Unable to reproduce aluminum bulk modulus

[shivang-22]

Hello all, I am trying to calculate the bulk modulus of aluminum using the code below. I used the DeformedStructureSet technique to create different strains on the material, and then using CHGNet predict the stresses corresponding to each strain. The aluminum cif file was taken directly from the Materials Project website.

from chgnet.model.model import CHGNet
from pymatgen.core import Structure
from pymatgen.analysis.elasticity.strain import Deformation, DeformedStructureSet, Strain
from pymatgen.analysis.elasticity.stress import Stress
from pymatgen.analysis.elasticity import ElasticTensor

chgnet = CHGNet.load()
structure = Structure.from_file('cif_files/Al.cif')
prediction = chgnet.predict_structure(structure)

eq_stress = Stress(prediction['s'])

deformed_structure = DeformedStructureSet(structure)

strain_list = [i.green_lagrange_strain for i in deformed_structure.deformations]
stress_list = []

for i in range(len(deformed_structure)):
    prediction = chgnet.predict_structure(deformed_structure[i])
    stress_list.append(Stress(prediction['s']))

elastic_tensor = ElasticTensor.from_independent_strains(strain_list, stress_list, eq_stress)

print(elastic_tensor.property_dict)

What I get is:
{'k_voigt': 9.845064083735359, 'k_reuss': 9.845063833881039, 'k_vrh': 9.845063958808199, 'g_voigt': 8.0023897488912, 'g_reuss': 4.5904549461157655, 'g_vrh': 6.2964223475034835, 'universal_anisotropy': 3.716336252207812, 'homogeneous_poisson': 0.23641618649298335, 'y_mod': 15569997014.898907}

This seems wrong since the bulk modulus of aluminum should be around 76 GPa, and not around 10 GPa. Can anyone point me in the correct direction, or tell me what have I missed? I would very much appreciate it.

[bowen-bd]

from chgnet.model.dynamics import EquationOfState
import numpy as np

eos = EquationOfState()
eos.fit(structure) # the structure get from mp-134 on Materials project
print(eos.get_bulk_modulus(unit='GPa'))

The above code gives me 62.086 GPa (or 0.3875 eV/A^3) using CHGNet 0.3.0

This is very similar to the bulk modulus of Al (mp-134) on Materials Project

[bowen-bd]

CHGNet should have reasonably good extrapolation for structures not in Materials Project, the expected error is shown in the validation and test set error (which we didn't use to train the model), and also Matbench Discovery.

[shivang-22]

Okay. I see that CHGNet has an inbuilt capability to calculate the bulk modulus using EOS, but is there a capability to evaluate other similar properties like shear modulus, Young's modulus, or yield strength?

[janosh]

Have a look at matcalc.

[shivang-22]

Thanks for the quick reply. But from what I understand, that too calculates only the bulk modulus using the Birch-Murnaghan EOS. I'm trying to evaluate other similar mechanical properties using CHGNet, if possible.

[janosh]

ElasticityCalc gives you the full elastic tensor.