auto vasper

this project is for using vasp(Vienna Ab initio Simulation Package) easily with python.

There are two aims of this project

Analysis the results of vasp calculation

The recent research has shown the ability of optimal combination algorithm in searching the ground state energy of given components. We can use these algorithms to avoid enumerating all the structures of a given supercell.

demoNiAl.py is a demo script for searching ground state energy with the auto-vasper. The brief workflow is as follows:

read the results from atat-vasp calculation. This will write the record to a given ASE atom database file.

database = connect("NiAl.db")

if not os.path.exists("NiAl.db"):
    VaspIO.archive_atat_record(
        "vasp_rundir", # demo results from atat-vasp
        database
    )

obtain the primitive structure
constuct the icet ClusterExpansion Object

model = LassoCV(fit_intercept=False)
y = sc.get_fit_data(key='dft_energy')[1]
X = sc.get_fit_data(key='dft_energy')[0]
model.fit(X, y)
ce = ClusterExpansion(cluster_space=cs, parameters=model.coef_)

search the ground state by EA solver

Notes that the referenced paper&code have already implement the function for searching the lowest energy on the whole phase diagram

here I also implement the part for searching ground state on specific component ratio. The mode can be switch by the parameter 'cross_element'.

get_ground_state(cluster_space=cs,
                cluster_expansion=ce,
                atoms=supercell,
                regression_model_weight=model.intercept_,
                elements = atom_type,
                elements_label = chemical_symbols,
                generation=100, 
                population=100,
                cross_element=False,
                element_ratio=[5,3])

Here the parameter 'element_ratio=[5,3]' means the atoms amount of 'Ni' and 'Al' in the superstructure is 5:3.

Use ICET with Vasp (testing)

ICET declares that their structure derivativing algorithm is purposed by Gus L. W. Hart and Rodney W. Forcade. It could be faster than the FWZ algorithm using by atat accroding to this article : Algorithm for generating derivative structures.

So, I want to test if ICET can work with Vasp instead of ATAT. In demoAgPd.py. There are two step in the script as follows:

Preparations. The POTCAR is search from 'template/Potentials/'. POSCAR and KPOINTS is generated automatically. The INCAR cannot be generated by now, the auto-vasper will copy it directly from 'template/INCAR'

mkdir template/Potentials
cp -r PATH_TO_PBW_POTENTIALS_IN_YOUR_COMPUTER/* template/Potentials/
vim template/INCAR
# set the content of the incar file

defining the primitive structure. (ASE Atom object)

atoms = ['Ag', 'Pd', 'Pd']

sites = np.array([
    [0.0, 0.0, 0.0],
    [0.33333333333333326, 0.3333333333333334, 0.6666666666666667],
    [0.6666666666666666, 0.6666666666666667, 0.3333333333333333]]
)
lattice = np.array([
    [0.0 ,2.045, 2.045], 
    [2.045, 0.0, 2.045], 
    [2.045, 2.045, 0.0]]
    )
lattice, atoms, sites = to_primitive_structure(lattice, atoms, sites)
primitive_structure = Atoms(symbols=atoms,
                        positions=sites @ lattice,
                        cell=Cell(lattice), pbc=[True, True, True]
                    )

loop the vasp calculation from size of primitive structure to max_size. when the calculated samples for vasp is equal to max_cal_sample, the size will increase.

StandardSolver.vasp_loop(
        primitive_structure, 
        vasp_workdir, 
        max_size, 
        max_cal_sample, 
        db,
        cutoffs=cutoffs)

irumeria/auto-vasper

auto vasper

Analysis the results of vasp calculation

Use ICET with Vasp (testing)