Model performance test: Failed to fit the equation of state

[walk-for-me]

I trained a titanium-hydrogen binary potential function with DeepMD-Kit v3.0.0, and the training ran normally.
The input file is as follows：
input.json.
The running log is as follows:
out.log.
I then used it to fit the equations of state for titanium and titanium hydrides, which include:
.
Calculate the energy of 'each structure obtained by scaling the POSCAR coefficient' and average it over each atom. Further comparison with DFT results shows that titanium hydride is in good agreement, but the results of pure titanium are not ideal.


What would you like to say about this problem?

[wanghan-iapcm]

The different Ti phases should be sufficiently represented in your training data, the ML potentials can not "extrapolate" to the phases that they have never been trained with.