Machine learning (ML) has been adopted by researchers to better comprehend the unconventional behavior of supercooled liquids. These are liquids that are cooled below their normal freezing point without undergoing a phase transition into a solid state. By using ML, research has been able to predict the future behavior of supercooled liquids to a certain extent.
The team of researchers adopted a system-level approach instead of focusing on individual particle trajectories. Instead of treating supercooled liquids as a collection of distinct particles, they represented them as a single function that captures the average arrangement of particles. This approach aimed to predict the trajectory of this collective function (a measure of viscosity) and comprehend the changes in its behavior approaching the glass transition. This decision was motivated by both theoretical and experimental considerations.
The team conducted a series of molecular dynamics simulations on both supercooled liquids and glass to gather a comprehensive dataset. These simulations allowed for detailed information about the behavior of individual particles in the system to be gathered. Next, the researchers transformed their particle-resolved description into a collective representation, a well-established framework in condensed matter theory that allowed for the average organization of particles in the supercooled liquid or glass to be captured. To predict the trajectory of the collective function, a neural network was employed. The neural network was trained using the initial conditions of the system as input and the desired collective trajectory was the target output.
The neural network demonstrated excellent performance in predicting the behavior of the collective function for both liquids and glasses. Subsequently, the team formulated a supercooled liquid theory in the form of an exact equation that encapsulated all the approximations and simplifications into a single unknown function (the memory term). This equation allowed for the dynamics of the system to be expressed in a more concise and manageable form. To explore the general features of the approximate memory term in the equation, an evolutionary strategy was employed that involved iteratively refining and optimizing the parameters of the approximate term to identify its key characteristics.
By leveraging this approach, the researchers gained insights into the underlying principles governing the behavior of the collective function and its relationship to the system’s initial conditions. It is hoped that refining the understanding of the memory term will provide deeper insights into the long-term dynamics and memory effects within supercooled liquids and glasses.
The researchers are also incorporating artificial intelligence (AI) techniques, particularly generative AI, to complement molecular dynamics simulations. This integration of generative AI with simulation methods aims to create hybrid approaches that provide more precise, authentic, and broad predictions, which could improve our ability to explore and understand the dynamics of supercooled liquids and glasses.
