A Generalized Machine-Learning Framework for Developing Alchemical Many-Body Interaction Models for Polymer-Grafted...

Area of Science:

• Nanomaterials Science
• Computational Chemistry
• Machine Learning

Background:

• Polymer-grafted nanoparticles (PGNs) are key building blocks for advanced nanomaterials.
• Designing self-assembled nanomaterials requires optimizing PGN attributes like polymer length and grafting density.
• Accurate modeling of PGN interactions is essential for physics-informed inverse design.

Purpose of the Study:

• To develop a novel machine-learned interatomic model (ML-IAM) for predicting PGN interactions.
• To create an "alchemical" ML-IAM (X-ChIMES) that captures interactions based on inter-particle distance and PGN attributes.
• To enhance inverse design strategies for self-assembled nanomaterials.

Main Methods:

• Developed an extended ChIMES (X-CHIMES) ML-IAM, building upon the physics-informed ChIMES model.
• Trained X-CHIMES using potential of mean force (PMF) data for PGNs with varying polymer lengths.
• Utilized steered molecular dynamics and grid sampling within HOOMD-blue for efficient data generation.

Main Results:

• Demonstrated the efficacy of ChIMES for coarse-grained (CG) modeling of PGNs.
• Successfully developed and applied the extended ChIMES (X-CHIMES) model.
• Integrated X-CHIMES with digital alchemy inverse design simulations.

Conclusions:

• X-CHIMES accurately models PGN interactions across varying distances and attributes.
• This work enables efficient inverse design for targeted self-assembled nanomaterials.
• This is the first application of ChIMES for CG systems coupled with inverse design.