I’m a newbie on ATAT and I’m trying to use maps to discover the wurtzite ZnOS alloy, in which partial O are substituted by S. I relaxed all structures generated by maps one by one and wrote the cell energy of each structure into the energy file.The force convergence threshold is 1.e-3 Ry/Bohr. What I confused is that the Crossvalidation Score didn’t decrease with including more structures. And the groud state for wurtzite ZnO and wurtzite ZnS show more divergence between fitted energy and calculated energy. Especially for ZnS, the difference between fitted and calculated energies becomes 80 meV, as shown in the figure. Is it possible to reduce Crossvalidation Score by including more structures or other way? The manual shows the weight.in file can be inputted to finely tune the ground states. But there isn’t more infomation about how to do that. the last question is that if I want to get the structure with the lowest energy predicted by maps in predstr.out file, I should use genstr to generate all structures. My lat.in is following. Please help me to solve my problems.
On issue that can slow down convergence is when structure relax far away from the starting structure.
Try the command
checkrelax
and do a
touch nn/error
for any structure nn with a relaxation value of more than, say 0.1 strain
Other than that, adding more structure is the only way to reduce cv score.
The weight.in file could help you fit specific structures better.
Just specify one weight per line (in the same order as in fit.out) and the code will use that. You would give a height weight the ZnS in your case. The CV score may increase however.
BTW, double check if any of your runs didn’t by any chance abort abnormally. the ZnS energy seems really out of the trend.
I’m not sure where is the problem. The calculation of ZnS is alright without any error. I redo the calculation of ZnS with double k mesh and it reduces the total energy of about 10 meV. Then I restarted maps and this time it gave the reasonbale ground states with CV of 0.0077 eV. I start maps -g=32 to make maps search ground states up to 32 atoms per cell and it gave some ground states with negative formation energy. What should I do if I want to comfirm these ground states with ab inition calculation? Should I just touch ready and wait maps to give out the structure? Will maps give out the ground states structure? Or I should use the other method to figure out it?