Heterogeneous catalyst design by generative adversarial network and first-principles based microkinetics.

Area of Science:

• Materials Science
• Chemical Engineering
• Computational Chemistry

Background:

• Developing efficient heterogeneous catalysts is crucial for industrial chemical processes.
• Discovering new catalyst materials often relies on extensive experimentation and computational screening.
• Ammonia (NH3) synthesis is a vital industrial process with ongoing efforts to improve catalyst efficiency.

Purpose of the Study:

• To develop an automated method for proposing novel heterogeneous catalysts.
• To leverage machine learning, specifically generative adversarial networks (GANs), in catalyst design.
• To enhance the performance of catalysts for ammonia (NH3) formation reactions.

Main Methods:

• Combined microkinetic analysis based on density functional theory (DFT) with generative adversarial network (GAN) for catalyst proposal.
• Calculated turnover frequencies (TOF) for NH3 formation on Rh-Ru alloy surfaces using DFT-based microkinetics.
• Explicitly considered six elementary reactions including N2 dissociation, H2 dissociation, NHx formation, and NH3 desorption.

Main Results:

• Successfully generated new alloy surfaces not present in the initial DFT dataset.
• The generated surfaces exhibited significantly higher NH3 formation turnover frequencies (TOF).
• Enhanced N2 dissociation exothermicity on the generated surfaces correlated with higher TOF.

Conclusions:

• Demonstrates the feasibility of automated catalyst material improvement using DFT and GANs.
• Highlights the potential of integrating computational chemistry with machine learning for accelerated materials discovery.
• Suggests a pathway for designing superior heterogeneous catalysts for reactions like ammonia synthesis.