Lawrence Livermore National Laboratory



John Moriarty, Randy Hood, and Lin Yang

(This project is a part of the DOE funded SciDAC-2 project: Hierarchical Petascale Simulation Framework for Stress Corrosion Cracking led by Priya Vashishta at USC)

Schematic representation of the proposed hybrid QMD-MD/MGPT simulation capability.

Schematic representation of the proposed hybrid QMD-MD/MGPT simulation capability. A real-time information feedback loop between the alternate QMD and MD/MGPT segments of the simulation will allow optimized and fully re-useable MGPT potentials to be obtained in the long time limit.


The 5-year goal of the SciDAC-2 (http://www.scidac.gov ) project is to provide a de novo hierachical petascale simulation framework to address the complex scientific and technological problem of corrosion-induced failure in vital industrial materials operating under stress loads. The goal of our work is to provide a key element of that multiscale framework, namely, the capability to perform ultra-scale atomistic simulations in complex materials with quantum-mechanical accuracy. For many materials, including complex d-electron transition-metal alloys of interest in our SciDAC-2 project, simulation methods based on density-functional-theory (DFT) quantum mechanics can provide the desired accuracy and predictive power of materials properties. First-principles DFT-based QMD simulations are inherently expensive, however, and the cost scales with the number of atoms as N^3. Even using large terascale computing platforms such as IBM BG/L, one can expect to treat only 100-1000 atoms for 1-10 ps of time in transition metals with direct QMD methods. To overcome this limitation, we will extend our existing work on developing advanced DFT-based interatomic potentials using model generalized pseudopotential theory (MGPT) to establish a direct real-time coupling to QMD simulations that can greatly increase the accuracy of the derived potentials. By developing this capability for small systems of 250-1000 atoms, one can dramatically extend the effective QMD time scale and harvest optimized temperature-dependent MGPT potentials in the long time limit, as shown schematically above in Fig. 1. In this process, there will be a real-time information feedback loop between the alternate QMD and MD/MGPT segments of the simulation, with both snapshot information on energies and forces and accumulated information on structure, total energy and pressure being used to achieve optimization of the potentials. The optimized potentials can then be re-used in ultra large (109-1012 atom) linear-scaling MD/MGPT atomistic simulations of materials properties, including the behavior of a crack tip under stress. Our initial goal will be to simulate one billion atoms at high temperature in prototype systems (e.g., Mo, TiAl) via MD/MGPT with comparable accuracy to a direct QMD simulation on 1000 atoms.


Maintained by   Randolph Q. Hood