NewtonNet model settings do not always get correctly passed to _add_stdev_and_hess or _get_hessian

[Andrew-S-Rosen]

Details about the quacc environment

• quacc version: 0.9.1
• Python version: 3.11

What is the issue?

In the NewtonNet recipes, the model path and settings paths for the NewtonNet model architecture are (by default) pulled from the global quacc settings. However, if the user passes model_path or settings_path as **calc_kwargs to the recipe, it will override the defaults.

quacc/src/quacc/recipes/newtonnet/core.py

Lines 64 to 68 in a73891d

	calc_defaults = {
	"model_path": SETTINGS.NEWTONNET_MODEL_PATH,
	"settings_path": SETTINGS.NEWTONNET_CONFIG_PATH,
	}
	calc_flags = recursive_dict_merge(calc_defaults, calc_kwargs)

This is all perfectly fine, but if a user passes a custom model_path or settings_path, this is not used by _add_stdev_and_hess because that function is hard-coded to use only the global settings:

quacc/src/quacc/recipes/newtonnet/core.py

Lines 212 to 215 in a73891d

	atoms.calc = NewtonNet(
	model_path=SETTINGS.NEWTONNET_MODEL_PATH,
	settings_path=SETTINGS.NEWTONNET_CONFIG_PATH,
	)

This can cause a potential conflict. Note that the same is true for _get_hessian():

quacc/src/quacc/recipes/newtonnet/ts.py

Lines 286 to 289 in a73891d

	ml_calculator = NewtonNet(
	model_path=SETTINGS.NEWTONNET_MODEL_PATH,
	settings_path=SETTINGS.NEWTONNET_CONFIG_PATH,
	)

Tagging @samblau and @kumaranu.

How can we easily reproduce the issue?

from quacc.recipes.newtonnet.core import relax_job
from ase.build import molecule

atoms = molecule("CH4")
model_path = "/my/custom/model/path" # not in settings
relax_job(atoms, model_path=model_path)