Ammonia is a key player in the agriculture and energy sectors, being a precursor and feedstock of fertilizers, and a clean and high energy density fuel^[1-3]. It is equally important in the industrial manufacture of chemicals such as hydrazine, explosives, and synthetic fibers, as a refrigerant, solvent and pH modulator in pharmaceutical products. Conventionally, it is produced by the Haber-Bosch process, which is not only energy intensive but also relies on fossil fuels, increasing the carbon footprint in the atmosphere^[4]. Carbon-free electrochemical reduction of nitrates is an excellent alternative to the conventional ammonia synthesis method. Nitrates accumulate in water bodies through agricultural runoff and industrial discharge, making aquatic life and human health vulnerable. Nitrate reduction to ammonia (NRA) is a green, sustainable, circular-economy-based method, that converts nitrate pollutants into useful ammonia^[5].

Different materials have been used as electrocatalysts for nitrate reduction, including metal-organic frameworks (MOFs)^[6,7], metals and metal alloys^[8,9], metal oxides^[10,11], electrified membranes^[12], various composites^[13], single-atom catalysts (SACs) and dual-atom catalysts (DACs)^[14,15]. SACs bridge heterogeneous and homogeneous catalysis by dispersing individual metal atoms on support materials^[16,17]. They are known for their high atomic utilization, as each atomic center is exposed instead of being buried in bulk, making them economical, especially for precious metals. All metal atoms have well-defined coordination environments and uniform active sites, imparting high selectivity and enabling facile optimization of the coordination environment. Furthermore, when single atoms are coordinatively unsaturated, they exhibit high affinity and lower activation energies for reactant species, enhancing the activity of the catalyst^[18-20]. These unique features make SACs outperforming in various electrochemical reactions^[21-25].

DAC is another advanced catalyst, sometimes referred to as a binuclear or diatomic catalyst. In DACs, two individual (same or different) metal atoms are coupled together as active sites, and scattered on the support material, similar to SACs^[26]. They offer dimeric active sites with synergistic mechanisms on both atoms, allowing the simultaneous adsorption and stabilization of multiple reaction intermediates. Different metal combinations create unique electronic structures with tunable properties, enhancing the reactivity, selectivity, and catalytic efficiency of DACs^[26-29]. These properties make them superior to SACs in many aspects, multi-electron reactions, where synergistic mechanisms on both metal sites are advantageous over individually scattered atoms^[28,30,31].

Numerous SACs and DACs have been reported for NRA. Though both are challenging to synthesize, yet they outperform other catalysts in Faradaic efficiency (FE), yield rate, and atomic utilization. SACs for NRA have been reviewed extensively, focusing on synthesis, microenvironment tuning, effects of the support material, and reaction mechanisms, and listing transition metal SACs for NRA^[32-35]. To the best of our knowledge, no review has provided a comparative analysis of SACs and DACs for NRA along with insights into DAC mechanism. DACs for NRA are growing rapidly, and a comprehensive review is needed not only to list all DACs but also to provide critical analysis and theoretical details in parallel with SACs. Our review provides mechanistic details of NRA on SACs and DACs, discussing the effects of different metals and metal combinations, microenvironments, their symmetry and saturation, performance parameters and descriptors, as well as density functional theory (DFT) screening of different catalysts, concluding with challenges in the field and a future roadmap for researchers. A comparison of SACs and DACs is summarized in Figure 1.

NRA is a multistep electrochemical process involving eight electrons and nine protons as hydrogen source. Understanding the reaction pathways, intermediates formed, and their kinetics is crucial for the logical design of efficient SACs. The general mechanism includes a series of deoxygenation steps followed by hydrogenation steps, involving six intermediates and finally yielding NH₃ and aqueous NH₄⁺. Each step involves proton-coupled electron transfer (PCET), which means that both proton and electron are transferred to the absorbed intermediate to drive the reaction forward^[36].

NO₃^- (aqueous) → *NO₃ → *NO₂ → *NO → *NOH → *N → *NH → *NH₂ → *NH₃ → NH₄⁺ (aqueous)

Usually, first step is the rate-determining step (RDS), but this may vary depending on the SAC, the support materials, the coordination environment of the SAC, and the reaction conditions^[37-39].

SACs can achieve 100% exposure of active sites, representing the theoretical limit of atom utilization, with highly tunable electronic structures, allowing for fine optimization of reaction mechanisms. Different metals offer different strengths for NRA. For example, Cu is the only non-noble metal with a d orbital energy close to the LUMO of nitrates. Cu (I) adsorbs nitrates more strongly than Cu (II). Cu SAC has been reported as the best performing SAC among Ni, Cu, Mn, Pd, Pt, Ru and Ga on nitrogen-doped carbon (NC), with 100% FE and a yield rate of 32,300 μg·h^-1·mg_cat^-1. Fe binds through oxygen, offering high selectivity for ammonia and stability over a wide pH range. Iron is used in NRA because of its natural role in nitrogen fixation to ammonia by the nitrogenase enzymes of nitrogen-fixing bacteria^[40,55]. Cobalt is known for its ability to enhance the N–O bond cleavage, favoring ammonia formation over nitrogen gas. Co is known for its poor HER activity, but doping with noble metals provides sufficient HER happen to promote the reduction of nitrates to ammonia without lowering the FE^[56,57]. Os is known for its low limiting potential of -0.42 V, as calculated by Wang et al. using a four-step screening high-throughput first-principle, highlighting the strong binding of Os with the oxygen of NO₃^- due to rapid charge transfer from Os to NO₃^-[58]. Shang et al. reported a Ru SAC in coordination with Bi for the efficient reduction of NO₂^-, with a FE towards ammonia^[59]. This suggests the use of Ru in DACs for NRA to adsorb intermediates and conduct tandem reduction after NO₂^- formation.

Similarly, the choice of metals in DACs is also important, and DACs can be homonuclear or heteronuclear. In homonuclear DACs, both metal atoms are identical. They offer well-defined symmetry in the active sites, enabling a better understanding of metal-metal interactions and their effects on NRA activity. Mechanistic understanding of homo-DACs is easy because of symmetric charge distribution and uniform active sites, which facilitates the optimization of catalytic activity^[60-62]. Pandiyan and co-worker computationally screened five DACs, all of which showed high FE reaching 100% for Rh-Rh at low limiting potentials ranging from -0.16 to -0.4 V vs. RHE. For the calculations, spin-polarized DFT was run using the Vienna ab initio simulation package (VASP)^[63]. In heteronuclear DACs, each metal overcomes the limitations of the other, and synergistic working optimizes the NRA performance of the catalyst. Heteronuclear DACs are more competitive because of their bifunctional active sites, which may offer a tandem mechanism. Both metals may belong to the 3d series, or one may belong to higher d series; combinations of only higher d series for NRA are less common. Asymmetric electronic distribution and non-uniform active sites with strong d-d interactions generate molecule-like electronic states near Fermi level, giving rise to a donor–acceptor mechanism. This makes the origin of catalytic activity less predictable, and therefore their optimization is comparatively more challenging than that of homo-DACs^[64]. Zhao et al. screened several metal atoms through calculation to identify the best heteroatom to form a DAC with Cu, based on NO₃^- adsorption energy, and free energy changes for OH* + H⁺ + e^- → H₂O^[45]. The results indicated that Fe, Pt, Ir, and Mn are promising. Additionally, it has been reported that Cu–Pd DAC have been applied for NRA, where Cu performs well in oxygen addition but is insufficient in hydrogenation^[47]. Introducing Pd as an active site can enhance the supply of atomic hydrogen H* by promoting water dissociation. Zhang et al. showed that a Pd-Cu DAC exhibits a higher ammonia yield rate than Pd SAC and Cu SAC at lower potentials of -0.1 and -0.2 V vs. reversible hydrogen electrode (RHE)^[65]. In a Fe-Mo DAC, Mo is chosen based upon its activity in nitrogenase, while Fe works to lower the thermodynamics barriers for NRA. Through their synergistic coupling, this catalyst achieves an ammonia yield rate of 13.56 mg·cm^-2·h^-1 and a FE as high as 94%, outperforming the respective SACs of Fe and Mo^[48]. Cu-Co DACs have also been reported as a successful couple in multiple coordination environments, with an ammonia yield of up to 482.56 mmol·g^-1·h^-1, maintaining performance for 150 h without significant activity loss. DACs with Co doping have shown promising DFT-backed results, but experimental verification is still required^[49,50]. Zhang et al. reported a Cu-Fe DAC on nitrogen-doped graphene which achieves a FE of 92.51% at a low potential of just -0.3 V vs. RHE^[51]. The study shows that the dual active sites form by coupling Fe with Cu facilitate the adsorption and discharge of nitrate ions, weakening the N–O bond and lowering the overall energy barrier of NRA. Additionally, Meng et al. further designed the same Fe-Cu DAC supported on NC as a selective catalyst for NRA, delivering a FE of 95% and a yield rate of 6.13 mg·h^-1·mg_cat^-1[66]. This DAC prevents the desorption of NO₂^- and other intermediates, shifting the reaction toward NRA and mitigating the accumulation of intermediates. In SACs, NO₃^- binds to the metal through O, while NO₂^- binds through N; thus, desorption is mandatory for the reaction to proceed. In DACs, when the two metal atoms are in proximity, NO₂^- does not need to desorb, but can simply shift to the adjacent atom, thereby skipping the desorption step and improving the FE^[67]. Liu et al. also prepared an asymmetric Ru-Cu DAC, achieving a FE of 95.7% at 0.4 V vs. RHE^[52]. Both studies have demonstrated the potential of Ru in DACs for NRA. Wei et al. prepared DAC combining Zn with various metals (Co, Fe, Mn, Pd, Pt, Ni, Ru) using a MOF as the precursor^[68]. Among them, the Zn-Co DAC with an asymmetric coordination environment stands out, achieving a FE of 98.95% for NRA at -0.4 V vs. RHE and maintaining this efficiency over seven consecutive cycles. A Ni-Co DAC on MXene has been reported with a FE toward NH₃ higher than Ni and Co SACs, reaching a maximum of 92.46% with minor changes over ten cycles. Here, both metals act synergistically by coupling electron interactions; empty d orbital attract NO₃^-, while occupied ones donate electrons for the reduction of NO_x species^[69]. Therefore, one site facilitates NO₃^- attraction and adsorption, and the other favors reduction of NO_x, leading to NH₃ production even at a low potential of -0.37 V vs. RHE. Feng et al. designed a metal-free DAC with Si and iodine on a Ni-O film, where iodine modified the electronic structure, promoting H* availability for nitrate reduction, while Si stabilized the catalyst, preventing the leaching of Ni from the supporting film^[70]. This opens another potential option for making DACs that are metal-free and more stable than the metal DACs. Shen et al. reported a Cu-WO₃ DAC for NRA, showing that the electron deficient structure of the synthesized DAC promotes nitrate adsorption and facilitates the dissociation of H₂O for supplying H*^[71]. Metal oxides are similar to metals in oxygen coordination environments, where oxygen optimizes the nitration reduction abilities. Ren et al. introduced a new direction by actively participating in support so that the single-atom and a support atom function as DACs^[72]. They anchored a Cu SAC on NiFe layer double hydroxide (LDH) using a special cation vacancy generation method, which modulated the adsorption and surface accumulation of intermediates, as well as the availability of H*. This results in a high FE of 92% and an ammonia yield of 2.08 mg·cm^-2·h^-1. A Cu-Ru DAC has also been reported as one of the highest performing catalysts for NRA, showing maintained activity for 108 h^[52]. It worked in both neutral and alkaline media, giving ammonia yield rates from 35.44 to 125.44 mg·h^-1·cm^-2 with a FE above 90%.

Computational studies, together with the experimental work, showed that the central metal in SAC and the metal combinations in DAC are not sole determining factors for NRA^[73]. The coordination environment, support material, and interaction of metals with these are also of prime importance, as the same metals can exhibit different activities in different coordination environments and on different supports. Os SACs were found to be the best in terms of FE and yield on nitrogen-doped graphene, whereas Ru SAC outperformed when the support material was a graphitic carbon nitride framework. This suggests that the interactions of the metal(s) with the surrounding coordination environment determine the NRA activity of catalysts^[32]. Oxyphilic metals like Cr, Mn, Fe, Ru form O-M-N₄ via oxygenation upon exposure to nitrate ions, while Co, Ni, Cu, Pd, and La sites run associative adsorption of NO₃^- to produce ammonia proceeding through NO₂ intermediates^[74].

The choice of metal in SACs/DACs presents the thermodynamic limits of NRA. Every metal has some limitations in SACs. For instance, Cu d-orbital is close to the LUMO of nitrate, but it delivers incomplete reduction; similarly, Fe binds nitrate so strongly that desorption is tough, leading to catalyst poisoning. DACs are better in this regard, as they use two metals to overcome these limitations, yet other critical gaps remain. Both metals may compete for the same reactants/intermediates, causing them to interfere with each other.

The main challenges for SACs/DACs lie in their stability during electrolysis. Aggregation, clustering, and leaching of metal are prevalent, causing the catalytic activity to decline with increasing cycle numbers. Moreover, all work on SACs/DACs remain at the laboratory scale; to advance them toward practical applications, scalable production and field testing are required. Computer screening is being used in designing SACs/DACs, and the use of artificial intelligence and machine learning is strongly recommended to achieve better designs, provided that all experimental factors are considered in the simulations. Triple metal atom catalysts represent another promising future direction worth exploring, yet their synthesis is more challenging^[75]. Additionally, stimulus responsive SACs/DACs can also be designed to alter their coordination in response to the concentration of intermediates in the medium, thereby thermodynamically and kinetically favoring specific reduction steps^[76].

The coordination environment is equally critical, influencing the electronic structure of active sites, the binding abilities, the stabilization of intermediates, and the overall catalysis mechanism. SACs/DACs provide exemplary precision in optimizing catalytic behavior through atomic level regulation of the coordination environment^[54,77]. In multistep NRA, several key intermediates are involved. They bind to active sites through different atoms, requiring different potentials and energies for each step. This makes the optimum adsorption of all intermediates on the same active site a challenge, that can be met by a delicate balance of the coordination environment. The coordination environment is a multilayer framework that drives the electrolytic performance of SACs and DACs. In addition to atom directly bonded to the SACs/DACs, three other structural features are important: (1) the symmetry and asymmetry of the coordination environment; which controls the distribution of electron density at the active sites; (2) the saturation or unsaturation of the metal coordination; which determines the number of available active sites and the Lewis acidity; (3) the second shell coordinating heteroatom (not directly bonded to SACs/DACs), which operates through long range electronic and electrostatic interactions^[78,79].

The type of first-shell coordinating atom directly regulates the electronic properties and the interaction of SACs with the reaction intermediates.

(i) M-N_x coordination: Many efficient SACs have been reported with central metal directly linked to N and supported on a carbon material. Zhao et al. designed a Cu-N₄ SAC supported on porous carbon, exhibiting a Faradaic efficiency of 87.2% and an ammonia selectivity of 94.1%, both of which are higher than those of Cu nanoparticles on the same substrate^[80]. The nitrogen coordination environment in Cu-N₄ favors the hydrogenation of NO*, leading to high ammonia selectivity over N₂. This N-Metal bonding also stabilizes the SAC, fine-tunes its d-orbital, and improves the absorption of the reaction intermediates.

(ii) M-O_x coordination: Oxygen can also play a role for NRA but it is less commonly used. Sun et al. synthesized Mn-(O-C₂)₄ from bacterial cellulose, which showed a FE of 89% at -0.5 V vs. RHE^[54]. In this catalyst, the oxygen coordination tunes the d orbitals of Mn atom, improving its interaction with the intermediates.

(iii) Mixed Ligand Environments: These involve two different atoms, coordinated to the metals. They further refine the electronic structures and the interactions with nitrate and other intermediates. Chen et al. reported a Fe SAC derived from ZIF-67 with sulfur coordination^[81]. The sulfur modification makes the coordination asymmetric, enhancing the adsorption of NO₃^-, and achieving a FE of 93.9% at a lower potential of -0.47 V vs. RHE. This demonstrates how different bonding environments can impact the initial nitrate activation. Phosphorus bonded to the metals in CoCu DAC plays a critical role as part of the coordinating environment. P changes the electronic density and facilitates charge transfer. It also helps to stabilize the intermediates that would otherwise poison the catalyst by blocking active sites. This DAC showed an ammonia yield of 1.59 mmol·h^-1 and a FE of 95.36% at -0.3 V vs. RHE^[50]. Li et al. designed an enzyme mimic DAC by anchoring Cu atoms on polymeric carbon nitride (PCN)^[82]. PCN provides an enzyme-like coordination environment, making the Cu centers more positive, so that they efficiently attract NO₃^- ions. Consequently, this DAC showed the highest current in linear sweep voltammetry (LSV) and the highest FE among bare PCN, SAC PCN, and Cu- nanoparticles on PCN, as shown in Figure 4A and B.

Defects in the coordination environment promote nitrate reduction. Unsaturated sites favor the adsorption of NO₃^- and overall activity, the same holds true for an asymmetric coordination environment. Understanding how these defects interact: symmetric vs. asymmetric geometries, saturated vs. unsaturated coordination, and first vs. second-shell engineering enables the rational design of catalysts to achieve high FE and ammonia yield rates.

Researchers compared Cu SACs in different coordination symmetries with nitrogen and oxygen, like Cu-N₄, Cu-N₃, Cu-2N₂O (cis and trans), through simulation-based studies for NRA. A study of radial distribution functions (RDFs) showed that cis Cu-2N₂O exhibits the highest FE and absorbs the largest amount of nitrate ions on its surface^[32]. The reason is that the cis geometry gives rise to local asymmetry, which in turn causes polarity in the structure. This polar structure attracts nitrate ions from electrolytes more strongly than other coordinating environments.

Wan et al. reported a Fe-N₄ SAC with axially coordinated Cl^[83]. The axially attached chlorine atom modifies the electronic structure on Fe, stabilizing charge fluctuations at the active site. It increases the adsorption of NO₃^- and key intermediates, leading to a high FE of 99.4% at exceptionally low potential of -0.28 V vs. RHE, and showed a high ammonia yield in comparison to symmetric Fe-N₄, as shown in Figure 4C. Xue et al. compared a Cu SAC with N₄ and N₃C₁ coordination environment^[84]. This work explains how breaking coordination symmetry creates spatial heterogeneity. The N₃C₁ configuration in Cu-N₃C₁ concentrates the charge density near Cu, leading to the adsorption of NO₃^- on Cu, while H* is absorbed on the nearby C. This adsorption behavior lowers the energy and makes the reduction reaction thermodynamically more favorable. This catalyst delivers a maximum FE of 94.8% and a stability of up to 84 h, both outperforming the symmetric Cu-N₄. Liu et al. prepared Cu SACs with different numbers of coordinating nitrogen: Cu-N₄, Cu-N₃, and Cu-N₂, and showed that Cu-N₄ better stabilizes the Cu active site, while Cu-N₂ is good at absorbing NO₂^-, thus preventing the accumulation of intermediates^[85]. Hence, saturated and unsaturated coordination environments have different impacts on NRA activity.

The NRA activity of DACs is also affected by the coordination environment in similar patterns, with greater asymmetry and coupling caused by two metal atom sites. Lv et al. reported a Pd-Cu DAC embedded in asymmetric C₃N₄, the porous structure facilitates the incorporation of the metal atoms and lowers the Gibbs free energy and the limiting potential of the catalyst^[15]. Cu-Co DACs with FLPs on NC showed relay kinetics and a push-pull mechanism for the adsorption and stabilization of NO₃^- and NO₂^-, forming Cu-O-N-O-N and Co-O-N-O structures in a nitrogen rich environment^[49]. Wei et al. reported another Zn-Co DAC in an asymmetrical coordinated environment^[68]. The metal-bonded nitrogen in the N₃Zn–CoN₂ configuration enhances the electronic coupling between the metal active sites, optimizes the adsorption of key intermediates especially NOH*, and reduces the energy required for the reduction reaction, thereby lowering the activation energy barrier.

Lian et al. reported an AgCu‐C₃N₄ DAC, in which Cu acts as a Lewis acid through coordination with nitrogen to adsorb NO₃^- ions, while Ag, as a noble metal, promotes the supply of H*^[86]. Under visible light irradiation, this DAC achieves an ammonia yield as high as 630.5 µmol·h^-1·g^-1. Yin et al. designed an N₂S₁Cu-RuN₃/SNC DAC featuring an asymmetrical coordinating environment and showing a FE of 98% at -0.6 V vs. RHE^[87]. Sulfur plays a crucial role in modulating the electronic structure, facilitating intermediate adsorption and lowering the energy barrier. As shown in Figure 4D, the sulfur modified DAC shows higher FE and NH₃ yield than the same DAC without S.

Critical analysis of the coordination environment shows a significant gap between theoretical predictions and experimental results for SACs and DACs. Breaking symmetry is regarded as a strategy to increase activity, but there is no solid explanation confirming whether this enhancement arises from a change in coordination number or can be attributed to a change in the oxidation state of the metal. It demands more systematic studies and future research. Furthermore, current characterization methods cannot determine whether asymmetry, unsaturation, and other features of the coordination environment are merely spectators or real participants responsible for the activity. Without time-resolved operando techniques with atomic resolution to capture coordination changes and the binding of intermediates during catalysis, the coordination-performance relationship remains more hypothetical than an established principle.

Second-shell heteroatom doping involves incorporating non-metal elements into the support material or the vicinity of active sites. These dopants alter the electronic structure of SACs modifying its catalytic performance, but they are not directly bonded to single metal atom. This approach is a transformative method for optimizing NRA activity. The dopant atoms are usually anchored on the support material (mostly carbon based), from where they modulate electronic structure by redistributing charge, strain engineering, and long-range electronic effects.

Chen et al. reported a P-doped Cu SAC in a nitrogen coordination environment; it lowered the energy barrier for the NO* to NOH* conversion, driving selectivity toward ammonia and increasing production rates^[81]. P induces polarization in the structure, which strengths NO₃^- binding and drives H* supply, and an overall increase in current is observed^[88]. Iodine as a dopant is known to alter adsorption preferences, favoring NO₃^- over H₂O/H*. Hence it suppresses HER competition and boosts ammonia yield^[89]. Cai et al. reported an iodine doped Co SAC, formulated as CoN₄, which showed an ammonia yield rate of 18.64 mg·h^-1·mg_cat^-1 and high selectivity for NH₃ over H₂^[90].

Doping in DACs is more complicated because of the interplay and constructive interaction of the two metal atoms, both of which may have different effects on the electronic structures of the dopants. Researchers reported boron doping on Ni-Co DAC; the boron favors the anchoring of Ni and Co on the carbon support and prevents the aggregation of metal atoms into clusters, which would otherwise decrease catalyst activity. Consequently, with B doping, a high NH₃ yield of 0.87 mmol·h^-1·cm^-2 at -0.3 V is achieved^[91]. Selenium has also been found useful in enhancing the adsorption of NO* on Fe SAC, so it can be used for NRA as well, either alone or with another dopant^[92]. The effect of dopants on NRA activity shows they are not merely a structural feature but an opportunity to fine-tune SACs/DACs for better performance.

Following the detailed discussion of SACs/DACs structural features, this section briefly examines synthesis methods, highlighting their pivotal role in achieving these specific structural features. While the synthesis techniques of SACs and DACs have been extensively reviewed by numerous researchers (Xiong et al.^[93], Xiang et al.^[37], Chao et al.^[94] and Xu et al.^[95]), the correlation between synthesis methods, structural features and NRA activity remained an underexplored area. This section provides an overview of the primary synthesis techniques, evaluating their influence on the resulting catalyst structure and their key role in NRA performance of SACs/DACs.

DFT and computational studies are indispensable in SACs/DACs research. Researchers employ DFT to support and elucidate experimental work and mechanisms, as well as to predict active metals under specific conditions.

Xue et al. used the VASP package along with the Perdew-Burke-Ernzerhof (PBE) exchange correlation functional for DFT calculations^[100]. Bader analysis was used to quantify charge, and thermal stability was assessed by ab initio molecular dynamics. They identified a PdCu catalyst that is effective in preventing HER and selective for NH₃.

Lv et al. studied catalysts under a wide pH range using the CASTEP software package^[15]. Projected density of states (PDOS) and surface Pourbaix diagrams were utilized to investigate electronic properties. Simulations predicted TiFe as the best-performing DAC among thirty-six combinations on the g-C₃N₄ surface. This provided a guide to identify optimal metal combinations for DACs.

Zhao et al. screened 80 transition metals using spin-polarized DFT and climbing-image nudged elastic band (CI-NEB) method^[60]. A three-step screening approach was employed to study energy barriers, and Cu homonuclear dual atom on N-doped graphene was demonstrated to be the optimal candidate catalyst for NRA. Experimental studies validated the theoretical results, showing 97.4% FE for the catalyst at low potential of -0.14 V vs. RHE. This method can be employed to avoid trial-and-error preparation of SACs and DACs for NRA. Furthermore, Zhao et al. also used spin-polarized DFT to study dual heteroatom doping on CoP SAC, and performed ab initio molecular dynamics simulations at 300 K, thus revealing the bidentate bonding mode of intermediates and the bidirectional current flow initiated by the double heteroatom, which is expected to lay the foundation for the theoretical regulation of the microenvironment^[67].

Wang et al. investigated 300 heteronuclear DACs by employing high-throughput first-principles calculations and a hierarchical four-step screening method^[64]. To simulate varying electrochemical conditions, a constant-potential model was utilized. The results indicated that Cr-Rh and Mn-Rh exhibit high selectivity and activity at extremely low potentials of -0.16 and -0.25 V, respectively. Furthermore, it is confirmed that the orbital interaction of two metals produces activity under an association-dissociation mechanism. Wang et al. also employed DFT simulations along with PDOS analysis^[101]. By coupling these calculations with in-situ Fourier-transform infrared spectroscopy and electron paramagnetic resonance reaction intermediates were identified. The experimental work verified computational results by exhibiting a high FE of 98% and successful implementation in Zn-NO₃ batteries. This method can be employed for all similar catalysts to get an in-depth, simulations-based understanding of the mechanism.

Li et al. used DFT to study solvation effects and dispersion correction to efficiently predict catalyst activity^[62]. DMol3 spin-unrestricted DFT-Sol and DFT-D were used for these purposes, respectively. An Os dimer was found to be the most promising catalyst on the CN surface at low potential of -0.15 V vs. RHE. Considering solvent effects gives a more refined value for limiting potential.

DFT has become essential in SACs/DAC research, and many simulation packages are available. However, no single simulation suite can study all aspects of a catalyst. Furthermore, the computational cost of such studies is remarkably high, taking hours to weeks. This also necessitates specialized experts; a wet chemistry scientist or material engineer cannot operate this complex software without extensive learning and training. Machine learning approaches are increasingly necessary. Well-trained machine learning models with user-friendly interfaces are highly in demand. Data scientists, computational chemists, and material experts need to collaborate to develop models that can be easily employed by lab scientists for prediction and mechanistic studies without requiring in-depth computational expertise.

The greatest challenge lies in the precise preparation of SACs and DACs, with meticulous control over the coordination environment, dopant positioning, asymmetry, and unsaturation^[114]. During catalysis process, the coordination environment is prone to change, which may cause migration, cluster or reorganization of ligands. Unsaturated sites may be saturated by adsorbing side products or intermediates. Furthermore, asymmetry and unsaturation in the coordinating environments, as well as specific heteroatom arrangement, are thermodynamically sensitive and less stable compared to saturated and symmetric systems. During electrolysis, the system tends to move toward higher stability at the expense of activity, resulting in a less active but more stable configuration. For instance, heteroatoms may migrate to energetically favorable positions, ligands may move to create a symmetrical coordination environment, or electrolyte species may bond on unsaturated sites^[85].

Furthermore, it is difficult to determine whether the key tuning parameter for catalytic activity is the metal atom, the first-shell coordination geometry, the second-shell heteroatom, or even the reaction medium, including its pH and operating potential. Since tuning effects vary significantly and may even be contradictory across different cases, the contribution of all these features to the mechanism needs detailed discussion and evaluation to rationally optimize performance and design SACs/DACs. If a heteroatom or vacancy concentration increases the activity, its optimum level and the interplay of other structural features need to be determined. It remains unexplored how different heteroatoms and coordination environments reinforce each other or act neutrally for NRA. This multilayered mapping of all parameters demands further studies^[115,116].

Sometimes, during DAC preparation, one or two types of SACs are also formed. It is challenging to distinguish the NRA activity of SACs and DACs present on the same support. The mechanistic understanding of SACs/DACs is highly complicated; the intermediates are short-lived and often cannot be separated. Researchers rely on DFT, assisted by limited experimental data, to propose mechanisms. To date, DFT and computational screening do not offer a complete alternative of experimental work. Therefore, improved simulations and programs are needed that can consider not only structure of SACs/DACs but also all reaction parameters, such as pH, concentration of electrolyte, composition. In-situ and operando techniques are used for quantifications and mechanistic studies; however, these are extremely expensive and not universally available, thereby restricting research due to cost factors^[62,103].

The absence of a standard framework for testing and reporting protocols severely restricts comparative studies and technology transfer. SACs/DACs are mostly tested under laboratory conditions. Their real-world field application and scaling present another significant challenge for researchers. A comprehensive cost and sustainability analysis is required to compare NRA using SACs/DACs with other technologies, assessing their real-world applicability and future potential.

SACs/DACs possess numerous structural details; each structural feature can be designed for specific steps of NRA. This systematic design approach will rationally optimize activity. For this purpose, a robust database needs to be designed, or machine learning models can be trained to correlate these structural features with reaction mechanisms, intermediate binding energies, and kinetics. Such models should be verified by experimental work. Utilizing such a trained and verified model, novel and unique combinations of metals, coordinating atoms, and dopants can be easily screened, and synergistic effects can be studied without the need for tedious synthesis. Machine learning can also be applied to inverse catalyst design. Instead of synthesizing SACs/DACs with features and then evaluating its performance computationally or experimentally, the focus should be on the potential of each step and the binding energies of intermediate. Subsequently, researchers can screen metal atoms and coordinating systems that can deliver the required features^[117,118].

It is highly recommended to build universal metrics that quantify the extent of asymmetry, degree of unsaturation, and second shell coupling. This can be achieved by studying the dipole moment and symmetry order, the coordination number and Lewis acidity, and the strain energy, respectively. Such descriptors would make performance comparison possible across diverse catalyst families, enabling applications of design rules in general instead of specific systems.

Bioinspired catalysts for NRA are receiving growing attention, but further advances will require the incorporation of nitrate-reductase like coordination environments into synthetic catalysts and their optimization under realistic experimental conditions^[82]. Similar to SACs and DACs, triple atom catalysts are also emerging, though they remain rare. There is a need to explore these for NRA and compare them with SACs/DACs. Multiple second shell dopants can also be incorporated to get various advantages^[119].

Time-resolved X-ray spectroscopy with specific element detection can capture changes in first shell coordinated atoms and second shell dopants by utilizing extended X-ray absorption fine structure at both the metal K-edge and the dopant K-edge, respectively. If researchers combine this with electrochemical measurements using a suitable interface, all structural changes can be tracked during catalysis^[120]. Material scientists, analytical chemists, and engineers need to work together in this direction.

The future lies in responsive materials. Instead of static structures, scientists should aim to design SACs/DACs that are responsive to pH, the concentration of specific species, applied potentials and temperature. Thus, variations in these factors can change the coordination environment of metal atoms, allowing them to adsorb intermediates accordingly at low energies^[76]. Furthermore, catalyst activity is often diffusion limited. To overcome this, autonomous self-propelled micro- and nanorobots are gaining popularity. SACs/DACs can be converted into nanorobots to self-propel in areas of high nitrate concentration and carry out NRA effectively^[121].

A further outlook includes the practical implementation of SACs/DACs for NRA on an industrial and commercial scale, through integration with membrane electrode assemblies (MEAs). Shen et al. designed an NRA continuous flow electrolysis cell integrated with a membrane separator^[71]. This achieved an ammonia production rate of 325.9 mg_N·h^-1·g_Cu^-1 with 98.3% collection efficiency at an energy consumption of 17.11 kWh·g_N^-1. This work highlights that coupling SACs/DACs with MEA is highly feasible and demonstrates potential for scalability.

During catalysis, structure reconstruction takes place, which decreases the activity of SACs/DACs. There is a need for in-operando regeneration methods to reverse these reconstructions. For instance, oxidative pulses can regenerate vacancies, and a reducing environment can relocate dopants. Such maintenance could sustain SACs/DACs over longer time periods. For example, Cu and Ti oxide-based nanofibers were designed to undergo in-situ electrochemical reduction to stabilize Cu⁺ active sites. This increased catalyst stability over 50 cycles with nitrate removal efficiency exceeding 90%^[122]. Such self-healing materials need to be explored for SACs/DACs as they can undergo reversible structural changes, thereby enabling in-situ modulation of intermediate adsorption and nitrate reduction kinetics under variable operating conditions. A mind-map of current challenges and future goals have been elaborated in Figure 7.

This review summarizes and critically discusses recent research on SACs and DACs for NRA, offering a parallel comparison of these catalysts. It focuses on the choice of metals and the coordination environment variables, and on how these factors tune the electronic structure, reaction mechanism, intermediate stabilization, and selectivity, activity, and stability of catalyst. This review also elaborates on how metal-metal synergistic effects in DACs, along with the complicated interplay of coordination asymmetry, unsaturation, heteroatom dopants, and experimental operators like potential and pH, control catalytic performance. The computational studies, software packages, and simulations used, as well as the future directions of DFT and machine learning in SACs/DACs, are also discussed. Furthermore, challenges in the field are highlighted, and viable solutions along with a clear roadmap for future research are critically suggested to advance work in this field.