Is the the following true: "Assume $f\in L^1[0,1]$ and $\hat{f}\in L^1(\mathbb{R})$, then $f$ is differentiable a.e"
-
it seams that whenever Fourier inversion theorem holds $f(x)=\int_{\mathbb{R}}\hat{f}(y)e^{2\pi ixy}dy$ a.e. Then $f$ Is differentiable a.e. – BigM Mar 14 '14 at 16:39
-
1So you require $\widehat f$ to be integrable on $\mathbb R$? If so, you should modify your post. – Etienne Mar 14 '14 at 16:50
-
Yes.thats one of my assumptions – BigM Mar 14 '14 at 16:54
1 Answers
Theorem (Bernstein). If $f:\mathbb R\to\mathbb R$ is compactly supported and $C^\alpha$ continuous for some $\alpha>1/2$, then $\hat f\in L^1(\mathbb R)$.
This theorem gives a negative answer to your question, since an appropriate version of the Weierstrass function satisfies the above conditions.
The proof is the same as the proof of the corresponding result for Fourier series. For the sake of completeness, I modified the proof and included it below.
Proof. For $h\in\mathbb R$ write $f_h(t)=f(t-h)$. The Fourier transform of $f_h-f$ is $$(e^{-i \xi h}-1)\hat f(\xi)$$ The $L^2$ norm of $f_h-f$ is $O(h^\alpha)$. By the Parseval identity $$\int_{\mathbb R}|e^{-i \xi h}-1|^2 |\hat f(\xi)|^2\,d\xi = O(h^{2\alpha})$$ We want to bound $ |e^{-i\xi h}-1|$ from below. There is no uniform bound for all $\xi$, but if we focus on some dyadic range $D_k=\{\xi: 2^k\le |\xi|< 2^{k+1}\}$, then choosing $h=\frac{2\pi}{3}\cdot 2^{-k}$ gives good result: $\frac{2\pi}{3}\le |\xi h|< \frac{4\pi}{3}$, which keeps $e^{-i\xi h}$ far away from $1$. Thus, $$ \int_{D_k} |\widehat{f}(\xi)|^2\,d\xi = O(2^{-2\alpha k}) $$ By the Cauchy-Schwarz inequality, $$ \int_{D_k} |\widehat{f}(\xi)| \,d\xi = O(2^{k/2})\, O(2^{-\alpha k})= O(2^{k(1/2-\alpha)}) $$ and the sum over $k$ converges.

- 9,683
-
Is there any converse to Bernstein's theorem? Or at least a proof that the $\frac 12$ is sharp? I found this for the case of the torus/Fourier series https://www.epfl.ch/labs/pde/wp-content/uploads/2020/03/Lecture3.pdf (Thm. 2.1) "We observe that the preceding theorem is actually quite sharp, in that there is no embedding $C^{1/2}(S^1) \hookrightarrow \mathcal A(S^1)$" where $\mathcal A(S^1)$ is the space of absolutely convergent Fourier series, i.e. $f\in L^1(S^1)$ with Fourier coefficients in $\ell^1(\mathbb Z)$)." – D.R. Mar 14 '24 at 00:58