Skip to main content

Potential energy surfaces for atomistic simulation (LJ + Metatomic multi-ABI dlopen; nanobind abi3)

Project description

rgpot

img

codecov

rgpot is a potential-energy library and Cap'n Proto RPC server for atomistic simulation codes. It gives clients one geometry carrier, one result carrier, and one extensible backend-configuration carrier:

  • ForceInput: positions, atomic numbers, box, and unit strings
  • PotentialResult: energy plus flat force array
  • PotentialConfig: backend setup, with arms such as none, nwchem, and cpmd

The public wire schema lives in CppCore/rgpot/rpc/Potentials.capnp and is shared by the C++ server, Python integration tests, and the Rust core crate. Native rgpot units are eV and Angstrom. RPC clients may request compatible units through ForceInput.lengthUnit and ForceInput.energyUnit.

Python (PyPI)

Install the manylinux wheel (nanobind abi3, Python >= 3.12 for the stable ABI extension; requires-python is >= 3.10 for pure metadata):

pip install rgpot
# Metatomic path (engine libs resolve from these packages at runtime):
pip install 'rgpot[metatomic]'
# or: pip install torch metatomic-torch metatensor-torch metatensor-core vesin

Lennard-Jones needs only numpy. Metatomic uses dlopen of a bundled engine:

  • layout: rgpot/lib/torch-X.Y/libmetatomic_engine.so
  • picker: installed torch major (same multi-ABI idea as metatomic-torch)
  • supported torch majors: 2.7 and newer (wheels ship 2.7–2.13 engines)
  • torch 2.6 and older are not supported for Metatomic dlopen
import numpy as np
import rgpot

print(rgpot.__version__)
print(rgpot.available_metatomic_engine_abis())  # e.g. 2.7 .. 2.13
# energy, forces, variance = rgpot.evaluate_metatomic(
#     positions, atom_types, box, model_path="model.pt")

See https://pypi.org/project/rgpot/.

Build And Test

Meson is the primary build path. Use the rpctest environment when running the RPC integration scripts because they need pycapnp.

pixi shell -e rpctest
meson setup bbdir -Dwith_tests=true -Dwith_rpc=true --buildtype=debug
meson compile -C bbdir
meson test -C bbdir --print-errorlogs
python tests/rpc_integ.py --server-bin ./bbdir/CppCore/potserv

CMake is supported as well:

cmake -B build \
  -DRGPOT_BUILD_TESTS=ON \
  -DRGPOT_BUILD_EXAMPLES=ON \
  -DRGPOT_WITH_RPC=ON
cmake --build build
ctest --test-dir build --output-on-failure

For client-only consumers that only need the Cap'n Proto schema and generated RPC client types:

cmake -B build_client \
  -DRGPOT_RPC_CLIENT_ONLY=ON \
  -DRGPOT_BUILD_TESTS=ON
cmake --build build_client
ctest --test-dir build_client --output-on-failure

Backends

Backend Server selector Build / runtime notes
LJ LJ Built-in 12-6 Lennard-Jones reference potential
CuH2Pot CuH2 Built-in Cu-H EAM potential
XTBPot XTB, GFNFF, GFN0xTB, GFN1xTB Enable with -Dwith_xtb=true; use pixi env xtbbld or tbbld; linked XTBPot + dlopen libxtb_engine.so (see docs/xtb_backends.md)
TBLitePot TBLite, TBLiteGFN1, TBLiteIPEA1 Enable with -Dwith_tblite=true; use pixi env tblitebld or tbbld
MetatomicPot Metatomic:<model_path> Enable with -Dwith_metatomic=true; use pixi env metatomicbld. Pip engines: torch 2.7+
NWChemPot NWChem Frontend always builds; load libnwchemc from the split nwchemc project at runtime
CPMDPot CPMD Frontend always builds; load libcpmdc from the split cpmdc project at runtime

Example server commands:

./bbdir/CppCore/potserv 12345 LJ
./bbdir/CppCore/potserv 12345 Metatomic:CppCore/tests/data/lj38/lennard-jones.pt
CPMDC_LIBRARY=/path/to/libcpmdc.so ./bbdir/CppCore/potserv 12345 CPMD
NWCHEMC_LIBRARY=/path/to/libnwchemc.so ./bbdir/CppCore/potserv 12345 NWChem

CPMD And NWChem Configuration

CPMD and NWChem do not use ad hoc rgpot config files. They use the PotentialConfig union over RPC:

  • PotentialConfig.cpmd carries CPMDParams for CPMDPot
  • PotentialConfig.nwchem carries NWChemParams for NWChemPot
  • geometry still arrives through ForceInput on every calculate() call

For CPMD, tests/cpmd_params.py contains pycapnp helpers for building CPMDParams and PotentialConfig.cpmd messages. The helper covers scalar fields, raw inputBlocks, and every structured CPMDInputSection arm. See CppCore/rgpot/CPMDPot/README.md for the engine lookup order and schema field mapping.

The CPMD runtime path is:

  1. Build libcpmdc.so in the split cpmdc repository (or use the in-tree libcpmdc_fake_engine.so from a -Dwith_tests=true build for CI).
  2. Start potserv <port> CPMD with CPMDC_LIBRARY, RGPOT_CPMDC_ENGINE, or RGPOT_CPMD_ENGINE pointing at that shared library.
  3. Send configure(PotentialConfig.cpmd) once for method setup.
  4. Send ForceInput on each calculate() call for geometry and requested output units.

For NWChem, tests/nwchem_params.py builds NWChemParams / PotentialConfig.nwchem. Engine lookup order is NWCHEMC_LIBRARY, RGPOT_NWCHEMC_ENGINE, RGPOT_NWCHEM_ENGINE, then enginePath on params. See CppCore/rgpot/NWChemPot/README.md. The NWChem runtime path mirrors CPMD:

  1. Build libnwchemc.so in the split nwchemc repository (or libnwchemc_fake_engine.so from a tests build).
  2. Start potserv <port> NWChem with the library env vars above.
  3. configure(PotentialConfig.nwchem) once, then calculate(ForceInput).

potctl drives the same Potential Cap'n Proto RPC against potserv (bridge stress and CI legs pass the potential name such as NWChem or CPMD).

Developer Tasks

Hooks use prek through prek.toml. Common commands:

pixi r prek-install
pixi r prek
pixi r rust-test
pixi r -e rpctest python tests/test_cpmd_params.py
pixi r -e rpctest python tests/test_nwchem_params.py
pixi r -e rpctest python tests/test_rpc_integ_cpmd.py
pixi r -e rpctest python tests/test_rpc_integ_nwchem.py
# Full RPC + C ABI E2E (point env at built fake engines + potserv)
export RGPOT_POTSERV=/path/to/bbdir/CppCore/potserv
export NWCHEMC_LIBRARY=/path/to/bbdir/CppCore/libnwchemc_fake_engine.so
export RGPOT_CPMD_ENGINE=/path/to/bbdir/CppCore/libcpmdc_fake_engine.so
pixi r -e rpctest python tests/test_rpc_e2e_c_abi.py
pixi r -e rpctest python tests/rpc_integ.py \
  --server-bin ./bbdir/CppCore/potserv \
  --nwchem-smoke \
  --cpmd-smoke

The root README is generated from readme_src.org. Project documentation sources live under docs/orgmode/.

License

MIT, with backend-specific notes: some potentials are adapted from eOn under BSD-3-Clause terms. The unit expression parser in CppCore/rgpot/units.cc is derived from metatomic-torch (BSD-3-Clause, metatensor developers).

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

rgpot-2.5.1.tar.gz (1.7 MB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

rgpot-2.5.1-cp312-abi3-manylinux_2_39_x86_64.whl (426.4 kB view details)

Uploaded CPython 3.12+manylinux: glibc 2.39+ x86-64

File details

Details for the file rgpot-2.5.1.tar.gz.

File metadata

  • Download URL: rgpot-2.5.1.tar.gz
  • Upload date:
  • Size: 1.7 MB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.2.0 CPython/3.14.6

File hashes

Hashes for rgpot-2.5.1.tar.gz
Algorithm Hash digest
SHA256 a6654573775ea9bd1a6aca2ccc6005f238fb8a93185e48c2a1528aad1789ee89
MD5 44072f43fdd711c2c17faaaa6c1c0392
BLAKE2b-256 c16b8f580e66b6a4aba839ec24ec99893b56d5563fbf95a7854422b8ddbef4d4

See more details on using hashes here.

File details

Details for the file rgpot-2.5.1-cp312-abi3-manylinux_2_39_x86_64.whl.

File metadata

File hashes

Hashes for rgpot-2.5.1-cp312-abi3-manylinux_2_39_x86_64.whl
Algorithm Hash digest
SHA256 944367cb90b7e3e528ec76aa3f08a338d75565c2bd9e3bc181f5c52b041dc4b6
MD5 4dbd3e568c723eb4656d1e525e7264bf
BLAKE2b-256 85bf1aa230355c0a498e939c37749a97b2f8827871916bf2669dcb8e8a367cb9

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page