Spectroscopic Limited Maximum Efficiency-First Progressive Screening of Halide Double Perovskites via First Principle-Constrained Machine Learning

The traditional data-driven screening framework for halide double perovskites (HDPs) used in optoelectronic devices suffers from key bottlenecks, including low efficiency, a large candidate space, and insufficient multi-property balancing. This study presents an ab-initio-gated, spectroscopic limited maximum efficiency (SLME)-first cascade sieve that achieves high-precision discovery of HDPs suited for photovoltaic materials of absorber layers through a progressive decision chain that strictly follows the order of space group symmetry, SLME, bandgap characteristics, and thermodynamic stability for screening. This strategy achieves a revolutionary 109-fold acceleration compared to an end-to-end density-functional theory (DFT) screening workflow for a single material while maintaining physical interpretability. The framework rapidly and accurately identified only 34 high-potential HDPs from 29,364 initial candidate materials. Through DFT validation, we confirm three outstanding materials: Rb2TIPBr6, Cs2TIPBr6, and Cs2GePbBr6, which exhibit suitable direct bandgaps (1.2−1.8 eV), high SLME (>28%), broad spectral absorption, and thermal stability up to 600 K. Beyond material discovery, this work establishes a scalable paradigm for targeted interface photovoltaic material design by integrating explainable machine learning with domain-specific physical constraints.