Mathematics - 2012-08-27 (page 1 of 3)

« first day (754 days earlier) ← previous day next day → last day (4266 days later) »

12:01 AM

@PeterTamaroff Thanks for being solicitous.

@MeAndMath That's just me, being me =)

@PeterTamaroff :-D

@MeAndMath But really, don't take advantage! =P

@PeterTamaroff Of course!I would never!

Gustavo Bandeira

12:34 AM

@MeAndMath I'm back!

Welcome back!

Gustavo Bandeira

Quase que me roubam tudo! Teve jogo de futebol, a torcida tá passando pra "captar recursos pra compra dos ingressos do próximo jogo"

@GustavoBandeira vish!

Gustavo Bandeira

What's up?

nothing.

Gustavo Bandeira

12:36 AM

What's your name? I'll call you Pizza - because I don't know your name... and as we're from math, we can use it in a symbolic way.

Then: What are you doing, pizza

?

call me mary

or meanmath.like everybody else

Gustavo Bandeira

My girlfriend has this name.

it's a common name

Gustavo Bandeira

Yep.

I solved another Project euler problem today, yay!

out for about 6 hours... see y'all later

Gustavo Bandeira

12:41 AM

I remembered from this song from the english course.

@zzk Welcome!

@GustavoBandeira Really?what problem?

Gustavo Bandeira

The third.

It's easy.

fascinating

do you like number theory?

Gustavo Bandeira

I still have only a basic conception of what it should be, but I think it's interesting. Why?

just asking

Gustavo Bandeira

12:53 AM

I have an Invitation to number theory, from Oystein Ore.

Did you read it?

Gustavo Bandeira

Nope, I've been instructed to study calculus first, but I'm gonna read a little now.

You like it?

like what?

Gustavo Bandeira

Number theory?

Who doesn't like it?

Gustavo Bandeira

12:56 AM

Oh, this invitation is really small: ~127 pages.

what do you know about calculus?what calculus are you in?

Gustavo Bandeira

I always thoght that "practical" people don't like it.

I'm in the beginning of calculus 1.

I got stuck, searching for a deeper understanding on trigonometric functions.

yeah.it1s funny.a few collegues in applied hate it.actually they don'nt know much about it...shame...anyway.I think people think I'm tooo applied for pure and too pure for applied.I just love math too much.

I'm in calculus 5 LOL

Gustavo Bandeira

Same, I'm just TD4I

I mean, I love it but I'm TD4I

sorry,what does mean TD4I?

Gustavo Bandeira

1:00 AM

"Too Dumb For It" =D

ok

Gustavo Bandeira

I neglected math while I was in high school.

When I tried to probe for deeper meanings on it, my teachers cut me down.
-Teacher, why is the sine function useful for?
-For passing the vestibular.

That wasn't the answer I expected...

...no words

Gustavo Bandeira

With ~18 years, I started to get curious on it, it was the only thing I had no idea of what it was.

Math = ?!

sad...

Gustavo Bandeira

1:04 AM

I still have no idea of what it is, but I have a "feeling"

no one knows

math has a meaning for every single person

Gustavo Bandeira

No one knows

jeez...please...

Gustavo Bandeira

Haha

Youre on calculus 5, how many calculus are there?

i think 6

Gustavo Bandeira

1:13 AM

Are there sausages made with meat and cheese?

WTF?

Gustavo Bandeira

It would be GREAT

?

Gustavo Bandeira

It would be a mix of this and sausage.

Btw, this cheese is incredible!

Hmnmnm!

there's a paradox:ham sandwich hmmm topology paradox

Gustavo Bandeira

1:19 AM

0

Q: Sausage made of cheese and meat?

Is there some kind of sausage that blends some kind of cheese with meat? In Brazil we have a kind of cheese that has almost the same shape of a sausage, I guess this could exist.

cheese sausages

Gosh...

Gustavo Bandeira

What?

nothing.

sausages reminds me math.

Gustavo Bandeira

Why?

Pizza reminds me more of it.

: For the food sometimes called a Wiener (sausage), see hot dog or Vienna sausage. In the mathematical field of probability, the Wiener sausage is a neighborhood of the trace of a Brownian motion up to a time t, given by taking all points within a fixed distance of Brownian motion. It can be visualized as a sausage of fixed radius whose centerline is Brownian motion. The Wiener sausage was named after Norbert Wiener by because of its relation to the Wiener process; the name is also a pun on Vienna sausage, as "Wiener" means "Viennese" in German. The Wiener sausage is one of the simp...

mathworld.wolfram.com/SausageConjecture.html

sausage conjecture

actually everything reminds me math

Gustavo Bandeira

1:27 AM

I'll see in a min.

take your time

Gustavo Bandeira

Nice name. Haha

Can you figure what is in this wikipedia article?

I still can't.

what article?

Gustavo Bandeira

On Wiener sausage.

no

If you let me,I'd better go to sleep...I'm faling over keyboard...

Guten nacht!

Good night!

Gustavo Bandeira

1:37 AM

Guten nacht. =)

Auf wiedersehen - i guess.

bye bye1

2 hours later…

4:03 AM

Is @Jasper now @WillHunting?

Gustavo Bandeira

4:31 AM

@JayeshBadwaik Yep.

Just Will Hunting? I thought he was good. :P

Gustavo Bandeira

lol

user19161

5:00 AM

Hey @rob! I see that you are back to the original mean square. That looks way better than the twisted version!

Till now he had an apparent dual personality (a benevolent one and a mean square one). Now he is back to his inherently mean square personality.

Especially since @WillHunting is not being good.

user19161

Have not seen Rajesh and Jordan in here for a while, hope they are OK.

hi folks. anyone in here a C developer?

5:16 AM

@Jeff yup, I am.

Got time for a stupid newbie question?

@Jeff shoot.

Am I not allowed to declare functions using #define variables? Like this:
#define N 8 /* Number of masses */
#define NEQ 2*N /* Number of equations */
....
static void PrintIntro(int npes, int N, int NOUT);

@Jeff Yes, you can define even complex functions using such macros. For a long time, mathematical functions were defined in the same way to save on performance. But, nowadays, compilers have become much much better and hence, it does not make any sense to define stuff as such, especially if performance is the only factor.

You would be better off using const variables.

Can you help me understand this error: FPU.c:78: error: expected ';', ',' or ')' before numeric constant FPU.c: In function 'main':.

Where line 78 is: static void PrintIntro(int npes, int N, int NOUT);

5:22 AM

You syntax is wrong.

you have effectively declaring "int 8" where you have written "int N"

oh. so then I cannot use a define variable in a declaration

you can use, but you have to use just N. Did not notice you were talking about declaration. No, just type int k in the declaration.

like this? static void PrintIntro(int npes, N, NOUT);

that gave me an error, too

Sorry, just edited.

You have to give a type and a dummy variable, if it is only declaration, not definition, there is no need to specify the variable itself except for readability purposes.

so: static void PrintIntro(int npes, int k, int l) ?? even though I don't have a k or l variable

5:25 AM

Yup

so even better, i can just say static void Printintro(int, int, int)

Yup, but then readability is gone.

I would suggest
static void PrintIntro(int npes, int number_of_masses, int number_of_equations)

well, C is not the most readable language anyway :D. and it compiled correctly :D

ok. i'll use your def. that's even better

But I went with: static void PrintIntro(int npes, int nMasses, int nTimeSteps); instead

...because I told you the wrong variable name at the very start (MX was not the var used).

thanks @Jay

Hmm, that is okay too i guess.

Welcome.

No. I'm copying another program. The program I'm copying is here: sundials.sourcearchive.com/documentation/2.4.0-2/…

whoops. i think you removed what i was responding to

5:32 AM

Just to be picky since you are a newbie: You know that using #define is not saving you any performance right?

@JayeshBadwaik response above :D

:-)

this is a rather advanced program for a C newbie. It uses MPI (do you know what that is?) and an advanced math library called CVODE, which is an ODE solver.

Yeah, I have used MPI in my work a lot.

and I used a different ODE library.

But I did most of my work in C++.

what! no way! I was beginning to think nobody had similar experience. I'm working alone and have had sooo much difficulty getting answers to questions. My prof. is a math professor and doesn't know computers. He selected me for this project because I have some programming experience - but not in this.

Yes, stackoverflow does, but there was nobody there right now. And also they are not nearly as friendly as here in math.

5:35 AM

@Jeff Well, what is your background?

I mean what equations are you solving?

I am here so ping me if you have any problems with the stuff.

my backround is programming, but it was application level programming (specifically, Lotus Domino...

Okay.

the first equation i'm solving is called the FPU equation (ever hear of it?).

When you first said FPU, I thought Floating Point Unit

Is there a way to friend people on math stackexchange? Or something like that

no. no... Fermi, Pasta, Ulam. It's an unresolved problem from the earliest days of programming and mathematical/scientific programming.

5:37 AM

Fermi-Pasta-Ulam right?

@JayeshBadwaik @jay yes

My prof wants me to program the FPU system. Then he wants me to do it in 2 dimensions - all so that I can get my master's degree! (in his defense, it is my master's thesis)

Yup, I did play around with that a couple of years ago. Just amateur stuff, nothing big.

no kidding! that's too funny.

Say, do you know where i can get documentation for MPI? I can't find it for the version of mpi we have

Almost all MPI is the same, there is no major difference. API is same. that is why it is MPI (I for interface)

But there are some differences. What platform do you have?

platform is Gentoo Linux. I forget the version of MPI (and how to find it).

5:42 AM

Hmm, see, basically there are many versions. Apart from the specialized vendor versions, two are widely supported.

mpich and openmpi

i think we have mpich 1.7

Okay. Nice.
see this https://computing.llnl.gov/tutorials/mpi/

Basically just google mpi tutorials

there will be thousands of them

and a lot of good tutorials are from us govt national lab site
argonne national lab, sandia national lab and similar

yeah. i've been reading tutorials about everything from linux, to bash to emacs and gdb... :D

the lab you linked me to is the same one that makes CVODE (the lib i'm using).

Okay.

You have a year I suppose?

we are using MPICH Version 1.2.7. The site for MPICH only has doc sets for v2 and above

5:45 AM

For the problem?

no. i'm a year behind. i already finished my classwork.

my professor is very intimidating (though he doesn't know it) and i had m hands full just trying to pass classes, so I didn't do much work at all on this until the last few months.

so now i have to get a job and finish this! :D

MPI v1 is a subset of MPI v2. so see other tuorials if possible.

Do you have linux experience?

nope

are you a linux (or unix) user?

Yes.

Ahh, then jumping into gentoo is linux equivalent of trying out a superbike for learning motorcycle riding.

yes. i found that out when i tried to install gentoo linux on a spare machine i had at home so i could learn on it!

i spent months trying to figure out how to install it and then install CVODE and, after all that, found out that since I'm using MPI, I can't do anything on my home system anyway! :D

5:50 AM

Use linux mint, fedora, ubuntu or something similar. Or probably even CentOS at home.

i have a virtual Ubuntu linux on the same system i'm using atm.

Okay, good.

So use that system for learning all the things

Ubuntu.

but i never use it. i have enough linux knowledge now to proceed with the programming (at least i hope i do).

How complex is your system?

8 entities?

entities? do you mean masses (or oscillators)?

5:53 AM

How many variables I mean.

it is specifically supposed to be variable number of masses.

Unknowns.

Still a range?

Because, the stuff is, directly attempting MPI with a shaky background is kind of a big risk.

i'm not sure. the governing equation for mass $n$ is this $\ddot{u}=u_{n-1}-2u_n+u_{n+1}$.

Whenever I wrote solvers, we first wrote solvers for a single machine.

Then and slowly parallelized differnet parts of my code.

Most of the ODE solvers would work without MPI anyway for the same reason.

Well, the ODE solver is a library, so i don't have to worry about that. i just have to write the function which is the RHS of the system.

Yeah, but we need multiple processors for big systems.

so how many variables would you call that equation?

5:56 AM

This one mass is three unknowns.

The first derivate, the second derivative and the variable itself.

So how many masses are you supposed to have? a range?

are you counting the mass to the left and to the right as a variable, too?

No,

the number of masses will vary each execution. the idea is to test a variety of levels masses and collect information.

so then it's 3 variables, i guess. multiply that by the number of masses (which we always call $N$). $N$ will have values $8, 16, 32, 64$, and $128$.

and it wil be in two dimensions (really $N \times N$ system).

Then, actually your system is not too big.

Has your professors mandated that the problem be solved using an MPI software?

that's not big?!

5:59 AM

Naah.

For my simulations we had around 1 million unknowns.

he has not mandated it. but he seems to think it's not soluble on a PC.

which is normal in FEM methods.

no kidding! well, then he will probably want to use even larger values of $N$

i've heard of FEM, but i don't know anything about it. a friend of mine's thesis has to do with FEM.

has he worked enough on it?

If he has, get his help, that will be a lot better.

my friend doing the FEM work? she is using some Windows based tool to do her FEM work. i offered to help her, because she is not computer savvy. But she continued on her own.

6:03 AM

@Jeff oh, so she must be using some commercial software (Ansys or similar)
Anyway, my point is, MPI is tricky, so most of the libraries provide option to run methods without MPI.

I am yet to see if CVODE does.

are the #define constants in scope inside of functions? the following line is giving me an error:
static void SetIC(N_Vector u) {
int n, N;
...

CVODE has an MPI option.

again you are definining a constant.

int 8;

is illegal

Put another way, $N$ is in scope inside that function. So I can just use it as is.

Yes.

#define scope is global

I was only declaring $N$ so that I could assign it the value of #define N, anyway.

6:06 AM

Okay, I just saw the library. It has serial examples too.

So, thing is generally there is a single switch or a simple method change from a serial to parallel.

So, first get your system running on a single machine.

yes, it does. wow, you went through its doc pretty fast.

I already had it working on single system. :D

but a single system is so different, that there is a lot of changes i have to make.

Now what you do is, you try to setup MPI using very small simple tutorials.

And after that you combine the two.

so here's a big problem i'm having. suppose i have this declaration: realtype *udata;

okay

and then this line: udata = NV_DATA_P(u);

how do i find out what was put in udata? how do i find out how many elements udata is? etc? (note: i'm using gdb)

6:10 AM

There is no way to find out the size without knowing what NV_DATA_P(u) does.

NV_DATA_P is a CVODE routine

Okay, so it must specify what does it return

and will most probably also specify the size of what it returns

yeah. i am having difficulty understanding that chapter of the docs. i'll have to go over it again.

no. the size depends on what was sent in somewhere else.... grr.....

Yup, that is true.

have you MPI programmed any second order ODE problems?

6:12 AM

Yup.

You know how you have to split a 2nd order ode up into two first order equations? And you know how you have to split the problem up into the number of processors it's going to run on?

@Jeff The splitting from second order to first is mathematical, depends on the exact equation you have to use.

in other words, if you have 20 equations and 4 processors, you have 5 equations on each processor. then each process has to send the left most number to the previous processor and the right most number to the next processor. understand what i mean?

Yes, I understand what you mean. But this type of stuff is handled by the library solver code.

See basically, at every stage, finally you will have a matrix.

not in the example program for CVODE. Look here: sundials.sourcearchive.com/documentation/2.4.0-2/…. and do a find for "Function Called by the Solver"

6:18 AM

And two vectors and one of the vector is to be determined by inverting the matrix.
So, now the first stage of parallelization is inverting the matrix.

Okay will see.

The function below that, called "f", is the RHS of the example code. About 20 lines into that function is the comment /* Pass needed data to processe...

That, and the next paragraph, are sending the results of each equation to the process to the left and the right

and then receiving info back from them.

Yup. Okay.

So that is quiet automatic. You don't have to do much there.

well, that works easily for a first order ODE. i have a second order one. so i'm trying to think of strategies for how to organize the system and what info to send to whom.

@robjohn heyo

For second order DE, my guess is you would have to send data to two machines on the left and two on the right.

@JonasTeuwen What do you use for plotting graphs?

6:23 AM

So this example program breaks the problem up into local and global vectors.

I mean not TikZ/pstricks type of plotting.

Yes

@Jeff Yes

so i think each local vector will have to include two equations for each mass.

two equations?

two equations for each mass in the local vector. that is, if there are 4 masses being processed by a thread - then there will be 8 equations.

yes. Each mass will have two equations.

But then, the two equations would also have to send the data to each other.

and then there will be a third equation between the two derivatives.

6:29 AM

right! not only that, but i have to make sure the equations are divvied up so that there is always an even number of equations on each processor.

A better way to deal with the stuff is to separate the association of masses from processor and instead, deal with the equations as associated with the processor.

So now you will have two levels of equations.

The first set is that of for the masses and the second is for the derivatives.

And now, apply similar strategies to each of them .Send a value to the previous and send a value to the next.

First solve the equation for the for masses, then for derivatives or reverse.

are you saying to put the first step of the equations on one set of processors, then the second set on another set of processors? i think that would be way too much communication, no?

Oh hello @Eugene, long time no sign of being alive!

@Jeff Communication will not be your bottleneck till quiet a large number of N.

@jay to clarify: if my system is $\ddot{u}_n=u_n$, I introduce variable $v_n$ and break it up into the first equation $\dot{u}_n=v_n$ and the second equation $\dot{v}_n=u_n$.

6:35 AM

Then now you have two sets of equations, and you will have to solve them one after other.

And yes, this is how you should do it.

Are you saying to put all the "first equations" on one set of processors and all the "second equations" on another processor?

No, I was going to comment on that statement. I was saying, treat them independently. Like suppose you have 1000 "first equatiosn" and 990 something "Second equations"

Can anyone upvote this question please?

@gigli ok. just did.

and you have 20 processors

then first solve the first set of equations 50 equations on each processor

and then solve the "second equations" set

on 50 on each processor.

6:38 AM

Thanks.

@jay i really think that would be a lot of communication, because the second equations depend on the firsts. i was thinking of grouping the equations so that if i have 1000 masses (that's 2000 equations), then if a processor handles 50 masses, it does two levels (100 equations) of processing, and pass the results right and left.

There will anyway be slowdown due to syncing.

the method you described means (I think, if i understand it right) that i would have to pass each result for every mass on every step.

Hmm.Yes my method does that.

i'm going to try dividing it up by masses. if that doesn't work,by then i'll have plenty of experience to fix it :D

6:46 AM

Yup, that seems reasonable.

so now i have to figure out how to make each local vector twice as long as the number of masses without overstepping the other vectors.

the reason i suggested the other alternative is you are then just basically solving a first order equation thrice. The code complexity reduces and the bugs too. You can reuse the first order code for any order you want.

so we don't have to worry about different data structures.

i see. why thrice? you mean twice, right?

sorry twice

I did the mistake of calculating "twice for the first equatiosn and once for the second"

which is basically once for the first equation and once for the second

i'm going to try a debug run right now (btw, I think debugging in MPI is hazardous to your health). be back in a few. (again, thanks for your help)

6:51 AM

Well, I am going off for lunch now. See ya later. After around an hour or so.

Welcome.

Lol (debugging) so true.

@jay i might be in bed by then. are you on linkedIn or anything where i might find you again (and, by "find", i mean "pester" :D ).

@Jeff My email is given on my profile.

jayesh[dot]badwaik90[at]gmail[dot]com

1 hour later…

user19161

8:14 AM

Does anyone know why air-conditioners leak water? Is the pipe choked and the water from the air?

@Will the basic answer is that it's sucking humidity out of the air. But good ACs don't leak. A good AC should be able to evaporate the water right back into the air, but on the outside.

Arrgh. This internet. I am going to kill the broadband guy.

8:35 AM

huh, low-hanging fruit

A fruiting body.

@JonasTeuwen

Hey

Can I ask you something

What does it mean to be uniformly bounded in $N$ and $x$ say?

@Gigili hi! long time no see!

@JonasTeuwen If I'm looking at $\sum_{0 < |n| \leq N} e^{inx}/n$ say

what does it mean to be uniformly bounded in $N$ and $x$? @JonasTeuwen

@JonasTeuwen You're getting really obsessed with the mushrooms...

8:41 AM

@J.M. You would for less! 8-).

@BenjaLim That there is a constant not depending on $N$ and $x$ such that...

@JonasTeuwen Ok.

@J.M. Candida just is hairy.

Because I understand uniform boundedness in the usual sense

Also hairy leukoplakia is odd.

but when they say "in ..." I get lost. @JonasTeuwen

Ok thanks

8:42 AM

@BenjaLim That means the appropriate constants are independent on that stuff.

@JonasTeuwen Yes, it makes eating a less enjoyable experience. Yeesh.

Nice day!

Guys, is it true that restriction of real or imaginary part of holomorphic function in $\mathbb{C}^2$ on riemann surface is harmonic on this surface?

@Ilya good day! Ты дарил научнику подарки на ДР, когда был студентом?)

@J.M. It makes it wonder wtf. Because for me it was associated with other things which I do not have. But it appears to be possible.

But whatever!

I love the phrasing empty and hollow, but not empty enough to not feel terror and hopelessness. Probably cited it wrong.

@JonasTeuwen Real life is strange that way...

Yep. Oh yeah, whatever.

But the guy was like: well... maybe we need to test again like punctions or whatever. I am like hell now what wait what? No. Fix this first please. Hmm.

Perhaps a bit like (but totally not like!) a friend of mine that had some extra metastasis and was like "what does it matter a few more tumors? Have plenty of them already" with a really big smile.

That was quite... interesting.

8:54 AM

@JonasTeuwen Now that's an interesting way to look at things, yes. :D

Well, yes. It somewhat nullifies all other problems.

As in: right, whine I should not. Nothing to do about it.

But, cool!

What is the thing in your gravatar?

Only two days of holiday left... noooo.

« first day (754 days earlier) ← previous day next day → last day (4266 days later) »

Transcript for

Aug '1227

$Mathematics$ Mathematics

Associated with Math.SE; for both general discussion & math qu...

join this room
about this room

00:00

06:00

12:00

18:00

all times are UTC

$Mathematics$

site design / logo © 2024 Stack Exchange Inc; legal

mobile