Blue Waters Graduate Fellow: Sherwood Richers

10.31.16 -

The Blue Waters Graduate Fellowship was awarded to ten outstanding Ph.D. students in computational science. In this series we're featuring brief introductions to who they are and what they're trying to accomplish. This program serves to prepare the next generation of science researchers to solve the world’s problems. Follow along as we highlight these young researchers. Read more profiles here.

Tell me a little bit about yourself—where are you studying now, where did you do your undergrad, what was your major, etc.

I study high-energy astrophysics, and in particular I study core-collapse supernova explosions and neutron star mergers. Even more specifically, my primary focus is methods for treating the motion of neutrinos around these systems, called "neutrino transport." I studied Astronomy-Physics at the University of Virginia. The research I did during my undergrad that sent me along this path was simulating accretion disks with John Hawley, but I also worked in an experimental ultrafast laser lab, spent a summer researching Carbon Nanotubes at Rice University, and did some data analysis of radio astronomical observations at the Remeis-Bamberg Observatory in Germany.

Tell me about your research—what are you trying/hoping to accomplish? What made you want to pursue this topic?

My primary interest in neutrino transport has a couple of objectives. Let's look at core-collapse supernovae first. The big problem in this field is that observers watch stars explode on a daily basis, but when we put the most complete set of physics possible in the largest simulations running on supercomputers (like Blue Waters), they don't consistently explode. Something is missing, and that something might be a proper treatment of neutrino transport. The equations describing neutrino transport are notoriously difficult to simulate, so they have to be heavily approximated, but I am trying to remove as much of the approximation as I can. The vast majority of the energy released from these explosions is in the form of neutrinos, and even a small correction could be the difference between an explosion and a dud. In the case of neutron star mergers (which have yet to be conclusively observed), the effects of neutrinos are expected to dramatically affect which kinds of atoms are released, and considering that mergers can release vast amounts of heavy elements (like gold, platinum, etc), properly treating neutrinos is very important for determining where all of these heavy elements we use every day come from. It also turns out that the neutrinos can change how these events look, and if we can do the simulation well enough, we can help observers know what to look for. Since all of these interesting things depend on an accurate treatment of neutrinos, I can attack both the problems with core-collapse supernovae and mergers with a single new neutrino transport method.

So what was your process like getting involved with Blue Waters? What made you want to apply for this fellowship?

As part of my previous research, I have run big simulations and data analysis/visualization on several supercomputers, including Blue Waters. Given that Blue Waters has been an invaluable tool for my group's research in the past, it was only sensible to apply to the fellowship to get additional support, in the form of funding, computer time, and personal assistance.

How will the ability to use Blue Waters impact your research?

There are two problems in simulations that make supercomputers like Blue Waters necessary. The first is that simulations take so long that if I don't get thousands of processors to work on them simultaneously, I won't live to see the results. The large 3D simulations I have done in the past take on the order of a month on upwards of 10,000 processors, meaning it would take about 1,000 years on a single processor. The second is that the simulations are so big that they don't fit on a regular computer. A single snapshot of the simulations I have run in the past requires several terabytes of RAM, and the addition of the neutrino transport method I am working on will only increase that requirement. So, I really need access to a world-class supercomputer to deal with these simulations.

Would you have been able to do this kind of research on any other machine? Why or why not?

There are a very small number of machines that would make the research possible. However, since I and my group have an extended history with Blue Waters, it makes Blue Waters much easier to use. The large simulation code I use has already been ported to and optimized on Blue Waters. In addition, the large allocations available for Blue Waters are not possible on other systems. The hardware setup is also optimal. The radiation transport method I am developing will increase the total memory requirement. Blue Waters has a very large amount of memory on each node, meaning that I will be able to run on fewer nodes than I would be able to on other computers. Fewer nodes means less time communicating information between nodes and better scalability. One of my future aims is also to accelerate the neutrino transport component of the simulations using GPUs, so even if I am able to run on a couple of other computers, Blue Waters has every resource I would want to help me move the science forward, and is the ideal system for my work.

Is there anything else you’d like people to know about your research/fellowship?

I am extremely grateful for the fellowship. The academic freedom it affords, the networking opportunities, and the computing resources are all amazing benefits that really make it possible for me to do such exciting work.

National Science Foundation

Blue Waters is supported by the National Science Foundation through awards ACI-0725070 and ACI-1238993.