Machine Learning for Improved Lattice Boltzmann Method (LBM) Simulations at High Reynolds Number

Research at the interface between classical numerical methods and modern data-driven algorithms tends to be referred to as Scientific Machine Learning (SML). There has been an enormous growth in interest in SML over recent years however the vast majority of this has been based upon classical numerical methods which solve macroscopic models of the physics (e.g. partial differential equations). There has been surprisingly little research that focuses on the use of the LBM within SML: the proposed topic of this project. This will make the literature easier to cover, however it will require becoming familiar with mesoscopic techniques such as LBM, as well as with standard techniques in machine learning. The expectation is that much of the computational work will be carried out using GPUs since both LBM and ML algorithms are ideally suited to exploiting such architectures.