Title of the talk: First-principles simulations of geological systems using machine learned interatomic potentials
Machine learned interatomic potentials have recently reached an accuracy level where they are being used as surrogates for first principles methods like DFT. This allows for large (millions of atoms) and long (nanoseconds to microseconds) molecular dynamics simulations of realistic systems with first principles accuracy, for example to compute thermodynamic properties like the phase diagram of water. In this talk, I will present how we train such interatomic potentials with an active learning approach, and some ways we apply them to dry and wet mineral systems undergoing mechanical failure.
Short bio: Henrik Andersen Sveinsson is s Postdoc at the Njord Centre. His main research topic is the molecular scale mechanisms involved in mechanical failure processes such as fracture and friction. For his PhD, he modeled mechanical failure processes in methane hydrates. Sveinsson is currently working on two EuroHPC projects focused on the fracture process of silicates, the CO2 storage potential of glassy basalts and the slipperiness of ice. He is the PI of an upcoming RCN funded project called nanoscale controls of reaction-induced fracturing from first principles.
To get news, invitations to seminars and more from Njord, please go here to subscribe to our newsletter.