Intrinsic structure of relaxor ferroelectrics from first principles

We introduce FIRE-Swap, a first-principles-based framework for sampling representative compositional structures in complex perovskites using machine-learning interatomic potentials. By combining Monte Carlo chemical swaps with geometric relaxation, the method treats chemical and geometric degrees of freedom concurrently. Applied to lead magnesium niobate (PMN), lead zirconate titanate (PZT), and lead strontium titanate (PST), FIRE-Swap robustly predicts rock-salt-like chemical order in PMN, but not in the isovalent solid solutions PZT and PST with the same mixing ratio. For PMN, it further reveals anchored Nb clusters that host interconnected, noncoarsened polar nanoregions. These results provide a mesoscale structural basis for understanding relaxor ferroelectricity and establish FIRE-Swap as a practical route toward systematic, first-principles modeling of complex perovskites and related substitutionally disordered materials.