Insight into Glass Problems from Machine-Learned Softness
We are taught that liquids flow while solids don’t. But as liquids are supercooled below their freezing temperatures, their dynamics become increasingly slow until their flow becomes undetectable on reasonable timescales—this is the glass transition. Conversely, when glasses are put under mechanical load, they can flow very sluggishly—this is plasticity. In both of these glass problems, flow involves microscopic hops, in which constituent particles change their neighbors. For decades, no link between local structure and hops was found for either of these two glass problems, but it is easily detected using machine learning. We review a machine-learned local structural quantity, softness, which is predictive of hops, describing what it teaches us about both glass problems.
Authors
- Daniel M. Sussman (ORCID: https://orcid.org/0000-0002-6164-9124)
- Sean A. Ridout (ORCID: https://orcid.org/0000-0003-2387-8361)
- Douglas J. Durian (ORCID: https://orcid.org/0000-0003-3240-2381)
- Andrea J. Liu (ORCID: https://orcid.org/0000-0002-2295-2729)
- Ge Zhang
- Hongyi Xiao
Institutions
- Emory University (US)
- City University of Hong Kong (HK)
- University of Michigan (US)
- University of Pennsylvania (US)
Publication Details
- Journal
- Annual Review of Condensed Matter Physics
- Published
- 2026-11-12
- DOI
- https://doi.org/10.1146/annurev-conmatphys-062226-113838
- Primary Topic
- Material Dynamics and Properties
- Type
- article
- Field-Weighted Citation Impact
- 0.00