Add possibility to use your own M3GNet potential (#911)

* allow the possibility to use your own M3GNet potential

allow the possibility to use your own M3GNet potential, instead of the pretrained model only.

* added a unit test

* added a unit test

* test_dir not needed

* change kwargs passing

src/atomate2/forcefields/utils.py

@@ -421,7 +421,8 @@ def ase_calculator(calculator_meta: str | dict, **kwargs: Any) -> Calculator | N
             import matgl
             from matgl.ext.ase import PESCalculator
 
-            potential = matgl.load_model("M3GNet-MP-2021.2.8-PES")
+            path = kwargs.get("path", "M3GNet-MP-2021.2.8-PES")
+            potential = matgl.load_model(path)
             calculator = PESCalculator(potential, **kwargs)
 
         elif calculator_name == MLFF.MACE:

tests/forcefields/test_utils.py

@@ -163,3 +163,25 @@ def test_fix_symmetry(fix_symmetry):
         assert symmetry_init["number"] == symmetry_final["number"] == 229
     else:
         assert symmetry_init["number"] != symmetry_final["number"] == 99
+
+
+def test_m3gnet_pot():
+    import matgl
+    from matgl.ext.ase import PESCalculator
+
+    kwargs_calc = {"path": "M3GNet-MP-2021.2.8-DIRECT-PES", "stress_weight": 2.0}
+    kwargs_default = {"stress_weight": 2.0}
+
+    m3gnet_calculator = ase_calculator(calculator_meta="MLFF.M3GNet", **kwargs_calc)
+
+    # uses "M3GNet-MP-2021.2.8-PES" per default
+    m3gnet_default = ase_calculator(calculator_meta="MLFF.M3GNet", **kwargs_default)
+
+    potential = matgl.load_model("M3GNet-MP-2021.2.8-DIRECT-PES")
+    m3gnet_pes_calc = PESCalculator(potential=potential, stress_weight=2.0)
+
+    assert str(m3gnet_pes_calc.potential) == str(m3gnet_calculator.potential)
+    # casting necessary because <class 'matgl.apps.pes.Potential'> can't be compared
+    assert str(m3gnet_pes_calc.potential) != str(m3gnet_default.potential)
+    assert m3gnet_pes_calc.stress_weight == m3gnet_calculator.stress_weight
+    assert m3gnet_pes_calc.stress_weight == m3gnet_default.stress_weight