From empirical formulas to machine learning: A comprehensive review of detonation velocity prediction for advanced energetic materials

• This article surveys detonation velocity prediction for ideal and non-ideal energetic materials. • It compares classical empirical formulas with modern machine learning frameworks. • It demonstrates how theoretical descriptors enable virtual screening of explosives. • It evaluates the impact of molecular architecture on predictive model accuracy. • It provides a practical decision tree for choosing the best modeling strategy. The predictive modeling of detonation velocity is a cornerstone of advanced materials design, offering a cost-effective and safe alternative to the hazardous experimental characterization of novel energetic compounds. This analysis bridges the gap between traditional empirical correlations and emerging machine learning architectures, assessing how structural descriptors and quantum-chemical parameters dictate macroscopic explosive performance. While established empirical formulas provide rapid utility for standard CHNO-based scaffolds, their reliability diminishes when applied to the complex, heterogeneous formulations required for modern aerospace and defense applications. In response, data-driven ensemble learning models have demonstrated superior capacity for the virtual screening of candidate molecules, effectively mapping molecular architecture to detonation kinetics with high fidelity. Integrating these predictive methodologies into a robust decision framework enables a more sustainable workflow for the rational design and structural optimization of high-performance energetics