Skip to content

Latest commit

 

History

History
97 lines (73 loc) · 3.24 KB

utilities.md

File metadata and controls

97 lines (73 loc) · 3.24 KB
Raw

Pypolymlp utilities

Structure generation for DFT calculations

  1. Prototype structure selection

Prepare prototype structures in POSCAR format.

  1. Random structure generation
  • Generation from prototype structures
> pypolymlp-structure --random --poscars prototypes/* --n_str 10 --low_density 2 --high_density 2

Structures for are listed in polymlp_str_samples.yaml. Structures are generated in "poscar" directory.

  • Generation from a given structure
> pypolymlp-structure --random --poscars POSCAR --n_str 10 --low_density 2 --high_density 2
  • Random displacements for phonon calculations (phonopy is required.)
> pypolymlp-structure --random_phonon --supercell 3 3 2 --n_str 20 --disp 0.03 -p POSCAR

Structures are generated in "poscar_phonon" directory.

  1. DFT calculations for structures

Compression of vasprun.xml files

> pypolymlp-utils --vasprun_compress vaspruns/vasprun-*.xml

Compressed vasprun.xml is generated as vasprun.xml.polymlp.

Automatic division of DFT dataset

> pypolymlp-utils --auto_dataset dataset1/*/vasprun.xml dataset2/*/vasprun.xml
> cat polymlp.in.append >> polymlp.in

A given DFT dataset is automatically divided into some sets, depending on the values of the energy, the forces acting on atoms, and the volume. A generated file "polymlp.in.append" can be appended in your polymlp.in, which will be used for developing MLPs. Datasets identified with "train1" and "test1" are composed of structures with low energy and small force values. The predictive power for them is more important than the other structures for the successive calculations using polynomial MLPs, so the prediction errors for "train1" and "test1" datasets should be accuracy measures for polynomial MLPs.

Atomic energies

(Experimental: Only for VASP calculations using PBE and PBEsol functionals)

> pypolymlp-utils --atomic_energy_elements Mg O --atomic_energy_functional PBE
> pypolymlp-utils --atomic_energy_formula MgO --atomic_energy_functional PBE
> pypolymlp-utils --atomic_energy_formula Al2O3 --atomic_energy_functional PBEsol

A standard output can be appended in your polymlp.in, which will be used for developing MLPs. The polynomial MLP has no constant term, which means that the energy for isolated atoms is set to zero. The energy values for the isolated atoms must be subtracted from the energy values for structures.

Estimation of computational costs

calc_cost option generates a file 'polymlp_cost.yaml', which is required for finding optimal MLPs.

  1. Single polynomial MLP
> pypolymlp-utils --calc_cost --pot polymlp.lammps

# hybrid polynomial MLP
> pypolymlp-utils --calc_cost --pot polymlp.lammps*
  1. Multiple polynomial MLPs
> pypolymlp-utils --calc_cost -d $(path_mlps)/polymlp-00*

Enumeration of optimal MLPs on convex hull

> pypolymlp-utils --find_optimal Ti-Pb/* --key test-disp1

Files 'polymlp_error.yaml' and 'polymlp_cost.yaml' are needed for each MLP.