Modeling Supernovae With a Supercomputer

A team of scientists at the University of Chicago will be using 22 million processor-hours to simulate the physics of exploding stars. The team will make use of the Blue Gene/P supercomputer at Argonne National Laboratory to analyze four different scenarios for type Ia supernovae. Included in the link is a video simulation of a thermonuclear flame busting its way out of a white dwarf. The processing time was made possible by the Department of Energy's INCITE program. "Burning in a white dwarf can occur as a deflagration or as a detonation. 'Imagine a pool of gasoline and throw a match on it. That kind of burning across the pool of gasoline is a deflagration,' Jordan said. 'A detonation is simply if you were to light a stick of dynamite and allow it to explode.' In the Flash Center scenario, deflagration starts off-center of the star's core. The burning creates a hot bubble of less dense ash that pops out the side due to buoyancy, like a piece of Styrofoam submerged in water."

Re:

[chanrobi]

Which is why it's called a simulation. Who the heck modded this insightful?

Re:flawed

[Aranykai]

Probably someone who actually knows what simulation means.

Simulation is something which simulates a system or environment in order to predict actual behavior.

To speculate on the other hand is to make an inference based on inconclusive evidence; to surmise or conjecture.

So, he was indeed insightful when he stated that the lack of understanding would render the results speculative at best.

(all definitions courtesy of wikitionary)

Re:

[Anonymous Coward]

Well, I'm a pyromaniac and I consider this stimulation.

Re:flawed

[DeadChobi]

You guys are missing the point of the modeling. You don't always model something to make predictions of its actual behavior. In this case modeling serves as an excellent way to test our models against empirical data collected from observations of supernovae. So, we do our best to construct a model, then compare this model to the real system in order to expose holes in our understanding of the phenomenon. This is good science.

Re:

[ScreamingCactus]

Agreed, I don't understand why more people don't see that. It's through simulations like these that we test our theories about the universe. Wasn't it computer simulations of galaxies that led to the discovery of dark matter? If we already perfectly understood the physics of supernovae, there would be no need to run the simulation in the first place.

Re:

[cheeseboy001]

Correct me if I'm wrong, but haven't we not actually found any dark matter yet?

Re:

[ScreamingCactus]

"the phenomenon that the majority of the gravitational effects within galaxies are unaccounted for, but are now commonly attributed to some kind of 'invisible matter'" just seemed like too much to write. Alas, I ended up writing it anyway.

Re:

[PingPongBoy]

Just wondering if this will lead to an alternative energy source, like building a dwarf star under the hood. That would cost about as much as a tank of gas anyways.

Re:

[maxume]

So if you have a model and you run what you think is a simulation and then it turns out that the prediction from your model was incorrect, you should have been calling your simulation speculation?

How is that not just hair splitting semantics?

Re:flawed

[hazem]

There are several steps in constructing useful models and the last, and most controversial is that of "model validation".

In building a computer model/simulation, you generally follow these steps:

1) problem formulation - what do you want to figure out, gather data, get the "reference behavior pattern"
2) formulate a mental model of the system - what are the entities involved and how are they related
3) build and debug your model
4) verification - this is where you ensure the model behaves as expected against specific sets of inputs - as you change inputs, does it do what you expect (I turn up the volume knob - and the sound gets louder)

5) validation - this is where you compare the results of the model with the reference data from the real world. If it doesn't match, you then have to back up and figure out what's wrong... was the implementation of the model incorrect? were your initial hypotheses incorrect? And if it does match, have you gathered enough real world data to know your model is functioning well? How confident are you of this model's ability to model the system you're interested in.

So suppose you've built a model that you can validate against gathered data, you still have to demonstrate that your model is valuable to the scientific community.

You're probably going out on a limb to make strong assertions when a model demonstrates/predicts behavior that has not been observed. However, it can serve a great role in helping determine what other things to look for, what conditions may exist, or help see relationships that you may not have seen before.

The controversy is over how much you can use a model to help prove a hypothesis.

Pfft

[Icarium]

Hookers are cheaper than models, and almost always do what you expect!

Re:

[palewook]

troll much?

Re:

[jr-slash]

You are also testing the math formula with this, if the visualized model looks different to what can be seen in nature, the formula is flawed. If it looks similar to nature it probably is a good model.

Re:

[wizardforce]

there wouldn't be any point in simulating this if we knew everything about them. it's a step on to better understanding of what is going on.

Re:flawed

[pclminion]

we understand little about it and the math formula used will be a half guess. supercomputer or not, results will be speculative at best.

I don't think you understand how experiments work... If the results of the computations are something other than what is observed in nature, then the methods and/or equations are proven wrong. That is most certainly a NON-speculative result.

Just because the model shows a burst of star stuff blowing out this way or that way in some particular configuration doesn't mean that scientists will leap up from their chairs and say "Stars do this, and we've proven it."

You can never know if your models are correct. All you can do is continually test them and try to prove them wrong. Maxwell's equations have not been proven to be correct -- they've just never been shown to be wrong. This simulation is just a step on the path of evidence.

Re:

[palewook]

"I don't think you understand how experiments work" sounds like you are making an assumption as well

Re:

[pclminion]

"I don't think you understand how experiments work" sounds like you are making an assumption as well

I'm not assuming anything. I've formed a model based on the evidence.

Re:

[madboson]

we understand little about it and the math formula used will be a half guess. supercomputer or not, results will be speculative at best.

There is a big difference between guessing at some thing and hoping for the best, verses a systematic analysis of a problem using all the available knowledge that man has right now. Yeah, sounds like a sugar coating on exactly the same thing but come on and think about it. Science is mix a little ingenuity in with previously known facts and a healthy dose of self cynicism. It's the self defeatist attitude like this that hampers progress.

Not to mention several "PhD life spans" probably were spent on e

Re:

[SL Baur]

Can you imagine a Beowolf Cluster of these things?

"The Argonne Blue Gene/P supercomputer is one of the largest and fastest supercomputers in the world," said Fisher, a Flash Center Research Scientist. "It has massive computational resources that are not available on smaller platforms elsewhere." Desktop computers typically contain only one or two processors; Blue Gene/P has more than 160,000 processors. What a desktop computer could accomplish in a thousand years, the Blue Gene/P supercomputer can perform in three days.

Oh wait ...

with a computer?

[nih]

blimey, i was expecting them to use prayer, because that has been proven to work unlike these so called 'computers', which according to Ben Stein the nazi's used

Re:

[Dirtside]

Hm, Stein's Corollary to Godwin's Law: Anything the Nazis did, used, or believed in is evil.

Kaboom

[Reader X]

Imagine a pool of gasoline and throw a match on it.

like a piece of Styrofoam submerged in water

Now I know what to do with myself on this slow Sunday morning.

Re:

[NemoinSpace]

ok, but don't try the other.

'A detonation is simply if you were to light a stick of dynamite and allow it to explode.'

At least till they figure out how to simulate lighting a stick of dynamite and *NOT* allowing it to explode?

Even then, I wouldn't test the results empirically.

They need to get Michael Bay involved..

[DanWS6]

I bet he could make a neater looking explosion without the use of a super computer. It'd even have audio!

Re:

[iknowcss]

Hey, the linked video DOES have audio! It's in space for christ sake!

crysis

[conan1989]

maybe, just maybe, it might pack enough grunt to play crysis

it would be interesting.......

[slydmin]

....to see the BlueGene explode in flames from all the complicated calculations....hope to see this on youtube!

Modeled it at home for forty-seven cents...

[Doctor Memory]

I modeled a Type 1A at home by taking an old 74LS00 IC and hooking power and ground up to a neon light power supply. I yelled "Don't cross the streams!" and flipped the switch. Glowing fragments flew in all directions, proving the inversion of the event horizon and validating my work in the field of glowing-particle physics.

There seems to be some quantum effect component also, because right after the simulation, my landlady appeared and went supernova too!

saw this on tv

[lucky130]

A little over a third of the way through s02e09 of The Universe here [tv.com] has these guys talking about their simulation.

honest question

[sunwolf]

What are processor-hours, exactly? I don't think it's utilization of the supercomputer for x-amount of hours, since that would mean they've booked the computer for almost 42 years.

Re:

[Anonymous Coward]

From TFA: "Blue Gene/P has more than 160,000 processors."

BTW, 22 million hours = 2500 years, not 42 years.

Re:honest question

[pclminion]

A processor-hour is a single processor being utilized for an hour. This supercomputer has a lot of processors (as do all supercomputers, really).

Re:

[Digi-John]

They said 22,000,000 hours on a 160,000 processor computer. That comes down to about 5.7 days.

!news

[ScreamingCactus]

I thought this story sounded familiar. Then I clicked the link, and lo and behold, there's the exact same video I remembered watching a year ago. I double checked -- the video was dated March of 2007. So why is this just now making headlines? I could understand if they re-ran the simulation with new physics that proved to be more accurate or something, but why link to the old video?

So....

[Kierthos]

was I the only one who initially read the headline as "Modeling Supermodels with a Supercomputer"?

Re:

[hoopshank]

was I the only one who initially read the headline as "Modeling Supermodels with a Supercomputer"?

Yes

Re:

[TeknoHog]

No. I was also thinking of supermodels after the headline.

Re:

[Anonymous Coward]

modeling with supercomputers should be called "supermodeling" from now on

Computing in Cloud

[JavaGenosse]

I think Scientific Modeling in a compute cloud is more sexy, since it is way cheaper than 42 millions of processor hours and allows spikes. If one doesn't see differences between lab grid and cloud, go read wikipedia or http://groups.google.ca/group/cloud-computing/browse_thread/thread/73e1030b18df3730?hl=en [google.ca]

Re:

[ceoyoyo]

That's just fine provided you don't need fast (or really, any) interprocess communication. Not all problems work that way.

Re:

[JavaGenosse]

Yeah I agree, latency might be terrible and can kill some MPI or like jobs. But for independent jobs it is perfect and cheap to overflow to cloud.

Re:

[ceoyoyo]

There are a few applications, like SETI that work that way. A lot don't. This is one of them. Someone else mentioned that this project uses a multibody problem algorithm. Multibody problems are tightly integrated. Not only latency, but synchronization issues mean you have to use a dedicated machine.

Architecture, language, details?

[munro]

Does anyone have any implementation level details about this? I'd love to hear what the software approach is, what programming language they'll be working in, how the parallelism will be handled, what sorts of problems are involved etc etc etc. We come here for news for nerds (well, cool graphics are OK too, but...)

Re:Architecture, language, details?

[mikael]

If you visit the webpages of the various research departments related to visualisation and parallel processing, then you can find many research papers related to this and other topics:

A study of parallel techniques for visualisation [ucdavis.edu].

A parallel visualization pipeline for Terascale earthquake simulation [ucdavis.edu]

Scientific Discovery through Advanced Visualization [ucdavis.edu]

A case study in Supernovae Simulation Data [uchicago.edu]

It's just amazing to find out how much is going on inside a star - not just the fusion of Hydrogen and Helium atoms, but intense magnetic fields that drive rivers of liquid Hydrogen and Helium through rising and falling convection cells, which in turn create new magnetic fields.

Re:Architecture, language, details?

[transonic_shock]

I had looked at ther work month ago when researching on writing my own N-body code. So, basically this is an implementation of Fast multipole Method (used in N-body computaion). Tradationaly (or rather in it's naive form) N-body codes are of the order N^2. Fast multipole algorithm (and Barnes-Hut and multitudes of their derivatives) does this at NLogN or better. You can have various kinds physical phenomenon occuring between two bodies/particles/points (graviational, electromagnetic etc) and this problem solves the physics for millions (or billions of such particles making up a supernovae) The entire 3d space is broken down into a oct-tree. You can traverse down to a group of particles (or one particle in Barnes-Hut), and traverse up calculating the force. The basic idea is to make a group of particles a large distance act like a single particle when calculating it's potential on another particle. Mind you the particle here is really a loose definition. It's really the most granular subdivion of space you could afford to calculate. Hence the need for bigger computers for better accuracy. The parallelism is MPI based. It's simpler to handle parallelism for nbody stuff compared to eulerian grid type problems.

Re:

[Anonymous Coward]

the code they are using, flash code, is not a simple n-body code. the flash code is a multi-physics, adaptive mesh, eularian hydrodynamics code. you may want to do a better job on your homework before posting.

p.s. - i'm one of the original authors of the flash code.

Use Those Excess Cycles

[Nom du Keyboard]

With all the excess CPU cycles coming on-line every year now, it's nice to see some significant uses for them.

We just need continually improving ways to make those excess cycles available.

Superfluous Supercomment

[Nautical Insanity]

"Modeling Supernovae with a Supercomputer controlled by the Aperture Science Supercolliding Superbutton"

with a Supercomputer?

[stymyx]

And here I thought that supernovae would be modeled by a supermodel.

Every so often the star seems to...

[exp(pi*sqrt(163))]

...stop 'exploding', do a 360 degree revolution, and carry on exploding. Are the physicists sure they have this right? I don't think that kind of process could preserve angular momentum, not to mention the vast amounts of energy that seem to be held at bay for significant periods of time.