BUG: elastic stress tensor instability

[chtchelkatchev]

After installing the latest version of atomate2, I noticed a big change in the results of elastic stress tensor calculations, compared to the results found in the previous version: diagonal elements of the elastic tensor increased by 15GPa in the latest automate2. MACE force field was used. 15GPa is a large mistake. Hopefully the issue has a solution. Files used for the calculation I attach. Of cause, I made other tests: the issue is stable.
results.tar.gz

[JaGeo]

Thank you very much for reporting the issue. We will try to track it down.

@gpetretto any ideas?

[JaGeo]

Or @esoteric-ephemera maybe? It could also be a stress problem again.

[chtchelkatchev]

Here for comparison I present for the same system VASP calculation of elastic stress using https://github.com/prnvrvs/elastic_vasp, PBE functional, pseudopotentials with "semicores" (INCAR and
vasp_elastic.tar.gz
KPOINTS made using https://muellergroup.jhu.edu/K-Points.html attached):

second order coefficients are as follow:
C11_C12_I/= 40.36400509687549
C12/= 539.822847581932
C44/= 51.67496079672538

===== Methodology =====
Bulk=(C11_C12_II/) x 2/9
shear=(C11_C12_I/) x 1/2
C11=(3xbulk+4xshear)/3
C12=(3xbulk-2xshear)/3
C44=2/3 x C44/

C11= 146.8699695272352
C12= 106.50596443035973
C44= 34.44997386448359
bulk_modulus= 119.96063279598489

==== stiffness matrix ====

[[146.87 106.51 106.51 0. 0. 0. ]
[106.51 146.87 106.51 0. 0. 0. ]
[106.51 106.51 146.87 0. 0. 0. ]
[ 0. 0. 0. 34.45 0. 0. ]
[ 0. 0. 0. 0. 34.45 0. ]
[ 0. 0. 0. 0. 0. 34.45]]

check stability criteria
C11−C12 > 0 ; C11+2C12 > 0 ; C44 > 0

�[1m�[1;32m structure is mechanically stable�[0m
===== Polycrystalline Average =====
K_voigt= 119.96 GPa
K_reuss= 119.96 GPa
K_VRH= 119.96 GPa

G_voigt= 28.74 GPa
G_reuss= 26.86 GPa
G_VRH= 27.80 GPa

E_VRH= 77.42 GPa
nu_VRH= 0.39
Pugh ratio= 4.32

[JaGeo]

I have another question:

• does this also happen with the default model in a similar way?

[chtchelkatchev]

The default mace model, trained on MP, is very inaccurate, producing diagonal elements of the elastic tensor ~80GPa (old atomate2 version). There are no HEA alloys in MP or smth similar...

[esoteric-ephemera]

Hey @chtchelkatchev: hard to say what the issue is when you're using a custom model. Is it possible that a concurrent update to mace-torch changed the behavior of your custom model?

For what it's worth, I get the same IEEE elastic tensor as your atomate15 results:

[[155.26919200503175, 116.78524306551743, 115.70601945984995, 0.321952086327638, -0.09031786462525584, 0.20932543394844486], [116.78524306551745, 155.75413141868546, 115.5687433938244, 0.2842888347738036, -0.1294715086539406, 0.41592329764601677], [115.70601945984998, 115.5687433938244, 156.92833870245556, 0.34971305863650415, 0.07430341175764595, 0.26944609523427265], [0.32195208632763817, 0.2842888347738036, 0.3497130586365042, 34.09329882576332, -0.3411876268440579, -0.4002136984038336], [-0.09031786462525584, -0.12947150865394053, 0.0743034117576459, -0.3411876268440579, 33.58993887719511, -0.2004357058455722], [0.20932543394844474, 0.4159232976460168, 0.2694460952342726, -0.4002136984038336, -0.20043570584557222, 33.523619543904275]]