Library Subscription: Guest
Begell Digital Portal Begell Digital Library eBooks Journals References & Proceedings Research Collections
International Journal for Uncertainty Quantification

Impact factor: 1.000

ISSN Print: 2152-5080
ISSN Online: 2152-5099

Open Access

International Journal for Uncertainty Quantification

DOI: 10.1615/Int.J.UncertaintyQuantification.2017013407
pages 99-115

CALIBRATION AND RANKING OF COARSE-GRAINED MODELS IN MOLECULAR SIMULATIONS USING BAYESIAN FORMALISM

Hadi Meidani
Department of Civil and Environmental Engineering, 1211 Newmark Civil Engineering Laboratory, University of Illinois at Urbana-Champaign, Urbana, IL 61822, USA
Justin B. Hooper
Department of Materials Science & Engineering, 206 Civil and Materials Engineering Building, University of Utah, Salt Lake City, UT 84112, USA
Dmitry Bedrov
Department of Materials Science & Engineering, 206 Civil and Materials Engineering Building, University of Utah, Salt Lake City, UT 84112, USA
Robert M. Kirby
Scientific Computing and Imaging Institute, University of Utah, Salt Lake City, UTAH 84112, USA; School of Computing, University of Utah, Salt Lake City, UTAH 84112, USA

ABSTRACT

Understanding and prediction of the performance of complex materials using molecular simulations, as well as the design of a new generation of materials with desired functionality, depend on the predictive capacity of the coarsegrained models that enable reduced computational cost. Depending on what aspect of the behavior is of interest, we often have at our disposal various coarse-grained models with different predictive capabilities. In this work, focusing on coarse-grained water models, we demonstrate how the plausibilities of these models are relatively compared, and how predictions can be made exploiting the ensemble. Using a Bayesian model ranking framework, we will show the plausibility results which are in agreement with experts' expectation on how these models rank in terms of predicting different quantities of interest, and how different models can be mixed and produce an ensemble prediction with higher accuracy.