Page 74 - Read Online

Meng et al. J. Mater. Inf. 2025, 5, 3 https://dx.doi.org/10.20517/jmi.2024.74 Page 15 of 25

properties for NRR applications. Lv et al. developed a “five-step” screening strategy [Figure 8A] using high-
throughput DFT calculations to evaluate TM /CN catalysts for NRR. This approach identified Fe /g-CN as a
2
2
highly active catalyst with low energy consumption and high selectivity, concluding that moderate electron
*
donation from TM to N is crucial for balancing N activation and NH formation steps . Similarly, Sun et
[116]
2
3
2
al. used high-throughput computations to screen TM TM @C N [TM , TM = 3(4) d TM atoms] and
9
4
1
2
2
1
[117]
identified five effective catalysts (TM , TM = NiRu, FeNi, TiFe, TiNi, NiZr) with strong activity . Sun et
1
2
al. further applied a “four-step” screening strategy [Figure 8B], predicting that six catalysts, including
W@V -V -BC , Re@V -V -BC , Mo@V -V -BC , Ti@V -V -BC , Mo@V -V -BC , and Ta@V -V -BC ,
B
3
2
3
1
B
3
1
B
C
3
3
1
3
3
B
B
3
exhibited superior NRR activity and selectivity out of 33 candidates . Pei et al. systematically assessed the
[118]
stability of 3d-5d TM trimers embedded in C N nanosheets (TM @C N ) and found that configurations
3
3
3
3
3
with Re, Ru, or Pt trimers demonstrated excellent catalytic activity for NRR .
[108]
ML
While high-throughput computing provides a structured approach for catalyst design, the massive data
generated can obscure key performance factors, making it challenging to pinpoint the critical influences on
catalytic performance. Advanced data analysis tools are crucial to fully leverage high-throughput screening
for next-generation NRR catalysts. Emerging ML techniques are providing us with such a solution, enabling
rapid and efficient data extraction, prediction, and analysis. By accelerating the identification of high-
performance materials from thousands of candidates, ML not only enhances our understanding of
structure-activity relationships but also sheds light on the fundamental mechanisms of catalysis
[Figure 9A] .
[118]
In recent years, ML has shown promise in catalyst design, with applications emerging in NRR research.
Wang et al. used ML techniques to explore the nitrogen reduction reactivity of novel, graphene-supported
SACs, based on their prior high-throughput calculations [Figure 9B]. Their study evaluated 29 TMs and 57
ligand structures, generating 1,626 unexplored catalyst configurations. Using a trained ML model, they
) of these configurations and identified 45
f
predicted four target properties (E, ∆E N 2 , ∆G N 2 -N 2 H , ∆G NH 2 -NH 3
promising candidates for NRR [Figure 9C]. Among these, Mo-B CN-(orthogonal-B) performs the best, with
2
exceptionally low free energy along the distal pathway. Through statistical analysis, the researchers
developed a predictive descriptor (∆E , ∆G , ∆G ) with high generalizability, applicable to
N 2 N 2 -N 2 H NH 2 -NH 3
[119]
untested NRR catalyst systems .
Zhang et al. also applied ML models to evaluate the catalytic activity of SACs by directly predicting reaction
Gibbs free energy. Their findings underscored the high predictive accuracy of the gradient boosting
regression (GBR) model for both ∆G ( N → NNH) and ∆G ( NH → NH ). Feature importance analysis
*
*
*
*
3
2
2
revealed that the accuracy of the GBR model was due to its effective identification of key characteristics
related to the active center and coordination environment, with the covalent radius emerging as a
particularly influential descriptor .
[120]
Despite its potential, ML-based catalyst selection is currently constrained by limited experimental data.
Nevertheless, integrating high-throughput screening with ML strategies offers a promising pathway for
catalyst discovery, opening new horizons for efficient and targeted materials design.
DESCRIPTOR-BASED SCREENING AND DESIGN OF NRR CATALYSTS
Given the complexity of multiple intermediates in the NRR process, developing simplified parameters for
evaluating catalyst activity is beneficial. Compared with heavy DFT computations, descriptors can quickly
predict catalyst performance, providing experimental researchers with a practical tool for screening high-