[Paper Review] Machine Learning Assisted Design and Optimization of Transition Metal-Incorporated Carbon Quantum Dot Catalysts for Hydrogen Evolution Reaction

Duong Nguyen Nguyen, Min‐Cheol Kim+8 more|arXiv (Cornell University)|Oct 26, 2022

Electrocatalysts for Energy Conversion4 citations

TL;DR

This study proposes a machine learning (ML)-assisted design strategy for transition metal-doped carbon quantum dots (M@CQD) to optimize hydrogen evolution reaction (HER) catalysts. Using a Bayesian genetic algorithm (BGA) ML model, Ni@CQD on 3D reduced graphene oxide was identified as the optimal catalyst, achieving 189 mV overpotential at 10 mA cm⁻², a Tafel slope of 52 mV dec⁻¹, and high durability in acidic media, validated by electrochemical testing and DFT simulations.

ABSTRACT

Development of cost-effective hydrogen evolution reaction (HER) catalysts with outstanding catalytic activity, replacing cost-prohibitive noble metal-based catalysts, is critical for practical green hydrogen production. A popular strategy for promoting the catalytic performance of noble metal-free catalysts is to incorporate earth-abundant transition metal (TM) atoms into nanocarbon platforms such as carbon quantum dots (CQDs). Although data-driven catalyst design methods can significantly accelerate the rational design of TM element-doped CQD (M@CQD) catalysts, they suffer from either a simplified theoretical model or the prohibitive cost and complexity of experimental data generation. In this study, we propose an effective and facile HER catalyst design strategy based on machine learning (ML) and ML model verification using electrochemical methods accompanied with density functional theory (DFT) simulations. Based on a Bayesian genetic algorithm (BGA) ML model, the Ni@CQD catalyst on a three-dimensional reduced graphene oxide (3D rGO) conductor is proposed as the best HER catalyst under the optimal conditions of catalyst loading, electrode type, and temperature and pH of electrolyte. We validate the ML results with electrochemical experiments, where the Ni@CQD catalyst exhibited superior HER activity, requiring an overpotential of 189 mV to achieve 10 mA cm-2 with a Tafel slope of 52 mV dec-1 and impressive durability in acidic media. We expect that this methodology and the excellent performance of the Ni@CQD catalyst provide an effective route for the rational design of highly active electrocatalysts for commercial applications.

Motivation & Objective

• To develop a cost-effective, noble-metal-free electrocatalyst for green hydrogen production via the hydrogen evolution reaction (HER).
• To overcome limitations in data-driven catalyst design by integrating machine learning with experimental validation and DFT simulations.
• To identify optimal transition metal-doped carbon quantum dot (M@CQD) configurations for enhanced HER activity.
• To establish a predictive and scalable design framework for high-performance electrocatalysts using Bayesian genetic algorithm (BGA) modeling.

Proposed method

• A Bayesian genetic algorithm (BGA) was employed to predict the optimal M@CQD configuration for HER based on theoretical descriptors.
• The BGA model evaluated catalyst loading, electrode type, temperature, and pH of the electrolyte to identify optimal operating conditions.
• Electrochemical experiments were conducted to validate the ML-predicted performance of the Ni@CQD catalyst on 3D reduced graphene oxide (3D rGO).
• Density functional theory (DFT) simulations were used to support the mechanistic understanding of the catalytic activity and electronic structure changes in Ni@CQD.
• The model was iteratively refined using experimental data to improve prediction accuracy and reliability.
• The integration of ML predictions with experimental and theoretical validation enabled a closed-loop design pipeline for catalyst optimization.

Experimental results

Research questions

• RQ1 Which transition metal-doped carbon quantum dot (M@CQD) configuration exhibits the highest HER activity among earth-abundant elements?
• RQ2 How do catalyst loading, electrode type, temperature, and pH influence the HER performance of M@CQD catalysts?
• RQ3 Can a Bayesian genetic algorithm (BGA) model accurately predict the optimal M@CQD configuration and operating conditions for HER?
• RQ4 What is the intrinsic electronic and surface mechanism behind the enhanced HER activity in Ni@CQD as predicted by DFT simulations?
• RQ5 To what extent does experimental validation confirm the ML-predicted performance of the Ni@CQD catalyst?

Key findings

• The Ni@CQD catalyst on 3D reduced graphene oxide (3D rGO) achieved an overpotential of 189 mV to reach a current density of 10 mA cm⁻² in acidic electrolyte.
• The Tafel slope of the Ni@CQD catalyst was measured at 52 mV dec⁻¹, indicating favorable Volmer-Tafel or Heyrovsky reaction kinetics.
• The Ni@CQD catalyst demonstrated excellent durability under acidic conditions, maintaining performance over extended operation.
• The BGA model successfully predicted the optimal catalyst configuration and operating conditions with high accuracy, validated by experimental results.
• DFT simulations confirmed that Ni doping significantly modifies the electronic structure of CQDs, enhancing hydrogen adsorption free energy and facilitating proton transfer.
• The integrated ML-experimental-DFT framework enabled rapid, rational design of high-performance, noble-metal-free HER electrocatalysts.