Question about the theory of free energy minimization

[biglinn]

Dear developers,

I have learned the theory of free energy minimization. I briefly summarize it as follows:

I believe that the final Phi is the dyn matrix output by dyn.save_qe.
Moreover, is the final R saved? Could it represent the equilibrium configuration at the specified temperature？

[mesonepigreco]

Hi, yes, the dyn.save_qe saves both $\Phi$ and $\mathcal R$, and, indeed, it represets the equilibrium configuration at the specified temperature.
To access the structure, you can load the dynamical matrices saved at the end of the minimization and export it as

dyn = CC.Phonons.Phonons("my_dyn")
dyn.structure.save_scf("my_structure.scf")

The structure will be saved in the standard espresso file format. You can also save it in cif via

dyn.structure.save_generic_file("my_structure.cif")

As the "CIF" extension, all standard file formats supported by ASE should work out of the box.

[biglinn]

Thank you! I get it.

And I have a question about the density matrix $\rho$.
I find that SSCHA directly give its expression, like this:

In this, the $\Psi$ is:

Could you recommend me an acticle with derivation of the expression?

[biglinn]

Hi, there is not function save_generic_file in the class Structure.Structure, but read_generic_file exits. Dose SSCHA not support cif format now?

[biglinn]

Hi, @mesonepigreco

I have a question about performing DFT calculations for minimization with different DFT packages.
I have used QE and VASP to perform DFT calculations respectively.

Using QE:


The gradient converges by 3 populations, each population with 20 configurations.
Although the effective sample size of each population have reached the threshold controlled by kong-liu parameter, it doesn't matter.

Using VASP:


When I use VASP, the free energy and FC gradient show great discontinuity between two populations. The convergence is not reached with 4 populations. Moreover, its frequency evolution is completely different from that by QE.

Why does this happen with VASP?

[biglinn]

When using VASP, I just convert the unit from eV to Ry (eV/A to Ry/Bohr, eV/A^3 to Ry/Bohr^3).
But I find that energies of QE and VASP have large difference.
Qe is like this:

VASP is like this:

Although the absolute value of energy is meaningless, is there something else to change?

[biglinn]

Oh my god! It's my fault!
I change the order of atoms when transform pw.in to POSCAR, so that the order of atomic forces in forces_populationX_X.dat don't correspond to the order of atomic displacement in u_populationX_X.dat

[Guodonglin-cqust]

Oh my god! It's my fault! I change the order of atoms when transform pw.in to POSCAR, so that the order of atomic forces in forces_populationX_X.dat don't correspond to the order of atomic displacement in u_populationX_X.dat

The atomic positions generated by QE are not in the same order as those in the POSCAR, leading to incorrect forces and an increase in free energy rather than a decrease. Do you have any methods to resolve this issue? Thank you.

[biglinn]

I dont have a direct method to solve it. So I write scripts to record the order of atomic positions in the pw.in and transform back after extracting energy, forces from OUTCAR.

[Guodonglin-cqust]

when you use the script, after VASP calculations, dose the free energy decrease with the steps? Would you mind sharing the script you wrote? Thank you.