LOGWEIGHTS having no effect on HISTOGRAM

[Aël Cador]

Dear Plumed users,

I have run short umbrella sampling simulations (25 trajectories of 20 000 steps each, amounting to 3 ps for each trajectory) and want to analyze them using the procedure described here in exercise 4: https://www.plumed-tutorials.org/lessons/21/003/data/Solution.html

I am using Plumed 2.10 (dev version). I was able to compute the logweights using the wham.py script used in the tutorial, though the values can be very negative, down to -350, which is rather disquieting if this is the logarithm of a quantity.

Using this plumed script to make the histograms:

diff: READ FILE=bias_multi3.dat VALUES=diff IGNORE_TIME
lw: READ FILE=bias_multi3.dat VALUES=logweights IGNORE_TIME

hh: HISTOGRAM ARG=diff GRID_MIN=-4.0 GRID_MAX=4.0 GRID_BIN=400 BANDWIDTH=0.08
DUMPGRID GRID=hh FILE=hist_diff800_cat3.dat
ff: CONVERT_TO_FES GRID=hh TEMP=300 MINTOZERO
DUMPGRID GRID=ff FILE=fes_diff800_cat3.dat

hhr: HISTOGRAM ARG=diff GRID_MIN=-4.0 GRID_MAX=4.0 GRID_BIN=400 BANDWIDTH=0.08 LOGWEIGHTS=lw
DUMPGRID GRID=hhr FILE=hist_diff800_catr3.dat
ffr: CONVERT_TO_FES GRID=hhr TEMP=300 MINTOZERO
DUMPGRID GRID=ffr FILE=fes_diff800_catr3.dat

with this command:

plumed driver --noatoms --plumed plumed_wham.dat --kt 2.49

(using default Plumed units and T = 300 K, so kBT = 2.49 kJ/mol)

which outputs a histogram and a FES either with or without logweights: I get the exact same FES, not only visually but also down to every decimal in the output files.

If I try to use directly WHAM_HISTOGRAM to perform this, the plumed setup runs without a hitch until the end then exits abruptly with a segmentation fault error.

By the way, I am using Plumed 2.10 because it was the only version of Plumed which I managed to compile together with CPMD after many attempts.
The histogram parameters in the Plumed file are set to see a TS around a CV value of 0. The bandwidth is equal to the standard deviation of the CV during the umbrella sampling runs. 

Does anyone have an idea of what I could look into to resolve this? I haven't found many options that seem useful to debug this yet... Hopefully this isn't a specificity of the dev version at the time I downloaded it.

There are probably other problems to address here such as the very negative values of the logweights (due to the low number of points?). I also used the entirety of the trajectories without considering relaxation times, but in all cases I would expect the logweights to change the profile even by a tiny bit.

Attached are the files bias_multi3.dat (basically a COLVAR file) containing the CV values and weights, as well as the histogram and FES files.

Thank you in advance for your help. Let me know if you need additional details about what I'm trying to do.

Aël Cador