Public defence in Engineering Physics, M.Sc. Jari Järvi

Title of the doctoral thesis: Structure search of molecular adsorbates with Bayesian inference and density-functional theory

Opponent: Professor Ruben Perez, Universidad Autónoma de Madrid / Condensed Matter Physics Center (IFIMAC), Spain
Custos: Professor Patrick Rinke, Aalto University School of Science, Department of Applied Physics

The doctoral thesis is publicly displayed 10 days before the defence in the publication archive Aaltodoc of Aalto University

Electronic thesis

Public defence announcement:

Modern electronic devices rely on materials that have precision-engineered functional properties. These properties depend sensitively on the atomic structure, which must be determined accurately to optimize device performance. The devices commonly employ hybrid organic-inorganic materials that contain hybrid interfaces. The interface structures can be studied via molecular adsorbates and organic thin films, but this remains challenging with current experimental and computational tools. State-of-the-art microscopy methods, such as atomic force microscopy (AFM), struggle in imaging non-planar structures conclusively. Global structure search requires accurate energy sampling with ab-initio atomistic simulations, which is prohibitively expensive for molecular adsorbates due to the large configurational space. 

Novel artificial intelligence tools can help to reduce the amount of costly calculations in structure search through strategic sampling of adsorbate configurations. In this thesis, I utilize a recently developed Bayesian Optimization Structure Search (BOSS) method that uses active learning to detect all stable structures accurately and efficiently. Moreover, the computational cost of configurational sampling is reduced on complex substrates with an approximate substrate model. I also investigate a human-in-the-loop approach, which performs adsorbate search with Bayesian inference using only chemical intuition of materials scientists. The results indicate that this approach is viable and could be later integrated with energy sampling for increasingly powerful structure search. 

The BOSS workflow is demonstrated with a camphor molecule on Cu(111) and F4TCNQ and TTF molecules on an approximated electronically decoupled graphene, Gr/O/Ir(111). The stable adsorbates of these materials were reliably identified at half or less of the computational cost compared to traditional methods. The discovered structures correspond well to AFM experiments and prior literature. The approximated graphene substrate reduced energy calculation time to only 1% while preserving adsorption properties. The BOSS workflow aids in interpreting adsorbate structures and thin film morphologies in microscopy images. The method is broadly applicable to other materials and disciplines where organic-inorganic interfaces are studied for novel applications, for example in electronics and green energy production.

Contact details of the doctoral student: [email protected], +358 40 560 9059