Internal Benchmarking of Semi-Empirical Methods: Bromine-Containing Crystals as a Sensitive Test Case

Selecting appropriate computational methods for organic crystals becomes particularly challenging for systems containing heavy halogens like bromine, whose complex electronic structures and diverse non-covalent interactions challenge approximate methods. Here we benchmark periodic DFT (PBE-D3BJ), CrystalExplorer (CE17/CE21), DFTB3-D3BJ, and PM7 against experimental stability data for 14 chlorine- and bromine-containing polymorphs across six CSD families. Chlorine systems show method-consistent performance, but bromine introduces large lattice energy variations (>10 kJ/mol) and, in several cases, qualitatively wrong stability rankings. Crucially, low mean absolute errors do not ensure correct thermodynamic ordering, and no semi-empirical method proves universally reliable for bromine. Only PBE-D3BJ achieves perfect experimental agreement. These results position bromine-containing crystals as exceptionally sensitive benchmarks and emphasize internal validation against experiment or reference DFT as essential before large-scale studies-particularly timely as machine-learning potentials emerge.