Extended Data Fig. 10: Linear propensity prediction.
From: Unveiling the predictive power of static structure in glassy systems
Linear regression on the number of neighbouring particles within a sphere of a given size (x-axis). Results are shown for different S(q,t) values, and different temperatures (colour). Linear prediction gets easier with time and lower temperature. Nearest neighbours matter most for the ballistic regime, whereas for glassy and diffusive regime, best discrimination is achieved for larger spheres. There, wavy pattern shifts with temperature. The shift gets removed by re-scaling the x-axis to match the box size of all four conditions.
