Design rules for glass formation from model molecules designed by a neural-network-biased genetic algorithm

• Published in: Soft matter 2019-10, Vol.15 (39), p.7795-788
• Main Authors: Meenakshisundaram, Venkatesh, Hung, Jui-Hsiang, Simmons, David S
• Format: Article
• Language: English
• Subjects: Algorithms; Asphericity; Colloids; Computer simulation; Condensed matter physics; Dependence; Design; Fragility; Genetic algorithms; Glass; Glass formation; Glass transition temperature; High temperature; Liquids; Materials science; Molecular dynamics; Molecular structure; Neural networks; Organic chemistry; Physical properties; Polymers; Solidification; Temperature effects; Transition temperatures
• ISSN: 1744-683X; 1744-6848
• DOI: 10.1039/c9sm01486a

The glass transition - an apparent amorphous solidification process - is a central feature of the physical properties of soft materials such as polymers and colloids. A key element of this phenomenon is the observation of a broad spectrum of deviations from an Arrhenius temperature of dynamics in glass-forming liquids, with the extent of deviation quantified by the "fragility" of glass formation. The underlying origin of "fragile" glass formation and its dependence on molecular structure remain major open questions in condensed matter physics and soft materials science. Here we employ molecular dynamics simulations, together with a neural-network-biased genetic algorithm, to design and study model rigid molecules spanning a broad range of fragilities of glass formation. Results indicate that fragility of glass formation can be controlled by tuning molecular asphericity, with extended molecules tending to exhibit low fragilities and compact molecules tending toward higher fragilities. The glass transition temperature itself, on the other hand, correlates well with high-temperature activation behavior and with density. These results point the way towards rational design of glass-forming liquids spanning a range of dynamical behavior, both via these physical insights and via future extensions of this evolutionary design strategy to real chemistries. Finally, we show that results compare well with predictions of the nonlinear Langevin theory of liquid dynamics, which is a precursor of the more recently developed elastically collective nonlinear Langevin equation theory of Mirigian and Schweizer, identifying this framework as a promising basis for molecular design of the glass transition. A neural-network-biased genetic algorithm is employed to design model glass formers exhibiting extremes of fragility of glass formation, elucidating connections between molecular geometry, thermodynamics, fragility, and glass-transition temperature.