Three-Dimensional Spatial Projections of Molecular Dynamics Simulation Data

• Activity Number: 416
• Type: Contributed
• Date/Time: Tuesday, August 5, 2014 : 2:00 PM to 3:50 PM
• Sponsor: Section on Physical and Engineering Sciences
• Author(s): Geoffrey Peterson*+
• Companies: North Carolina State University
• Keywords: Spatial ; Bayesian ; Material Science

Abstract

Molecular dynamics simulations model the movement of atoms in materials, allowing for the measurement of structural defects within solids as it changes under various experimental inputs. The three-dimensional spatial distribution of all of the particles is reduced to a com- putationally manageable set of summary responses in cubic subsections of the solid, called voxels. The response is regressed against functional expansions of the experimental input, leading to coefficients that are projected into a spatially-independent space. When predicting defects under changing tilt tension angles, a conditional autoregressive model performs well under cross-validation, providing high predictive coverage and low predictive mean-squared error.