EPW Forum

In my work,I have to do structure relaxation using vc-relax .Before that i converge k-ponits and ecutwfc.But during vc-relax i use more denser k-point.But if i set the convergenge threshold like below,
forc_conv_thr = 1.0d-6
etot_conv_thr = 1.0d-6
conv_thr = 1.0d-9
/

&IONS
ion_dynamics = 'bfgs' ! Algorithm to move atoms
/

&CELL
cell_dynamics = 'bfgs' ! Algorithm to change cell shape
press = 0.0 ! Target pressure (usually 0 kbar)
press_conv_thr = 0.0 ! Target pressure convergence (in kbar)

The bfgs is always failed .
Then using the converged k-point ,cecutwfc and relax position.I almost in 90% ,i got negative phonon.

Is there any method available to relax the structure?So I get positive phonon.

Hi,

First would be helpful if you can share inputs/outputs to see what potential issues there might be. What you mentioned seems possible to be either calculation parameter related, or system related, or both.

Regarding the relaxation:
-When you say BFGS always fail, do you mean the cell relaxation never finish? How large are the forces/stress for the structure you used to evaluate phonons?
-It is possible that negative phonon can be related to underconverged phonon calculations - for example, could be related to tr2_ph in ph.x input. Again, it would be helpful to see the inputs/outputs.
-If you are seeing the negative frequencies for acoustic modes at q=0, it might just be related to acoustic sum rule.

Otherwise if it is not convergence issues, it is also completely possible that your result is correct, but your structure is a local minimum. For example this is very common in perovskites. Atomic relaxation (zero force/stress at equilibrium) does not guarantee dynamical stability (meaning you atoms stably vibrates near the equilibrium position). If that's the case, something to look at would be the special displacement method, see the last tutorial of Friday: https://epw2024.oden.utexas.edu/74-schedule.

Best!
Xiao