Molecular Dynamics and Monte Carlo simulations of atomistic systems provide a unique view on the microscopic world. An essential ingredient is the model for the potential energy surface (PES) felt by the atoms in a simulation. Given a PES model, these simulations can be used as a computational microscope to observe the properties and dynamics of complex systems with atomic resolution. Visualizing these simulations is merely a first step. One also derives relevant macroscopic properties by post-processing them with methods founded in statistical physics. The success of this post-processing builds on two main assumptions: (i) the quality of the PES model, i.e. use of suitable approximations, and (ii) the computational efficiency of the PES model, i.e. to collect sufficient data for a statistically relevant post-processing. The tension between those two requirements resulted in a broad spectrum of PES models, ranging from very approximate and fast to very accurate and expensive. Newly developed PES models generally aim for an improved compromise between accuracy and efficiency.
A central theme in Toon Verstraelen's research is the development of PES models, mostly so-called force-field models. Historically, force fields emerged as a purely empirical substitute for more demanding electronic-structure calculations. They enabled simulations on million-atom systems, making our computational microscope much more exciting and widely applicable. The downside of force fields, and actually of all PES models, is that they always make some approximation, i.e. none of them is exact, and that the impact of said approximations on final predictions is hard to estimate a priori. Improving accuracy and controlling the uncertainties of PES models is a major long-term challenge in our field. Several strategies are considered, such as building more physically motivated force fields and exploiting large data sets and machine learning to reduce and quantify remaining uncertainties. TV's research interests also extend to other domains, e.g. where the same strategies are employed to address similar challenges, or where smarter sampling methods are used to explore potential energy surfaces more efficiently.