This talk will begin with an overview of laboratory astrophysics at high energy density, with attention to various aspects that are computationally challenging. From this I will transition into a discussion more focused on radiation hydrodynamics. For the last four years, I have managed a project in which, among other things, we produced a radiation hydrodynamic code for modeling high-energy-density experiments. For one who is not a numericist, this has been an informative experience. In this presentation, I will share some of the challenges, as seen through my eyes.
The specific context of our work is as follows. We had the good fortune to begin with a very well established Space Weather code, BATSRUS. We extended this code to have a set of features we label CRASH. CRASH is a block-adaptive, Eulerian code that scales well to about 1,000 processors. It includes hydrodynamics, radiation transport via (flux limited) multigroup diffusion, electron physics and electron heat transport via flux-limited diffusion, material identification via interface tracking, dynamic adaptive mesh refinement, and tabular equations of state and opacities. CRASH also has a laser package that traces rays in 3D and deposits energy in 3D or 2D. We are now using the code to model a number of high-energy-density experiments. I will summarize the code and some applications.
Having been a close witness of the process of creating this code has left several impressions. Firstly, “knowing the equations” though necessary is an almost negligible aspect of having a viable code. One is constantly subject to the tradeoff between fidelity and feasibility. One has no choice but to compromise the solution, at least in some regimes, and every specific method has its limitations. I will discuss some examples and offer some commentary.
One comes face to face with issues of fidelity when one undertakes uncertainty quantification. What has struck me about this is that serious uncertainty quantification is expensive and invasive. One cannot hope to succeed by making this an “add-on” activity. I will offer some thoughts about this issue.
I will briefly comment on two other challenging aspects of such computational work. There are sociological challenges. Code developers, being people, tend to like methods whose behavior matches their personalities. Second, is deeply ironic that rules once intended to protect US leadership in certain areas now impede the ability of the US to keep up and to train the next generation.
This work is funded by the Predictive Sciences Academic Alliances Program in NNSA-ASC via grant DEFC52- 08NA28616.