What options are available for pursuing a PhD in computational physics?

[Jakeness]

Undergrad here. Does anyone on this forum have experience working in a computational physics research group at the graduate level? I noticed Carnegie Mellon has "computational physics" as a research area for its physics PhD students. It sounds like it would be right up my alley but it's unclear to me what exactly it means. I have a few questions:

1) I thought computational physics was a set of tools to be applied to another area of physics, not really a field in and of itself. What exactly would you learn doing a PhD in computational physics?

2) Would Carnegie Mellon be a good choice for this research area? What schools are well established in this field?

Thanks.

 

[DrClaude]

I can't answer your second question, but I can comment on the first one. Saying that computational physics is a "set of tools to be applied to another area of physics" is like saying that mathematics is "set of tools to be applied to another area of physics." It is not wrong to say they are tools, but some people need to build such tools! Computational physics is concerned by the inner workings of algorithms and computations, in order to build better tools. Hot topics right now include things like developing methods that work efficiently on GPU clusters and multi-scale simulations.

To go in the field of computational physics, you have to have more than a passing interest in computers, but be ready to dive into more basics aspects of how computers work and how to get the most of them to solve physics problems.

 

[Jakeness]

I can't answer your second question, but I can comment on the first one. Saying that computational physics is a "set of tools to be applied to another area of physics" is like saying that mathematics is "set of tools to be applied to another area of physics." It is not wrong to say they are tools, but some people need to build such tools!

Good point.

Computational physics is concerned by the inner workings of algorithms and computations, in order to build better tools. Hot topics right now include things like developing methods that work efficiently on GPU clusters and multi-scale simulations.

To go in the field of computational physics, you have to have more than a passing interest in computers, but be ready to dive into more basics aspects of how computers work and how to get the most of them to solve physics problems.

Then this is definitely something I'll want to look into. If you don't mind me asking, are there any textbooks on the subject that you would recommend? (Assuming prior reasonable competence with parallel and GPU computing.)

 

[Arsenic&Lace]

I have worked on a graduate level project (according to my advisor anyway) in computational biophysics but am an undergraduate.

Disclaimer aside, here are my answers:
1). Some problems in physics are more about computation than theory, although they still interface with experiment in more or less the same manner. In my case, for computational biophysics, while there are theoretical biophysicists, most problems of characterizing a protein don't necessarily call for new theory; it's really about developing algorithms to produce accurate data at speed and then analyze said data.

2). Computational physics is in the same category as theoretical or experimental physics with overlap between the two. In other words, asking "Is school X good at computational physics?" is like asking if school X is good at experimental physics. The actual field of physics is important! For instance, UIUC is still probably the best school in the nation, to my knolwedge, for computational condensed matter physics. School Y might be better than UIUC at computational astrophysics.

Some schools do seem to emphasize computation though, and the only two which really spring to mind immediately are U of Washington and UIUC although I'm sure there are others.

 

[ZombieFeynman]

In my experience, computational science is a tool that is used, just like an experimental apparatus or mathematics, to attack different areas of physical problems. In my group, we do extensive quantum monte carlo calculations and large-scale exact diagonalization, as well as some DFT and DMFT to study strongly correlated condensed matter systems.

 

[babagoslow]

Computational physics is a huge field that can range from statistical analysis (for instance, time series of the stock markets), to agent based models (traffic control, economic simulations, and other complex systems), to doing the heavy duty work in various theoretical branches of the natural sciences (molecular dynamics, DFT, Monte Carlo, Navier-Stokes etc). I couldn't tell you which particular areas CMU are good at, but my suggestion to you is that you should not be too seduced by the name of the school. It's the research group that's important -- many excellent groups can be found in relatively low-ranking universities.

It would be better for you to decide exactly where your interests lie, find a good group, and then work towards entering that university, in that order.

A PhD in computational physics will probably ensure you will be able to find some kind of employment, provided your research work requires you to develop skills in some of the more mainstream programming languages like C++ or Python. The vast majority of PhDs do not end up as tenured professors, so it might be a good thing also to find out exactly what skills (in terms of programming languages, architectures like MPI, etc) you can expect to pick up during your PhD. That will serve you in good stead post PhD. A quick chat with a professor or PI should give you a few solid ideas.

 

[AlephZero]

Mod note: The post to which AlephZero responded is of dubious quality and has been deleted. Aleph's response is a good quality response and contains good information. I don't want to delete this response just because he responded to a deleted post. There will be a few more such seemingly out of place posts below.I would disagree with that, based on my own experience doing "computational engineering" in industry. Unless you have a good understanding of the the application, you won't get far in producing some useful (read, efficient) software. In other words, you need to know some physics, not necessarily at PhD level, but well enough to understand what the researchers are actually trying to do.

At least, that applies to the sort of computational physics DrClaude was talking about. There is a certain amount of "important trivia" like reformatting data sets and doing simple statistics that doesn't need much application knowledge, but that sort of work probably won't hold your interest for more than a few months. And even those tasks are more interesting if you understand WHY they are being done, rather than just being another (often fairly simple) programming puzzle that you have to solve.

 

[Vanadium 50]

Mod Note: Response to deleted post

I would also disagree with this. Computational physics is generally about solving problems numerically at very large scales: combustion, climate modeling, cosmology, complex materials, accelerator orbits, etc. Essentially none of these problems are limited by the computer science - they are limited by machine capacity and the domain science. Getting a PhD in computational physics is not the equivalent of a BS in CS hired by a professor.

 

[Jakeness]

Thanks for all the replies, they've been very helpful to me.

I couldn't tell you which particular areas CMU are good at, but my suggestion to you is that you should not be too seduced by the name of the school. It's the research group that's important -- many excellent groups can be found in relatively low-ranking universities.

It would be better for you to decide exactly where your interests lie, find a good group, and then work towards entering that university, in that order.

Admittedly I singled them out partially because of brand bias but mostly because they are the first result when you Google search the phrase "computational physics phd." I'm just beginning the process of compiling a list of schools I want to apply to.

A PhD in computational physics will probably ensure you will be able to find some kind of employment, provided your research work requires you to develop skills in some of the more mainstream programming languages like C++ or Python. The vast majority of PhDs do not end up as tenured professors, so it might be a good thing also to find out exactly what skills (in terms of programming languages, architectures like MPI, etc) you can expect to pick up during your PhD. That will serve you in good stead post PhD. A quick chat with a professor or PI should give you a few solid ideas.

I would prefer to go into industry post-graduation rather than academia, so I'll definitely keep this in mind.Mod note: Deleted reference to a deleted post.

I understand the point you're making, but I don't really see how it translates into a long-term plan. Sure, this may be possible to do in undergrad, but how many scientific computing jobs are available in industry for someone with a bachelors in CS? I'm doing a minor in CS and teaching myself C++ and Python in my spare time. It seems easier to teach yourself programming to an acceptable level than to teach yourself climate modeling, cosmology, etc to an acceptable level.

Unless you have a good understanding of the the application, you won't get far in producing some useful (read, efficient) software. In other words, you need to know some physics, not necessarily at PhD level, but well enough to understand what the researchers are actually trying to do.

Would it be more pragmatic (with respect to finding employment in this field) to attempt a Master's degree?

Computational physics is generally about solving problems numerically at very large scales: combustion, climate modeling, cosmology, complex materials, accelerator orbits, etc. Essentially none of these problems are limited by the computer science - they are limited by machine capacity and the domain science. Getting a PhD in computational physics is not the equivalent of a BS in CS hired by a professor.

I found this comforting. I was worried, probably irrationally, that I would be signing up to be a "code-monkey."

 

[babagoslow]

Jakeness, I was glad to read your posts and see your thinking. I think you are sensible and you'll definitely avoid some of the rather horrible fates that await some PhD graduates.

One tip I can give you is to go to Google Scholar, and in the search box, type in, for example "label:complex networks" and click Search Authors, you get a list of the top people in the chosen field. To get useful results (it's rather unlikely that professors with upwards of 50-100k citations are still accepting students) you will probably have to use a label that is more specific than just "computational physics". Read their papers to get a flavour of what these guys do. There are usually a few free copies of their papers online, or your school may give you free access to journals (most of the bigger universities do).

I would advise you to ignore those people who ask you to study CS. You'd be a fool to do a PhD that would require a CS background, but at the same time you have value as a physics graduate. It's quite telling that many companies in the financial industry prefer to hire physicists and mathematicians who can code, as opposed to pure CS grads who are good at coding but lack exposure to the mathematics.

Don't worry about being a code monkey or not. If your future boss needed a genuine CS expert, he'd get one already, instead of you. Professors don't take in students blindly, they know what they're getting - or at least they will be well aware about what your background is and what you can offer.

 

[BruceW]

Undergrad here. Does anyone on this forum have experience working in a computational physics research group at the graduate level? I noticed Carnegie Mellon has "computational physics" as a research area for its physics PhD students. It sounds like it would be right up my alley but it's unclear to me what exactly it means. I have a few questions:

This is very rare to find a "computational physics" research group, as far as I am aware. Usually, physics research groups are made up of people with the same physics interest. i.e. there might be a "condensed matter" physics research group, which could use a lot of computational tools. But still, they all have an interest in condensed matter in common. (which is why they form a group).

1) I thought computational physics was a set of tools to be applied to another area of physics, not really a field in and of itself. What exactly would you learn doing a PhD in computational physics?

well, I just looked at the Carnegie Mellon website, and you're right, they do have a "computational physics" group. But if you look at the webpage, each of the members of the group are from a different area of physics. i.e. they are each part of another physics group which is specific to the physics that they are researching. So it seems that they also formed a "computational physics" group so that they could share knowledge and resources related to the computational aspect.

2) Would Carnegie Mellon be a good choice for this research area? What schools are well established in this field?

Thanks.

The very fact that they have formed a "computational physics" research group is probably a good indicator that they have several physics research groups which rely heavily on computational tools. So yes, most likely. But you should keep in mind that the most important thing is the actual physics group which they are in. The "Computational physics" group will be a secondary thing, not their primary group.

 

[Arsenic&Lace]

I think there are some classes of problems which can be thought of strictly as computational physics, though. The underlying physics of protein folding is not controversial to my knowledge (classical/statistical mechanics mainly). The question is, how do I solve the problem when the computational requirements are enormous (e.g. for 200 reside proteins which extremely high dimensional phase spaces)?

Some groups work on DFT software and apply it to a wide variety of systems, from proteins (small ones, granted) to crystal growth to other topics. Their interest is a computational tool which can then be used to study a variety of problems, rather than a specific problem.

 

[ZombieFeynman]

Mod note again: Yet another link to a deleted post has been deleted.

I've rarely seen a series of posts more filled with exaggerated generalizations or with as many sweeping innaccuracies as those which you have spewed forth in this thread. But I do agree with the fact that you should only pursue graduate education if you truly love the subject.

 

[SteamKing]

6 years at graduate school in CS? That seems awfully slow, even for computer science. Don't MDs complete their undergrad & graduate education with the residency requirement in less than 13 years?

 

[D H]

Jakeness: You might want to consider computational science as an option to a program that specializes in computational physics. A number of colleges either offer degrees in computational sciences or have a multidisciplinary computational sciences center (in which case you would get a PhD in physics).

Here's a link to some computational sciences programs maintained by SIAM: http://www.siam.org/students/resources/cse_programs.php. It's obviously not up-to-date because the Carnegie-Mellon program isn't listed.

Computation science, along with specific offshoots such as computational physics, is a fairly new specialty. Right now it looks a bit Wild West-like. There isn't anything close to a standard curriculum, for example.

 

[tolove]

6 years at graduate school in CS? That seems awfully slow, even for computer science. Don't MDs complete their undergrad & graduate education with the residency requirement in less than 13 years?

6 years, believe it or not, is a little under the average time (page 6):
http://www.aps.org/programs/educati...-Physics-Joint-APS-AAPT-Task-Force-Report.pdf

edit: for physics

 

[Jakeness]

(Apologies if this thread is too old and this bump unwanted, but I figure the relatively high amount of views this thread received means many others are interested in the topic.)

I wanted to gain more hands-on experience in computational methods than my curriculum gives me. My search led me to these "lecture notes" (essentially textbooks). I'm leaving these links here with hopes that those of you with more experience will comment on them, and/or so others in a situation similar to mine will benefit from them.

Computational Physics With Python - Eric Ayars (Pretty basic; assumes no prior knowledge of Python.)
Computational Physics - Morten Hjorth-Jensen (More advanced; uses mostly C++ as well as some Fortran and Python; explores topics from above book in greater detail; unfortunately the last chapters are incomplete.)
And, Statistical Mechanics: Algorithms and Computations (Online Coursera course; uses Python; this, the second lecture in particular, is what originally got me interested in statistical and computational physics.)

Jakeness, I was glad to read your posts and see your thinking. I think you are sensible and you'll definitely avoid some of the rather horrible fates that await some PhD graduates.

I blushed.

One tip I can give you is to go to Google Scholar, and in the search box, type in, for example "label:complex networks" and click Search Authors, you get a list of the top people in the chosen field. To get useful results (it's rather unlikely that professors with upwards of 50-100k citations are still accepting students) you will probably have to use a label that is more specific than just "computational physics". Read their papers to get a flavour of what these guys do. There are usually a few free copies of their papers online, or your school may give you free access to journals (most of the bigger universities do).

This sounds like an excellent little method, I'm surprised that I have not heard of it before.

I would advise you to ignore those people who ask you to study CS. You'd be a fool to do a PhD that would require a CS background, but at the same time you have value as a physics graduate. It's quite telling that many companies in the financial industry prefer to hire physicists and mathematicians who can code, as opposed to pure CS grads who are good at coding but lack exposure to the mathematics.

My plan C – if for some reason, A) I am rejected from everywhere I apply and B) can't find a job by graduation – is staying an extra year at my undergrad institution to receive a second bachelors in CS. My [naive] guess is that dual bachelor's in physics and CS would give me much greater advantage than either alone for software development in finance or engineering.

But I have no intention to drop physics for CS, or anything for that matter.

well, I just looked at the Carnegie Mellon website, and you're right, they do have a "computational physics" group. But if you look at the webpage, each of the members of the group are from a different area of physics. i.e. they are each part of another physics group which is specific to the physics that they are researching. So it seems that they also formed a "computational physics" group so that they could share knowledge and resources related to the computational aspect.
...
But you should keep in mind that the most important thing is the actual physics group which they are in. The "Computational physics" group will be a secondary thing, not their primary group.

In hindsight, I'm not sure why I overlooked this. Thank you for pointing this out, I'll most certainly keep this in mind.

Jakeness: You might want to consider computational science as an option to a program that specializes in computational physics. A number of colleges either offer degrees in computational sciences or have a multidisciplinary computational sciences center (in which case you would get a PhD in physics).

Here's a link to some computational sciences programs maintained by SIAM: http://www.siam.org/students/resources/cse_programs.php. It's obviously not up-to-date because the Carnegie-Mellon program isn't listed.

Great idea. Thanks for posting that list, I thought my grad schools options were much, much more limited than that.

Computation science, along with specific offshoots such as computational physics, is a fairly new specialty. Right now it looks a bit Wild West-like. There isn't anything close to a standard curriculum, for example.

That's somewhat worrying, yet very intriguing.