« first day (240 days earlier)      last day (1258 days later) » 

12:24 AM
@SimplyBeautifulArt
Im here is better
 
@RichardClare =P Okay
 
What I have down stairs dividing the integral is $1 - e^{\frac{6\pi}{5}}$ as it should be
note in the formula it says that you will have 2(n + 1)/m
that's what I get after using the wedge of 2pi/5
however at the end is not the same result I dont know where they obtain the 3 and the minus
I have the 6, because when you substitute in the integral the parametrization you should multiply by the derivative which is indeed the point.
you can erase the previous comments if you like im doing it.
 
 
1 hour later…
1:40 AM
@S
@SimplyBeautifulArt Thank you for everything.
 
 
3 hours later…
4:53 AM
@amWhy @SimplyBeautifulArt: Heheh see these:

Censorship 2

53 mins ago, 7 minutes total – 6 messages, 1 user, 0 stars

Bookmarked 41 secs ago by user21820

Interesting conversation

6 hours ago, 5 hours 34 minutes total – 5 messages, 3 users, 0 stars

Bookmarked 4 mins ago by user21820

 
 
5 hours later…
9:57 AM
@amWhy Doing good. Thanks for asking. Hope all of you are well. :)
When you want the room to be private and public at the same time. Woe to me and my indecision.
 
10:13 AM
@Mithrandir o/
 
@shredalert the room is a public room
If you don't see a lock at the top right, it's public.
Otherwise people would need special permission to talk here.
 
@Mithrandir thanks for the tip
@Mithrandir things been going good?
 
Yep
Bit tired though
 
Same here
 
Now that school started I get to do the summer vacation stuff without all the crowds ;)
 
10:32 AM
@Mithrandir excellent :D
 
11:07 AM
@amWhy: Hello!
 
 
2 hours later…
12:46 PM
@user21820 Doesn't exist?
I assume it was put to the trash
 
Yea I guess so.
I've a new 'theorem': Given any two cranks P and Q, if P is more cranky than Q then P is more persistent than Q. 'Proof': Obvious.
2
Anyway, how are things? =)
 
Meh, alright
Nice to have the day off
 
Day off? Is it a holiday?
 
@user21820 Labor day in America
 
Ah I see.
 
1:06 PM
Nifty, a self-answer after a few comments.
 
@SimplyBeautifulArt It's not yet correct.
How did he get 2 upvotes?!
 
xD
@user21820 are you familiar with regularizing a divergent series?
 
@SimplyBeautifulArt I've seen some explanations of it before, but I don't see the point of it. I know some physicists claim that it has physical implications, but as of now I weigh the evidence at about 90% confidence that they are talking nonsense.
 
@user21820 Hm, okay. I had a nifty problem I was facing concerning it, if you are interested
Hello @user170039
 
Ah if it's a purely mathematical problem for fun, then it's a different story. =)
But I probably can't help much with it except poke logical holes (if any). =P
@user170039: Hi there!
 
1:15 PM
Well, the challenge is to regularize the following series:
$$\sum_{k=1}^\infty(-1)^{k+1}\left\lceil\frac k2\right\rceil$$
 
user131753
Hello @SimplyBeautifulArt, @user21820.
 
Where $\lceil x\rceil$ is the ceiling function/least greater integer
 
I know, but what do you mean by regularize? There is no one single definition.
You can arbitrarily choose some kind of cutoff function as a so-called regularizer, and then it just comes down to a question of whether some other summation converges.
 
@user21820 To apply a summation method that returns the original series if it converges.
@user21820 I assume you mean adding on a term and analytically continuing?
I am, for the most part, wanting an approach involving a well-known summation method, such as Cesaro summation, Abel sum, Ramanujan, Borel, etc.
 
@SimplyBeautifulArt But as you know there are many many different methods of assigning values to divergent series that extends standard convergent summation, and they don't even always agree.
So you'd have to pick one.
Also, in case you haven't seen this: terrytao.wordpress.com/2010/04/10/…
 
1:19 PM
@user21820 Well, you can pick any, so long as the end result is finite
 
I'll just assign the value 0 to any divergent series then.
Tada done! =D
I assign the usual value to convergent series.
 
=P Well, then you only miss out on the fun
I only mention this because most of the common summation methods do not work
 
So what is your goal? Clearly you want a new assignment that has some properties other than what you've mentioned, namely agreement on convergent series.
 
@user21820 Meh, to just do it for heck's sake
Hello and welcome to my realm @AnantSaxena
 
Ah .. thanks .. I was planning to be a lurker
but nevermind
 
1:23 PM
lol
As of yet, I've only found one approach to doing this series
 
Thanks for commenting on my question: math.stackexchange.com/questions/2416269/…
 
And if you haven't written out the first few terms, its$$S=1-1+2-2+3-3+\dots$$
 
Yeap I wrote out the first few terms.
 
@AnantSaxena Oh, no problem lol
Funny we also happen to be talking about series regularizations in here right now
 
Haha ... nice ... Any book recommendations?
 
1:25 PM
I'll be away for a while. Back later.
 
@user21820 cya
@AnantSaxena On series regularization?
I mean, idk...
 
Yea .. Preferably ones that might be useful to a physicist talking to a mathematician
So we can atleast speak the same language
 
I'm not entirely familiar with what a physicist would use, but I'd assume zeta regularization to be in their pockets.
The approach is fairly simple.
$$S=\sum a_n$$
$$f(s)=\sum a_nn^{-s}$$
And you want to analytically continue to $f(0)$
Surprisingly, this is the only approach I've found that's able to do my problem, and it provides 1/8
 
I see ... your problem = ???
 
Btw, do you have mathjax turned on?
 
1:30 PM
nope ..
Sorry Im new to the chatrooms
 
If you are using chrome, see the picture guide I made here: chat.stackexchange.com/transcript/message/38299513#38299513
So I basically want to regularize the series S = 1 - 1 + 2 - 2 + 3 - 3 + ...
 
I see ... Use Ceasaro summation?
 
Nope, Cesaro summation gives +∞
The partial sums are 1, 0, 2, 0, 3, 0, ...
which "average out" to +∞
 
I see
 
You might be interested in Euler summations
Euler summations have the same kind of "averaging" approach that Cesaro summations do
To make it intuitive, I sort of made what I'll call a partial Euler sum.
If S(n) is the sum of the first n terms, and S(n) converges as n → ∞, then one might expect that [S(n)+S(n+1)]/2 converges to the same limit
 
1:38 PM
yea ...
 
Let S(n,1) = [S(n)+S(n+1)]/2
And S(n,k+1) = [S(n,k)+S(n+1,k)]/2
Then we would expect S(n,k) → S(n) → L, as n → ∞
 
yups
 
For example, consider S(n) = 1 - 1 + 1 - ...
S(1) = 1
S(2) = 0
S(3) = 1
...
S(n,1) = [S(n)+S(n+1)]/2 = 1/2, always
 
Cool
 
So the partial Euler sum of Grandi's series is 1/2
 
1:40 PM
Btw I wanted to ask can we analytically continue the partial sums ... "1, 0, 2, 0, 3, 0, ..."?
Like in the zeta fashion or something?
 
The full Euler sum is a bit crazier though. It claims that S(n,k) converges to the original series as k → ∞ for any n.
@AnantSaxena Sure
$$f(s)=\sum_{k=1}^\infty(-1)^{k+1}\left\lceil\frac k2\right\rceil k^{-s},\quad\Re(s)\ge2$$
(may require MathJax here)
\begin{align}f(s)&=\sum_{k=1}^\infty\frac k{(2k-1)^s}-\frac k{(2k)^s}\\&=\frac12\sum_{k=1}^\infty\frac1{(2k-1)^{s-1}}\\&= \frac12\sum_{k=1}^\infty\frac{(-1)^{k+1}}{k^{s-1}}+\frac1{k^s}-\frac1{(2k)^s}\\&=‌​\frac12\eta(s-1)+\frac12\left(1-\frac1{2^s}\right)\zeta(s)\\\implies f(0)&=\frac12\eta(-1)=\frac18\end{align}
 
yea ...
Im gonna get it ... one moment
 
Or just copy/paste onto the main site lol
 
Good idea
 
=P For the lazy people
 
1:44 PM
Haha ... or the computer challenged
 
Euler summing also allows weighting terms differently
 
How so? (example)
 
For example, we could've had S(n,k+1) = [a*S(n,k)+b*S(n+1,k)]/(a+b)
 
ah ...
 
Fairly simple :-)
 
1:45 PM
I think Ive been doing regularization for a long time without knowing its been called that
 
Euler summing also works really well for numerical approximations of an alternating series
 
cool ... Can I share with you one the series I've regularized?
 
I think its an interesting result
$$ \lim_{k \to \infty} \lim_{n \to \infty}\ \sum_{r=1}^n d_r(s) \left( f(\frac{k}{n}r)\frac{k}{n} \right) \sim \underbrace{\frac{1}{\zeta(s)} \sum_{r=1}^\infty \frac{d_r(s)}{r^s}}_{\text{removable singularity}} \times \int_{0^+}^\infty f(x) \, dx $$
If u want ill share the proof too
 
Haha, I've never really liked the removable singularity approach, it seems so cheasy
 
1:48 PM
y so?
 
@AnantSaxena Idk, I just never really liked it
 
Hmm ... k ... But I think the formula I shared is pretty cool
its like limit of a sum but with weighting certain elements
 
What's d_r(s) supposed to be?
 
the weight of the element ... (Its a variable)
s tends to 1
 
It's kind of like an analytic continuation + Abel summation + Cesaro summation?
Hm, idk, but yeah
 
1:51 PM
Im not sure ... I can share the proof and then u can decide urself
 
Okay, sure.
 
Im really not a mathematician ... but apparently ive been playing with series regularizations without knowing it
Consider an integral such that $$ \int_{0^+}^\infty f(x) \, dx = C,$$where, $f(x)$ is a smooth and continuous function and absolutely converges.

Now we raise both sides to the power s:

$$\left(\int_{0^+}^\infty f(x) \, dx\right)^s = C^s $$

We substitute $x$ with $rx$ to get:

$$\left(\int_{0^+}^\infty f(rx) \, dx\right)^s = (C/r)^s $$

Multiplying both sides by an arbitrary coefficient:

$$ (b_r)\left(\int_{0^+}^\infty f(rx) \, dx\right)^s = (b_r)( C/r)^s $$

Taking their sum:

$$ \sum_{r=1}^\infty b_r \left(\int_{0^+}^\infty f(rx) \, dx\right)^s = C^s \underbrace{\sum_{r=1}^\infty \frac
 
@AnantSaxena Ah, very nifty
 
Cool right!!
 
But should it be $s\to1^+$ at the end? Just for obvious convergence issues.
 
1:54 PM
Yea ...
 
=P
Still, that's pretty nifty :-)
 
o/
Nice to see the realm still going
 
@shredalert yeup, we're still here :D
Except for @user21820, who just left
 
assume series is absolutely convergent
 
@AnantSaxena then yeah, it doesn't matter :P
 
1:56 PM
Yay!!
 
Btw, would you be interested in numerically approximating an integral?
 
which integral?
 
$$\int_0^\infty \frac{\sin(\pi x)}{\ln(x)} ~\mathrm dx$$
Try approximating this out at least 5 places
 
Im not good at programming
Ill try pen and paper
 
The minimum natural $n$ such that $\left|\int_n^\infty \dots ~\mathrm dx\right|<10^{-6}$ has $n>10^{4,000,000}$
 
2:00 PM
Wait cant I use complex analysis or something?
 
@AnantSaxena xD Okay then, good luck. I personally can't do it without a computer
@AnantSaxena Sure I suppose :-)
 
nevermind ... Just substituted $x$ with $e^x $ ... And Im stuck
 
Maybe I need some more time ... Unless its one of those series where a trick to solve it doesnt even exist
*integrals
 
I personally brute forced the integral
Though it is possible to do in a nice analytic manner
But it doesn't have any good closed form AFAIK
 
2:04 PM
really? (my instincts were right?)
 
=) I won't tell you, else you'll miss out on all the fun of trial and error
 
haha ... k ... i'll try it out
 
@SimplyBeautifulArt (this looks like gibberish without MathJax in chat)
 
@SimplyBeautifulArt you might be interested in reading about functional analysis for applications to numerical approximation techniques. They go hand in hand together.
 
@SimplyBeautifulArt At first I thought it diverges... then I realized it ought to converge.
 
2:09 PM
o/ @user21820
 
@Mithrandir There's a ChatJax bookmarklet but I state no guarantees that it is clean (no malware) or that it works (I haven't tried it).
@shredalert: Hi there! =)
 
@user21820 Hello :)
 
I gotta go ... Dinner waiting ... Catch up with you all another time :)
 
@Mithrandir =P
@user21820 Terribly slowly though
@AnantSaxena cya
 
cya @AnantSaxena
 
2:15 PM
I can easily approximate it numerically using Graph via a series acceleration trick.
I get −3.2192, assuming that Graph is not numerically unstable. I have no idea what that is supposed to be.
Is my numerical value correct? Curious to know the trick? =)
 
@user21820 Correct... but I meant 5 places after the decimal
@user21820 The trick is to replace $\sin$ with $e^{\pi ix}$ and do a quarter circle contour: math.stackexchange.com/a/2407623/272831
@user21820 but yeah, series acceleration also works
 
2:43 PM
 
@SimplyBeautifulArt −3.219195 Unfortunately, attempting your challenge made me find out that Graph cannot perform numerical integration to high precision, so I in fact have no choice but to use series acceleration to do it...
Actually I give up. I cannot be certain of the last digit "5" using my software.
I could plug into WA though.
Hmm WA doesn't allow more than 5dp???
It seems that series converges too quickly for series acceleration to be accurate!
Try finding integrate[sin(pi*x)/ln(1+ln(1+x)),x,0,∞].
@SimplyBeautifulArt: Where are you? =)
 
Lol
@user21820 I may have found a bug in Mathematica? math.stackexchange.com/a/2416484/272831
 
2:59 PM
I seem to be able to use acceleration to find the numerical value for the above integral to much higher precision than for yours.
 
@user21820 Meh, I didn't use many terms. What acceleration method are you using?
 
Well the trivial one; let g(x) be the definite integral from 0 to x, and h1(x) = (g(x)+g(x+1))/2, and h2(x) = (h1(x)+h1(x+1))/2 = (g(x)+2·g(x+1)+g(x+2))/4, and so on. h2 is too slow. h3 and h4 converge fast but not fast enough before Graph get numerically inaccurate.
 
@user21820 Lol, did you actually try and implement my approach?
 
Hence for your integral I only can confirm 6sf. But for my integral I get 11sf with just h4(80).
Does your approach work for my integral? I didn't try because I wasn't looking for an analytic method.
 
@user21820 Oh, you want me to try and do that?
 
3:05 PM
Yeap!
 
*scratches head* And you used my averaging partial Euler sum approach? Am I expected to tackle it differently?
I wonder if Euler was originally thinking about what I had laid out
 
I assumed you were only using your averaging approach for divergent series...
@SimplyBeautifulArt <− I thought you asked me to try this.
 
@user21820 =P What?! Nah, its useful, and terribly easy to program
 
But would you also have done to higher order as I did?
 
@user21820 You mean up to h4?
 
3:08 PM
Yea haha..
 
Ah. Then no different.
I thought you might have a different method.
 
Lemme think
 
I think there must be a better way to accelerate for the kind of integral like yours that converges fast, but lazy to think right now.
 
For my integral? Did you see robjohn's approach?
 
3:11 PM
I didn't notice any link from you, so nope.
Link please? =)
 
He replaces sin with e^iπx, then did a quarter circle contour
This makes the resulting integral converge very fast
3
A: Improper convergent integral: differentiation of parameter

robjohnUsing the contour contour integration gives $$ \begin{align} \overbrace{\color{#C00}{\text{PV}\int_0^\infty\frac{e^{i\pi x}}{\log(x)}\,\mathrm{d}x}}^{\substack{\text{integral along the line}\\\text{minus an infinitesimal}\\\text{interval centered at $1$}}}+\overbrace{\vphantom{\int_0^\infty}\ ...

 
Thanks! Aha I was right in doubting my last digit!
I'm sure I can find a purely numerical method for such quickly converging integrals. I don't quite like special methods that only work for special cases hahaha..
 
@SimplyBeautifulArt: I'm beginning to suspect that perhaps it's the Graph software that has inaccurate numerical integration, rather than the acceleration method. Do you mind computing h4(100) and h4(101) in any other software you have?
Which means you just need the 6 definite integrals over [0,n] for n in [100..105].
 
3:34 PM
lol, gimme a bit
@user21820 which integral? Mine, or yours?
 
Yours
It's the one I can't get accurate.
h4(100) and h4(101) lie on the same side of robjohn's numerical answer. That's why I suspect it's not the acceleration method that's the problem.
 
How many places you want?
 
10sf.
WA only gives me 6sf. So sad no more "more digits" feature.
 
@user21820 Just type 15 places in front of the integrate...
 
=O does it really work?
 
3:37 PM
Hm... idk, don't think WA is accurate enough on that
@user21820 it might just return the first 15 places it computes, but I'd only trust the first 6
I don't really have a good integration software :-/
 
I'm trying this now haha..
It fails when lower limit is set to 0, I think due to division by zero...
 
why am I trialing and erroring with these things when I can ask smarter calculator people?
 
I just finished using numberempire and spreadsheet to compute.
The method works flawlessly.
I get h4(100) = -3.21919002089425 and h(101) = -3.21919002068104.
 
Oh wait... the "2" is wrong.
 
3:46 PM
I wonder, is it more accurate to tackle h4(100) or h100(4)?
(obviously the second requires a lot more effort)
 
Guess what? Graph can't even compute g(1) correctly to 10sf...
That's why it's wrong.
I used numberempire with lower limit 1e−10 now.
 
-3.21919003874785
-3.21919003853464
Tallies with robjohn's value.
Yay!
 
So this method works.
 
3:47 PM
Ofc it works lol
 
No it's not obvious that it should work well. The series needs to converge smoothly, otherwise it won't work.
 
It actually sucks against series that aren't alternating, since averaging gets you farther from the limit than the next term.
 
For monotonic series I would extrapolate instead.
 
@user21820 :-/ Well if the series is alternating and absolute value of terms are monotonically decreasing, then its obviously bounded towards the limit.
 
What I mean is that the method won't allow you to numerically compute limits of jerkily converging alternating series to high precision, despite the theoretical limit being preserved by the acceleration.
Assuming linear rate of convergence usually works well for monotonically converging sequences (unless it actually converges faster than that).
 
3:51 PM
Ah, yeah, I suppose.
@user21820 How do you set the error on the integral calculator?
 
@SimplyBeautifulArt You're referring to numerempire right? I didn't set anything. I just took whatever it gave me.
 
@user21820 Ah, okay
 
It's not bad, being correct to 10sf.
Now I know I can't rely on Graph for integrals.. =(
 
Assuming it is correct, then yeah, WA does not do so well
@user21820 =(
 
Oh that's interesting. I would have thought WA would do well.
I used to use numberempire many years ago until WA came along..
I clean forgotten the website name and had to wade through many Google searches to find it again...
 
3:55 PM
Understandable
 
However, it gets the original question wrong. It says that the integral from 1e−10 to inf fails to converge. =P
 
@user21820 =P
 
Anyway okay enough playing for me. I shall be off soon. See you next time! =)
 
Cya @user21820
 
4:33 PM
@user21820 Wrote a small little thing
 
 
3 hours later…
7:33 PM
Will you share with me now what your class schedule(s) are this term? Last I asked, they were yet to be confirmed. Of course, no need to share all, but what math class(es) are you enrolled in this fall? (Again, only if you want to share.)
 
7:57 PM
@SimplyBeautifulArt ^
 
@amWhy Oh, sure. I'm taking AP Physics Electro-Magnetism, AP Macroeconomics, AP Computer science A, AP US gov, chinese, and then some English + electives involving the use of some Adobe products
 
 
1 hour later…
9:08 PM
@SimplyBeautifulArt Impressive!
Do you have a keyboard that accommodates chinese characters?
 
@amWhy No sadly
 
Well, I still believe that learning to write in any language is best started with a pencil (or pen) and paper. I know Spanish, but can utilize on my computer the entire alphabet, and in particular, accents! But it is doable. So I am quite intrigued about Chinese.
 
Well, I know you can type in pinyin into online keyboards, which then allow you to select the various characters.
 

« first day (240 days earlier)      last day (1258 days later) »