Forecasting System of Computational Time of DFT/TDDFT Calculations under the Multiverse Ansatz via Machine Learning and Cheminformatics

With the view of achieving a better performance in task assignment and load-balancing, top-level designed forecasting system for predicting computational times density-functional theory (DFT)/time-dependent DFT (TDDFT) calculations is presented. The time assumed as intrinsic property molecule. Based on this assumption, established using “reinforced concrete”, which combines cheminformatics, several machine-learning (ML) models, framework many-world interpretation (MWI) multiverse ansatz. Herein, cheminformatics used to recognize topological structure molecules, ML models are build relationships between topology cost, MWI hold various combinations functionals basis sets DFT/TDDFT calculations. Calculated results molecules from DrugBank dataset show that (1) it can give quantitative predictions costs, typical mean relative errors be less than 0.2 with derivations ±25% exactly pretrained (2) also employed functional set cases without while only slightly enlarge errors.