Copper oxide-based catalysts are widely used in electrocatalytic CO₂reduction reactions due to their unique surface chemical properties. During the reaction, copper oxide catalysts can rely on their inherent oxidation state stability, avoiding structural degradation caused by surface redox changes, which is one of the common deactivation mechanisms of metal catalysts. As a metal oxide, copper oxide exhibits typical alkaline characteristics, effectively enhancing CO₂adsorption and further improving catalytic activity. Cu₂O and CuO are the two most common forms of copper oxide catalysts, exhibiting different catalytic activities at different reduction potentials, and the adsorption strength of *CO increases with the increase of copper oxidation state^[75].

Synergistic effects can significantly enhance catalyst performance beyond that of individual components. Roy et al.^[51]prepared Cu₂O/CuO thin-film catalysts on conductive nickel foam via electrodeposition and investigated their performance in the CO₂reduction reaction. When subjected to a 120-minute reduction reaction at -1.3 V (vs Ag/AgCl) in KHCO₃solution, Cu₂O/CuO exhibited high electrocatalytic activity, with a maximum current density reaching 46 mA/cm², and demonstrated high selectivity toward methanol. The high activity is likely due to the synergistic effect arising from the Cu₂O/CuO heterostructure formed by Cu(I) and Cu(II), which helps enhance the functional properties of the material. The surface copper oxide provides an appropriate hydrogen overpotential and CO adsorption energy, facilitating efficient methanol production.

The specific catalyst structure can significantly enhance its performance. Azenha et al.^[52]studied a CuO nanowire catalyst, achieving high selectivity for methanol production without the use of precious metals, and effectively suppressing the competitive hydrogen evolution reaction. This nanowire structure not only provides a large specific surface area, increasing the number of active sites, but also its unique electronic properties further reduce the overpotential of the reaction. The methanol Faraday efficiency reached 66.4%, with a yield of 1.27×10^-4 mol·m^-2·s^-1, representing a 6.7% improvement over previous catalysts and demonstrating excellent selectivity under mild conditions.

The facet effect plays an important role in the regulation of catalyst performance. Liu et al.^[53]synthesized Cu₂O nanostructures with tunable facets via a hydrothermal method (Figure 2a) and investigated the influence of different crystal facets on the CO₂reduction reaction. The octahedral Cu₂O (Cu₂O-o) exhibited the optimal CO₂reduction performance, achieving a total Faraday efficiency of 35.4% for alcohol products, with methanol accounting for 4.9%. This study indicated that differences in oxygen vacancy defects and Cu—O bond lengths on various crystal facets are key factors influencing catalytic activity. Additionally, doping and defect engineering are effective strategies for enhancing the CO₂reduction performance of copper oxide-based catalysts, further improving their electrochemical properties and boosting their stability and efficiency in long-term reactions. Guo et al.^[54]developed a novel catalyst composed of atomically dispersed tin and defective CuO, achieving a methanol Faraday efficiency as high as 88.6% at a current density of 67.0 mA/cm². The introduction of defects significantly enhanced CO₂adsorption and electron transfer processes, thereby improving catalytic efficiency and demonstrating that the tin-doped CuO catalyst exhibits excellent stability and selectivity in the CO₂reduction reaction.

Copper oxide-based catalysts exhibit high methanol selectivity due to their abundant oxygen vacancies and tunable oxidation states (Cu⁺/Cu²⁺). However, they still suffer from issues such as low electrical conductivity, easy loss of active sites, and susceptibility to reduction to Cu⁰under strongly reducing conditions, leading to decreased selectivity. Therefore, optimizing the electrical conductivity of copper oxide catalysts, enhancing their resistance to reduction, and improving their interfacial stability are important directions for future research.

Copper alloys regulate the position of the d-band center by introducing a second or multiple metals (such as Ag, Co, etc.), optimizing CO₂ adsorption, activation, and the subsequent stability of intermediates. A notable feature of these catalysts is their ability to enhance the adsorption capacity for specific reaction intermediates through synergistic effects by adjusting alloy composition and surface structure, thereby improving methanol Faraday efficiency.

Han Buxing's team^[55]employed an in-situ dual-doping strategy, introducing silver and sulfur elements to optimize the CO₂reduction performance of the Cu₂O/Cu catalyst. As shown in the left panel of Figure 2b, the dual-doping strategy regulates the electronic structure of the catalyst through synergistic effects, enabling the Ag,S-Cu₂O/Cu catalyst to achieve a methanol Faraday efficiency as high as 67.4% and a current density of 122.7 mA/cm². Both theoretical calculations and experimental results indicate that the introduction of Ag significantly increases the reaction energy barrier for HER, effectively suppressing the HER process and enhancing methanol selectivity and catalyst activity. Meanwhile, the doping of S further modulates the surface electronic structure of the catalyst, improving the adsorption and conversion efficiency of key intermediates and synergistically boosting overall catalytic performance. The right panel shows that when ΔG _*CHO-ΔG _*COis close to 0.6 eV, the Ag,S-Cu₂O/Cu exhibits the highest methanol current density, indicating that this free energy difference favors the preferential formation of CH₃OH.

The synergistic effect of copper alloys can also enhance the stability of the catalyst, enabling it to exhibit superior resistance to deactivation during prolonged electrolysis. Peng et al.^[56]synthesized a Cu-Co PBA-V_CNcatalyst using wet chemical methods and hydrogen-cooled plasma etching, demonstrating outstanding CO₂electroreduction performance. At -0.9 V (vs RHE), the total Faraday efficiency for methanol and ethanol reached 83.8%. During a 100-hour long-term test, the low-carbon alcohol FE remained at approximately 75%, indicating excellent stability and durability. This study provides new insights into the development of highly efficient and durable ECO₂R catalysts.

In addition to the commonly used copper oxide-based catalysts and copper alloy catalysts, other copper-based catalysts (such as copper single atoms, copper nanoclusters, copper intermetallic compounds, etc.) have also made significant progress in electrocatalytic CO₂reduction reactions. By employing methods such as preparing single atoms, elemental doping, defect engineering, and surface reconstruction, these catalysts demonstrate potential advantages in catalytic efficiency and selectivity.

(1) Single-atom catalysts (SACs)

SACs, with their maximized atom utilization and uniform active sites, have emerged as an important strategy for enhancing methanol selectivity. Zhao et al.^[34]prepared a single-atom copper-immobilized MXene (Ti₃C₂Cl _x) catalyst by selectively etching a quaternary MAX phase, as illustrated in the structural diagram shown in Figure 2c. As seen in the right panel of Figure 2c, this catalyst achieves a Faraday efficiency (FE) for methanol of 59.1% at -1.4 V (vs RHE). The strong interaction between the single-atom copper sites and the MXene support material is considered key to its high efficiency, promoting the highly selective generation of methanol.

(2) Composite material catalyst

Designing composite catalysts aims to further enhance catalytic performance through synergistic interactions between materials. Hussain et al.^[57]reported the synthesis of a Cu-g-C₃N₄/MoS₂composite catalyst via a hydrothermal method. This composite catalyst exhibits superior electrocatalytic performance due to the synergistic effects among its components. At a potential of -1.4 V (vs Ag/AgCl), the current density reaches 78 mA/cm². Electrochemical impedance spectroscopy and chronoamperometry tests demonstrate that the composite material possesses better charge transfer capability and greater stability compared to individual Cu-g-C₃N₄and Cu-MoS₂, highlighting its advantages in long-term reactions. In addition, the ternary composite catalyst CuO-ZnO-MoS₂synthesized by them^[58]exhibits a higher current density (121 mA/cm²) at a lower onset potential. The introduction of MoS₂not only prevents the aggregation of CuO and ZnO but also enhances the overall catalytic performance through synergistic effects, demonstrating high Faraday efficiency in the production of methanol and ethanol. Banerjee et al.^[59]prepared a copper oxide/nitrogen-doped carbon (Cu _xO/NC) composite catalyst via a microwave-assisted synthesis method, achieving a Faraday efficiency as high as 95% (primarily for formic acid and methanol) at -0.55 V (vs RHE). The synergistic interaction between nitrogen-doped carbon and copper oxide significantly boosts the electrochemical activity of the catalyst, showcasing the potential of the Cu _xO/NC composite catalyst in CO₂reduction.

(3) Intermetallic compound catalyst

Bagchi et al^[60]designed a novel Cu-Ga intermetallic compound catalyst (CuGa₂and Cu₉Ga₄). At a low potential of -0.3 V (vs RHE), the FE of methanol over the CuGa₂catalyst can reach 77.26%. This high selectivity is attributed to the unique two-dimensional structure of CuGa₂, where surface and subsurface oxide species (Ga₂O₃) are retained under reducing conditions, thereby promoting the selective formation of methanol.

(4) Surface-reconstructed catalyst

Lin et al.^[61]developed a nanoporous Cu_{2- x}Se-derived np-Cu (Se-5%) catalyst, which enhances catalytic performance through a surface reconstruction mechanism. In-situ and quasi-in-situ spectroscopic analyses indicate that the Cu (Se-5%) layer serves as the active phase, achieving a FE of 58% for methanol at -0.5 V (vs RHE). Furthermore, density functional theory calculations demonstrate that the reconstructed Cu (Se-5%) facilitates charge transfer and COOH adsorption, reducing the energy barrier for the CO₂electroreduction intermediate CHO and promoting the highly selective generation of methanol.

(5) Defect-engineered catalyst

Biswas et al^[62]reported a defect-engineered copper nanocluster (Cu NCs) catalyst. By removing Cu atoms at the vertices of Cu NCs to introduce defects, the geometric structure and edge modifications of the catalyst were altered, effectively improving the product selectivity of electrochemical reactions. At -0.7 V (vs RHE), the FE of Cu NCs for methanol reached 54%.

(6) Electron Delocalization Catalyst

Kong et al.^[63]proposed a novel Cu₂NCN catalyst, which achieves selective control over the CO₂reduction pathway by regulating the electron delocalization state. The surface interactions of the catalyst are illustrated in Figure 2d. In this catalyst, Cu(I) ions strongly coordinate with NCN₂ ^-, resulting in a highly delocalized electronic state. This leads to a methanol selectivity of up to 70%, a methanol production rate of 0.160 μmol·s^-1·cm^-2, and an outstanding methanol partial current density (92.3 mA·cm^-2)).

Overall, copper-based catalysts have emerged as a core material system for electrocatalytic CO₂reduction to methanol, owing to their excellent CO₂activation capability, moderate intermediate adsorption properties, diverse structural modulation approaches, abundant resource availability, and strong resistance to deactivation. Among these, copper oxide-based catalysts have demonstrated remarkable effectiveness in optimizing CO₂adsorption and enhancing methanol production efficiency. Copper alloy catalysts improve the Faraday efficiency of methanol generation by tuning alloy composition and surface structure, particularly excelling in long-term stability. Meanwhile, other copper-based catalysts further enhance catalytic efficiency and selectivity through defect engineering and composite material design. Therefore, each type of copper-based catalyst has its unique advantages and challenges. The key to research lies in optimizing the electronic structure, stability, and reaction pathways of the catalysts to achieve efficient and stable methanol synthesis.

Precious metal catalysts exhibit significant advantages in the field of CO₂electroreduction due to their high catalytic activity. The structure and interfaces of nanoalloys play a crucial role in the catalytic performance of CO₂reduction reactions. For instance, Payra et al.^[64]reported a C-supported PtZn nanoalloy catalyst, where different structures of PtZn/C, Pt₃Zn/C, and Pt _xZn/C series catalysts were constructed by adjusting the ratio of Pt to Zn, achieving highly efficient catalytic reduction of CO₂to methanol. DFT calculations indicated that compared to single-phase alloys, mixed-phase Pt _xZn catalysts with cubic and tetragonal structures facilitate single-electron transfer to adsorbed CO₂and exhibit better binding energies for carbon dioxide intermediates. At -0.90 V, the Pt _xZn/C catalyst achieved a methanol Faraday efficiency of 81.4%. Furthermore, by tuning the electronic structure between different catalyst components and providing additional active sites, the formation of key reaction intermediates can be promoted, thereby enhancing methanol yield. Zhu et al.^[65]developed a composite catalyst consisting of MnO₂nanosheets and Pd nanoparticles (Pd NPs/MnO₂ NSs). The interaction between Pd nanoparticles and MnO₂nanosheets at the interface effectively modulated the electronic structure of MnO₂, introduced additional active sites, and reduced the activation energy of CO₂, promoting the formation of *CO intermediates. As shown in Figure 3a, this catalyst achieved a methanol Faraday efficiency as high as 77.6% and exhibited a partial current density of 250.8 mA/cm²during electrolysis. The study also highlighted that the morphology of the catalyst significantly influences the selectivity of CO₂reduction products. Using nickel foam or copper foam as substrates helps form well-ordered MnO₂nanosheets, which in turn promotes uniform deposition of Pd nanoparticles, improving methanol selectivity and catalytic activity.

Heterostructured catalysts enhance catalytic performance by optimizing the electronic distribution on the catalyst surface and the synergistic effects between components. Wang et al.^[66]developed a novel ternary heterostructured catalyst CoO/CN/Ni. Figure 3billustrates the structural diagrams and reaction mechanisms of the two catalysts. The main product over the CoO/CN catalyst is formic acid, while methanol is the primary product over the CoO/CN/Ni catalyst. The introduction of nickel alters the surface electronic state of the cobalt species at the catalytic center by donating electrons to the CN layer, further enhancing catalytic activity and suppressing hydrogen generation. Ultimately, the CoO/CN/Ni catalyst achieves a methanol Faraday efficiency of 70.7%. Huang et al.^[67]reported a novel composite heterogeneous catalyst C-Py-Sn-Zn, which successfully immobilizes 4-aminopyridine (Py) and bimetallic Sn-Zn onto pure carbon paper through electrochemical oxidation and electrodeposition. The synergistic effect between Sn and Zn, as well as the interaction between Sn-Zn and Py, plays a crucial role in the electrochemical reduction of CO₂. At a potential of -0.5 V (vs RHE), the C-Py-Sn-Zn catalyst exhibits a methanol Faraday efficiency as high as 59.9%, maintaining stable catalytic activity over an extended electrolysis period of up to 26 hours.

Carbon materials, as supports, offer high conductivity and dispersibility. Surface modification of these materials can further enhance the adsorption of active sites and intermediates, thereby improving catalytic performance. To reduce particle aggregation and increase the number of active sites, the team led by Gong Jinlong at Tianjin University^[68]loaded Mo₂C particles onto nitrogen-doped carbon nanotubes, constructing a Mo₂C/N-CNT catalyst (Figure 3c). This structure leverages the strong metal-support interaction between Mo and N sites, effectively achieving spatial dispersion of particles and thus enhancing catalytic activity. Moreover, electron hybridization between the d-orbitals of molybdenum and the s-orbitals of carbon strengthens the catalyst's surface adsorption capacity for oxygen species, promoting the conversion of oxygen-containing intermediates along the methanol formation pathway rather than participating in the hydrogen evolution reaction. In other words, by optimizing the electronic structure of the catalyst surface, the adsorption energy and configuration of key intermediates are regulated, thereby suppressing HER and improving methanol selectivity. In a high-pressure continuous CO₂reduction system, the Mo₂C/N-CNT catalyst achieved a methanol Faraday efficiency of 80.4% at -1.1 V (vs SHE). Chongdar et al.^[69]reported a nickel-based hollow zero-dimensional (0D) carbon superstructure catalyst for low-potential CO₂reduction to methanol. As shown in Figure 3d, this catalyst was successfully synthesized by pyrolyzing Ni-MOF precursors, resulting in 0D porous hollow carbon superstructures Pyr-CP-600 and Pyr-CP-800 anchored with Ni nanoparticles. This unique structure provides the material with a high specific surface area and surface roughness, effectively enhancing its catalytic performance for the selective electroreduction of CO₂to CH₃OH. In a 1.0 mol/L KOH solution, Pyr-CP-800 achieved a methanol Faraday efficiency of 32.46% at -0.60 V (vs RHE), the highest among currently known nickel-based electrocatalysts.

Self-assembled monolayer-modified electrodes exhibit numerous advantages in electrocatalytic carbon dioxide reduction reactions due to their molecular-level precision control, optimized electronic structure, and interfacial regulation effects. Akter et al.^[70]reported a metal and metal oxide electrode modified by a self-assembled monolayer. The study found that the thiol self-assembled monolayer-modified ZnO electrode (C₃SH-ZnO) significantly enhanced the selectivity of the catalyst through synergistic effects between surface sites. In a tandem flow electrolyzer, the C₃SH-ZnO electrode achieved a methanol Faraday efficiency as high as 92%.

In summary, non-copper-based catalysts have demonstrated significant catalytic potential, exhibiting high activity and stability during electrolysis by adjusting the catalyst's structure or surface properties.

Different crystal facets expose different atomic arrangements, leading to variations in intermediate adsorption behavior and reaction pathways. For example, the Cu(100) facet favors the formation of C—C bonds and is more inclined to produce C₂products^[76],while the Cu(111) facet tends to favor the production of C₁products^[40].Therefore, by constructing specific crystal morphologies (such as nanocubes, octahedra, etc.), it is possible to achieve controllable regulation of crystal facet exposure.

The particle size of a catalyst significantly influences its catalytic activity. Smaller-sized catalysts expose more low-coordinated metal atom sites, resulting in stronger adsorption capacity for *CO intermediates; however, they may also promote the formation of by-products such as hydrogen and carbon monoxide, and excessively small sizes could compromise catalyst stability. Conversely, excessively large particles reduce the number of edge-active sites, thereby diminishing reaction activity. Therefore, appropriately optimizing catalyst size is crucial for the efficient production of target products.

The morphology of the catalyst also significantly influences its performance. By constructing high-surface-area, porous structures such as nanowires, nanosheets, or foam-like architectures, not only is the diffusion of CO₂ in the electrolyte and the release of products facilitated, but local concentration gradients can also be effectively reduced, thereby enhancing reaction efficiency at high current densities. Meanwhile, these structures may expose specific crystal facets and modulate the local reaction environment, positively impacting product selectivity.

The introduction of defects can effectively regulate the electronic structure of catalysts, thereby adjusting the adsorption strength of intermediates. An appropriate amount of defects helps stabilize intermediates, preventing excessive adsorption or premature desorption, and thus promoting the selective generation of target product pathways. Defect engineering can also modulate the local electronic environment, providing new means to enhance catalytic performance.

In addition, strategies such as alloying, elemental doping, and valence state engineering also demonstrate great potential in optimizing catalytic performance. In the process of catalyst design, achieving a balance between activity and product selectivity remains a critical challenge that urgently needs to be addressed. Although introducing highly active sites (such as low-coordinated atoms or defect structures) can enhance CO₂activation and increase the reaction rate, it often also strengthens the adsorption of reaction intermediates, thereby inhibiting subsequent hydrogenation or desorption steps and reducing the efficiency of target product formation. Therefore, achieving a dynamic balance between activity and selectivity is a key objective in catalyst design. By employing various approaches, including alloying, crystal facet engineering, strain modulation, and localized reaction environment adjustment, it is possible to precisely control the adsorption energy of intermediates and the reaction pathways, enhancing catalytic activity while ensuring high methanol selectivity, thus realizing an efficient and synergistic CO₂reduction catalytic process.

In summary, whether copper-based catalysts, non-copper-based catalysts, or phthalocyanine-based catalysts, certain breakthroughs have been achieved in improving methanol Faraday efficiency and catalytic stability. However, the current systems still face key challenges such as insufficient catalytic activity, complex reaction pathways, limited structural stability, and competition from side reactions. Future research can focus on multi-scale catalyst design, interface engineering optimization, precise regulation of reaction pathways, and machine learning-assisted screening, aiming to develop an efficient, low-energy-consumption, and stable CO₂reduction system for methanol production. The introduction of data-driven methods such as machine learning is expected to accelerate the rational design of high-performance catalysts, providing new insights for the resource utilization of CO₂and promoting the field toward efficient and sustainable development.

In the study of electrocatalytic CO₂reduction to methanol, accurately predicting catalyst activity is crucial for developing highly efficient catalysts. In recent years, researchers have developed various machine learning models to identify key descriptors for CO₂conversion, such as binding energy, free energy, limiting potential, and electronegativity. Among these, binding energy is one of the most widely studied descriptors for predicting catalyst activity. Yohannes et al.^[81]obtained 10,300 sets of adsorption energy data through DFT calculations and used this data to train machine learning regression models, with the extreme gradient boosting regression model (XGBR) performing the best. By predicting the binding energies of key intermediates (*CHO, *OH, *CO, *H) on transition metal nitride catalysts, the screening process for catalysts has been accelerated. Furthermore, feature importance analysis indicates that the group number of transition metals significantly affects the *OH binding energy, thereby determining the stability of the catalyst. In this study, DFT primarily revealed the excellent catalytic performance of Co-, Cr-, and Ti-based transition metal nitrides, while machine learning played a supporting and accelerating role in the screening process.

In addition to binding energy, free energy is also a key parameter for predicting catalytic activity. For example, Song et al.^[82]used active learning combined with DFT calculations to achieve efficient screening of single-atom alloy catalysts. Through four rounds of active learning iterations, a high-precision machine learning model was successfully trained, with gradient boosting regression (GBR) and XGBR performing best. This model predicted the free energies of key surface intermediates (*CHO, *COH, *CO, *H) for 380 catalysts, significantly accelerating the catalyst screening process and ultimately identifying 8 high-performance catalysts. As shown in Figure 6a,the orange region represents the highly active area for CO₂reduction, where Ti@Cu(100), Au@Pt(100), and Ag@Pt(100) exhibit the optimal catalytic activity for producing methanol or methane.

Compared to binding energy and free energy predictions, the limiting potential can more intuitively reflect the catalyst's activity and has higher experimental comparability. Mou et al.^[83]used machine learning to predict the limiting potential, enabling rapid screening of novel SACs@MOF catalysts. This study employed physicochemical features obtained from DFT calculations as input, training a series of GBR models to accurately predict the limiting potential of C₁products (Figure 6b). Different background colors indicate catalyst grouping based on their preferred C₁products, with yellow, blue, and pink corresponding to formic acid, carbon monoxide, methane, or methanol, respectively. The model identified W@UiO-66-H as the most favorable catalyst for methanol production. This model can be extrapolated to 48 new systems with X = CH₃, OH, and NO₂, and the reliability of the predicted results was verified through DFT calculations.

Predicting the limiting potential provides an efficient strategy for catalyst screening; however, constructing a universal descriptor to reveal the structure-activity relationship in complex electrocatalytic systems remains a key breakthrough for enhancing the predictive capability of catalyst activity. Ren et al.^[84]used the GBR algorithm to assess feature importance and developed a general-purpose descriptor φ, which is composed of intrinsic atomic properties such as electronegativity, electron type, and quantity—properties that can be directly obtained in the laboratory—and is used to predict the catalytic activity of diatomic catalysts (DACs@2D). As shown in Figure 7a, c, the higher the onset potential (U _HCOOH ^onset), i.e., the closer it is to 0, the better the catalytic activity typically is. Figures 7b and dillustrate the relationship between the onset potential and φ, with the volcano peak corresponding to the region of optimal catalytic activity. This method not only accelerates the discovery of high-performance catalysts but can also be extended to other electrocatalytic reaction fields such as nitrogen reduction and oxygen reduction.

Catalyst activity prediction provides important guidance for the screening of efficient electrocatalytic CO₂reduction catalysts for methanol production. However, to truly achieve the design and optimization of high-performance catalysts, further exploration and innovation are still needed. Machine learning can reveal the deep-level relationships between catalyst performance and its structural and electronic properties, offering a more efficient and precise pathway for catalyst optimization. In this context, Zhang et al.^[85]conducted feature importance analysis on graphene-supported binary catalysts using machine learning (random forest regression), identifying electronegativity, atomic number, and the number of d-electrons as the three most critical factors influencing catalyst stability. These three factors aid in catalyst design and optimization, accelerating the optimization process. By analyzing the adsorption energies of key intermediates, exploring linear scaling relationships, and calculating the minimum potentials for different catalysts, catalyst screening was performed. The results indicated that CrRh and MnIr are most favorable for the efficient generation of CH₃OH.

For multicomponent catalyst systems, Roy et al.^[86]employed a high-throughput screening approach assisted by machine learning (Gaussian process regression, support vector regression, kernel ridge regression, random forest regression), combined with DFT calculations, to systematically screen alloy catalysts composed of Cu, Co, Ni, Zn, and Mg. Figure 8aclearly illustrates the workflow for screening active and selective catalysts, where the best-fit algorithm trained on the dataset was used to predict the adsorption energies of *H, *O, *CO, *HCO, *H₂CO, and *H₃CO intermediates. Ultimately, seven highly active catalysts were identified, among which the CuCoNiZn quaternary alloy exhibited excellent methanol selectivity. Similarly, as shown in Figure 8b,they predicted the adsorption energies of various high-entropy alloys composed of Cu, Co, Ni, Zn, and Sn in another study^[87]. From 36,750 catalysts, 35 candidates with high activity and selectivity for methanol synthesis were screened out, providing important references for catalyst design.

In addition, Rittiruam et al.^[88]used machine learning to predict adsorption energy (linear regression, Gaussian process regression, support vector regression, kernel ridge regression, gradient boosting regression, and extreme gradient boosting regression), efficiently screening out Cu-Mn-Ni-Zn high-entropy alloy catalysts superior to Cu(111) from 11,920 data sets. Combined with Gibbs free energy calculations, machine learning not only accelerated the screening process (reducing DFT calculation time by approximately 635,880 hours) but also revealed that Mn is the optimal active site, enabling rational design and precise control of the catalyst. The above research indicates that gradient boosting regression is the most widely used machine learning model for electrochemical CO₂reduction to methanol. Machine learning has expanded beyond simple activity prediction to include rational catalyst design, providing a powerful tool for the development of highly efficient catalytic materials.

In terms of catalyst activity prediction, machine learning has achieved efficient prediction of intermediate adsorption energies, free energies, and limiting potentials by training models based on experimental data or DFT calculations. In particular, the development of data-driven universal descriptors has significantly improved the efficiency and accuracy of catalyst screening. Moreover, by integrating active learning strategies, machine learning can still efficiently optimize models under limited data conditions, accelerating the discovery of novel catalysts. Regarding catalyst design, machine learning further combines generative models to accomplish tasks such as catalyst composition design and optimization, providing design principles across different material categories for various catalytic systems. Therefore, the deep integration of machine learning with high-throughput computing will further promote the intelligent development of CO₂reduction catalysts, enabling efficient and cost-effective optimization of catalytic systems.