Machine learning the computational cost of quantum chemistry

Abstract

Computational quantum mechanics based molecular and materials design campaigns consume increasingly more high-performance compute resources, making improved job scheduling efficiency desirable in order to reduce carbon footprint or wasteful spending. We introduce quantum machine learning (QML) models of the computational cost of common quantum chemistry tasks. For single point, geometry optimization, and transition state calculations the out of sample prediction error of QML models of wall times decays systematically with training set size. We present numerical evidence for thousands of organic molecular systems including closed and open shell equilibrium structures, as well as transition states. Levels of electronic structure theory considered include B3LYP/def2-TZVP, MP2/6-311G(d), local CCSD(T)/VTZ-F12, CASSCF/VDZ-F12, and MRCISD+Q-F12/VDZ-F12. In comparison to conventional indiscriminate job treatment, QML based wall time predictions significantly improve job scheduling efficiency for all tasks after training on just thousands of molecules. Resulting reductions in CPU time overhead range from 10% to 90%.

Ab Initio Machine Learning in Chemical Compound Space

“…For the QML models of the computational cost of typical quantum chemistry computations (measured by the CPU wall time), Heinen et al. reported the QMt data set, 380 consisting of timings of various tasks (single point energy, geometry optimization, and transition state search) for thousands of QM9 molecules at several levels of theory including B3LYP/def2-TZVP, MP2/6-311G(d), LCCSD(T)/VTZ-F12, CASSCF/VDZ-F12, and MRCISD+Q-F12/VDZ-F12.…”

Machine Learning na Física, Química, e Ciência de Materiais: Descoberta e Design de Materiais

“…Como mostrado na seção anterior, o MLé uma ferramenta matemática e estatística, podendo então ser aplicada aos mais diferentes tipos de problemas. Nesse sentido, não apenas os problemas científicos em si podem ser estudados, como também tarefas que usualmente não são estudadas mas que existem dados disponíveis, tal como estimar o tempo que um processo computacional vai levar, permitindo sua otimização [47]. No contexto específico de materais, o ML pode ser usado para a descoberta, design e otimização de propriedades tanto partindo de dados experimentais [48][49][50] como de simulação, e esta pode ser atomística (clássica) ou ab initio (quântica).…”

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

“…However, as far as we know, there is nearly no related work concerning to the prediction of computation cost in the field of quantum chemistry, except in the area of quantum machine learning (QML) models that very recently introduced by Heinen and co-workers. 14 They demonstrated that QML-based wall time predictions significantly improve job scheduling efficiency by reducing CPU time overhead ranging from 10 to 90%. Until now, there has been no universal solution for predicting the computational cost.…”

“…The current QML solution is restricted to a specified computational approach and specified parameters, and training of a corresponding ML model is essential before practical predictions. 14 However, it may not be convenient for training the specific model each time before the practical calculations. Thus, generalization ability should be one of the essential elements for an universal solution when predicting the computational cost.…”

“…Additionally, the traditional ML models including the ensemble learning, 38 recurrent, 39 or graph-based neural network 40 may also be the good alternatives to the reported QML solutions. 14 …”