Include Grüneisen parameter (thermal expansion) in benchmark?

[ltalirz]

The latest version of the method paper for $k_{SRME}$ (July 2025) includes an analysis of the Grüneisen parameter $\gamma$ that enters the coefficient of thermal expansion.

It notes

having accurate conductivity predictions does not necessarily imply having accuracy in predictions for the Grüneisen parameter, and vice-versa

and

while fMLPs generally tend to systematically underestimate the conductivity, there is no such bias for the Grüneisen parameter. Overall, our findings highlight that the macroscopic κ and γ have different sensitivity to atomistic vibrational properties

The CTE is relevant for many technological applications - would it be possible / make sense to include $\gamma$ in the benchmark?

[janosh]

thanks @ltalirz! i think that's a great suggestion that also crossed my mind and which i'm very open to 👍

don't want to act on this unilaterally though. plus there are several ideas floating around on how to bring MLIP benchmarking to the next level so it's really up to @MSimoncelli @bpota @gabor1 how to best integrate a Grüneisen benchmark into that

[gabor1]

We already have a phonon tab in the new system, I'd really like that to grow to include thermal conductivity and Grüneisen

[JaGeo]

Outsider view: i do like the idea of Grüneisen parameters as well. Reference data is also much cheaper as for thermal conductivity. 😃

[MSimoncelli]

Thanks everyone for the feedback! Indeed it could be integrated, just we should discuss how to combine it into CPS ecc

[JaGeo]

Actually, I am wondering one more thing: Could we have mean/max phonon frequencies as well as part of the benchmark (like we did in the mace-mp-0 paper)? While I assume that most of the models get this right but it would add some additional information and might be useful for physical model development.

[janosh]

surprisingly, many models do not get $\omega_\text{min}$ right. i think that's an interesting metric because it exposes frequent spurious dynamic instabilities predicted by several models

[JaGeo]

Ah, that's very true! Maybe also possible to include "min" or at least dynamic stability!