What are descriptors?

Note

In materials science, descriptors are representations of the qualitative or quantitative composition, and/or structure of compounds or alloys, which play an important role in machine learning of materials data.

Warning

Descriptors are usually associated with the classic definition of machine learning features. However, there is a slight discrepancy between the definitions of materials scientists and data analysts.

In materials science, descriptors of certain properties of materials are understood as physical or statistical indicators of materials associated with a certain property. Often, these indicators have a characteristic functional form strictly for certain materials, for example, only for cubic crystals or only for insulators.

The first more or less universal descriptors for different properties of materials is a concentrational weighted average of atomic properties over chemical composition:

\[\begin{aligned} f(composition) = \sum^n_i c_i \cdot P_i \end{aligned}\]

To explain we need a portion of code. First, we are to parse chemical formula:

from pymatgen.core import Composition

f2c = lambda x: Composition(x).as_dict()

formula = 'Ni3Al'
composition_as_dict = f2c(formula)
print(composition_as_dict)
  

Output: {'Ni': 3.0, 'Al': 1.0}

def get_atomic_concentrations(formula: dict=None):
    n = sum(list(formula.values()))
    concs = formula.copy()
    for key in formula.keys():
        concs[key] = concs[key] / n
    return concs
  

def get_avg_prop(concentrations: dict=None, prop: str=None):
    ''' concentrations = dict obj '''
    avg_prop = 0
    try:
        for symbol, concentraion in concentrations.items():
            avg_prop += get_elem_props(symbol)[prop]*concentraion
        return avg_prop
    except Exception as e:
        print(f'while {symbol} is processed {e} is occured. NaN value will be returned!')
        return np.nan
  