A major bottleneck in low-scaling energy-based fragmentation methods is the need for manual intervention in the fragmentation step, which is time-consuming, inconsistent, and hard to generalize across diverse molecular systems. To address this challenge, we develop a template-based automatic fragmentation algorithm that extends the generalized energy-based fragmentation (GEBF) approach to a wide range of large and complex molecules. A hierarchical SMILES-encoded GEBF template library for both cyclic and acyclic functional groups enables chemically meaningful and efficient partitioning via structure conversion, macrocycle detection, substructure matching, and small-fragment merging. Controlling fragment sizes ensures a balance between accuracy and computational cost, while user-defined templates offer enhanced flexibility. Benchmarks on biomacromolecules, macrocycles, porous organic cages, polyamide oligomers, and ionic liquids reproduce conventional quantum-chemistry results within a few kcal · mol−1 (or sub-meV/atom), while reducing the largest subsystem basis size to less than one-third of the full system. The accuracy of the GEBF forces is further validated, enabling reliable geometry optimizations and spectroscopic predictions with near-experimental agreement. Large polyamide oligomers with ≈1500 atoms can be computed within practical timeframes. The method also predicts reaction barriers and reaction energies for enzyme-catalyzed reactions at the level of electron correlation. This work paves the way for fully automated, scalable, low-cost, high-accuracy quantum chemistry, bridging theory and large-scale real-world applications.
Wang et al. (Mon,) studied this question.