Parallel computing for PhD in Computational Astrophysics

[simha]

Hi,

I wanted to know if someone who has a degree in parallel computing and related work experience will be able to get a PhD project in Computational Astrophysics? What other skill sets would the person require?

I have seen many PhD positions that accept people from a parallel computing background into Computational Biology. Does the same hold good for Astronomy as well?

Thanks in advance.

 

[zhermes]

Parallel computing is huge in astrophysics (more so than astronomy, per se), and becoming a bigger and bigger part of it. Essentially every large astrophysics institution has a group that works with parallel computing in cosmological, n-body, stellar evolution, markov chain monte carlo simulations, etc.
There should be many opportunities for such work, but usually institutions will lean towards people with strong physics backgrounds.

I'm curious as to the specific Computational Biology applications; do you know particular institutions and projects using parallel computing?

 

[twofish-quant]

I wanted to know if someone who has a degree in parallel computing and related work experience will be able to get a PhD project in Computational Astrophysics? What other skill sets would the person require?

Maybe. If you have the ability to do partial differential equations and undergraduate physics ability, then you have the basic skill sets to do computational astrophysics. Whether you can find a program to admit you is another question.

Let me know exactly what you are interested in, and I can give you some ideas.

I have seen many PhD positions that accept people from a parallel computing background into Computational Biology. Does the same hold good for Astronomy as well?

I think it could. If you have undergraduate physics coursework and practical HPC experience, it would make a very strong graduate school application.

Here is the basic text that will help you

https://www.amazon.com/dp/0521540623/?tag=pfamazon01-20

Parallel methods are much less used in astrophysics than in biophysics and there is a chapter in Castor that explains exactly why that is. The trouble is that do use monte carlo with radiation hydrodynamics you have to run a lot more simulations to get a convergent values, because astrophysical systems are tightly coupled. Now, if you are interested, I'd really like for you to stare at that chapter because 1) there may be a clever way of getting around that problems or 2) it *might* be possible that you just overwhelm the problem with throwing CPU. Or maybe parallel methods just won't work.

Also if you have work experience and are interested in working on astrophysics problems, then applying to astrophysics graduate schools may not be the best approach. Something that might be better is to either get a job at an HPC center or maybe apply to for a Ph.D. in high performance computing.

 

[twofish-quant]

Parallel computing is huge in astrophysics (more so than astronomy, per se), and becoming a bigger and bigger part of it. Essentially every large astrophysics institution has a group that works with parallel computing in cosmological, n-body, stellar evolution, markov chain monte carlo simulations, etc.

The other issue is that there are some fundamental things that "we do not know how to do". For example, there is not currently a "cookbook method" of solving GR equations or CFD equations with a large cluster of GPU's. There probably will be in ten years.

 

[Simfish]

The trouble is that do use monte carlo with radiation hydrodynamics you have to run a lot more simulations to get a convergent values, because astrophysical systems are tightly coupled

Hm that's interesting. Is that just relevant for radiation hydrodynamics or for the entire field too? (I'll try to get the book when I have some time). I'd expect biological systems to be more tightly coupled than astrophysical systems.

Edit: Just looked at the book - it doesn't seem to contain anything about parallel computing even in its numerical methods chapter.

==

Well, as one example, there was this astrophysics professor (at UMass) who made the news for using Playstation 3s for his parallel computing (since Playstation 3s are much cheaper than the alternatives).

 

[simha]

Many thanks for everyone's replies. It was really helpful. I have had some Physics during my Engineering, plus I did an Astronomy related project during my Masters.

There are many places that do Computational Biology. The MOSAIC group at ETH, Zurich for example. I find that Computational Biology has more funding than Computational Physics or Chemistry, plus its a relatively new field. I have also seen adverts that require parallel computing for simulating brain surgery. I give a link below.

http://www.scholarshipnet.info/postgraduate/uk-msc-and-phd-studentships-in-brain-surgery-simulation-high-performance-computing/

I would probably have a better chance at doing a PhD in HPC itself, but I want to do it in an application related area. PhDs in HPC mostly seem to concentrate only on the computer science part which I do not prefer much.

 

[twofish-quant]

Hm that's interesting. Is that just relevant for radiation hydrodynamics or for the entire field too? (I'll try to get the book when I have some time).

Much of astrophysics boils down to radiation hydrodynamics, but the section on why monte carlo works badly in radiation hydrodynamics applies to a lot of other problems. Of course, a lot of this will change wildly over the next few years.

I'd expect biological systems to be more tightly coupled than astrophysical systems.

Sometimes. The thing that you can do in biological systems is that you can run a simulation for one organism, change things, run it for another, and then run statistics. This is hard to do with the simulations that people have traditionally done in astrophysics.

Edit: Just looked at the book - it doesn't seem to contain anything about parallel computing even in its numerical methods chapter.

*BIG GRIN*

The cool thing about research is that you can't find the answer in the textbook. If you could, then it wouldn't be research.

So what you need to do is to read up on radiation hydrodynamics, read up on parallel computing, read up on anything that you think might be relevant, talk to people, try some things, and then write the chapter on parallel computing.

 

[Simfish]

Oh okay I see. Thanks for the reply!

Ah yes, that's actually a pretty good research idea.