1

Which conditions must fulfill $f(t)$ to have an absolute-integrable Fourier Transform $F(w)$: $\int\limits_{-\infty}^\infty |F(w)| dw < \infty$?

  1. At first, thinking in $f(t)$ as an arbitrary one variable function which has a Fourier Transform
  2. Secondly, for a specific $f(t)$ that is time-limited and continuous, so its also compact-supported and bounded ($\sup_t |f(t)| < \infty$), and has unlimited bandwidth.
  3. Third, adding to the conditions of point (2) that it has a Fourier Transform that fulfill the Riemann-Lebesgue Lemma, and also has finite-energy $\int_{t_0}^{t_F} |f(t)|^2\,dt < \infty$, is absolute-integrable $\int_{t_0}^{t_F} |f(t)|\,dt < \infty$, and has zero values at the edges of its domain $f(t_0)=f(t_F)=0$.
  4. Same as third, but requiring now that $f(t_0)\neq 0$ and/or $f(t_F) \neq 0$.

The actual order of requirement following how restrictive they are is $(3)\rightarrow (4)\rightarrow (2)\rightarrow(1)$.

Joako
  • 1,380
  • This is no simple condition for defining the Fourier transform of $L^1$. – reuns Nov 29 '21 at 19:39
  • @reuns But there are any combinations of requirements? Which ones?... by myself, adding to (3) that $f(t)$ is of bounded variation $V(t) < \infty$ and also has bounded derivative $||f'(t)||\infty < \infty$ will bound the $||F(w)||_1 < \infty$ if I had made right some manipulations through the Cauchy-Schwarz Inequality, but this is requiring too many restrictions I think since, as example, $f(t) = t,\log(t^2)/2,,|t|\leq 1$ has unbounded $||f'(t)||\infty$ but bounded $||F(w)||_1$... – Joako Nov 29 '21 at 19:53
  • @reuns (...) looking for answers I found someone saying that bounded $||F(w)||_1$ implies $f(t)$ continuous, but I don´t know if the converse is true. – Joako Nov 29 '21 at 19:54
  • @reuns PS: I have not included my attempts (yet), because I don´t want to skew/misdirect answers in any way. Hope you can elaborate in your first comment as an answer. Beforehand, thanks you very much. – Joako Nov 29 '21 at 20:17

1 Answers1

3
    • Notice first that if $f$ is such that its Fourier transform is positive, then $$ \int_{-\infty}^\infty |F| = \int_{-\infty}^\infty F = f(0) \leq \|f\|_{L^\infty} $$ so in this case it is sufficient to be bounded. But this is a rather special case.
  • In general, at least $1/2$ derivative has to be in $L^2$. More precisely, by the Cauchy-Schwarz inequality, for any $s>1/2$ $$ \int_{-\infty}^\infty |F| \leq \left(\int_{-\infty}^\infty (1+|2\pi w|^2)^s\,|F(w)|^2\,\mathrm d w\right)^{1/2} \left(\int_{-\infty}^\infty (1+|2\pi w|^2)^{-s}\,\mathrm d w\right)^{1/2}. $$ Since $s>1/2$, the last integral $C = \int_{-\infty}^\infty (1+|w|^2)^{-s}\,\mathrm d w$ is finite, so as $(1+|w|^2)^{s/2}\,F(w)$ is the Fourier transform of $(1-\Delta)^s\,f$ and the Fourier transform preserves the $l^2$ norm $$ \int_{-\infty}^\infty |F| \leq C\, \|(1-\Delta)^{s/2}f\|_{L^2} = C\, \|f\|_{H^s}. $$ This space $H^s$ is the Sobolev space of order $s$. Since every function with integrable function transform is bounded, the order $s>1/2$ is optimal in the sense that one can find unbounded functions in $H^{1/2}$. Of course it is sufficient to take for example $s=1$, in which case the above inequality becomes $$ \int_{-\infty}^\infty |F| \leq C\, \left(\int_{-\infty}^\infty |f|^2 + |f'|^2\right)^{1/2}. $$

  • For the reader that knows Besov spaces, one can get slightly more precise by replacing $H^s$ with $s>1/2$ by $B^{1/2}_{2,1}$ which is an intermediate space between $H^s$ and $H^{1/2}$. But this is a subject with a huge literature, you can find papers easily on the subject by typing "Integrability of Fourier transform" in Google for instance. Also, the space of functions with integrable Fourier transforms is sometimes called a Wiener algebra. You can find a survey on these questions called "The Wiener algebra of absolutely convergent Fourier integrals: an overview".

  • Being continuous and converging to $0$ at infinity is also not sufficient (i.e. the converse of Riemann-Lebesgue Lemma is false), see A Fourier transform of a continuous $L^1$ function.

  1. The fact that the function is compactly supported will not change anything as what is needed is regularity.

  2. Similarly, to have a bounded integral or to have a bounded energy will not help.

  3. Worse than 3, so still not sufficient.

LL 3.14
  • 12,457
  • On the inequality of the second point, the integral $\int |f'|^2$ could been split by hölder inequality (with $p=1$) as $||f'||_\infty \cdot \int |f'|$, so, since I believe that $\int |f'| \cong V(t)$ its total variation, so, if I add that the functions have bounded derivative and bounded variation (5), it will make integrable $|F(w)|$, Is this right?... If the function $f(t)$ fulfill (5), Does it implies is Differentiable?... Or could have integrable $|F(w)|$ limited by fulfilling (5) but allowing finite-size jump-discontinuities on $f'(t)$? (as example, $f(t) = \sin(\pi|t|)$) – Joako Nov 29 '21 at 23:30
  • 1
    Yes of course if $f'$ is integrable and bounded, then it is in $L^2$, so it works too. And yes, $f'$ can be bounded with finite size jump-discontinuities. Notice that $f'∈ L^2$ is locally much weaker than $f'∈ L^\infty$ as it allows infinite size discontinuities for $f'$ – LL 3.14 Nov 30 '21 at 01:03