Hybrid perovskites are a promising material for solar cells, but they easily break down from moisture. To protect their surface, scientists apply special molecules, much like patches on a frayed cloth. A single patch lies flat and keeps water out. But sew on too many patches overlapping, and they bulge, leaving gaps, and the protection fails. Scientists found a similar effect in solar cells: when there are too many protective molecules, they start getting in each other's way, disrupting the crystal structure and reducing stability.

Understanding the details required machine learning. Researchers at Oxford created a program that predicts how atoms behave. They trained it on a host of examples, then fine-tuned it for the specific protective molecules containing silicon, carbon, oxygen, and hydrogen. Numerical simulation showed that when too many molecules crowd in, they shift and deform neighboring atoms. Thus the paradox: extra protection turns into harm.

To model the behavior of just one molecule, they had to simulate over three thousand atoms — an entire nanocrystal, visible only under an electron microscope. This knowledge will help precisely tailor the composition of protective coatings without lengthy trial and error.