Identifying Useful Nanocrystal Morphologies Using Advanced Sampling Techniques and Machine Learning

We applied dimensionality reduction,
together with both supervised
and unsupervised machine learning (ML), to classify the temperature-dependent
equilibrium shapes of Ag nanoparticles containing 100–200 atoms
in 1–2 nm size range. The Ag nanocrystal shapes were generated
using parallel tempering molecular dynamics. Using ML techniques,
we identified five unique particle shape classes with 14 underlying
subclasses. We considered the ramifications of our results for catalysis
by characterizing the strain and coordination of the surface atoms
for different particle subclasses. These studies revealed that icosahedra
and hybrid decahedra-icosahedra possess the widest variety of under-coordinated
surface atoms and the highest strain. This work helps to forge the
link between structure and function and identify processing strategies
for achieving beneficial nanocrystal morphologies.