kbarros
commented
2 years ago Hi Matt, I think we're good to merge. Want to do the honors (push the button)?
Closed mswwilson closed 2 years ago
kbarros
commented
2 years ago Hi Matt, I think we're good to merge. Want to do the honors (push the button)?
PT/FBOPT and its examples are removed and replaced by REMC/FBO.
Changes:
User must initialize MPI in their script before instantiating replicas -- a function "init_MPI()" is provided for this
Instead of tempering using kT, an arbitrary control parameter α is used
User must construct their own schedule for α before running REMC (using process rank)
Control parameter schedule must start with smallest value as first entry (ex: largest kT <-> smallest β)
User must define a function "set_α!(::Replica, ::Float64)" to assign updated α when using FBO --- This "set_α!" should set a new SpinSystem in the sampler with updated Hamiltonian parameters, or can update --- parameters of the replica's AbstractSampler
The integral in feedback updates has been debugged (to more reliably enforce normalization)
After running FBO, the optimized α schedule is contained in the replicas and does not need to be set again
Configurations are exchanged between replicas during exchange proposal step instead of after acceptance
Net magnetization is measured as vector and histogram is added for magnetization norm.
Replica flow, f(α), can be measured during REMC for diagnostic purposes