GAP Training tutorial fail to make potential

[ladyteam]

Hi, I was trying to follow instructions in Training a GAP tutorial, but when trying to generate potential with gap_fit energy_parameter_name=energy force_parameter_name=forces do_copy_at_file=F sparse_separate_file=T gp_file=GAP.xml at_file=train.xyz default_sigma={0.008 0.04 0 0} gap={distance_2b cutoff=4.0 covariance_type=ard_se delta=0.5 theta_uniform=1.0 sparse_method=uniform add_species=T n_sparse=10}
This results in error

================================= End of report ================================
SYSTEM ABORT: Traceback (most recent call last)
File "/work/QUIP-build-gcc/src/GAP/gap_fit_module.f95", line 1252
kind unspecified
fit_data_from_xyz: too small virial_sigma (.00000000000000000E+000), should be greater than zero
STOP 1
I compiled current git version with linux_x86_64_gfortran arch description. It seems like a problem reading energies E0/atom = array mostly filled with zero values.

[albapa]

It looks we haven't updated the tutorial when we added a check for the sigma values. The code is complaining about default_sigma={0.008 0.04 0 0} - in particular the last two zeros. Replace them with positive numbers. Do you have any virial stress targets? If yes, choose an appropriate regularisation (sigma) value for them. This is the third number in the default_sigma. If not, any positive number will do. For the 4th number, which represents the regularisation used for Hessian targets, which is an experimental feature, use any positive number. I hope this helps.