W911NF-17-S-0002: Stochastic Modeling and Multiscale Propagation of Model-Form Uncertainties Arising in Molecular Dynamics Simulations
Abstract
Molecular dynamics simulations are a cornerstone in computational solid mechanics where they are commonly employed to explore conformational space, predict macroscopic physical properties, and understand subscale physical mechanisms for materials design and discovery. In practice, numerical outcomes can be strongly aÀected by uncertainties arising from model parameter calibration, based on first-principles calculations or physical experiments, and by modeling errors in the definition or selection of the (e.g., coarse-grained) force fields. While the consideration of the former type of parametric uncertainties is fairly classical in the literature, the rigorous treatment of model-form uncertainties mostly remains an open question due to numerous challenges pertaining, for instance, to the functional structure of the problem. The aim of this research eÀort is to fill in this gap and to develop a new stochastic modeling framework enabling the systematic description and integration of model uncertainties in molecular dynamics simulations. The formulation will rely on a stochastic reduced-order model, formulated using a random matrix representation on the Stiefel manifold. Model inference will be explored by means of projection techniques preserving local and global structures, and optimal parameterization will be proposed under constraints related to identifiability based on computational or physical experiments. Propagation of model uncertainties will be investigated in graphene-based systems, with a particular emphasis on microscopic processes and macroscopic parameters including eÀective mechanical and fracture properties. Methodologies adapted to the proposed representations will finally be developed to perform sensitivity analysis and quantify the influence of model inadequacy on multiscale predictions. By endowing atomistic simulation results with proper measures of confidence with respect to modeling errors, this work is expected to advance the simulation-based predictive capabilities at the Army Research O
Document Details
- Document Type
- DoD Grant Award
- Publication Date
- Apr 26, 2023
- Source ID
- W911NF2310125
Entities
People
- Johann Guilleminot
Organizations
- Army Contracting Command
- Duke University
- United States Army