Xiaowen
Sun
a,
Rafael B.
Araujo
b,
Egon Campos
dos Santos
c,
Yuanhua
Sang
a,
Hong
Liu
*ad and
Xiaowen
Yu
*a
aState Key Laboratory of Crystal Materials, Shandong University, Jinan 250100, China. E-mail: hongliu@sdu.edu.cn; xiaowen.yu@sdu.edu.cn
bDepartment of Materials Science and Engineering, The Ångstrom Laboratory, Uppsala University, SE-751 03 Uppsala, Sweden
cDepartamento de Física dos Materials e Mecânica, Instituto de Física, Universidade de SãoPaulo, 05508-090, São Paulo, Brazil
dJinan Institute of Quantum Technology, Jinan Branch, Hefei National Laboratory, Jinan, 250101, China
First published on 19th June 2024
Descriptors play a crucial role in electrocatalysis as they can provide valuable insights into the electrochemical performance of energy conversion and storage processes. They allow for the understanding of different catalytic activities and enable the prediction of better catalysts without relying on the time-consuming trial-and-error approaches. Hence, this comprehensive review focuses on highlighting the significant advancements in commonly used descriptors for critical electrocatalytic reactions. First, the fundamental reaction processes and key intermediates involved in several electrocatalytic reactions are summarized. Subsequently, three types of descriptors are classified and introduced based on different reactions and catalysts. These include d-band center descriptors, readily accessible intrinsic property descriptors, and spin-related descriptors, all of which contribute to a profound understanding of catalytic behavior. Furthermore, multi-type descriptors that collectively determine the catalytic performance are also summarized. Finally, we discuss the future of descriptors, envisioning their potential to integrate multiple factors, broaden application scopes, and synergize with artificial intelligence for more efficient catalyst design and discovery.
Understanding the interactions between catalysts and reaction intermediates is pivotal for improving the catalytic activity of certain catalysts. These interactions occur at specific sites on the catalysts’ surface, known as active sites, whose concept was first introduced by Taylor in 1925,28 laying the foundation for subsequent research on the electronic structure of catalysts and the binding between catalysts and reaction intermediates. The Sabatier principle, as proposed by Nørskov,29 states that for optimal catalytic performance, the adsorption of reaction intermediates on the catalyst surface should neither be too strong nor too weak. Instead, a moderate level of adsorption is preferred. Accordingly, a volcano plot visualizes the relationship between catalytic activity and adsorption energy of reaction intermediates for different catalysts.30–32 Catalysts located near the peak of the volcano diagram have an optimal balance of adsorption energy and thereby the highest catalytic activity. On the other hand, catalysts with weak adsorption energy have lower catalytic activity due to insufficient interaction with the reaction intermediates. Similarly, catalysts with excessively strong adsorption energy may hinder the release of the final products, leading to reduced catalytic efficiency. This suggests that modifying the adsorption strength of key intermediates can optimize the catalytic activity of materials. Various approaches, such as doping heteroatoms, introducing defects, forming alloys, employing strain/stress, and designing heterojunctions on catalysts, have been employed to alter the electronic structure of active sites. By doing so, the adsorption strength of catalyst surfaces and intermediates can be modulated, leading to significant changes in catalytic performance.33–37 Especially, if the readily accessible intrinsic properties or electronic structure features of a catalyst material can be used as descriptors to correlate with the adsorption energy of key intermediates in reactions or its catalytic activity in a linear or volcano plot, it allows for a physical or chemical mechanistic explanation of the origin of the catalysts’ activity. This understanding can then be utilized to efficiently screen and design other catalysts with desired properties.38–40
The design of catalysts based on descriptors offers a more efficient and targeted approach to catalyst development. By utilizing descriptors, researchers can significantly reduce the need for extensive theoretical calculations and experimental tests, and thereby save valuable time, resources, and efforts in the search for highly active catalysts.27,41–46 One of the simplest descriptors used in catalysts is based on the adsorption energy of key intermediates. For instance, the adsorption Gibbs free energy of H atoms (ΔG*H) is commonly employed to directly describe the catalytic activity of the considered catalysts for HER.47–49 Additionally, the adsorption energy of O atoms (ΔE*O) usually exhibits a linear relationship with the adsorption energies of OH (ΔE*OH) and OOH (ΔE*OOH), which also provides valuable information for evaluating catalytic activities in OER and ORR.50–52 However, for multi-electron complex processes such as NRR and CO2RR, multiple reaction intermediates are involved, making it challenging to evaluate catalytic activity using a single descriptor based on a particular intermediate. In 1987, Nørskov proposed a fitting of the adsorption energy of nitrogen atoms (ΔE*N) to the limiting potential (UL) of NRR on metal surfaces, successfully replacing the adsorption energies of various N-related intermediates (e.g., NxHy) in NRR by ΔE*N, resulting in a volcano diagram relationship between ΔE*N and UL, which can effectively illustrate how ΔE*N can serve as a descriptor for evaluating the catalytic activity of transition metals (TMs) for NRR.53 In addition, some descriptors have been utilized to predict the trend of electrocatalytic activity, including the widely used metal d-band center, O-2p band center for oxides, eg filling, electron transfer, number of active sites, electronegativity, bond lengths, etc.54–60 In the past decade, there has been increasing interest in investigating the catalytic activity of spin-regulated systems. It has been observed that changing the spin state of the active site can influence the binding interactions of the active centers with intermediates, thereby modulating the catalytic activity.61,62
Although there have been several reviews on descriptors in recent years, many of them focused on specific types of descriptors or descriptors for particular reactions,63–65 and there are relatively fewer comprehensive summaries available that encompass descriptors related to readily accessible intrinsic properties, spin-related characteristics, the application of comprehensive d-band centers, and the multi-type descriptors. In this review, we aim to provide a systematic and comprehensive summary of the relevance of physical and chemical features mapping key reaction intermediates and their applications as descriptors to quantify the catalytic activity of reactions (see overview in Fig. 1). First, we describe the critical electrochemical reaction processes such as HER, OER, ORR, NRR, and CO2RR under different conditions and identify the available key reaction intermediates that can be correlated to quantify the catalytic activity. Then, by combining the results of density functional theory (DFT) calculations and experimental data, we summarize the descriptors for the above-mentioned electrochemical reactions. The descriptors are categorized into d-band center descriptors, readily accessible intrinsic properties of the catalysts that can be obtained directly from databases, spin-related property descriptors, and multi-type descriptors. Finally, by summarizing the current state of descriptor-related research, we provide a comprehensive perspective on this field and highlight potential directions for future descriptor studies.
![]() | ||
Fig. 2 (a) Mechanism diagram of HER on the surface of an electrode. Reproduced with permission. Copyright 2020, American Chemical Society.73 (b) HER volcano curve of metal carbides (red) compared to the previously reported O-terminated MXenes (black) and Pt (blue) in acidic media. Reproduced with permission. Copyright 2020, Wiley-VCH.76 (c) HER volcano curve of transition metal carbides in alkaline media. Reproduced with permission. Copyright 2019, American Chemical Society.78 |
The strength of the interaction between the H atom and the catalyst surface, as determined by ΔG*H, plays a crucial role in determining the catalytic activity of catalysts for HER. According to the Sabatier principle,29 an optimal catalyst for HER should exhibit a moderate ΔG*H. If the interaction between *H and the catalyst surface is too strong, the active sites will be poisoned, thus hindering the desorption of molecular H2 from the catalyst surface. On the other hand, if the *H adsorption is too weak, the reaction will require higher η, meaning that an additional energy input is needed to drive the reaction forward, which will result in an increase in energy consumption and a decrease in overall efficiency. Therefore, an ideal HER catalyst should strike a balance between *H uptake and release, and thereby a moderate adsorption strength for *H. Experimental studies have identified Pt and Pt-based metals as the most active catalysts for the HER. Theoretical calculations have revealed that these catalysts exhibit moderate H adsorption, indicated by a moderate ΔG*H value. This suggests that the ΔG*H parameter can effectively characterize the catalytic activity of catalysts. For example, in an acidic electrolyte, ΔG*H is considered as the simplest and most intuitive descriptor for HER; its absolute value |ΔG*H| can serve as an indicator of the catalytic activity of the catalyst. Generally, a smaller value of |ΔG*H| indicates better catalytic performance of the catalyst, as shown in Fig. 2b.76 In an alkaline electrolyte, although the binding energy of OH− and the kinetics barrier for water dissociation have significant impacts on the catalytic activity of the catalysts for HER, ΔG*H still plays important roles, which can also exhibit a volcano-type diagram as a function of the logarithm of theoretical exchange current density i0 (lg(i0)) (Fig. 2c).78 Therefore, in both acidic and alkaline electrolytes, wherein the adsorbed *H is a critical intermediate for HER, ΔG*H associated with *H is a key factor in determining the catalytic activity of catalysts for HER. Furthermore, it is important to note that the kinetics barriers also play a significant role in HER. For instance, many metal catalysts exhibit H binding strength comparable to that of Pt, but display significantly lower activity in catalyzing HER. This discrepancy may arise from the distinction between the kinetically active H species and the inactive surface H species with binding energies as descriptors. For Pt and Au, the H atoms with G ∼ 0 (hollow site) are inert in the kinetic correlation region, whereas the top H atoms are active and show weaker H adsorption. For the HER on Au, the H atoms should transit through a less strongly bound top site before diffusing into the stable hollow site. This diffusion encounters a high potential barrier, which contributes to the observed low HER activity on Au. In contrast, for the HER on Pt, H atoms located at the top site with particularly low barriers facilitate higher reaction rates. Therefore, it is essential to recognize that the HER activity cannot be accurately assessed solely based on G ∼ 0.79 Thorough analysis of kinetically relevant H and inactive H species is crucial for a comprehensive understanding of HER catalysts.
![]() | ||
Fig. 3 (a) The mechanism representation of OER on the surface of a catalyst in acidic and alkaline electrolytes. (b) The volcano curve between ΔE*O and OER activity. Reproduced with permission. Copyright 2007, Elsevier.86 (c) The 3D volcano plot showing the “standard” scaling relationship (ΔG*OOH = ΔG*OH + 3.20) for ΔG*OOH and ΔG*OH in ORR for a wide variety of two-dimensional materials. Reproduced with permission. Copyright 2018, American Chemical Society.87 |
In the assessment of the OER catalytic activity, the ΔG of each reaction step serves as a critical parameter. There are two commonly utilized indexes to evaluate the OER activity of the catalyst: (i) UL is defined as UL = −ΔGmax/e, where ΔGmax represents the maximum increase in the Gibbs free energy of the reaction. (ii) ηOER is defined as the difference between UL and equilibrium potential (U0); this difference represents the additional potential required to drive the OER beyond the thermodynamically balanced potential, i.e., ηOER = UL–U0. For the reaction 2H2O → H2 + O2, U0 is known as 1.23 V.82 Therefore, ηOER can also be expressed as ηOER = UL − 1.23 V.85,88 A lower ηOER generally indicates better catalytic activity of the catalysts being considered. Currently, Ru and Ir oxides are considered as the best metal-based OER catalysts. For instance, through DFT calculations, Rossmeisl et al. identified that RuO2 and IrO2 are located at the apex of the ΔG*O volcano diagram for the OER, suggesting that these catalysts are the most effective catalysts for this reaction, and highlighting the significance of *O as a crucial reaction intermediate in this process (Fig. 3b).86 Additionally, the adsorbed *OH and *OOH species are also considered as key intermediates in OER. For example, in 2011, Nørskov et al. studied a large number of oxide surfaces for OER and found interesting results regarding the relationships among ΔG*OH, ΔG*O, and ΔG*OOH. They found a linear relationship between ΔG*OOH and ΔG*OH with a slope of approximately 1; this suggests that both *OOH and *OH species are bound to the catalyst site through a single bond. On the other hand, the slope for the linear relationship between ΔG*O and ΔG*OH is approximately 2, indicating that *O is bound to the catalyst site through a double bond.89–91 Additionally, they established a volcano curve to describe the OER catalytic activity of various oxide surfaces by correlating −ηOER as a function of ΔG*O − ΔG*OH.89
The electrochemical ORR occurring at the cathode is a determining reaction that limits the development of Zn–air batteries and proton exchange membrane fuel cells.92 It should be noted that ORR can be considered as the reverse reaction of OER, and they share the same intermediates (e.g., *O, *OH, *OOH). The ORR can proceed through two main pathways, depending on the number of electrons transferred during the overall reaction. One pathway involves the reduction of O2 to H2O by transferring four electrons. Another pathway involves the reduction of O2 to hydrogen peroxide (H2O2) by transferring two electrons.93–97 The overall reaction for the four-electron transfer during the ORR is O2 + 4H+ + 4e− → H2O in acidic media or O2 + 2H2O + 4e− → 4OH− in alkaline media. In the ORR process, the initial step involves the activation of O2. Generally, it is believed that the adsorption of molecular O2 onto a surface is a chemical process and that electron transfer is unlikely to limit the rate of this process. However, through the use of ab initio molecular dynamics (AIMD) coupled with a solvation model, Dudzinski et al. revealed that O2 adsorption on Au(111) during the ORR is significantly influenced by the applied (absolute) potential due to local electric field effects. This observed dependency implies that O2 adsorption should be viewed as an electrochemical step rather than purely a chemical one, underscoring the importance of O2 adsorption in the overall process.98 Additionally, the strength of *OH adsorption on the catalyst surface, represented by ΔG*OH, is recognized to play a crucial role in determining the ORR kinetics. There is a commonly accepted standard scaling relationship between ΔG*OOH and ΔG*OH: ΔG*OOH = ΔG*OH + 3.20 eV. This scaling relationship implies that the energy required for adsorption of *OOH is approximately 3.20 eV higher than that of *OH (Fig. 3c).87 From the analysis mentioned above, it is apparent that the ORR activity of the catalyst is limited by this scaling relationship. In theory, an ideal ORR catalyst would operate at the equilibrium potential of 1.23 V, where all the reaction free energies would be zero.
Electrochemical NRR has two possible mechanisms: the associative mechanism and the dissociative mechanism (Fig. 4a).99,100 The N2 activation process includes the end-on and side-on configuration, and to simplify, the subsequent hydrogenation of the side-on configuration is not displayed. The associative mechanism involves the reduction of the adsorbed N2 molecule through generating NxHy intermediates, followed by the breakage of the NN bond and further reduction to form NH3 eventually. The dissociative mechanism involves the direct breaking of the N
N triple bond in N2, resulting in two separated N atoms which then go through individual reductions to form NH3. In general, breaking the N
N triple bond is challenging. As a result, the associative mechanism is typically observed in NRR.101 For most of the catalysts, the conversion of *N2 to *NNH, *NH to *NH2 or *NH2 to *NH3 is identified to be the potential-determining step (PDS) of NRR in the associative mechanism.102,103 Considering all the possible elementary steps in the NRR, the Gibbs free energy change of key intermediates (such as ΔG*N2, ΔG*NNH, ΔG*NH2 and ΔG*NH3, or ΔG*N2
to
*NNH and ΔG*N2
to
*NH3) is an important factor in determining the NRR performance. In NRR, the atom that interacts with the catalyst surface is always N. Therefore, the interaction strength between the catalyst and the N atom can provide valuable information regarding the efficiency of NRR or NORR; thus the adsorption Gibbs free energy of N (ΔG*N) can be utilized as a descriptor to assess the catalytic activity of the catalysts (Fig. 4b and c).53
![]() | ||
Fig. 4 (a) Schematic depiction of three mechanisms for NRR. Reproduced with permission. Copyright 2020, Wiley-VCH.99 (b) UL volcano curves of HER and NRR on metal (111) surfaces. (c) UL volcano curves of HER and NRR on metal (211) surfaces. Reproduced with permission. Copyright 2015, Chemistry Europe.53 |
The research on NORR has received significant attention recently, because the NO molecule has relatively low activation energy (204 kJ mol−1) compared to N2 (941 kJ mol−1) and thus would be more easily activated than N2. By performing NORR, NO removal and NH3 synthesis can be achieved simultaneously, making the process attractive for environmental applications.69,104,105 Among pure metal surfaces, Cu is considered as one of the best catalysts for NORR since it exhibits both high activity and selectivity.106 The reaction intermediates and mechanisms observed in NORR are similar to those in NRR. In NORR, there are several possible steps that can serve as the PDS, including *NO + H+ + e− →*NOH or *NH2 + H+ + e− → *NH3. Additionally, the production of H2O step (*OH + e− + H+ → H2*O) can also act as the PDS in NORR.107 The key intermediates involved in these steps can have a great influence on the catalytic efficiency. Recent research conducted by Wu et al. has demonstrated that graphene-based single atom catalysts (SACs) are promising electrocatalysts for NORR. The adsorption of NO on such catalysts provided valuable insight into the catalytic activity of NORR. In particular, catalysts with a moderate ΔG*NO between 0 and −1 eV have been found to be effective in promoting protonation and preventing poisoning.108 Catalysts with metal-doped α-borophene monolayers (M@BM-α) have been found to exhibit a volcano-shape relationship when correlating UL with ΔG*N for NORR. This suggests that there is an optimal range of ΔG*N that leads to the best performance of the catalysts for such reaction.107 According to the Sabatier principle, the optimal catalytic activity is achieved when there is neither too strong nor too weak binding between the key intermediates and the catalysts; the balanced interaction ensures that the intermediates undergo the necessary chemical transformations without too strong adsorption or desorption. Furthermore, the physical or chemical mechanisms involved in the interaction between the key intermediates and the catalysts play decisive roles in determining the overall reaction efficiency.
![]() | ||
Fig. 5 (a) Mechanism illustration for the CO2RR process. Reproduced with permission. Copyright 2022, The Royal Society of Chemistry.116 (b) Free energy diagram at 0 V vs. RHE for the energetics of the dimerization of *CO to form O*C*CO. Reproduced with permission. Copyright 2018, Nature.121 (c) The volcano curve between UL and ΔG*CO. Reproduced with permission. Copyright 2023, American Chemical Society.122 (d) *C*CH as a key intermediate in the production of ethylene and *CHCHOH and CH3CH*O as key intermediates in the production of ethanol. |
Effective CO2 adsorption is typically considered as a crucial step in the CO2RR, although its exact nature is subject to debate. One perspective suggests that CO2 adsorption leads to the formation of CO2− species carrying a unit charge on the adsorbed surface. Alternatively, it has been proposed that *CO2 adsorption is driven by the interaction between the dipole of *CO2 and the interfacial electric field, without the existence of *CO2− species. Through an analysis of adsorbate-induced density of states widths, Vijay et al. found that electron transfer to CO2 was facilitated on metal and metal nitrogen-doped carbon catalysts (MNCs), indicating the important role played by the dipole of *CO2. In addition, their research further revealed that the *CO2 formation tends to be the rate-limiting step for TMs, while MNCs may encounter limitations related to *CO2 adsorption or *COOH formation.123 Therefore, it is important to optimize *CO2 dipoles in the design of CO2RR catalysts. Different products observed in CO2RR can be attributed to the variations in the key intermediates for different catalysts. For instance, late TMs (such as Au, Ag, Zn) tend to bind *COOH more strongly than *CO, thus leading to the production of CO as the main product. Sn, Pb, and Hg tend to interact with O*CHO rather than with the *COOH intermediate, resulting in the HCOOH as the primary product.124,125 In most instances, the adsorption strength of *CO plays a decisive role in determining the activity and selectivity of CO2RR, because too strong adsorption of *CO will lead to poisoning of the active sites, while too weak adsorption will cause the desorption of *CO and thus result in lower yield of C2 products. To achieve optimal selectivity, it is desirable to have a moderate *CO adsorption, which can promote C–C coupling and favor the formation of valuable C2 products (Fig. 5b).121 This can be achieved by tuning the geometric or electronic properties of the catalysts. For example, doping lanthanide (Ln) elements into Cu2O can change the lattice constant of the catalyst, which subsequently affected the adsorption strength of *CO. This modification reduced ΔG*CO–*COCO and finally promoted the catalytic activity of the catalyst (Fig. 5c).122 In addition to the study of the adsorption strength of CO, the effect of CO coverage on C–C coupling is a topic of ongoing debate. Many works argue that high CO coverage promotes C–C coupling, leading to the production of more multi-carbon products. Experimental strategies often involve increasing the concentration of CO or enhancing the adsorption strength of CO through improved catalysts. However, DFT calculations on the Cu(111) and Cu(100) surfaces indicate that high CO coverage leads to significantly weaker CO adsorption, making it unfavorable for the generation of C2 products.126 C–C coupling is a prerequisite step for the generation of C2+ products in the CO2RR process. It is typically categorized into three main pathways: *CO direct dimerization, *CO and *CHO coupling, and *CHO direct dimerization. In all these pathways, the specific type of *CO hydride formed after C–C coupling plays a crucial role in determining the resulting products.127 For example, *C*CH serves as a key intermediate in ethylene production, while *CHCHOH and CH3CH*O are crucial for ethanol production (Fig. 5d). In addition, the protonation of the intermediate O*C*CH is an important step that influences the formation of acetic acid and ethylene. By precisely controlling site-selective protonation through adjusting the H affinity, it is possible to stabilize O*C*CH, protect the CO-bonded intermediates from H attack, and selectively generate the acetate product.128
In summary, intermediates play critical roles in chemical reactions and their Gibbs free energies can serve as fundamental descriptors to correlate catalytic activity. Nevertheless, when considering the same reaction on different catalysts, the key intermediates may be different. Moreover, the key intermediates corresponding to different products may also differ significantly for the same reactants. Therefore, thorough consideration of all factors and meticulous investigation of different reaction processes are essential in the study of key intermediates in electrocatalytic reactions.
![]() | ||
Fig. 6 (a) Schematic illustration of the interaction between two electronic states. Reproduced with permission. Copyright 1995, Springer Nature.133 (b) Electrochemically measured potential (ESCE) for Pd overlayers on a number of metals as a function of the calculated shift of the d-band center (δεd). Reproduced with permission. Copyright 2011, PNAS.134 (c) Variations in the semielliptical width of the density of states distribution as a function of the d-band center for the 4d transition metals. Reproduced with permission. Copyright 2013, Springer Link.135 (d) Adsorption energies of C, N, and O on the (111) surface (atop site) of Pd and PdX alloys as a function of the εd of surface Pd atoms. (e) Adsorption energies of C, N, and O on Pd and Pd alloys with late TMs at 3![]() ![]() ![]() ![]() ![]() ![]() |
The d-band model was first proposed by Nørskov et al. and is particularly beneficial for understanding the trends in bond formation and reactivity of TMs.133 In 2010, Nørskov et al. proved that the experimental results can be well described by a theoretical volcano plot based on the d-band model (Fig. 6b).134 However, when describing the interaction between absorbates and metal atoms, only considering the location of the d-band center is not sufficient; it is important to take into account the width and shape of the d-band (Fig. 6c).135 Hence, based on DFT and the Newns–Anderson model, Nørskov et al. extended the d-band model and showed that bond strength can be accurately defined by the location of the upper edge of the d state. The TMs Ni, Pd, and Pt were identified to be particularly affected by the d-band shape, and the generalized electronic structure descriptor εd + Wd/2 was introduced to account for these effects (Fig. 6d).136 For some TM alloys, the d-band center, even after taking into account the corrections for the d-band width and s/p electrons, may not fully capture the reactivity change from one surface to another. Thus, by studying the high moments of the d-band, the upper d-band edge εu, defined as Hilbert's highest peak position of the Hilbert transform of the density of states projected onto the d orbitals of the active metal sites, was determined for the electron surface activity descriptor (Fig. 6e).136 The d-band center theory has highly favored and extensively applied in the field of electrocatalysis to explain the interaction between active sites and adsorbates. It has proven to be a valuable framework for understanding and designing electrocatalysts.
Typically, the determination of d-band centers is carried out through DFT calculations; however, in practical applications, specific absolute values are not necessarily required to assess the adsorption strength of intermediates. The shift of the valence band of an element relative to the Ef level following a catalyst change can be ascertained through techniques such as X-ray absorption near-edge spectroscopy (XANES) and ultraviolet photoemission spectroscopy (UPS), where the shift of the valence band corresponds to the shift of the d-band center, providing insights into the adsorption strength of the intermediates.137,138 Nevertheless, practical challenges may arise from the discrepancies between theoretical calculations and experimental findings. Generally, DFT calculations often simplify catalysts and ignore the external influences, which may lead to deviations in experimental results. In this case, one can consider corrections to the DFT model to account for solvation effects, etc. Alternatively, one can speculate about changes in the d-band center based on the difference in electronegativity between the target atom and the surrounding atoms.
![]() | ||
Fig. 7 (a) The optimized structure of Pt1/TMDs. (b) The linear scaling relation between |ΔG*H| and df of Pt1/TMDs. Reproduced with permission. Copyright 2022, Wiley-VCH.141 (c) The structure of Co/Fe/Cu-doped Ni5P4. (d) Schematic illustrations of bonding formation between the catalyst surface and the adsorbate. (e) ΔG*H values of doped Ni5P4 electrocatalysts. Reproduced with permission. Copyright 2022, Wiley-VCH.142 |
Non-noble metal Ni is generally considered inert for HER due to its too strong adsorption for *H. However, by synthesizing the Ni–Cu alloy and changing the surface composition ratio of Ni to Cu to control the ligand effect, the position of the d-band center of the Ni–Cu alloy can be adjusted to be comparable to that of Pt (−1.929 eV with respect to EF), making the Ni–Cu alloy a potential alternative to Pt for HER.143 Another work on Ni–Cu alloys was conducted by Wang et al., which focused on the improved d-band center, especially considering the hollow adsorption sites for *H on Ni–M alloys. In their work, the conventional d-band center is not suitable for describing *H adsorbed on hollow sites, because *H interacts with three metal atoms, forming three M–H bonds. Therefore, they quantified the effect of these three M–H bonds, and established a linear relationship between the improved d-band center and ΔG*H. The improved d-band center is defined as εd = A εNi + (1 − A) εM, where A is the ratio between the numbers of Ni–H and M–H bond. In addition, the proposed d-band center descriptor is also applicable for trimetallic alloys (Ni–Cu–M).144
Recently, high entropy alloys (HEAs), comprised of five or more metal elements in roughly equal proportions, have received a lot of attention due to their high thermal stability, good resistance to high temperature oxidation, and adjustable electronic structure.145–151 Kwon et al. successfully synthesized nanoporous Ir-based HEAs (ZnNiCoIrX, X = Fe, Mn) through dealloying in the vacuum system. They found that when Mn is doped into HEAs, the electronic structure of the Ir site is modified, leading to the d-band center being far away from EF, and thus weakening the adsorption strength of the adsorbed intermediates and consequently improving the catalytic activity. Experimental results confirmed that the ZnNiCoIrMn HEA has lower η for HER and OER than its counterparts, suggesting that tuning the d-band center of HEAs through doping with different metals can effectively enhance the catalytic activity.152 Perovskites have been considered as a promising candidate for electrocatalysis, solar cells, etc. A recent report demonstrated that the LnBaCo2O5+δ perovskites exhibited excellent catalytic HER properties. Notably, among these perovskites, the one with the O p/Co d-band center that is located far from EF showed the best activity for HER. More importantly, the report highlighted the importance of designing perovskite-based catalysts with appropriate O p/Co d-band centers for enhancing the HER performance.153
In HER, transition metal carbides, nitrides, sulfides, phosphides, borides, etc. are considered as promising candidates to replace the precious metal Pt.154,155 It was reported that Co4N has high activity for OER, while its catalytic HER activity is poor. By introducing heteroatom doping, the d-band center of the Co atom can be tuned and therefore result in different HER catalytic activities. The V-atom-doped Co4N nanosheets displayed an overpotential of 37 mV at 10 mA cm−2 for HER, and the increased catalytic activity was attributed to the decreased d-band center.137 Similarly, Zhang et al. improved the HER performance of Fe3C in alkaline media by adjusting the d-band center of Fe atoms. They grew Cr-doped Fe3N structures on nitrogen-doped graphene and observed a lower η compared to the pristine Fe3C. DFT calculations illustrated that Cr doping reduced the d-band center of the Fe atom, thus optimizing the HER performance.156 Furthermore, introduction of holes in the CoP nanosheets can also optimize their catalytic performance for HER. DFT calculations revealed that the valence electron numbers of the d-band center and Co sites around the holes are shifted upward, resulting in a change of the ΔG*H of the Co atoms surrounding the holes.157 In 2019, Li et al. demonstrated that ruthenium diboride (RuB2) is the most competitive HER catalyst among diborides through theoretical calculations; it exhibited a catalytic activity comparable to that of Pt. The high HER activity of RuB2 was further experimentally verified in both acidic and alkaline environments. Furthermore, the study revealed a linear relationship between the catalytic activities of various metal diborides and their d-band centers.158 Another study investigated the catalytic performance of Ni5P4 nanosheets manipulated by heteroatom doping, especially with Fe and Co atoms (Fig. 7c). The doped Ni5P4 catalysts were found to exhibit different HER performance according to theoretical calculation results (Fig. 7d). Compared with Fe-doped, Cu-doped, and pristine Ni5P4, the Ni5P4 nanosheets with appropriate Co doping demonstrated the best electrocatalytic performance, with the lowest overpotential of 100.5 mV at 10 mA cm−2 for HER. The enhanced activity was attributed to its moderate position of the d-band center, which contributes to moderate anti-bonding filling (Fig. 7e) and thus promotes the adsorption and desorption processes of H atoms on the catalyst surface.142 Wang et al. designed and synthesized a series of lattice-strained homogeneous NiSxSe1−x nanosheets@nanorods hybrids for HER/ORR.159 Among various compositions, the NiS0.5Se0.5 sample with approximately 2.7% lattice strain exhibited the best catalytic performance for both HER and OER, with an extremely low overpotential of 70 mV for HER and 257 mV for OER at 10 mA cm−2. In addition, it also displayed a good long-term durability for 300 h at a high current density of 100 mA cm−2. Experiments and theoretical calculations revealed that the lattice strain reduced the d-orbital overlap of the metals, resulting in narrowing of the bandwidth of the d-band and bringing the d-band center closer to EF. Therefore, NiS0.5Se0.5 has the optimal *H adsorption kinetics for HER and the lowest energy barrier for OER. All the above results demonstrate that the adsorption strength of *H on metal active sites can be optimized by adjusting the d-band center. The crystalline–amorphous CoSe2/CoP heterojunction has also been demonstrated to be active for HER. This heterojunction showed a reduced HER kinetics barrier, which was attributed to the strong electronic coupling at the interface that lowered the d-band center of the Co atom, leading to the feasible adsorption of *H.160 The MoS2 basal plane is generally inert for HER catalysis; however, its electronic structure can be tuned by introducing dopants, defects, or strain, which can significantly enhance the HER catalytic activity. Yang et al. studied the defective MoS2/graphene catalysts and demonstrated that the d-band center of the metal is a good descriptor for evaluating the adsorption strength of *H under different strain conditions.161 Kwon et al. synthesized ternary alloy nanosheets MoSe2–VSe2–NbSe2 through a hot-injection colloidal reaction. They found that these ternary alloys exhibited higher electrocatalytic activity for HER compared to binary alloys when tested in an acidic electrolyte. They also revealed a volcano-type relationship between the HER performance and the d-band center of the alloys, where the d-band center was determined based on synchrotron radiation X-ray photoelectron spectroscopy valence band spectra.162
The electron transfer process during the OER can occur through two mechanisms: adsorbate evolution mechanism (AEM) and lattice oxygen oxidation mechanism (LOM) (Fig. 8a).65,170,171 In the AEM, the reaction occurs at the surface of the catalyst, where oxygen species are adsorbed on the metal surface and undergo a series of oxidation reactions. In the LOM, the lattice oxygen within the bulk of the catalyst participates in the electron transfer process.172,173 Balancing the adsorption and desorption of oxygen-containing intermediates is essential for optimizing the OER process following the AEM mechanism. This balance is crucial for maximizing the effectiveness of OER catalysts, necessitating the presence of the optimal d-band center in the electrocatalysts.174 In contrast, for reactions following the LOM mechanism, metal atoms with minimal electronegativity difference from oxygen are usually preferred. This characteristic enables the feasible intramolecular electron transfer from the oxygen ligand to the TM cation, leading to a negative metal d-band center relative to the O-2p band center (εTM–d–εO–2p). This provides a prerequisite for lattice oxygen redox chemistry to occur.175 Thus, the influence of the d-band center on the OER activity of catalysts under the LOM mechanism primarily manifests through alterations in the band distribution of TMs to influence the lattice oxygen activation.
![]() | ||
Fig. 8 (a) The AEM and LOM mechanisms for OER. Reproduced with permission. Copyright 2016, Nature.170 (b) The reduced gap between d- and p-band centers (ΔEd−p) and correspondingly increased OER/ORR performance on perovskite oxides of Sm0.5Sr0.5CoO3−δ. Reproduced with permission. Copyright 2020, American Chemical Society.176 (c) Schematic illustration of the effect of Mo doping and O vacancies on the d-band center of Co3O4. Reproduced with permission. Copyright 2022, Wiley-VCH.177 |
Therefore, oxygen-containing catalysts thus play an important role in facilitating the OER.178–183 There have been extensive studies on the oxide catalysts for OER. In the research on V-doped LaCoO3 (V-LCO), it was observed that the closer the d-band center to the EF of the Co atom after doping, the stronger the bond strength between the adsorbate and the catalyst and thereby the better OER activity.178 Luo et al. theoretically calculated and experimentally verified that Fe-doped Ni3Ge2O5(OH)4 is a superior OER catalyst among serpentine germanane materials. Doping Fe3+ allowed for fine-tuning of the d-band center of the Ni3Ge2O5(OH)4 serpentine, which optimized the binding strength of the intermediate with the surface Ni and Fe atoms, leading to an effective alteration in the OER catalytic activity.184 Some studies have suggested that the difference between the metal d-band center and the oxygen p-band center (ΔEd–p) can serve as a descriptor in describing the OER catalytic activity of transition metal oxides.185–187 For example, Lee et al. investigated the OER and ORR catalytic activities of the oxygen vacancy modified perovskite catalysts. They found that the metal d-band center (Md) values of the perovskites with early TMs (Cr, V, and Mn) are higher than their O p-band center (Op) values, while the perovskites with late TMs (Co, Ni, and Cu) have lower Md values than the Op values. This implies that changing the d-band center can modulate ΔEd–p and thereby alter the OER/ORR performance (Fig. 8b). DFT calculations verified that introducing the surface oxygen vacancies can also modulate ΔEd–p and thereby effectively tune the adsorption strength of the intermediates and contribute significantly to the OER activity.176 Xu et al. proposed that the relative positions of the O p-band center and the MOh (Co and Ni in octahedra) d-band center in spinel oxides ZnCo2−xNixO4 (x = 0–2) are correlated with their stability and different reaction mechanisms. Based on this, they predicted ZnCo1.2Ni0.8O4 as a high-activity catalyst, which has been experimentally verified. In addition, they found that Ni substitution can adjust the relative position between the O p-band center and the MOh d-band center in ZnCo2O4, leading to changes in the reaction mechanism and enhancement in the OER activity.188
The rational introduction of oxygen vacancies can alter the electronic properties of the active sites, which in turn affects the binding strength of the OER and ORR intermediates to the active sites. Zhang et al. studied the amorphous/crystalline RuO2 with rich oxygen vacancies for OER. They attributed the high activity of the catalyst to the downshifting d-band center relative to EF, which leads to a weakening of the chemical bond between the oxygen-containing intermediate and the RuO2 surface, thus lowering the activation barrier for the OER.189 Yang et al. comprehensively studied the bifunctionality of LaCoO3 with oxygen vacancies for both OER and ORR through DFT calculations. They confirmed three criteria: (i) oxygen vacancies can change the d-band center of Co atoms and the p-band center of O atoms, i.e., ΔEd–p, (ii) the AEM reaction mechanism usually occurs at the Co active site when the oxygen vacancy locates at the sublayer, and (iii) the reduced desorption energy for *O leads to increased ORR activity and the cobalt and oxygen vacancies react synergistically in promoting the ORR process.190 Huang et al. synthesized oxygen vacancy enriched Mo-substituted Co3O4 hierarchical porous nanoneedle arrays on carbon cloth (P-Mo-Co3O4@CC) by an Ar-plasma-assisted method. They mentioned that the Mo atom substitution can effectively regulate the atomic ratio of Co3+/Co2+ in the structure and increase the number of oxygen vacancies, leading to electron transfer and regulating the d-band center of Co3O4 (Fig. 8c), subsequently improving the adsorption energy of oxygen intermediates (e.g., *OOH) on P-Mo-Co3O4@CC.177
Apart from the oxygen vacancies, strain engineering also plays a key role in tuning the electronic structure of materials. Extensive works have shown that employing strains can indirectly change the binding strength between the active site and the intermediate, which in turn promotes the activation of molecules and thereby the catalytic activity of materials. Zhang et al. proved that the increased compressive strain induced by the folded nanostructures can weaken the binding strength of the adsorbed oxygen-containing substances and accelerate the kinetics process of the ORR, which in turn leads to a high catalytic activity for highly wrinkled Pd nanosheets.191 P, N, and S can be used as non-metal dopant atoms in oxides, because they have similar properties to the O element. Cui et al. studied P-doped MoO2 and demonstrated that the P-doping raises the d-band center of MoO2, which enhances the interaction between MoO2 and the associated oxygen-containing intermediates during ORR/OER, and thereby improves the catalytic activity of the material. Furthermore, they assembled zinc–air batteries with P-MoO2 as the cathode material, which exhibited excellent charge and discharge performance.192
In recent years, SACs have received increasing attention in catalysis due to their wide range of active sites and large specific surface areas. In addition, non-planar axial ligand atoms are also creeping into the limelight, which can change the electronic properties of the reaction sites and improve the catalytic activity. Wang et al. studied the atomically dispersed penta-coordinated-Zr catalyst with the axial oxygen ligand for ORR and confirmed the effect of oxygen ligands on the active center. They uncovered that the reduced d-band center of Zr caused by the oxygen axial ligand can result in a stable local structure and proper intermediate adsorption capacity at the Zr site. As a result, the ORR performance of Zr–N–C–O significantly exceeds that of the commercial Pt/C catalyst, achieving a half-wave potential of 0.91 V and excellent durability (92% current retention after 130 h) for ORR.193 Li et al. studied the Nd-doped RuO2 catalyst for OER under an acidic condition. Theoretical calculations showed that the d-band center of the Ru atom in the doped structure was shifted downward, resulting in a balance in the adsorption and desorption of OER intermediates. In addition, experiments showed that the dissolution of RuO2 was suppressed after the introduction of Nd, indicating that Nd can improve the acid resistance of RuO2.194
![]() | ||
Fig. 9 (a) The schematic diagram of the NRR for CrN3@graphene. (b) and (c) The interaction between Cr-3d and N2-2p on (b) pristine and (c) lattice strained (2.5%) CrN3@graphene. Reproduced with permission. Copyright 2021, The Royal Society of Chemistry.199 (d) and (e) The linear relationship between the d-band center of M and the ΔG of key reaction steps in NRR on the proposed M3C12X12. (f) The volcano curve between the d-band center and the ΔG of PDS. Reproduced with permission. Copyright 2021, The Royal Society of Chemistry.202 |
The doping of graphene with N atoms to form a graphene-like CxNy structure as a catalyst has received considerable attention in recent years. Two commonly used examples of such catalysts are g@CN and g-C3N4. Recently, Yu et al. developed a novel Ru-loaded g-C3N4 catalyst. Through experimental characterization, they demonstrated that the considered catalyst has high activity for NRR in strong alkaline electrolytes with an NH3 yield of 23.0 μg mg cat−1 h−1 at 0.05 V and a Faraday efficiency (FE) of 8.3%. Through DFT calculations, they compared the NRR process of the synthesized material with bulk Ru, and revealed that the activity of the Ru-loaded g-C3N4 catalyst originates from the altered d-orbital energy, which results in moving the d-band center toward EF.203 In 2021, Li et al. constructed three Mn-based SACs, namely Mn@g-CN, Mn@C3N4, and Mn@g-N4. Among these, Mn@g-N4 has the largest η for electrocatalytic NRR (0.75 V). After introducing heteroatoms (F, P, and S) and imposing strain engineering, the d-band center energy of the central Mn atom significantly reduced, resulting in the improved performance for electrocatalytic NRR.204 Recently, researchers have designed several materials with structure similar to traditional graphene-like carbon and nitrogen compounds. For instance, Zhang et al. designed a series of TM SACs anchored on a π–π conjugated graphitic carbon nitride (g-C10N3) substrate with Dirac dispersion and good conductivity through DFT calculations. They applied a high throughput screening from 27 candidates to evaluate their NRR activity and predicted that W@g-C10N3 is a promising catalyst with a UL of only −0.46 V. They explained the origin of its catalytic activity in terms of d electron number and electronegativity.205 In recent years, the development of machine learning (ML) techniques has revolutionized catalyst screening and descriptor selection. ML algorithms can efficiently analyze large amounts of data and establish correlations between catalytic activity and various parameters. This allows researchers to rapidly identify the most influential factors that govern the catalytic performance. Mukherjee et al. applied a combination of machine learning and DFT calculations to predict Ta-based C6N6 as the best candidate material for NRR. In their machine learning regression model, they identified the covalent radius and d-band center of TM as the most relevant descriptors to correlate the N2 adsorption energy.206 Another new class of electrocatalysts, TM@g-C4N3, is also used to assess the NRR activity and unveil the underlying mechanism of nitrogen fixation. Through high throughput screening and DFT calculations, Wang et al. systematically investigated the NRR activity of 3d, 4d, and 5d TM embedded g-C4N3 catalysts and identified four good catalysts through a five-step screening. Interestingly, they found that the onset potential (Uonset) and the adsorption energy of N2 (ΔEads(*N2)) showed a volcano-shape relationship, and thus ΔEads(*N2) can be used as a descriptor to correlate the catalytic activity of NRR. Through an in-depth analysis of the readily accessible intrinsic properties of the best four selected catalysts, they revealed that the order of increase in activity is consistent with the order of increase in the d-band center of the catalyst.207 Quan et al. designed 12 TM SACs on g-C3N5 nanosheets (TM-g-C3N5) to explore their underlying mechanism for NRR. Through a three-step screening process based on stability, activity and selectivity, V-g-C3N5 was identified as the best electrocatalytic NRR catalyst among 12 TM-g-C3N5 catalysts with an onset potential of −0.30 V. For V-g-C3N5, the process of *N2 → *N2H is considered as the PDS, meaning that ΔG*N2H is the key factor in evaluating the overall NRR activity. Interestingly, the d-band center is linearly correlated with ΔG*N2H, indicating that the d orbit of the considered catalysts is also an important factor in evaluating the NRR activity.208
2D metal–organic frameworks (MOFs) have attracted significant attention in the field of electrochemical reactions due to their unique properties and advantages, including high conductivity, chemical tunability, easy synthesis, and facile accessibility to active sites. M3C12N12, also known as MOF-235, is a commonly studied MOF with a well-defined structure. Researchers have explored different variations of M3C12N12 by varying the central metal atoms and the coordinating atoms. In 2021, Wang et al. constructed various M3C12X12 catalysts with X = O, S, and Se, and evaluated their NRR performance (Fig. 9d). In their research, *N2 → *N2H or *NHH → *NH3 is considered as the PDS, and interestingly, ΔGN2→N2H and ΔGNHH → NH3 exhibited an inversely proportional relationship. In addition, the d-band center of M3C12X12 has a linear relationship with the ΔG of the PDS for NRR (Fig. 9d and e). Therefore, the d-band center can be taken as a good descriptor to evaluate the NRR activity of these catalysts (Fig. 9f).202 In 2023, a study focused on investigating the influence of p–d orbital hybridization on electrocatalytic NRR was conducted using M3C12X12 MOFs with X = O, S, Se, and NH and with different central metal atoms including Os, Ru, and Ir.209 The study reported that the strength of the hybridization interaction between the appropriate metal–d and intermediate atom-p orbitals is a key indicator of the NRR catalytic activity. Additionally, the strength of hybridization interactions can be quantitatively calculated by the p–d band central energy difference (Δd–p), which is a powerful descriptor to correlate the p–d hybridization strength with NRR performance. As a result, a volcano relationship between the maximum free energy change (ΔGmax) and Δd–p was obtained, and the OsC4(Se)4, IrC4O4, and RuC4(Se)4 catalysts located at the top of the volcano curves were proved to have better NRR performance.209
TMDs, abbreviated as MX2 (M = Mo, W, etc.; X = S, Se, Te), have attracted a lot of attention worldwide due to their 2D layered structure, which is similar to graphene. Doping TMs into ReS2 nanosheets is also a reliable method to simultaneously improve nitrogen fixation and suppress competitive HER. The catalyst with Fe-doped ReS2 nanosheets wrapped around nitrogen-doped carbon nanofibers (Fe–ReS2@N-CNF) has the best electrochemical activity with an NH3 yield of 80.4 μg h−1 mgcat−1, along with an overpotential as low as 146 mV and a Tafel slope of 63 mV dec−1. Through DFT calculations, it was interpreted that the dopant atoms provided more active sites and electron transfer for ReS2, which caused the d-band centers move towards EF, thereby changing the adsorption strength of the intermediates, and consequently achieving different catalytic activities.210 Similarly, VSe2 with abundant Se vacancies also exhibited high NRR performance, which is attributed to the increase of the d-band center of V atoms induced by the Se vacancy. This can promote the binding of key *N2/*NNH intermediates to the active site, resulting in an energetically favorable NRR process.211
The construction of heterojunctions allows for different interfacial charge transfers for the purpose of changing the electronic properties of the surface. In 2021, Liu et al. reported the FeNi2S4/NiS heterogeneous structure, which effectively promoted N2 adsorption in NRR and inhibited the competitive HER. Its NH3 yield and FE reached up to 128.398 ± 1.32 μg−1 h−1 cm−2 and 28.64 ± 0.18%, respectively. Both experimental results and DFT calculations indicated that the enhanced electrocatalytic performance and selectivity of the FeNi2S4/NiS heterostructure for NRR were attributed to the changes in the d-band center and the electronic structure of the catalyst surface.212 In 2022, Guo et al. designed a novel structure of WO3 in contact with TMDs, including MoS2, MoSe2, WS2, and WSe2, and validated their potential as the NRR electrocatalyst. The WO3/MoS2 heterostructure was identified as an optimal catalyst with a low onset potential of 0.25 V, which is attributed to its elevated d-band center that effectively modulated the adsorption strength of intermediate adsorbates.213 In 2022, Biswas developed a Lewis acid-dominated aqueous electrolyte as a co-catalyst to overcome the N2 activation problem on the catalyst surface. N2 adsorption is the first and the most critical step in NRR. DFT calculations have shown that the BF3-induced charge polarization can shift the metal d-band center of the MnN4 moiety closer to EF, and thereby facilitate the N2 activation. The free BF3 molecule displays Lewis acidity, further suggesting that it can facilitate the NN bond polarization as well as the first protonation step. In addition, the change in the d-band center can effectively reduce the PDS energy of the NRR. As a result, a high NH3 productivity can be achieved using the MnN4/BF3 catalyst, with a rate of 2.45 × 10−9 mol s−1 cm−2, which is comparable to the yields achieved in the industrial NH3 production process.214
For CO2RR, selectivity is by far the most significant and urgent challenge in this field. Among pure metals, Cu has attracted much attention due to its unique selectivity for CO2RR. In addition, various Cu-based alloys have been developed to further improve the catalytic activity and selectivity towards specific products. Rasul et al. prepared copper–indium (Cu11In9) alloys by electrochemical deposition of In on rough Cu surfaces and achieved efficient and selective reduction of CO2 to CO. The upward shifted d-band center of the Cu atoms induced by doping In atoms attributed to the improved catalytic activity.217 In 2017, Clark et al. reported electrochemical CO2RR at a CuAg bimetallic electrode, which exhibited higher selectivity towards multi-carbon products compared to a pure Cu electrode. The introduction of Ag resulted in a compressive strain on the neighboring Cu atoms, which shifted the d-band distribution of Cu to a deeper energy level. This modified electronic structure of Cu was found to be beneficial for reducing the binding energies of *H and *O compared with *CO intermediates, thus effectively suppressing the competitive HER and increasing the selectivity towards CO-derived products. Furthermore, the strain effects induced by Ag introduction can also reduce the adsorption of *H species and decrease the bond strength of the Cu–O bond, consequently contributing to enhanced selectivity towards multi-carbon products. This work provided an effective way to modify the CO2RR selectivity in order to generate multi-carbon oxygenated products.218 Ren et al. synthesized isolated Sn sites with a high surface density (up to 8%) anchored on Cu hosts (denoted as Cu–Sn NPs) (Fig. 10a) through a simple one-step reduction method. In an alkaline electrolytic cell, the Cu97Sn3 catalyst achieved a FE of 98% for CO production at a tiny overpotential of 30 mV. A high CO current density of 100 mA cm−2 can be achieved on this catalyst at an overpotential of 340 mV. DFT calculations demonstrated that the PDOS of the d-orbitals of Cu in Cu–Sn alloys shifted to deeper energy levels and the d-band centers for Cu–Sn alloys are farther away from EF, making the *COOH and *OCHO species interact with the metal surface weakly, which improves the UL(CO2)–UL(H2) of CO production, and therefore achieves a better CO2RR activity (Fig. 10b and c).219
![]() | ||
Fig. 10 (a) The structures of pure Cu and its alloys. (b) The free energy diagram of the CO2RR pathway. (c) The d-band center of Cu and Cu–Sn alloy. Reproduced with permission. Copyright 2021, Nature.219 (d) The d-band centers of Rh NP, Rh NS, Rh73W27 NP, and Rh75W25 NS catalysts. (e) Their corresponding products for CO2RR. Reproduced with permission. Copyright 2017, American Chemical Society.220 (f) The optimized structure of N4Fe–CuN3. (g) The linear sweep voltammetry profiles of Fe/Cu–N–C catalysts in a CO2-saturated 0.1 M KHCO3 electrolyte. (h) PDOS and d-band center of Fe on N4Fe–CuN3 and FeN4. Reproduced with permission. Copyright 2021, The Royal Society of Chemistry.221 |
At present, Cu-based alloys are considered as the state-of-the-art CO2RR catalysts; some other non-noble metal alloys have also been explored as promising catalysts for CO2RR. For example, ultra-thin Rh75W25 alloy NPs have been successfully prepared by Zhang et al., which exhibited better performance than the Rh nanosheets (NSs) and Rh73W27 NPs for catalytic CO2 hydrogenation to CH3OH. According to DFT calculations, the Rh75W25 NSs exhibit higher d-band centers (Fig. 10d), thus enhancing the adsorption and activation of CO2. Correspondingly, the experimental results showed that Rh75W25 NSs exhibited the highest methanol yield (Fig. 10e).220
For pure metallic materials, strain engineering is often employed to change the lattice spacing and adjust the position of the d-band center. Huang et al. investigated the electrochemical performance of CO2RR with octahedral and icosahedral Pd catalysts. Experimental results demonstrated that the maximum FE of the Pd icosahedral/C catalyst was 91.1% at 0.8 V, which was 1.7 times higher than the maximum FE of the Pd octahedral/C catalyst at 0.7 V. Theoretical calculations showed that the Pd icosahedra have tensile strain relative to the Pd octahedra, leading to the d-band center close to EF, which enhances the adsorption of the key intermediate *COOH on the metal surface, and consequently improves the catalytic activity.222 Au NPs with compression-rich strain were synthesized via a liquid laser ablation strategy (denoted as Au-LAL) by Zhang et al. and were used for electrochemical CO2RR. The experimental results indicated that the presence of compressive strain in Au-LAL resulted in higher current densities and FEs compared to plain Au NPs. In situ Raman characterization and theoretical calculations confirmed that the d-band center of the strained Au is changed, which increased the binding strength between the key intermediate *COOH and Au, and thus modulated the catalytic activity and selectivity of the Au-LAL catalyst.223
SACs are one of the commonly used 2D catalysts. The strong interaction of CO on the FeN4 moiety limits CO desorption and therefore reduces the CO2 to CO efficiency. The FE of CO2RR can be significantly improved by constructing holes on the graphene substrate, because the introduction of hole edges decreases the d-band center of the active Fe atom site and weakens the Fe–C bond strength, which leads to a moderate *CO adsorption strength and promotes CO desorption.224 Feng et al. designed and synthesized a nitrogen-doped carbon substrate with precisely controlled Fe–Cu di-atom sites at the atomic level and characterized its catalytic performance for CO2RR (Fig. 10f). Experiments demonstrated that Fe/Cu–N–C has higher current density (Fig. 10g), excellent FE, and high turnover frequency, which was attributed to the synergistic interaction between Fe and Cu that enables fast charge transfer, optimizes the position of the d-band center (Fig. 10h), and lowers the energy barriers for *COOH formation and *CO desorption.221
Briefly, the d-band center stands as the most widely used descriptor due to its quantitative interpretation of the interaction between the adsorbate and the catalyst surface. In recent years, significant advancements have been made to enhance the d-band center concept, aiming to gain a better understanding of the unique properties of the catalyst surface and the adsorbates involved in catalytic reactions. In future applications of the d-band center theory, a comprehensive approach incorporating the specific adsorbed atoms or molecules, along with the attributes of the active sites (top, bridge, or hollow), will be crucial in precisely predicting the activity of catalysts.
In catalytic reactions, the active center, which comprises both active adsorption sites and their coordination environment, plays a significant role in determining the overall catalytic activity and selectivity. Therefore, a careful consideration must be exercised on the synergy between active sites and their coordination environment to achieve the desired catalytic performance.85 Recently, a general design principle was proposed by Xu et al. to evaluate the activity of graphene-based SACs for ORR, OER, and HER. This design principle takes into account multiple internal factors, including coordination number, electronegativity of active sites and the nearest neighboring atoms, and the elemental periodic correlation coefficient. By studying single vacancy with three carbon atoms (SV-C3), double vacancy with four carbon atoms (DV-C4), four pyridine nitrogen atoms (pyridine-N4), and four pyrrole nitrogen atoms (pyrrole-N4) (Fig. 11a), they established a descriptor: , where α represents the correction coefficient, EM, EN, and EC are the electronegativity of different atoms, and nN and nC are the number of nearest-neighbor N and C atoms. Interestingly, this descriptor showed a volcano relationship between ηHER and UonsetOER (Fig. 11b and c). Furthermore, the applicability of this descriptor can also be extended to macrocyclic molecules.231 Furthermore, molybdenum disulfide (MoS2), featuring unique electronic and optoelectronic properties, has attracted intensive attention. In particular, Mo-terminated edge sites of MoS2 have demonstrated remarkable catalytic activities comparable to that of Pt-based catalysts. However, there is still a lack of uniform descriptors to avoid the dependence on traditional trial-and-error methods in catalyst design. Ran et al. proposed a universal descriptor based on the theoretical study of HER on various MoS2 structures such as TM-substitute, S-vacancy, Mo-edge, and S-edge. The descriptor exhibited a linear relationship with ΔG*H, which can be defined as
, in which NBTM−S is the number of TM–S bonds. XA−B is defined as
, where XA and NCA are the electronegativity and the corresponding coordination number of atom A. NVTM–S is the covalent electron number of the TM–S bond. RTM is defined as
. The results highlighted the significance of bond properties within the coordination microenvironment for catalytic reactions. More interestingly, a similar volcano curve between the descriptor and ΔG*H can be found in the WS2 system with different defects.232
![]() | ||
Fig. 11 (a) Four geometric configurations of graphene-based SACs. (b) UonsetOER and (c) ηHER of graphene-based SACs as a function of the descriptor φ′. Reproduced with permission. Copyright 2024, Springer Nature.231 |
The electronegativity and the number of valence electrons can also be utilized to describe the catalytic NRR activities. For example, Guo et al. studied the single TM decorated defective MoS2 for NRR.233 They discovered that the catalytic NRR activity was limited by the hydrogenation of the second N atom and the desorption of the second NH3 molecule. These processes appear to be affected by the strong interaction between N and the decorated metal atom. By introducing the descriptor, the correlation between ΔE*N and the readily accessible intrinsic property of the active site is established. The descriptor is denoted as φ = θd × EM, where EM and θd are the electronegativity and the number of valence electrons respectively for the embedded TM atom. The greater value of φ corresponds to the weaker bonding of N to the active site.
Recently, 2D transition metal borides, known as MBenes (boron analogues of MXenes), have emerged as a new kind of promising catalyst due to their large specific surface area and high density of active sites. Yao et al. evaluated the NRR performance of TMs in Mo vacancies of the MBene nanosheet (Mo2B2O2). They identified the first hydrogenation of N2 as the PDS and highlighted *NNH as the key intermediate. Furthermore, they proposed a binary descriptor based on the number of d electrons and the period of the doped metal atom to establish a connection between the readily accessible intrinsic properties and catalytic activities of the catalyst.234 Among the various metal catalysts for NRR, Au is often regarded as the best choice due to its high catalytic performance for NRR and low activity for competitive HER.235,236 In order to further improve the activity of the Au catalyst and explore its nature in NRR, Zheng et al. screened a series of Au-based single-atom alloys and proposed three descriptors, including key intermediates and readily accessible intrinsic properties of the doped metal. They identified the reaction step *N2 → *NNH as the PDS because the triple bond in N2 is relatively inert. Interestingly, they observed a volcano relationship between ΔG(*N2–*NNH) and the readily accessible intrinsic descriptor , where d and Em represent the number of d electrons and electronegativity of the doped metal atom (Fig. 12a and b). Based on the volcano relationship, the Mo- and W-doped Au(111), located at the peak of the volcano curve, was identified as the best catalyst for NRR.237 Likewise, the descriptor
can also be used in single-atom-embedded artificial holey SnN3 monolayers (Fig. 12c) for NRR, and a volcano relationship was found between UL and φ (Fig. 12d).238 In addition, the first ion energy of the active center can also influence the bonding strength between active sites and the absorbates due to the electron transfer during the interaction process. Recently, Ouyang et al. demonstrated that boron-containing co-doped graphene showed improved catalytic activity for NRR compared to carbon-based catalysts. Based on the calculations, ΔG*NNH has a linear correlation with ΔGmax. In order to reveal the intrinsic characteristics of these doped structures, the researchers proposed a descriptor defined as Φ = (EX/EB) × IX1/IB1, where E and I are the electronegativity and the first ion energy of the atom, respectively. X and B are the doped element and the boron atom, respectively. The study highlighted that the intrinsic difference between doped elements and the carbon atom has a significant impact on the electron transfer and bonding strength of the intermediates, which ultimately influences the catalytic activity of the catalyst in NRR.39
![]() | ||
Fig. 12 (a) Models of M/Au(111) single-atom alloys. (b) A volcano relationship between ΔG(*N2 → *NNH) and the descriptor φ. Reproduced with permission. Copyright 2021, American Chemical Society.237 (c) Top view and side view of TM/d-SnN3 with a single TM atom embedded in the central cavity. (d) Volcano plot for the UL of TM/d-SnN3 as a function of the descriptor Φ. Reproduced with permission. Copyright 2022, Wiley-VCH.238 |
Electrocatalytic reduction of nitrate (NO3−) or NO to NH3 (NO3RR) has attracted intensive attention in recent years. This process offers the advantages of simultaneous elimination of harmful substances and production of valuable products.239–241 In order to address the lack of information on the underlying catalytic mechanism and the impact of readily accessible intrinsic factors on NO3RR, Wang et al. systematically investigated the performance of NO3RR/NORR on TMN3 and TMN4 materials and used the activity descriptor . This descriptor aims to elucidate the relationship of readily accessible intrinsic properties of TM centers and their local environments with regard to the catalytic activities of NO3RR. The adsorption of NO3− is linearly correlated with the activity descriptor for both TMN3 and TMN4 materials. Additionally, a volcano-type diagram was discovered between the descriptor and the UL of NORR on TMN3 and TMN4 materials. More importantly, this descriptor also allows for the identification of other graphene-based SACs with excellent NO3RR activity.242
The electrochemical CO2RR presents several challenges, including multiple electron transfer processes, complex reaction intermediates, and formation of multiple reaction products. These factors have posed significant limitations on developing efficient catalysts for CO2RR. To overcome this bottleneck and identify high-performance catalysts, it is essential to search for descriptors that can accurately predict the catalytic activity of catalysts. The descriptor used for CO2RR may include multi-type readily accessible intrinsic properties such as electronegativity, atom radius, and coordination microenvironment feature due to its complex reaction intermediates. For instance, Gong et al. have conducted a theoretical study on M–N–C (porphyrin)-contained SAC structures with 3d TMs for the CO2RR (Fig. 13a), and revealed a volcano-shaped relationship between η required for CO formation and the bonding energy of the *CO intermediate (Fig. 13b). Further studies that focused on the readily accessible intrinsic properties of catalysts have demonstrated that the ηCO for CO2RR can be described as a segmented function of the descriptor Φ = VM × EM/rM depending on different reaction pathways, where VM, EM, and rM represent the valence electron number, the electronegativity of metal atoms, and the radius of metal ions, respectively (Fig. 13c). In addition to the 3d TMs, the descriptor can also be extended to predict the catalytic behavior of CO2RR for 4d and 5d TMs.243 Wang et al. improved the efficiency of CO2 to CH4 conversion by introducing axial O atoms in TM-coordinated nitrogen-doped carbon SAC species (M–N4O–C) (Fig. 13d).244 For both M–N4–C and M–N4O–C, η can be expressed as a function of ΔG(*OH) and ΔG(*CHO), which are strongly correlated with the proposed descriptor ; the relationship diagram of ΔG(*OH) and η on M–N4O–C is shown in Fig. 13e. Here, VM is the d-shell electron number of a metal atom; N is the coordination number of the central atom; XM, XL, XC, and XO represent the electronegativity of the M atom, its nearest atom L, adsorbate C and O atoms connected by metal atoms, respectively; and d is the bond length between the central atom and the nearest neighbor atom. Finally, the volcano curve for η and φ is obtained (Fig. 13f). Correspondingly, the relationship between the descriptor and catalytic CO2RR activities is linked by the key intermediates in the CO2RR reaction pathway.244
![]() | ||
Fig. 13 (a) The top view of TM-based SACs. (b) The volcano curve between the overpotential and ΔG*CO. (c) The volcano curve between the overpotential and the descriptor φ. Reproduced with permission. Copyright 2019, Wiley-VCH.243 (d) The side view and top view of the M−N4O−C catalyst. (e) The linear relationship between ΔG(*OH) and the descriptor φ. (f) The volcano relationship between UL and the descriptor φ. Reproduced with permission. Copyright 2022, American Chemical Society.244 |
In CO2RR, effective suppression of competitive HER and side products is of great importance to improve the efficiency of target products. Existing works on porous covalent organic frameworks (COF) showed an extremely low η (0.34 V) for the selective reduction of CO2 to CO, while preventing the formation of the unwanted by-products such as H2 and HCOOH. In the mechanistic explanation of different catalytic activities on these COF catalysts, a concise and clear descriptor was put forward. In this descriptor, N, rTM, and n represent the number of d electrons, atomic radius, and periodic number of TM, respectively. By using this descriptor, a volcano relationship between ηCO and Φ was established.245 Achieving a comprehensive understanding of the underlying mechanism and improving the selectivity of Cu catalysts for the CO2RR remain a challenging task. In this regard, Zhu et al. conducted a study employing grazing incidence X-ray diffraction and electron back-scattered diffraction techniques to track the top-surface reconstructions of Cu catalysts during CO2RR. Building upon their observations, they proposed three types of descriptors (crystal facet, atomic coordination number, and step-terrace angle) to establish an intrinsic structure–performance relationship. These descriptors provided valuable insight into the mechanism of CO2RR and aid in the design of more efficient CO2RR catalysts.246 In a recent DFT study, Gao et al. quantified the effect of the readily accessible intrinsic properties of adsorbates and substrates on their bonding behavior through step-by-step correction of descriptors. The studied adsorbates encompassed key intermediates involved in the reaction, while the chosen substrates included a diverse range of TMs, NPs, intermetallics, and oxides. Finally, the researcher innovatively developed a model for determining the adsorption energy (Ead) using an electronic descriptor (Ψ). This relationship was expressed as
. Here, Ead represents the adsorption energy of the intermediate, Xm denotes the maximum number of bonds that the central atom of a given adsorbate can form with the specific coordinating group, X represents the actual number of formed bonds, Ψ is related to the number of electrons and electronegativity,
indicates the generalized coordination number of surfaces,247 and θ stands for a constant associated with each adsorbate. Importantly, this model covers a broad range of intermediates, as well as the physical and chemical properties of various materials, providing a long-sought guidance for future materials design.248
Using the readily accessible intrinsic nature of a catalyst as a descriptor for predicting its catalytic activity offers a cost-effective and practical approach that avoids the need for extensive DFT calculations and experimental tests. When considering readily accessible intrinsic properties, it is essential to meticulously examine certain physical and chemical properties, such as the atomic characteristics of the active site, the distance to the surrounding ligand atoms, and the variance in electronegativity. In addition, accurately assigning weights to each of these factors is a complex issue that requires in-depth study. This is due to the fact that different catalysts exhibit significantly varied influencing factors on catalytic activity, which are interrelated with both the catalyst itself and the adsorbate.
![]() | ||
Fig. 14 (a) Magnetic susceptibility of d-FeN4 or FeN4 and their orbital interactions with O2. Reproduced with permission. Copyright 2022, ScienceDirect.251 (b) The proposed ORR mechanism for the Fe1–N4SC catalyst. (c) and (d) The overpotential as a function of (c) ΔG*OH and (d) magnetic moment for the considered catalysts. Reproduced with permission. Copyright 2021, Wiley-VCH.254 (e) The structures of Se- and Te-doped C catalysts. (f) The correlation between ΔG*NNH and the spin moment of the active C atom. (g) Free-energy diagrams of NRR on Se- and Te-doped C catalysts. Reproduced with permission. Copyright 2020, Wiley-VCH.255 |
In 2021, Yang et al. rationally designed and synthesized an atomically dispersed bimetallic Fe, Mn/N–C catalyst for ORR. They aimed to modulate the catalytic activity of the Fe–N structure by altering the spin state of the Fe atoms. Through measurements of zero-field cooling temperature-dependent magnetization, they found that the FeIII ions in the intermediate spin state of the FeN4/C system, with a single electron configuration (t2g4eg1), were preferentially involved in OER. This allowed for the occupation of the antibonding π-orbital of oxygen, resulting in high catalytic activity of the designed Fe, Mn/N–C catalyst.259 Li et al. demonstrated that strain engineering can modify the spin states of the active site, thus impacting the catalytic activity of catalysts. By applying a 9% strain to CoOOH, they were able to effectively adjust the electronic arrangement of Co atoms from a low-spin configuration (t2g6eg0) to a high-spin configuration (t2g4eg2). This adjustment significantly reduced the energy (from 1.35 to 0.03 eV) required for releasing O2 from adsorbed intermediates during the OER.260
Numerous studies have emphasized the significance of the first protonation step (resulting in the formation of *NNH) as the decisive PDS in the NRR. In 2020, Qiao et al. conducted an extensive computational investigation on carbon materials doped with oxygen-group elements (O, S, Se, Te) as potential catalysts for NRR (Fig. 14e). They highlighted the significant influence of spin polarization on the adsorption of the *NNH intermediate and identified a linear relationship between the adsorption energy of *NNH (ΔG*NHH) and the magnetic moment (Fig. 14f). Through comprehensive mechanistic studies, they demonstrated that the heteroatom doping-induced charge accumulation enhanced the adsorption of N2 on carbon atoms. Moreover, the heteroatom doping-induced spin polarization promoted the formation of *NNH through the first protonation step, reducing the energy barrier associated with the decisive velocity step and enabling more efficient NRR. The final results indicated that the Se-doped carbon catalyst exhibited favorable electronic and spin properties, enhancing the adsorption and activation of N2, thus displaying high intrinsic NRR activity (Fig. 14g).255 In 2021, Zhang et al. conducted further research on Co-doped LaNiO3 (denoted as Co-LNO) with a focus on investigating the relationship between the effective spin magnetic moment of the catalyst and its catalytic performance in NRR. The study demonstrated that LaNi0.995Co0.005O with a 3d high-spin configuration exhibited remarkable NRR performance. Specifically, it achieved a high NH3 yield of 14.57 mg h−1 mg−1, an outstanding FE of 26.44%, and a remarkable energy efficiency of 21.35% (at 0.1 V vs. RHE). Further investigation into the mechanism revealed that Co-LNO provided additional catalytically active sites comprising Co and accompanying oxygen vacancies. These sites played a crucial role in modulating the electronic structure of LNO, promoting N2 adsorption, and facilitating the first protonation step during the NRR process.264
Cao et al. summarized a vast amount of experimental data on NH3 synthesis and demonstrated two crucial factors that impact the catalytic activity of the NRR, providing a new pathway for the development of advanced catalysts for NH3 synthesis. The first factor identified is the electrostatic interaction between the adsorbed promoter and the N–N dissociative transition state. The second factor is the magnetic catalyst promoter effect, which involves the manipulation of the spin properties of the catalyst, resulting in a decrease in the activation energy for NRR.265 Building upon this research, Guo et al. proposed a strategy to modulate the NRR activity of magnetic catalysts by regulating their spin magnetic moments. Through DFT calculations, they observed significant changes in the magnetic moments of Fe atoms based on the coordination environment of neighboring N atoms. Notably, the magnetic moment of Fe atoms exhibited a linear correlation with both the applied potential U versus SHE and the adsorption energy of N atoms. By controlling the magnetic moments of Fe atoms, it becomes possible to adjust the adsorption strength of intermediates, fine-tuning the catalytic activity, and ultimately lowering the UL of NRR.266 In a recent study, Huang et al. employed B-doped carbon dots to modulate the electron spin density of the magnetic Co3O4 catalyst, resulting in an exceptional FE for production of NH4+ (94.6 ± 0.9%). The change of spin configuration from t2g6eg0 of Co3+ to t2g5eg2 of Co2+ was identified as a key factor in enhancing the electrocatalytic NO3RR. Experimental results and theoretical calculations further supported that the modulation of spin-density affected the PDS, reducing the energy barrier that needed to be overcome, and ultimately optimizing the catalytic activity.267
The spin arrangement and magnetic moment of the magnetic atom within the active site of a catalyst have been demonstrated to exert a significant influence on the binding strength with crucial intermediates, and consequently impact its catalytic activity in specific reactions. Altering the type or quantity of ligand atoms surrounding the magnetic atom at the active site can modulate its electron arrangement, leading to changes in the occupancy of various orbitals. In studies related to spin-catalysis, emphasis is placed on acquiring the orbital distribution and electron occupation of the central atoms, which can be correlated with the adsorption of key intermediates to predict the catalytic activity.
In general, the process of using ML methods (Fig. 15a)271 to identify the parameters affecting the catalytic activity and the extent of each parameter entails the following steps: (i) data preparation: this involves acquiring the essential multidimensional features from the database, including readily accessible intrinsic properties of catalysts, spin properties, d-band centers, and other relevant factors. Additionally, the inclusion of experimental or DFT results is essential as part of the training data. (ii) Model selection: the optimal ML model is selected by comparing the output results and mean absolute error (MAE) of different models to ensure the most suitable model for the specific application. Generally, a smaller MAE indicates a more accurate ML model. (iii) Descriptor determination: this involves utilizing the selected ML model to assess the impact of various input parameters on catalytic activity, thereby identifying the most significant descriptors. It is important to include as many input parameters as possible to ensure a more precise search for descriptors, as the catalytic performance of a catalyst is generally determined by multiple factors. (iv) Catalyst prediction: this step is performed using the recognized factors and the identified descriptors from the ML model to enable the screening of known catalysts or the design of new catalysts with improved performance.
![]() | ||
Fig. 15 (a) Workflow of machine learning methods. Reproduced with permission. Copyright 2022, American Chemical Society.271 (b) The feature importance calculated by random forest regression methods for ΔG*H, Copyright 2023, Wiley-VCH.272 (c) The feature importance analysis in the random forest model for CO2RR on graphdiyne and holey graphyne-based SACs. Reproduced with permission. Copyright 2023, Wiley-VCH.273 (d) The feature importance of the most important 12 features got by ML methods. Reproduced with permission. Copyright 2021, American Chemical Society.274 |
Taking HER as an example, Fung et al. utilized *H as a key intermediate to assess the catalytic effectiveness of single-atom N-doped graphene catalysts. By employing the compressed sensing method, SISSO (Sure Independence Screening and Sparsifying Operator), they determined a multidimensional descriptor . In this descriptor, the intrinsic properties such as electronegativity (EN) and covalent radius (rcov), d-state center (εd), Bader charge (q), and the number of occupied d-states (docc) for spin-related properties are all included to collectively evaluate the catalytic activity.275 For the g-CN structure, it has been documented that certain anchored metal atoms (such as Sc, Ti, V, Ga, Sn, and Au) on large holes can serve as excellent catalysts for HER,276 while the strong *H adsorption energy by other metals still poses challenges for the release of *H. A recently reported new class of hybrid double atom catalysts (HDACs), featuring both metal and non-metal atoms embedded in the holes of the g-CN structure, effectively mitigate this issue by altering the geometric and electronic properties between the active site atoms. Through the application of ML, it has been demonstrated that the hydride formation enthalpy (Hxf), the bond angle between H atoms and catalyst surface atoms (θM-H-NM), the distance between H atoms and active center atoms (dM/NM-H), and the p-band center of non-metal atoms (εp) all play crucial roles in determining the catalytic activity of these HDACs for HER (Fig. 15b). In addition, a catalytic activity map of HDACs has been established, providing a predictive framework to accurately forecast the activity of HDACs in unexplored chemical spaces.272 In a recent development, Abraham et al. integrated ML strategies with DFT calculations to screen 2D MXene-based catalysts with exceptional HER catalytic activity. They established nine distinct ML models and employed random forest regression (RFR) and gradient boosting regression (GBR) to reduce the number of features through recursive feature elimination. Ultimately, they identified the valence electron number (VT) and electron affinity (EAT) of the termination group, along with the d-band center relative to the mean variance (dbcs2), as predictive descriptors to correlate ΔG*H.269
In the context of multi-electron reaction processes, such as CO2RR, identifying the primary factors influencing the catalytic activity of specific catalysts is particularly challenging. Ren et al. designed graphdiyne and holey graphyne-based single atom catalysts and identified *CO2 and *OH as crucial intermediates in CO2RR. By leveraging a ML model, they successfully pinpointed the intrinsic properties, including the polarization charge (Q) of the central metal, the interaction between the central metal and the carrier (ICOHP), and relevant magnetic properties such as the magnetic moment of the central metal atom, as the key descriptors (Fig. 15c). Furthermore, they quantified the relative influence of these descriptors on catalytic performance. Interestingly, this ML model can also be applied to quickly predict other candidate materials for CO2RR.273 Another study focuses on dual-metal-site catalysts (DMSCs) for CO2RR, where dual-TM atoms are supported on a phthalocyanine (Pc) ring. This study demonstrated that the gradient boosting regression (GBR) model proved to be the most effective method for predicting the UL of CO2RR for 249 Pc DMSCs. Moreover, the model successfully identified Ag-MoPc and Ag-CoPc as superior catalysts for CO2RR. In the quest to uncover the origin of catalytic activity in these catalysts, *COOH was identified as a key intermediate in the reaction. This study also unveiled that the electronegativity of the active atoms of the catalyst, along with the charge transfer (Q) from the TM atom to the Pc ring, all play pivotal roles in determining the catalytic activity of the catalyst (Fig. 15d).274 Furthermore, due to significant constrains on the kinetics of the CO2RR process and the successful industrial-level conversion of CO2 into CO, catalytic CORR has become a focal point of research. In a DFT calculation study on CORR using single TM atom anchored nitrogen-doped γ-graphene (N-GY) as the catalyst, it was found that the position of metal atoms in the periodic table of elements influences the final product. Specifically, metal atoms in the front and middle positions of the periodic table favor the production of CH4, while TMs are conducive to methanol production. In addition, five key characteristics—Pauli electronegativity (Ne), first ionization energy (Im), atomic radius (Rm), d orbital, and atomic number (Zm) of the TM atom—are utilized as input data to correlate ΔECHO* through ML methods, with Zm identified as the predominant descriptor.277
Despite the significant advances in ML, the training of reliable ML models requires large amounts of high-quality data as input. However, the cost of collecting such data, whether through computational simulations or experimental work, can be prohibitively expensive. As a result, there is an urgent need to develop more efficient methods that can reduce resource waste and maximize data utility. Wang's group has developed an innovative approach that combines automatic DFT calculations with an improved reinforcement learning algorithm, especially the Monte Carlo Tree Search with Policy Gradient (MCTS-PG) algorithm. This integration has enabled the rapid identification of materials with desired properties starting from minimal initial data, thereby reducing the need for extensive DFT calculations. By utilizing this approach, they successfully screened seven reliable catalysts (CuAl3, SbCaCu, ZrInCu2, Si2CaCu2, AlV6Ga, CAlTi2, and ZrInCu2) for CO2 activation and methanation through 200 MCTS-PG steps.278 In summary, ML plays a pivotal role in (i) efficiently screening and selecting eligible catalysts based on their readily accessible intrinsic properties; (ii) predicting crucial properties that connect activities such as adsorption energy; and (iii) determining key descriptors that influence the catalytic performance. These capabilities greatly facilitate the design and optimization of catalysts, contributing to the development of more effective catalysts for diverse chemical processes.
This journal is © The Royal Society of Chemistry 2024 |