Research Projects

Monolayer-protected nanoclusters (MPCs) are an emerging class of nanomaterials having atomically precise structures – a metal core (Au, Ag, and Cu) surrounded by a monolayer of organic ligands (thiolates, phosphines, and peptides). Their unique size-dependent opto-electronic properties make them exceptional candidates for potential applications in catalysis, bio-imaging, sensing, and energy storage. However, their full exploitation requires a comprehensive understanding of MPCs' structural and dynamic evolution and reactivity across various scales, from molecular-level to macroscopic structures. This proposal aims to develop an integrated machine learning (ML) enhanced sampling (ES) simulation and a multi-scale computational framework to model MPCs from automatic to coarse-grained models. Understanding the structure-property relationship is of paramount interest in the technology-specific use of a material. There exists a large number of experimentally synthesized and characterized MPCs with various combinations of metal cores and ligands. Our first part of the project will be devoted to unveiling the relationship between MPCs structures and their stability and opto-electronic properties using graph neural network (GNN)-based approaches. GNNs are well-suited for this task, as they can directly represent the atomic connectivity and local environments within MPCs as graphs with the heavy atoms as nodes and chemical bonds as edges. We will use features like atom types, atomic numbers, and charges for the nodes, and bond types and bond lengths for the edges. Furthermore, we will integrate generative AI models that will be trained on existing MPC data and will be capable of designing new MPC structures and predicting their properties. This will guide experimental synthesis of novel MPCs for specific applications. The second part of the proposal will be devoted to understanding the structural and dynamic evolution of various small MPCs in the gas phase, in solution, and on solid supports. Simulating such systems poses significant challenges since traditional force fields are often unavailable or insufficiently accurate for MPCs due to their complex structure and charge transfer characteristics. To address this, we will develop machine learning interatomic potentials (MLIP) for a library of MPCs with Au, Ag, and Cu cores and various ligands. MLIPs are data-driven potentials trained on high-quality quantum chemical datasets such as those obtained from density functional theory (DFT). We will utilize the DeepMD code, which is specifically designed for training neural network potentials for systems ranging from simple fluids to complex solids. The training dataset will incorporate a diverse set of configurations from the phase space, including reactive configurations. The obtained MLIPs will then be utilized to run MD and ES simulations for nanoseconds and beyond timescales, required to study ligand exchange between MPCs, core reconstruction, and MPC-mediated catalysis. The third part will focus on investigating the self-assembly of MPCs in the solution phase. While atomistic MD simulations are useful for studying individual MPCs or a few MPCs in solution, they are computationally prohibitive in simulating systems containing hundreds to thousands of MPCs in explicit solvents. To circumvent the time- and length-scale limitations, we will develop coarse-grained (CG) models of various MPCs and study their self-assembly forming superstructures. At first, the MARTINI models will be used for rapid screening of assembly thermodynamics. These models will provide preliminary insights into the thermodynamic driving forces for the self-assembly processes. In the subsequent phase, we will leverage Graph Neural Network (GNN)-based to develop more accurate CG potentials. The CG models will be obtained from training a GNN with atomistic simulation data. The CG models thus obtained will enable the simulation of self-assembly with greater fidelity.