subsampling energies in somd-freenrg

[ElSmaele]

Hello, I have been running somd-freenrg calculations I often obtained the following warning when analyzing my results "#WARNING SUBSAMPLING ENERGIES RESULTED IN LESS THAN 50 SAMPLES, CONSIDER RERUN WITHOUT SUBSAMPLE OPTION".
What I understand is that I should run my simulations for a longer time. But I am not quite sure what the subsampling option actually does.

• how is the statistical inefficiency evaluated?
• is the subsampling also applied for the MBAR analysis?
• what are these 50 samples? are they gradient values per lambda or the whole number of gradient values considered for the TI analysis?
  Thank you