Enabling Microscopic Simulators to Perform System Level Tasks: A System-Identification Based, "Closure-on-Demand" Toolkit for Multiscale Simulation Stability/Bifurcation Analysis, Optimization and Control
PRINCETON UNIV NJ
Pagination or Media Count:
This project developed computational and mathematical tools, based extensively on elements of system theory, that enable microscopicistochastic simulators to perform system-level tasks analysis, design, control, optimization. In current engineering modeling we are often faced with situations where the system model is given at an atomistic I stochastic I agent based level, while the modeling tasks simulation, controller design, optimization must be performed at a much larger, macroscopic in space and time level. The objective was to construct a bridge between existing and future microscopic simulation codes kMC, MD, MC, BD, LB etc. and traditional, continuum numerical analysis. To accomplish this, we traded function evaluations in the continuum computations for appropriately initialized bursts of microscopic simulations, executed over short space and time intervals, followed by post-processing based on system-identification techniques. Separation of time scales is another important ingredient of the bridge between micro-simulation and macro-modeling. We stress controller design and optimization tasks this year. We also explored how this model on demand approach can evolve into experimental design protocols. Several examples, from small molecule folding to complex fluid rheology and computational materials science were pursued in order to validate the approach.
- Numerical Mathematics
- Computer Programming and Software