Photocatalytic nitrogen reduction to ammonia: Insights into the role of defect engineering in photocatalysts

Engineering of defects in semiconductors provides an effective protocol for improving photocatalytic N2 conversion efficiency. This review focuses on the state-of-the-art progress in defect engineering of photocatalysts for the N2 reduction toward ammonia. The basic principles and mechanisms of thermal catalyzed and photon-induced N2 reduction are first concisely recapped, including relevant properties of the N2 molecule, reaction pathways, and NH3 quantification methods. Subsequently, defect classification, synthesis strategies, and identification techniques are compendiously summarized. Advances of in situ characterization techniques for monitoring defect state during the N2 reduction process are also described. Especially, various surface defect strategies and their critical roles in improving the N2 photoreduction performance are highlighted, including surface vacancies (i.e., anionic vacancies and cationic vacancies), heteroatom doping (i.e., metal element doping and nonmetal element doping), and atomically defined surface sites. Finally, future opportunities and challenges as well as perspectives on further development of defect-engineered photocatalysts for the nitrogen reduction to ammonia are presented. It is expected that this review can provide a profound guidance for more specialized design of defect-engineered catalysts with high activity and stability for nitrogen photochemical fixation.


Introduction
Ammonia is an indispensable raw material which is widely applied in the production of agricultural fertilizers, industrial and household chemicals [1][2][3][4]. Use of ammonia to produce fertilizers quadrupled crop yield and global population [2]. In addition, ammonia is also proposed as a potential hydrogen carrier and distribution medium of the future, because it possesses high hydrogen content (17.6 wt.%), large energy density (4.3 kWh·L −1 at −33.3 °C and 1 bar, 6.25 kWh·kg −1 ), and COx-free emissions [5][6][7][8]. Equally importantly, ammonia is readily liquified (−33 °C) at atmospheric pressure [9,10]. The technology of liquefaction, storage, and pipeline transport of ammonia has been well achieved in existing industries compared to liquid hydrogen [11].
In nature, nitrogen from the atmosphere is transformed into ammonia via nitrogenase enzymes under mild conditions (< 40 °C, atmospheric pressure), called biological nitrogen fixation [12]. The most common nitrogenases enzymes mainly contain FeMo nitrogenases, which are composed of two components including Fe proteins as electron-transfer media and FeMo proteins as N2-binding and reduction active sites [13][14][15]. Unfortunately, enzyme nitrogenases are susceptible to oxygen [16,17]. Additionally, biological nitrogen fixation to ammonia has low space-time yield and hardly meets the demands of modern societies with rapidly growing population worldwide [18,19]. One of the greatest scientific achievements in the 20 th century was accredited to the discovery and implementation of industrial ammonia synthesis, i.e., the Haber-Bosch process named according to its primary inventors Fritz Haber and Carl Bosch [20,21]. The Haber-Bosch process involves the reaction of nitrogen and hydrogen on iron-based catalysts to synthesize ammonia [4,22]. The heterogeneous catalyst, already developed in 1910 by Alwin Mittasch, is still used today with only minor alterations [21,23]. However, the classical Haber-Bosch process takes place under very harsh reaction conditions (400-500 °C, [15][16][17][18][19][20][21][22][23][24][25]. Annually, nearly 200 million tons of ammonia are synthesized through the Haber process. Apart from nitrogen, hydrogen is used as a feedstock which is usually obtained from methane steam reforming, and as a result, industrial ammonia synthesis consumes about 1%-3% of the world's total energy and leads to more than 300 million tons of carbon dioxide emissions [20,24,25]. From these scenarios, it is highly desirable to explore and develop an environmentally-friendly, ambient, and sustainable N2 fixation approach. To date, various alternative strategies have been exploited for nitrogen fixation under mild conditions (< 300 °C, < 1 MPa), including biomimetic, thermo-catalytic, plasma-catalytic, photocatalytic, electrocatalytic, and chemical looping methods [26][27][28][29][30][31][32]. Among these routes, photocatalytic nitrogen reduction into ammonia using water as a coreactant instead of hydrogen (Overall reaction: N2(g) + 3H2O(l) → 2NH3(g) + 3/2O2(g)), mimicking natural photo-synthesis, shows potential for clean and sustainable NH3 fabrication [33,34]. It solely requires inexhaustible solar energy, water as proton source, and N2 reactant [35,36]. Photocatalytic NH3 synthesis correlates with two coupled redox half reactions, including the oxidation of water (3H2O(l) + 6h + → 6H + (aq) + 3/2O2(g)) with photogenerated holes in the valence band (VB) and the reduction of nitrogen (N2(g) + 6H + (aq) + 6e -→ 2NH3(g)) with photogenerated electrons in the conduction band (CB) [37][38][39]. Pioneering work for nitrogen fixation was reported by Schrauzer et al. on TiO2-based photocatalysts under ultraviolet (UV) light in 1977 [40]. Since then, significant research efforts have been made to optimize the photocatalytic performance for N2 fixation [41]. The nature of the photocatalyst plays a vital role in photo-catalyzing N2 reduction. Diversified photocatalysts have been explored for N2 reduction reaction, such as metal oxides [42][43][44][45], metal sulfides [46,47], g-C3N4 [48,49], and MXene (Ti3C2) [50,51], among others. Nonetheless, the overall solar-to-chemical conversion efficiency for N2 reduction is still as low as 0.1% which is far from the value required in industry or even any reasonable technological interest [2,[52][53][54]. This is essentially associated with the issues of low light utilization, lack of effective active sites, rapid recombination of photoexcited electron-hole pairs, in addition to the inertness and stability of the nonpolar N≡N triple bond and poor adsorption/activation capability [55,56]. To address these challenges, several photocatalyst modification strategies have been explored to improve the performance of photocatalytic nitrogen fixation [57].
Defect engineering is a simple, useful, and appealing approach to enhance photocatalytic N2 fixation by altering the electronic structure and chemical properties of semiconductors [58][59][60][61][62][63][64]. Generally, knowledge-driven and well-planned defect engineering not only improves light absorption ability and accelerates charge carrier separation, but also provides active sites for N2 adsorption and activation [65][66][67]. Furthermore, introduction of defects on the surface of photocatalysts can endow an electron-rich state and high surface energy, thereby enriching the lowest unoccupied molecular orbital (LUMO) electron density of N2 via e − → π*-orbital (N) transition and favoring N2 adsorption and activation [39]. To elucidate the relationship between the structure of defects and catalytic performance, major endeavors have been attempted by combining experiments and theoretical calculations [38,[68][69][70]. Despite recent progress that has been achieved in this regard, many challenges remain to be solved including precise quantification, fine structure tuning, and stability enhancement of defects [71,72].
Herein, we provide an up-to-date review on defect engineering over semiconductor-based photocatalysts for the nitrogen reduction towards ammonia. Fundamentals such as the properties of N2 molecules, photoreduction principles of N2, reaction pathways, and NH3 measurement methods are initially elaborated. In the following, the defect classification, synthesis strategies, and identification techniques are reviewed. Especially, the critical roles of defect engineering on photocatalytic nitrogen reduction are discussed. Additionally, recent achievements of defect-engineered photocatalysts for nitrogen reduction to ammonia are presented. Finally, we provide perspectives on future opportunities and challenges on defect-engineering of photocatalysts to promote the nitrogen reduction to ammonia. We expect that this review can provide profound guidance for more specialized design of defect-engineered photocatalysts with high activity, stability, and selectivity for ammonia synthesis.

Fundamental properties of N2 molecules
It is generally known that nitrogen molecules are formed by two nitrogen atoms connected by a strong nonpolar N≡N triple bond. Each atom possesses a pair of electrons in the 2s orbital with opposite spin direction and three lone-pair electrons dispersed in the 2p orbitals with the same spin direction [41]. Hybridization of the s-p atomic orbitals leads to the formation of four bonding orbitals (two σ and two π orbitals) and four antibonding orbitals (two σ* and two π* orbitals), with the shared electrons in the π and 2σ orbitals forming an N≡N bond [73]. The large energy gap between the highest occupied molecular orbital (HOMO) and LUMO is 10.82 eV, which severely impedes electron transfer [74]. The N2 molecule is extremely stable and kinetically inert with a high N≡N cleavage energy (945 kJ·mol −1 ) and first-bond breaking energy (410 kJ·mol −1 ) [37]. Among others, an N2 molecule has both large ionization potential (15.85 eV) and negative electron affinity (−1.9 eV), making it difficult to be oxidized or reduced [41]. Indeed, activating N2 molecules at ambient conditions is a formidable challenge.

Conventional heterogeneous catalysts for ammonia synthesis
Although the classical heterogeneous ammonia synthesis process is named as the Haber-Bosch process, the development of the catalyst was primarily the achievement of Alwin Mittasch at BASF, who tested in the early 20 th century more than 2,000 different catalyst formulations in more than 6,000 experiments [75]. It is even more remarkable that the multiply promoted iron catalyst resulting from Mittasch's experiments is still used today with only minor alterations. The composition and nanostructure of this catalyst have been described most conclusively by R. Schlögl [21,22]. Early on, it was already established that the catalyst would only show high activity, if a particular starting mineral was used, or if the compositions of this mineral was artificially mixed. Only when magnetite (Fe3O4) was promoted with defined amounts of K2O, Al2O3, and potentially CaO in an oxide melt, a particular active structure would form under reaction conditions, which is termed as "ammonia iron". Iron can easily form nitrides in presence of ammonia, and the formation of different surface nitrides has been proposed. The major bulk part of the iron, however, remains in the likely defective α-Fe phase. In conclusion, the microstructure of the industrial catalyst, the "ammonia iron", is highly complex and defective, being likely composed of bulk iron particles, kept apart from each other by the irreducible promoter oxides, and surrounded by partially nitride defective iron platelets [21,22]. This structure is schematically depicted in Fig. 1(a).
Nitrogen dissociation is the rate-determining step in ammonia synthesis over iron [21,22]. N2 is first adsorbed molecularly in a precursor state before dissociation occurs. From surface science experiments with model single crystal surfaces, important influences of the potassium promoter on accelerating nitrogen dissociation and ammonia desorption have been identified [21,76]. After N2 has dissociated into nitrogen atoms, it is successively hydrogenated, forming adsorbed *NH, *NH2, and *NH3, before ammonia desorption occurs [21].
The only other industrially used catalyst for ammonia synthesis is based on ruthenium. The high price of Ru prohibits the use of a bulk material, so Ru is used in nanoparticulate form on oxide or carbon supports [75]. It may also be beneficial for catalytic reasons to use Ru nanoparticles instead of larger structures, because such "rough" particles may exhibit more active sites [21]. For Ru-based catalysts, nitrogen dissociation is also the rate-determining step, but it proceeds without a molecular precursor state [21,77]. It is by now well known that nitrogen activation critically depends on the exposure of surface steps ( Fig. 1(b)) [21,78]. Due to the better stabilization of adsorbed N2 at the step compared to the bare terrace, nitrogen dissociation can proceed nine orders of magnitude faster at the step site compared to the bare terrace. This has been verified by poisoning the steps sites by gold deposition [78]. The interaction of hydrogen with Ru is much stronger, but desorption limitation of ammonia is less of an issue than in case of iron. Except for the lack of a molecularly adsorbed nitrogen precursor state, the proposed reaction mechanism on Ru is similar to the mechanism on Fe, although the rate of the individual elementary steps differs. Since hydrogen adsorbs very strongly on Ru, it is beneficial to use a synthesis gas with nitrogen excess rather than a stoichiometric one [77].
It is interesting to observe the commonalities of the two systems: In both cases, nitrogen dissociation is the ratedetermining step, and only a small fraction of all exposed sites can efficiently dissociate nitrogen [21,78,79]. Alkali promoters are used in both cases in order to optimize adsorption energetics of reactants and products [21,75]. And finally, highly defective structures with steps (Ru) and partial nitridation (Fe) are needed in order to form the active sites. Guiding design principles may be derived here which may also be relevant for the photocatalytic process: Even if reaction pathways differ (see below), the photocatalyst eventually must also dissociate N2, and NH3 desorption at room temperature must be feasible. Knowing that N2 dissociation might be more easily accomplished at step sites or partially nitrided surfaces, for example, is then very valuable.

Basic principles of N2 photoreduction to NH3
According to energy band theory, semiconductor materials possess a CB and a VB. The potential difference between the CB and VB is referred to as bandgap energy (Eg) (Fig. 2(a)) [80]. Photocatalytic N2 reduction to NH3 proceeds through three fundamental steps ( Fig. 2(b)): (1) generation of charge carriers (i.e., electrons and holes) by photon excitation (Ehv ≥ Eg), (2) separation and migration of electrons and holes to the catalyst surface, (3) redox reactions between surface-adsorbed species and electron-hole pairs, i.e., N2 reduction to produce NH3 and water oxidation to O2 and H + [81][82][83]. The capability of a semiconductor to absorb light and its photoreaction thermodynamics are dependent on its bandgap and the CB/VB potential value [84]. Therefore, in designing photocatalysts, it is essential to consider the redox potentials of photoexcited conduction band electrons and valence band holes to satisfy corresponding N2 reduction and water oxidation.
The photocatalytic reduction of nitrogen to ammonia generally involves a dissociative mechanism and an associative mechanism ( Fig. 3) [88]. For the dissociative mechanism ( Fig. 3(a)), the N≡N triple bond is split into two nitrogen atoms followed by subsequent hydrogenation of individual nitrogen atoms to form NH3. This mechanism resembles the classical Haber-Bosch process [89,90]. It is known that breaking the N≡N triple bond (945 kJ·mol −1 ) requires extremely high energy inputs, so only very limited photocatalysts can successfully drive this reaction process. If experiments indicate that this mechanism is the dominant one for a certain type of photocatalyst, the design principles from thermal catalysis outlined above might be worth to try for improvement of the catalytic function. In the associative mechanism, the hydrogenation of adsorbed N2 molecules occurs without cleavage of the N≡N bond, analogous to biological nitrogen fixation. It is regarded as the dominated mechanism for photocatalytic N2 conversion to NH3. Two possible pathways i.e., distal and alternating pathways, may be involved, leading to distinct intermediates [91,92]. For the distal pathway, continuous protonation is carried out on the nitrogen atom farthest from the surface of the catalyst to generate an NH3, which is then released, leaving another nitrogen atom on the surface for further hydrogenation (Fig. 3(b)). In regard to the alternating pathway, two nitrogen atoms are hydrogenated alternately to yield N≡NH*, NH=NH*, NH-NH2*, NH2-NH2*, and finally to NH3 (Fig. 3(c)).

Oxidation half-reaction and effect of O2/CO2 on N2 reduction
Photocatalytic N2 reduction to NH3 comprises two coupled redox half reactions, and the overall reaction entails transfer of six electrons. However, most studies only focused on the reduction reaction, the oxidation half-reaction is usually ignored. The photocatalytic redox reaction is determined by the relative position of the reduction potential of the reactants and the energy band structure of the semiconductor. From the perspective of thermodynamics, if the VB potential of a semiconductor is higher than the oxidation potential of water, it is thermodynamically feasible for photo-generated holes to react with water to generate O2, or possibly even strongly oxidizing hydroxyl radicals ·OH (2H2O + 4h + → O2 + 4H + , E 0 redox = 0.81 V vs. NHE at pH = 7 or H2O + h + → ·OH + H + , E 0 redox = 2.32 V vs. NHE at pH = 7) [93]. The impact of O2 on the photoreduction of N2 is still open for further exploration. Recently, Hirakawa and co-workers demonstrated that bubbling with air could suppress NH3 formation over TiO2 [67]. This was supposed to arise from the easier reduction of O2 compared to N2 using the electrons in the CB (O2 + e − → O2 − , E 0 = −0.137 V vs. NHE at pH = 7) [52,93]. Furthermore, the ·OH originating from self-oxidation of H2O could oxidize the photo-formed NH3 to yield nitrite or nitrate. Consequently, the O2 or ·OH resulting from the oxidation half-reaction is unfavorable for the reduction of N2.
Some semiconductors exhibit poor ability for water oxidation, hence a hole sacrificial reagent (i.e., Na2SO3, amines, alcohols, and ethylene diamine tetraacetic acid) is always needed to consume accumulated holes and improve the N2 conversion to NH3 [94]. Methanol is demonstrated to be more suitable and effective than the other hole sacrificial reagents, because it loses electrons more easily owing to its lower HOMO [95]. The impact of CO2 on the photoreduction of N2 also requires further investigation. It was recently reported that methanol added during a photocatalytic process could be easily oxidized to form formic acid and CO2 by photogenerated holes (h + ) on Figure 3 Proposed mechanisms for N2 reduction to produce ammonia [88].
Au/HCNS-NV (Au nanoparticle-embedded hollow mesoporous carbon nitride spheres with abundant nitrogen vacancy) [96]. CO2 was demonstrated to be readily adsorbed on the surface of g-C3N4 nanosheets to generate CO2 •intermediates, which accelerated the N2 photoreduction to NH3 owing to its strong reducing capability (ECO2/CO2•− = 1.8 V) (5N2 + 2CO2 •-+ 4H2O → 2NH3 + 2CO2 + 2OH -) [97,98]. However, the introduction of hole sacrificial reagents is not a sustainable process, and it may also interfere subsequent NH3 quantification and cause photocatalyst corrosion [99,100]. In this regard, the use of hole sacrificial reagents is a double-edged sword and should be carefully considered.

Photon-driven disproportionation reaction of N2
As described in section 2.5, the N2 reduction to NH3 involves two coupled redox half reactions and disproportionation reaction may concurrently take place (overall reaction: 4N2 + 9H2O → 5NH3 + 3HNO3), leading to the formation of different products (such as N2H4, NO3 -) [101,102]. The selectivity of N2 reduction to NH3 is substantially reduced as a result of the disproportionation reaction. It is thus desirable to identify the oxidation products, especially when using water as a proton source, because the oxidation half reaction may generate oxidizing substances, such as HClO, ·OH, H2O2, or others, which may oxidize ammonia to nitrate [103]. For example, the by-product nitrate (NO3 -) was observed during photocatalytic N2 reduction in pure water over W18O49 nanowires rich in oxygen vacancies (OVs) under simulated solar light from 427 to 515 nm [104]. The production of NH4 + and NO3resulting from the disproportionation reaction severely limits the N2 photo-reduction efficiency and selectivity. Note that, the exact mechanism on the disproportionation of N2 remains elusive thus far. Alternatively, Fe-doped TiO2 microspheres with OVs were also found to disproportionate nitrogen into nitrate under photon illumination [105]. It was speculated that Fe doping introduced energy levels in the TiO2 band gap, thus altering the redox ability of photogenerated carriers [106], which favored the formation of H2O2 from H2O adsorbed on the catalyst surface. The formed H2O2 then oxidized NH3 into nitrate. Using catalysts such as plasmonic Au nanoparticles that are effective for H2O2 decomposition can combat the further oxidation of ammonia in favor of the reductive reaction [107]. Consequently, elemental doping or/and plasmon modification appears a method of choice to suppress photon-driven disproportionation reaction of N2 and improve the selectivity towards NH3.
To inhibit the oxidation of N2 reduction products and improve the NH3 selectivity, a three-phase suspension system can be applied for the N2 photoreduction [27]. Specifically, the semiconductor material is suspended at the gas-liquid interface towing to the presence of surface tension, while N2 gas is continuously introduced to the water/photocatalyst interface. The N2 concentration in the gas phase above the photocatalyst can be about 140 times that of water saturation, so the N2 concentration on the photocatalyst surface of the three-phase system is dramatically higher than in the two-phase system, thus remarkably improving the NH3 production rate [27]. Perhaps more importantly, O2 and NH3 are separated by the gas phase and the water phase on both sides of the photocatalyst in the three-phase system. This can effectively suppress the oxidation of NH3.

Measurement and quantification of NH3
A variety of techniques have been attempted to determine the amount of NH3 produced during nitrogen photoreduction reaction and can be mainly divided into six types, including (1) spectrophotometry (or colorimetry), (2) ion chromatography (IC), (3) ion-selective electrode (ISE), (4) fluorescence, (5) 1 H NMR spectroscopy, and (6) ultrahigh performance liquid chromatography-mass spectrometry (UPLC-MS) [108,109]. Currently, photocatalytic nitrogen reduction studies are heavily reliant on the spectrophotometric/colorimetric methods using indophenol blue [110] and Nessler's reagents [111]. The methods have been well established with advantages of good sensitivity (0-0.6 mgNH 3 -N·L −1 ) and low cost. Recently, a frequencyselective pulse nuclear magnetic resonance (NMR) technique was proposed to detect the micromolar concentration of NH3 (present in the assay as NH4 + ) in an electrolyte after electrocatalysis [112]. This NMR technique was supposed to be suitable for various conditions, including nondeuterated, nonaqueous and aqueous electrolytes, and would not require separation of NH3 from the electrolyte. Its sensitivity to NH3 can reach 1 μM with isotopic and chemical specificity. An alternative gas chromatographic (GC) method was attempted for in situ ammonia detection [113]. The in situ GC method was shown to quantify ammonia present in the gas phase in less than 5 min analysis time with a detection limit of about 150 ppb v/v (~ 110 ng/L NH3 or 6.5 nM NH3 in the gas phase), a threshold level relevant to studying and screening (electro)catalysts. This GC method can also determine the ammonia concentration in the electrolyte solution, enabling full quantification of the analyte under consideration. More recently, a nonperturbative approach to ammonia detection was presented based on all-optical detection of surfaceenhanced Raman signals (SERS) [114]. This approach was claimed to feature with chemical selectivity to ammonia, allowing rapid detection of sub-1 ppm ammonia in under 1 s, which shows potential for ultra-sensitive in situ/operando chemical experiments. Another advantage is that SERS detection does not have any restrictions on the identity or morphology of electrodes. Ammonia close to the electrode on a micro-and macro-scale can be detected, providing a local reading of ammonia closest to the reaction site.
It should be pointed out that each of these methods has advantages and limits for assaying ammonia. To attain an overall level of accuracy and accountability, it is strongly recommended to use a combination of different approaches [115]. Moreover, it is imperative to develop more selective, sensitive, accurate, and robust protocols for ammonia quantification, as well as in situ and continuous processes for monitoring the nitrogen photoreduction process.

Impact of impurities on NH3 detection
It becomes more and more recognized that the very small yields nowadays achievable in photo-and electrocatalysis bear the danger that impurities present in the catalyst or in the reactor construction materials may be liberated under reaction conditions, participate in the target reaction, and be consequently falsely detected as reaction products. This is by now well established for photocatalytic CO2 reduction [93,116,117]. It is easily conceivable that organic impurities, such as leftovers from solvents and precursors, or rubber-and grease-based sealing materials, may be present that could contribute to the formation of organic products such as methane or methanol. For nitrogen, being predominantly present as gaseous dinitrogen that one is anyway intending to convert, this is not so straightforward. However, in 2019 potential sources of error have been discussed extensively for electrochemical ammonia synthesis, leading to a complicated, but rigorous, measurement protocol capable of detecting false positive results (Fig. 4) [109,118].
Adventitious sources of nitrogen were proposed to be (1) membranes in the electrochemical setup, for example made from Nafion which can accumulate ammonium ions, (2) small amounts of ammonia present in air (< 1 to 250 ppm) of human breath (~ few ppm), or (3) NOx impurities in the nitrogen gas. Consequently, as already suggested in 2018 [119,120], control experiments using Ar instead of N2 are advised, a procedure also adopted in photocatalysis [121]. Furthermore, 15 N2 should be used whenever possible to confirm that yields of 15 NH3 are as high as for the unlabeled case. But even then, the 15 N2 may similarly contain 15 NOx or 15 NH3, so careful monitoring and purification of the gas supply is also required for the labeled gas.
Other researchers [122] particularly highlighted the potential presence of nitrogen-containing species in the electrolyte, for example resulting from nitrate impurities in lithium salts. Control experiments with pure electrolyte in an Ar atmosphere are required to identify these impurities. For photocatalytic nitrogen reduction to ammonia, care must also be taken that carbon impurities are excluded, because they too may contribute to the reaction [123]. It has even been suggested that activation of nitrogen on titania can take place exclusively in presence of carbon on the surface, because carbon radicals are involved in the process. By a combination of ambient pressure X-ray photoelectron spectroscopy (XPS) and density functional theory (DFT) calculations, a reaction pathway has been proposed in which a CN2 species is hydrogenated to liberate ammonia in the end [124].
The above discussion highlights the complexity of the potential sources and influences of impurities. In order to correctly determine the mode of action of carefully designed defects in photocatalytic ammonia formation, care must be taken to address these issues in their entirety.

Aspects to be taken into consideration
Although encouraging advances have been made in ambient photocatalytic ammonia synthesis, there are several factors that need to be taken into consideration [57]. (1) The amounts of ammonia evolved are usually determined by ion chromatography, aqueous-based spectrophotometric/colorimetric assays, ion-selective electrode, and fluorescence methods [99,125]. However, in many photocatalytic nitrogen reduction processes, sacrificial agents are used, which may disturb the measurements [100]. Multiple combined quantification methods are thus recommended to further confirm the reliability of the test results. (2) The diffusion of nitrogen molecules to the active sites of heterogeneous photocatalytic materials is a critical step in the N2 reduction reaction [41]. The extremely low solubility of dinitrogen gas in water-based electrolytes (0.66 mmol·L −1 under room temperature and atmospheric pressure) [126], however, drastically restricts the diffusion and local nitrogen concentration on the catalyst surface, thus affecting ammonia yield. This issue can be ameliorated by increasing the N2 chamber pressure to enhance dissolution of nitrogen [127]. (3) The origin of nitrogen in ammonia should be carefully tracked because it may come from potential exogenous nitrogen contaminants which may be present in the air, environment, human respiration, sample tubing, lab coat, latex glove, stale Milli-Q water, etc [122,128]. Any possible contaminants could give rise to uncertainty and even false positives data on the photocatalytic nitrogen reduction performance. Therefore, it is urgently desired to establish reliable and accurate methods for quantifying NH3 production and rule out other extraneous contamination sources [129]. Performing isotope labeling using 15 N2 both qualitatively and quantitatively is strongly suggested to verify the N source of the detected NH3 and elucidate reaction mechanisms especially for metal nitride semiconductors [118]. (4) Side reactions may happen during the photocatalytic nitrogen reduction, yielding products such as N2H4, NO3 -, and H2, which thus lowers the selectivity of N2 reduction to NH3 [105]. Also, H2O is used as a source of protons. The oxidation half reaction may generate oxidizing substances, such as O2, ·OH, H2O2, or others, which may oxidize ammonia to nitrate. (5) The stability of photocatalyst is a crucial metrics in evaluating the performance for practical applications. An efficient and robust photocatalyst is expected to remain excellent stability during continuous photocatalytic processes. Nevertheless, most current stability tests of the photocatalyst only involve a few cycles and the total reaction time is a few hours, which are far from practical industrial applications [130]. The poisoning and deactivation of photocatalysts remain to be further explored and clarified. Overall, the photocatalytic ammonia synthesis is still in the initial stage, and deserves enormous research efforts to address the above-mentioned issues and upgrade this technology.

Defect engineering of semiconductor photocatalysts
Defects are usually present in semiconductor photocatalysts. Defect engineering enables one to modulate the local surface microstructure, electronic band structure, and chemical properties of a photocatalyst to enhance its performance for the nitrogen reduction to ammonia [72,131]. The activity of a photocatalyst for ammonia synthesis is closely related to the nature of defects. It must be pointed out that the role of defects is complicated in photocatalysis, as they may not be associated exclusively with beneficial effects but may also reduce photocatalytic activity. As will be discussed below, they may also act as recombination centers or undesired trap states. As such, understanding the structure and property of defects would help rationally design high-performance semiconductor photocatalysts. In this section, we will provide brief discussions on the classification, synthesis strategies, and characterization of defects.

Categories of defects
Defects can be classified according to their atomic structure and location in semiconductor photocatalysts. (1) Depending on their dimensions, defects can be categorized as zerodimensional (0D) point defects (e.g., vacancy and doping), one-dimensional (1D) line defects (e.g., screw dislocation and edge dislocation), two-dimensional (2D) planar defects (e.g., grain boundary and twin boundary), and three-dimensional (3D) volume defects (e.g., lattice disorder and void) (Fig. 5) [132,133]. At present, 1D and high-dimensional defects are relatively less discussed in photocatalytic nitrogen reduction. 0D point defects (i.e., vacancies and impurities) were observed to play predominant roles in improving nitrogen reduction. Hence, we mainly correlate the photocatalytic N2 reduction performance with the point defects. (2) Depending on their location, defects are grouped into surface/interface defects, subsurface and bulk defects [134,135]. Recent results revealed that bulk defects could act as recombination centers for photogenerated electrons and holes, being unfavorable for photocatalytic activity [135,136]. In contrast, surface defects may offer new active sites for transfer of photogenerated carriers to the adsorbate, accelerating carrier separation and adsorption of small target molecules on the photocatalyst, benefiting photocatalysis [67]. A spatial and electronic synergy was put forth by introducing surface defects and bulk defects. This not only promotes bulk separation of electrons and holes but also can efficiently lower the conduction band and serve as a capture center for electrons [137]. Subsurface oxygen vacancies were found to enhance conductivity and electron transfer of TiO2, benefiting photocatalytic activity [138]. Interface defects are regarded as the most complex defects, in which the defects exist between the interfacial contacts. Interface defects can promote the interaction between different semiconductors. Due to the differences in the electronic structure of the defects, there may be a strong interaction between the crystals and the defects, which is conducive to the transfer of photogenerated carriers [134]. It has been demonstrated that supersaturated vacancies are prone to precipitate into voids or defect clusters [139]. The aggregation of defects, therefore, may induce the process of phase separation and further evolve to interface/ grain boundaries. Moreover, simulations indicated that energy barriers for defect migration were lower near the interface in composites [139]. This can explain the higher activity on the interface, where the density of defects should be higher than the bulk [140]. On the other hand, defects in semiconductors can optimize the contact interface, transforming the type II heterojunction into a direct Z-scheme structure, thereby enhancing redox ability [141,142]. (3) According to the involved elements, the defects can be subdivided into OVs, nitrogen vacancies (NVs), sulfur vacancies (SVs), carbon vacancies (CVs), fluorine vacancies (FVs), and other elements vacancies. Additionally, starting from a broader definition, single-atom catalysts is also considered as defects [38,143].
As already briefly addressed above, the type and location of a defect may decide whether it has beneficial or detrimental effects. The knowledge base in photocatalysis obtained so far makes it difficult to define generally valid design rules, because a defect beneficial in one material may be detrimental in another. Yet, the following general rules have been developed until today: (1) Surface defects are usually preferred over bulk defects when they are supposed to act as trap states, because charge carriers trapped in the bulk cannot participate in the catalytic process [135,136]; (2) charge carriers located at defects must still provide sufficient oxidation or reduction potential to allow for the target reaction to occur, so shallow traps are usually preferred over deep traps; and (3) catalytic active sites often need defective structures, because low-coordinated sites are usually needed for reactant activation [144].

Synthetic strategies for defects
The synthetic strategy determines the defect type and consequently impacts the performance of photocatalytic materials [145,146]. Diverse synthetic protocols for creating defects have been developed for the N2 photoreduction, which can be mainly split into post-treatment methods and in situ synthetic strategies.

Post-treatment methods
A post-treatment method to introduce defects involves two steps, including photocatalyst synthesis and further partial reduction of semiconductor photocatalysts.

Chemical reduction
Chemical reduction is believed to be an effective strategy to introduce defects in semiconductors. This can be achieved by thermal treatment in reducing gas atmospheres (H2, CO, NH3, and H2S) or through an electro-reduction route. The concentrations of different defects in semiconducting materials can be readily tuned by manipulation of the treatment temperature, time, and reducing gas composition. H2 atmosphere is able to reduce Bi5O7I, Bi2MoO6, and TiO2, creating large amounts of exposed OVs [147][148][149]. NVs were also reported through H2 treatment [52,150,151]. NH3 is also employed as a reducing gas to treat g-C3N4 and generate CVs [152,153].
Compared to thermal treatments in reducing atmospheres, the electrochemical reduction method is more convenient, greener, and energy-saving. Defective TiO2 nanobamboo arrays (DTiO2 NBAs) were synthesized by using the electro-reduction strategy and were shown to effectively catalyze the N2 reduction to NH3 in the visible and near infrared light range [154].

Thermal treatment
Conventional thermal processing can create various defects as a result of atom escape accelerated by high temperature [155][156][157]. As an example, hollow porous prismatic g-C3N4 with NVs and oxygen doping was attained by combining low-temperature hydrothermal treatment and a subsequent annealing process [84]. It was proposed that the photogenerated electrons in the CB quickly migrated to the NVs inducing a mid-gap state, which thus promoted the separation and transfer of photogenerated carriers and boosted the photocatalytic N2 reduction to NH3. Likewise, OVs rich-TiO2 nanosheets decorated with Au nanocrystals were obtained through low-temperature hydrothermal synthesis followed by thermal treatment in Ar atmosphere [158]. The OVs and Au nanocrystals in the hybrid were hypothesized to contribute to the N2 photoreduction by a "working-in-tandem" mechanism, akin to the nitrogenase enzyme system. NVs were also introduced in g-C3N4 using a similar facile thermal treatment approach either in air [159] or N2 atmosphere [160] for enhanced N2 photoreduction.

Force-induced strategies
Defects on semiconductor materials can be induced after subject to plasma etching, alkali assisted etching, and ultrasound irradiation. Plasma etching can construct various intrinsic defects into the semiconductors [161]. Nitride-based photosensitizing semiconductors (i.e., p-GaN, i-GaN, and n-GaN) with NVs have been prepared by plasma-assisted molecular beam epitaxy on commercially available Si(111) wafer [162]. Dielectric barrier discharge plasma treatment was demonstrated to construct NVs and sulfur co-doped g-C3N4 for active photocatalytic N2 reduction to NH3 [163].

In situ synthetic strategies
As opposed to post-treatment methods, the type and density of intrinsic defects in semiconductors are directly regulated during the synthesis process.

Hydro/solvothermal method
The hydro/solvothermal method using water, alcohol, or a mixture of both as solvents has been intensively applied to introduce various defects on semiconductor materials with features of convenience, simple operation, and cost-effectiveness [145]. MIIMIII-LDH (MII = Mg, Zn, Ni, Cu; MIII = Al, Cr) nanosheet photocatalysts with OV defects have been synthesized based on a one-step hydrothermal method [169]. The introduction of OVs in the CuCr-LDH nanosheets was surmised to induce distortion of MO6 octahedrons, contributing to the superior N2 photoreduction to NH3. Analogously, SVs and O-doping were collaboratively introduced into 1T-MoS2 nanosheets, significantly enhancing the N2 adsorption and activation [170]. Hydrothermal treatment by microwave was also reported to construct 1D attapulgite (ATP) mineral supported Pr 3+ : CeF3 nanocomposite with abundant FVs [171]. The FVs along with Pr 3+ doping was supposed to expand light absorption range and also provide abundant active sites, jointly facilitating the adsorption of N2 and weakening the N≡N triple bond. Moreover, BiOBr nanosheets with desirable OVs and dominant exposed (001) facets were fabricated by controlling the addition of polyvinylpyrrolidone (PVP) during the solvothermal process [172]. Single-unit-cell Bi3O4Br nanosheets with tunable surface defects were also prepared and showed boosted photocatalytic nitrogen fixation by mediating electron-hole separation [173].

Low-valence metal doping
Introduction of low-valence metals can alter redistribution of the electronic structures of semiconductors, defects are hence generated to balance the positive and negative charges. Lowvalence metal species can serve as the coordinatively unsaturated sites with electron-rich properties, conducive to adsorption and activation of reactants on the catalyst surface, thereby improving photocatalytic activity [41,174,175]. For instance, W18O49 ultrathin nanowires with OVs were synthesized via subtle Mo doping [66]. The doped low-valence Mo species was calculated to polarize the chemisorbed N2 molecules and facilitate the electron transfer from coordinatively unsaturated sites to N2 adsorbates, making dissociation of the N≡N bond more feasible. In addition, the defect-band center was elevated toward the Fermi level, preserving the energy of photoexcited electrons for N2 reduction. TiO2 nanosheets with OVs and intrinsic compressive strain were obtained by employing a Cu-doping strategy [176]. Based on a similar route, OVs were introduced in ZnAl-LDH nanosheets [177]. It was supposed that Cu addition gave rise to OVs and coordinatively electronrich unsaturated Cu δ+ (δ < 2). This accelerated separation of photogenerated carriers and promoted N2 adsorption and activation, thereby enhancing photocatalytic N2 reduction to NH3.

Light irradiation
Light irradiation enables one to create defects in some semiconductors during a photocatalytic process [108,178,179]. Creation of OVs on the surface of Bi5O7Br nanotubes was observed upon visible-light photoirradiation [180,181]. In light of the high O atom concentration, low bond energy, and long bond length of the Bi-O bond in Bi5O7Br, the Bi-O bonds were easily dissociated by UV light irradiation. O atoms were dragged away through strong interactions with solvent ligands, leaving the OVs on the Bi5O7Br surface. However, this method is unsuitable to break the strong metal-oxygen bonds to generate OVs for metal oxides.
Apart from the above-mentioned methods, there are also several promising synthesis strategies that haven't been explored in photocatalytic N2 reduction reaction. For instance, ball milling deforms the material structure, substantially reduces the volume, and exposes defects, providing an economical, effective, and reliable approach to creating vacancy edge defects [182]. Molten salt (MS) synthesis can induce vacancy defects in metal compounds and g-C3N4 [183,184]. An alternative lithium reduction strategy can implant defects into a series of metal oxides, such as TiO2, ZnO, SnO2, CeO2 and their mixtures at ambient conditions [185][186][187]. Among others, vapor diffusion can allow one to construct defects on photocatalysts with desired location and density [188].

Identification of defects
Although the introduction of defects by suitable defect engineering has a positive effect on photocatalysis, the recognition and quantification of defects at the atomic level is still a grand challenge. The further one clarifies the atomic structures of defects and specifies their concentrations by advanced characterization technologies, the more helpful it is to establish the relationship between defect structure and catalytic behavior. This can then guide researchers towards rational design of photocatalysts. To date, several characterization techniques including ex situ and in situ/operando methods have been attempted to identify and quantify the defects in semiconductors for the N2 photoreduction to NH3.

Transmission electron microscopy (TEM)
TEM provides the most direct tool for defect investigation, through which surface defects can be visually observed to acquire information with respect to the type and density of vacancies over semiconductor materials. By high-resolution TEM, it is possible to vividly see some lattice defects [168]. Nevertheless, it is still challenging to distinguish the difference between surface defects and bulk defects by TEM. Another issue is that only a small amount of samples is probed by electron microscopy. To confirm the TEM results, it is necessary to couple with other spectroscopic or resonance methods that can examine larger amounts and areas of a particular sample.

X-ray powder diffraction (XRD)
XRD is commonly used to identify catalyst defects given that the material diffraction signals may change upon introduction of defects [189,190]. It has been recognized that doping with other metal elements can broaden the XRD reflection peaks [191]. For example, the sharpness and relative intensity of XRD peaks of VO-BiOBr nanosheets became significantly reduced compared to BiOBr nanoplates, implying the presence of OVs in VO-BiOBr nanosheets [172]. Additionally, the XRD peaks may shifted, indicating variation of lattice parameters of a catalyst due to lattice expansion or distortion caused by defects [192]. It was noticed that the (110) Bragg reflections for both CuCr-NS and ZnAl-NS shifted to higher 2θ angles relative to bulk CuCr and ZnAl. This suggested an in-plane biaxial compressive strain in the plane, induced by surface OVs [169]. Although the XRD technique can qualitatively reflect the existence of defects, it fails to provide information on the type and exact location of defects and cannot quantitatively determine the concentration of defects.

XPS
XPS is a well-established comparatively surface-sensitive (≈ 10 nm depth) method. It has been intensively applied to detect the surface defect type of a semiconductor photocatalyst [193,194]. Commonly, the existing defects in materials can change the electronic structure and chemical environment of the elements, giving rise to differences in the XP spectrum, namely peak shift, intensity variation, or the appearance of new peaks [181,195].
Several papers have appeared nowadays, in which a signal in an X-ray photoelectron spectrum (XP spectrum) in the range of the O 1s orbital around 530 eV has been assigned to an oxygen vacancy [181,196,197]. However, thinking about this logically, it is plainly impossible that an atom that is not there liberates a photoelectron upon irradiation with X rays. Furthermore, an electron trapped in a vacancy, if liberated, would certainly not show up in an XP spectrum in the range of the O 1s orbital, because it is much more loosely bound. More reasonable assignments refer to changes in the electronic state of (surface) oxygen atoms caused by the presence of the OVs and associated charge imbalances. For example, a signal at ~ 530.5 eV has been assigned to highly oxidative oxygen species in a perovskite-type mixed lanthanum cerium ferrite, whereby these particular oxygen species are closely associated with the presence of surface OVs [198]. In a related manner, a signal at 530.55 eV has been assigned to surface oxygen atoms bound to Ti 3+ in TiO2 [199]. Indirectly, this signal can also be associated with OVs, because Ti 3+ is indicative of a reduced sample. Many other examples may be found, which is outside the scope of this review article. In summary, it has to be kept in mind that OVs cannot be monitored directly by means of XPS, but can be monitored by secondary effects they have on the electronic structure of the sample of interest.

Electron paramagnetic resonance (EPR)
EPR is a powerful magnetic resonance technology fundamentally based on the magnetic moment of unpaired electrons, which can be used to qualitatively and quantitatively detect the unpaired electrons contained in the atoms or molecules of substances (e.g., vacancy defects), and to explore the structural characteristics of their chemical environments [200]. Unpaired electrons in various chemical environments can be monitored by EPR in terms of the g value. Different g values and signal intensities are closely related to the type and relative concentration of defects [201,202]. In general, for oxide-based materials, signals with g values smaller than 2 are usually associated with foreign dopant atoms, or with pretreatment-induced self-doping (e.g., Ti 3+ or Zn + in TiO2 or ZnO caused by reductive pretreatment) [203][204][205][206]. Isotropic signals appearing in very close proximity of the free electron (g = 2.0023) are usually attributed to unpaired electrons trapped in (anion) vacancies, such as OVs [203,205]. Care must be taken to properly distinguish them from anisotropic signals near g = 2, because they may originate from oxygen-centered radicals, such as superoxide [207,208]. For instance, an EPR peak with a g-value of 1.999 was observed for BiOBr, verifying the presence of OVs. After Fe doping, the EPR peak signal increased apparently. This can be mainly ascribed to the size difference between Fe and Bi atoms, which brought about structural distortion, and thus created abundant OVs [209]. Apart from identification of anion vacancies, EPR can also probe cation vacancies. Recently, Qiu and co-workers utilized the EPR to detect Ti defect of the NH2-MIL-125 (Ti) [210].

Raman spectroscopy
Raman spectroscopy has been widely employed to characterize and analyze chemical bonding energy on the surface of semiconductor [211,212]. It is worthwhile mentioning that Raman spectroscopy frequently helps to obtain more detailed information on intrinsic defects of carbon materials [212]. For instance, Li and co-workers used Raman spectroscopy to identify intrinsic defects of graphdiyne@-Fe3O4 (GDY@Fe-A and GDY@Fe-B) [213]. The level of defects was reflected by the relative intensity ratio of D to G-band (ID/IG) [212]. Compared with pure GDY (ID/IG = 0.80), both GDY@Fe-A (ID/IG = 0.83) and GDY@Fe-B (ID/IG = 0.86) displayed larger ID/IG, indicating the existence of much more defects. The Raman peak position of the anatase Eg mode was observed to have a linear red shift with the concentration of OVs [214]. Raman spectroscopy was also applied to elucidate the dependence of N2 photoreduction rate on OV concentration [158]. A decrease in the concentration of OVs in TiO2 was observed to induce an increase in the lattice spacing, hence resulting in a red shift of the Eg mode of anatase in the Raman spectrum. It was further demonstrated that the N2 photoreduction rate decreased nearly linearly with the reduction of OV concentration.

Spectroscopic ellipsometry (SE)
SE is regarded as a versatile and nondestructive technique, which can be used to characterize thin films and heterostructures, with benefits of high sensitivity to material properties and surface morphology [215,216]. For example, SE has been applied to confirm the OVs on a-TiO2 (Fig. 6) [217]. Figures 6(a) and 6(b) illustrate that several intraband absorptions exist in TiO2, which were attributed to the defects of a-TiO2, resulting in nonzero density of states in the band gap.

X-ray absorption spectroscopy (XAS)
XAS has emerged as an effective and indispensable analytical technique which is extensively employed to obtain quantitative structural information for semiconductor at the atomic scale (i.e., the oxidation state, bond length, coordination number, and atomic species) [218]. Furthermore, XAS can also enable identification and analysis of the energy-dependent fine structure of the X-ray absorption coefficient (μ(E)) near the absorption edge of a particular element [219]. The μ(E) represents a smooth function of the photo energy, which can be calculated by μ(E) = dZ4/mE3 where d, Z, and m are the target density, atomic number, and atomic mass, respectively. Generally, when the energy of the photon exceeds the binding energy of the core electron, a new absorption channel will be created, resulting in a sharp increase in the absorption coefficient. When the energy is higher than the energy gap between the unoccupied bound state and the core energy level, the photoelectron is promoted to a continuous state, which generates a wave that propagates outward and is scattered on adjacent atoms (Figs. 7(a) and 7(b)) [220]. The interference mode of the emitted and scattered waves is determined by the geometry of the absorbing environment and the wavelength of the photoelectron.
Hence, XAS is commonly categorized as XANES (X-ray absorption near-edge structure, approximately 40 eV and below) and EXAFS (extended X-ray absorption fine structure beyond the XANES region) dependent on the difference in the relative absorption threshold of energy ( Fig. 7(c)). For instance, the local coordination of copper ions in 0.5%-ZnAl-LDH and CuZnAl-LDH has been explored by XANES (Figs. 7(d)-7(f)) [177]. The Cu K-edge oscillations in the range of 0-10 Å were found to be different (Figs. 7(d) and 7(e)), suggesting subtle variations in the local environments around the Cu atoms in the two samples ( Fig. 7(f)). The EXAFS R-space spectra exhibited two peaks, corresponding to the first Cu-O shell and Cu-metal shell. The 0.5%-ZnAl-LDH samples showed a longer Cu-O distance (1.99 Å) and a lower coordination number (5.38) compared with CuZnAl-LDH (an average Cu-O distance of 1.93 Å and a coordination number of 6.0 for the first Cu-O coordination sphere). This reveals that Cu in 0.5%-ZnAl-LDH possessed a high degree of coordinative unsaturation and abundant OVs, causing structural distortion and strain within the LDH nanosheets. The formation of surface OVs on TiO2 was also characterized by XAS with increased relative intensity of pre-edge peaks in the VO-TiO2, an indication of distortion from the octahedral TiO6 unit ( Fig. 7(g)) [221]. While modification by F atoms led to decrease of peak intensity, manifesting that the partial distortion might be recovered. The partial OVs were supposed to be taken by F atoms. Likewise, N-defects in the heptazine rings of g-C3N4 and mCNN can be probed by performing C K-edge and N K-edge synchrotron-based XANES measurements (Figs. 7(h) and 7(i)) [222].

Positron annihilation spectroscopy (PAS)
PAS, a novel, sensitive, and non-destructive spectroscopy, has recently sparked extensive attention since it can provide direct information on defect structures including defect types and relative concentrations in photocatalysts, based on analyzing the positron lifetime and intensity [223][224][225]. After entering the material, positrons are converted into γ photons with electrons. When there is a lattice defect (e.g., vacancies and vacant clusters), the positrons will be trapped in the bound state of this defect, thereby altering the positron annihilation spectrum [225,226]. As an example, the positron lifetime spectra of defect-rich SUC Bi3O4Br and defect-deficient Bi3O4Br were observed to exhibit three lifetime components, confirming the existence of Bi vacancy related clusters (Fig. 8(a)) [173].
The Bi vacancy concentration appeared to increase with the increase in the relative intensity of the positron lifetime (Figs. 8(b) and 8(c)).

In situ or operando technique studies
Apart from the above ex situ characterization techniques, in situ or operando technologies can provide deeper insight into the defects of photocatalysts although this area is still nascent. Indeed, the emerging in situ or operando XPS, EPR, and XAS can identify the type and concentration of defects and allow real-time monitoring of the defect state and its evolution during the N2 reduction process. These techniques unlikely influence the type and amount of the intermediates due to their not strong energy conditions. In the following, we provide a brief discussion on recent progresses attained regarding in situ or operando investigation of defects during the N2 reduction process and make attempts to understand the structure-performance relationship of defects. This would guide further exploration and development of more efficient defect-engineered photocatalysts for practical applications.

In situ XPS
By virtue of quasi in situ XPS, the formation and chemical states of defects can be probed during photocatalysis at the atomic level. In situ XPS has been applied to track the defect structure of WO3 during the N2 photoreduction process (Fig. 9) [179]. Quasi in situ XPS measurements were performed before and after the treatment of a flowing H2O vapor of N2 at 25 °C  with light for 30 min in a cell reactor connected with a photoemission end station. The in situ O 1s XP spectra (Fig. 9) of WO3/W-400, WO3-600, and WO3-NPs before and after the treatment displayed three main peaks at 530.2, 531.7, and 532.8 eV, which were assigned to lattice oxygen, OVs, and surface hydroxyl group, respectively. The peak intensity of OVs for WO3/W-400 was observed to slightly decrease after the treatment. This was supposed to originate from oxidation of surface OVs by nearby O atoms from H2O ( Fig. 9(a)). For WO3-600, the signal of OVs did not emerge before treatments, while a strong signal for OVs appeared after exposure to N2 and H2O under light irradiation ( Fig. 9(b)). Whereas for WO3-NPs, no obvious variation of the O 1s spectra was detected before and after the treatment (Fig. 9(c)).

In situ EPR
The attenuation and generation of defects during the N2 photoreduction process have also been verified by in situ EPR measurements [179]. For example, the role of defects and electron-rich Cu δ+ in promoting the N2 photoreduction was recently revealed by in situ EPR experiments under a constant N2 flow (Fig. 10(a)) [177]. An external magnetic field was applied to the sample to excite those unpaired free electrons on the atomic orbital or molecular orbital to a higher energy level, resulting in a characteristic EPR signal ( Fig. 10(b)). Relative to CuZnAl-LDH and ZnAl-LDH, the 0.5%-ZnAl-LDH sample exhibited a more intense EPR signal of Cu 2+ (3d 9 ). This implied that the generation of oxygen defects increased the number of free electrons on the Cu center in 0.5%-ZnAl-LDH (Fig. 10(c)). The intensity of the Cu 2+ EPR signal for the 0.5%-ZnAl-LDH decreased under UV-vis illumination (Fig. 10(d)), indicating that the photogenerated electrons were transferred to the Cu centers, thereby reducing some Cu 2+ to low-valence Cu δ+ .

In situ XAS
In situ XAS is able to verify the generation of defects during the N2 photoreduction process at the atomic level [227]. The electronic properties of W element in WO3-600 during N2 photocatalytic reduction were studied by in situ XANES [179]. It was found that the W L3-edge variation of absorption edge position of WO3-600 was almost negligible when being immersed in water under N2 atmosphere without irradiation ( Fig. 11(a)). While the absorption edge position of WO3-600 shifted to lower energy under light irradiation, suggesting that the W element in WO3-600 was partially reduced to generate OVs during the photocatalytic N2 reduction reaction. Additionally, the coordination number for W-O shell in WO3-600 decreased from 5.4 to 4.4 during the photocatalysis (Fig. 11(b)), consistent with the variation of electronic structures of W element as well as the formation of NH3 on the surface of catalysts due to the generation of operando OVs.
Advances of spectroelectrochemical techniques also make it possible for imaging nano-scale defects and detection of their changes during the N2 reduction [228,229]. To obtain real time data regarding morphological and microstructural evolution of defect-engineered photocatalysts at the atomic level during the N2 reduction process, in situ scanning electron microscopy (SEM) and TEM provide a unique possibility [230][231][232]. More intriguingly, the disorder of vacancy channels and their local rearrangement at different temperatures can be observed by in situ heating TEM [233].
Other techniques that haven't been attempted for photocatalytic N2 reduction include in situ Raman, which enables real-time  detection of chemical bond changes, microstructural evolution and its formation conditions of defective photocatalysts at the atomic level [234,235]. Note that laser sources in Raman spectroscopy may affect the evolution and detection of N2 reduction intermediates. To minimize such impact and also avoid other optical responses (such as fluorescence) from catalysts, near-infrared laser is preferred. In situ XRD can monitor the phase transformation and structural evolution of catalysts [236]. Combined operando electrochemical impedance spectroscopy (EIS) with cyclic voltammetry (CV) provides a potentially useful experimental tool to gain insight into the role of defects during catalytic reactions in a dynamic way [237]. Despite these recent advances, further development of advanced and powerful techniques is still urgently demanded to track the evolution of defects for the sake of gaining insight into active sites and reaction mechanism of photocatalytic N2 reduction.

Roles of defects in photocatalytic nitrogen reduction
In this section, we attempt to summarize and discuss the positive roles of defects toward photocatalytic N2 reduction reaction. For each of the possible aspects that the defect may address, e.g., extension of the light response range, acceleration of charge transfer rates etc., we briefly explain the fundamental physical necessities for this effect to occur, and then give recent examples from literature. It always depends on the specific semiconductor and specific defect, which mode of action is in operation, because the physical characteristics such as reduction and oxidation potential of conduction and valence band edge are different in each case. It is, to the best of our knowledge, not yet possible to unveil general correlations that would allow to predict the exact function in detail.

Extending light response range
The photo response ability of a semiconductor photocatalyst is largely determined by its band gap structure. Narrowing the band gap structure of a semiconductor can effectively extend the range of light absorption. Appropriate engineering of defects enables one to modulate the band gap structure of a semiconductor [238,239]. Generation of defects is inferred to introduce electronic states within the bandgap, which helps to narrow the bandgap and serve as mid-gap states for the photoexcited electron, thereby expanding the light response range [240]. The position of mid-gap states can be tuned by tailoring the types and concentrations of defects. On the one hand, the intermediate energy level may be situated below the edge of the CB, decreasing the band gap of semiconductors [121,130,241]. As a matter of fact, manipulation of NVs enabled tuning of bandgap of cyano-group-modified g-C3N4 nanoribbons (mCNN) from 2.77 to 2.0 eV by creating a sub-gap state (2.57 eV) under the edge of CB, thus extending the absorption of g-C3N4 to the entire visible-light and even near infrared regions (Figs. 12(a) and 12(b)) [222]. By controlling the concentration of OVs in TiO2 nanosheets via Cu-doping [176], the light absorption range was adjusted from 400 to 800 nm (Fig. 12(c)). A bandgap of 3.0 eV with the valence band maximum (VBM) located at 2.75 eV was attained favoring the N2 photoreduction ( Fig. 12(d)). On the other hand, the intermediate energy level may be alternatively above the edge of the VB, also lowering the bandgap of a semiconductor. This is usually at the expense of its oxidation potential [242]. Mid-gap states, namely the surface oxygen-defect states, above and partly overlapped with the VB of Bi2MoO6 could be introduced by hydrogen treatments. The VBM was shifted toward the vacuum level by about 0.73 eV, reducing the bandgap of the Bi2MoO6 (Fig. 12(e)) [148]. By introducing the mid-gap states through defect engineering [243], the bandgap of BiOBr nanosheets was lowered from 2.69 to 2.43 eV. Correspondingly, the absorption range of its absorption edge was extended lying between 450 and 750 nm (Figs. 12(f) and 12(g)) [172]. Likewise, the bandgap of Bi2O2CO3 was decreased from 2.42 to 2.24 eV by increasing OVs concentration (Fig. 12(h)) [244].
Recently, it was discovered that defects can also concurrently introduce mid-gap states (i.e., defect states) and minimize bandgaps to extend the absorption edge of a semiconductor [245][246][247]. For instance, Fe doping was shown to effectively narrow the intrinsic bandgap of SrMoO4, extending the light absorption edge from UV light to visible light region (Fig. 12(i)) [174]. With the increase of Fe dopant concentration, the intrinsic bandgap was lowered accordingly (Figs. 12(j) and 12(k)). Independently, Fe doped SrWO4 nanoparticles were also synthesized with bandgap tuned in the range from 4.78 to 3.39 eV [245]. Upon Fe doping, the conduction band minimum (CBM) was observed to shift downward while the VBM shifted upward, resulting in narrowed bandgap. Consequently, the absorption range was extended by shifting its absorption edge. The narrowed bandgap achieved either by the shift of the CBM downward or the shift of the VBM upward could both reduce the redox ability correspondingly on the reduction side or the oxidation side. A balance in this regard should be considered when modulating the type and concentration of defects.

Modulating charge carrier transfer kinetics
Defects can also act as reactive sites and capture photoinduced electrons, which can effectively restrain the photoinduced electron/hole pair recombination [154,[248][249][250][251]. Several advanced techniques have been employed to characterize charge transfer and separation in defect-engineered photocatalysts, including EIS, photocurrent response, room-temperature photoluminescence (PL) and time-resolved photoluminescence (TR-PL) spectroscopy as well as transient absorption (TA) spectroscopy [252]. It was proposed that the OVs in BiOBr nanosheets played a key role in promoting photo-generated carrier migration and separation [253]. The interfacial electrons transferred from the excited BiOBr nanosheets facilitated activation of N2 and further reduction to NH3. Further PL, TR-PL, and photocurrent response results provided evidences that the OVs inhibited the recombination of photogenerated electron/hole pairs and increased the lifetime of charge carriers. Another interesting work from Zhang and coworkers showed that the introduction of OVs and Ru single atoms (SAs) in TiO2 cooperatively accelerated electron and hole separation and suppressed their recombination, prolonging the lifetime of photogenerated carriers to 0.73 ns ± 0.03 ns (Figs. 13(a)-13(f)) [254]. Note that the defects in semiconducting photocatalysts is a double-edged sword for charge-carrier separation. An appropriate level of defects introduced in semiconductors is favorable to extend light response range, promote photogenerated carrier separation, and hamper photogenerated electron-hole pair recombination. While excessive defects can serve as recombination centers, resulting in a substantial drop in carrier separation efficiency.

Improving N 2 adsorption
Materials that can provide a high density of N2 adsorption sites with adequate strength, are promising for good catalytic performance given that the surface coverage (at given surface area) controls the number of N2 molecules converted per time  to a great extent [41]. After N2 molecules approach the catalyst surface, adsorption on the surface might occur, either by physisorption or chemisorption. Physisorption of nitrogen, which is purely based on weak forces, e.g., on van der Waals interactions, is commonly exploited in heterogeneous catalysis for surface area measurements according to the Brunnauer-Emmett-Teller (BET) method at temperatures near the boiling point of liquid nitrogen [255]. Compared with physical adsorption, chemical adsorption involves electron transfer between the catalyst and nitrogen molecule, which is linked to N2 activation. Defects have been shown to enhance the N2 adsorption on the photocatalyst surface because of the following aspects: (1) the formation of thermodynamically unstable atoms with low coordination induced by defects, favoring N2 adsorption [148]; (2) some surface defects are frequently positively or negatively charged, which is conducive to promote the adsorption of oppositely charged reactant molecules by electrostatic interactions [133]; (3) some defects (i.e., NVs) can selectively adsorb N2 since their size and shape resemble the nitrogen atoms in N2 [256].
Experiments (such as in situ diffuse reflectance infrared Fourier-transform spectroscopy (DRIFTS), temperatureprogrammed desorption (TPD), isotopic labeling measurements) combined with theory (such as DFT calculations) have been employed to verify the roles of defects (OVs, NVs, SVs) in facilitating adsorption and activation of N2 in photocatalytic N2 reduction. As revealed by DRIFTS (Fig. 14(a)), a series of infrared (IR) peaks located at 3,555, 3,360, 2,874, 2,359, 1,624, and 1,557 cm −1 were detected for N2 adsorption on OVs enriched Bi5O7Br-40 in the dark for up to 40 min, which were ascribed to N−H stretch, adsorbed −OH, NH4 + stretching vibration, adsorbed CO2, adsorbed N2, and adsorbed NH3, respectively [181]. The intensity of these vibration peaks increased with the prolongation of adsorption time. Under visible light, peaks of 3,555, 2,874, and 2,359 cm −1 were extremely weakened, whereas peaks at 3,360, 1,624, and 1,557 cm −1 were enhanced ( Fig. 14(b)), indicating promoted adsorption of N2 and the conversion of N−H to NH3. With increasing photoirradiation time (up to 35 min), the IR signals centered at 3,555 and 2,874 cm −1 enhanced, while the peak at 1,624 cm −1 became weakened (Fig. 14(c)). This originated from the activation and cleavage of N≡N bond and the subsequent formation of NH4 + . Upon switching off the light, the IR signals at 3,555 and 3,360 cm −1 remained nearly unchanged, whereas the peak at 1,557 cm −1 enhanced (Fig. 14(d)). This suggested that the N−H bond concentration increased and a large number of N2 molecules were adsorbed on the Bi5O7Br catalyst surface. Based on DFT calculations, Bi5O7Br−O containing OVs was found to possess more negative adsorption free energy for the initial N2 adsorption and activation (*N2) and larger adsorption free energy of the *NH2 product, compared with neat Bi5O7Br and pristine Bi5O7Br+O (Figs. 14(e)-14(h)).

Enhancing N2 activation
It has been proposed that the coordination between N2 and defects such as anion vacancies [253] could promote the dissociation of nitrogen and electron transfer, enhancing N2 transformation. For example, OVs with abundant localized electrons have been shown to act as nitrogen trapping sites that can effectively capture and activate N2 by prolongating the N≡N triple bond [257]. For defective metal-based semiconductors, adsorbed N2 binds with surrounded metallic atoms to generate a metal-N2 complex. This can activate dinitrogen where N2 molecules donate electrons from their occupied σ orbitals to empty σ orbitals of the metal. Reversibly, back-donation of electrons from the metal to unoccupied π* orbital of N2 molecules takes place, leading to cleavage of dinitrogen. Such electron back-donation mechanism is likely applicable to metal-free systems by forming coordinated nonmetal-N2 such as carbon or boron.

Surface defect engineering
Semiconductor materials have high surface energy and abundant exposed atoms to form dangling bonds, thus causing the formation of surface defects [134]. Surface defect engineering of semiconductors (i.e., surface vacancy and doping) has been employed to modify photocatalysts toward effective photocatalytic N2 reduction for NH3 synthesis [38,82]. In the subsequent section, we will provide discussions on the most common surface defects including OVs, NVs, SVs, CVs, and FVs. A summary of defect-engineered photocatalysts for the N2 reduction is given in Table 1.

Oxygen vacancies (anion defects)
A considerable number of metal oxides, such as Ti-based [191,248,258], Bi-based [41,259,260], layered-double-hydroxide (LDH) [169], W 6+ [245], Mo 6+ [261,262], In 3+ [121,263], Ce 4+ [47] oxides have been reported to be active for the N2 photoreduction. Among these oxides, oxygen-vacancy-rich TiO2 is the most investigated semiconductor material for the N2 photoreduction. A commercially available TiO2 with abundant OVs was successfully fabricated, which provided coordinatively unsaturated Ti 3+ species as active sites on OVs [67]. OVs as photoexcited electron traps inherently created surface Ti 3+ species that served as adsorption sites for N2 and promoted the photocatalytic N2 reduction to NH3 (Fig. 15(a)). This semiconductor displayed high activity with a solar-to-chemical energy conversion efficiency of 0.02%. Analogously, OVs and fluorine (F)-modified TiO2 photocatalyst could further improve the photocatalytic N2 reduction activity with an NH3 yield rate of 206 μmol·h −1 ·g −1 [221]. The impressive performance was ascribed to the synergy of OVs and F, in which the OVs facilitated N2 chemical adsorption via electron transfer and F sites altered the TiO2 surface properties from hydrophilic to aerophilic ( Fig. 15(b)). A "working-in-tandem" pathway was proposed for Au nanocrystals and OVs co-modified TiO2 nanosheets (Au/TiO2-OV) toward photo-driven N2 reduction [158]. The OVs acted as active sites to promote chemisorption and activation of N2 molecules while hot electrons originating from plasmon excitation of the Au nanocrystals facilitated NH3 formation (Figs. 15(c) and 15(d)). An NH3 yield rate of 78.6 μmol·h −1 ·g −1 accompanied by an apparent quantum effciency (AQE) of 0.82% was achieved at 550 nm, exceeding many reported systems for N2 photoreduction. However, the purity of both N2 and Ar used in this work was not provided. Additionally, the origin and quantity of NH3 formed were not investigated by using 15 N2 isotope labeling.
Apart from Ti-based photocatalysts, Bi-based compounds have recently attracted extensive attentions for N2 reduction because of their suitable energy band gap structures and large exposed surface areas [172,264]. OV-rich BiOBr nanosheets with exposed (001) facets (BOB-001-OV) were shown to be active for visible light-driven N2 reduction to NH3. The defect states caused by OVs on the surface of BiOBr were supposed to accelerate interface charge transfer from the excited nanosheets to N2 adsorbed on OVs [253]. Alternatively, an  [180].
Combining plasmon excitation to afford hot electrons and enhanced N2 adsorption, near infared (NIR)-driven N2 reduction to NH3 was attained over crystalline VO-rich CeO2 grown at the ends of the gold nanorods (Au/end-CeO2) [280]. An NH3 yield rate of 114.3 μmol·h −1 ·g −1 was obtained under 808 nm laser illumination. The performance was mostly credited to the spatially-separated architecture, in which the Au nanorods absorbed NIR photons to produce plasmon-excited hot electrons and injected them into the CB of VO-rich CeO2. This boosted  [279] (2020) a The AQE was calculated [180]. b The apparent quantum yield (AQY) was calculated [96].
c Ar-PC represents photocatalytic experiments performed in Ar. d LDH stands for layered-double-hydroxide. broad-spectrum N2 reduction on the OVs sites (Figs. 15(g) and 15(h)). It is noted that the purity of N2 feed gas in this work was unknown. The source and amount of NH3 evolved were not probed by 15 N2 isotope labeling experiments.

Nitrogen vacancies (anion defects)
NVs have primarily been investigated in g-C3N4 with abundant nitrogen atoms [84,160,241,281]. The type of surface NVs is usually grouped into N vacancy defects and surface functional groups (i.e., cyano groups and amine), which could effectively improve the photocatalytic N2 reduction to NH3 [94,282]. The surface NVs possess similar size and shape with the nitrogen atom in N2, thus facilitating selectively adsorb and activate N2.
Elemental analysis (Fig. 16(a)) in combination with XPS ( Figs. 16(b) and 16(c)) was employed to estimate the C/N molar ratio for g-C3N4 and V-g-C3N4 (with NVs) and verify the formation of NVs [256]. Based on N 1s and C 1s XP spectra, the peak-area ratio of N-C3 to C-N-C was found to decrease from 3.25 to 0.76, suggesting that NVs were primarily located at the tertiary nitrogen lattice sites. The disappearance of threecoordinate nitrogen was accompanied by the gradual formation of two-coordinate carbon (N-C-N), further confirming the generation of NVs (Fig. 16(d)). Notably, an NH4 + amount of 2.4 mM and an NH4 + production rate of 1.24 mmol·h −1 ·g −1 were attained over V-g-C3N4 in the N2 atmosphere under visible light irradiation (Figs. 16(e) and 16(f)). Moreover, when using a mixed gas (N2, O2, and CO2 with the volume ratio of 1:1:1) in place of N2, the amount of NH4 + formed remained unchanged. This indicated that the photocatalytic N2 fixation activity of V-g-C3N4 was unlikely interfered by other gases. Besides g-C3N4, NVs were also created in some nitride semiconductors [283,284]. For example, gallium nitride nanowires (NWs) with NVs were designed and applied for N2 photoreduction [162]. The presence of NVs in GaN nanowires was validated by EPR measurements revealing a peak at approximately 3,450 G. After treating the GaN nanowire photocatalyst with N2, the EPR signal apparently decreased, indicating the interaction between NVs and N2. The assynthesized GaN NWs with NVs displayed an NH3 yield rate of 340 μmol·gcat −1 ·h −1 under visible light (> 400 nm) with good stability and reusability ( Fig. 16(g)). 15 N2 isotope labeling experiments were further performed to examine the origin of nitrogen in the formed NH3 (Fig. 16(h)). It was found that when using 15 N2 (98% purity) as a feedstock, 14 NH3 was still generated as the main product over GaN NWs. 15 NH3 did not become the main product until the catalyst was evacuated under high temperature and vacuum (> 400 °C under < 5 × 10 −2 mbar for 12 h). This supported that the 14 NH3 product was mainly ascribed to adsorption of atmospheric N2 to the catalyst and not from GaN decomposition.
The surface NVs coupled with extra species including cyano groups and amine can jointly regulate the electronic structure and hence improve photocatalytic efficiency for N2 reduction [97,271]. As an example, g-C3N4 photocatalyst was modified with cyano groups and also intercalated with K + (mCNN), which provided a high NH3 yield of 3.42 mmol·g −1 ·h −1 under visible-light irradiation (Fig. 16(i)) [222]. When Ar was applied as the feed gas, the quantity of NH3 also increased with the extension of irradiation time but the value was substantially reduced compared to the process using N2 feed gas (Fig. 16(j)). This observation suggested that under Ar, the nitrogen source in the formed NH3 stemmed from the photocatalyst. 15 N2 isotope labeling analysis was carried out. The ratio of the collective integral area of 15 NH4 + peaks centered at chemical shifts of d = 6.88 and 7.06 ppm after 4 and 8 h of reaction was 1.65 showing continuous 15 NH4 + production, whereas the ratio of the collective integral area of 14 NH4 + peaks centered at chemical shifts of d = 6.84, 6.97, and 7.10 ppm after 4 and 8 h of reaction was 1.07, indicating that the generation of 14 NH4 + discontinued. This result suggested that the utilizable N from mCNN for NH3 formation was exhausted after some time.

Sulfur vacancies (anion defects)
Introducing SVs onto photocatalyst surfaces can adjust the electronic structure of sulfides, which is conducive to enhance the photocatalytic N2 reduction [278]. Until now, sulfides such as In2S3 [274], MoS2 [46], Mo0.1Ni0.1Cd0.8S [275], ZnMoCdS [277], Zn0.1Sn0.1Cd0.8S [276], and Zn0.11Sn0.12Cd0.88S1.12 [285] have been demonstrated to be capable of catalyzing ambient photochemical N2 reduction to NH3. For example, CdS nanorods modified with SVs-rich O-doped 1T-MoS2 nanosheets (denoted as SV-1T-MoS2) (as a cocatalyst) could effectively catalyze photochemical N2 reduction [170]. The optimized composites presented a remarkable NH3 rate as high as 8,220.83 μmol·L −1 ·h −1 ·g −1 under simulated solar light irradiation ( Fig. 17(a)). The outstanding photocatalytic activity was speculated to result from the synergy of SVs, O-doping, and more metallic 1T phase, which extended visible light absorption and also promoted the separation and migration of photogenerated electrons (Fig. 17(b)). Nevertheless, the purity of N2 used in this work was not provided. In addition, 15 N2 isotope labeling experiments were not conducted to confirm the origin of NH3 formed. An Ru-Co bimetal center at the interface of Ru/CoSx with S-vacancy on graphitic carbon nitride nanosheets (Ru-VS-CoS/CN) was designed for N2 reduction, delivering an NH3 yield rate of 0.438 mmol·g −1 ·h −1 with an apparent quantum efficiency of 1.28% at 400 nm and solar-toammonia efficiency of 0.042% in pure water (Fig. 17(c)) [168]. The two N atoms in N2 were posited to be bridged to the Ru-Co center. The asymmetrical electron donation from Ru and Co atoms to the N2 adsorbate polarized the N≡N bond to double bond order. The Schottky barrier between Ru and CoSx endowed an interface with plasmonic electron transfer from CoSx to Ru, favoring hydrogenation of the Ru-end bound N at the Ru-Co center (Fig. 17(d)).

Carbon vacancies (anion defects)
CVs are the most common anion vacancy present in carbonbased materials [286]. Like OVs, NVs, and SVs, manipulation of CVs enables tuning of the electronic structure of materials. Meanwhile, CVs can act as active sites to enhance adsorption and activation of N2 [153]. For instance, integration of CVs and iodine doping led to lowered band gap and enhanced separation efficiency of the photo-generated charge carriers over g-C3N4, promoting the N2 reduction to NH3 with an NH3 yield rate of 200.8 mg·L -1 ·gcat. -1 under simulated sunlight irradiation within 3 h, 2.8 times as high as that of bulk g-C3N4 (Figs. 18(a) and 18(b)) [272]. The CVs on the surface of g-C3N4 were proposed as active centers for N2 adsorption and N≡N triple bond activation, affording an NH3 production rate of up to 54 mmol·L -1 within 100 min (Fig. 18(c)) [287].

Fluorine vacancy (anion defects)
FVs are recognized as an effective anion defect while they are rarely discussed especially for photocatalytic N2 reduction. Pr 3+ :CeF3 supported on one-dimensional ATP mineral was demonstrated to be active for N2 photoreduction [171]. The prominent activity was attributed to the abundant FVs which served as active sites to facilitate N2 adsorption and weaken the N≡N triple bond during the photocatalytic reaction. Moreover, FVs were postulated to act as a mediator to facilitate the formation of a Z-scheme structure [279]. Yb 3+ and Tm 3+ co-doping of LaF3 anchored on palygorskite (Pal) (LaF3:Yb 3+ , Tm 3+ /Pal) was shown to promote the photocatalytic N2 reduction [288]. An NH3 formation rate of 5.7 mg·L -1 within 3 h was attained even under NIR light irradiation. The good photocatalytic performance was supposed to arise from the upconversion capability of LaF3:Yb 3+ Tm3+, which converted NIR into visible and UV light, drastically increasing the utilization of sunlight. Furthermore, the indirect Z-scheme heterostructure comprising LaF3:Yb 3+ , Tm 3+ and Pal could be effectively mediated by FVs to promote photogenerated carrier separation and migration and also retain the redox capacity.

Cation defects
In addition to anion vacancies, metal cation vacancies can also affect the physicochemical and electronic properties of metal compounds, originating from their specific orbital distribution with characteristic electronic configurations [289]. Introduction of cation vacancies can change the semiconductor conductivity from n-type to p-type and facilitate the separation of photogenerated carriers. Compared with anion vacancies, clarification of the role of metal cation vacancies is more challenging given their high formation energy and seldomly reported experimental protocols [59]. Ni12P5/ZnIn2S4 hybrids with zinc vacancies (ZnVs) were fabricated and used as a dualfunctional photoredox catalyst with holes to oxidize benzyl alcohol and electrons to reduce N2 [290]. The as-prepared Ni12P5/ZnIn2S4 heterostructures with ZnVs showed promising capability for photocatalytic N2 fixation coupled with benzyl alcohol oxidation. Recently, a Ti-based metal organic framework Ti8O8(OH)4(BDC)6 (MIL-125 (Ti)) was firstly constructed and employed for photocatalytic N2 fixation [210]. Ligand functionalization was demonstrated to extend light harvesting of the metal organic framework (MOF) to visible region of up to 550 nm ( Fig. 19(a)). The electron transfer from ligand to metal induced Ti 3+ species in the form of Ti8 clusters with defect sites as proved by EPR analysis (Fig. 19(b)). Enriched photogenerated electrons were observed to inject from the Ti 3+ to N2 by time-resolved photoluminescence decay measurements (Fig. 19(c)). Furthermore, integration of Ti sites and aminefunctionalized linkers enabled an enhanced ammonia evolution rate of 12.3 μmol·g −1 ·h −1 (Fig. 19(d)). 15 N2 isotopic labeling experiments confirmed the true reduction of N2 (Fig. 19(e)). It was envisioned that electrons were transferred from the organic ligand to Ti 4+ forming Ti 3+ , while the Ti8 clusters with defect sites adsorbed and activated N2 molecules to yield NH3 and then reacted with Ti 3+ to reproduce Ti 4+ under visible light irradiation ( Fig. 19(f)).

Synergism of cation and anion defects
It is discovered that multifold defects, such as cation and anion defects, cation and coordinatively unsaturated metal atoms, exhibit synergistic effects for the photocatalytic N2 reduction to NH3 [148,291]. The synergistic effects not only extend the range of light response and promote efficient separation and migration of photogenerated electron-hole pairs, but also enhance the adsorption and activation of N2 molecules [292]. For instance, two-dimensional oxidized Sb nanosheets with Sb and OVs were reported through controlled liquid exfoliation of bulk Sb [167]. The 2D defective Sb contributed to an NH3 formation rate of 388.5 μgNH 3 ·h −1 ·gcat. −1 , nearly 8 times higher than that for bulk Sb. Further DFT calculations revealed that the anion-cation vacancy pairs (Sb and OVs) acted as abundant surface catalytic sites, which significantly promoted the formation of *NNH, thereby facilitating the photocatalytic N2 reduction. Single-unit-cell Bi3O4Br nanosheets with "Bi-O" vacancy pairs were synthesized and applied for N2 photoreduction (Figs. 20(a) and 20(b)) [173]. A number of point defects associated with Bi vacancies were observed on defect-rich   Bi3O4Br. The Bi3O4Br nanosheets with "Bi-O" vacancy pairs were active for photocatalytic N2 reduction with an NH3 yield rate of 50.8 μmol·g −1 ·h −1 , more than 9.2 and 30.9 times compared to the defect-deficient Bi3O4Br and bulk Bi3O4Br, respectively ( Fig. 20(a)). The enhanced photocatalytic activity was supposed to be related with the increased charge separation efficiency based on density of state (DOS) calculations and ultrafast transient absorption (TA) spectroscopy. DFT calculations uncovered that the defective structure of Bi3O4Br could enhance the adsorption of N2. Furthermore, by virtue of 15 N2 isotope labeling experiments, it was verified that the evolved NH3 stemmed from the actual reduction of N2. Likewise, ZnCr-LDH nanosheets with OVs were fabricated [165]. The nanosheets contained abundant coordinately unsaturated metal active sites (zinc vacancies) with OVs which served as the sites for N2 chemisorption. Unsaturated Zn sites were calculated to have a more negative adsorption free energy value (−0.45 eV) than saturated Zn sites (−0.17 eV) (Fig. 20(c)), suggesting that the unsaturated Zn sites promoted both N2 adsorption and activation. DFT calculations further illustrated that the unsaturated Zn sites could be more easily generated after introducing zinc vacancies ( Fig. 20(d)). Consequently, the ZnCr-1 h nanosheets rich in oxygen and cation vacancies provided an NH3 yield rate of 33.19 μmol·g −1 ·h −1 (Fig. 20(e)) with quantum efficiency of 0.95% at 380 nm ( Fig. 20(f)). In addition to enhanced adsorption and activation of N2 due to the oxygen and cation vacancies as revealed by XAFS and DFT calculations, separation, and migration of photogenerated carriers were also promoted evidenced by photocurrent and TR-PL measurements (Figs. 20(g) and 20(h)).

Heteroatom doping
Heteroatom doping has emerged as another effective and feasible strategy for enhancing the N2 reduction. The introduction of foreign atoms with different physical (such as types, concentrations, and distribution locations, etc.) and chemical properties (such as atom radius, electron density, and chemical valence, etc.) can modulate the local electronic structure and electronic properties of semiconductor materials [293][294][295][296]. Heteroatom doping can be mainly divided into metal element doping and nonmetal element doping, depending on the types of dopants [82].

Metal doping
Metal doping mainly focuses on transition metals (TMs) because TM species can bind with N2 at low temperatures. The nature of the interaction between the TMs and N2 primarily depends on the "acceptance-donation" of electrons, where the combination of empty and occupied d orbitals plays an indispensable role [297,298]. The TM center can accept the lone pair of electrons of N2 through the empty d orbital [299]. The backdonation of d electrons from the TMs to N2 can strengthen the N-TM bonds and also simultaneously weaken the N≡N bonds, facilitating effective N2 binding and activation [300]. TMs such as Fe, Mo, Mn, etc. have been widely employed as dopants to enhance N2 reduction [301][302][303]. For example, doping of ultrathin W18O49 nanowires by Mo atoms showed efficiency in improving photocatalytic N2 reduction [66]. EPR spectroscopy revealed co-existence of coordinatively unsaturated Mo (Mo 5+ ) and oxygen defects in the samples (Figs. 21(a) and 21(b)). The optimized Mo-W18O49 with 1 mol.% Mo was observed to possess effective photocatalytic N2 reduction activity with an NH3 production rate of 195.5 μmol·gcat. −1 ·h −1 under full spectrum (Fig. 21(c)). Two aspects associated with such high performance were inferred: (1) elevation of defect-band center toward the Fermi level, thus preserving the energy of photoexcited electrons for N2 reduction (Fig. 21(d)); (2) polarization of chemisorbed N2 molecules and acceleration of electron transfer from coordinatively unsaturated Mo sites to N2 adsorbates, hence boosting dissociation of N≡N bonds through proton coupling (Figs. 21(e) and 21(f)). Despite this efficiency, the true origin of NH3 formed was not examined by performing 15 N2 isotope labeling. Analogously, Mn-doped W18O49 microspheres (Mn-W18O49) were synthesized as a photocatalyst for N2 reduction [304]. The Mn 2+ ions were supposed to replace some W sites in the W18O49 lattice and intervene the formation of microsphere-like structure. The 3.0% Mn-doped W18O49 exhibited an optimal N2 reduction activity affording an NH3 production rate of 97.9 μmol·g −1 ·h −1 under Xe-lamp's full spectrum irradiation (Fig. 21(g)).
It appears that Mn 2+ -doping extended light response range (Fig. 21(h)) and promoted photoinduced charge separation and migration. Meanwhile, the dopant served as chemisorption and activation centers of N2 and H2O molecules, boosting the transfer of photogenerated electrons to adsorbed N2 molecules ( Fig. 21(i)). Alternatively, Fe was incorporated into Bi2MoO6 to enhance the NH3 production rate from 28.2 μmol·g −1 ·h −1 to 106.5 μmol·g −1 ·h −1 under visible light irradiation ( Fig. 21(j)) [305]. The enhanced photocatalytic activity was mainly ascribed to lowering of surface work function after Fe doping, thereby facilitating transport of photoinduced charges and prolonging the lifetime of photogenerated carriers (Figs. 21(k) and 21(l)).

Nonmetal doping
Nonmetal doping provides a promising and efficient method to boost the N2 reduction by extending light response range and accelerating carrier transport [306,307]. C doping was shown to narrow the bandgaps of anatase titanium oxide nanosheets and also facilitate the generation of highconcentration Ti 3+ sites, thereby promoting the photocatalytic N2 reduction for NH3 synthesis (Fig. 22) [308]. An optimal C4-TiOx with a Ti 3+ /Ti 4+ ratio of 72.1% was obtained and showed an NH3 production rate of 109.3 μmol·g −1 ·h −1 and an apparent quantum efficiency of 1.1% at 400 nm ( Fig. 22(b)). Likewise, doping of oxygen vacancy-rich anatase nanoplatelets by S (VO-S-rich TiO2−xSy) was demonstrated to improve ammonia yield rate under visible-NIR irradiation [309]. The as-made VO-S-rich TiO2−xSy exhibited a narrowed bandgap of 1.18 eV and stronger light trapping that extended from UV light to the near-infrared (NIR) region ( Fig. 22(d)). This was attributed to the co-presence of oxygen vacancies and sulfur dopants. Upon further increasing the concentrations of oxygen vacancy and sulfur anion, the N2 adsorption energy was markedly enhanced from −0.76 to −2.1 eV, indicating that the N2 adsorption on VO-S-rich TiO2−xSy was energetically more favorable than that on VO-S-poor TiO2−xSy (Fig. 22(e)). The effects of OVs and sulfur anion dopants on the bandgap and adsorption energy of N2 molecule on different TiO2−xSy- (101) surfaces are summarized in Fig. 22(f). These results assuredly suggest that the synergistic effect of OVs and sulfur dopants could extend the light absorption, narrow the bandgap of titania, and also promote the adsorption of inert N2, contributing to improved photocatalytic N2 reduction.
Additionally, nonmetal doping can help improve the stability of photocatalytic materials for N2 reduction. For instance, B-doped g-C3N4 (BCN) with B-N-C coordination was synthesized [159]. Based on redshift of UV-vis peaks and strong PL quenching as well as enhanced photocurrent, it was speculated that B-doping both effectively improved the visible-light absorption and inhibited the recombination of photogenerated electron-hole pairs in g-C3N4 (Figs. 22(g)-22(i)). The as-synthesized BCN delivered an NH3 yield rate of 27.3 μg·h −1 ·mgcat. −1 (Fig. 22(j)). The exposed active N atoms in g-C3N4 nanosheets were posited to participate in the NH3 formation during the N2 reduction process. These exposed N atoms were stabilized by forming B-N-C coordination in g-C3N4 nanosheets, as confirmed by 15 N2 isotopic labeling experiments (Figs. 22(k) and 22(l)). Incorporation of S into g-C3N4 porous nanosheets (SCNNSs) with CVs was reported to enhance N2 reduction [273]. The adsorption energy of SCNNSs-550 was calculated to be −0.665 eV, 1.99 times as high as that of bulk SCN (−0.335 eV). This suggests that sulfur doping and CVs provided abundant active sites, thus promoting the N2 adsorption and activation.

Single atom catalysts
Single-atom catalysts (SACs), defined as a kind of supported catalyst that only contains a single metal site isolated from each other as the catalytic active center [310], display distinct catalytic activities because of low-coordination environment of the metal center, quantum size effects, and metal-support interactions. SACs could be proposed as a general "defect" definition [38,311]. A single atom is commonly assumed to be a "defect" of the substrate material, where new chemical bonds would be formed between the atoms and substrate material, significantly regulating their electronic structures, SACs show promise in accelerating photocatalytic N2 reduction [54,[312][313][314][315].
Single Cu atoms were anchored on the defects of carbon nitride nanosheets (Cu-CN) and used as photocatalysts for N2 reduction [316]. Based on EPR measurements, isolated valence electrons were found to be easily excited to generate free electrons under photo-illumination, which originated from the abundant defects or edge amino group. The single Cu sites afforded an NH3 production rate of 186 μmol·g −1 ·h −1 under visible light irradiation with a quantum efficiency reaching 1.01% at 420 nm monochromatic light illumination. The good performance was hypothesized to result from the suppression of recombination of photogenerated excitons and efficient separation of the photoexcited electron-hole pairs. Single Ru atoms were recently decorated on TiO2 nanosheets rich in OVs for photocatalytic N2 reduction ( Fig. 23(a)) [266]. Isolated Ru atoms were shown to effectively promote ammonia generation with a yield rate of 56.3 μg·g −1 ·h −1 (Fig. 23(b)). Experiments in combination with DFT calculations revealed that atomically dispersed Ru greatly promoted chemisorption of N2, and improved charge carrier separation, hence enhancing N2 photoreduction to NH3 (Fig. 23(c)).
Single B atoms attached to g-C3N4 (on H-vacancy; a kind of C vacancy) were predicted as a potential metal-free photocatalyst for N2 reduction [317]. The main group IIIA B atoms were proposed to serve as catalytic sites for N2 bonding and an N-B-N active center was formed within g-C3N4 for photocatalytic N2 reduction (Figs. 23(d) and 23(e)). It was envisioned that N2 couldbond with B on g-C3N4 either by side-on or end-on mode, corresponding to an adsorption energy of −1.04 and −1.28 eV, respectively. The charge density difference revealed that the single B atom functioned through electron acceptance or backdonation within the B-N2 configuration, akin to TMs (Figs. 23(f) and 23(g)). Additionally, the decoration of B on g-C3N4 lowered the band gap of g-C3N4 and enhanced the visible light and infrared light harvesting (Fig. 23(h)).

Combined strategies
Relative to a single protocol, a combination of different design schemes casts considerable potentials for jointly/collaboratively improving the photocatalytic N2 reduction activity. This area has stimulated heightened research interest.

Coupling defect engineering and morphology control
The morphology of semiconductors plays a critical role in their photocatalytic performance. Nanostructuring by controlling the desired morphologies (e.g., particle size, shape, geometrical feature, and surface texture) can maximize the exposure of active sites, accelerate carrier transportation and migration, being beneficial to photocatalysis [80]. Defect engineering in combination with morphology control allows one to further improve photocatalytic N2 reduction activity, taking full advantages of joint/synergistic effects [260]. Such concept has been demonstrated by construction of light-switchable oxygen vacancies in ultrafine Bi5O7Br nanotubes, which realized active and stable photocatalytic N2 reduction [180]. The enhanced photocatalytic activity was postulated to stem from a synergy of oxygen vacancies and ultrafine nanotubes, which provided large surface areas and simultaneously promoted efficient separation and migration of photogenerated carriers, in addition to enhanced N2 adsorption and activation.

Coupling defect engineering and amorphization
Amorphization can introduce dangling bonds and unsaturated coordination sites in semiconductors, endowing active sites for N2 reduction reactions [318]. Combining defect engineering and amorphization can strikingly modulate the electronic structure of photocatalysts, thereby promoting N2 adsorption and activation. However, the stability and longevity of such catalysts remain to be addressed.

Coupling defect engineering and cocatalyst
Incorporation of suitable cocatalysts can improve the photocatalytic efficiency by lowering activation energy or overpotential, facilitating electron-hole pairs separation, and increasing the photocatalyst stability [319]. Defect engineering along with cocatalyst modification has been extensively employed to further boost the N2 reduction photocatalysis [168].

Coupling defect engineering and heterojunction construction
Formation of a heterostructure can extend spectral range of light absorption, and also induce a built-in electric field to facilitate the separation and migration of electron-hole pairs, favoring photocatalysis [320]. Integration of defect engineering and heterostructuring can present a strong synergistic effect to drastically enhance the N2 photoreduction [321].

Coupling defect engineering and external field
Aside from the inherent driving force of photocatalysis, introduction of an external field can provide additional energy to promote light generation and charge separation and migration, thereby improving the overall catalytic efficiency [322]. This emerging scheme has been recently combined with defect engineering in photocatalytic N2 reduction. A remarkably high NH3 yield rate exceeding 1.93 mg·L −1 ·h −1 was attained over defective BaTiO3 under an applied magnetic field [323]. Such significant enhancement was mainly attributed to boosted N2 adsorption by oxygen vacancies and suppression of photoexcited carriers recombination arising from an electromagnetic synergistic effect between the internal electric field of the photocatalyst and an external magnetic field.

Summary, challenges, and future perspectives in defect-engineered N 2 photocatalysis
Defect engineering of a semiconductor facilitates N2 photocatalysis by virtue of extending the range of light response, promoting the efficient separation and migration of photogenerated electrons-hole pairs, and improving the N2 adsorption and activation, all of which are the critical steps in N2 reduction to NH3. Although significant and encouraging progress has been acquired in N2 photocatalysis in the past five years, the attained solar energy conversion efficiency (SCE) of N2 photoreduction is still less than 1%, far below the minimum standard of 10% needed to realize industrialization [324]. The NH3 yield rates of certain photocatalysts have increased from μmol·g −1 ·h −1 to mmol·g −1 ·h −1 , which is however still much below the requirements for practical applications. There are a number of scientific and technological challenges that need to be overcome to accomplish large-scale application of defectengineered N2 photocatalysis.
Only by accurately controlling the type, concentration, and location of defects can one rationally design and construct defect-engineered photocatalysts. However, the existing synthetic methods are incapable of achieving the defect control at such a high level [217]. In situ creation of OVs on the catalyst surface (such as caused by light irradiation method) during the photoreduction process is considered as a feasible and promising strategy [180]. Introduction of specific defects with different concentrations could be achieved by accurate amounts of heteroatom doping into photocatalysts. Controllable production and tuning of defects require more future research efforts. It is known that excessive amounts of defects in a photocatalyst lead to reunion of excited carriers. To avoid such issue, it is important to precisely quantify the concentration of defects. Equally importantly, the cooperative mechanisms among multiple defects and role of each individual defect in N2 photocatalysis remain to be elucidated.
To date, major endeavors toward defect-engineered N2 photocatalysis have been made on traditional semiconductors. Some emerging photocatalytic materials with defects such as MOFs and covalent organic frameworks (COFs) have not been fully explored for photocatalytic N2 reduction. Meanwhile, the design of cation defects and further combined defects and investigation of their roles in N2 reduction are rarely conducted, which deserves further studies. To synergistically promote N2 photocatalysis, integration of multiple design strategies (e.g., defect engineering and other modification strategies such as creation of Z-scheme heterostructures to separate ammonia production and water oxidation sites in space) is preferred. Additionally, combination of defect engineering and external fields (e.g., microwaves, mechanical stress, temperature gradient, electric field, magnetic field, and coupled fields) is another promising strategy to further boost photocatalytic N2 reduction reactions [322].
Although a range of techniques have been reported to probe defects (e.g., HRTEM, XRD, SE, etc.), most of them are limited to qualitative characterization and cannot quantitatively determine the defect concentrations. To gain insight into the defect structure-performance relationships for the N2 reduction process, it is urgently desired to exploit and develop more advanced in situ/operando characterization techniques which is however still in its early stage. In situ electrochemical tip-enhanced Raman technique (EC-TERS) enables imaging of changes in the surface defect structure of photocatalysts. We expect that this emerging technique may provide a platform for real-time monitoring of defect state and its concentration during the N2 reduction process in the near future after some technical advances.
The impurity composition of N2 supply must be specified. Highly pure N2 and Ar with purity over 99.999% are strongly suggested for use in the N2 photoreduction. Before starting the N2 photoreduction, the N2 feed gas and photocatalysts should be subjected to purification treatments to exclude artifacts from exogenous nitrogen contaminants (NH3, nitrate/nitrite, and nitrous oxide). Particularly, for photocatalysts that either contain structural/lattice nitrogen or are synthesized from nitrates or ammonium precursors, 15 N2 isotope labeling should be performed to verify genuine N2 reduction. Currently, the spectrophotometric methods including indophenol blue and Nessler's reagent spectrophotometry are commonly used for quantification of NH3. However, the accuracy of these spectrophotometric methods is usually interfered with several factors, such as pH, ionic strength, and sacrificial reagents. Isotope labeling experiments to analyze and confirm mechanistic studies have been applied too rarely up to now. To provide unequivocal evidences for photocatalytic N2 reduction to ammonia and obtain reliable NH3 yield, quantitative isotope measurements by making essential use of 15 N2 along with multiple control experiments are required. It is also necessary to develop more selective, precise, and in situ methods for NH3 determination.
Although theoretical simulations are a powerful approach for study of defect-engineered photocatalysts, the simulated model cannot predict the true structure of the catalyst. Greater advances are expected through the collective knowledge and insights to be gained from fundamental research that integrates experiments and theory. Development of in situ/operando techniques will help build up more valid theoretical models and deepen clarification of reaction pathways, degradation modes as well as catalytic structure-property relationships specific to given materials, further guiding new direction for catalyst design.
format, as long as you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons licence, and indicate if changes were made.
The images or other third party material in this article are included in the article's Creative Commons licence, unless indicated otherwise in a credit line to the material. If material is not included in the article's Creative Commons licence and your intended use is not permitted by statutory regulation or exceeds the permitted use, you will need to obtain permission directly from the copyright holder.