Topological Machine Learning Unveils Hidden Reaction Pathways in Nanocrystal Synthesis

Area of Science:

• Materials Science
• Chemical Engineering
• Computational Chemistry

Background:

• Analyzing reaction pathways in nanocrystal synthesis is challenging due to transient intermediates and limitations of conventional methods.
• Manual spectral analysis is subjective and may miss crucial reaction events.
• Developing objective, data-driven approaches is essential for mechanistic understanding.

Purpose of the Study:

• To present a novel machine learning framework for elucidating reaction pathways directly from raw spectroscopic data.
• To overcome the limitations of manual interpretation in mechanistic analysis.
• To enable objective discovery of reaction intermediates and pathway selection.

Main Methods:

• Integration of transformer-based data augmentation with topological manifold learning.
• Direct analysis of high-dimensional, raw spectroscopic data (UV-vis).
• Application to ex-situ datasets from indium arsenide nanocrystal synthesis.

Main Results:

• Reconstruction of the complete reaction landscape for nanocrystal synthesis.
• Identification of previously unreported metastable intermediates.
• Demonstration of how chemical additives influence intermediate formation and pathway selection.

Conclusions:

• The topological learning framework objectively elucidates reaction pathways from spectroscopic data.
• This approach provides a generalizable strategy for mechanistic discovery and predictive control in complex chemical systems.
• The method successfully identified new intermediates and pathway modulation in indium arsenide nanocrystal synthesis.