How to perform free energy minimization well?

[biglinn]

Hi, @mesonepigreco:
I do SSCHA iteration with manually submission. But the result of minimization is not satisfactory and even strange. When I plot the info during the minimization with sscha-plot-data.py, the figure is confusing, as follows:

It's very different from Au's figure in the first tutorial, which is as follows:

Is it the reason for few minimization steps? Or is there other parameters about free energy minimization?

[mesonepigreco]

Hi, what is free energy, and what is its gradient? (Plotted in the other figure of sscha-plot-data.py)
If the gradient is small, the result converges close to the starting point. If the starting point was the harmonic dynamical matrix, it is the signature that your system is harmonic; thus, the SSCHA result coincides with the harmonic result.

[biglinn]

This is the first iteration:

This is the 5th iteration:

So, according to the gradient, does it mean that the result converges in the 5th iteration?
That's fantastic ! I used to always judge convergence by plotting phonon dispersion and observing change between two iterations.