Robust Computational Design of Catalysts for Hydrogen Fuel Cells

Low temperature proton-exchange membrane fuel cells (PEMFCs) have received considerable attention as a sustainable route of energy conversion in the transportation sector. It is widely recognized that a major bottleneck for PEMFCs is related to a high overpotential due to sluggish kinetics of the oxygen reduction reaction (ORR). Currently platinum (Pt) and Pt-containing alloys are among the most active catalysts for this reaction. However, use of Pt-based electrocatalysts creates a cost barrier to long-term operation while still not reaching the highest activity thermodynamically possible from ORR. For example to reach the highest thermodynamically possible activity for ORR, an ideal catalyst must bind oxygen reaction intermediates by ∿ 0.2 eV weaker than Pt(111). Density functional theory (DFT) led catalyst discovery has attempted to address this issue by exploring a large sample space of materials and allowing a detailed fundamental understanding of surface dynamics. However, predictions using DFT suffer from inherent uncertainty imposed by assumptions in the exchange-correlation (XC) functional parameter in each calculation. This affects energy calculations and may yield contradicting results when comparing from among different XC-functionals. In this case higher order computation methods may help, but ultimately become too costly for all purposes. This thesis explores several computational tools to aid in the understanding catalyst behavior. To balance between minimizing computational cost and maximizing prediction robustness, these tools are globally framed around uncertainty in calculations due to the choice of the XC-functional. Next a general error estimation approach is defined in terms of prediction confidence which is propagated through surface Pourbaix diagrams, activity volcano relationships, and thermodynamic property predictions at finite temperature and pressure using the Debye-Grüneisen model for lattice vibrations. Error estimation as represented by prediction confidence is further used to study the alloy catalyst system of PtTi for the ORR. Choice of catalyst material and composition design was done in coordination with experimental collaborators to ensure that catalysts described are experimentally and industrially accessible within practical synthesis capabilities and have a potential for industrial scale-up. In particular the PtTi work focuses on Pt-skin models that were used to guide experiments and design a catalyst with higher activity than pure Pt. We likewise explore the trade-off between strain and ligand effect, as well as the effect of sparse alloy content on catalyst performance. Together the methods in uncertainty quantification and catalyst design provide a framework for accelerating material understanding and design.