No Headache for PIPs: A PIP Potential for Aspirin Runs Much Faster and with Similar Precision Than Other Machine-Learned Potentials

Abstract

Assessments of machine-learning (ML) potentials are an important aspect of the rapid development of this field. We recently reported an assessment of the linear-regression permutationally invariant polynomial (PIP) method for ethanol, using the widely used (revised) rMD17 data set. We demonstrated that the PIP approach outperformed numerous other methods, e.g., ANI, PhysNet, sGDML, and p-KRR, with respect to precision and notably with respect to speed [Houston et al., J. Chem. Phys. 2022, 156, 044120]. Here, we extend this assessment to the 21-atom aspirin molecule, using the rMD17 data set, with a focus on the speed of evaluation. Both energies and forces are used for training, and the precision of several PIPs is examined for both. Normal mode frequencies, the methyl torsional potential, and 1d vibrational energies for an OH stretch are presented. We show that the PIP approach achieves the level of precision obtained from other ML methods, e.g., atom-centered neural network methods, linear regression ACE, and kernel methods, as reported by Kovaćs et al. in J. Chem. Theory Comput. 2021, 17, 7696−7711. More significantly, we show that the PIP PESs run much faster than all other ML methods, whose timings were evaluated in that paper. We also show that the PIP PES extrapolates well enough to describe several internal motions of aspirin, including an OH stretch.

Assessing Permutationally Invariant Polynomial and Symmetric Gradient Domain Machine Learning Potential Energy Surfaces for H3O2–

“…The two PESs are similarly precise; however, the PIP PES is much faster to evaluate for energies and energies plus gradient than the sGDML one, with factors of 200 and 70, respectively. This factor of roughly 2 orders of magnitude is consistent with a similar factor found in a previous assessment for ethanol and a very new one for 21-atom aspirin . DMC calculations of the ground vibrational state wave function were done using both PESs.…”

“…This factor of roughly 2 orders of magnitude is consistent with a similar factor found in a previous assessment for ethanol 29 and a very new one for 21-atom aspirin. 35 DMC calculations of the ground vibrational state wave function were done using both PESs. Since these require just energies, the calculation using the PIP PES took roughly 300 times less CPU time than the sGDML one.…”

Fundamental Invariant Neural Network (FI-NN) Potential Energy Surface for the OH + CH3OH Reaction with Analytical Forces

“…It is noteworthy that the increasing number of monomials and polynomials with the system size has a smaller impact on the PIP method compared to the PIP-NN method. , For a single fit, the PIP method has been applied to the same class of systems as the PIP-NN and FI-NN methods and even to larger systems, − where PIP-NN and FI-NN methods have not yet been applied. Generally, the neural network (NN) architecture is more complex and has more parameters than the linear regression used in the PIP method.…”

Tell Machine Learning Potentials What They Are Needed For: Simulation-Oriented Training Exemplified for Glycine

“…Previous comparisons of ANI and PIP for similarly precise fits for ethanol indicate that the PIP PES runs roughly 10–100 times faster than the ANI PES based on the evaluation with the old version of MLatom . This factor is consistent with a more recent assessment of speed and precision for ANI and PIP PESs for aspirin …”

“…Performance enhancements of PIPs are also possible. One that has already been reported for the PIP PES for aspirin is the reduction of the PIP basis with very little loss in precision but a significant acceleration of both training and prediction.…”