Conspectus
Multimetallic nanoparticles
(NPs) have highly tunable properties
due to the synergy between the different metals and the wide variety
of NP structural parameters such as size, shape, composition, and
chemical ordering. The major problem with studying multimetallic NPs
is that as the number of different metals increases, the number of
possible chemical orderings (placements of different metals) for a
NP of fixed size explodes. Thus, it becomes infeasible to explore
NP energetic differences with highly accurate computational methods,
such as density functional theory (DFT), which has a high computational
cost and is typically applied to up to a couple of hundred metal atoms.
Here, we demonstrate a methodology advancing NP simulations by effectively
exploring the vast materials space of multimetallic NPs and accurately
identifying the ones with the most thermodynamically preferred chemical
orderings. With accuracies reaching that of DFT, our methodology is
applicable to practically any NP size, shape, and metal composition.
We achieve this by significantly advancing the bond-centric (BC) model,
a physics-based model that has been previously shown to rapidly predict
bimetallic NP cohesive energies (CEs). Specifically, the BC model
is trained in a way to understand how the bimetallic bond strength
changes under different coordination environments present on a NP
and how the metal composition of every site affects the detailed coordination
environment using fractional coordination numbers. This newly modified
BC model leads to an improvement from 0.331 (original model) to 0.089
eV/atom in CE predictions when compared to DFT values on a robust
data set of 90 different NPs consisting of PtPd, AuPt, and AuPd NPs
with varying compositions and chemical orderings. By incorporating
the modified BC model into an in-house-developed genetic algorithm
(GA) we can effectively and accurately predict the most stable chemical
orderings of large, realistic bimetallic NPs consisting of thousands
of metal atoms. This is demonstrated on AuPd bimetallic NPs, a challenging
system due to the similarity in the cohesion of the two metals. By
training our BC model using a unique DFT calculation on a bimetallic
NP (one calculation for two metals combining together), we expand
to explore the chemical ordering of multimetallic NPs. We first demonstrate
the application of our methodology on a AuPdPt NP and validate our
stability predictions with literature data. Then, we effectively explore
the vast materials space of multimetallic NPs consisting of combinations
of Au, Pt, and Pd as a function of metal composition. Our thermodynamic
stability trends are presented in a ternary diagram revealing detailed,
and yet, unexpected chemical ordering trends. Our computational framework
can aid both experimental and computational researchers toward effectively
screening multimetallic NP stability. Moreover, we provide an outlook
of how this framework can be applied to catalyst discovery, high-entropy
alloys, and single-atom alloys.