Session C53: Massively Parallel Simulations of Chemical, Materials and Biological Systems

First-principles temperature-dependent phonons and elastic constants

Calculations of thermodynamic properties of materials from first-principles are critical for equation of state and materials strength modeling. ~Here we present the thermodynamic properties of a select set of metals based on density functional theory. In particular, we present elastic constants and lattice dynamics for body-centered cubic metals obtained from first-principles molecular dynamics and a self-consistent phonon approach. In order to calculate the thermodynamic properties, we make use of fluctuation formulas associated with the canonical ensemble form of \textit{ab initio} molecular dynamics (AIMD). This procedure is efficient and takes into account the anharmonic contributions to the equilibrium thermodynamic properties. In the self-consistent lattice dynamics approach, the phonon dispersions at finite temperature are determined from small displacements along normal modes associated with the chosen temperature.~ This method provides an efficient and accurate technique for phonon spectrum and finite-temperature acoustic sound speeds and elastic constants. We found that both methods provide consistent results for the temperature- and pressure-dependent elastic moduli. The AIMD include full anharmonicity but suffers from statistical errors of the order of 5{\%}. The self-consistent phonon method, on the other hand, has less statistical uncertainty but does not explicitly account for electron-phonon coupling. At ambient pressure, our calculations (both methods) agree quite well with experimental data.\\ \\This work performed under the auspices of the U.S. DOE by LLNL under Contract DE-AC52-07NA27344.   

The ReaxFF method - new applications and developments

The ReaxFF method provides a highly transferable simulation method for atomistic scale simulations on chemical reactions at the nanosecond and nanometer scale. It combines concepts of bond-order based potentials with a polarizable charge distribution. Since it initial development for hydrocarbons in 2001, we have found that this concept is transferable to applications to elements all across the periodic table, including all first row elements, metals, ceramics and ionic materials. For all these elements and associated materials we have demonstrated that ReaxFF can reproduce quantum mechanics-based structures, reaction energies and reaction barriers with reasonable accuracy, enabling the method to predict reaction kinetics in complicated, multi-material environments at a relatively modest computational expense. This presentation will describe the current concepts of the ReaxFF method, the current status of the various ReaxFF codes, including parallel implementations and recently developed hybrid Grand Canonical Monte Carlo options - which significantly increase its application areas. Also, we will present and overview of recent applications to a range of materials of increasing complexity, with a focus on applications to combustion, biomaterials, batteries, tribology and catalysis.   

Discovering chemistry with an ab initio nanoreactor

Traditional approaches for modeling chemical reaction networks such as those involved in combustion have focused on identifying individual reactions and using theoretical approaches to explore the underlying mechanisms. Recent advances involving graphical processing units (GPUs), commodity products developed for the videogaming industry, have made it possible to consider a distinct approach wherein one attempts to discover chemical reactions and mechanisms. We provide a brief summary of these developments and then discuss the concept behind the ``ab initio nanoreactor'' which explores the space of possible chemical reactions and molecular species for a given stoichiometry. The nanoreactor concept is exemplified with an example to the Urey-Miller reaction network which has been previously advanced as a potential model for prebiotic chemistry. We briefly discuss some of the future directions envisioned for the development of this nanoreactor concept.   

Multi-petaflop/s quantum and reactive molecular dynamics simulations

We have developed a divide-conquer-recombine algorithmic framework for large quantum molecular dynamics (QMD) and reactive molecular dynamics (RMD) simulations. The algorithms have achieved parallel efficiency over 0.98 on 786,432 IBM Blue Gene/Q processors for 39.8 trillion electronic degrees-of-freedom QMD in the framework of density functional theory and 67.6 billion-atom RMD. We will discuss several applications including (1) 16,616-atom QMD simulation of rapid hydrogen production from water using metallic alloy nanoparticles, (2) 6,400-atom nonadiabatic QMD simulation of exciton dynamics for efficient solar cells, and (3) 112 million-atom RMD simulation of nanocarbon synthesis by high temperature oxidation of SiC nanoparticles.