What are we accomplishing?

[jasong]

I'm not sure what to post here, I'm bored and need something to do, but I don't want it to be totally worthless. So I'll ask the following questions:

What is Eon looking to accomplish? I don't know a heck of a lot of people who care about the long-term dynamics of solids, much less ice growth, so what is the purpose of the project? Is there some sort of question being asked, or is it just "pure science" that doesn't have a perceivable use at the moment, but may in the future. Is it the result of a science project idea that's main intention is to give someone a really good GPA?(I'd crunch for 2-4 weeks for a mild-mannered college student trying to get an A :D )

It's a cool project, but a more dumbed-down explanation and some verbal theatrics might be helpful.

[graeme]

Excellent question. I'll write a simple explaination about the project, what we are trying to do, and why people are interested in these kind of simulations. I'll find time for this tomorrow.

[graeme]

The eon project is about simulating dynamics of systems at the atomic scale. We are working on developing new computational algorithms and using distributed computing to solve this problem.

A fundamental challenge of simulating dynamics of atoms is one of time scale. The most direct way to calculate how atoms move over time is to simulate the motion directly -- in the same way you would calculate the trajectory of a baseball using Newton's laws (F=ma). This is not hard to do, but a problem arises because the atoms vibrate very quickly. The atoms in a solid typically move back and forth 10^13 times per second. If you were to spend years worth of computational time on a single processor, you could manage simulating 10^7 vibrations, or 1 microsecond. It is also not easy to simulate this in parallel. We tend to be interested in dynamics on a human time scale of seconds or longer. This leaves a time scale gap of about 6 orders of magnitude between what we can simulate and what we are most often interested in simulating. Eon attempts to bridge this gap.

The deposition and growth of thin films (that we have been simulating) is quite important for many practical applications. Control of deposition and diffusion during the semiconductor manufacturing process, for example, is increasingly important as chips become smaller. But we're not out to make a better chip; our goal is to start by developing a basic understanding of how atoms move on surfaces. This kind of understanding is needed to be able to make new materials from the ground, or the atomic scale, upward. Most materials engineering is currently done by trial and error, but as computers and algorithms improve, we are moving towards a more intelligent design.

Nanoparticle catalysts are a neat idea that we are working towards being able to simulate with eon. Catalysts are used to make chemical reactions happen faster, without being consumed. An example of an everyday catalyst is the platinum (Pt) metal in the catalytic converter in a car. This is a can in the exhaust system filled with some beads which are coated in the precious Pt metal. The Pt is present to speed up the conversion of carbon monoxide (CO) to carbon dioxide (CO2), and generally to reduce any nasties left by incomplete combustion. It is only the surface of the Pt metal that speeds up these reactions, so it is spread as a very thin layer onto the supporting beads (over 100m^2/g). But over time, these metal atoms move along the surface, and ball up. At this point, you can fail your emissions test and will need to get a new one, with the metal thinly spread again. This is exactly the kind of atomic scale dynamics over long times that we can understand with eon. Even more interesting than current catalysts, is the recent discovery that nanoscale metal clusters can make extremely efficient catalysts. The major challenge that needs to be overcome, before they can be used, is to prevent the little clusters from balling up into larger, less efficient clusters. Developments in this area will likely change the nature of modern catalysis.

So, yes, we are looking at idealized systems, and the algorithms are in both in their infancy and under development. But the problem we are working on is a hard and fundamental one. Progress in this area will be an important step towards the goal of being able to simulate how atoms move on a human time scale.

Thanks for all your help!

[jasong]

Originally posted by graeme
The deposition and growth of thin films (that we have been simulating) is quite important for many practical applications. Control of deposition and diffusion during the semiconductor manufacturing process, for example, is increasingly important as chips become smaller.
I suspected this type of thing could be accomplished with your program, now that it's a confirmed fact, I'm back online with you guys.

[CaptainMooseInc]

Not to trivialize the eOn project but this seems like it may be a good analogy...

Basically eOn is the DHEP (Distributed Hardware Evolution Project) on steroids???

DHEP looks for ways to make circuits better and more efficient.

eOn looks for ways to make atoms play nice together for a longer period of time by making the way catalysts are constructed more efficient???

So doesn't that mean eOn could be considered an offshoot of DHEP with a different set of "perfection" in mind???

I know it's not affiliated with DHEP at all, but it's kinda a new sub-category of Science projects that both seem to fit under.

-Jeff

[graeme]

Although I'm not familiar with the Distributed Hardware Evolution Project, I think that there are a couple of aspects of eon which set it apart from other distributed computing projects. One is that everyone is working on the same calculation at the same time. This is because eon is trying to simulate how a system evolves in time. The clients discover events which can take place, and calculate their likelihood. After about 10 minutes, the server chooses one process, and the system moves forward in (simulated) time. At this point, any information about how the system could move from the old state is irrelevant. The clients must do the calculations in step with one another.

In chemisty and material science, we are always trying to scale up atomic simulations. Making systems larger is fairly easy to do in parallel. The techniques basically involve putting different regions of the system on different processors. The tricky part is figuring out how to deal with the edges of the regions. Time, on the other hand, is intrinsically harder to parallelize. If you want to run a dynamics simulation 10 times as long, you can't just divide up time and put it on 10 processors. You can make the first processor simulate the first 1/10th of the time interval, but you can't start the second one until you know the final state of the first one. This is what makes the time scale problem so challenging.