NERSC Initiative for Scientific Exploration (NISE) 2011 Awards
High performance simulation of reactive transport processes at the pore scale
Brian Van Straalen, Lawrence Berkeley National Lab
Associated NERSC Project: APDEC: Applied Differential Equations Center (m1055)
Principal Investigator: Phillip Colella, Lawrence Berkeley National Lab
|NISE Award:||1,200,000 Hours|
|Award Date:||June 2011|
We have combined high performance CFD capabilities in Chombo--incompressible flow plus advection-diffusion in complex geometry--with the complex geochemistry package in CrunchFlow to model reactive transport processes in resolved pore space. This work is being funded by a recent ARRA supplement to the SciDAC project, APDEC, upon which our current allocation, MP111, is based. In this new work we are partnering with the Energy Frontier Research Center for Nanoscale Control of Geologic Carbon at LBL to model emergent processes in geologic subsurface porous medium where the pore space is resolved. The goal is to achieve mesh spacing resolved down to 1 micron, which is the resolution of the image data obtained from validating experiments. For a 1 cm long capillary tube tightly packed with spheres and an aspect ratio of approximately 18.5:1 we have been able to achieve 1024x64x64, which yields just under 10 micron resolution. These runs have been performed on a 96 core cluster. In order to attain 1 micron resolution our problem size will increase by 8-fold. As intermediate targets we will attempt 2048x128x128 with 1024 processors on Franklin, and 4096x256x256 with 8192 processors on Franklin. Ultimately, we will attempt to run the finest resolution of 8192x512x512 with over 65,000 processors on Hopper. In each of these runs we are seeking steady-state solutions, first by running the CFD velocity code to steady-state and then advancing the transport to steady-state while holding the velocity fixed. This is permissible since we are not changing the geometry of the pore space yet as a result of dissolution or precipitation.
Also, we have found that the transport code completes 3x more timesteps per minute than the velocity code. The velocity can take on the order of 1000 timesteps to achieve steady-state; the transport usually takes another 1000-2000 steps depending on the process--whether dissolution or precipitation--and the rate constants. Furthermore, there are algorithmic issues that will result from higher resolution of these tightly packed complex geometries that we will have to work out during the scaling process for our elliptic solvers.
This work is important to the modeling needs of the Center for Nanoscale Control of Geologic Carbon, one the DOE Energy Frontier Research Centers, which is focused on modeling the fate of super critical CO2 injected into the earth as it pertains to carbon sequestration. The ability to model reactive transport processes at the pore scale will allow for better parameterizations of field scale models which can in turn predict the fate of CO2 in the subsurface over the long periods of time relevant to carbon sequestration.