Atomistic simulations offer an unparalleled insight into fundamental mechanistic processes such as plastic deformation and fracture - like a super-microscope with super slow motion. While the quantum mechanics of atoms and particularly their electrons tells us how the atoms move on this scale, the computational complexity of this method restricts its application to the tiniest systems (a few thousand atoms). For larger problems, classical effective potentials or force fields are the workhorses of atomistic simulations. By eliminating electronic degrees of freedom, they describe the energy of a system as a function of the atomic positions only, thus allowing fast and highly parallelisable simulations. Unfortunately, the bias, approximations and uncertainties incurred in representing the energy landscape created by the electronic interactions between atoms by a potential with a limited number of parameters are badly controlled, with an a priori unknown impact on quantities of interest of a simulation.
In this talk, I present a strategy to encapsulate the uncertainty in the potential model in a way that allows us to propagate it to the simulations outcomes. In this way, we offer a tool to judge the the validity and reliability of simulation results with potentials that we know are deficient, but do not know to what degree. This is an important step towards predictive simulations, and help answer the question whether our computational super-microscope shows us the reality or just an artefact of the model we are using.