Naslov
Neural Network Modelling of Potential Energy Surface of Methanoic Acid Spanned by Normal Coordinates

Autori
Sović, Karlo ; Pišonić, Zrinka ; Hrenar, Tomica

Vrsta, podvrsta i kategorija rada
Sažeci sa skupova, sažetak, znanstveni

Izvornik
Computational Chemistry Day 2022 : Book of abstracts / - Zagreb, 2022, 37-37

ISBN
978-953-6076-94-9

Skup
Computational Chemistry Day 2023

Mjesto i datum
Zagreb, Hrvatska, 24.09.2022

Vrsta sudjelovanja
Poster

Vrsta recenzije
Međunarodna recenzija

Ključne riječi
neural networks, potential energy surface, normal coordinates

Sažetak
Molecular dynamics and Monte Carlo calculations require a description of potential energy surface (PES) of an investigated system at arbitrary configurations. Accurate information at the higher levels of theory is computationally expensive to obtain and limited to a certain number of geometries. An ideal PES should have the accuracy of ab initio calculations and be as fast to evaluate as empirical or semiempirical models. As an alternative to standard interpolation methods for constructing PES from the results of first- principles energy calculations, several efforts to use artificial neural networks to describe PES have been reported. In our study, we used reinforcement learning to train multilayer neural network, implemented in program moonee, for describing PES of methanoic acid spanned by normal coordinates. A particular strength of neural networks is that it can fit any real-valued, continuous function of n-dimensions to arbitrary accuracy using a finite number of parameters and it can effectively model data with noise. For all examined 1D and 2D PESs of methanoic acid, neural networks have shown that they can reproduce accurate description of investigated PESs using only a smaller number of total data points. The computational cost of training the neural network is small and we expect that this method will be useful in modeling a wide variety of PESs at the high level of theory for further molecular dynamics calculations.

Izvorni jezik
Engleski

Znanstvena područja
Kemija

POVEZANOST RADA