Implement Magnetostatics in TorchPME #133
Conversation
E-Rum
force-pushed
the
magnetostatics
branch
2 times, most recently
from
615c3cd to
a73d69e
Compare
| r_outer = torch.einsum( | ||
| "bi,bj->bij", vector, vector |
There was a problem hiding this comment.
Might be worth for speed later to replace the einsum by outer. The latter is usually much faster.
| from ..potentials import PotentialDipole | ||
|
|
||
|
|
||
| class CalculatorDipole(torch.nn.Module): |
There was a problem hiding this comment.
We should the replace the "current" Calculator with "CalculatorMonopole" ?
There was a problem hiding this comment.
Good question! We can discuss it after we merge.
| return scalar_potential.unsqueeze(-1) * torch.eye(3).to(r_outer).unsqueeze( | ||
| 0 | ||
| ) - r_outer * ( | ||
| 3.0 / (r_mag**5) | ||
| - 3.0 * torch.erfc(torch.sqrt(alpha) * r_mag) / r_mag**5 | ||
| - 2 | ||
| * torch.sqrt(alpha / torch.pi) | ||
| * (2 * alpha + 3 / r_mag**2) | ||
| * torch.exp(-alpha * r_mag**2) | ||
| / r_mag**2 | ||
| ).unsqueeze(-1) |
There was a problem hiding this comment.
Can you break this up into more readable lines. 10 lines for a return statement might be a but long.
There was a problem hiding this comment.
Makes sense! I will do that after we finish, and we’ll make sure we get the correct physical values. Then, we should do the “linting.”
| raise ValueError( | ||
| "Cannot compute background correction without specifying `smearing`." | ||
| ) | ||
| return self.smearing * 0.0 |
There was a problem hiding this comment.
Always 0?
| return self.smearing * 0.0 | |
| return 0.0 |
then we can even remove the value error?
There was a problem hiding this comment.
Here, it’s done this way to stay consistent with the pipeline. This ensures that 0.0 is returned as a torch.tensor with the correct device and dtype; otherwise, we would just return a float number.
There was a problem hiding this comment.
In general, I have no idea whether it should always be zero or not. One would need to perform a theoretical derivation to determine whether dipole neutrality is a crucial aspect for the dipole-Ewald calculator
…le classes for improved performance and clarity
…or PotentialDipole class
…r in PotentialDipole
|
Ok, from that point on, it seems to be working. Both the direct and Ewald calculators give reasonable values for two dipoles with a fixed distance. Next, I will add documentation, improve the code structure, and add tests. Meanwhile, I would love to get some help in calculating reference values for comparisons. @PicoCentauri and I started testing it with the espressomd P3MDipole code, but it seems to tune its parameters for each calculation. We will investigate how to prevent this or find an alternative method for testing. |
This branch is linked to #54 and introduces the initial implementation of magnetostatics in the existing TorchPME library. The commit provides a basic implementation for direct potential calculations along with a simple test case.
With this implementation, it’s a good time to start discussing how to structure the current magnetostatic functionality and align it with the broader scope of the library. Notable changes to the main library include:
1. Dipole Potential Components:
Unlike charge-based potentials (e.g., Coulomb), dipole potentials consist of two components: a scalar and a tensor part. Currently, these components are explicitly returned separately and later combined in the calculator. While this approach works for now, we should evaluate whether it remains viable as we transition to more complex Ewald-based calculators.
2. Calculator Structure for Dipoles:
The dipole calculator significantly differs from charge-based calculators, potentially necessitating a new base calculator tailored for magnetostatics. For now, the implementation resides in the calculators folder, but it may need restructuring.
Key differences include:
• Passing dipoles instead of charges.
• Utilizing neighbor vectors instead of neighbor distances.
• Returning potentials in a vectorial form, which appears more convenient from an ML perspective.
📚 Documentation preview 📚: https://torch-pme--133.org.readthedocs.build/en/133/