Computational Math Seminar: Wayne Mitchell
Developing a Learning Algorithm-Generated Empirical Relaxer (LAGER) 听 听 听 听 听
Wayne Mitchell
Applied Mathematics,听
Date and time:听
Tuesday, November 11, 2014 - 10:30am
尝辞肠补迟颈辞苍:听
GRVW 105
础产蝉迟谤补肠迟:听
Modern hydrodynamics simulations are commonly performed on a mesh which moves with the fluid. This Lagrangian mesh movement causes problems, however, when zones in a mesh tangle: i.e. they become severely distorted such that calculations begin to yield non-physical values (negative zone volumes, densities, etc.). This problem can be addressed by using an Arbitrary Lagrangian Eulerian (ALE) approach. This involves performing an Eulerian relaxation step (moving mesh nodes and advecting fluid through the mesh) after the Lagrangian step. Currently it is up to the user to decide how much relaxation is appropriate for a given problem. This 鈥渂y hand鈥 method of relaxing requires significant expertise on the part of the user and results in relaxation criteria which are suited only to individual problems. Our goal is to automate the relaxation process through the use of a Learning Algorithm-Generated Empirical Relaxer (LAGER). This relaxer will utilize a regressive random forest algorithm to decide the appropriate amount of relaxation at each step during a simulation.