How to show for $alphain (0,1)$, any $fin C^alpha([0,1]/{sim})$ has a Fourier series $S_nf$ uniformly...
$begingroup$
Technically homework(a midterm) but its over and I'm itching to know the solution. I know how to show it for $alpha>1/2$ (the Fourier series will converge absolutely), but apparently its true for any $alpha$; the question guided me as follows:
- Show that if a equicontinuous sequence of functions ($f_n$) converges pointwise to $f$, then $f_n$ converges uniformly to $f$.
- Show for $f∈ C^alpha([0,1]/{sim})$ that $S_nf → f$ pointwise.
- Show that the sequence $(S_nf)$ is equicontinuous and conclude.
1 and 2 posed no problems to me but I could not do 3. Any help? In addition, I would not mind other ways to prove the result.
analysis fourier-analysis
$endgroup$
add a comment |
$begingroup$
Technically homework(a midterm) but its over and I'm itching to know the solution. I know how to show it for $alpha>1/2$ (the Fourier series will converge absolutely), but apparently its true for any $alpha$; the question guided me as follows:
- Show that if a equicontinuous sequence of functions ($f_n$) converges pointwise to $f$, then $f_n$ converges uniformly to $f$.
- Show for $f∈ C^alpha([0,1]/{sim})$ that $S_nf → f$ pointwise.
- Show that the sequence $(S_nf)$ is equicontinuous and conclude.
1 and 2 posed no problems to me but I could not do 3. Any help? In addition, I would not mind other ways to prove the result.
analysis fourier-analysis
$endgroup$
add a comment |
$begingroup$
Technically homework(a midterm) but its over and I'm itching to know the solution. I know how to show it for $alpha>1/2$ (the Fourier series will converge absolutely), but apparently its true for any $alpha$; the question guided me as follows:
- Show that if a equicontinuous sequence of functions ($f_n$) converges pointwise to $f$, then $f_n$ converges uniformly to $f$.
- Show for $f∈ C^alpha([0,1]/{sim})$ that $S_nf → f$ pointwise.
- Show that the sequence $(S_nf)$ is equicontinuous and conclude.
1 and 2 posed no problems to me but I could not do 3. Any help? In addition, I would not mind other ways to prove the result.
analysis fourier-analysis
$endgroup$
Technically homework(a midterm) but its over and I'm itching to know the solution. I know how to show it for $alpha>1/2$ (the Fourier series will converge absolutely), but apparently its true for any $alpha$; the question guided me as follows:
- Show that if a equicontinuous sequence of functions ($f_n$) converges pointwise to $f$, then $f_n$ converges uniformly to $f$.
- Show for $f∈ C^alpha([0,1]/{sim})$ that $S_nf → f$ pointwise.
- Show that the sequence $(S_nf)$ is equicontinuous and conclude.
1 and 2 posed no problems to me but I could not do 3. Any help? In addition, I would not mind other ways to prove the result.
analysis fourier-analysis
analysis fourier-analysis
edited Jan 24 '18 at 14:10
Calvin Khor
asked Nov 8 '14 at 19:07
Calvin KhorCalvin Khor
12.3k21438
12.3k21438
add a comment |
add a comment |
3 Answers
3
active
oldest
votes
$begingroup$
Suppose that $|f(x)|le C$ and $|f(x)-f(y)|le C|x-y|^alpha$.
Express the Difference Using the Dirichlet Kernel
Using the Dirichlet Kernel, we get
$$
begin{align}
|S_nf(x)-f(x)|
&=left|,int_{-1/2}^{1/2}frac{sin((2n+1)pi y)}{sin(pi y)}[f(x-y)-f(x)],mathrm{d}y,right|\
&=left|,sum_{k=-n}^nint_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}frac{sin((2n+1)pi y)}{sin(pi y)}[f(x-y)-f(x)],mathrm{d}y,right|tag{1}
end{align}
$$
Estimate each Integral Using the smoothness of $boldsymbol{f}$
Since $left|,frac{sin((2n+1)pi y)}{sin(pi y)},right|lefrac{2n+1}{big|2|k|-1big|}$ and each interval is $frac1{2n+1}$ wide, we can bound
$$
begin{align}
left|,int_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}frac{sin((2n+1)pi y)}{sin(pi y)}[f(x-y)-f(x)],mathrm{d}y,right|
&lefrac{C}{big|2|k|-1big|}left(frac{2|k|+1}{4n+2}right)^alphatag{2}
end{align}
$$
Estimate each Integral Using Cancellation from $boldsymbol{sin((2n+1)pi x)}$
For $|y|lefrac12$, we have $|2y|le|sin(pi y)|le|pi y|$, and because
$$
int_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}sin((2n+1)pi y),mathrm{d}y=0tag{3}
$$
and
$$
int_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}|sin((2n+1)pi y)|,mathrm{d}y=frac2{(2n+1)pi}tag{4}
$$
if we let $m_k$ be the middle of the range of $frac{f(x-y)-f(x)}{sin(pi y)}$ on $left[frac{2k-1}{4n+2},frac{2k+1}{4n+2}right]$, for $kne0$, we can bound
$$
begin{align}
&left|,int_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}sin((2n+1)pi y)frac{f(x-y)-f(x)}{sin(pi y)},mathrm{d}y,right|\
&=left|,int_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}sin((2n+1)pi y)left[frac{f(x-y)-f(x)}{sin(pi y)}-m_kright],mathrm{d}y,right|\
&lefrac1{(2n+1)pi}frac{overbrace{pifrac{2|k|+1}{4n+2}}^{sin(pi y)}overbrace{C(2n+1)^{-alpha}vphantom{frac{|}2}}^{Delta (f(x-y)-f(x))}+overbrace{2Cvphantom{()^1}}^{f(x-y)-f(x)}overbrace{pi(2n+1)^{-1}}^{Deltasin(pi y)}}{underbrace{frac{4k^2-1}{(2n+1)^2}}_{sin^2(pi y)}}\
&=frac{C(2n+1)^{-alpha}}{4|k|-2}+frac{2C}{4k^2-1}tag{5}
end{align}
$$
Use each Estimate in its Proper Place
If we use estimate $(2)$ for $kle m=n^{frac{alpha}{alpha+1}}$ and estimate $(5)$ for $kgt m$, then we get
$$
begin{align}
sum_{|k|le m}frac{C}{big|2|k|-1big|}left(frac{2|k|+1}{4n+2}right)^alpha
&lefrac{C}{(4n+2)^alpha}left[1+6sum_{k=1}^m(2k+1)^{alpha-1}right]\
&lefrac{C}{(4n+2)^alpha}frac3alpha(2m+1)^alpha\
&simfrac{3C}{alpha2^alpha}n^{-fracalpha{alpha+1}}tag{6}
end{align}
$$
and
$$
begin{align}
sum_{mlt|k|le n}frac{C(2n+1)^{-alpha}}{4|k|-2}
&lefrac{C}{2^{alpha+1}}frac{H_n}{n^alpha}\
&simfrac{C}{2^{alpha+1}}frac{log(n)}{n^alpha}\
&=oleft(n^{-frac{alpha}{alpha+1}}right)tag{7}
end{align}
$$
and
$$
begin{align}
sum_{mlt|k|le n}frac{2C}{4k^2-1}
&le Csum_{k=m}^inftyfrac1{k^2-1}\
&=frac{C}{2}sum_{k=m}^inftyleft(frac1{k-1}-frac1{k+1}right)\
&=frac{C}{2}left(frac1{m-1}+frac1mright)\
&sim Cn^{-frac{alpha}{alpha+1}}tag{8}
end{align}
$$
Put Everything Together
Therefore, we have uniform convergence:
$$
|S_nf(x)-f(x)|leleft(1+frac3{alpha2^alpha}right)Cn^{-frac{alpha}{alpha+1}}tag{9}
$$
$endgroup$
$begingroup$
In Wikipedia, it is claimed that we should be able to get uniform convergence on the order of $O(log(n)n^{-alpha})$; that is, the order of the term in $(7)$ that is ignored as too small. I will have to look into this.
$endgroup$
– robjohn♦
Nov 11 '14 at 20:30
$begingroup$
Thanks, I will read through your answer later today (the formulae are cut off on my mobile phone)
$endgroup$
– Calvin Khor
Nov 12 '14 at 8:07
$begingroup$
This, of course, is a very good answer, but I think that it shows more than it is desired. The question only wants to prove equicontinuity of $S_nf$, which then yields uniform convergence (without the asymptotic (9), of course). I guess that the proof of such equicontinuity should be something along the same lines as this answer, only simpler.
$endgroup$
– Giuseppe Negro
Nov 12 '14 at 18:24
add a comment |
$begingroup$
While I accepted the answer above, this is how my lecturer (and later my friend) explained it to me (the exam is tomorrow). We first define $$g_n(x):=f(x) - S_n f(x)$$ just to remind ourselves that we need to be careful of cancellations. Then uniform convergence of $S_nf$ to $f$ is equivalent to showing $g_n→ 0$ uniformly; since we know (part 2) that $g_n(x) → 0$ pointwise, it suffices to show $g_n$ is uniformly continuous (by part 1).
Since $newcommand{d}{text{d}}newcommand{intT}{∫_{-1/2}^{1/2}}g_n(x) = f(x)times 1 - intT f(z-x) D_n(z) d z = intT [f(x) -f(z-x)] D_n(z) d z$ ,
begin{align}
|g_n(x) - g_n(y)| ≤ intT |D_n(z)|underbrace{|f(x) - f(z-x) - f(y) + f(z-y)|}_{(star)} d z
end{align}
We now need to find bounds independent of $n$ . We use a simple bound for the Dirichlet kernel $D_n$: as there is $C_0$ such that $|sin(2π z)|>C_0|z|$ on $[-1/2,1/2]$,
$$|D_n(z)| < frac{C_1}{|z|} $$
Since we don't gain too much form a simple bound we need to bound $(star)$. The trick is to use two different bounds, each good on different sets:
begin{align} |color{red}{f(x) - f(z-x)} - color{blue}{f(y) + f(z-y)}| &leq C_3|z|^alpha \
|color{red}{f(x)} - color{blue}{f(z-x)} - color{red}{f(y)} + color{blue}{f(z-y)}|
&leq C_3|x-y|^alpha \
end{align}
Thus $$|g_n(x) - g_n(y)| leq ∫_{|z|leq|x-y|} C_4|z|^{alpha-1} d z + |x-y|^alpha ∫_{|x-y|<|z|<1/2}frac{C_5}{z} d z = I_1 + I_2 $$
Now $I_1$ is $mathcal{O}(|x-y|)$ because $|z|^{alpha-1}$ is $L^1([-1/2,1/2])$. The second we compute,
$$I_2 = C_5 |x-y|^alphaleft(logfrac{1}{2} + logfrac{1}{|x-y|}right) $$
And we win because polynomials beat logarithms.
$endgroup$
$begingroup$
Nice answer. Do you still remember how to show the pointwise convergence?
$endgroup$
– PhoemueX
Jan 26 at 10:25
$begingroup$
@PhoemueX Thanks. I believe the "Dirichlet–Dini Criterion" is sufficient, see math.stackexchange.com/questions/1109758/… . Does that make sense? I recall proving it directly, probably by using the $2pi$ periodicity to introduce a $f(x)-f(x-y)$ integrand somewhere in $S_nf $ but my quick attempts weren't enough to bring this memory back.
$endgroup$
– Calvin Khor
Jan 26 at 13:54
add a comment |
$begingroup$
Actually I just used estimates in the same spirit of the ones used in the answer of Calvin Khor, proving directly the result without appelling to Ascoli-Arzelà theorem and obtaining also an estimate of the rate of convergence, so I think that it makes sense to post this answer.
First, get $f:mathbb{R}to mathbb{C}$ be a $2pi$-periodic $alpha$-Hölder continuous function and for each $xinmathbb{R}$ define $f_x(t):=f(x+t)-f(x)$.
We want to prove that:
$$sup_{xin[-pi,pi]}left|sum_{n=-N}^Nhat{f}(n)e^{inx}-f(x)right|to0, Nto+infty.$$
Now:
$$sum_{n=-N}^Nhat{f}(n)e^{inx}-f(x)=int_{-pi}^pi (f(x+t)-f(x))frac{sinleft((N+frac{1}{2})tright)}{sinleft(frac{t}{2}right)}frac{operatorname{d}t}{2pi}=int_{-pi}^pi f_x(t)frac{sinleft((N+frac{1}{2})tright)}{sinleft(frac{t}{2}right)}frac{operatorname{d}t}{2pi} \ = int_{-pi}^pi f_x(t)cotleft(frac{t}{2}right)sin(Nt)frac{operatorname{d}t}{2pi}+int_{-pi}^pi f_x(t)cosleft(Ntright)frac{operatorname{d}t}{2pi}= \ frac{1}{pi}int_{-pi}^pi frac{f_x(t)}{t}sin(Nt)operatorname{d}t+int_{-pi}^pi f_x(t)left(cotleft(frac{t}{2}right)-frac{2}{t}right)sin(Nt)frac{operatorname{d}t}{2pi}+int_{-pi}^pi f_x(t)cosleft(Ntright)frac{operatorname{d}t}{2pi}.$$
Now, the only problematic integral is the first, so let's estimate only this one.
We have that:
$$int_{-pi}^pi f_x(t)cotleft(frac{t}{2}right)sin(Nt)frac{operatorname{d}t}{2pi} = -int_{-pi}^pi f_x(t+frac{pi}{N})cotleft(frac{t+frac{pi}{N}}{2}right)sin(Nt)frac{operatorname{d}t}{2pi},$$
so:
$$left|int_{-pi}^pi f_x(t)cotleft(frac{t}{2}right)sin(Nt)frac{operatorname{d}t}{2pi}right| = left|frac{1}{2} int_{-pi}^pi left(f_x(t)cotleft(frac{t}{2}right)-f_x(t+frac{pi}{N})cotleft(frac{t+frac{pi}{N}}{2}right)right)sin(Nt)frac{operatorname{d}t}{2pi}right| \ le frac{1}{4pi} int_{-pi}^pi left|f_x(t)cotleft(frac{t}{2}right)-f_x(t+frac{pi}{N})cotleft(frac{t+frac{pi}{N}}{2}right)right|operatorname{d}t \ = frac{1}{4pi} int_{-pi}^pi left|f_x(t-frac{pi}{2N})cotleft(frac{t-frac{pi}{2N}}{2}right)-f_x(t+frac{pi}{2N})cotleft(frac{t+frac{pi}{2N}}{2}right)right|operatorname{d}t$$
So we need an uniform estimate in $x$ for the quantity:
$$int_{-pi}^pi left|f_x(t-h)cotleft(frac{t-h}{2}right)-f_x(t+h)cotleft(frac{t+h}{2}right)right|operatorname{d}t$$
for $hto 0$.
Split the integral for $|t|<2|h|$ and for $2|h|le|t|le pi$ for $|h|<1$. So:
$$int_{|t|<2|h|} left|f_x(t-h)cotleft(frac{t-h}{2}right)-f_x(t+h)cotleft(frac{t+h}{2}right)right|operatorname{d}t\ le int_{|t|<2|h|} left|f_x(t-h)cotleft(frac{t-h}{2}right)right|operatorname{d}t+int_{|t|<2|h|} left|f_x(t+h)cotleft(frac{t+h}{2}right)right|operatorname{d}t le \ int_{|t|<2|h|} |t+h|^alphaleft|cotleft(frac{t+h}{2}right)right|operatorname{d}t+int_{|t|<2|h|} |t-h|^alphaleft|cotleft(frac{t-h}{2}right)right|operatorname{d}t\ le c_1 int_{|t|<4|h|} |t|^{alpha-1}operatorname{d}t = c_2 |h|^alpha.$$
While, for $2|h|le|t|le pi$ we have:
$$int_{2|h|le|t|le pi} left|f_x(t-h)cotleft(frac{t-h}{2}right)-f_x(t+h)cotleft(frac{t+h}{2}right)right|operatorname{d}t \ le int_{2|h|le|t|le pi} left|f_x(t-h)left(cotleft(frac{t-h}{2}right)-frac{2}{t-h}right)right|operatorname{d}t+int_{2|h|le|t|le pi} left|f_x(t+h)left(cotleft(frac{t+h}{2}right)-frac{2}{t+h}right)right|operatorname{d}t+int_{2|h|le|t|le pi} left|f_x(t-h)frac{2}{t-h}-f_x(t+h)frac{2}{t+h}right|operatorname{d}t.$$
Now it is clear that the first two integrals aren't a problem, while for the last:
$$int_{2|h|le|t|le pi} left|f_x(t-h)frac{1}{t-h}-f_x(t+h)frac{1}{t+h}right|operatorname{d}t \ le int_{2|h|le|t|le pi} |t|left|frac{f_x(t-h)-f_x(t+h)}{(t-h)(t+h)}right|operatorname{d}t + |h|^2int_{2|h|le|t|le pi} left|frac{f_x(t-h)+f_x(t+h)}{(t-h)(t+h)}right|operatorname{d}t \ le |2h|^alpha int_{2|h|le|t|le pi} left|frac{t}{(t-h)(t+h)}right|operatorname{d}t + |h|int_{2|h|le|t|le pi} left|t-hright|^{alpha-1}operatorname{d}t+|h|int_{2|h|le|t|le pi} left|t+hright|^{alpha-1}operatorname{d}t \ le c_3|h|^{alpha}(1+log|h|).$$
Putting all together we get:
$$sup_{xin[-pi,pi]}left|sum_{n=-N}^Nhat{f}(n)e^{inx}-f(x)right|le Cleft|frac{pi}{N}right|^{alpha}log left|frac{pi}{N}right|to 0, Nto+infty$$
$endgroup$
add a comment |
Your Answer
StackExchange.ifUsing("editor", function () {
return StackExchange.using("mathjaxEditing", function () {
StackExchange.MarkdownEditor.creationCallbacks.add(function (editor, postfix) {
StackExchange.mathjaxEditing.prepareWmdForMathJax(editor, postfix, [["$", "$"], ["\\(","\\)"]]);
});
});
}, "mathjax-editing");
StackExchange.ready(function() {
var channelOptions = {
tags: "".split(" "),
id: "69"
};
initTagRenderer("".split(" "), "".split(" "), channelOptions);
StackExchange.using("externalEditor", function() {
// Have to fire editor after snippets, if snippets enabled
if (StackExchange.settings.snippets.snippetsEnabled) {
StackExchange.using("snippets", function() {
createEditor();
});
}
else {
createEditor();
}
});
function createEditor() {
StackExchange.prepareEditor({
heartbeatType: 'answer',
autoActivateHeartbeat: false,
convertImagesToLinks: true,
noModals: true,
showLowRepImageUploadWarning: true,
reputationToPostImages: 10,
bindNavPrevention: true,
postfix: "",
imageUploader: {
brandingHtml: "Powered by u003ca class="icon-imgur-white" href="https://imgur.com/"u003eu003c/au003e",
contentPolicyHtml: "User contributions licensed under u003ca href="https://creativecommons.org/licenses/by-sa/3.0/"u003ecc by-sa 3.0 with attribution requiredu003c/au003e u003ca href="https://stackoverflow.com/legal/content-policy"u003e(content policy)u003c/au003e",
allowUrls: true
},
noCode: true, onDemand: true,
discardSelector: ".discard-answer"
,immediatelyShowMarkdownHelp:true
});
}
});
Sign up or log in
StackExchange.ready(function () {
StackExchange.helpers.onClickDraftSave('#login-link');
});
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
StackExchange.ready(
function () {
StackExchange.openid.initPostLogin('.new-post-login', 'https%3a%2f%2fmath.stackexchange.com%2fquestions%2f1012365%2fhow-to-show-for-alpha-in-0-1-any-f-in-c-alpha0-1-sim-has-a-fouri%23new-answer', 'question_page');
}
);
Post as a guest
Required, but never shown
3 Answers
3
active
oldest
votes
3 Answers
3
active
oldest
votes
active
oldest
votes
active
oldest
votes
$begingroup$
Suppose that $|f(x)|le C$ and $|f(x)-f(y)|le C|x-y|^alpha$.
Express the Difference Using the Dirichlet Kernel
Using the Dirichlet Kernel, we get
$$
begin{align}
|S_nf(x)-f(x)|
&=left|,int_{-1/2}^{1/2}frac{sin((2n+1)pi y)}{sin(pi y)}[f(x-y)-f(x)],mathrm{d}y,right|\
&=left|,sum_{k=-n}^nint_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}frac{sin((2n+1)pi y)}{sin(pi y)}[f(x-y)-f(x)],mathrm{d}y,right|tag{1}
end{align}
$$
Estimate each Integral Using the smoothness of $boldsymbol{f}$
Since $left|,frac{sin((2n+1)pi y)}{sin(pi y)},right|lefrac{2n+1}{big|2|k|-1big|}$ and each interval is $frac1{2n+1}$ wide, we can bound
$$
begin{align}
left|,int_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}frac{sin((2n+1)pi y)}{sin(pi y)}[f(x-y)-f(x)],mathrm{d}y,right|
&lefrac{C}{big|2|k|-1big|}left(frac{2|k|+1}{4n+2}right)^alphatag{2}
end{align}
$$
Estimate each Integral Using Cancellation from $boldsymbol{sin((2n+1)pi x)}$
For $|y|lefrac12$, we have $|2y|le|sin(pi y)|le|pi y|$, and because
$$
int_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}sin((2n+1)pi y),mathrm{d}y=0tag{3}
$$
and
$$
int_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}|sin((2n+1)pi y)|,mathrm{d}y=frac2{(2n+1)pi}tag{4}
$$
if we let $m_k$ be the middle of the range of $frac{f(x-y)-f(x)}{sin(pi y)}$ on $left[frac{2k-1}{4n+2},frac{2k+1}{4n+2}right]$, for $kne0$, we can bound
$$
begin{align}
&left|,int_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}sin((2n+1)pi y)frac{f(x-y)-f(x)}{sin(pi y)},mathrm{d}y,right|\
&=left|,int_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}sin((2n+1)pi y)left[frac{f(x-y)-f(x)}{sin(pi y)}-m_kright],mathrm{d}y,right|\
&lefrac1{(2n+1)pi}frac{overbrace{pifrac{2|k|+1}{4n+2}}^{sin(pi y)}overbrace{C(2n+1)^{-alpha}vphantom{frac{|}2}}^{Delta (f(x-y)-f(x))}+overbrace{2Cvphantom{()^1}}^{f(x-y)-f(x)}overbrace{pi(2n+1)^{-1}}^{Deltasin(pi y)}}{underbrace{frac{4k^2-1}{(2n+1)^2}}_{sin^2(pi y)}}\
&=frac{C(2n+1)^{-alpha}}{4|k|-2}+frac{2C}{4k^2-1}tag{5}
end{align}
$$
Use each Estimate in its Proper Place
If we use estimate $(2)$ for $kle m=n^{frac{alpha}{alpha+1}}$ and estimate $(5)$ for $kgt m$, then we get
$$
begin{align}
sum_{|k|le m}frac{C}{big|2|k|-1big|}left(frac{2|k|+1}{4n+2}right)^alpha
&lefrac{C}{(4n+2)^alpha}left[1+6sum_{k=1}^m(2k+1)^{alpha-1}right]\
&lefrac{C}{(4n+2)^alpha}frac3alpha(2m+1)^alpha\
&simfrac{3C}{alpha2^alpha}n^{-fracalpha{alpha+1}}tag{6}
end{align}
$$
and
$$
begin{align}
sum_{mlt|k|le n}frac{C(2n+1)^{-alpha}}{4|k|-2}
&lefrac{C}{2^{alpha+1}}frac{H_n}{n^alpha}\
&simfrac{C}{2^{alpha+1}}frac{log(n)}{n^alpha}\
&=oleft(n^{-frac{alpha}{alpha+1}}right)tag{7}
end{align}
$$
and
$$
begin{align}
sum_{mlt|k|le n}frac{2C}{4k^2-1}
&le Csum_{k=m}^inftyfrac1{k^2-1}\
&=frac{C}{2}sum_{k=m}^inftyleft(frac1{k-1}-frac1{k+1}right)\
&=frac{C}{2}left(frac1{m-1}+frac1mright)\
&sim Cn^{-frac{alpha}{alpha+1}}tag{8}
end{align}
$$
Put Everything Together
Therefore, we have uniform convergence:
$$
|S_nf(x)-f(x)|leleft(1+frac3{alpha2^alpha}right)Cn^{-frac{alpha}{alpha+1}}tag{9}
$$
$endgroup$
$begingroup$
In Wikipedia, it is claimed that we should be able to get uniform convergence on the order of $O(log(n)n^{-alpha})$; that is, the order of the term in $(7)$ that is ignored as too small. I will have to look into this.
$endgroup$
– robjohn♦
Nov 11 '14 at 20:30
$begingroup$
Thanks, I will read through your answer later today (the formulae are cut off on my mobile phone)
$endgroup$
– Calvin Khor
Nov 12 '14 at 8:07
$begingroup$
This, of course, is a very good answer, but I think that it shows more than it is desired. The question only wants to prove equicontinuity of $S_nf$, which then yields uniform convergence (without the asymptotic (9), of course). I guess that the proof of such equicontinuity should be something along the same lines as this answer, only simpler.
$endgroup$
– Giuseppe Negro
Nov 12 '14 at 18:24
add a comment |
$begingroup$
Suppose that $|f(x)|le C$ and $|f(x)-f(y)|le C|x-y|^alpha$.
Express the Difference Using the Dirichlet Kernel
Using the Dirichlet Kernel, we get
$$
begin{align}
|S_nf(x)-f(x)|
&=left|,int_{-1/2}^{1/2}frac{sin((2n+1)pi y)}{sin(pi y)}[f(x-y)-f(x)],mathrm{d}y,right|\
&=left|,sum_{k=-n}^nint_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}frac{sin((2n+1)pi y)}{sin(pi y)}[f(x-y)-f(x)],mathrm{d}y,right|tag{1}
end{align}
$$
Estimate each Integral Using the smoothness of $boldsymbol{f}$
Since $left|,frac{sin((2n+1)pi y)}{sin(pi y)},right|lefrac{2n+1}{big|2|k|-1big|}$ and each interval is $frac1{2n+1}$ wide, we can bound
$$
begin{align}
left|,int_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}frac{sin((2n+1)pi y)}{sin(pi y)}[f(x-y)-f(x)],mathrm{d}y,right|
&lefrac{C}{big|2|k|-1big|}left(frac{2|k|+1}{4n+2}right)^alphatag{2}
end{align}
$$
Estimate each Integral Using Cancellation from $boldsymbol{sin((2n+1)pi x)}$
For $|y|lefrac12$, we have $|2y|le|sin(pi y)|le|pi y|$, and because
$$
int_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}sin((2n+1)pi y),mathrm{d}y=0tag{3}
$$
and
$$
int_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}|sin((2n+1)pi y)|,mathrm{d}y=frac2{(2n+1)pi}tag{4}
$$
if we let $m_k$ be the middle of the range of $frac{f(x-y)-f(x)}{sin(pi y)}$ on $left[frac{2k-1}{4n+2},frac{2k+1}{4n+2}right]$, for $kne0$, we can bound
$$
begin{align}
&left|,int_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}sin((2n+1)pi y)frac{f(x-y)-f(x)}{sin(pi y)},mathrm{d}y,right|\
&=left|,int_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}sin((2n+1)pi y)left[frac{f(x-y)-f(x)}{sin(pi y)}-m_kright],mathrm{d}y,right|\
&lefrac1{(2n+1)pi}frac{overbrace{pifrac{2|k|+1}{4n+2}}^{sin(pi y)}overbrace{C(2n+1)^{-alpha}vphantom{frac{|}2}}^{Delta (f(x-y)-f(x))}+overbrace{2Cvphantom{()^1}}^{f(x-y)-f(x)}overbrace{pi(2n+1)^{-1}}^{Deltasin(pi y)}}{underbrace{frac{4k^2-1}{(2n+1)^2}}_{sin^2(pi y)}}\
&=frac{C(2n+1)^{-alpha}}{4|k|-2}+frac{2C}{4k^2-1}tag{5}
end{align}
$$
Use each Estimate in its Proper Place
If we use estimate $(2)$ for $kle m=n^{frac{alpha}{alpha+1}}$ and estimate $(5)$ for $kgt m$, then we get
$$
begin{align}
sum_{|k|le m}frac{C}{big|2|k|-1big|}left(frac{2|k|+1}{4n+2}right)^alpha
&lefrac{C}{(4n+2)^alpha}left[1+6sum_{k=1}^m(2k+1)^{alpha-1}right]\
&lefrac{C}{(4n+2)^alpha}frac3alpha(2m+1)^alpha\
&simfrac{3C}{alpha2^alpha}n^{-fracalpha{alpha+1}}tag{6}
end{align}
$$
and
$$
begin{align}
sum_{mlt|k|le n}frac{C(2n+1)^{-alpha}}{4|k|-2}
&lefrac{C}{2^{alpha+1}}frac{H_n}{n^alpha}\
&simfrac{C}{2^{alpha+1}}frac{log(n)}{n^alpha}\
&=oleft(n^{-frac{alpha}{alpha+1}}right)tag{7}
end{align}
$$
and
$$
begin{align}
sum_{mlt|k|le n}frac{2C}{4k^2-1}
&le Csum_{k=m}^inftyfrac1{k^2-1}\
&=frac{C}{2}sum_{k=m}^inftyleft(frac1{k-1}-frac1{k+1}right)\
&=frac{C}{2}left(frac1{m-1}+frac1mright)\
&sim Cn^{-frac{alpha}{alpha+1}}tag{8}
end{align}
$$
Put Everything Together
Therefore, we have uniform convergence:
$$
|S_nf(x)-f(x)|leleft(1+frac3{alpha2^alpha}right)Cn^{-frac{alpha}{alpha+1}}tag{9}
$$
$endgroup$
$begingroup$
In Wikipedia, it is claimed that we should be able to get uniform convergence on the order of $O(log(n)n^{-alpha})$; that is, the order of the term in $(7)$ that is ignored as too small. I will have to look into this.
$endgroup$
– robjohn♦
Nov 11 '14 at 20:30
$begingroup$
Thanks, I will read through your answer later today (the formulae are cut off on my mobile phone)
$endgroup$
– Calvin Khor
Nov 12 '14 at 8:07
$begingroup$
This, of course, is a very good answer, but I think that it shows more than it is desired. The question only wants to prove equicontinuity of $S_nf$, which then yields uniform convergence (without the asymptotic (9), of course). I guess that the proof of such equicontinuity should be something along the same lines as this answer, only simpler.
$endgroup$
– Giuseppe Negro
Nov 12 '14 at 18:24
add a comment |
$begingroup$
Suppose that $|f(x)|le C$ and $|f(x)-f(y)|le C|x-y|^alpha$.
Express the Difference Using the Dirichlet Kernel
Using the Dirichlet Kernel, we get
$$
begin{align}
|S_nf(x)-f(x)|
&=left|,int_{-1/2}^{1/2}frac{sin((2n+1)pi y)}{sin(pi y)}[f(x-y)-f(x)],mathrm{d}y,right|\
&=left|,sum_{k=-n}^nint_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}frac{sin((2n+1)pi y)}{sin(pi y)}[f(x-y)-f(x)],mathrm{d}y,right|tag{1}
end{align}
$$
Estimate each Integral Using the smoothness of $boldsymbol{f}$
Since $left|,frac{sin((2n+1)pi y)}{sin(pi y)},right|lefrac{2n+1}{big|2|k|-1big|}$ and each interval is $frac1{2n+1}$ wide, we can bound
$$
begin{align}
left|,int_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}frac{sin((2n+1)pi y)}{sin(pi y)}[f(x-y)-f(x)],mathrm{d}y,right|
&lefrac{C}{big|2|k|-1big|}left(frac{2|k|+1}{4n+2}right)^alphatag{2}
end{align}
$$
Estimate each Integral Using Cancellation from $boldsymbol{sin((2n+1)pi x)}$
For $|y|lefrac12$, we have $|2y|le|sin(pi y)|le|pi y|$, and because
$$
int_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}sin((2n+1)pi y),mathrm{d}y=0tag{3}
$$
and
$$
int_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}|sin((2n+1)pi y)|,mathrm{d}y=frac2{(2n+1)pi}tag{4}
$$
if we let $m_k$ be the middle of the range of $frac{f(x-y)-f(x)}{sin(pi y)}$ on $left[frac{2k-1}{4n+2},frac{2k+1}{4n+2}right]$, for $kne0$, we can bound
$$
begin{align}
&left|,int_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}sin((2n+1)pi y)frac{f(x-y)-f(x)}{sin(pi y)},mathrm{d}y,right|\
&=left|,int_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}sin((2n+1)pi y)left[frac{f(x-y)-f(x)}{sin(pi y)}-m_kright],mathrm{d}y,right|\
&lefrac1{(2n+1)pi}frac{overbrace{pifrac{2|k|+1}{4n+2}}^{sin(pi y)}overbrace{C(2n+1)^{-alpha}vphantom{frac{|}2}}^{Delta (f(x-y)-f(x))}+overbrace{2Cvphantom{()^1}}^{f(x-y)-f(x)}overbrace{pi(2n+1)^{-1}}^{Deltasin(pi y)}}{underbrace{frac{4k^2-1}{(2n+1)^2}}_{sin^2(pi y)}}\
&=frac{C(2n+1)^{-alpha}}{4|k|-2}+frac{2C}{4k^2-1}tag{5}
end{align}
$$
Use each Estimate in its Proper Place
If we use estimate $(2)$ for $kle m=n^{frac{alpha}{alpha+1}}$ and estimate $(5)$ for $kgt m$, then we get
$$
begin{align}
sum_{|k|le m}frac{C}{big|2|k|-1big|}left(frac{2|k|+1}{4n+2}right)^alpha
&lefrac{C}{(4n+2)^alpha}left[1+6sum_{k=1}^m(2k+1)^{alpha-1}right]\
&lefrac{C}{(4n+2)^alpha}frac3alpha(2m+1)^alpha\
&simfrac{3C}{alpha2^alpha}n^{-fracalpha{alpha+1}}tag{6}
end{align}
$$
and
$$
begin{align}
sum_{mlt|k|le n}frac{C(2n+1)^{-alpha}}{4|k|-2}
&lefrac{C}{2^{alpha+1}}frac{H_n}{n^alpha}\
&simfrac{C}{2^{alpha+1}}frac{log(n)}{n^alpha}\
&=oleft(n^{-frac{alpha}{alpha+1}}right)tag{7}
end{align}
$$
and
$$
begin{align}
sum_{mlt|k|le n}frac{2C}{4k^2-1}
&le Csum_{k=m}^inftyfrac1{k^2-1}\
&=frac{C}{2}sum_{k=m}^inftyleft(frac1{k-1}-frac1{k+1}right)\
&=frac{C}{2}left(frac1{m-1}+frac1mright)\
&sim Cn^{-frac{alpha}{alpha+1}}tag{8}
end{align}
$$
Put Everything Together
Therefore, we have uniform convergence:
$$
|S_nf(x)-f(x)|leleft(1+frac3{alpha2^alpha}right)Cn^{-frac{alpha}{alpha+1}}tag{9}
$$
$endgroup$
Suppose that $|f(x)|le C$ and $|f(x)-f(y)|le C|x-y|^alpha$.
Express the Difference Using the Dirichlet Kernel
Using the Dirichlet Kernel, we get
$$
begin{align}
|S_nf(x)-f(x)|
&=left|,int_{-1/2}^{1/2}frac{sin((2n+1)pi y)}{sin(pi y)}[f(x-y)-f(x)],mathrm{d}y,right|\
&=left|,sum_{k=-n}^nint_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}frac{sin((2n+1)pi y)}{sin(pi y)}[f(x-y)-f(x)],mathrm{d}y,right|tag{1}
end{align}
$$
Estimate each Integral Using the smoothness of $boldsymbol{f}$
Since $left|,frac{sin((2n+1)pi y)}{sin(pi y)},right|lefrac{2n+1}{big|2|k|-1big|}$ and each interval is $frac1{2n+1}$ wide, we can bound
$$
begin{align}
left|,int_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}frac{sin((2n+1)pi y)}{sin(pi y)}[f(x-y)-f(x)],mathrm{d}y,right|
&lefrac{C}{big|2|k|-1big|}left(frac{2|k|+1}{4n+2}right)^alphatag{2}
end{align}
$$
Estimate each Integral Using Cancellation from $boldsymbol{sin((2n+1)pi x)}$
For $|y|lefrac12$, we have $|2y|le|sin(pi y)|le|pi y|$, and because
$$
int_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}sin((2n+1)pi y),mathrm{d}y=0tag{3}
$$
and
$$
int_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}|sin((2n+1)pi y)|,mathrm{d}y=frac2{(2n+1)pi}tag{4}
$$
if we let $m_k$ be the middle of the range of $frac{f(x-y)-f(x)}{sin(pi y)}$ on $left[frac{2k-1}{4n+2},frac{2k+1}{4n+2}right]$, for $kne0$, we can bound
$$
begin{align}
&left|,int_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}sin((2n+1)pi y)frac{f(x-y)-f(x)}{sin(pi y)},mathrm{d}y,right|\
&=left|,int_{frac{2k-1}{4n+2}}^{frac{2k+1}{4n+2}}sin((2n+1)pi y)left[frac{f(x-y)-f(x)}{sin(pi y)}-m_kright],mathrm{d}y,right|\
&lefrac1{(2n+1)pi}frac{overbrace{pifrac{2|k|+1}{4n+2}}^{sin(pi y)}overbrace{C(2n+1)^{-alpha}vphantom{frac{|}2}}^{Delta (f(x-y)-f(x))}+overbrace{2Cvphantom{()^1}}^{f(x-y)-f(x)}overbrace{pi(2n+1)^{-1}}^{Deltasin(pi y)}}{underbrace{frac{4k^2-1}{(2n+1)^2}}_{sin^2(pi y)}}\
&=frac{C(2n+1)^{-alpha}}{4|k|-2}+frac{2C}{4k^2-1}tag{5}
end{align}
$$
Use each Estimate in its Proper Place
If we use estimate $(2)$ for $kle m=n^{frac{alpha}{alpha+1}}$ and estimate $(5)$ for $kgt m$, then we get
$$
begin{align}
sum_{|k|le m}frac{C}{big|2|k|-1big|}left(frac{2|k|+1}{4n+2}right)^alpha
&lefrac{C}{(4n+2)^alpha}left[1+6sum_{k=1}^m(2k+1)^{alpha-1}right]\
&lefrac{C}{(4n+2)^alpha}frac3alpha(2m+1)^alpha\
&simfrac{3C}{alpha2^alpha}n^{-fracalpha{alpha+1}}tag{6}
end{align}
$$
and
$$
begin{align}
sum_{mlt|k|le n}frac{C(2n+1)^{-alpha}}{4|k|-2}
&lefrac{C}{2^{alpha+1}}frac{H_n}{n^alpha}\
&simfrac{C}{2^{alpha+1}}frac{log(n)}{n^alpha}\
&=oleft(n^{-frac{alpha}{alpha+1}}right)tag{7}
end{align}
$$
and
$$
begin{align}
sum_{mlt|k|le n}frac{2C}{4k^2-1}
&le Csum_{k=m}^inftyfrac1{k^2-1}\
&=frac{C}{2}sum_{k=m}^inftyleft(frac1{k-1}-frac1{k+1}right)\
&=frac{C}{2}left(frac1{m-1}+frac1mright)\
&sim Cn^{-frac{alpha}{alpha+1}}tag{8}
end{align}
$$
Put Everything Together
Therefore, we have uniform convergence:
$$
|S_nf(x)-f(x)|leleft(1+frac3{alpha2^alpha}right)Cn^{-frac{alpha}{alpha+1}}tag{9}
$$
edited Nov 12 '14 at 11:10
answered Nov 11 '14 at 10:15
robjohn♦robjohn
269k27309635
269k27309635
$begingroup$
In Wikipedia, it is claimed that we should be able to get uniform convergence on the order of $O(log(n)n^{-alpha})$; that is, the order of the term in $(7)$ that is ignored as too small. I will have to look into this.
$endgroup$
– robjohn♦
Nov 11 '14 at 20:30
$begingroup$
Thanks, I will read through your answer later today (the formulae are cut off on my mobile phone)
$endgroup$
– Calvin Khor
Nov 12 '14 at 8:07
$begingroup$
This, of course, is a very good answer, but I think that it shows more than it is desired. The question only wants to prove equicontinuity of $S_nf$, which then yields uniform convergence (without the asymptotic (9), of course). I guess that the proof of such equicontinuity should be something along the same lines as this answer, only simpler.
$endgroup$
– Giuseppe Negro
Nov 12 '14 at 18:24
add a comment |
$begingroup$
In Wikipedia, it is claimed that we should be able to get uniform convergence on the order of $O(log(n)n^{-alpha})$; that is, the order of the term in $(7)$ that is ignored as too small. I will have to look into this.
$endgroup$
– robjohn♦
Nov 11 '14 at 20:30
$begingroup$
Thanks, I will read through your answer later today (the formulae are cut off on my mobile phone)
$endgroup$
– Calvin Khor
Nov 12 '14 at 8:07
$begingroup$
This, of course, is a very good answer, but I think that it shows more than it is desired. The question only wants to prove equicontinuity of $S_nf$, which then yields uniform convergence (without the asymptotic (9), of course). I guess that the proof of such equicontinuity should be something along the same lines as this answer, only simpler.
$endgroup$
– Giuseppe Negro
Nov 12 '14 at 18:24
$begingroup$
In Wikipedia, it is claimed that we should be able to get uniform convergence on the order of $O(log(n)n^{-alpha})$; that is, the order of the term in $(7)$ that is ignored as too small. I will have to look into this.
$endgroup$
– robjohn♦
Nov 11 '14 at 20:30
$begingroup$
In Wikipedia, it is claimed that we should be able to get uniform convergence on the order of $O(log(n)n^{-alpha})$; that is, the order of the term in $(7)$ that is ignored as too small. I will have to look into this.
$endgroup$
– robjohn♦
Nov 11 '14 at 20:30
$begingroup$
Thanks, I will read through your answer later today (the formulae are cut off on my mobile phone)
$endgroup$
– Calvin Khor
Nov 12 '14 at 8:07
$begingroup$
Thanks, I will read through your answer later today (the formulae are cut off on my mobile phone)
$endgroup$
– Calvin Khor
Nov 12 '14 at 8:07
$begingroup$
This, of course, is a very good answer, but I think that it shows more than it is desired. The question only wants to prove equicontinuity of $S_nf$, which then yields uniform convergence (without the asymptotic (9), of course). I guess that the proof of such equicontinuity should be something along the same lines as this answer, only simpler.
$endgroup$
– Giuseppe Negro
Nov 12 '14 at 18:24
$begingroup$
This, of course, is a very good answer, but I think that it shows more than it is desired. The question only wants to prove equicontinuity of $S_nf$, which then yields uniform convergence (without the asymptotic (9), of course). I guess that the proof of such equicontinuity should be something along the same lines as this answer, only simpler.
$endgroup$
– Giuseppe Negro
Nov 12 '14 at 18:24
add a comment |
$begingroup$
While I accepted the answer above, this is how my lecturer (and later my friend) explained it to me (the exam is tomorrow). We first define $$g_n(x):=f(x) - S_n f(x)$$ just to remind ourselves that we need to be careful of cancellations. Then uniform convergence of $S_nf$ to $f$ is equivalent to showing $g_n→ 0$ uniformly; since we know (part 2) that $g_n(x) → 0$ pointwise, it suffices to show $g_n$ is uniformly continuous (by part 1).
Since $newcommand{d}{text{d}}newcommand{intT}{∫_{-1/2}^{1/2}}g_n(x) = f(x)times 1 - intT f(z-x) D_n(z) d z = intT [f(x) -f(z-x)] D_n(z) d z$ ,
begin{align}
|g_n(x) - g_n(y)| ≤ intT |D_n(z)|underbrace{|f(x) - f(z-x) - f(y) + f(z-y)|}_{(star)} d z
end{align}
We now need to find bounds independent of $n$ . We use a simple bound for the Dirichlet kernel $D_n$: as there is $C_0$ such that $|sin(2π z)|>C_0|z|$ on $[-1/2,1/2]$,
$$|D_n(z)| < frac{C_1}{|z|} $$
Since we don't gain too much form a simple bound we need to bound $(star)$. The trick is to use two different bounds, each good on different sets:
begin{align} |color{red}{f(x) - f(z-x)} - color{blue}{f(y) + f(z-y)}| &leq C_3|z|^alpha \
|color{red}{f(x)} - color{blue}{f(z-x)} - color{red}{f(y)} + color{blue}{f(z-y)}|
&leq C_3|x-y|^alpha \
end{align}
Thus $$|g_n(x) - g_n(y)| leq ∫_{|z|leq|x-y|} C_4|z|^{alpha-1} d z + |x-y|^alpha ∫_{|x-y|<|z|<1/2}frac{C_5}{z} d z = I_1 + I_2 $$
Now $I_1$ is $mathcal{O}(|x-y|)$ because $|z|^{alpha-1}$ is $L^1([-1/2,1/2])$. The second we compute,
$$I_2 = C_5 |x-y|^alphaleft(logfrac{1}{2} + logfrac{1}{|x-y|}right) $$
And we win because polynomials beat logarithms.
$endgroup$
$begingroup$
Nice answer. Do you still remember how to show the pointwise convergence?
$endgroup$
– PhoemueX
Jan 26 at 10:25
$begingroup$
@PhoemueX Thanks. I believe the "Dirichlet–Dini Criterion" is sufficient, see math.stackexchange.com/questions/1109758/… . Does that make sense? I recall proving it directly, probably by using the $2pi$ periodicity to introduce a $f(x)-f(x-y)$ integrand somewhere in $S_nf $ but my quick attempts weren't enough to bring this memory back.
$endgroup$
– Calvin Khor
Jan 26 at 13:54
add a comment |
$begingroup$
While I accepted the answer above, this is how my lecturer (and later my friend) explained it to me (the exam is tomorrow). We first define $$g_n(x):=f(x) - S_n f(x)$$ just to remind ourselves that we need to be careful of cancellations. Then uniform convergence of $S_nf$ to $f$ is equivalent to showing $g_n→ 0$ uniformly; since we know (part 2) that $g_n(x) → 0$ pointwise, it suffices to show $g_n$ is uniformly continuous (by part 1).
Since $newcommand{d}{text{d}}newcommand{intT}{∫_{-1/2}^{1/2}}g_n(x) = f(x)times 1 - intT f(z-x) D_n(z) d z = intT [f(x) -f(z-x)] D_n(z) d z$ ,
begin{align}
|g_n(x) - g_n(y)| ≤ intT |D_n(z)|underbrace{|f(x) - f(z-x) - f(y) + f(z-y)|}_{(star)} d z
end{align}
We now need to find bounds independent of $n$ . We use a simple bound for the Dirichlet kernel $D_n$: as there is $C_0$ such that $|sin(2π z)|>C_0|z|$ on $[-1/2,1/2]$,
$$|D_n(z)| < frac{C_1}{|z|} $$
Since we don't gain too much form a simple bound we need to bound $(star)$. The trick is to use two different bounds, each good on different sets:
begin{align} |color{red}{f(x) - f(z-x)} - color{blue}{f(y) + f(z-y)}| &leq C_3|z|^alpha \
|color{red}{f(x)} - color{blue}{f(z-x)} - color{red}{f(y)} + color{blue}{f(z-y)}|
&leq C_3|x-y|^alpha \
end{align}
Thus $$|g_n(x) - g_n(y)| leq ∫_{|z|leq|x-y|} C_4|z|^{alpha-1} d z + |x-y|^alpha ∫_{|x-y|<|z|<1/2}frac{C_5}{z} d z = I_1 + I_2 $$
Now $I_1$ is $mathcal{O}(|x-y|)$ because $|z|^{alpha-1}$ is $L^1([-1/2,1/2])$. The second we compute,
$$I_2 = C_5 |x-y|^alphaleft(logfrac{1}{2} + logfrac{1}{|x-y|}right) $$
And we win because polynomials beat logarithms.
$endgroup$
$begingroup$
Nice answer. Do you still remember how to show the pointwise convergence?
$endgroup$
– PhoemueX
Jan 26 at 10:25
$begingroup$
@PhoemueX Thanks. I believe the "Dirichlet–Dini Criterion" is sufficient, see math.stackexchange.com/questions/1109758/… . Does that make sense? I recall proving it directly, probably by using the $2pi$ periodicity to introduce a $f(x)-f(x-y)$ integrand somewhere in $S_nf $ but my quick attempts weren't enough to bring this memory back.
$endgroup$
– Calvin Khor
Jan 26 at 13:54
add a comment |
$begingroup$
While I accepted the answer above, this is how my lecturer (and later my friend) explained it to me (the exam is tomorrow). We first define $$g_n(x):=f(x) - S_n f(x)$$ just to remind ourselves that we need to be careful of cancellations. Then uniform convergence of $S_nf$ to $f$ is equivalent to showing $g_n→ 0$ uniformly; since we know (part 2) that $g_n(x) → 0$ pointwise, it suffices to show $g_n$ is uniformly continuous (by part 1).
Since $newcommand{d}{text{d}}newcommand{intT}{∫_{-1/2}^{1/2}}g_n(x) = f(x)times 1 - intT f(z-x) D_n(z) d z = intT [f(x) -f(z-x)] D_n(z) d z$ ,
begin{align}
|g_n(x) - g_n(y)| ≤ intT |D_n(z)|underbrace{|f(x) - f(z-x) - f(y) + f(z-y)|}_{(star)} d z
end{align}
We now need to find bounds independent of $n$ . We use a simple bound for the Dirichlet kernel $D_n$: as there is $C_0$ such that $|sin(2π z)|>C_0|z|$ on $[-1/2,1/2]$,
$$|D_n(z)| < frac{C_1}{|z|} $$
Since we don't gain too much form a simple bound we need to bound $(star)$. The trick is to use two different bounds, each good on different sets:
begin{align} |color{red}{f(x) - f(z-x)} - color{blue}{f(y) + f(z-y)}| &leq C_3|z|^alpha \
|color{red}{f(x)} - color{blue}{f(z-x)} - color{red}{f(y)} + color{blue}{f(z-y)}|
&leq C_3|x-y|^alpha \
end{align}
Thus $$|g_n(x) - g_n(y)| leq ∫_{|z|leq|x-y|} C_4|z|^{alpha-1} d z + |x-y|^alpha ∫_{|x-y|<|z|<1/2}frac{C_5}{z} d z = I_1 + I_2 $$
Now $I_1$ is $mathcal{O}(|x-y|)$ because $|z|^{alpha-1}$ is $L^1([-1/2,1/2])$. The second we compute,
$$I_2 = C_5 |x-y|^alphaleft(logfrac{1}{2} + logfrac{1}{|x-y|}right) $$
And we win because polynomials beat logarithms.
$endgroup$
While I accepted the answer above, this is how my lecturer (and later my friend) explained it to me (the exam is tomorrow). We first define $$g_n(x):=f(x) - S_n f(x)$$ just to remind ourselves that we need to be careful of cancellations. Then uniform convergence of $S_nf$ to $f$ is equivalent to showing $g_n→ 0$ uniformly; since we know (part 2) that $g_n(x) → 0$ pointwise, it suffices to show $g_n$ is uniformly continuous (by part 1).
Since $newcommand{d}{text{d}}newcommand{intT}{∫_{-1/2}^{1/2}}g_n(x) = f(x)times 1 - intT f(z-x) D_n(z) d z = intT [f(x) -f(z-x)] D_n(z) d z$ ,
begin{align}
|g_n(x) - g_n(y)| ≤ intT |D_n(z)|underbrace{|f(x) - f(z-x) - f(y) + f(z-y)|}_{(star)} d z
end{align}
We now need to find bounds independent of $n$ . We use a simple bound for the Dirichlet kernel $D_n$: as there is $C_0$ such that $|sin(2π z)|>C_0|z|$ on $[-1/2,1/2]$,
$$|D_n(z)| < frac{C_1}{|z|} $$
Since we don't gain too much form a simple bound we need to bound $(star)$. The trick is to use two different bounds, each good on different sets:
begin{align} |color{red}{f(x) - f(z-x)} - color{blue}{f(y) + f(z-y)}| &leq C_3|z|^alpha \
|color{red}{f(x)} - color{blue}{f(z-x)} - color{red}{f(y)} + color{blue}{f(z-y)}|
&leq C_3|x-y|^alpha \
end{align}
Thus $$|g_n(x) - g_n(y)| leq ∫_{|z|leq|x-y|} C_4|z|^{alpha-1} d z + |x-y|^alpha ∫_{|x-y|<|z|<1/2}frac{C_5}{z} d z = I_1 + I_2 $$
Now $I_1$ is $mathcal{O}(|x-y|)$ because $|z|^{alpha-1}$ is $L^1([-1/2,1/2])$. The second we compute,
$$I_2 = C_5 |x-y|^alphaleft(logfrac{1}{2} + logfrac{1}{|x-y|}right) $$
And we win because polynomials beat logarithms.
answered Apr 20 '15 at 0:26
Calvin KhorCalvin Khor
12.3k21438
12.3k21438
$begingroup$
Nice answer. Do you still remember how to show the pointwise convergence?
$endgroup$
– PhoemueX
Jan 26 at 10:25
$begingroup$
@PhoemueX Thanks. I believe the "Dirichlet–Dini Criterion" is sufficient, see math.stackexchange.com/questions/1109758/… . Does that make sense? I recall proving it directly, probably by using the $2pi$ periodicity to introduce a $f(x)-f(x-y)$ integrand somewhere in $S_nf $ but my quick attempts weren't enough to bring this memory back.
$endgroup$
– Calvin Khor
Jan 26 at 13:54
add a comment |
$begingroup$
Nice answer. Do you still remember how to show the pointwise convergence?
$endgroup$
– PhoemueX
Jan 26 at 10:25
$begingroup$
@PhoemueX Thanks. I believe the "Dirichlet–Dini Criterion" is sufficient, see math.stackexchange.com/questions/1109758/… . Does that make sense? I recall proving it directly, probably by using the $2pi$ periodicity to introduce a $f(x)-f(x-y)$ integrand somewhere in $S_nf $ but my quick attempts weren't enough to bring this memory back.
$endgroup$
– Calvin Khor
Jan 26 at 13:54
$begingroup$
Nice answer. Do you still remember how to show the pointwise convergence?
$endgroup$
– PhoemueX
Jan 26 at 10:25
$begingroup$
Nice answer. Do you still remember how to show the pointwise convergence?
$endgroup$
– PhoemueX
Jan 26 at 10:25
$begingroup$
@PhoemueX Thanks. I believe the "Dirichlet–Dini Criterion" is sufficient, see math.stackexchange.com/questions/1109758/… . Does that make sense? I recall proving it directly, probably by using the $2pi$ periodicity to introduce a $f(x)-f(x-y)$ integrand somewhere in $S_nf $ but my quick attempts weren't enough to bring this memory back.
$endgroup$
– Calvin Khor
Jan 26 at 13:54
$begingroup$
@PhoemueX Thanks. I believe the "Dirichlet–Dini Criterion" is sufficient, see math.stackexchange.com/questions/1109758/… . Does that make sense? I recall proving it directly, probably by using the $2pi$ periodicity to introduce a $f(x)-f(x-y)$ integrand somewhere in $S_nf $ but my quick attempts weren't enough to bring this memory back.
$endgroup$
– Calvin Khor
Jan 26 at 13:54
add a comment |
$begingroup$
Actually I just used estimates in the same spirit of the ones used in the answer of Calvin Khor, proving directly the result without appelling to Ascoli-Arzelà theorem and obtaining also an estimate of the rate of convergence, so I think that it makes sense to post this answer.
First, get $f:mathbb{R}to mathbb{C}$ be a $2pi$-periodic $alpha$-Hölder continuous function and for each $xinmathbb{R}$ define $f_x(t):=f(x+t)-f(x)$.
We want to prove that:
$$sup_{xin[-pi,pi]}left|sum_{n=-N}^Nhat{f}(n)e^{inx}-f(x)right|to0, Nto+infty.$$
Now:
$$sum_{n=-N}^Nhat{f}(n)e^{inx}-f(x)=int_{-pi}^pi (f(x+t)-f(x))frac{sinleft((N+frac{1}{2})tright)}{sinleft(frac{t}{2}right)}frac{operatorname{d}t}{2pi}=int_{-pi}^pi f_x(t)frac{sinleft((N+frac{1}{2})tright)}{sinleft(frac{t}{2}right)}frac{operatorname{d}t}{2pi} \ = int_{-pi}^pi f_x(t)cotleft(frac{t}{2}right)sin(Nt)frac{operatorname{d}t}{2pi}+int_{-pi}^pi f_x(t)cosleft(Ntright)frac{operatorname{d}t}{2pi}= \ frac{1}{pi}int_{-pi}^pi frac{f_x(t)}{t}sin(Nt)operatorname{d}t+int_{-pi}^pi f_x(t)left(cotleft(frac{t}{2}right)-frac{2}{t}right)sin(Nt)frac{operatorname{d}t}{2pi}+int_{-pi}^pi f_x(t)cosleft(Ntright)frac{operatorname{d}t}{2pi}.$$
Now, the only problematic integral is the first, so let's estimate only this one.
We have that:
$$int_{-pi}^pi f_x(t)cotleft(frac{t}{2}right)sin(Nt)frac{operatorname{d}t}{2pi} = -int_{-pi}^pi f_x(t+frac{pi}{N})cotleft(frac{t+frac{pi}{N}}{2}right)sin(Nt)frac{operatorname{d}t}{2pi},$$
so:
$$left|int_{-pi}^pi f_x(t)cotleft(frac{t}{2}right)sin(Nt)frac{operatorname{d}t}{2pi}right| = left|frac{1}{2} int_{-pi}^pi left(f_x(t)cotleft(frac{t}{2}right)-f_x(t+frac{pi}{N})cotleft(frac{t+frac{pi}{N}}{2}right)right)sin(Nt)frac{operatorname{d}t}{2pi}right| \ le frac{1}{4pi} int_{-pi}^pi left|f_x(t)cotleft(frac{t}{2}right)-f_x(t+frac{pi}{N})cotleft(frac{t+frac{pi}{N}}{2}right)right|operatorname{d}t \ = frac{1}{4pi} int_{-pi}^pi left|f_x(t-frac{pi}{2N})cotleft(frac{t-frac{pi}{2N}}{2}right)-f_x(t+frac{pi}{2N})cotleft(frac{t+frac{pi}{2N}}{2}right)right|operatorname{d}t$$
So we need an uniform estimate in $x$ for the quantity:
$$int_{-pi}^pi left|f_x(t-h)cotleft(frac{t-h}{2}right)-f_x(t+h)cotleft(frac{t+h}{2}right)right|operatorname{d}t$$
for $hto 0$.
Split the integral for $|t|<2|h|$ and for $2|h|le|t|le pi$ for $|h|<1$. So:
$$int_{|t|<2|h|} left|f_x(t-h)cotleft(frac{t-h}{2}right)-f_x(t+h)cotleft(frac{t+h}{2}right)right|operatorname{d}t\ le int_{|t|<2|h|} left|f_x(t-h)cotleft(frac{t-h}{2}right)right|operatorname{d}t+int_{|t|<2|h|} left|f_x(t+h)cotleft(frac{t+h}{2}right)right|operatorname{d}t le \ int_{|t|<2|h|} |t+h|^alphaleft|cotleft(frac{t+h}{2}right)right|operatorname{d}t+int_{|t|<2|h|} |t-h|^alphaleft|cotleft(frac{t-h}{2}right)right|operatorname{d}t\ le c_1 int_{|t|<4|h|} |t|^{alpha-1}operatorname{d}t = c_2 |h|^alpha.$$
While, for $2|h|le|t|le pi$ we have:
$$int_{2|h|le|t|le pi} left|f_x(t-h)cotleft(frac{t-h}{2}right)-f_x(t+h)cotleft(frac{t+h}{2}right)right|operatorname{d}t \ le int_{2|h|le|t|le pi} left|f_x(t-h)left(cotleft(frac{t-h}{2}right)-frac{2}{t-h}right)right|operatorname{d}t+int_{2|h|le|t|le pi} left|f_x(t+h)left(cotleft(frac{t+h}{2}right)-frac{2}{t+h}right)right|operatorname{d}t+int_{2|h|le|t|le pi} left|f_x(t-h)frac{2}{t-h}-f_x(t+h)frac{2}{t+h}right|operatorname{d}t.$$
Now it is clear that the first two integrals aren't a problem, while for the last:
$$int_{2|h|le|t|le pi} left|f_x(t-h)frac{1}{t-h}-f_x(t+h)frac{1}{t+h}right|operatorname{d}t \ le int_{2|h|le|t|le pi} |t|left|frac{f_x(t-h)-f_x(t+h)}{(t-h)(t+h)}right|operatorname{d}t + |h|^2int_{2|h|le|t|le pi} left|frac{f_x(t-h)+f_x(t+h)}{(t-h)(t+h)}right|operatorname{d}t \ le |2h|^alpha int_{2|h|le|t|le pi} left|frac{t}{(t-h)(t+h)}right|operatorname{d}t + |h|int_{2|h|le|t|le pi} left|t-hright|^{alpha-1}operatorname{d}t+|h|int_{2|h|le|t|le pi} left|t+hright|^{alpha-1}operatorname{d}t \ le c_3|h|^{alpha}(1+log|h|).$$
Putting all together we get:
$$sup_{xin[-pi,pi]}left|sum_{n=-N}^Nhat{f}(n)e^{inx}-f(x)right|le Cleft|frac{pi}{N}right|^{alpha}log left|frac{pi}{N}right|to 0, Nto+infty$$
$endgroup$
add a comment |
$begingroup$
Actually I just used estimates in the same spirit of the ones used in the answer of Calvin Khor, proving directly the result without appelling to Ascoli-Arzelà theorem and obtaining also an estimate of the rate of convergence, so I think that it makes sense to post this answer.
First, get $f:mathbb{R}to mathbb{C}$ be a $2pi$-periodic $alpha$-Hölder continuous function and for each $xinmathbb{R}$ define $f_x(t):=f(x+t)-f(x)$.
We want to prove that:
$$sup_{xin[-pi,pi]}left|sum_{n=-N}^Nhat{f}(n)e^{inx}-f(x)right|to0, Nto+infty.$$
Now:
$$sum_{n=-N}^Nhat{f}(n)e^{inx}-f(x)=int_{-pi}^pi (f(x+t)-f(x))frac{sinleft((N+frac{1}{2})tright)}{sinleft(frac{t}{2}right)}frac{operatorname{d}t}{2pi}=int_{-pi}^pi f_x(t)frac{sinleft((N+frac{1}{2})tright)}{sinleft(frac{t}{2}right)}frac{operatorname{d}t}{2pi} \ = int_{-pi}^pi f_x(t)cotleft(frac{t}{2}right)sin(Nt)frac{operatorname{d}t}{2pi}+int_{-pi}^pi f_x(t)cosleft(Ntright)frac{operatorname{d}t}{2pi}= \ frac{1}{pi}int_{-pi}^pi frac{f_x(t)}{t}sin(Nt)operatorname{d}t+int_{-pi}^pi f_x(t)left(cotleft(frac{t}{2}right)-frac{2}{t}right)sin(Nt)frac{operatorname{d}t}{2pi}+int_{-pi}^pi f_x(t)cosleft(Ntright)frac{operatorname{d}t}{2pi}.$$
Now, the only problematic integral is the first, so let's estimate only this one.
We have that:
$$int_{-pi}^pi f_x(t)cotleft(frac{t}{2}right)sin(Nt)frac{operatorname{d}t}{2pi} = -int_{-pi}^pi f_x(t+frac{pi}{N})cotleft(frac{t+frac{pi}{N}}{2}right)sin(Nt)frac{operatorname{d}t}{2pi},$$
so:
$$left|int_{-pi}^pi f_x(t)cotleft(frac{t}{2}right)sin(Nt)frac{operatorname{d}t}{2pi}right| = left|frac{1}{2} int_{-pi}^pi left(f_x(t)cotleft(frac{t}{2}right)-f_x(t+frac{pi}{N})cotleft(frac{t+frac{pi}{N}}{2}right)right)sin(Nt)frac{operatorname{d}t}{2pi}right| \ le frac{1}{4pi} int_{-pi}^pi left|f_x(t)cotleft(frac{t}{2}right)-f_x(t+frac{pi}{N})cotleft(frac{t+frac{pi}{N}}{2}right)right|operatorname{d}t \ = frac{1}{4pi} int_{-pi}^pi left|f_x(t-frac{pi}{2N})cotleft(frac{t-frac{pi}{2N}}{2}right)-f_x(t+frac{pi}{2N})cotleft(frac{t+frac{pi}{2N}}{2}right)right|operatorname{d}t$$
So we need an uniform estimate in $x$ for the quantity:
$$int_{-pi}^pi left|f_x(t-h)cotleft(frac{t-h}{2}right)-f_x(t+h)cotleft(frac{t+h}{2}right)right|operatorname{d}t$$
for $hto 0$.
Split the integral for $|t|<2|h|$ and for $2|h|le|t|le pi$ for $|h|<1$. So:
$$int_{|t|<2|h|} left|f_x(t-h)cotleft(frac{t-h}{2}right)-f_x(t+h)cotleft(frac{t+h}{2}right)right|operatorname{d}t\ le int_{|t|<2|h|} left|f_x(t-h)cotleft(frac{t-h}{2}right)right|operatorname{d}t+int_{|t|<2|h|} left|f_x(t+h)cotleft(frac{t+h}{2}right)right|operatorname{d}t le \ int_{|t|<2|h|} |t+h|^alphaleft|cotleft(frac{t+h}{2}right)right|operatorname{d}t+int_{|t|<2|h|} |t-h|^alphaleft|cotleft(frac{t-h}{2}right)right|operatorname{d}t\ le c_1 int_{|t|<4|h|} |t|^{alpha-1}operatorname{d}t = c_2 |h|^alpha.$$
While, for $2|h|le|t|le pi$ we have:
$$int_{2|h|le|t|le pi} left|f_x(t-h)cotleft(frac{t-h}{2}right)-f_x(t+h)cotleft(frac{t+h}{2}right)right|operatorname{d}t \ le int_{2|h|le|t|le pi} left|f_x(t-h)left(cotleft(frac{t-h}{2}right)-frac{2}{t-h}right)right|operatorname{d}t+int_{2|h|le|t|le pi} left|f_x(t+h)left(cotleft(frac{t+h}{2}right)-frac{2}{t+h}right)right|operatorname{d}t+int_{2|h|le|t|le pi} left|f_x(t-h)frac{2}{t-h}-f_x(t+h)frac{2}{t+h}right|operatorname{d}t.$$
Now it is clear that the first two integrals aren't a problem, while for the last:
$$int_{2|h|le|t|le pi} left|f_x(t-h)frac{1}{t-h}-f_x(t+h)frac{1}{t+h}right|operatorname{d}t \ le int_{2|h|le|t|le pi} |t|left|frac{f_x(t-h)-f_x(t+h)}{(t-h)(t+h)}right|operatorname{d}t + |h|^2int_{2|h|le|t|le pi} left|frac{f_x(t-h)+f_x(t+h)}{(t-h)(t+h)}right|operatorname{d}t \ le |2h|^alpha int_{2|h|le|t|le pi} left|frac{t}{(t-h)(t+h)}right|operatorname{d}t + |h|int_{2|h|le|t|le pi} left|t-hright|^{alpha-1}operatorname{d}t+|h|int_{2|h|le|t|le pi} left|t+hright|^{alpha-1}operatorname{d}t \ le c_3|h|^{alpha}(1+log|h|).$$
Putting all together we get:
$$sup_{xin[-pi,pi]}left|sum_{n=-N}^Nhat{f}(n)e^{inx}-f(x)right|le Cleft|frac{pi}{N}right|^{alpha}log left|frac{pi}{N}right|to 0, Nto+infty$$
$endgroup$
add a comment |
$begingroup$
Actually I just used estimates in the same spirit of the ones used in the answer of Calvin Khor, proving directly the result without appelling to Ascoli-Arzelà theorem and obtaining also an estimate of the rate of convergence, so I think that it makes sense to post this answer.
First, get $f:mathbb{R}to mathbb{C}$ be a $2pi$-periodic $alpha$-Hölder continuous function and for each $xinmathbb{R}$ define $f_x(t):=f(x+t)-f(x)$.
We want to prove that:
$$sup_{xin[-pi,pi]}left|sum_{n=-N}^Nhat{f}(n)e^{inx}-f(x)right|to0, Nto+infty.$$
Now:
$$sum_{n=-N}^Nhat{f}(n)e^{inx}-f(x)=int_{-pi}^pi (f(x+t)-f(x))frac{sinleft((N+frac{1}{2})tright)}{sinleft(frac{t}{2}right)}frac{operatorname{d}t}{2pi}=int_{-pi}^pi f_x(t)frac{sinleft((N+frac{1}{2})tright)}{sinleft(frac{t}{2}right)}frac{operatorname{d}t}{2pi} \ = int_{-pi}^pi f_x(t)cotleft(frac{t}{2}right)sin(Nt)frac{operatorname{d}t}{2pi}+int_{-pi}^pi f_x(t)cosleft(Ntright)frac{operatorname{d}t}{2pi}= \ frac{1}{pi}int_{-pi}^pi frac{f_x(t)}{t}sin(Nt)operatorname{d}t+int_{-pi}^pi f_x(t)left(cotleft(frac{t}{2}right)-frac{2}{t}right)sin(Nt)frac{operatorname{d}t}{2pi}+int_{-pi}^pi f_x(t)cosleft(Ntright)frac{operatorname{d}t}{2pi}.$$
Now, the only problematic integral is the first, so let's estimate only this one.
We have that:
$$int_{-pi}^pi f_x(t)cotleft(frac{t}{2}right)sin(Nt)frac{operatorname{d}t}{2pi} = -int_{-pi}^pi f_x(t+frac{pi}{N})cotleft(frac{t+frac{pi}{N}}{2}right)sin(Nt)frac{operatorname{d}t}{2pi},$$
so:
$$left|int_{-pi}^pi f_x(t)cotleft(frac{t}{2}right)sin(Nt)frac{operatorname{d}t}{2pi}right| = left|frac{1}{2} int_{-pi}^pi left(f_x(t)cotleft(frac{t}{2}right)-f_x(t+frac{pi}{N})cotleft(frac{t+frac{pi}{N}}{2}right)right)sin(Nt)frac{operatorname{d}t}{2pi}right| \ le frac{1}{4pi} int_{-pi}^pi left|f_x(t)cotleft(frac{t}{2}right)-f_x(t+frac{pi}{N})cotleft(frac{t+frac{pi}{N}}{2}right)right|operatorname{d}t \ = frac{1}{4pi} int_{-pi}^pi left|f_x(t-frac{pi}{2N})cotleft(frac{t-frac{pi}{2N}}{2}right)-f_x(t+frac{pi}{2N})cotleft(frac{t+frac{pi}{2N}}{2}right)right|operatorname{d}t$$
So we need an uniform estimate in $x$ for the quantity:
$$int_{-pi}^pi left|f_x(t-h)cotleft(frac{t-h}{2}right)-f_x(t+h)cotleft(frac{t+h}{2}right)right|operatorname{d}t$$
for $hto 0$.
Split the integral for $|t|<2|h|$ and for $2|h|le|t|le pi$ for $|h|<1$. So:
$$int_{|t|<2|h|} left|f_x(t-h)cotleft(frac{t-h}{2}right)-f_x(t+h)cotleft(frac{t+h}{2}right)right|operatorname{d}t\ le int_{|t|<2|h|} left|f_x(t-h)cotleft(frac{t-h}{2}right)right|operatorname{d}t+int_{|t|<2|h|} left|f_x(t+h)cotleft(frac{t+h}{2}right)right|operatorname{d}t le \ int_{|t|<2|h|} |t+h|^alphaleft|cotleft(frac{t+h}{2}right)right|operatorname{d}t+int_{|t|<2|h|} |t-h|^alphaleft|cotleft(frac{t-h}{2}right)right|operatorname{d}t\ le c_1 int_{|t|<4|h|} |t|^{alpha-1}operatorname{d}t = c_2 |h|^alpha.$$
While, for $2|h|le|t|le pi$ we have:
$$int_{2|h|le|t|le pi} left|f_x(t-h)cotleft(frac{t-h}{2}right)-f_x(t+h)cotleft(frac{t+h}{2}right)right|operatorname{d}t \ le int_{2|h|le|t|le pi} left|f_x(t-h)left(cotleft(frac{t-h}{2}right)-frac{2}{t-h}right)right|operatorname{d}t+int_{2|h|le|t|le pi} left|f_x(t+h)left(cotleft(frac{t+h}{2}right)-frac{2}{t+h}right)right|operatorname{d}t+int_{2|h|le|t|le pi} left|f_x(t-h)frac{2}{t-h}-f_x(t+h)frac{2}{t+h}right|operatorname{d}t.$$
Now it is clear that the first two integrals aren't a problem, while for the last:
$$int_{2|h|le|t|le pi} left|f_x(t-h)frac{1}{t-h}-f_x(t+h)frac{1}{t+h}right|operatorname{d}t \ le int_{2|h|le|t|le pi} |t|left|frac{f_x(t-h)-f_x(t+h)}{(t-h)(t+h)}right|operatorname{d}t + |h|^2int_{2|h|le|t|le pi} left|frac{f_x(t-h)+f_x(t+h)}{(t-h)(t+h)}right|operatorname{d}t \ le |2h|^alpha int_{2|h|le|t|le pi} left|frac{t}{(t-h)(t+h)}right|operatorname{d}t + |h|int_{2|h|le|t|le pi} left|t-hright|^{alpha-1}operatorname{d}t+|h|int_{2|h|le|t|le pi} left|t+hright|^{alpha-1}operatorname{d}t \ le c_3|h|^{alpha}(1+log|h|).$$
Putting all together we get:
$$sup_{xin[-pi,pi]}left|sum_{n=-N}^Nhat{f}(n)e^{inx}-f(x)right|le Cleft|frac{pi}{N}right|^{alpha}log left|frac{pi}{N}right|to 0, Nto+infty$$
$endgroup$
Actually I just used estimates in the same spirit of the ones used in the answer of Calvin Khor, proving directly the result without appelling to Ascoli-Arzelà theorem and obtaining also an estimate of the rate of convergence, so I think that it makes sense to post this answer.
First, get $f:mathbb{R}to mathbb{C}$ be a $2pi$-periodic $alpha$-Hölder continuous function and for each $xinmathbb{R}$ define $f_x(t):=f(x+t)-f(x)$.
We want to prove that:
$$sup_{xin[-pi,pi]}left|sum_{n=-N}^Nhat{f}(n)e^{inx}-f(x)right|to0, Nto+infty.$$
Now:
$$sum_{n=-N}^Nhat{f}(n)e^{inx}-f(x)=int_{-pi}^pi (f(x+t)-f(x))frac{sinleft((N+frac{1}{2})tright)}{sinleft(frac{t}{2}right)}frac{operatorname{d}t}{2pi}=int_{-pi}^pi f_x(t)frac{sinleft((N+frac{1}{2})tright)}{sinleft(frac{t}{2}right)}frac{operatorname{d}t}{2pi} \ = int_{-pi}^pi f_x(t)cotleft(frac{t}{2}right)sin(Nt)frac{operatorname{d}t}{2pi}+int_{-pi}^pi f_x(t)cosleft(Ntright)frac{operatorname{d}t}{2pi}= \ frac{1}{pi}int_{-pi}^pi frac{f_x(t)}{t}sin(Nt)operatorname{d}t+int_{-pi}^pi f_x(t)left(cotleft(frac{t}{2}right)-frac{2}{t}right)sin(Nt)frac{operatorname{d}t}{2pi}+int_{-pi}^pi f_x(t)cosleft(Ntright)frac{operatorname{d}t}{2pi}.$$
Now, the only problematic integral is the first, so let's estimate only this one.
We have that:
$$int_{-pi}^pi f_x(t)cotleft(frac{t}{2}right)sin(Nt)frac{operatorname{d}t}{2pi} = -int_{-pi}^pi f_x(t+frac{pi}{N})cotleft(frac{t+frac{pi}{N}}{2}right)sin(Nt)frac{operatorname{d}t}{2pi},$$
so:
$$left|int_{-pi}^pi f_x(t)cotleft(frac{t}{2}right)sin(Nt)frac{operatorname{d}t}{2pi}right| = left|frac{1}{2} int_{-pi}^pi left(f_x(t)cotleft(frac{t}{2}right)-f_x(t+frac{pi}{N})cotleft(frac{t+frac{pi}{N}}{2}right)right)sin(Nt)frac{operatorname{d}t}{2pi}right| \ le frac{1}{4pi} int_{-pi}^pi left|f_x(t)cotleft(frac{t}{2}right)-f_x(t+frac{pi}{N})cotleft(frac{t+frac{pi}{N}}{2}right)right|operatorname{d}t \ = frac{1}{4pi} int_{-pi}^pi left|f_x(t-frac{pi}{2N})cotleft(frac{t-frac{pi}{2N}}{2}right)-f_x(t+frac{pi}{2N})cotleft(frac{t+frac{pi}{2N}}{2}right)right|operatorname{d}t$$
So we need an uniform estimate in $x$ for the quantity:
$$int_{-pi}^pi left|f_x(t-h)cotleft(frac{t-h}{2}right)-f_x(t+h)cotleft(frac{t+h}{2}right)right|operatorname{d}t$$
for $hto 0$.
Split the integral for $|t|<2|h|$ and for $2|h|le|t|le pi$ for $|h|<1$. So:
$$int_{|t|<2|h|} left|f_x(t-h)cotleft(frac{t-h}{2}right)-f_x(t+h)cotleft(frac{t+h}{2}right)right|operatorname{d}t\ le int_{|t|<2|h|} left|f_x(t-h)cotleft(frac{t-h}{2}right)right|operatorname{d}t+int_{|t|<2|h|} left|f_x(t+h)cotleft(frac{t+h}{2}right)right|operatorname{d}t le \ int_{|t|<2|h|} |t+h|^alphaleft|cotleft(frac{t+h}{2}right)right|operatorname{d}t+int_{|t|<2|h|} |t-h|^alphaleft|cotleft(frac{t-h}{2}right)right|operatorname{d}t\ le c_1 int_{|t|<4|h|} |t|^{alpha-1}operatorname{d}t = c_2 |h|^alpha.$$
While, for $2|h|le|t|le pi$ we have:
$$int_{2|h|le|t|le pi} left|f_x(t-h)cotleft(frac{t-h}{2}right)-f_x(t+h)cotleft(frac{t+h}{2}right)right|operatorname{d}t \ le int_{2|h|le|t|le pi} left|f_x(t-h)left(cotleft(frac{t-h}{2}right)-frac{2}{t-h}right)right|operatorname{d}t+int_{2|h|le|t|le pi} left|f_x(t+h)left(cotleft(frac{t+h}{2}right)-frac{2}{t+h}right)right|operatorname{d}t+int_{2|h|le|t|le pi} left|f_x(t-h)frac{2}{t-h}-f_x(t+h)frac{2}{t+h}right|operatorname{d}t.$$
Now it is clear that the first two integrals aren't a problem, while for the last:
$$int_{2|h|le|t|le pi} left|f_x(t-h)frac{1}{t-h}-f_x(t+h)frac{1}{t+h}right|operatorname{d}t \ le int_{2|h|le|t|le pi} |t|left|frac{f_x(t-h)-f_x(t+h)}{(t-h)(t+h)}right|operatorname{d}t + |h|^2int_{2|h|le|t|le pi} left|frac{f_x(t-h)+f_x(t+h)}{(t-h)(t+h)}right|operatorname{d}t \ le |2h|^alpha int_{2|h|le|t|le pi} left|frac{t}{(t-h)(t+h)}right|operatorname{d}t + |h|int_{2|h|le|t|le pi} left|t-hright|^{alpha-1}operatorname{d}t+|h|int_{2|h|le|t|le pi} left|t+hright|^{alpha-1}operatorname{d}t \ le c_3|h|^{alpha}(1+log|h|).$$
Putting all together we get:
$$sup_{xin[-pi,pi]}left|sum_{n=-N}^Nhat{f}(n)e^{inx}-f(x)right|le Cleft|frac{pi}{N}right|^{alpha}log left|frac{pi}{N}right|to 0, Nto+infty$$
answered Jan 24 at 19:38
BobBob
1,7011725
1,7011725
add a comment |
add a comment |
Thanks for contributing an answer to Mathematics Stack Exchange!
- Please be sure to answer the question. Provide details and share your research!
But avoid …
- Asking for help, clarification, or responding to other answers.
- Making statements based on opinion; back them up with references or personal experience.
Use MathJax to format equations. MathJax reference.
To learn more, see our tips on writing great answers.
Sign up or log in
StackExchange.ready(function () {
StackExchange.helpers.onClickDraftSave('#login-link');
});
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
StackExchange.ready(
function () {
StackExchange.openid.initPostLogin('.new-post-login', 'https%3a%2f%2fmath.stackexchange.com%2fquestions%2f1012365%2fhow-to-show-for-alpha-in-0-1-any-f-in-c-alpha0-1-sim-has-a-fouri%23new-answer', 'question_page');
}
);
Post as a guest
Required, but never shown
Sign up or log in
StackExchange.ready(function () {
StackExchange.helpers.onClickDraftSave('#login-link');
});
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
Sign up or log in
StackExchange.ready(function () {
StackExchange.helpers.onClickDraftSave('#login-link');
});
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
Sign up or log in
StackExchange.ready(function () {
StackExchange.helpers.onClickDraftSave('#login-link');
});
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Sign up using Google
Sign up using Facebook
Sign up using Email and Password
Post as a guest
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown
Required, but never shown