I poked around to try to find it after posting here, and all I could find is something in arxiv. Perhaps the refs didn't like the paper... maybe there was something wrong with it? I'm not sure -- I'm not an astronomer. Looks like my memory is a bit fuzzy...:)
As a preface, "How much physics do you know?": A lot.
Anyway, there was a paper in Astrophysical Journal a few years back, in either 2005 or 2006 from the University of Victoria. They got nice rotation curves for galaxies just from general relativity, without invoking dark matter. Kind of neat.
How would Havard publishing online prevent them getting published in Nature or Science. Do you have any citations that say this?
No, I don't have any citations that say that. I was just throwing some high impact factor journal names out there. Furthermore, you misinterpreted the article I think. Harvard would not allow the authors to submit their papers to a journal whose policy didn't allow Harvard to post the papers, unless the authors apply for a waiver (as somebody later pointed out in this thread). So it's a Harvard policy, not a journal policy.
But since I threw those names out, and you came up with citations as to their specific policies we can address them in terms of my original post. Regarding Science.... absolutely. It has also been my experience that I retain the copyright on my papers (which are not in Science -- haha). However, you'll notice "the author grants AAAS exclusive rights to use and authorize use of the work" -- and one of the granted rights by Science is to post it on the author's website. Does that include some general Harvard repository? I don't know. Like I said, I was just throwing out the name "Science".
As for Nature, you'll notice that those are services provided by Nature. I doubt Nature would appreciate Harvard taking money out of their pocket by posting the papers for free on their site.
Ok, that's fine and well that if the journal allows it, Harvard makes a copy of the article freely available.
What about those journals (Nature and Science, maybe?) that do not allow this. Does this mean that Harvard faculty will not publish in Nature and Science? Somehow I doubt that. Does this mean that Harvard will break copyright agreements? Maybe?
The article doesn't quite say.
Something was wrong with their old iBook, and Apple wasn't doing anything. She sent a pretty angry e-mail to Steve Jobs' address, and got a personal reply, apology, some free stuff, and a new iBook.
OOP belongs in business... not in Molecular Dynamics.
Yes, because in physics, nothing is like anything else. (rolls eyes)
Nah, that's not what I'm saying. It's just that OOP can sometimes obscure what's going on and just add unneeded complexity to a program. I had a friend do some Monte Carlo stuff, and he used all kinds of OOP in his code, and it was done in a sane way. But trying to debug that code was hell. There's a place for everything, and I'm not convinced that simulations are the place for it.
I am wondering, as a college student who is beginning to get into a financial profession typically dealing with immense datasets, how much value will this be to scientists/mathematicians/statisticians? Wouldn't it make more sense to prototype something in Maple, PERL, etc... and hand it off to programmers to implement a full model at top speed? In other words, is there any value to learning FORTRESS analagously to the value of learning FORTRAN in the 70s-80s? Haven't advances in computing rendered these languages obscure?
Anyone doing modeling with extremely large datasets wish to comment? Should I bother to pursue low-level programming languages?
Looks a lot like Pascal, too.
Let me speak as a scientist. Looking at the example in Fortress, this is beneficial for some of my colleagues who are not programmers per se. It allows them to maybe focus on the math instead of the logic and overhead (small as it is) of actually writing a code. Somethings cannot be modeled in Maple, for example, and then handed off to programmers to implement. Secondly, there is the chance not all programmers would understand the underlying science that would allow them to make sure their implementation is producing correct results. If one were to prototype something in Perl... well, you may as well just use C (though I prefer Fortran!;) ) and write the program since the syntax is so similar.
I model large datasets -- sometimes with 15 million "particles" (which corresponds to a lot of data... that would be 45 million coordinates, 45 million forces, 45 million velocities, etc.). Fortran is very fast. I couldn't use Mathematica or Maple for what I'm doing, and for my particular task, a C implementation had noticeable slower performance (and yucky syntax when it came to doing mathematical stuff).
I think it is definitely in your best interest to learn a low-level language -- and I would readily recommend either C or Fortran, since a lot of the high-performance libraries out there mainly support these languages only (e.g., IMSL from Visual Numerics, FFTW, and friends). Fortran is easier to start off with. C is arguably more powerful overall.
I used fortran quite a lot around 25 years ago. Sure it had some oddball limitations and wierdness, but it is damn fast and quite efficient for some coding purposes.
I wrote a Fortran program that printed out a calendar with the year in a banner font at the top. It took 57 cards (no library calls etc, beyound PRINT). Try do anything useful in 57 lines with today's languages.
It shouldn't surprise you... but I use fortran quite a lot today. A lot of the oddball limitations and weirdness are gone in Fortran 90/95... though I still use Fortran 77. It is still an amazingly useful and fast language to code numerical stuff in. Yes indeed. No need for any math.h in this programmer's world! I just hope this Fortress business is just as fast as fortran, because if it's not... you won't see many fortran guys switching over. The main reason (as far as I'm concerned) that we're still using it is that it is fast, and simple. OOP belongs in business... not in Molecular Dynamics.
Or? How often you do compute volume of irregular 3D objects, or do matrix determinants or solve system of equations of third power so you can meet the month budget?
As a physicist, everyday.;)
I agree with you that for kids it's a good idea to show them how they can apply the math, etc. Leave the hard theory for college or whatever. But math is certainly not overrated, which was the point I was trying to make, and I'm sure you agree with that.
Really? Didn't anyone stop to think that maybe math is overrated?
Why might that be? Math is critical to functioning in today's world -- from figuring out whether you have enough cash in your bank account to both pay the rent AND buy that new shiny iPod to determining how to modify your recipe to accommodate the unexpected arrival of 5 more dinner guests to averaging the scores of your english students to... you name it.
I really doubt math could ever be overrated. At least not in general.
Now... TOPOLOGY (as in the branch of mathematics)... might be considered overrated in many circles... but your average high schooler isn't taking topology or any other specialized area... at least as far as I know.
I'm not a lawyer so I'm not too clear on the GPL. I thought you could modify the software under it and release it without ever being forced to hand out the source code. I could be wrong though.
Well, they're certainly free to write their own proprietary modules and have their own GUI, that's for sure.
Believe me -- I am! An 80-core machine would mean that I could take some of my older, non-MPI-based simulations and make them run in parallel very easily using OpenMP. Right now, I am using 8- and 16-CPU IBM machines for these older programs. They're so complex that modifying them to use MPI is a headache and could take a good year of my time. On the other hand, using OpenMP, I can parallelize these codes in under an hour.
A processor like this could mean that for maybe $5-6,000 I could have huge performance increases. Add to that the fact that I *COULD* make a Beowulf out of them, and the amount of work I could do is just amazing... and it can certainly benefit all areas of science.
I used to have BSOD as my screen saver for an earlier version of Fedora (IIRC). It was always amusing when people would stop by to chat, a little while later, they'd see my PC suddenly BSOD! The looks I'd see (on other people's faces) makes me laugh just remembering.
I used to have that screensaver going in X. I actually flipped out once because I got the kernel panic (wasn't thinking) and I was in the middle of something rather important and I just went on about "damn damn damn" in true Life of Brian fashion... only to realize... yes. It was a screensaver.
Is this IBM's new approach to clusters/supercomputers? We just bought a relatively small (i.e. 176 processors) system from them... half is Power5+ and the other half is Xeon.
42 thousands particle? you have any idea what is size of electrons vs atoms? interaction between nuclei? do you understand what you talking about? sines and cosines.... adn this is 5 score comment! rated informative!
I'll bite this time. Just once. I have a very good idea what I'm talking about. I said sines and cosines because to a first approximation the wavefunctions of the atoms probably resemble that, so I'd assume the interactions be built off of them in some fashion. I'm sure you took elementary quantum mechanics. Undergraduates learn that stuff. Notice I didn't say what specifically what the equations would like like because I have no idea. I have a master's degree in computational physics -- what're your qualifications?
So in essence, it takes about.2 teraflops per atom... And that was only after spending a lot of time condensing the algorithms. This makes me wonder two things. First, what do these equations look like such that it takes 200 gigaflops just to model one atom. Second, over what timeframe does this simulation take place? Are we talking real-time, calculating for 50 years, what?
0.2 TFlops per atom, yes. But there are 1000 atoms, and it's molybdenum which has 42 eletrons... so that's 42,000 particles that all interact with each other. Still... that's not too many. But maybe they're considering interactions between nuclei, too. Who knows...
As for your question about what the equations look like? They're probably very nasty integrals of sines and cosines and what not to various odd (read: strange) powers and stuff. I do fairly computationally intensive simulations on some big IBM machines and just simple equations can amount to quite a bit of calculations. Nothing like what these guys are doing, though.
Finally... what time frame is the simulation over? I'd wager VERY SHORT times, maybe nanoseconds or something like that. Even casual "molecular dynamics" simulations can only probe very short timeframes. Their coarse-grained cousins can maybe do microseconds or milliseconds.
Slashdot Mirror
Here's the link to the manuscript, though.
As a preface, "How much physics do you know?": A lot.
Anyway, there was a paper in Astrophysical Journal a few years back, in either 2005 or 2006 from the University of Victoria. They got nice rotation curves for galaxies just from general relativity, without invoking dark matter. Kind of neat.
Mike.
Yeah but the question was about how "a Windows HPC cluster" presents itself, regardless of version.
From the FAQ command line submission is supported, which to me sort of implies that there must also be a graphical submission method.
M.
M.
No, I don't have any citations that say that. I was just throwing some high impact factor journal names out there. Furthermore, you misinterpreted the article I think. Harvard would not allow the authors to submit their papers to a journal whose policy didn't allow Harvard to post the papers, unless the authors apply for a waiver (as somebody later pointed out in this thread). So it's a Harvard policy, not a journal policy.
But since I threw those names out, and you came up with citations as to their specific policies we can address them in terms of my original post. Regarding Science.... absolutely. It has also been my experience that I retain the copyright on my papers (which are not in Science -- haha). However, you'll notice "the author grants AAAS exclusive rights to use and authorize use of the work" -- and one of the granted rights by Science is to post it on the author's website. Does that include some general Harvard repository? I don't know. Like I said, I was just throwing out the name "Science".
As for Nature, you'll notice that those are services provided by Nature. I doubt Nature would appreciate Harvard taking money out of their pocket by posting the papers for free on their site.
Ok, that's fine and well that if the journal allows it, Harvard makes a copy of the article freely available. What about those journals (Nature and Science, maybe?) that do not allow this. Does this mean that Harvard faculty will not publish in Nature and Science? Somehow I doubt that. Does this mean that Harvard will break copyright agreements? Maybe? The article doesn't quite say.
Playstation 3.
:) )
xoxoxo,
Sony
(Yes, I'm kidding.
Sounds like the making of an urban legend!
Mike.
Yes, because in physics, nothing is like anything else. (rolls eyes)
Nah, that's not what I'm saying. It's just that OOP can sometimes obscure what's going on and just add unneeded complexity to a program. I had a friend do some Monte Carlo stuff, and he used all kinds of OOP in his code, and it was done in a sane way. But trying to debug that code was hell. There's a place for everything, and I'm not convinced that simulations are the place for it.Mike.
I am wondering, as a college student who is beginning to get into a financial profession typically dealing with immense datasets, how much value will this be to scientists/mathematicians/statisticians? Wouldn't it make more sense to prototype something in Maple, PERL, etc... and hand it off to programmers to implement a full model at top speed? In other words, is there any value to learning FORTRESS analagously to the value of learning FORTRAN in the 70s-80s? Haven't advances in computing rendered these languages obscure?
Anyone doing modeling with extremely large datasets wish to comment? Should I bother to pursue low-level programming languages?
Looks a lot like Pascal, too.
Let me speak as a scientist. Looking at the example in Fortress, this is beneficial for some of my colleagues who are not programmers per se. It allows them to maybe focus on the math instead of the logic and overhead (small as it is) of actually writing a code. Somethings cannot be modeled in Maple, for example, and then handed off to programmers to implement. Secondly, there is the chance not all programmers would understand the underlying science that would allow them to make sure their implementation is producing correct results. If one were to prototype something in Perl... well, you may as well just use C (though I prefer Fortran! ;) ) and write the program since the syntax is so similar.
I model large datasets -- sometimes with 15 million "particles" (which corresponds to a lot of data... that would be 45 million coordinates, 45 million forces, 45 million velocities, etc.). Fortran is very fast. I couldn't use Mathematica or Maple for what I'm doing, and for my particular task, a C implementation had noticeable slower performance (and yucky syntax when it came to doing mathematical stuff).
I think it is definitely in your best interest to learn a low-level language -- and I would readily recommend either C or Fortran, since a lot of the high-performance libraries out there mainly support these languages only (e.g., IMSL from Visual Numerics, FFTW, and friends). Fortran is easier to start off with. C is arguably more powerful overall.
Mike.
Mike.
I wrote a Fortran program that printed out a calendar with the year in a banner font at the top. It took 57 cards (no library calls etc, beyound PRINT). Try do anything useful in 57 lines with today's languages.
It shouldn't surprise you... but I use fortran quite a lot today. A lot of the oddball limitations and weirdness are gone in Fortran 90/95... though I still use Fortran 77. It is still an amazingly useful and fast language to code numerical stuff in. Yes indeed. No need for any math.h in this programmer's world! I just hope this Fortress business is just as fast as fortran, because if it's not... you won't see many fortran guys switching over. The main reason (as far as I'm concerned) that we're still using it is that it is fast, and simple. OOP belongs in business... not in Molecular Dynamics.Mike.
No. The term you're thinking of is Penile Stimulation Engineer, or something to that effect.
Mike.
...it was obviously captured by aliens.
As a physicist, everyday. ;)
I agree with you that for kids it's a good idea to show them how they can apply the math, etc. Leave the hard theory for college or whatever. But math is certainly not overrated, which was the point I was trying to make, and I'm sure you agree with that.
Mike.
Why might that be? Math is critical to functioning in today's world -- from figuring out whether you have enough cash in your bank account to both pay the rent AND buy that new shiny iPod to determining how to modify your recipe to accommodate the unexpected arrival of 5 more dinner guests to averaging the scores of your english students to... you name it.
I really doubt math could ever be overrated. At least not in general.
Now... TOPOLOGY (as in the branch of mathematics)... might be considered overrated in many circles... but your average high schooler isn't taking topology or any other specialized area... at least as far as I know.
Mike.
Well, they're certainly free to write their own proprietary modules and have their own GUI, that's for sure.
Mike.
Mike.
Believe me -- I am! An 80-core machine would mean that I could take some of my older, non-MPI-based simulations and make them run in parallel very easily using OpenMP. Right now, I am using 8- and 16-CPU IBM machines for these older programs. They're so complex that modifying them to use MPI is a headache and could take a good year of my time. On the other hand, using OpenMP, I can parallelize these codes in under an hour.
A processor like this could mean that for maybe $5-6,000 I could have huge performance increases. Add to that the fact that I *COULD* make a Beowulf out of them, and the amount of work I could do is just amazing... and it can certainly benefit all areas of science.
Mike.
I used to have that screensaver going in X. I actually flipped out once because I got the kernel panic (wasn't thinking) and I was in the middle of something rather important and I just went on about "damn damn damn" in true Life of Brian fashion... only to realize... yes. It was a screensaver.
(Score:-10, Stupid) for me that day, for sure.
Mike.
I'll bite this time. Just once. I have a very good idea what I'm talking about. I said sines and cosines because to a first approximation the wavefunctions of the atoms probably resemble that, so I'd assume the interactions be built off of them in some fashion. I'm sure you took elementary quantum mechanics. Undergraduates learn that stuff. Notice I didn't say what specifically what the equations would like like because I have no idea. I have a master's degree in computational physics -- what're your qualifications?
Mike.
0.2 TFlops per atom, yes. But there are 1000 atoms, and it's molybdenum which has 42 eletrons... so that's 42,000 particles that all interact with each other. Still... that's not too many. But maybe they're considering interactions between nuclei, too. Who knows...
As for your question about what the equations look like? They're probably very nasty integrals of sines and cosines and what not to various odd (read: strange) powers and stuff. I do fairly computationally intensive simulations on some big IBM machines and just simple equations can amount to quite a bit of calculations. Nothing like what these guys are doing, though.
Finally... what time frame is the simulation over? I'd wager VERY SHORT times, maybe nanoseconds or something like that. Even casual "molecular dynamics" simulations can only probe very short timeframes. Their coarse-grained cousins can maybe do microseconds or milliseconds.
Mike.