| Contents |
| Many molecular dynamics sampling algorithms rely on efficiently generating trajectories that sample the canonical (NVT) ensemble. A popular choice for creating suitable trajectories is to use Langevin dynamics to provide a thermostatted dynamics whose solutions are consistent with the desired Boltzmann-Gibbs distribution. These dynamics then require numerical discretization in order to approximate the solution in time, which in turn introduces a bias in the sampling.
In this talk, we will discuss some results on the invariant distributions sampled by Langevin dynamics discretization schemes defined through exactly solving parts of the SDE vector field in sequence. We focus on the case where purely configurational sampling of the ensemble is of interest. |
|