Representation of Potential Energy Surfaces using Neural Networks

  Umme Kulsum,  Raza Imam,  Mohd Abdullah Khan,  Asra Ansari

Deep learning is ideally suited for modelling nonlinear potential-energy surfaces, expressing quantum-mechanical interactions, and expanding chemical compound space research. Given the presence of hidden layers, neural networks do more effective predictive analyses as the neural network employs the multiple hidden layers to improve prediction accuracy. There is a requirement for precise potentials that can swiftly repeat high-quality results since the interactions in force fields are represented by a variety of different functions. In this work, we strive to investigate the representation of Potential Energy Surfaces, a crucial component of chemical dynamics, using neural networks. We developed neural network models that can be applied widely to fit one-dimensional data and two-dimensional potential energy surfaces separately. Our methodology concludes different key analytical outcomes as well as crucial future directions that aim to strengthen the potential of chemical dynamics and machine learning.

Keywords- Potential Energy Surfaces, Neural Networks, Morse Potential, Activation Function, Dimensional Curves

Unique Identification Number - IJEDR2204004
Page Number(s) - 32-42
Pubished in - Volume 10 | Issue 4 | November 2022
DOI (Digital Object Identifier) -    http://doi.one/10.1729/Journal.31971
Publisher - IJEDR (ISSN - 2321-9939)

  Umme Kulsum,  Raza Imam,  Mohd Abdullah Khan,  Asra Ansari,   "Representation of Potential Energy Surfaces using Neural Networks", International Journal of Engineering Development and Research (IJEDR), ISSN:2321-9939, Volume.10, Issue 4, pp.32-42, November 2022, Available at :http://www.ijedr.org/papers/IJEDR2204004.pdf