Compact multipolar representation of the electrostatic potential for flexible molecules

A new method for generating a compact multipolar representation of the electrostatic potential (EP) for flexible molecules is presented. The method is based on a constrained minimization of the difference between the quantum mechanical and the classical EP. The fitting procedure used adopts the least absolute shrinkage and selection operator technique [R. Tibshirani, J. Roy. Stat. Soc. B 58, 267 (1996)] which can be seen as penalized ordinary least squares. The penalty function optimized for the particular molecule of interest effectively removes redundant multipoles. It is shown that the use of multiple conformations is crucial for the predictive ability of the EP model for flexible molecules. The multipole local coordinate systems are chosen in a way that best reflects the key conformational changes. It was demonstrated that such an approach improves the predictive ability of EP models. It also allows to exploit equivalence of atoms in the calculation of multipoles components. In the case of polar flexible molecules, the augmentation of the EP model based on charges by higher multipoles decreases the relative root mean square error by a factor of 1.5-5. The corresponding effect of enlargement of the set of multipoles was significantly reduced.