Executive Summary : | The atomistic simulations are potential theoretical tool for investigating the mechanical and physical behaviour of the materials. The quantum mechanical simulations are found to most accurate methods, but computationally more involved as complete atomic structure of atoms need to model for simulations and thus expensive for large scale atomistic simulation. On the other hand, molecular dynamics (MD) simulations through classical interatomic potential are relatively faster as compared to quantum mechanical calculations, but these usually fail to map the complete potential energy surface (PES) of quantum mechanical calculation as these have fewer tailoring empirical parameters to accommodate the high-dimensional PES along with interatomic forces and stresses. The proposed research deals with developing machine learning interatomic potentials (MLIPs) for atomistic simulation of two-dimensional nanomaterials. The MLIPs are found to be more accurate when compared to that for the classical interatomic potentials. In MLIPs the functional form of interatomic potential is trained to represent the potential energy surface (PES) obtained through quantum mechanical simulations along with data for forces and stresses in the molecular system. In developing MLIPs, the data corresponding to PES, lattice parameters, interatomic forces and stresses is first obtained through DFT calculations and then this data will be used to train the machine learning (ML) models to map the same PES through different polynomials, for which weights will be obtained through supervised nonlinear regression algorithm. In training the ML models, 60% of the total data for PES will be used to train ML models and remaining 40% will used to examined accuracy of trained model to access any oscillations in the fitted nonlinear regression models. The MLIPs will further be used to investigate nonlinear constitutive and vibration behaviour of the nanosheets, and nanotubes access their accuracy for large scale atomistic simulations. The obtained MLIPs will also be integrated with finite element method to propose MLIPs-based multiscale modelling for large deformation mechanics of nanotubes and nanosheets under different kind of loadings. |