search

learningmatter-mit / geom

GEOM: Energy-annotated molecular conformations
193 stars 24 forks source link

Boltzmann weights #7

Open hansen7 opened 2 years ago

hansen7 commented 2 years ago

Hi, there, thanks for sharing this work!

I've noticed that for some molecules (~265 in total), the summation of Boltzmann weights are larger than 1 (roughly doubles), any ideas?

The code to reproduce the above observations:

import json, numpy as np
mols = json.load(open("summarise.json"))

for mol in mols:
    if sum(mol['boltzmann_weight']) > 1.1:
        print(sum(mol['boltzmann_weight']))
1.9999999999999998
2.0000200000000015
2.0000199999999997
2.0000199999999997
2.00004
1.9999600000000002
2.0000200000000006
2.00008
1.9996600000000038
1.9999999999999998
1.999860000000001
1.9994800000000013
1.999880000000001
1.9998800000000023
2.0000199999999997
2.0
...
simonaxelrod commented 2 years ago

Well, I have no idea how some of the sums became 2. For now I'd suggest ignoring these molecules or dividing the Boltzmann weights by their sums. In the meantime I'll look into it and update the dataset accordingly