Note that this event will be held online.

Many soft-matter systems, such as fluids and glasses, are clearly disordered materials, lacking the long-ranged translational order of crystals. However, hiding in the apparent structural disorder of fluids and glasses can be an impressive degree of local order, where similar structural features are repeated again and again. Research over the last decades has revealed that (at least some of) these repeating structural motifs are intimately connected to the dynamic evolution of the system. While this holds for atomic and molecular systems as well, colloidal soft matter, where particles can be observed in real-space and real-time, forms the ideal playground for exploring this connection. Arguably the two main areas of soft matter science where such local structure is likely to play a role are the fields of crystal nucleation and glassy dynamics. However, while there is at least a partial consensus on the importance of such motifs, characterizing local structure remains a complex problem.

In crystal nucleation research, system- and phase- specific algorithms that classify local structure (e.g. based on local bond order [1, 2]) have traditionally been used to separate particles that have crystallized from those that remain in the fluid phase. While actively used to probe the structure of the nucleating crystal (including changes in time linked to e.g. two step nucleation, and interfacial structures) and its propensity to form [3, 4, 5], these algorithms were typically designed by trial and error, and are system and structure specific. Machine learning techniques, which excel at finding order hidden in complex datasets, have recently started to be employed to identify local structures associated with crystal nucleation. Inspired, at least in part, by the successful application of machine learning to efficiently approximate molecular interaction potentials [6, 7, 8, 9, 10], ML has just recently proven highly successful in the development of new order parameters, allowing one to train algorithms to recognize specific crystalline environments [11, 12], and even to autonomously pick out groups of particles in different phases [13, 14].

In the study of the glass transition, the link between local structure and dynamics is a long-standing question [15, 16, 17]. While structural features clearly play a key role in the relaxation of glassy fluids [18, 19, 20, 21, 22,23], pinpointing a structural explanation for e.g. the length scale of dynamical heterogeneities remains a challenge [24]. To help address this, a multitude of methods have been developed to detect different local structural features in glassy fluids, in order to correlate these with dynamics [17, 25, 26]. Here, ML has also made a significant impact, leading to new algorithms that can be trained to identify soft spots in glasses [27, 28, 29, 30], or even spot structural variations in glasses without relying on dynamical information [31].

This workshop will act as a platform to bring together scientists from different communities interested in quantifying or understanding local structure, and will give them the opportunity to exchange ideas, knowledge, and methodologies. The main aims of this workshop include:

1. To compare experimental, theoretical and computational perspectives on the role of local structure in crystal nucleation and glassy dynamics.
2. To discuss the most recent advances in local structure detection and characterization.
3. To identify new collaborative pathways for using these methods to unravel the correlation between local structure and dynamics.