Current Research :
Model Vitrimers: First I give overview of the work related to polymer system. My research focuses modelling reversible crosslink polymers which show many potential applications.
Permanently crosslinked polymers are extensively used in consumer products due to their very high mechanical properties. However, they cannot be reprocessed and reshaped once synthesized; therefore, they are unsustainable. Reversibly crosslinked polymer networks – polymer networks that can undergo bond association and dissociation reactions – rearrange their structures while maintaining their overall integrity, thus resulting in unique properties. These systems are called ‘Vitrimers’. It show potential application in preparing self-healing, reprocessable, shape memory polymers. So reversible crosslink polymer may be next step for sustainable material for future.
Another major problem of this century is management of plastics. The global plastics problem is a trifecta, greatly affecting environment, energy and climate. To achieve a circular plastics economy, one must overcome a central barrier of reusing mixed-plastics waste because a major fraction of plastics waste is in the form of mixed plastics. However, most polymer mixtures are immiscible, primarily because the entropy of mixing is dramatically reduced. From a recycling view point there is thus a pressing need to compatibilize generally immiscible mixed-plastic chains. Recent studies also show reversible crosslinker can help in mixed plastic mixing and recycling by keeping intact of its mechanical properties.
Here, we show from fundamental statistical mechanics that the introduction of crosslinks, whether reversible or permanent, directly impacts the equilibrium polymer density and hence the material’s surface tension using molecular simulation. For a limiting case where the bonds are the same size as the polymer chain bonds, simulations, Flory hypotheses and thermodynamic calculations show that the crosslinks induce an increased entropic cohesion in the liquid. These findings implicate density as a key variable in polymers with (dynamic) crosslinkers, one that can be used to facilely tune their properties.
Next we focus on dynamical aspect of reversible crosslink system. We observe rich chain dynamics for these systems depending on the magnitude of bond lifetimes. We also show there is a rich difference in chain dynamics for long lived bonds compared to short lived bonds. Chain dynamics is important for material design. Simultaneously, we are also examining role of energetics of the crosslinkers which show aggregation in vitrimer systems and glass transition of these systems. We are also studying mechanical properties through stress strain curve for reversible crosslink system.
Along with this in our group we are trying to study these systems from 1st principle quantum mechanics and reactive force filed method to capture better chemistry which will help to design dynamic crosslinker.
Machine Learning In polymer system: Machine learning models have been progressively used for predicting materials’ properties. These models can be built using pre-existing data and are useful for rapidly screening the physicochemical space of a material. However, ML models are inherently interpolative, and their efficacy for searching candidates outside the known range of property of a material is unresolved. Moreover, the performance of an ML model is intricately connected to its learning strategy and the volume of training data. Here, we determine the relationship between the extrapolation ability of an ML model, the size and range of its training dataset, and its learning approach. We focus on a canonical problem of predicting the properties of a copolymer as a function of the sequence of its monomers. Tree search algorithms, which learn the similarity between polymer structures, are found to be inefficient for extrapolation. Conversely, the extrapolation capability of neural networks and XGBoost models, which attempt to learn the underlying functional correlation between the structure and property of polymers, show strong correlations with the volume and range of training data. These findings have important implications for efficient ML models development.
Machine Learning to predict RDF of binary Mixture:
There are potential application for binary mixture from glass to high entropy alloys. Structural rearrangment of mixture can be captured through the Radial distribution function, which is directly coupled with thermodynamic parameters like temperature, pressure, composition, interaction among particles. There are huge thermodynamic parameter space which shows wide range of phases in mixture. It is a challenging to experiment or do molecular simulation to explore all types of phases. Machine Learning are capable to predict complex relation depending on data. In our current study we explore prediciton of RDF for different thermodynamic parameters. We generate dataset of RDF for different temperature, composition, size ratio of mixture and interaction. From the available data we build a Model to predict RDF and we also predict RDF for extrapolating region that is we train the Model in a limited range and our Model can predict for out of range with significant high accuracy. We do some feature engineering to build a better model for RDF prediction and compare with other features.
Generating new molecule using Language Model: LLM(Large Language model) is a very hot topic in recent times, which is a promising approach for generating text. We are using LMs in chemistry to create new molecule from a trained dataset. We are using modified k-gram LM to sample new molecule. We observe importance of byte-pair encoding to sample variety of molecules. We plan to screen new molecules and use DFT for geometry optimization.