isayev / comp6 Goto Github PK

I found that the sign of forces in the dataset is incorrect. Take the first frame of ANI-MD dataset as an example:
Below is the forces given by gaussian (unit: Hartree/Bohr)

and here is the forces in the dataset (extracted using example_data_sampler.py, unit: Hartree/Angstrom):

Considering the conversion from Bohr to Angstrom (1 Bohr = 0.53 Angstrom), it can be deduced that the sign is wrong.

Apologies if this information is available somewhere and I missed it. What are the units of various quantities in this data set?

My best guesses are:

• coordinates: Bohr
• energy: Hartree
• force: Hartree/Bohr
• charge/density: a.u.
• dipole: debye

Hi @isayev, I am using the ANI1x model included with torchani and I am unable to reproduce the values you have reported. Could there be something simple I am missing?