Graph Neural Network-Accelerated Multitasking Genetic Algorithm for Optimizing PdTi_1-H Surfaces under Various CO₂...

Area of Science:

• Materials Science
• Catalysis
• Computational Chemistry

Background:

• Palladium (Pd) hydride catalysts show promise for CO2 reduction reaction (CO2RR) and hydrogen evolution reaction (HER).
• Previous research indicated Ti-doped and Ti-alloyed Pd hydrides enhance CO2RR performance over pure Pd hydride.
• Surface composition, ordering, and reaction conditions significantly influence catalyst stability, activity, and selectivity.

Purpose of the Study:

• To screen for stable and active Pd-Ti-H surfaces with multiple adsorbates for CO2RR under diverse reaction conditions.
• To overcome the complexity of multi-adsorbate systems and vast search spaces in theoretical catalyst screening.
• To identify novel catalyst compositions for efficient CO2 conversion and syngas generation.

Main Methods:

• Application of a deep learning-assisted multitasking genetic algorithm for catalyst screening.
• Utilizing an ensemble deep learning model to accelerate structure relaxation and ensure accuracy of energy and force calculations.
• Simultaneously identifying globally stable surface structures across various reaction conditions.

Main Results:

• Screened 23 stable Pd-Ti-H surface structures under different reaction conditions.
• Identified specific compositions (e.g., Pd0.56Ti0.44H1.06 + 25%CO, Pd0.31Ti0.69H1.25 + 50%CO) exhibiting high activity for CO2RR.
• Determined that certain Pd-Ti-H structures are suitable for generating syngas (CO and H2).

Conclusions:

• The deep learning-guided genetic algorithm effectively screens complex catalytic systems.
• Optimized Pd-Ti-H compositions demonstrate significant potential for CO2RR and syngas production.
• This approach provides a pathway for discovering advanced catalysts under realistic reaction conditions.