This poster aims to summarize my studies in Direct Simulation Monte Carlo (DSMC), which is a method of Computation Fluid Dynamics (CFDs) for rarefied gas dynamics. A great challenge in flow simulation is when a continuum can no longer be assumed and there is a need to directly simulate “superparticles” that represent a macroscopic neighborhood of real particles. However, yielding accurate results comes at the tradeoff of computational cost. There are three main families of algorithms that I have implemented in order to increase accuracy and to reduce computational time. 

The first controllable factor is how to pair the particles. It’s not feasible to simulate every collision. A major family of pairing algorithms are the Bernoulli Trials (BT) schemes that reduce the amount of pair considerations and increase a statistical weight to maintain accuracy. Second, the scattering behavior of the particles. I have worked with Variable Hard Sphere (VHS), which is a macroscopic post-collision model, and Ab-Initio (ABI), which is a quantum based microscopic post-collision model. ABI is a more accurate model, but also at the cost of more expensive computations. Thus, the third algorithm I will implement is Neural Networks (NN). The aim is to use NNs for predictive fluid flow to reduce computational time while maintaining the high accuracy ABI offers. 

I aim to generalize my results to many fluid problems. These include relaxation, Couette flow, Fourier flow, and hypersonic cavity problems. I will conclude my poster with comparisons of different combinations of methods with the goal of implementing a high-efficiency-high-accuracy DSMC model.