Yuming Gu
Research Subjects
Yuming Gu has a total of 97 co-authors. Here are the top co-authors by number of shared publications:
- Jing Ma
State Key Laboratory of Coordination Chemistry MOE Key Laboratory of Mesoscopic Chemistry MOE Key Laboratory of High Performance Polymer Materials and Technology Jiangsu Key Laboratory of Green Energy Catalysis and Intelligent Chemical Engineering Suzhou Key Laboratory of Green Intelligent Manufacturing of New Energy Materials and Devices Tianchang New Materials and Energy Technologies Research Center Institute of Green Chemistry and Engineering School of Chemistry Nanjing University Nanjing...
7 shared publications - Jing-Lin Zuo
State Key Laboratory of Coordination Chemistry School of Chemistry and Chemical Engineering Nanjing University Nanjing P. R. China
4 shared publications - Jian Su
School of Chemistry and Chemical Engineering, Nanjing University of Science and Technology, Nanjing, Jiangsu 210094, China
3 shared publications - Shuai Yuan
State Key Laboratory of Coordination Chemistry, School of Chemistry and Chemical Engineering, Nanjing University, Nanjing 210023, P. R. China
3 shared publications Jiawei Chen
State Key Laboratory of Coordination Chemistry, Key Laboratory of Mesoscopic Chemistry of Ministry of Education, School of Chemistry and Chemical Engineering, Nanjing University, Nanjing 210093, China
2 shared publications- Mengning Ding
Nanjing University
2 shared publications
Latest Publications
- 9 February 2026
- Published by American Chemical Society (ACS) in Journal of the American Chemical Society
- Vol. 148 (7) , 7495-7504
Auxiliary Linker-Enabled Local Protonation Boosts the Ammonia Electrosynthesis in Heterometallic Metal–Organic Frameworks
Controlling the local proton environment is a powerful strategy for accelerating proton-coupled electron transfer reactions, but realizing this within stable heterogeneous catalysts remains challenging. Herein, we report a series of metal-organic frameworks (MOFs), PCN-800M-L, featuring precisely placed proton relays adjacent to catalytic sites for boosting electrocatalytic nitrate reduction to ammonia (NO3RR). PCN-800M-L series was constructed by sequentially installing catalytically active metal sites (M = Ni2+ or Co2+) and auxiliary nitrogen-donor linkers (L = L1-L5) into a robust Zr4+-based framework. Auxiliary linkers reversibly dissociate upon protonation, transiently exposing active metal sites while functioning as localized proton relays. By systematically varying the metals and linkers, we uncover a volcano-type dependence of NO3RR activity on linker pKa, with the PCN-800Co-L3 achieving the highest turnover frequency (1863 h-1) and Faradaic efficiency (97.9%). This work establishes a molecular-level design principle that harnesses auxiliary linkers as programmable proton sources in MOFs, enabling efficient multielectron electrocatalysis.
- 9 February 2026
- Published by American Chemical Society (ACS) in Journal of the American Chemical Society
- Vol. 148 (7) , 7495-7504
Auxiliary Linker-Enabled Local Protonation Boosts the Ammonia Electrosynthesis in Heterometallic Metal–Organic Frameworks
Controlling the local proton environment is a powerful strategy for accelerating proton-coupled electron transfer reactions, but realizing this within stable heterogeneous catalysts remains challenging. Herein, we report a series of metal-organic frameworks (MOFs), PCN-800M-L, featuring precisely placed proton relays adjacent to catalytic sites for boosting electrocatalytic nitrate reduction to ammonia (NO3RR). PCN-800M-L series was constructed by sequentially installing catalytically active metal sites (M = Ni2+ or Co2+) and auxiliary nitrogen-donor linkers (L = L1-L5) into a robust Zr4+-based framework. Auxiliary linkers reversibly dissociate upon protonation, transiently exposing active metal sites while functioning as localized proton relays. By systematically varying the metals and linkers, we uncover a volcano-type dependence of NO3RR activity on linker pKa, with the PCN-800Co-L3 achieving the highest turnover frequency (1863 h-1) and Faradaic efficiency (97.9%). This work establishes a molecular-level design principle that harnesses auxiliary linkers as programmable proton sources in MOFs, enabling efficient multielectron electrocatalysis.
- Open Access
- 3 September 2025
- Published by American Chemical Society (ACS) in ChemRxiv
LightChem: A Lightweight Domain-Specific Language Model for Molecular Property and Reaction Prediction in Chemistry
Large language models (LLMs) have achieved breakthroughs in natural language processing, yet their use in chemistry is limited by insufficient domain knowledge, high computational demands, and poor scalability. To overcome these challenges, we developed LightChem, a lightweight model that integrates retrieval-augmented generation (RAG) and tool-calling capabilities, providing a platform driven by chemical knowledge and theoretical calculations. LightChem adopts a three-layer architecture, data aggregation, intelligent retrieval, and specialized tool integration, supported by the self-developed RAPTOR system for real-time literature updates. The platform incorporates a suite of computational modules, including automated molecular property prediction (e.g., PoLogP), retrosynthetic planning via ReSynZ, and scalable quantum chemistry packages such as CIM and GEBF, enabling high-accuracy simulations of complex systems. Case studies demonstrate successful predictions of lipophilicity, excitation/emission spectra, and synthetic pathways for drug-like molecules, while large-scale calculations reproduce binding energies in zeolite clusters and excitation energies in GFP chromophores with accuracy comparable to experimental values. Beyond core computations, LightChem features a user-friendly web interface that connects sophisticated algorithms with practical laboratory workflows. It provides modules for reagent management, synthesis assessment, safety evaluation, and experimental design, thereby lowering the barrier for non-specialists and supporting systematic research in photosensitive and functional molecules. Applications to gold nanoclusters and zeolites further highlight the platform’s potential in photocatalysis and optical materials, while also revealing current limitations in predicting HOMO–LUMO gaps and optical properties of complex aggregates. LightChem demonstrates good performance in molecular property prediction, retrosynthetic planning, and large-scale quantum simulations, making it a versatile assistant for chemical research, bridging knowledge-driven reasoning with first-principles accuracy.
- 3 September 2025
- Published by American Chemical Society (ACS) in ChemRxiv
Open AccessLightChem: A Lightweight Domain-Specific Language Model for Molecular Property and Reaction Prediction in Chemistry
Large language models (LLMs) have achieved breakthroughs in natural language processing, yet their use in chemistry is limited by insufficient domain knowledge, high computational demands, and poor scalability. To overcome these challenges, we developed LightChem, a lightweight model that integrates retrieval-augmented generation (RAG) and tool-calling capabilities, providing a platform driven by chemical knowledge and theoretical calculations. LightChem adopts a three-layer architecture, data aggregation, intelligent retrieval, and specialized tool integration, supported by the self-developed RAPTOR system for real-time literature updates. The platform incorporates a suite of computational modules, including automated molecular property prediction (e.g., PoLogP), retrosynthetic planning via ReSynZ, and scalable quantum chemistry packages such as CIM and GEBF, enabling high-accuracy simulations of complex systems. Case studies demonstrate successful predictions of lipophilicity, excitation/emission spectra, and synthetic pathways for drug-like molecules, while large-scale calculations reproduce binding energies in zeolite clusters and excitation energies in GFP chromophores with accuracy comparable to experimental values. Beyond core computations, LightChem features a user-friendly web interface that connects sophisticated algorithms with practical laboratory workflows. It provides modules for reagent management, synthesis assessment, safety evaluation, and experimental design, thereby lowering the barrier for non-specialists and supporting systematic research in photosensitive and functional molecules. Applications to gold nanoclusters and zeolites further highlight the platform’s potential in photocatalysis and optical materials, while also revealing current limitations in predicting HOMO–LUMO gaps and optical properties of complex aggregates. LightChem demonstrates good performance in molecular property prediction, retrosynthetic planning, and large-scale quantum simulations, making it a versatile assistant for chemical research, bridging knowledge-driven reasoning with first-principles accuracy.
- Open Access
- 21 May 2025
- Published by American Chemical Society (ACS) in JACS Au
- Vol. 5 (6) , 2523-2532
Dual-Enzyme-Mimicking Sites in Covalent Organic Frameworks Enable Highly Efficient Relay Electrosynthesis of Ammonia
Electrocatalytic nitrate reduction poses significant importance in green ammonia synthesis and water denitrification. To date, high-performance nitrate reduction still relies on noble-metal-based catalysts, while transition metal compound and framework catalysts generally suffer from low activities (in terms of turnover frequency values), leading to insufficient ability to rapidly process large amounts of nitrate at the industrial level. To this end, enzyme-mimicking catalytic systems that integrate the high TOF efficiency of enzymatic sites (typically with nonprecious metal centers) into the crystalline, orderly assembled, and porous molecular frameworks hold great theoretical promises, yet the search for any single enzyme-site mimics has failed to overcome current limitations. Herein, we demonstrate the rational design of a series of covalent organic frameworks (COFs) to incorporate periodically alternating metalloporphyrin and metal-bis(dithiolene), respectively mimicking nitrate reductase and nitrite reductase enzymes, as a dual functional assembly. The dual-enzyme-mimicking Ni-TAPP-Cu boosts the electrocatalytic activity of NO3RR to a comparable level to the noble-metal catalysts via a relay pathway to individually boost the two kinetically significant nitrate-to-nitrite and nitrite-to-ammonia conversion steps, consequently achieving a remarkable TOF value of 235.7 mg·h–1·mg–1 site, 86.13% NH3 faradaic efficiency, nearly 100% NH3-selectivity, and excellent durability at typical acidic wastewater conditions (pH = 3). Mechanistic investigations indicate that Cu sites effectively suppress the HER competition, enhance the adsorption of NO3 –, and selectively generate NO2 –, while Ni sites facilitate the subsequent hydrogenation and conversion of *NO2. The Cu–Ni–COF exemplifies a promising strategy for integrating dual enzyme-mimicking sites within an ordered, porous framework to facilitate relay catalysis, thereby offering valuable insights for the design of high-performance, sustainable catalytic materials.
- 21 May 2025
- Published by American Chemical Society (ACS) in JACS Au
- Vol. 5 (6) , 2523-2532
Open AccessDual-Enzyme-Mimicking Sites in Covalent Organic Frameworks Enable Highly Efficient Relay Electrosynthesis of Ammonia
Electrocatalytic nitrate reduction poses significant importance in green ammonia synthesis and water denitrification. To date, high-performance nitrate reduction still relies on noble-metal-based catalysts, while transition metal compound and framework catalysts generally suffer from low activities (in terms of turnover frequency values), leading to insufficient ability to rapidly process large amounts of nitrate at the industrial level. To this end, enzyme-mimicking catalytic systems that integrate the high TOF efficiency of enzymatic sites (typically with nonprecious metal centers) into the crystalline, orderly assembled, and porous molecular frameworks hold great theoretical promises, yet the search for any single enzyme-site mimics has failed to overcome current limitations. Herein, we demonstrate the rational design of a series of covalent organic frameworks (COFs) to incorporate periodically alternating metalloporphyrin and metal-bis(dithiolene), respectively mimicking nitrate reductase and nitrite reductase enzymes, as a dual functional assembly. The dual-enzyme-mimicking Ni-TAPP-Cu boosts the electrocatalytic activity of NO3RR to a comparable level to the noble-metal catalysts via a relay pathway to individually boost the two kinetically significant nitrate-to-nitrite and nitrite-to-ammonia conversion steps, consequently achieving a remarkable TOF value of 235.7 mg·h–1·mg–1 site, 86.13% NH3 faradaic efficiency, nearly 100% NH3-selectivity, and excellent durability at typical acidic wastewater conditions (pH = 3). Mechanistic investigations indicate that Cu sites effectively suppress the HER competition, enhance the adsorption of NO3 –, and selectively generate NO2 –, while Ni sites facilitate the subsequent hydrogenation and conversion of *NO2. The Cu–Ni–COF exemplifies a promising strategy for integrating dual enzyme-mimicking sites within an ordered, porous framework to facilitate relay catalysis, thereby offering valuable insights for the design of high-performance, sustainable catalytic materials.
- 14 May 2025
- Published by American Chemical Society (ACS) in Journal of the American Chemical Society
- Vol. 147 (21) , 18327-18337
Square-Planar Tetranuclear Cluster-Based High-Symmetry Coordination Metal–Organic Polymers for Efficient Electrochemical Nitrate Reduction to Ammonia
Metal-organic polymers (MOPs) are gaining booming attention as atomically precise single-site catalysts for electrochemical nitrate-to-ammonia conversion owing to their regular structures and tunable functionalities. However, a molecular-level understanding is still lacking for the design of more efficient MOP electrocatalysts. Here, we report the construction of high-symmetry coordination MOPs (Mn-TATB, Fe-TATB, and Co-TATB), utilizing square-planar tetranuclear building units [M4(μ4-O)(CO2)8] (M = Mn, Fe, or Co) bridged by 2,4,6-tris(4-carboxyphenyl)-1,3,5-triazine (H3TATB) ligands. These MOPs possess distinct coordination motifs with well-defined porosity, high-density catalytic sites, accessible mass transfer channels, and nanoconfined chemical environments. Benefited from the unique metal-organic coordination framework, Co-TATB demonstrated a remarkable ammonia production Faradaic efficiency (FENH3) of ∼98% across a wide potential range (-0.7 to -1.0 V (vs RHE)) in the electrocatalytic nitrate reduction reaction (NITRR) and maintained stable performance over a long duration when tested in a flow cell at an industrially relevant current density of ∼332.1 mA cm-2. Furthermore, in situ spectroscopic analyses, combined with theoretical calculations, elucidate the intrinsic reaction pathway of the Co-TATB model during the NITRR process. These findings offer insightful perspectives on the strategic design of electrocatalysts with symmetrical configurations for the purification of nitrate-containing wastewater and the green synthesis of ammonia.
- 14 May 2025
- Published by American Chemical Society (ACS) in Journal of the American Chemical Society
- Vol. 147 (21) , 18327-18337
Square-Planar Tetranuclear Cluster-Based High-Symmetry Coordination Metal–Organic Polymers for Efficient Electrochemical Nitrate Reduction to Ammonia
Metal-organic polymers (MOPs) are gaining booming attention as atomically precise single-site catalysts for electrochemical nitrate-to-ammonia conversion owing to their regular structures and tunable functionalities. However, a molecular-level understanding is still lacking for the design of more efficient MOP electrocatalysts. Here, we report the construction of high-symmetry coordination MOPs (Mn-TATB, Fe-TATB, and Co-TATB), utilizing square-planar tetranuclear building units [M4(μ4-O)(CO2)8] (M = Mn, Fe, or Co) bridged by 2,4,6-tris(4-carboxyphenyl)-1,3,5-triazine (H3TATB) ligands. These MOPs possess distinct coordination motifs with well-defined porosity, high-density catalytic sites, accessible mass transfer channels, and nanoconfined chemical environments. Benefited from the unique metal-organic coordination framework, Co-TATB demonstrated a remarkable ammonia production Faradaic efficiency (FENH3) of ∼98% across a wide potential range (-0.7 to -1.0 V (vs RHE)) in the electrocatalytic nitrate reduction reaction (NITRR) and maintained stable performance over a long duration when tested in a flow cell at an industrially relevant current density of ∼332.1 mA cm-2. Furthermore, in situ spectroscopic analyses, combined with theoretical calculations, elucidate the intrinsic reaction pathway of the Co-TATB model during the NITRR process. These findings offer insightful perspectives on the strategic design of electrocatalysts with symmetrical configurations for the purification of nitrate-containing wastewater and the green synthesis of ammonia.
- 26 February 2025
- Published by American Chemical Society (ACS) in The Journal of Physical Chemistry Letters
- Vol. 16 (9) , 2357-2368
Digital Descriptors in Predicting Catalysis Reaction Efficiency and Selectivity
Accurately controlling the interactions and dynamic changes between multiple active sites (e.g., metals, vacancies, and lone pairs of heteroatoms) to achieve efficient catalytic performance is a key issue and challenge in the design of complex catalytic reactions involving 2D metal-supported catalysts, metal-zeolites, metal-organic catalysts, and metalloenzymes. With the aid of machine learning (ML), descriptors play a central role in optimizing the electrochemical performance of catalysts, elucidating the essence of catalytic activity, and predicting more efficient catalysts, thereby avoiding time-consuming trial-and-error processes. Three kinds of descriptors─active center descriptors, interfacial descriptors, and reaction pathway descriptors─are crucial for understanding and designing metal-supported catalysts. Specifically, vacancies, as active sites, synergize with metals to significantly promote the reduction reactions of energy-relevant small molecules. By combining some physical descriptors, interpretable descriptors can be constructed to evaluate catalytic performance. Future development of descriptors and ML models faces the challenge of constructing descriptors for vacancies in multicatalysis systems to rationally design the activity, selectivity, and stability of catalysts. Utilization of generative artificial intelligence and multimodal ML to automatically extract descriptors would accelerate the exploration of dynamic reaction mechanisms. The transferable descriptors from metal-supported catalysts to artificial metalloenzymes provide innovative solutions for energy conversion and environmental protection.
- 26 February 2025
- Published by American Chemical Society (ACS) in The Journal of Physical Chemistry Letters
- Vol. 16 (9) , 2357-2368
Digital Descriptors in Predicting Catalysis Reaction Efficiency and Selectivity
Accurately controlling the interactions and dynamic changes between multiple active sites (e.g., metals, vacancies, and lone pairs of heteroatoms) to achieve efficient catalytic performance is a key issue and challenge in the design of complex catalytic reactions involving 2D metal-supported catalysts, metal-zeolites, metal-organic catalysts, and metalloenzymes. With the aid of machine learning (ML), descriptors play a central role in optimizing the electrochemical performance of catalysts, elucidating the essence of catalytic activity, and predicting more efficient catalysts, thereby avoiding time-consuming trial-and-error processes. Three kinds of descriptors─active center descriptors, interfacial descriptors, and reaction pathway descriptors─are crucial for understanding and designing metal-supported catalysts. Specifically, vacancies, as active sites, synergize with metals to significantly promote the reduction reactions of energy-relevant small molecules. By combining some physical descriptors, interpretable descriptors can be constructed to evaluate catalytic performance. Future development of descriptors and ML models faces the challenge of constructing descriptors for vacancies in multicatalysis systems to rationally design the activity, selectivity, and stability of catalysts. Utilization of generative artificial intelligence and multimodal ML to automatically extract descriptors would accelerate the exploration of dynamic reaction mechanisms. The transferable descriptors from metal-supported catalysts to artificial metalloenzymes provide innovative solutions for energy conversion and environmental protection.
- 31 January 2025
- Published by American Chemical Society (ACS) in Langmuir
- Vol. 41 (5) , 3490-3502
Automated Machine Learning of Interfacial Interaction Descriptors and Energies in Metal-Catalyzed N2 and CO2 Reduction Reactions
The applications of machine learning (ML) in complex interfacial interactions are hindered by the time-consuming process of manual feature selection and model construction. An automated ML program was implemented with four subsequent steps: data distribution analysis, dimensionality reduction and clustering, feature selection, and model optimization. Without the need of manual intervention, the descriptors of metal charge variance (ΔQCT) and electronegativity of substrate (χsub) and metal (δχM) were raised up with good performance in predicting electrochemical reaction energies for both nitrogen reduction reaction (NRR) and CO2 reduction reaction (CO2RR) on metal-zeolites and MoS2 surfaces. The important role of interfacial interactions in tuning the catalytic reactivity in NRR and CO2RR was highlighted from SHAP analysis. It was proposed that Fe-, Cr-, Zn-, Nb-, and Ta-zeolites are favorable catalysts for NRR, while Ni-zeolite showed a preference for CO2RR. An elongated bond of N2 or a bent configuration of CO2 was shown in V-, Co-, and Mo-zeolites, indicating that the molecule could be activated after the adsorption in both NRR and CO2RR pathways. The generalizability of the automatically built ML model is demonstrated from applications to other catalytic systems such as metal-organic frameworks and SiO2 surfaces. The automated ML program is a useful tool to accelerate the data-driven exploration of relationship between structures and material properties without the need of manual feature selection.
- 31 January 2025
- Published by American Chemical Society (ACS) in Langmuir
- Vol. 41 (5) , 3490-3502
Automated Machine Learning of Interfacial Interaction Descriptors and Energies in Metal-Catalyzed N2 and CO2 Reduction Reactions
The applications of machine learning (ML) in complex interfacial interactions are hindered by the time-consuming process of manual feature selection and model construction. An automated ML program was implemented with four subsequent steps: data distribution analysis, dimensionality reduction and clustering, feature selection, and model optimization. Without the need of manual intervention, the descriptors of metal charge variance (ΔQCT) and electronegativity of substrate (χsub) and metal (δχM) were raised up with good performance in predicting electrochemical reaction energies for both nitrogen reduction reaction (NRR) and CO2 reduction reaction (CO2RR) on metal-zeolites and MoS2 surfaces. The important role of interfacial interactions in tuning the catalytic reactivity in NRR and CO2RR was highlighted from SHAP analysis. It was proposed that Fe-, Cr-, Zn-, Nb-, and Ta-zeolites are favorable catalysts for NRR, while Ni-zeolite showed a preference for CO2RR. An elongated bond of N2 or a bent configuration of CO2 was shown in V-, Co-, and Mo-zeolites, indicating that the molecule could be activated after the adsorption in both NRR and CO2RR pathways. The generalizability of the automatically built ML model is demonstrated from applications to other catalytic systems such as metal-organic frameworks and SiO2 surfaces. The automated ML program is a useful tool to accelerate the data-driven exploration of relationship between structures and material properties without the need of manual feature selection.
- 3 February 2023
- Published by American Association for the Advancement of Science (AAAS) in Science Advances
- Vol. 9 (5) , eadf2398
Covalent organic frameworks with Ni-Bis(dithiolene) and Co-porphyrin units as bifunctional catalysts for Li-O 2 batteries
The rational design of efficient and stable catalysts for the oxygen reduction reaction and oxygen evolution reaction (ORR/OER) is the key to improving Li-O 2 battery performance. Here, we report the construction of ORR/OER bifunctional cathode catalysts in a covalent organic framework (COF) platform by simultaneously incorporating Ni-bis(dithiolene) and Co-porphyrin units. The resulting bimetallic Ni/Co-COF exhibits high surface area, fairly good electrical conductivity, and excellent chemical stability. Li-O 2 batteries with the Ni/Co-COF–based cathode show a low discharge/charge potential gap (1.0 V) and stable cycling (200 cycles) at a current density of 500 mA g −1 , rivaling that of PtAu nanocrystals. Density functional theory computations and control experiments using nonmetal or single metal–based isostructural COFs reveal the critical role of Ni and Co sites in reducing the discharge/charge overpotentials and regulating the Li 2 O 2 deposition. This work highlights the advantage of bimetallic COFs in the rational design of efficient and stable Li-O 2 batteries.
- 3 February 2023
- Published by American Association for the Advancement of Science (AAAS) in Science Advances
- Vol. 9 (5) , eadf2398
Covalent organic frameworks with Ni-Bis(dithiolene) and Co-porphyrin units as bifunctional catalysts for Li-O 2 batteries
The rational design of efficient and stable catalysts for the oxygen reduction reaction and oxygen evolution reaction (ORR/OER) is the key to improving Li-O 2 battery performance. Here, we report the construction of ORR/OER bifunctional cathode catalysts in a covalent organic framework (COF) platform by simultaneously incorporating Ni-bis(dithiolene) and Co-porphyrin units. The resulting bimetallic Ni/Co-COF exhibits high surface area, fairly good electrical conductivity, and excellent chemical stability. Li-O 2 batteries with the Ni/Co-COF–based cathode show a low discharge/charge potential gap (1.0 V) and stable cycling (200 cycles) at a current density of 500 mA g −1 , rivaling that of PtAu nanocrystals. Density functional theory computations and control experiments using nonmetal or single metal–based isostructural COFs reveal the critical role of Ni and Co sites in reducing the discharge/charge overpotentials and regulating the Li 2 O 2 deposition. This work highlights the advantage of bimetallic COFs in the rational design of efficient and stable Li-O 2 batteries.
- 20 June 2022
- Published by Royal Society of Chemistry (RSC) in Journal of Materials Chemistry A
- Vol. 10 (28) , 14976-14988
Nitrogen reduction reaction energy and pathways in metal-zeolites: deep learning and explainable machine learning with local acidity and hydrogen bonding features
Conversion from very inert nitrogen to ammonia is crucial to the industrial production of fertilizers and other fundamental chemicals. To predict the possible reaction pathways and barriers of the nitrogen reduction reaction (NRR) in metal-zeolites, we set up a dataset consisting of electronic structures and reaction energies for 492 metal-zeolite structures with 27 metal elements (including 3d, 4d, and 5d transition metal and main group Pb and Bi metal atoms). The important role of hydrogen bonding (HB) interaction in facilitating the hydrogenation step (N2* → NNH*) but disfavoring the dehydrogenation step (NNH3* → N*) of the NRR is revealed by using a graph convolutional neural network (GCNN) model with the multilevel attention mechanism. A local acidity (LA) descriptor is defined by the combination of the electronegativity, first ionization energy, and atom radius of the embedded metal in various substrates (including zeolites, 2D materials, etc.), from which good predictivity of reaction energies was gained by using explainable machine learning (ML) models (e.g., XGBoost: R2 = 0.84) with an addition of geometry distortion and charge transfer descriptors along reaction pathways. The pathway probability of distal, alternating, and enzymatic processes is also predicted from the geometry and charge features with satisfactory performance. The promising zeolites with embedded Ti, Co and Nb atoms were recommended to favor the NRR with lower energy inputs of 0.13, 0.04, and 0.54 eV. For experimentally synthesized photocatalytic Ti-exchanged zeolites, the ML predicted maximum energy inputs are decreased with decreasing LA values, in line with the increased ammonia yield from 4A, 3A, to 5A of pore sizes. The predictive model for energy and pathway prediction and the LA descriptor could also be transferred to metal decorated 2D materials and non-metal B-containing systems.
- 20 June 2022
- Published by Royal Society of Chemistry (RSC) in Journal of Materials Chemistry A
- Vol. 10 (28) , 14976-14988
Nitrogen reduction reaction energy and pathways in metal-zeolites: deep learning and explainable machine learning with local acidity and hydrogen bonding features
Conversion from very inert nitrogen to ammonia is crucial to the industrial production of fertilizers and other fundamental chemicals. To predict the possible reaction pathways and barriers of the nitrogen reduction reaction (NRR) in metal-zeolites, we set up a dataset consisting of electronic structures and reaction energies for 492 metal-zeolite structures with 27 metal elements (including 3d, 4d, and 5d transition metal and main group Pb and Bi metal atoms). The important role of hydrogen bonding (HB) interaction in facilitating the hydrogenation step (N2* → NNH*) but disfavoring the dehydrogenation step (NNH3* → N*) of the NRR is revealed by using a graph convolutional neural network (GCNN) model with the multilevel attention mechanism. A local acidity (LA) descriptor is defined by the combination of the electronegativity, first ionization energy, and atom radius of the embedded metal in various substrates (including zeolites, 2D materials, etc.), from which good predictivity of reaction energies was gained by using explainable machine learning (ML) models (e.g., XGBoost: R2 = 0.84) with an addition of geometry distortion and charge transfer descriptors along reaction pathways. The pathway probability of distal, alternating, and enzymatic processes is also predicted from the geometry and charge features with satisfactory performance. The promising zeolites with embedded Ti, Co and Nb atoms were recommended to favor the NRR with lower energy inputs of 0.13, 0.04, and 0.54 eV. For experimentally synthesized photocatalytic Ti-exchanged zeolites, the ML predicted maximum energy inputs are decreased with decreasing LA values, in line with the increased ammonia yield from 4A, 3A, to 5A of pore sizes. The predictive model for energy and pathway prediction and the LA descriptor could also be transferred to metal decorated 2D materials and non-metal B-containing systems.