Glassy materials are ubiquitous in human life, with applications ranging from composites in computing chips to dense emulsions in the everyday handsoap. The complex physics and dynamics of glassy materials has puzzled scientists for long. The inability of glasses to equilibrate and find their lowest energy configurations is often attributed to their potential energy landscapes: rugged, barrier-filled surfaces in high-dimensional space that seem incredibly difficult to navigate. Further, the characteristic properties of system configurations deep in the energy landscape are fundamental to understanding questions surrounding the glass transition. In this study, we employ a modified metadynamics based algorithm (MIMSE), and report the surprising observation of canyon-like structures in high-dimensional glassy energy landscapes. Our algorithm successfully leverages these canyons as navigational aids to access low energy states deep within the potential energy landscape. The earlier-noted rugged surfaces line the canyon floors with smooth canyon walls. The algorithm successfully overcomes local ruggedness and explores low-energy regions of the landscape. Further, we observe such canyons in three vastly different glassy systems, successfully sampling low energy configurations in each - while revealing strikingly similar fractal signatures in their landscape structures. Moreover, analyzing local minima via high-dimensional angular clustering shows that these canyons taper in hypervolume as we descend to lower energies. Finally, we contrast our results with ones obtained using conventionally studied techniques like molecular dynamics and Monte Carlo based schemes, discussing the pros and cons of each.
Funding Acknowledgement: NSF DMR-1609525
My research interests span the fundamental study of physical systems using computational, physics-based and data driven tools. I have experience developing my own simulation packages, algorithms and programs, and using different standard packages like LAMMPS, hoomd-blue, MoSDeF, Quantum ESPRESSO and ML-based data driven tools to perform multi-scale analysis of various systems. As apart of my PhD, I initially worked on simulating 3-D ripening foams and their micro-structural and rheological properties, before switching to studying energy landscapes and dynamics of glassy systems. As a part of this work, I have analyzed different glassy, organic and polymeric model systems by developing a
modified algorithm (MIMSE) that finds descending canyons in the landscapes of such systems. Iâm interested in studying the contribution of the energy landscapes towards properties of systems and deciphering the different fundamental scientific principles underlying such physical and biological systems.