Towards accurate computational experimentation: machine-learning-driven simulation of nanocarbon synthesis

Project Details


In this project, we will extend the accuracy and range of applicability of the Gaussian approximation potential framework for multispecies simulation. We will develop algorithmic improvements towards more efficient computational evaluation of many-body atomic descriptors which will allow us to speed up the simulations. These improvements will be used to develop a new machine-learning-based carbon-metal interaction potential with accuracy close to quantum chemistry methods (such as density functional theory). The new potential will allow us to reconstruct and fully understand (down to the atomic resolution) the role of metal catalysts on carbon nanostructure formation, such as carbon nanotubes and nanofibers, as well as metal-functionalized versions of them. The new inexpensive simulation framework will enable in silico testing and design of new catalysts for cost-effective production of carbon nanostructures and will be used to explore and propose new nanocarbon forms.
Short titleCOMPEX
Effective start/end date01/09/201931/08/2023