(18k) Generating Nanomaterial Design Surfaces with Physics-Informed Machine Learning Models

Polymer-grafted nanoparticles (PGNs) have a manifold of applications in industries spanning photovoltaics, catalysis, drug delivery, and environmental remediation. Their increasing technological prominence in scientific research is a result of the expansive associated design space, which allows for finely tuned interparticle interaction. However, constructing a design space of PGNs for target applications remains a grand challenge due to difficulties in efficiently navigating such design space. Simulations provide a promising path forward, by enabling predictions of properties that emerge due to inter-PGN interactions. However, trial and error are still needed to determine how these interactions should be tuned to meet a given design objective. To aid and expedite these efforts, we have developed ChIMES-Design, a framework for constructing computationally efficient design surfaces describing how inter-PGN interactions are modulated by inter-PGN distance and PGN functionality. This framework can be deployed for forward design to predict emergent properties or inverse design to identify PGN functionality that yields a target emergent behavior.

In this presentation, we describe our initial study on prototypical coarse-grained PGNs. To generate training data for design surface generation, many-body potential of mean force (PMF) surfaces are generated via steered molecular dynamics. Subsequently, we parametrize the design space using ChIMES-Design, a new extension to the ChIMES, a physics-informed many-body machine-learned interatomic model (ML-IAM) framework. Beyond allowing for descriptive and precise back-mapping to the original PGN structures, resulting design surfaces also serve as a testbed for novel nanomaterial discovery techniques such as Bayesian optimization and inverse design. Notably, although ChIMES was initially tailored for modeling condensed phase reacting systems at atomic resolutions, our method has validated its efficacy in coarse-grained nanoparticle systems, paving the way for modeling more complicated nanoparticle structures.