search

axsk / ISOKANN.jl

Julia implementation of the ISOKANN algorithm for the computation of invariant subspaces of Koopman operators
MIT License
5 stars 5 forks source link

Save momentum in Langevin simulations #4

Closed axsk closed 9 months ago

axsk commented 1 year ago

Currently we re-seed momentum on each (short) tractory, which leads to unphysical results.

This requires to decouple NN input/data from the underlying simulation data, as the former doesn't require the momentum.

Look at the work done in IsoMu.jl

axsk commented 11 months ago

Actually sampling from Boltzmann is not really wrong, so maybe there is no need for this after all?

axsk commented 9 months ago

The current opinion on this is: Resampling from Boltzmann anew everytime is statisically correct and results in faster mixing