Related Experiment Video
Updated: Jun 28, 2026
08:09
A Technique to Functionalize and Self-assemble Macroscopic Nanoparticle-ligand Monolayer Films onto Template-free Substrates
Published on: May 9, 2014
Interpretable Machine Learning of Nanoparticle Stability through Topological Layer Embeddings
Felipe Hawthorne1,2, Leandro Seixas3, James M Almeida4
1Department of Physics, Federal University of Paraná, 81530-015 Curitiba, Paraná, Brazil.
The Journal of Physical Chemistry. A
|June 27, 2026
Summary
We developed a data-efficient machine learning model to predict stable nanoparticle configurations. This approach uses a layer-resolved descriptor and requires minimal reference calculations for materials discovery.
Area of Science:
- Materials Science
- Computational Chemistry
- Machine Learning
Background:
- Nanoparticle stability is complex due to diverse atomic environments.
- Identifying stable configurations is challenging with limited data for first-principles methods.
Purpose of the Study:
- To introduce a data-efficient machine learning framework for predicting nanoparticle stability.
- To enable accurate identification of low-energy configurations using limited reference data.
Main Methods:
- A fragmented, layer-resolved descriptor decomposing nanoparticles into surface, intermediate, and core environments.
- Gradient-boosted decision-tree models with a ranking-based learning strategy.
- Physically motivated weighting schemes and SHAP-based interpretability analyses.
Main Results:
- Accurate identification of stable nanoparticle configurations using a few hundred density functional theory calculations.
- Demonstrated high data efficiency with near-saturation of correlation and high top-k recall.
- Revealed contributions of surface segregation, coordination topology, and local disorder to stability.
Conclusions:
- The framework accurately predicts nanoparticle stability and is highly data-efficient.
- Provides physical insights into factors governing stability across different nanoparticle regions.
- The system and code-agnostic approach is transferable to other nanostructures and methods.
