Machine learning meets volcano plots: computational discovery of cross-coupling catalysts

Area of Science:

• Computational Chemistry
• Materials Science
• Catalysis

Background:

• Atomistic simulations are crucial for understanding chemical reactions.
• Machine learning (ML) offers powerful tools for accelerating scientific discovery.
• Predicting catalyst activity is key to developing efficient chemical processes.

Purpose of the Study:

• To develop ML models for predicting oxidative addition energy in C-C cross-coupling reactions.
• To utilize these predictions for identifying active homogeneous catalysts via molecular volcano plots.
• To discover readily available and cost-effective catalyst candidates.

Main Methods:

• Developed novel ML models to predict reaction energies for transition metal complexes.
• Screened large libraries of organometallic catalysts (Pt, Pd, Ni, Cu, Ag, Au) with diverse ligands.
• Applied ML predictions to identify catalysts within an optimal thermodynamic window and below a cost threshold.

Main Results:

• Successfully predicted energies for 18,062 compounds, identifying 557 potential catalyst candidates.
• Refined the candidate list to 37 finalists based on thermodynamic and cost criteria.
• Identified palladium-phosphine complexes and earth-abundant copper catalysts as promising.

Conclusions:

• Modern statistical learning techniques are effective for computational catalyst discovery.
• ML can significantly accelerate the identification of practical and economical catalyst candidates.
• This approach facilitates the discovery of novel homogeneous catalysts for C-C cross-coupling.