A DCAMM seminar will be presented by
Professor Kurt Anderson
Department of Mechanical, Aerospace, and Nuclear Engineering
Rensselaer Polytechnic Institute
Troy, NY 12180, USA
Abstract:
Predicting important structural properties of large biomolecular systems such as RNA which play a critical role in various biological processes has gained increasing importance in recent years. In the modeling process of many of these systems, adaptive multiscale techniques may be desired to realize efficiency through an optimal combination of simulation speed and accuracy. For instance, consider the simulations of biomolecular systems with thousands of atoms and associated degrees of freedom which involve temporal domains ranging from sub-femto seconds for covalently bonded atomic motions, to milliseconds for important conformational motions to emerge. The traditional approach for molecular modeling involves fully atomistic models which result in kinematically decoupled equations of motion. In spite of the simplicity of producing and solving the fully atomistic models, they become intractable for large systems due to the crippling cost of their associated forcing term calculation and the necessity of using exceedingly small integration step sizes.
Treating groups of atoms as rigid or flexible bodies connected via kinematic joints provides coarse grain structures to model a particular molecule. For larger systems, often a finest coarse grain structure is used for simulation. For most systems of interest, these coarse grained models are still prohibitively slow. Consequently, it is desired to generate and use the coarsest molecular system model which still provides results of acceptable accuracy. Due to the significant nonlinear coupling of the system equations of motion in the state variables and their derivatives, even modest changes in system state can render the existing coarsened model inadequate. Consequently, a coarse grain model which may perform well at one point of the simulation may perform poorly as the simulation progresses. Therefore, providing machinery to implement the simulation in an adaptive multiscale scheme is necessary. In this approach, the system model changes locally or globally as the simulation circumstances demand. The goal is to provide the appropriate model resolution when and where it is needed. This leads to an articulated multibody representation of the system in which the degrees of freedom are added or taken away during the course of the simulation. Such adaptive simulations add several new computational challenges which are discussed in this talk.
Danish pastry, coffee and tea will be served 15 minutes before the seminar starts.
All interested persons are invited.