Metropolis Samplers

These samplers include a Metropolis-Hastings style step that ensure that the Boltzmann distribution $\mu(x) \propto e^{-\beta V(x)}$ is exactly targeted; there is no bias associated with, for instance, a finite time step Δt.

Zeroth Order Methods

These are samplers which do not require the gradienet of the potential, ∇V.

BasicMD.RWM — Method
RWM(V, β, Δt)

Set up the RWM sampler for Boltzmann.

Fields

  • V - Potential
  • β - Inverse temperature
  • Δt - Time step

First Order Methods

These are samplers which require the gradienet of the potential, ∇V, and are in the spirit of first order in time discretizations.

BasicMD.MALA — Method
MALA(V, ∇V!, β, Δt)

Set up the MALA sampler for overdamped Langevin.

Fields

  • V - Potential
  • ∇V! - In place gradient of the potential
  • β - Inverse temperature
  • Δt - Time step

Second Order Methods

These are samplers which require the gradienet of the potential, ∇V, and are in the spirit of second order in time discretizations.

BasicMD.HMC — Method
HMC(V, ∇V!, β, M, Δt, nΔt)

Set up the HMC sampler for Boltzmann.

Fields

  • V - Potential
  • ∇V! - In place gradient of the potential
  • β - Inverse temperature
  • M - Mass matrix
  • Δt - Time step
  • nΔt - Number of time steps to use in each Verlet run