Efficient Numerical Methods for Brownian Dynamics
The Brownian (Langevin) dynamics model underpins a wide variety of applications in biology, chemistry, physics, finance and data science, and the efficient numerical solution of the equations of motion, the key to scalable (high-dimensional) simulation, is thus of great importance. In this talk I will discuss the design of numerical methods, including their analysis, implementation and application in various real-world settings, such as in molecular dynamics and for parameter inference in the training of neural networks. In many cases an order of magnitude improvement in computational efficiency is possible simply by minor rearrangement of the order in which various terms are computed.