Abstract
Advances in electron and probe microscopies allow 10 pm or higher precision in measurements
of atomic positions. This level of fidelity is sufficient to correlate the length (and hence
energy) of bonds, as well as bond angles to functional properties of materials. Traditionally,
this relied on mapping locally measured parameters to macroscopic variables, for example,
average unit cell. This description effectively ignores the information contained in the
microscopic degrees of freedom available in a high-resolution image. Here we introduce an
approach for local analysis of material structure based on statistical analysis of individual
atomic neighbourhoods. Clustering and multivariate algorithms such as principal component
analysis explore the connectivity of lattice and bond structure, as well as identify minute
structural distortions, thus allowing for chemical description and identification of phases. This
analysis lays the framework for building image genomes and structure–property libraries,
based on conjoining structural and spectral realms through local atomic behaviour.
