed $N,n,d$ is closed, so the set that obeys this for some $N$ and all $d$ (with $n$ held fixed) is an $F_\sigma$. Thus for fixed $m,n$ the set of $g$ for which $T_ng, T_mg\in L^1$ is an $F_\sigma$. Additionally, for all integers $d\geq 1$, we need the condition

\begin{equation} || {\bf 1}\_{[-d,d]}(x)T_{mn}g(x)||\_1 \leq \frac{1}{k}. \end{equation}

But similarly, the set of $g$ obeying this for fixed $m,n,k,d$ is closed in $C_0(\mathbb R)$, and we conclude that each $E_{m,n,k}$ is an $F_\sigma$.

share|cite|improve this answer
For some reason I can't get the last displayed equation to parse. – Mike Jury Apr 9 '13 at 12:58
replaced _ by _ in the last displayed equation. – Peter Michor Apr 9 '13 at 13:05
@Peter Michor: Thank you! – Mike Jury Apr 9 '13 at 13:38

It seems that Mike's argument shows that in fact the range $R$ of the Fourier transform is $F_{\sigma\delta}$ in $C_0(\mathbb R)$. Define the $T_ng$ as above, $$T_ng(x)=\int_{\mathbb R} e^{-a_n\pi\vert t\vert}g(t) e^{2i\pi tx}dt\;.$$ Then a function $g\in C_0(\mathbb R)$ is in $R$ iff two things hold:

(i) all $T_ng$ are in $L^1$;

(ii) the sequence $(T_ng)$ is Cauchy in $L^1$.

Condition (i) can be written as follows: $$ \forall n\;\exists N\in\mathbb N\; \left ( \forall d\in\mathbb R_+ \;:\;\int_{-d}^d \vert T_ng(x)\vert dx\leq N\right)$$
By dominated convergence, the condition under brackets is closed with respect to $g$; so (i) defines an $F_{\sigma\delta}$ subset of $C_0(\mathbb R)$.

Condition (ii) reads $$\forall k\in\mathbb N\;\exists N \;\left(\forall p,q\geq N\; \forall d \;:\; \int_{-d}^d \vert T_pg(x)-T_qg(x)\vert dx\leq \frac 1k\right)$$ By dominated convergence again, the condition under brackets is closed wrt $g$, so (ii) defines an $F_{\sigma\delta}$ subset of $C_0(\mathbb R)$.

Altogether, $R$ is the intersection of two $F_{\sigma\delta}$ sets, hence an $F_{\sigma\delta}$ subset of $C_0(\mathbb R)$. I would be extremely surprised if it were better than that; i.e. I "conjecture" that it is not $G_{\delta\sigma}$.

share|cite|improve this answer
Yes, that's much cleaner. And I would also suspect it is not $G_{\delta\sigma}$. – Mike Jury Apr 13 '13 at 12:53

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.