Conversation started Jul 29, 2012 at 13:09.
Jul 29, 2012 13:09
Hey @JonasTeuwen, may I ask you a question?
@MattN. Yes. (Don't ask to ask. Well, not you 8-)).
@JonasTeuwen Thanks! Actually, now I'm not sure anymore whether I still have a question. I was going to ask you why the Dirichlet kernel is interesting.
Wait, before you say anything, let me add something:
From my notes I gather two things we can do with it:
(1) We can take an $f$ in $C(\mathbb R / \mathbb Z)$. If $D_n$ denotes the $n$-th Dirichlet kernel of $f$ then we know that $T_n (f) = \int_0^1 f(x) D_n(x) dx $ is a continuous linear functional.
Well, I can already make a very general remark.
(2) For $n \to \infty$ we know that $\int_0^1 | D_n(x) | dx$ goes to zero.
And you missed an $(x - y)$.
Jul 29, 2012 13:14
So why are these two facts useful?
First of all, it is a so-called integral kernel right?
@JonasTeuwen Where?
@MattN. Here $D_n(y - x)$.
That is, if $T$ is an operator, then $Tf$ can sometimes be given as $Tf(x) = \int K(x, y) f(y) \, \text{d}y$.
@JonasTeuwen Well, I'm not comfy with these yet. The only time I heard of this before was in Pseudo-differential operators class and I've not figured out yet what I want to do with them.
@JonasTeuwen Not in my notes:
Where $K(x, y)$ is some kind of function.
Yes, sure. Just explaining. You wanted that, right? Wanted to start with a general remark.
So often you have this abstract operator $T$ which can concretely given (perhaps for a subclass of functions) as integration against a so-called integral kernel.
Jul 29, 2012 13:17
@JonasTeuwen Sure!
Meh, they want to be an idiot.
What do you mean?
If you replace the second $x$ by $y - x$ or so you get the partial sum up to $n$. Of the Fourier series.
Anyway, this integral form can be very useful to analyse your operator. Sure you agree.
It is even nicer if it is a so-called convolution operator! Then we have $K(x - y)$ instead of $K(x, y)$.
More properties (yay!).
Wait, let me see if I understand what you're saying:
If I have any linear operator $T$ then I can write it as $T(f) = \int f K$ where $K$ is some kernel?
Not always, but often.
Jul 29, 2012 13:20
Does $T$ have to be continuous for this?
Check out Hilbert-Schmidt operator, that is one of that form. You will probably also see it in the FA course.
Depends on what norm you take of course. Don't worry about that for now.
I just want to give you the feeling it is not just some academic useless exercise.
And how do I find $K$?
Well, it actually is but not that useless.
That is for a later discussion (in general: very hard).
@MattN. how is my answer?
@MattN. Here you can see the form I propose: en.wikipedia.org/wiki/Dirichlet_kernel
Jul 29, 2012 13:21
@BenjaLim Sorry haven't had time to look and am busy talking to Jonas about important stuff right now. So this is a bad time.
In that way it is much more insightful what it actually does. You want to find 'good kernels' which give you nice properties.
The Dirichlet kernel is just an integral kernel like above, but it has very nice properties.
@MattN. ok. But do read it so I can improve if anything :D
It gives you a way to express the partial sums of your Fourier series in the form of an integral (which is useful).
So then you can derive properties from the integral form.
Wait, let me finish looking at Wikipedia.
That is basically the idea, but a very fruitful one.
Jul 29, 2012 13:24
Hey, Dirichlet kernel is sum over some characters (the ones centered at zero) and by magic, if I convolute $f$ with it I get the partial Fourier series of $f$ over those characters.
Ah not magic!
The sum is finite so I just swap sum and integral.
Wait no, let me do it on paper, one sec.
@JonasTeuwen Sorry but this is already not clear to me. The definition of convolution of two functions is over all of $\mathbb R$: $f \ast g = \int_{-\infty}^\infty f(x) g(y-x) dx $ (ugh I can never remember this formula : /). But the convolution of the Dirichlet kernel is on $\int_{-\pi}^\pi$.
@MattN. Not on a finite measure space :-).
Would you feel more comfortable if it said $\int_{S^1}$?
@JonasTeuwen I see. It's just over all of the domain.
@JonasTeuwen Not sure. Because I'm not sure I finally understand the difference (and consequences) between $S^1$ and $\mathbb R / \mathbb Z$.
@MattN. Basically the same.
I thought that if we write $S^1$, the group operation is $e^{ia} \cdot e^{ib} = e^{i (a+b)}$.
Don't worry, just think of periodic functions.
Jul 29, 2012 13:33
Versus the group operation on $\mathbb T$: $arg(e^{ia}) + arg(e^{ib}) = a + b$.
@JonasTeuwen I suspect the consequence is that on $\mathbb T$ my characters are $e^{2 \pi i k x}$ whereas on $S^1$ my characters are $e^{ikx}$. Is this correct?
@MattN. Yes. Just a $2 \pi$ scaling.
@JonasTeuwen Yay! This means I finally understand what $\mathbb T$ and $S^1$ mean!
Good.
Basically... nothing special.
Instead of considering periodic function on $\mathbf R$ you just consider functions on $S^1$ (non-periodic).
Ok. I can see now. It's not magic at all it's just natural, that $D_n \ast f$ gives me $\sum_{-n}^n \hat{f}_k \chi_k$.
@JonasTeuwen I know but that was not what I was confused about. I used to not see any difference between $S^1$ and $\mathbb T$.
Oh, the scaling?
@MattN. Yep. You've got it bro.
Jul 29, 2012 13:43
Ok, back to the original question. (Let me re-read your previous comments)
@JonasTeuwen Ok. Good, I think I understand.
I mean, what we can do with it: express a partial Fourier series as an integral.
But that doesn't seem very useful.
But ok, let's just take that as it is an accept it.
The lemma after that tells me that $$ \int_0^1 | D_n (x) | dx \to \infty$$ as $n \to \infty$. Is that more useful?
Isn't this the same as saying that the integral of the Fourier series of $f$ over $\mathbb T$ goes to $\infty$?
Yes, it is. Check out the book by Stein and Shakarchi.
@MattN. We have the absolute value.
How is this possible? That should be finite since the space has finite measure and hence every $L^2$ function is also $L^1$. By assumption $f$ is $L^2$ and by Parseval, $\|f\|_2 = \|\hat{f}\|_2$...
You would expert a bit that the higher frequencies damp out somewhat.
@MattN. Yeah, but... $|\cdot|$.
@JonasTeuwen Yes: $\| f\|_1 = \int |f| d \mu$, no?
@MattN. But what is $|\hat f|$?
Jul 29, 2012 13:53
Oops, I think I am mixing up $f$s and $D_n$s.
@JonasTeuwen Hey, I think I have it. These two lemmas are used to show that there exists $f \in C(\mathbb T)$ such that $\hat{f}(e^{ix0}) = \hat{f}(1) = \infty$.
 
Conversation ended Jul 29, 2012 at 13:56.