Metadata-Version: 2.1
Name: torch-nl
Version: 0.3
Summary: TorchScript-able neighbor lists implementations (linear and quadratic scaling) for molecular modeling
Author: Félix Musil
License: MIT
Classifier: Programming Language :: Python :: 3.7
Classifier: Topic :: Scientific/Engineering :: Chemistry
Classifier: Topic :: Scientific/Engineering :: Physics
Requires-Python: >=3.8
Description-Content-Type: text/markdown
License-File: LICENSE

# torch_nl

Provide a pytorch implementation of a naive (`compute_neighborlist_n2`) and a linked cell (`compute_neighborlist`) neighbor list that are compatible with TorchScript.

Their correctness is tested against ASE's implementation.

Note that contrary to ASE, the atoms positions are assumed to be wrapped inside the unit cell.
# How to

## instal with pip

```bash
pip install torch-nl
```

## use the neighborlist

```python
from torch_nl import compute_neighborlist, ase2data
from ase.build import bulk, molecule

frames = [bulk("Si", "diamond", a=6, cubic=True), molecule("CH3CH2NH2")]
pos, cell, pbc, batch, n_atoms = ase2data(frames)

mapping, batch_mapping, shifts_idx = compute_neighborlist(
    cutoff, pos, cell, pbc, batch, self_interaction
)
```

# Benchmarks

## Periodic structure

![](benchmark/diamond_benchmark.png)
