Position-dependent charge for electrostatistics

[smlee99]

Hi OpenMM devs!
I have a very similar question as previously mentioned on here. The one who opened this issue just adopt the charge during the simulation using updateParametersInContext method. As pointed by @jchodera , the position-dependence of the charges should also contribute to the force. Is it possible to account for this using a TorchForce object along with OpenMM PME electrostatistics? There are some articles (for instance, see https://www.tandfonline.com/doi/abs/10.1080/13642819208207645) mentioning the technical aspects of implementing charge transfer process.

[peastman]

As long as gradients can be computed correctly for your force, this will be handled correctly. Your model takes positions and returns energy. TorchForce calls backward() on it to get the gradient with respect to positions.