Last week, I visited Saint‑Malo, France, and presented a poster at the 7th International Conference on Geometric Science of Information, outlining our work on the application of the close‑packing principle to molecular crystal structure prediction. The theme of the 2025 edition of GSI was
From Classical to Quantum Information Geometry: Geometric Structures of Statistical & Quantum Physics, Information Geometry, and Machine Learning.
See the poster title and abstract below.
Title: Geometric Modelling in the Prediction of Molecular Crystal Structures: The Close‑Packing Principle Revisited
Abstract: A. I. Kitaygorodsky’s Close‑Packing Principle states that “the mutual arrangement of the molecules in a crystal is always such that the ‘projections’ of one molecule fit into the ‘hollows’ of adjacent molecules,” highlighting the underlying geometries of molecular interactions in crystalline ground states. We revisit this principle to demonstrate that a geometric approach can substantially reduce the computational effort required to predict stable phases of molecular crystals. By leveraging the duality between lattice energy and packing density via a crystallographic space‑group‑induced, fibre‑orbifold‑constrained programming approach, we generate first‑approximation structures during the initial stages of crystal structure prediction (CSP) within minutes on a standard laptop. Focusing on close packings of volumetric molecular models—where interactions are predominantly van der Waals forces, with minimal charge or electron exchange—we apply this purely geometric method to various rigid organic molecules. Our results show that integrating the Close‑Packing Principle as a geometric prior reduces computational costs in CSP workflows, enabling more efficient and accurate predictions of molecular crystal structures and promoting ethical and sustainable materials discovery.
Supplementary materials, including the poster handout, are available here: