UQ in Atomistic Simulations

Uncertainty Quantification in Atomistic-to-Continuum Simulations

Given the wide range of applications where microscale phenomena control the macroscale operation of engineering devices (e.g. nanoscale surface electrochemical processes that are at the core of fuel cells and batteries), there is a strong need for multiscale simulation approaches. Multiscale simulations resolve the relevant phenomena at the scales in which they occur and share information across the range of time and length scales to arrive at a full system simulation. As such, multiscale simulations form an essential tool to get a detailed understanding of and enable the science-based design of multiscale devices.

The coupling across the wide range of time and length scales does introduce uncertainties into the simulation, primarily due to differences in representations on different scales and approximations needed when transferring information across the interfaces between different scales. Sandia computational researchers are developing methods to assess and quantify such uncertainties in order to assess the overall predictive fidelity of multiscale simulations. Of particular interest is the quantification of uncertainties in coupled atomistic-to-continuum simulations. An approach has been developed to quantify uncertainties due to finite sampling on the atomistic level (e.g., in molecular dynamics simulations) and propagate this uncertainty, along with uncertainties in model inputs throughout the multiscale simulation.