The interfacial ionic space charge effect has been used to explain various phenomena including interfacial resistance in solid-state batteries [1] and conductivity enhancement in nano-spaced superlattices [2]. Many such phenomena have later been better explained by, e.g., secondary phases [1], but in principle, there must be a space charge region to align the electrochemical potential across any interface. However, direct experimental observation of the space charge has turned out to be extremely challenging [3], and theoretical modeling has either relied on phenomenological Poisson-Boltzmann models [4,5] or first-principles calculations of interface slab models considering a limited number of defect configurations [6]. For these reasons, the atomistic picture of the interfacial space charge and its role in determining interfacial properties remain, to a large extent, unclear.

Direct first-principles modeling of the space charge is hindered by the combinatorial explosion in the number of possible defect configurations and slow relaxation times in molecular dynamics simulations. To tackle this issue using modern-day supercomputing resources, we have been developing an open-source framework abICS (for ab Initio Configuration Sampling), which combines high-throughput ab initio calculations, machine learning, and parallel statistical physics methods to enable thermodynamic sampling of millions of configurations on a lattice [7,8].

In this work, we applied abICS to the Pt/YSZ cermet interface (a prototypical high-temperature fuel cell system) to examine the segregation of dopants and oxygen vacancies at surfaces and interfaces. Comparison with calculations at varying p_O2 showed that the thermodynamic conditions drastically modify the segregation behavior, with accumulation of negatively charged defects in oxidizing conditions and positive ones in reducing conditions. The space charge region was limited up to only a few atomic layers. These observations are generally in good agreement with those from Poisson-Boltzmann-type modeling [4]. From this, we may conclude that the impact of the space charge on the macroscopic properties such as ionic conductivity may be limited. On the other hand, the local structure at the surface/interface is drastically modified, implying a huge impact on the catalytic activity of oxide surfaces.