cause of non-convergence in SCPH calculation.

[Gi-HoonKang]

Hello. I'm sorry to bother you, but I have a question about the cause of non-convergence in SCPH calculation.
I obtained non-convergence result from my SCPH calculation. The input was as below.

&general
PREFIX = WH-sc4-V1H2-meta_scph
MODE = SCPH
NKD = 2; KD = W H
FCSXML = WH-sc4-V1H2-meta-random.xml
TMIN =0; TMAX = 2000; DT = 100
/

&scph
SELF_OFFDIAG = 0
MAXITER = 500
MIXALPHA = 0.2
KMESH_INTERPOLATE = 1 1 1 #same with the supercell size
KMESH_SCPH = 1 1 1 # should be multiple of KMESH_INTERPOLATE
/

&cell
1.8897259886 # factor in Bohr unit
1.2686920000000001e+01 0.0 0.0 # a1
0.0 1.2686920000000001e+01 0.0 # a2
0.0 0.0 1.2686920000000001e+01 # a3
/

&kpoint
2
1 1 1
/

I already knew that decreasing "MIXALPHA" or "DT" can improve the convergence from ALAMODE seminar. However, I am also curious about the other reasons. Is there any possibilities that inaccurate force constants or low training dataset (related to compressive sensing method) cause the non-convergence?
I attached the output of SCPH calculation just in case. I always appreciate your help.

scph.log

[ttadano]

The bad convergence may result from the inaccuracy of fourth-order IFCs. I would suggest increasing the training data for compressive sensing and checking if the convergence rate improves.

From your scph output, the calculation converged at 0 K and 200 K, but not at other temperatures. If you set LOWER_TEMP = 0 in the &scph field, the convergence can improve hopefully.

[Lamed-git]

I recently met a problem with non-convergence in SCPH calculations too. I double-checked the input file and found that I had written the wrong cell parameters when generating the force constants with alm program, so I hope you can check the correctness of the force constant file.