Modeling and Control of State-Affine Probabilistic Systems for Atomic-Scale Dynamics

Abstract

Public reporting burden for this collection of information is estimated to average 1 hour per response, including the time for reviewng instructions, searching data sources. gatherng and maintaining the data needed. and completing and reviewing the collection of information Send comments regarding this burden estimate or any other aspect of this collection of information, induding suggestiors for reducing this burden to Washington Headquarters Service, Directorate for Information Operations and Reports, 1215 Jefferson Davis Highway, Suite 1204, rlhington, VA 22202-4302 Sf. WORK UNIT NUMBER PERFORMING ORGANIZATION NAME(S) AND ADDRESS(ES) 8. PERFORMING ORGANIZATION Georgia Institute of Technology REPORT NUMBERAtlanta, GA 30332 AFRL-SR-AR-R-07-0 4 2 8 SPONSORING/MONITORING AGENCY NAME(S) AND ADDRESS(ES) 10, SPONSORIMONITOR'S ACRONYM(S)AFOSR SUPPLEMENTARY NOTES14. ABSTRACT Under this research grant, the framework and tools were developed for model reduction of atomic-scale many body systems. The state-affine mathematical structure of the original and reduced order models enabled the implementation of control through dynamic programming and estimation through an extended Kalman filter, The state of the high-dimensional stochastic system is first quantified using a high-dimensional pair correlation function. This state is then reduced using linear and nonlinear principal component analysis and is discretized using self-organizing maps. To create the dynamic model, a cell map is constructed using short simulations to quantify input-dependent transitions between the discrete states. The error associated with the model reduction was quantified and analyzed, and a method for predicting this error was proposed. Specific applications in materials processing were considered, which motivated and guided the development of the model reduction framework and tools. An existing model of gallium arsenide deposition was used to demonstrate the model reduction framework. A second modeling study in the molecular architecture of hyperbranched polymers was performed, and enabled a comparison of common themes and system specific features between the two different applications. SUBJECT TERMSMonte Carlo simulation, model reduction, gallium arsenide, hyperbranched polymers ABSTRACTUnder this research grant, the framework and tools were developed for model reduction of atomic-scale many body systems. The state-affine mathematical structure of the original and reduced-order models enabled the implementation of control through dynamic programming and estimation through an extended Kalman filter. In the framework developed here, the state of the high-dimensional stochastic system is first quantified using a high-dimensional pair correlation function. This state is then reduced using linear and nonlinear principal component analysis and is discretized using self-organizing maps. To create the dynamic model, a cell map is constructed using short simulations to quantify input-dependent transitions between the discrete states. The error associa...