Fitting GAP

In this tutorial, you will learn to write a bash-script that fits GAP on your DFT data. GAP learns the energies and forces as a function of the atomic positions. It is implemented as a command line tool gap_fit which takes many parameters. The parameters are specified in the following, and then passed to gap_fit

First, we define our parameters which we want to vary later In this case, that’s the Cut-off of the atomic neighborhood, the output file and the input file.

CUT_OFF=4.0
GAP_FILE=cut_off_4A/SOAP.xml
INPUT_FILE=train.xyz

Next, we have the parameters for GAP.

GAP_PARAMS=(

First, we have the parameters for the 2 body distance descriptor

distance_2b
cutoff=$CUT_OFF
covariance_type=ard_se
delta=1
theta_uniform=1.0
sparse_method=uniform
n_sparse=300
Z1=18
Z2=18 :

Then we have the parameters for the SOAP descriptor

soap

n_max and l_max is the order of expansion in spherical harmonics

l_max=6
n_max=6

atom_sigma is the smearing of the atomic position

atom_sigma=0.5
zeta=4

Then we specify the cut-off for the neighborhood around the centered atom

cutoff=$CUT_OFF
cutoff_transition_width=0.5

and some magic settings

covariance_type=dot_product
n_sparse=300
sparse_method=random
delta=1.0

Finally, we specify the number of different species and the atomic charge number of the species

n_Z=1 Z={18}

)

Next, we have some general parameters

GEN_PARAMS=(
e0_method=average
gap={"${GAP_PARAMS[@]}"}

Here we specify the input file (the atomic coordinates and forces) and the output file (the machine learned potential)

at_file=$INPUT_FILE
gp_file=$GAP_FILE
default_sigma={0.003 0.15 0 0}
sparse_jitter=1.0e-10

and then we have some parsing options for the input file

force_parameter_name=forces
energy_parameter_name=energy
)

Finally, we print our parameters and then run the gap_fit command

echo ${GEN_PARAMS[@]}

gap_fit "${GEN_PARAMS[@]}"

To run this script open the terminal, navigate to the folder your_project/gap and execute the command bash gap_fit.sh.