Abstract
Lattice constants in GdFeO
3-type ABO
3 perovskites are correlated to their constituent elemental properties by using linear regression (LR) and artificial neural networks (ANN) techniques and a sample set of 157 known GdFeO
3-type ABO
3 perovskites. LR models are first obtained using two elemental ionic radii only and ANN models, using five elemental properties; ionic radii, electronegativities of cation A and B, and the valence of ion A, are further developed to improve the model predictability, which reaches an error limits of less than 2%. It is shown that lattice constants of these compounds only roughly correlate to their ionic radii, and for a good prediction model 3 more elemental properties (electronegativity and valence) are necessary. In new materials research, where lattice constant is one of the key design target, the developed LR and ANN models may be used to screen and shortlist promising perovskites from a large pool of all possible candidates. These selected compounds may undergo further test using relatively more expensive experiments or quantum mechanics computations.