Methods in computational materials science have now evolved to have sufficient fidelity at different length and time scales that simulations results can often be directly compared with experiments and contribute to materials engineering. Moreover, in many cases a simulation at a given length and time scale receives information from experiments and simulations at other length and time scales that guide the form of the model and imply values of model parameters. The inherent data synthesis in this process influences the fidelity of the resulting model. For materials simulations to provide integral support of materials design and qualification processes, it is important to understand the implications of this coupling of scales in order to make sufficiently accurate predictions with quantified uncertainty. This symposium will focus on four critical aspects of multiscale simulation.
Verification (V) addresses the issue of the correct discretization and implementation of the algorithms and equations in a code. Approaches to verification include ensuring that code delivers intended functionality and quantifying numerical and discretization error by comparing to known solutions and fundamental physical laws. Both qualitative and quantitative verification methods are of interest.
Validation (V) is the assessment of the ability of a code or specific simulation to describe salient aspects of physical behavior under study. It primarily involves comparison between simulation results and experiments but may also involve comparison to accepted benchmark solutions and higher fidelity simulations.
Sensitivity analysis (SA) is the determination of the relative importance of different mechanisms and model parameters on the predictions of simulations. While methods for sensitivity on one parameter are well-established, there is significant interest in multivariate sensitivity analysis.
Uncertainty quantification (UQ) is the estimation of prediction uncertainty due to factors such as model-form error, information transfer between models and insufficient experimental data for both simulations at a single scale and especially challenges associated with the coupling of models at multiple scales. There is significant interest in both aleatoric and epistemic forms of uncertainty.
Contributions are solicited for V&V, SA and UQ within a single computational method, and for applications linking multiple methods between disparate length and time scales. Both advances in methodology and specific case studies focused on V&V, SA and UQ are of interest.