1
$\begingroup$

Let $f,g: \mathbb R \to \Bbb R$ be two $2\pi$ periodic functions.

(a) If $f$ is $C^{\infty}$ prove that for each $n$ there is $C_n \in \Bbb R$ such that$$\left|\hat{f}_k\right| \leq \frac{C_n}{\left|k\right|^n}\quad \forall k \in \Bbb Z \setminus \{0\}$$ where $\hat{f_k} = \frac{1}{2\pi}\int_0^{2\pi} f(t) e^{ikt} dt$ is the $k$-th Fourier coefficient of $f$.

(b) For $f \in C^{\infty}$ and $g \in L^{\infty}$ prove that $$\lim_{n\to \infty} \int_0^{2\pi} f(t) g(nt) dt = 2\pi \hat{f_0}\hat{g_0}$$

(c) Prove that (b) holds also if $f \in L^1$.

Now, I solved (a) using standard repeated integration by parts $n$ times, but what about the last two points? Can someone give me a hint? Thanks!

$\endgroup$
  • 1
    $\begingroup$ Hint: show the result for $f$ a trigonometric polynomial (where it is easy since if $m$ is its degree all except the constant $\hat{g_0}$ terms of the Fourier terms of $g((m+k)t, k \ge 1$ are orthogonal on $f$), and use that such approximate $L^1$ in norm plus the dominated convergence theorem $\endgroup$ – Conrad Sep 14 at 14:57
  • $\begingroup$ Thanks! If you want to, i would really appreciate if you could check out my answer below. In all the cases where we needed the dominated convergence theorem I used the fact that the fourier series converges in $L^2$ in norm, and $f \in L^2$ because of course is $C^{\infty}$. But now what about point(c) when $f$ is only in $L^1$? $\endgroup$ – astrobarrel Sep 15 at 9:37
0
$\begingroup$

So, I think I understood how to do point (b). As suggested in the comments, first prove it for a function $f$ like $$f_m(t)= \sum_{\left| k\right|\leq m} \hat{f_k} e^{ikt}$$ Therefore, if $n \geq m$ we can write $$\int_0^{2\pi}f(t)g(nt) dt= \sum_{\left| k\right|\leq m} \hat{f_k} \int_0^{2\pi} e^{ikt}g(nt)dt = \sum_{\left| k\right|\leq m} \hat{f_k} \int_0^{2\pi} \sum_{l \in \Bbb Z}e^{ikt} e^{inlt}\hat{g}_l dt$$ Now all the terms are zero except the one with $l = k = 0$ so all that stuff is equal to $$2\pi\hat{f_0}\hat{g_0}$$ as wanted. So now for each $n$ we get

$$\int_0^{2\pi}f(t)g(nt) dt = \sum_ k \hat{f_k} \int_0^{2\pi} e^{ikt}g(nt)dt = \sum_ { k\geq n } \hat{f_k} \int_0^{2\pi} e^{ikt}g(nt)dt +\hat{f_0}\hat{g_0} 2\pi $$

Letting $n$ go to infinity we see that the first term in the last member goes to zero so we get the result.

| cite | improve this answer | |
$\endgroup$
  • $\begingroup$ the last part needs a bit more rigor since it involves a remainder (infinite) sum $R_n$ in $k \ge n$ and then letting $n \to \infty$ and it is not obvious why the sum $R_n \to 0$ (each term does by Riemann Lebesgue and the boundness of $g$ but why the infinite sum does?); here one uses that for each $f \in L^1$ there are $f_k$ trigonometric polynomials $|f_k-f|_1 \to 0$; this means in particular $\hat{f_{k,0}} \to \hat{f_0}$; but $|\int_0^{2\pi}f_k(t)g(nt)dt-\int_0^{2\pi}f(t)g(nt)|dt \le |f-f_k|_{1}||g||_{\infty} \to 0$ with $k$, uniform in $n$ so conclude $\endgroup$ – Conrad Sep 15 at 11:33
  • $\begingroup$ For the reimainder what if I just estimate $\left|R_n\right| \leq \Vert g\Vert_{\infty} \sum_{k\geq n}\left|\hat{f_k}\right|$ so that this goes to zero when $n$ goes to infinity because $\sum_k\left|\hat{f_k}\right|$ is a convergent series? (by point (a)). Does it make sense? $\endgroup$ – astrobarrel Sep 15 at 13:43
  • $\begingroup$ Ok, I see what you mean. My argument is ok for point (b) but not (c) and we need yours to take care of point (c). Thanks a lot! $\endgroup$ – astrobarrel Sep 15 at 13:50
  • 1
    $\begingroup$ yes - point b can be addressed directly, because you have good estimates but point c needs approximation in norm $\endgroup$ – Conrad Sep 15 at 14:24
  • $\begingroup$ very clear, thank you very much $\endgroup$ – astrobarrel Sep 15 at 14:25

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.