fix: non-periodic systems in TorchSim graph construction (#161)

* fix: normalize non-periodic systems in TorchSim graph construction

The TorchSim code path (build_graph_from_simstate) was passing raw
pbc=False and a zero cell for non-periodic molecules, while the ASE
calculator path (_normalize_atoms) creates a large fake periodic cell
and wraps positions into it. This caused the M3GNet model to receive
different graph inputs for the same molecule, producing inconsistent
energies and forces.

Add _normalize_nonperiodic_systems() that replicates the same fake-cell
normalization for the TorchSim path, ensuring both code paths feed
identical inputs to the model.

Fixes #160

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* test: add regression tests for non-periodic TorchSim normalization

Add unit tests for _normalize_nonperiodic_systems() verifying:
- Periodic systems pass through unchanged
- Non-periodic molecules get a large fake periodic cell
- Cell and positions match the ASE _normalize_atoms code path
- Original SimState tensors are not mutated

Add end-to-end test (test_wrapper_molecule_consistency) comparing
TorchSimWrapper vs MatterSimCalculator on a water molecule.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* fix: request full FC format for phono3py v4 compatibility

Phono3py v4 defaults to compact force constants format, changing
fc2 shape from (n_satom, n_satom, 3, 3) to (n_patom, n_satom, 3, 3).
Explicitly pass is_compact_fc=False to produce_fc2() and produce_fc3()
to maintain the full format expected by downstream code.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

---------

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

feat: add double precision (float64) inference support (#156)

* feat: add double precision (float64) inference support

Add dtype='float64' option to MatterSimCalculator for double-precision
inference. This improves numerical stability for sensitive properties
like thermal conductivity.

Changes:
- MatterSimCalculator: new dtype parameter ('float32' or 'float64'),
  converts model weights via model.double() and creates inputs in the
  matching dtype
- m3gnet.py: replace .float() with .to(pos.dtype) for pbc_offsets so
  it respects the model's precision
- calculate(): upcast legacy-path float tensors to match model dtype

On equilibrium Si diamond, float64 reduces force noise from ~1e-6 to
~1e-15 eV/Å compared to float32.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* test: use float64 precision in BTE thermal conductivity tests

Switch BTE tests to dtype='float64' for more stable force constants.
Update reference kappa values to float64 means verified across 3 runs
(rel_std < 0.03%). Tighten tolerance from 8% to 2%.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* fix: reject float64 on MPS devices and skip float64 BTE tests on MPS

MPS does not support float64 tensors. This commit:
- Validates dtype+device combo early in MatterSimCalculator.__init__,
  raising ValueError before loading the model
- Falls back to float32 in the basic BTE test on MPS
- Skips the strict BTE test on MPS (reference values require float64)

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* test: allow strict BTE test to run on MPS with relaxed tolerances

Use float32 with 10% kappa tolerance and 0.5% FC tolerance on MPS,
while keeping float64 with tight tolerances (2% / 0.1%) on CUDA.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* fix: alias torch._inductor.config import to avoid shadowing torch

The local 'import torch._inductor.config' statement caused Python to
treat 'torch' as a local variable in __init__, making earlier references
to torch.float32/torch.float64 fail with UnboundLocalError.

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

* fix: preserve float64 dtype after calculator pickle roundtrip

Agent-Logs-Url: https://github.com/microsoft/mattersim/sessions/b2ac740c-f04e-4e6b-a5c0-b3389a4caaf3

Co-authored-by: yanghan234 <32856137+yanghan234@users.noreply.github.com>

---------

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>
Co-authored-by: copilot-swe-agent[bot] <198982749+Copilot@users.noreply.github.com>

fix: make MatterSimCalculator picklable and remove DeepCalculator (#83)…

… (#153)

Co-authored-by: Copilot <223556219+Copilot@users.noreply.github.com>

Merge pull request #143 from microsoft/fix-stress-import

fix stress import import

fix: fixing import failure with setuptools>=82 (#141)

* fix the issue with pkg_resources to get version

* add test case for the fix

ci: remove py39 for cibuildwheel (#123)

* update the cibuildwheel configuration

* remove building for py3.9

Update python-publish.yaml (#94)

Update python-publish.yaml (#94)

ci: Update python-publish.yaml (#93)

* Update python-publish.yaml

bump-to-py313

* Update build.yaml

Update the versions of python to test on and include python 3.13

* Update build.yaml

* Update environment.yaml

unpin torch<2.5.0 (#89)