Optimizing Sparse Representations of Kinetic Distributions via Information Theory
Abstract
This project is on the use of ideas from information theory in the kinetic simulation of a gas or plasma. A kinetic simulation describes the interactions (i.e., collisions and convection) of particles that constitute a gas or plasma. Since the number of physical particles is often much too large (e.g., 1020) for direct molecular dynamics computations, kinetic simulation often uses a moderate number, N (e.g.,105-107), representative "computational macro-particles" which act as surrogates for the particle interactions. The particle positions, xn, and velocities, vn, for n ranging from 1 to N, are a representative sample of a probability distribution function f(x; v). Traditionally, these macro-particles have all represented a constant number of real particles with a particle "shape" which is a single (Dirac-delta function) velocity and either delta functions in space or low order splines dependent on the spatial resolution sought as described in more detail in Bridsalls classic reference [1]. This sparse sampling of f results in a direct trade-off between spatial accuracy and statistical noise for key flow-field parameters such as mass, momentum, energy, and physical entropy.
Document Details
- Document Type
- Technical Report
- Publication Date
- Jul 31, 2017
- Accession Number
- AD1044589
Entities
People
- Daniel Eckhardt
- Robert R. S. Martin
Organizations
- Air Force Research Laboratory