Determine the Number of Crystallographic Sites from Chemical Formula Alone

Determine the number of crystallographic sites directly from a chemical formula for disordered crystal structures, where the summed non-vacancy stoichiometry may not equal the number of crystallographic sites.

Background

EP-Flow performs formula-conditioned generation for disordered crystals, but its generation process depends not only on global stoichiometry but also on the number of crystallographic sites. In the current framework, the site count is either specified explicitly or sampled from the empirical distribution of the benchmark setting.

For disordered structures, fractional occupancies and vacancies mean that summing the non-vacancy stoichiometric coefficients does not necessarily recover the number of crystallographic sites. The paper therefore identifies determining the site count from the formula alone as an unresolved problem, which would remove an additional conditioning requirement from disordered crystal structure prediction.