Class HMCStepRENS

source code

     object --+                        
              |                        
AbstractSampler --+                    
                  |                    
         AbstractMC --+                
                      |                
     AbstractEnsembleMC --+            
                          |            
         AbstractExchangeMC --+        
                              |        
                   AbstractRENS --+    
                                  |    
                   AbstractStepRENS --+
                                      |
                                     HMCStepRENS

Replica Exchange with Nonequilibrium Switches (RENS, Ballard & Jarzynski 2009) with stepwise trajectories as described in Nilmeier et al., "Nonequilibrium candidate Monte Carlo is an efficient tool for equilibrium simulation", PNAS 2011. The switching parameter dependence of the Hamiltonian is a linear interpolation between the PDFs of the sampler objects, H(λ) = H_2 * λ + (1 - λ) * H_1. The perturbation kernel is a thermodynamic perturbation and the propagation is done using HMC.

Note that due to the linear interpolations between the two Hamiltonians, the log-probability and its gradient has to be evaluated four times per perturbation step which can be costly. In this case it is advisable to define the intermediate log probabilities in _run_traj_generator differently.