Design high-entropy electrocatalyst via interpretable deep graph attention learning

Zhang, Jun; Wang, Chaohui; Huang, Shasha; Xiang, Xuepeng; Xiong, Yaoxu; Xu, Biao; Ma, Shunchao; Fu, Haijun; Kai, Ji‐Jung; Kang, Xiongwu; Zhao, Shijun

doi:10.1016/j.joule.2023.06.003

Cited by 24 publications

(5 citation statements)

References 77 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…As shown in Figure S6, we obtain MAEs of 0.057 eV on on-top OH* and 0.064 eV on fcc O* for the test split, where on-top OH* and fcc O* are jointly trained but separately visualized. It is noteworthy that state-of-the-art GNN models ,,, and conventional ML methods − typically show MAEs in range of 0.1–0.2 eV for adsorption enthalpy predictions. Our results indicate an excellent interpolation performance.…”

Section: Resultsmentioning

confidence: 99%

“…The main remaining limitation of our multiobjective BO framework is the simplified assumption of our surface model. Although the fcc(111) surface with a random arrangement of atoms is a good first approximation of HEA electrocatalyst surfaces that has been successfully applied in many theoretical studies, ,− ,− recent experiments have observed the formation of multiphases, which suggests that these assumptions may not be completely accurate for describing the true catalyst surface. At the present, this is a frontier area of HEA research.…”

Section: Resultsmentioning

confidence: 99%

“…Despite recent advances, these efforts are restricted to relatively low-dimensional compositions, barely scratching the surface of the vast design space available for this class of materials. For instance, recent theoretical ,,− and experimental ,, studies limit the optimization to 5-element HEA spaces, and the selected constituent five elements are primarily based on experimental observations or chemical intuition. This hinders the understanding of higher-dimensional landscapes and limits exploration to regions that have been extensively experimentally tested.…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Discovering High Entropy Alloy Electrocatalysts in Vast Composition Spaces with Multiobjective Optimization

Xu,

Diesen,

et al. 2024

J. Am. Chem. Soc.

View full text Add to dashboard Cite

High entropy alloys (HEAs) are a highly promising class of materials for electrocatalysis as their unique active site distributions break the scaling relations that limit the activity of conventional transition metal catalysts. Existing Bayesian optimization (BO)-based virtual screening approaches focus on catalytic activity as the sole objective and correspondingly tend to identify promising materials that are unlikely to be entropically stabilized. Here, we overcome this limitation with a multiobjective BO framework for HEAs that simultaneously targets activity, costeffectiveness, and entropic stabilization. With diversity-guided batch selection further boosting its data efficiency, the framework readily identifies numerous promising candidates for the oxygen reduction reaction that strike the balance between all three objectives in hitherto unchartered HEA design spaces comprising up to 10 elements.

show abstract

Section: Resultsmentioning

confidence: 99%

Section: Resultsmentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Discovering High Entropy Alloy Electrocatalysts in Vast Composition Spaces with Multiobjective Optimization

Xu,

Diesen,

et al. 2024

J. Am. Chem. Soc.

View full text Add to dashboard Cite

show abstract

“…207 Addressing this, Kang et al developed an accurate and efficient atomic graph attention (AGAT) network to expedite the creation of high-performance HEA electrocatalysts. 208 In a statistically rigorous framework, the dependability of scaling relations and classical d-band theory is affirmed across the surfaces of HEA electrocatalysts. The AGAT network, meticulously trained, coupled with electronic structure analysis, gauges the viability of linear scaling and d-band theory within the intricate confines of HEA surfaces.…”

Section: Design Strategy Of Heasmentioning

confidence: 98%

High-entropy alloys in electrocatalysis: from fundamentals to applications

Ren,

Chen,

Wang

et al. 2023

Chem. Soc. Rev.

112

View full text Add to dashboard Cite

show abstract

“…It was demonstrated that the pre-trained GNN model for Ru catalyst could be fine-tuned using a limited amount of DFT data for Fe catalyst to efficiently transfer acquired knowledge from Ru to Fe, while maintaining a high prediction accuracy in predicting adsorption energy. Another effort in this direction is that of Zhang et al [30] wherein an attention-based GNN was developed to investigate the compositional space comprising Ni-Co-Fe-Pd-Pt for high-entropy electrocatalysis. The model was used to identify new catalyst candidates, which were subsequently validated using experiments.…”

Section: Introductionmentioning

confidence: 99%

Interpretable Attention-based Transfer Learning in Plasma Catalysis: A Study on the Role of Surface Charge

Mesbah,

Shao,

Lele

et al. 2024

Preprint

View full text Add to dashboard Cite

Low-temperature plasma catalysis holds promise for electrification of energy-intensive chemical processes such as methane reforming and ammonia synthesis. However, fundamental understanding of plasma-catalyst interactions, essential for catalyst design and screening for plasma catalysts, remains largely limited. Recent work has demonstrated the importance of first-principles studies, including density functional theory (DFT), for elucidating the role of electro- and photo-effects such as electric field and charge in plasma catalysis. The availability of increasing amounts of DFT data in thermal catalysis presents a unique opportunity for plasma catalysis research to efficiently leverage this existing first-principles knowledge of thermal catalysis towards investigating plasma-catalyst interactions. To this end, this paper investigates interpretable transfer learning from thermal to plasma catalysis, with a focus on the role of surface charge. Pre-trained attention-based graph neural networks (GNNs) from the Open Catalysis Project, trained using millions of thermal catalysis DFT data points, are structurally adapted to account for surface charge effects and fine-tuned using plasma catalysis DFT data of single metal atoms on Al$_2$O$_3$ support and adsorbates involved in plasma-catalytic ammonia synthesis. Not only the fine-tuned attention-based GNN model provides a high test accuracy for predicting adsorption energies and atomic forces in plasma catalysis, but also shows adequate extrapolation for unseen single metal atoms in the plasma catalysis data used for the model fine-tuning. To distinguish the effects of surface charge from other dissimilarities in DFT data of thermal and plasma catalysis, a dual-model framework is presented that relies on two pre-trained GNNs, one of which is specifically tasked to capture surface charge effects using an attention mechanism that provides interpretable insights into their role. Lastly, it is demonstrated how the attention-based GNNs developed for single metal atoms can be efficiently adapted for predicting adsorption energies and atomic forces for metal clusters in plasma catalysis. This work highlights the vast potential of interpretable transfer learning from thermal catalysis to plasma catalysis to mitigate excessive computational requirements of first-principles studies in plasma catalysis, towards accelerating fundamental research in this domain.

show abstract

Design high-entropy electrocatalyst via interpretable deep graph attention learning

Cited by 24 publications

References 77 publications

Discovering High Entropy Alloy Electrocatalysts in Vast Composition Spaces with Multiobjective Optimization

Discovering High Entropy Alloy Electrocatalysts in Vast Composition Spaces with Multiobjective Optimization

High-entropy alloys in electrocatalysis: from fundamentals to applications

Interpretable Attention-based Transfer Learning in Plasma Catalysis: A Study on the Role of Surface Charge

Contact Info

Product

Resources

About