Organic thin films are characterized by a rich and intricate polymorphism, often forming structures with many molecules per unit cell. The large configurational complexity that arises from the combination of these many, partly independent degrees of freedom defies most established structure search methods. In the present contribution, we show that when thin films form commensurate interfaces with their support, the search space can be significantly reduced. Because the substrate acts as a registry to the adsorbate, it is possible to discretize the potential energy surface such a way that well-justified educated guesses for the minima are obtained. Local geometry optimization from these guesses towards the actual minimum structures are highly efficient. Moreover, because it is possible to label and enumerate the possible minima, we can efficiently traverse and scan the configurational space using Monte-Carlo sampling techniques.
A particularly intriguing option arises when combining the discretization with machine-learning techniques. Training only on a few, inexpensive, highly-periodic calculations (with few molecules per unit cell), it is possible to predict the energies of larger, more complicated unit cells extremely accurate. The prediction itself does not require significantly computational effort, allowing to predict, in principle, the energy of all minima at the same time. This allows us to study not only polymorphism, but also to investigate the propensity for and impact of defects in organic thin films.
The advantages, but also the drawbacks of our method will be demonstrated for the cases of tetracyanoethylene adsorbed on Ag(100) and Au(111) .
 V. Obersteiner, M. Scherbela, L. hörmann, D. Wegner, O.T. Hofmann, Nano Lett., 2017, 17 (7), pp 4453–4460
Back to Workshop I: Optimization and Optimal Control for Complex Energy and Property Landscapes