How to lower the RESIDUAL and fix the problem with α selection #326

dear developer,

During my calculation of anharmonic force constants using ALAMODE combined with AIMD, I encountered an issue of excessively high RESIDUAL values. I attempted to extend the AIMD simulation steps: the number of sampling snapshots was increased from the original 80 to 160, with a sampling interval of 50 fs and random displacement amplitude set to 0.04 Å. This adjustment reduced the RESIDUAL from 62.2275% to 40.5289%. I further expanded the cutoff radius to 30 Bohr and included fifth-order anharmonic force constants in the fitting.

However,` the cutoff radius of 30 Bohr now exceeds the lattice length of the supercell (24.216819646 Bohr). I am unsure whether this setup is physically valid.

In` addition, during the CV cross-validation stage, repeated calculations consistently yielded the minimum CVSCORE at alpha = 1.58489e-05. In theory, this alpha value should produce the minimal subsequent RESIDUAL, yet the actual result contradicts this expectation. After obtaining the optimal alpha of 1.58489e-05 via CV cross-validation, I manually modified the corresponding parameter in the opt.in file to 1.58489e-06; the resulting RESIDUAL dropped to 24.071%. This inconsistency confuses me greatly.

Could` you advise on the subsequent workflow to obtain physically reliable anharmonic force constants?

I am also attaching input files and output files below.

1opt.log

2opt.log

3opt.log
Best regards,
Tao