Geometry optimization

Python

The DE algorithm here is rewritten from scratch in what we call a generation mixing form. It allows us to save memory by storing the parameter space vectors from different generation in the same array. Tests have shown that this does not affect the performance of the DE algorithm. In addition, the DE iterations stop when the convergence is achieved to avoid extra evaluations. In the given case, convergence is defined as the spread of the cost function values for all parameter space vectors being less than some threshold value.

The python file contains the main functions QEGenrun(cost function) and QEDE, and the cell for a test run of the code on bulk Si followed by a few other cells that are used for the output results analysis such as visualizing and animating the DE convergence. The QEDE can perform optimization for an arbitrary number of parameters including integer ones, i.e. those for which only integer part is meaningful. The integer parameters are set by int_param_num option and, by convention, they are being counted from the end of a parameter space vector. Therefore, integer parameters must be placed accordingly in the parameter limits and placeholders arrays. A template file for the bulk Si with parameter placeholders starting by convention from '@' is provided.

Bash

This release also contains a bash script that creates the template of QE input file and a python script and then runs the python script. Once the python script run is finished, the temporarily created files are deleted leaving only the resulting QEDE*.out and QEDE*.png output files. The bulk Si is taken as an example with the three optimization parameters: ecutwfc, celldm and kpt. For kpt parameter only its integer part is meaningful.