Applied Catalysis A, General Minimizing energy demand and environmental impact for sustainable NH 3 and H 2 O 2 production — A perspective on contributions from thermal, electro-, and photo-catalysis

and of the of an in the synthesis of peroxide that is used in decontamination and that is used as However, these are currently produced with processes that are either very energy-intensive or unfriendly. This the perspectives of the and opportunities in the production of these chemicals, focusing on the roles of catalysis in more sustainable, alternative production methods that minimize energy consumption and environmental impact. While not intended to be a compre- review, the article provides a critical review of selected literature relevant to its objectives, discusses areas needed for further research, and potential new directions inspired by new developments in related ﬁ elds. For each chemical, production by thermal, electro-, and photo-excited processes are discussed. Problems that are common to these approaches and their di ﬀ erences are identi ﬁ ed and possible solutions suggested.


Introduction
Adequate supplies of water, energy, and food to satisfy an expanding population are essential to sustainable global development and maintaining world peace. These needs are intertwined in a complex manner. For example, the food sector accounts for about 30 % of total global energy consumption and 20 % of greenhouse gas (GHG) emissions [1], even though almost 800 million people still suffer from hunger [2]. In addition, agriculture is responsible for 69 % of global freshwater withdrawal [3]. Without question, there is an urgent need to implement environmentally benign processes that could reduce the demands for energy and natural resources in meeting the dual needs for food and water. Catalysis will play important roles in these processes.
This perspective article is intended as a springboard for discussion of the future roles of catalysis in the syntheses of two high volume commodity chemicals: ammonia and hydrogen peroxide. In addition to the fact that these two chemicals are critical in sustaining the global supply of food and water via fertilizer production and wastewater clean-up, respectively, they are also playing increasingly prominent roles in the move towards greener synthesis of chemicals. H 2 O 2 is a preferred oxidant in many chemical syntheses as it produces water as the benign byproduct. An example is the commercial process to produce propylene oxide (the HPPO process) using H 2 O 2 with a titanium silicalite catalyst [4]. As another example, ammonia together with hydrogen peroxide are being actively explored for ammoximation to produce oximes using Ti silicalite catalysts [5]. There are recent reports of using these two chemicals to react with small ketones to form lowcarbon-chain nitroalkanes [6].
The current processes to produce these two chemicals, unfortunately, have significant environmental impact. Ammonia is manufactured by the energy intensive Haber-Bosch process using catalysts that facilitate the reaction of nitrogen with hydrogen. Although direct combination of N 2 with H 2 to form NH 3 is thermodynamically favorable at standard state (ΔG°f = -16.2 kJ mol −1 at 25°C), the slow kinetics necessitates operation at elevated temperatures of 350-500°C where the thermodynamics becomes less favorable (ΔG°f = 27.1 kJ mol −1 at 427°C), and high pressures ranging from 20 to 40 MPa. The H 2 source is derived from reforming of hydrocarbons using processes that also entail high temperatures. For example, the principal route for hydrogen production in the U.S. is steam reforming of methane, which is thermodynamically favorable only above ∼530°C, when the reaction is highly endothermic (ΔH°r xn = 222 kJ mol −1 ). The U.S. Department of Energy estimates recently that the on-site energy consumption for ammonia production in the country amounts to ∼200 × 10 15 J a -1 (J/year) based on a production of 15 × 10 6 t (metric tonnes) in 2018 (14 × 10 3 kJ/kg) [7]. For a global annual production of 170 × 10 6 t [8], the global annual process energy consumption would be 10 18 -10 19 J, which is 1-2 % of the total global energy consumption of 5.9 × 10 20 J. H 2 O 2 is a versatile and powerful oxidant used as an environmentally friendly oxidant for chemical synthesis as well as in the advanced oxidation process (AOP) for decontamination of water by mineralizing dissolved organics through oxidation. The global production of H 2 O 2 is 4.5 × 10 6 t in 2014 [9], and about 10 % of it is produced in the U.S. (3.9 × 10 5 t in 2010 [10]). More than 95 % of H 2 O 2 is synthesized by the anthraquinone process that involves auto-oxidation of hydrogenated anthraquinone or its derivatives [11]. The process poses significant environmental burden due to the need to regenerate hydrogenated anthraquinone and separation and purification of the produced H 2 O 2 from large quantities of organic solvents and catalysts. It is also energy-intensive, consuming 16 × 10 6 J/kg even though the overall process (H 2 +O 2 = H 2 O 2 ) is thermodynamically favorable and requires no energy input (ΔG°f = −2.6 × 10 6 J kg -1 , ΔH°f = −5.5 × 10 6 J kg -1 ) [10].
Thus, reducing the energy demand and decreasing the ecological footprint in the production of these two chemicals would contribute to sustainable global development. In this perspective article, both thermal and photo/electrochemical processes in ammonia and hydrogen peroxide production are discussed. Although thermal processes are well developed, there is room for further improvement with the advent of recent new understanding in catalyst structure-property relationships. The historic economic disadvantages of electrocatalytic synthesis due to high cost of electricity relative to petrochemical fuels and feedstocks are no longer valid as the cost of renewable electricity has lowered consequent of advances in materials, engineering, and efficiencies in capturing solar and wind energy [12,13].
This article is not intended to be a comprehensive review of catalysis on the production of these two chemicals, since there are many excellent recent reviews on the direct synthesis of hydrogen peroxide [11,[14][15][16][17] and ammonia synthesis [16,[18][19][20][21][22][23][24][25][26][27][28]. Instead, the article focuses on potential roles and research opportunities to improve existing catalysts and production methods that have not been commercialized. Thus, review of the literature will be critical but brief and confined to those directly relevant to the ensuing discussion. For H 2 O 2 , production by direct synthesis with H 2 and O 2 (i.e. reduction of O 2 with H 2 ), reduction of O 2 with simultaneous oxidation of another molecule which may form a useful co-product, and photoelectrochemical (PEC) oxidation of H 2 O will be discussed. For NH 3 production, the discussion centers on improving the catalyst in the Haber-Bosch process and exploring the alternative nitride looping process, and the potential for photo/electrochemical production.

H 2 O 2 production
2.1. Thermal catalysis 2.1.1. Direct H 2 O 2 production from H 2 and O 2 The direct synthesis of hydrogen peroxide (DSHP) from H 2 and O 2 offers the simplest route and the highest atom utilization efficiency of all known processes. It has been explored as an alternative to the commercial anthraquinone process. One particularly interesting aspect is the direct, in situ production in which the H 2 O 2 formed is used immediately without separation or purification, and thus offering potentially significant cost reduction. For example, in situ production of H 2 O 2 could reduce the raw material cost for propylene oxidation to propylene oxide by up to 45 %, and the total production cost by 24 % compared with using purified H 2 O 2 [29]. However, there are significant barriers that need to be overcome before the benefits of DSHP can be realized. These barriers include the danger of explosion with a H 2 /O 2 mixture and the competition by various facile side reactions, such as decomposition of H 2 O 2 , hydrogenation of H 2 O 2 , and oxidation of H 2 to H 2 O that adversely affect the H 2 O 2 yield (Fig. 1). If H 2 O 2 is produced in the presence of organic compounds, H 2 O 2 oxidation of the organics also needs to be minimized. To overcome these issues, there is intensive ongoing research that focuses on the design of efficient catalysts and processes to accommodate the requirements for commercial applications [11,15,[30][31][32].
It is widely accepted that the key reaction steps of the DSHP include (i) dissociation of adsorbed H 2 molecule to H atoms on an active metal surface, (ii) formation of a *OOH intermediate by reaction of the H atom with an adsorbed O 2 molecule (where * denotes an active site), and (iii) formation and desorption of H 2 O 2 by reaction of *OOH intermediate with another H atom. In order to attain high H 2 O 2 selectivity, it is essential to suppress the dissociation of the OeO bond, which results in reactive intermediates such as *O or *OH that eventually leads to wasteful H 2 O formation. It is equally important to prevent the subsequent, thermodynamically feasible side reactions of decomposition and hydrogenation of produced H 2 O 2 . For example, it has been reported that for Pd/C, H 2 O 2 hydrogenation is as much as 3-4 times faster than H 2 O 2 decomposition, decreasing the H 2 O 2 selectivity by 70-85 % [33]. Therefore, an ideal catalyst should selectively promote H 2 O 2 formation without also promoting these unwanted side-reactions. The approaches being investigated include fine-tuning the physicochemical properties of the active metals via bimetal formation, addition of modifiers, and/or modifying the particle size, phase, and electronic structure. For practical purposes, improving the stability of the catalyst and understanding the effect of the reaction medium are equally important.
There has been impressive improvement in recent years in the catalyst performance for DSHP. However, there is insufficient detailed understanding of the underlying reasons for the improvement because experiments performed to identify the best catalysts generally utilize metal particles of a distribution of sizes with different exposed crystal planes and, for bimetallic catalysts, particles with a range of compositions. Model catalysts that are structurally and compositionally defined have been employed to generate in-depth understanding. Since Pd is considered the most active metal for DSHP, it is the most heavily studied.
2.1.1.1. Nature of the active sites. It has been shown that H 2 O 2 production activity depends on the exposed crystal planes of Pd. The premise is that crystal planes where OeO bond scission is suppressed favors high selectivity for H 2 O 2 . Zhou and Lee synthesized colloidal Pd particles in the presence of ionic polymers to obtain particles with different preferentially exposed crystal planes, and demonstrated that the (110) planes exhibited higher H 2 O 2 selectivity compared with the (100) and (111) planes [34]. Kim, et al. compared SiO 2 -supported Pd nanocubes with exposed (100) planes and nanooctahedra with (111) exposed planes, and found the nanooctahedra to be more selective for H 2 O 2 production than the nanocubes at similar conversions [35]. They noted that the observation of a higher selectivity for the (111) than the (100) surface planes agrees with the DFT computational results of O 2 adsorption and dissociation on low index planes of Pd, which showed that the barrier for O 2 dissociation was higher on a Pd (111) surface than on (110) or (100) [36]. Unlike adsorption on the (100) or (110) planes, adsorption of O 2 on Pd (111) leads mostly to placing two surface O atoms that are closer to each other than the molecular dissociation distance, thus encouraging associative adsorption which favors production of H 2 O 2 [36].
It should be noted that although it is established that the atomic structure of the exposed plane of an active metal critically influences catalytic performance, the detailed arrangement of metal atoms of a desired active site for H 2 O 2 formation remains unresolved. Shape control of an active metal particle can be challenging in practice. The most stable surface under reaction conditions may be different from the one most selective for H 2 O 2 formation. The effect of support, especially on the geometric and electronic properties of the active metal needs to be examined and understood.
The dependence of selectivity on the structure of the exposed plane likely contributes to the observed particle size dependence. It was reported that on a series of Pd/hydroxyapatite, samples with 1−2 nm particles were much more selective than those with larger particles, whereas samples with isolated Pd were inactive [37]. The higher selectivity of the 1−2 nm particles was attributed to the presence of higher ratios of Pd δ+ /Pd o with the support stabilizing the partially charged Pd atoms. However, Burch et al. [38] observed that for supported Pd catalysts, generally the reduced catalysts exhibited higher H 2 conversion activity and H 2 O 2 selectivity. Flaherty et al. [15] noted an absence of an induction period for H 2 O 2 formation with Pd nanoparticles whereas a noticeable induction period existed for PdO particles, suggesting a need to reduce the PdO for activity. These apparent variant in observations suggests the potential that there exists an optimal ratio of ionic and metallic Pd and their distribution/location on the surface. The challenge is to identify this optimum, devise a system to maintain it under reaction conditions, and understand the underlying reaction mechanism. Such an understanding should also improve our ability to identify beneficial modifiers to Pd, such as halides and a second metal.

Effect of catalyst modifiers.
A mechanism for H 2 O 2 production involves formation of η 2 -adsorbed O 2 and its subsequent hydrogenation to *OOH [39]. Thus, catalyst modifiers that alter the stability and/or barrier to the formation of these species would affect the reaction. DFT computational results suggest that the energy barrier for hydrogenation of *OOH is lower than for OeO bond scission on isolated Pd atoms but not on ensembles [40]. Thus, modifiers that break up Pd ensembles should be beneficial. In addition, modifiers that can reduce the extent of electron back-donation to 2π* orbitals of adsorbed O 2 would also reduce the tendency for OeO bond cleavage [15]. This mechanistic interpretation may explain in part the effect of chloride and bromide modifiers. In an acidic solution, these halides greatly suppress H 2 O 2 decomposition or hydrogenation while exhibiting much milder effects on H 2 O 2 formation [41]. It should be noted that other halides, such as iodide or fluoride are much less effective. That is, additional study is needed to fully understand the effect of halide modifiers.
Ligands and bimetallics are used also to modify the Pd active sites. Ligands or surfactants are often used in colloidal preparation to control the shape of metal particles. In one example, hexadecyl-2-hydroxyethyldimethyl ammonium dihydrogen phosphate (C 20 H 46 NO 5 P) was used as a stabilizing and reducing agent to prepare Pd colloidal particles deposited onto a carbon support. Compared to naked Pd particles on carbon, the stabilized Pd catalyst showed a higher H 2 O 2 selectivity [42]. It was suggested that on the surface modified with surfactant molecules, the peroxyl intermediate adsorbed in a vertical configuration instead of lying flat on the Pd surface, making cleavage of the OeO bond more difficult. The activation barrier for the hydrogenation of the *OOH intermediate to H 2 O 2 was lower than the barrier for over-hydrogenation to water, and the desorption of the product was energetically advantageous over its further hydrogenation and/or decomposition.
A number of modifying metals have been reported to be beneficial for Pd, including Au, Pt, Ag, Sn, Zn, and others [43][44][45]. The pioneering work of Hutchings and coworkers highlighted the potential for bimetallic catalysts [43], and AuPd seems to be one of the most efficient active metal combinations. In this system, Au acts as an electronic modifier of Pd. It decreases the rates of H 2 O formation much more than H 2 O 2 formation, resulting in an increase in H 2 O 2 selectivity [46]. In the case of a Pd/C catalyst, modifiers that block H 2 O 2 decomposition sites on the carbon support surface also contribute to selectivity enhancement [47]. These sites could originate from impurities. Density functional theory (DFT) calculations suggest that Au may inhibit the activation of H 2 and dissociation of O 2 , *OOH, and H 2 O 2 , and the AuPd may be an optimum bimetallic in terms of activity and selectivity [17,44]. Because the binding energies of O, OH, and OOH on a metal surface scale proportionally to each other [48][49][50], an optimal catalyst for H 2 O 2 formation activity would need to have an intermediate binding strength of H and O. Other metal modifiers such as Ni, Zn, or Sn that are beneficial probably exert similar geometric and electronic effects on the active sites of Pd [51][52][53].
Recently, Freakley et al. reported that in the Pd system, small metallic particles (< 4 nm) could be encapsulated more readily than larger particles with a thin (< 2 nm) layer of an amorphous oxide, such as Sn oxide, that are inactive for H 2 O 2 decomposition or hydrogenation. The encapsulation greatly suppressed the H 2 O 2 degradation activity of these particles [54]. Larger particles, though not encapsulated, could be made to produce H 2 O 2 with high activity and selectivity by partial oxidation of the surface Pd. Thus, by designing an oxidation-reduction-oxidation cycle to achieve both encapsulation of small particles and partial oxidation of larger ones, they obtained active and selective catalysts. This positive effect by encapsulation with oxide clusters was also observed with Ni, Ga, Zn, Co, and In modifiers. Interestingly, unlike the PdSn system, the oxidation-reduction-oxidation cycle was not beneficial for the AuPd/TiO 2 system [54], possibly because Au oxide is unstable at high temperatures and would not form an inactive Au oxide layer similar to Sn.
The above examples point to the importance of understanding the transformation of catalysts upon pretreatment and during reaction. Instead of comparing catalysts at a fixed reaction time, collecting temporal information would facilitate identification of the nature of the active sites, an important piece of information for catalyst development that is not commonly collected. For example, information from in situ characterization or, at least, after reaction would help explain the changes observed on a AuPd/TiO 2 catalyst that the initial rate of H 2 O 2 production was 114 mol h −1 kg −1 catalyst and the initial H 2 O 2 selectivity was 93 %, but deteriorated to 66 and 60 %, respectively, after 30 min [55].
In addition to modifiers and bimetallics, the reaction medium also plays an important role. The presence of gaseous H 2 suppresses significantly the rates of decomposition of H 2 O 2 to H 2 O and O 2 on both Au-Pd [56] and Pd catalysts [41]. The presence of protons is also very important. Protic acids are essential for the beneficial effect of Cl − and Br − [41]. Under identical conditions, H 2 O 2 formation rate is much lower in an aprotic than a protic solvent [46]. This suggests the possibility that hydrogenation of either *OO or *OOH may involve simultaneous proton and electron transfer from the catalyst to the adsorbate [39].
2.1.1.3. Catalyst stability. During operation, metal leaching is a concern. The primary cause of leaching is the strong ligation of surface metal atoms by species in the reaction medium. Metal ions and atoms at highly coordination unsaturation sites such as corners are particularly vulnerable. Metal leaching can be serious in strong acids and the presence of halide ions [41]. It was reported that stacking polyelectrolyte multilayers (PEMs) on a sulfonated resin can mitigate active metal leaching and enhance the H 2 O 2 yield [57,58]. Continuous production up to 100 h of 8-10 wt% H 2 O 2 was achieved by using Pd nanoparticle catalysts either encapsulated in or immobilized on the PEMs (Fig. 2).

Electrocatalytic synthesis
Direct synthesis of H 2 O 2 from H 2 and O 2 can also be achieved electrochemically at room temperature or below by combining two half-cell reactions: Because the overall reaction is thermodynamically favorable, the system can operate as a fuel cell to generate electricity during H 2 O 2 production ( Fig. 3) [59]. In early work, H 2 O 2 was generated under short-circuit conditions, using Au mesh and graphite membrane cathodes [60]. However, the resulting H 2 O 2 concentration was limited to 59 mM (0.2 wt%) and the current efficiency (that is, the selectivity for H 2 O 2 based on the quantity of H 2 reacted) was 10 % over a span of 10 h. This relatively poor performance was due to the low activity of the cathodes and low concentrations of O 2 in the catholyte. Subsequent work improved the catalyst and the delivery of O 2 [61]. For example, a cathodic catalyst prepared by pyrolysis of 0.05 wt.% Co-porphyrins (e.g. Co-TPP; TPP: 5,10,15,20-tetrakis(phenyl)-21H,23H-porphyrin) on vapor-grown carbon-fiber (VGCF) substrates achieved a production of pH-neutral, pure, 4.0 M (13.5 wt%) aqueous H 2 O 2 solutions [62]. Another report using an optimized carbon-based support gave a H 2 O 2 concentration of 5.5 M (18.7 wt%) in conjunction with a current efficiency of 55

Formation during parallel catalytic oxidation
The formation of H 2 O 2 in the liquid phase in parallel with oxidation of organics has been reported some time ago. Instead of reducing the O 2 with H 2 as in DSHP, an organic reductant is used with concomitant formation of an oxidized organic product. Although in principle this can be accomplished in the gas phase, available literature has been confined only to the liquid phase. Because energy-demanding separation would be needed to produce pure solutions of H 2 O 2 , the parallel-production process is best applied in the form of in situ generation of H 2 O 2 that is used immediately for selective oxidation to produce high-value products as the end goal, e.g. for selective epoxidation.
An obsolete industrial process to produce H 2 O 2 in parallel with oxidation of an organic compound involved autooxidation of 2-propanol [64,65]. Currently, the predominant industrial process for H 2 O 2 production is the anthraquinone autooxidation process that involves the cycle of hydrogenation of anthraquinone to anthrahydroquinone and its oxidation back to anthraquinone, while producing H 2 O 2 [66]. On the other hand, catalytic selective oxidation of hydrocarbons or other reductant other than H 2 with parallel formation of H 2 O 2 is much less studied. Only a limited number of catalytic systems have been reported in the literature, which include palladium (II) acetate-catalyzed oxidation of terminal alkenes to ketones in an organic solvent [67,68], 2-octanol oxidation [69], aqueous phase oxidation of glycerol catalyzed by Au/C [70], and oxidation of CO catalyzed by Au/TiO 2 [71] and Au/ Calgon [72].
The proposed mechanism for the Pd-complex catalyzed reaction involves first activation of O 2 to form a peroxy ligand on the metal. Insertion of the peroxy into a metal-hydride bond forms a hydroperoxy ligand [73,74], which is hydrogenated or hydrolyzed to form H 2 O 2 [75]. Similar to DSHP, retaining the OeO bond in the O 2 molecule is believed to be essential. There is little mechanistic information for heterogeneous catalytic processes, but a mechanism that includes retention of the OeO bond and the formation of adsorbed peroxy intermediate, leading to the subsequent formation of peroxide is consistent with isotopic labelling results of the Au-catalyzed CO oxidation-assisted selective oxidation of propylene to propylene oxide. It was observed that 18 O-labelled propylene oxide was formed exclusively when 18 O 2 was used but none when H 2 18 O was used [71]. Since it is reasonable to assume that propylene oxide is formed by reaction of propylene with H 2 O 2 that is formed in situ, the labelling experiments suggest that H 2 O 2 is formed directly from O 2 . Various mechanisms that retain these essential features have been postulated [76]. Interestingly, when H 2 O 2 was formed during oxidation of glycerol, 18

Opportunities and challenges for thermal and electrocatalytic processes
Among the processes discussed in this section, the DSHP process is the most studied and much more information is available for this catalytic system than the others. Thus, our discussion will center on the DSHP process while commenting on extension to other processes and focus on possible directions to explore and the knowledge gap for improving catalyst activity, selectivity, and stability. The important catalytic steps for discussion include activation of O 2 , selective hydrogenation of O 2 , and degradation of H 2 O 2 .
For processes that rely on reduction of O 2 , which include DSHP, formation in parallel with oxidation of organics or CO, and electrochemical reduction of O 2 , it is important to activate the O 2 molecule without dissociating the OeO bond. Thus, they would benefit from modifiers and surface atomic structures that reduce the MeO bond   strength and weakens the tendency of electron donation from the catalyst to the π* orbital of the adsorbed O 2 molecule. These processes may differ in the hydrogenation step of the adsorbed O 2 . In DSHP, H 2 is dissociatively adsorbed on a catalyst and both H atoms are utilized in H 2 O 2 formation by insertion into the surface MeO bond without dissociating the OeO bond. For production by parallel oxidation, the H atoms are derived from the organic molecules directly. Thus, the ability of a catalyst to cleave a CeH bond and facilitate hydride transfer to the *O 2 without cleaving the OeO bond is essential. If the H atom is derived from a proton and the ancillary electron, a situation that applies to electrochemical processes, the electronegativity of the surface that determines the potential for electron transfer to the adsorbed O 2 and proton is important, and the potential is expected to depend on the chemical nature of the surface and the morphology at the atomic scale [77,78]. Understanding this dependence is needed to predict the effects of catalyst/electrocatalyst modifiers. In all cases, the need to be able to transfer a H atom to O 2 without cleaving the molecule remains an essential criterion. It is important to better understand how these properties depend on the surface morphology, metal particle size, and the metal oxidation state distribution. More detailed characterization under catalytically relevant conditions would be highly desirable, especially in situ characterization.
There has been significant effort to optimize the addition of acid and halide promoters [31,38,39,79,80]. Chloride and bromide ions are generally used to suppress the degradative decomposition and hydrogenation of H 2 O 2 . Protons are necessary for the observed effects of halide ions on H 2 O 2 formation activity [15,31]. Unfortunately, rules for the effects of these promoters are not yet available, and at times combination of catalysts, additives, and reaction conditions may lead to adverse effects. Apparently, the reaction mechanism becomes more complex and therefore difficult to predict upon addition of multiple promoters.
The support material plays an important role in these reactions and offers an avenue to catalyst improvement. Ideal support materials for H 2 O 2 production should either be sufficiently inert to retain the intrinsic characteristics of the active metal or, preferably, offer synergistic effects with the active metal, as well as be inactive for any of the undesirable side reactions. Therefore, judicious selection of a suitable supporting material is as important as fine-tuning the active metal properties. In this regard, it is not understood why a carbon-supported Pd, Au, or Au-Pd catalyst is less active than their counterparts on titania, magnesia and alumina supports [81]. In the literature, a large number of support materials have been explored that possess: (i) wellordered textural properties to minimize the mass transfer resistance of reactants, (ii) selective adsorption sites for anionic/cationic metal precursors to increase the metal dispersion and stability, and/or (iii) acidic functional groups or immobilized halide ions to eliminate the use of inorganic acids or halides [31,81,82].
Carbon material is perhaps the most popular and widely studied support for a variety of catalyst designs for DSHP, due to its low cost, ready availability, inertness, and stability in the presence of caustic additives. Carbon-supported catalysts are also employed for the parallel oxidation process, and due to its high electrical conductivity, for the electrochemical processes. To overcome the shortcomings of conventional activated carbons, such as the heterogeneous nature of the surface and irregular microporous structures, ordered mesoporous carbons have been tested, and their superior textural properties led to improved catalytic activities by reducing the diffusion resistance of the reactants [83].
Other efforts to improve catalytic performance of catalysts associated with carbon materials include modifying their properties by incorporating nitrogen or oxygen groups. N-incorporated carbon supports provide selective adsorption sites for the anionic Pd precursors, which led to the formation of small, monodispersed metal nanoparticles. Heterocyclic N species around the supported Pd nanoparticles withdraw electron density and modify the electronic properties of Pd species. By virtue of these two beneficial effects, the Pd catalyst supported on N-containing carbon nanotubes showed promising activity in DSHP [84]. However, it is important to minimize formation of amine/ imine groups during the N-incorporation on the carbon supports because the basic sites may accelerate H 2 O 2 decomposition [41,85]. Likewise, care needs to be exercised when modifying the surface of carbon materials by acid treatments. Although the acid-treated carbons may act as a solid acid support and be efficient in suppressing H 2 O 2 decomposition [47], excessive acid treatment could have adverse effects such as destruction of the pore structure, enhanced hydrophilicity of the surface, and active site poisoning by carboxyl groups [86,87]. It should be cautioned that the relationship between H 2 O 2 decomposition activity and pH has not been firmly established.
Solid acid supports have been extensively investigated as means to prevent equipment corrosion and metal leaching caused by aqueous acids. These include inorganic oxides, carbon materials, organic/inorganic molecular sieves, heteropoly acids, metal-organic frameworks (MOFs), and resins [87][88][89][90][91]. For industrial applications, sulfonated resins offer several advantages: large-scale production with competitive price and easy control of the acid concentration. Active site blocking or diffusion limitation to molecular fluxes caused by the irregular pore structure of the highly crosslinked resin structure can be reduced by swelling the matrix in a polar solvent. As illustrated in Fig. 5, however, the matrix may become de-crosslinked slowly by the produced H 2 O 2 . Decomposition of the resin releasing polysulfonic acid (PSA) may also occur. The latter decreases the catalytic activity, and the PSA is difficult to remove via conventional solid-liquid separation techniques. In propylene epoxidation, PSA may significantly decrease the PO selectivity by catalyzing the ring opening of epoxide when in situ generated H 2 O 2 is used.
MOFs are an emerging, important class of solid acid support materials, because they can be prepared to possess high concentrations of robust Lewis acid sites at the metal nodes and Brønsted acid sites at the organic linkers introduced by post-synthesis modifications [92]. Indeed, Pd catalysts supported on acidic and highly stable MOFs such as MIL-101-SO 3 H and UiO-66-SO 3 H showed superior catalytic activity to Pd/HBEA (SiO 2 /Al 2 O 3 = 25) [91]. However, prolonged contact of the MOF with generated H 2 O 2 during reaction leads to partial and gradual oxidative decomposition of the MOF structure. Partial deformation of the MOF structure by H 2 O 2 was also reported [93].
It is unlikely that modifying the acidity/basicity of or addition of halide to the support could be applied to the parallel oxidation process involving organic alcohols unless the highly basic reaction condition can be eliminated. Likewise, MOFs might not be useful unless structures immune to destruction in a strongly basic medium or new chemistry can be found that can proceed at near neutral conditions. Innovative support designs should be pursued to prevent the subsequent degradation reaction of the generated H 2 O 2 . For DSHP, one approach is to encapsulate a metal particle with a thin, size-selective carbon coating that permits hydrogen diffusion through to the metal but not oxygen. In this manner, the metal dissociates the H 2 molecule and the H atoms formed spillover onto the carbon layer, where hydrogenation of O 2 to H 2 O 2 occurs. By preventing O 2 from accessing the metal, the degradative pathway initiated with dissociation of O 2 on the metal and the accompanying side reactions are avoided. Preliminary investigations demonstrated that a thin carbon coating on Pt nanoparticles allows preferential diffusion of H 2 to the Pt surface over O 2 , thereby greatly reducing OeO dissociation (Fig. 6). Consequently, the O 2 molecules react with the spilt-over hydrogen that diffuses from the Pt to the carbon coating to form H 2 O 2. Moreover, the carbon layer is also efficient in suppressing the access of generated H 2 O 2 to the Pt surface. Therefore, this thin carbon coating increases the H 2 O 2 selectivity at the expense of rate of H 2 conversion due to the increased mass transfer resistance [94].
In addition to activity and selectivity, the stability of a catalyst is also of critical importance. In the liquid phase, metal leaching resulting in performance degradation is a common issue. Coating also offers the potential benefit of enhancing catalyst stability by reducing metal leaching for processes in an aqueous medium as well as metal sintering. This is particularly useful when the DSHP process is used in tandem processes for chemical production in the presence of base metal catalysts. In general, leaching of base metals such as iron, cobalt, copper, and manganese is more severe than noble metals. For example, leaching of Cu [95], Fe [96], and Co [97][98][99] during catalytic oxidation has been reported, and the resulting liquid containing the leached metal ions may become toxic. On the other hand, no leaching of Au was detected in the oxidation of phenol [100], although there was leaching during oxidation of neat cyclooctene [101]. It has been reported that an overcoat of Al 2 O 3 deposited on Cu/Al 2 O 3 prevented Cu leaching because the overcoat interacted strongly with the under-coordinated copper atoms that are more susceptible to leaching [102]. Similarly, a yolk-shell structure of Co@C-N prevented Co leaching because of the stabilizing interaction between the Co nanoparticles and the C-N nanosheets [103].
Metal sintering is another source of deactivation of catalysts that can be mediated by coating. For example, encapsulating Au nanoparticles with a carbonaceous shell hinders crystallite growth [104] and there are many literature examples of other combinations of components. Another strategy for catalyst stability utilizes strong interactions between catalyst components to anchor the active phase (such as metal  particles). For example, the particle growth of Ru on supported catalysts in aqueous solutions was found to be related to the bulk electronegativity of the support, with sintering rate decreasing in the order of SiO 2 > C≈TiO 2 > γ Al 2 O 3 [105]. This is attributed to the Ru-O-M (M = support cation) interaction that is influenced by the electron density on the oxygen which in turn is affected by the electronegativity of the support cation. Another observation is that cobalt leaching was less severe with CoFe 2 O 4 than cobalt oxide due to the strong Fe-Co interactions in CoFe 2 O 4 [106]. For oxide-supported catalysts, hydrolysis and redeposition of the support oxide could result in phase transformation and/or loss of surface area and cause sintering and/or encapsulation of the supported metal particles. This has been documented with common supports such as γ-Al 2 O 3 [107]. Understanding the chemistry of leaching and degradation and devising preventive methods are research opportunities. Our current understanding is that the susceptibility of a support oxide to dissolution may be related to the density of defects. Thus, stability can be enhanced by reducing the defect density. This has been observed with zeolites where capping the silanol defects with organosilanes improved the ability of the zeolite to maintain crystallinity in hot liquid water [108]. Coating of an oxide surface with a layer of more stable material is also effective. This has been observed with samples comprising an alumina coating on amorphous silica-alumina [109] and carbon coating of SBA-15 [110].
The effect of the reaction medium is not well understood. Whereas acidic solutions are beneficial for effective action of halide modifiers in DSHP, known examples of production in parallel with oxidation require highly basic solutions where the high hydroxide concentrations should make the system insensitive to halide modification. At present, the mechanistic implication of these differences has yet to be explored.
The nature of diluent gas for the H 2 /O 2 mixture also exerts great influence on the catalytic activity for DSHP. CO 2 seems preferable to N 2 as measured by higher H 2 O 2 selectivity and productivity, because it may act as a mild-acid promoter and increase the surface coverage of active sites with oxygen by increasing the amount of dissolved oxygen under high pressure [83,111]. However, the possibility of equipment corrosion by the carbonic acid, which is produced from CO 2 in the presence of water-containing solvent, is of concern for industrial applications.
For the DSHP reaction to be a real breakthrough for selective oxidation processes, the catalyst performance without using halide additives needs to be significantly improved. To develop an improved catalyst, a better understanding of the fundamental reaction kinetics and mechanism of the DSHP is needed. In this context, a standard test reaction protocol with a benchmarking catalyst will be desirable for meaningful comparison of the performance of different catalysts. In addition, advanced computational methods and in situ characterization techniques would help improve our understanding of the reaction mechanisms that depend on the catalyst and reaction conditions. In that regard, there have been noticeable advances in recent years [44,112]. However, the simple molecular structures and similar rotational/vibrational behaviors of the involved molecules limit the possibilities in in situ characterization, in particular the vibrational spectroscopic approaches.
The economic feasibility of the hydrogen peroxide-propylene oxide (HPPO) process depends on the cost of production and the market price of H 2 O 2 . Therefore, successful implementation of an integrated direct synthesis DS-HPPO process requires not only the continuous production of approximately 8-12 wt% H 2 O 2 without additional purification steps, but also direct supply of the in situ generated H 2 O 2 without any separation. Although successful development of the DS-HPPO process may open a new avenue for related industries, i.e., PO, polyols, and polyurethane by substantially lowering the raw material cost, the performance of reported state of the art catalysts is still too poor for the DS-HPPO process in the absence of caustic additives. Therefore, at present, the use of caustic additives seems inevitable to meet the industrial target for H 2 O 2 concentration, which obviously increases the fixed and operating costs of the process.
For H 2 O 2 production in the parallel oxidation scheme, there are practical challenges: (1) For commercial viability, if H 2 O 2 is to be sold as a purified product, the oxidation product of the parallel process needs to have appropriate economic value after separation. An alternative is to use the H 2 O 2 in situ to produce a high-value product. (2) Ideally, every molecule in the parallel reaction is involved in H 2 O 2 generation. (3) None of the H 2 O 2 produced is degraded by decomposition to O 2 or hydrogenation to water. (4) The catalyst needs to be stable in the reaction solution mixture. The very limited information available thus far suggests that it is possible to achieve the limit of forming one H 2 O 2 molecule for every molecule of alcohol oxidized in the parallel reaction absence of the subsequent decomposition of the formed H 2 O 2 . It would be even more attractive if every oxidation equivalent in the parallel reaction can be utilized for H 2 O 2 production. Another serious problem with the very few systems studied is the significant degradation of the formed H 2 O 2 , thus limiting the overall productivity. Thus, there is much to be understood and improved for this process.
Other than CO, alcohols are the other reductant studied for H 2 O 2 production in the parallel reaction. Anecdotal information suggests that aldehydes but not acids are also very effective. It is known that different CeH bonds have different ability to reduce O 2 to peroxy and peroxide, but a more comprehensive understanding would allow prediction of how effective a molecule could be to produce H 2 O 2 . Ideally, the parallel oxidation reaction would make use of low-cost renewable reactant and produce a high-value product. For example, coupling the dehydrogenation of propane to propene with the production of H 2 O 2 (ΔG 300K = 119 kJ mol −1 ) is thermodynamically feasible. In this case, propane functions as a hydrogen source to reduce O 2 to H 2 O 2 . An even more desirable hydrogen source is H 2 O. However, the reaction O 2 + 2H 2 O = 2H 2 O 2 is thermodynamically uphill (116.7 kJ mol −1 ). Thus, it needs to be coupled with another reaction to overcome the thermodynamics. Table 1 lists some possibilities, unfortunately, the highly desirable oxidation of alkene to epoxide (ΔG⁰ rxn ≈ −115 kJ mol −1 ) is not among them.
Similar to DSHP, suppressing the undesired H 2 O 2 decomposition reaction is also very important during co-oxidation. At present it is not known whether formation and decomposition of H 2 O 2 share the same active sites, and if there are methods to suppress decomposition without affecting formation. This could be addressed with more detailed understanding of the reaction mechanism.

Photoelectrochemical (PEC) synthesis
Photocatalytic and PEC approaches have the potential to eliminate the need to supply H 2 as a fuel or electrical energy because electrons and holes excited by the energy of light are available. They convert light energy to chemical energy. Similar to electrochemical methods, they both have two half-cell reactions, and the difference among them rests on the source of energy to achieve electron-hole charge separation: electrical potential for the electrochemical approach, and photon energy for the photocatalytic and PEC approaches. Photocatalytic production of H 2 O 2 has been demonstrated for a long time. For example, excitation of ZnO or TiO 2 with UV radiation produces H 2 O 2 in an aqueous medium, albeit with a rather low efficiency [113]. In order for this approach to be commercially attractive, higher efficiencies and use of visible light excitation are highly desirable.

Photocatalytic processes
One area of current interest is to discover/develop photocatalytic systems that utilize visible light. One such system is based on C 3 N 4 . Shiraishi et al. reported the production of H 2 O 2 via photocatalytic reduction of O 2 on graphitic carbon nitride (C 3 N 4 ) and related materials [114]. Earlier work showed that the conduction band potential of C 3 N 4 was sufficiently negative to reduce O 2 to H 2 O 2 , but the valence band edge was not sufficiently positive to oxidize H 2 O to O 2 efficiently. Thus, isopropanol was needed to serve as a sacrificial electron donor. Under these conditions, the Faradaic efficiency of the H 2 O 2 production process reached 90 %, based on the mass balance of the oxidation products of isopropanol (acetic acid and CO 2 ). Modifying the C 3 N 4 network by doping with the electron-deficient aromatic compound pyromellitic diimide (PDI) shifts the valence band edge to a more positive value [115], and the resulting PDI-doped C 3 N 4 (C 3 N 4 /PDI) generated H 2 O 2 from O 2 -saturated water under visible light irradiation with an apparent quantum yield of 2.6 % at 420 nm. Further modification with reduced graphene oxide (RGO) [116] and boron nitride (BN) [117] as electron and hole collectors, respectively, improved the charge separation efficiency (Fig. 7), and the photocatalyst produced H 2 O 2 at a solar-to-chemical conversion efficiency of 0.27 %. In this system, O 2 reduction and water oxidation were presumed to proceed on the RGO and BN, respectively, however, O 2 evolution by water splitting could not be confirmed because O 2 was also bubbled through the reaction solution.
Charge separation in C 3 N 4 can also be enhanced by doping with biphenyl diimide (BDI) instead of PDI but with a significant difference [118]. Based on the calculated energy positions of the highest occupied and the lowest unoccupied molecular orbitals using DFT, both photoelectrons and holes in the C 3 N 4 /PDI system are generated in the same melem unit. In contrast, photoexcited holes are localized in the BDI unit while electrons remain in the melem unit in the C 3 N 4 /BDI system. For this reason, it is not necessary to modify C 3 N 4 /BDI with RGO and BN to promote charge separation. It has also been reported that mellitic triimide (MTI), having a three-directional planar structure, can be used in place of PDI (which has a two-directional planar structure) to produce dense sheets [119]. Compared to the PDI system, the C 3 N 4 /MTI system contains a higher density of melem units in its sheet structure and exhibits improved conductivity owing to the enhanced intra-and interlayer transport of photogenerated charge carriers. These C 3 N 4 -related photocatalysts have generated H 2 O 2 to concentrations > 4 mM (140 ppm) [116]. The production of H 2 O 2 from pure water using a BiVO 4 photocatalyst loaded with Au nanoparticles serving as active sites for the two-electron O 2 reduction reaction has also been demonstrated [120].

Photoelectrochemical processes
An emerging PEC system is based on BiVO 4 . In their pioneering work [121], Fuku and Sayama examined PEC water oxidation using BiVO 4 /WO 3 bilayer photoanodes prepared on glass substrates coated with fluorine-doped tin oxide (FTO) in solutions of various electrolytes, including sulfate, phosphate, borate and bicarbonate. These BiVO 4 / WO 3 /FTO photoanodes (hereafter referred to simply as BiVO 4 photoanodes) were found to generate exceptionally high photocurrents in an aqueous bicarbonate solution. In addition, H 2 O 2 was produced at an appreciable concentration in an anolyte when the catholyte and anolyte were separated by a proton exchange membrane and bicarbonate was employed as a supporting electrolyte. Bicarbonate ions were reported to be electrochemically oxidized to percarbonate ions, after which these ions would undergo spontaneous hydrolysis to regenerate bicarbonate, giving H 2 O 2 as the overall product [122,123]. A similar reaction mechanism would be feasible on BiVO 4 photoanodes in response to photoexcitation, where bicarbonate ions are adsorbed and oxidized to percarbonate, followed by hydrolysis to produce H 2 O 2 and regeneration of bicarbonate in the electrolyte solution. Accordingly, bicarbonate ions can be regarded as a catalyst for the two-electron oxidation of water to H 2 O 2 . Interestingly, BiVO 4 itself has also been shown to function as an electrocatalyst for H 2 O 2 production by water oxidation in the absence of light irradiation [123], while Bi 2 O 3 , V 2 O 5 and many other binary oxides (except for Al 2 O 3 and TiO 2 ) lower the H 2 O 2 production rate.
Thus, it appears that the unique electrochemical characteristics of the BiVO 4 surface may have specific functions in promoting the PEC production of H 2 O 2 . It was suggested by DFT calculation that the surface of BiVO 4 had a moderate formation energy for adsorbed OH (*OH) species and was favorable for H 2 O 2 production [50]. Too strong or too weak binding of *OH would result in the sequential oxidation of H 2 O 2 into O 2 via the four-electron oxidation or the formation of OH radical (OH % ) species via the one-electron oxidation, respectively By increasing the bicarbonate electrolyte concentration and bubbling CO 2 through the electrolyte solution (to lower the pH), the Faradaic efficiency of H 2 O 2 production on BiVO 4 photoanodes reached a maximum of 54 % and a H 2 O 2 concentration of 2 mM (68 ppm).
The Faradaic efficiency of H 2 O 2 production on BiVO 4 photoanodes can also be improved by surface modification of the anode with oxide coatings, such as SiO 2 , ZrO 2 , TiO 2 and Al 2 O 3 , formed by spin coating of metal-organic solutions and subsequent calcination (Fig. 8) [124]. Among these oxides, Al 2 O 3 is the most effective, and has been shown to increase the Faradaic efficiency from 54 % to 80 %. The deposited Al 2 O 3 forms a mesoporous layer on the surface of the BiVO 4 photoanode and suppresses the sequential PEC oxidation of H 2 O 2 to O 2 on the anode. It has been suggested that H 2 O 2 generated on the BiVO 4 surface diffuses into the electrolyte solution through this mesoporous Al 2 O 3 layer, likely by concentration gradient. The weak acidity of the Al 2 O 3 surface may also enhance adsorption of the weakly basic bicarbonate ions. Thus, BiVO 4 photoanodes modified with Al 2 O 3 are able to evolve H 2 O 2 from dilute aqueous solutions of bicarbonate. It was found further that using chemical vapor deposition (CVD) to apply a thinner conformal Al 2 O 3 coating than by dip-coating impeded the photocurrent less while maintaining the beneficial effects of the Al 2 O 3 [125]. The potential of the conduction band edge for BiVO 4 is lower than the reversible hydrogen evolution potential (0 V vs. RHE). Therefore, it is necessary to apply an external voltage or combine a BiVO 4 photoanode with photovoltaic cells or photocathodes to drive the PEC H 2 O 2 and H 2 evolution reactions. Alternatively, O 2 can be reduced to H 2 O 2 by two-electron reduction on the cathode [126]. The reversible potential for this reaction is +0.68 V vs. RHE, which is more positive than the conduction band edge potential for BiVO 4 . Thus, photoexcitation of a BiVO 4 photoanode in conjunction with a H 2 O 2 -evolving cathode can spontaneously produce H 2 O 2 by water oxidation with the simultaneous reduction of O 2 (Fig. 9), achieving the overall reaction: O 2 + 2 H 2 O → 2 H 2 O 2 . The reported faradaic efficiencies for H 2 O 2 production on a BiVO 4 photoanode was 50 % and on a cathode 90 % based on an FTO-coated electrode modified with Au nanoparticles under short-circuit conditions.

Challenges and opportunities
The production of H 2 is perhaps the most useful process for solar-tochemical energy conversion as H 2 can be used as a fuel and chemical feedstock. Solar splitting of water also produces O 2 , which has a rather low market value (about $0.1 kg −1 , less than one-tenth that of H 2 ) [127]. In contrast, the price of H 2 O 2 is much higher ($0.7-1.2 kg −1 in 2009) [9]. Therefore, co-production of H 2 and H 2 O 2 by water oxidation instead of O 2 is much more attractive economically [128]. The oxidation of water into H 2 O 2 also has other benefits. For example, H 2 O 2 is produced in the liquid phase and spontaneously separated from the coproduced gaseous H 2 . The reversible potential for the two-electron oxidation of water to H 2 O 2 is +1.77 V vs. RHE, higher than the +1.23 V for the four-electron oxidation to O 2 . If PEC water splitting into H 2 and H 2 O 2 proceeds with the same quantum efficiency and the product H 2 O 2 can be collected efficiently, this reaction would yield a solar-to-chemical energy conversion efficiency 1.4 times greater than water splitting into H 2 and O 2 . Materials with wider band gaps (i.e., shorter absorption edge wavelengths) are necessary to split H 2 O into H 2 and H 2 O 2 , although this will not be a major problem because most common oxide semiconductors have valence band edges at potentials more positive than the reversible potential for the two-electron oxidation of water [129].
To date, the semiconductors applied to H 2 O 2 production under visible light have been limited to C 3 N 4 -related materials and BiVO 4 , which have relatively large bandgap and can utilize only the higher energy portion of the solar spectrum. Thus, it would be helpful to explore the applicability of smaller band-gap semiconductor materials to determine the surface properties essential for efficient H 2 O 2 evolution, and to allow the development of more efficient PEC and photocatalytic systems. For both PEC and photocatalytic systems, it is especially vital to improve the physical electronic properties of the semiconductors so as to efficiently utilize photoexcited carriers [130].
Particle transfer technology can be used to facilitate the survey of various well-crystallized particulate semiconductors in PEC reactions [131]. As an example, photoelectrodes comprising particulate semiconductors fabricated by particle transfer have been applied to the direct PEC production of methylcyclohexane, a promising H 2 carrier, from toluene ( Fig. 10) [132]. Suppression of the sequential decomposition of H 2 O 2 on the catalyst and support remains an important challenge because H 2 O 2 is an intermediate product regardless of whether it is generated reductively or oxidatively. Thus, catalysts and surface modification layers should be developed for the current PEC and photocatalytic systems to promote H 2 O 2 production and suppress sequential decomposition. Sayama et al. developed an Al 2 O 3 coating that blocked the access of H 2 O 2 in the electrolyte solution to the BiVO 4 photoanode [124,125], and a similar effect has been obtained by applying an amorphous transition metal oxide coating to overall water splitting photocatalysts [133]. Such coatings allow O 2 evolved on the photocatalyst surface to diffuse into the reaction solution while blocking the access of H 2 O 2 in the reaction solution to the photocatalyst. The modification of photocatalysts and photoelectrodes with such coating layers may improve selectivity for H 2 O 2 production, It is equally essential that these semiconducting materials are stable in the aqueous and acidic or basic environment under light irradiation. Oxidation of the photoelectrode material facilitated by light is another challenge. Coating layers could also serve as a protection layer. Some oxynitride photocatalysts evolve not only O 2 but N 2 as an oxidation product during the overall water splitting reaction because of the selfoxidation of nitride ions (N 3− ). Fortunately, modification with amorphous oxide layer can suppress it and enable oxynitride photocatalysts to split water into H 2 and O 2 at the stoichiometric ratio [134]. Note that H 2 O 2 is an efficient hole scavenger and is readily oxidized to water [135]. It is therefore essential to use surface modification layers with

Ammonia synthesis
The development of the Haber-Bosch Process, which currently accounts for the production of around 174 million tons of ammonia annually, was a landmark achievement of the 20th Century [136]. Over 85 % of the ammonia produced is used in synthetic fertilizers which are credited with the sustenance of around 40 % of the global population. The process involves the reaction of hydrogen with nitrogen in the presence of a promoted iron catalyst and is operated at high pressure (> 100 atm) favoring ammonia formation. Whilst thermodynamically, ammonia synthesis is favored at low reaction temperatures (ΔG°f = −16.2 kJ mol −1 at 25°C), kinetics dictate operation at ca 350-500°C and elevated pressures are needed to achieve acceptable process yields (Fig. 11). The hydrogen feed for the process is derived from fossil fuel sources such as natural gas. It has been recently estimated that the current operation of industrial ammonia synthesis accounts for ∼2.5 % of global fossil fuel-based CO 2 emissions [136]. Furthermore, when considered in its entirety, including the generation of feed streams, the Haber-Bosch Process is believed to be responsible for ∼2 % of the world-wide energy consumption [136]. Whilst the industrial scale synthesis has evolved to be a relatively mature technology which is operated on the large scale in highly integrated and very efficient plants, the iron-based catalysts employed continue to be similar to those developed over a century ago. Recently there has been a resurgent interest in the discovery of new catalytic materials due to three, to an extent inter-related, motivations: (i) The increasing availability of electricity derived from renewable sources such as wind and tidal power, which makes the large-scale, economic and fossil-free generation of hydrogen via water electrolysis an increasingly realistic prospect. In this context, ammonia might also be an appropriate energy storage molecule for intermittent periodic oversupply and peak demands of electricity necessitating reactors which can be started up quickly; (ii) The increasing interest in the application of ammonia as a fuel.
Whilst ammonia synthesis is currently growing at a rate of around 1.5 %-2.0 % per annum [137], its application as a fuel would require a significant growth in the synthesis capacity for which the ability to generate "green ammonia" as detailed in (i) above would be desirable in terms of sustainability and for more localized production; (iii) Expansion in the ability to produce ammonia at localized smallerscale facilities which is of potential importance for establishing processes in locations where there is little infrastructure, such as in remote global locations or on farmland for the production of "ondemand" locally derived fertilizer.
All of the above considerations impact upon current, or projected, anthropogenic CO 2 footprints. For example, (ii) is viewed as a replacement for current fossil fuel requirements and indeed there were buses driven on ammonia as fuel operational in Belgium in 1943, and as a partial replacement for the large scale CO 2 -intensive production facilities employed today, (iii) offers potential additional savings in offsetting fossil fuel-driven transportation of ammonia-based fertilizers to their points of application. As can be anticipated, heterogeneous catalysis has a potential major role to play in the development of more sustainable localized ammonia production. Here we focus on discussion of potential routes and challenges to improve the energy and CO 2 footprint. The emphasis is on catalyst design in heterogeneous thermal catalytic processes, electrocatalytic, and photocatalytic processes. Although ammonia synthesis is possible with homogeneous catalytic routes [139], this approach will not be discussed in this perspective. Fig. 10. Schematic drawings of (a) a membrane photocathode assembly (MPA) consisting of a commercially-available membrane filter and self-supported ZnSe:CIGS photocathode prepared by a particle transfer method and (b) a two-chamber PEC cell composed of aqueous and toluene/methylcyclohexane (TL/MCH) chambers separated by the ZnSe:CIGS MPA. Reprinted with permission from Ref. [132]. © 2018, American Chemical Society. Fig. 11. The percentage of ammonia at equilibrium for a 3/1 H 2 /N 2 mixture as a function of temperature and pressure. Data taken from Reference [138].

Thermal catalytic (Haber-Bosch) process
In industrial practice, the predominantly applied catalyst is based on promoted Fe systems which differ only in minor ways from the original formulations. Such formulations include the presence of alumina as a structural promoter along with electronic promoters based upon alkali and alkaline earth metals such as K and Ca [140]. Whilst the system may seem simple, it is indeed more complex and is, for example, enhanced by structural distortion. Fortunately, fundamental surface science studies have underpinned much of the fundamental understanding. The reaction is structure sensitive with the rate-determining N 2 activation step being favored on C7 sites (iron atoms with seven nearest neighbours) associated with surfaces such as the (111) surface plane [141], and only about 5 % of the geometric surface in the industrial catalyst participating in reaction [142]. More recently, a commercial carbon-supported Ru based catalyst has been developed which is about 20 times more active than the commercial iron-based Haber-Bosch catalyst. The higher activity of this catalyst permits less demanding process operational parameters. This catalyst forms the basis of the Kellogg Advanced Ammonia Process (KAAP) which was announced in 1994. The background to the origin of this catalyst is detailed in a very nice overview [143]. A significant challenge in relation to this catalyst was the development of hydrogenation-resistant mesoporous graphitic support. In the Ru system, the B5 site has been identified as being particularly active [144,145]. This site is associated with steps and comprises five Ru atoms. The abundance of these B5 sites is dependent upon catalyst metal dispersion and crystallite morphology and therefore the reaction is strongly structure sensitive with a strong particle size and morphology dependence. Such sites have a maximum occurrence for particle size in the range 1.8-2.5 nm with only (001) and (100) surfaces exposed [146]. Whilst the commercial catalyst uses a carbon-based support, BN (which is isoelectronic with C and which can exist in a graphite as well as a diamond-like structure) has been reported as a very effective and hydrogenation-resistant support [147]. Basic supports such as MgO and Pr 2 O 3 are also reported to generate highly active Ru-based catalysts [148].
Mechanistically, the reaction in both the Fe and Ru-based systems is widely believed to be of Langmuir-Hinshelwood form with N 2 dissociation being the rate-determining step. Unlike Fe where *N inhibits the synthesis reaction, in the case of Ru, the reaction is inhibited by adsorbed *H instead [28]. Alkali promoters reduce the effect meaning that the reaction rate is not adversely affected under a high concentration of ammonia as might be the case close to the reactor exit [28].
Of all the chemical elements, Ru exhibits the highest activity. In this context it is interesting to note that Os, the third chemical element in the same group as Fe and Ru, was also recognized to be highly active for ammonia synthesis, particularly in the early work of Haber [149]. In a general context, the catalytic performance of metals for ammonia synthesis has been rationalized on the basis of a scaling relationship by Nørskov and co-workers [150,151], which states that the strength of nitrogen binding on a metal surface is related to ammonia synthesis activity: too weak binding results in difficulties in activating the N 2 molecule, and too strong binding results in slow removal of *N. This explanation results in a Sabatier volcano type relationship, as presented in Fig. 12 [150], which demonstrates that there is an optimal performance as a function of *N binding energy.
For catalysts based on pure metals, this optimum corresponds to Ru. In terms of the discussion above, it is interesting to note the relatively high activities of Ru, Fe and Os. Whilst many studies have addressed the performance of Fe and Ru, the two metals which form the basis of commercial ammonia synthesis catalysts, little attention seems to have been directed towards Os. Originally its scarcity was a concern and it is necessary to exercise care in its handling due to potential toxicity. In a recent DFT based study, the barrier for N 2 activation for Os nanoparticles was determined to be similar to that for Ru nanoparticles, with Ru, when existing as 2-4 nm diameter particles, being a better catalyst taking into account activation energy, surface vacancies and density of step sites [151]. Os has also been reported to be very effective for the non-steady state production of NH 3 using sequential nitrogen and hydrogen pulses [152], although this approach does not seem to have been studied further.
Another aspect of the scaling relationship and the volcano plot presented is related to the possibility of tuning catalytic performance by applying appropriate combinations of elements. A specific example of this relates to the catalytic performance of the ternary nitride Co 3 Mo 3 N which, particularly when promoted with low levels of Cs + , has been reported to be highly active for ammonia synthesis and competitive with commercial formulations [150,[153][154][155][156][157][158]. The activity of this material has been rationalised on the basis of Fig. 12 with the combination of Mo (which binds nitrogen strongly) with Co (which binds nitrogen weakly) leading to a material that has an activity close to that of the optimal elemental catalyst, Ru [150]. Implicit in this approach is the requirement for structure sensitivity of the Co 3 Mo 3 N system, with the (111) plane which exposes both Co and Mo being active [150], and no assumed role for the interstitial lattice N in this system beyond ensuring the expression of the active surface lattice plane. Such structure sensitivity has yet to be demonstrated, although there are recent reports that dispersed CoMo catalysts on CeO 2 support are very effective and efficient particularly when considered from the viewpoint of Co and Mo utilization [159,160].
There is recent evidence to suggest that interstitial N may participate in the reaction mechanism in the form of a nitrogen-based Marsvan Krevelen mechanism. The reducibility of Co 3 Mo 3 N, forming ammonia, was demonstrated with the formation of the previously unreported Co 6 Mo 6 N phase [161][162][163]. Isotopic N exchange experiments demonstrated that the lattice N is exchangeable in this system [164] and computational modelling indicated the N vacancy concentration in the Co 3 Mo 3 N to be potentially significant in the temperature regime of interest for ammonia synthesis [165]. Further modelling has also probed the activation steps for N 2 and H 2 in the system [166] as well as the energy profiles of potential mechanisms [167]. Fig. 13 presents one such mechanism that is energetically favorable. Interestingly, this mechanism is associative in nature, in that there is no N 2 dissociation preceding initial hydrogenation, which is more akin to enzymatic systems. It is also notable that the homomolecular exchange of N 2 (i.e. 14 N 2 / 15 N 2 isotopic scrambling) which can be a probe of N 2 dissociation, was not observed over Co 3 Mo 3 N at 400°C [164], although the material catalyzes ammonia synthesis at this temperature. The suggestion of ammonia synthesis occurring via an associative mechanism is rare in the heterogeneous catalytic literature, although it has been proposed as a minor pathway for the Ru catalyst [168]. The link that such a pathway produces with respect to the much slower but ambient condition conversion of N 2 by nitrogenase [169] could be potentially significant in the development of heterogeneous catalysts active at lower reaction temperatures.
It should be noted that there are reports in the literature which propose that the Co 2 Mo 3 N phase, present as an impurity along with Co 3 Mo 3 N, is a very active catalyst [170,171]. To date, this has not yet been studied in detail. In terms of ternary nitride catalysts, Ni 2 Mo 3 N and Fe 3 Mo 3 N demonstrate high activity [146,156], although they have been less widely investigated than Co 3 Mo 3 N which generally seems more active. It is possible to rationalize the performance of Ni 2 Mo 3 N on the basis of Fig. 12. For these nitride systems, the Mars-van Krevelen mechanism may also apply and the lattice nitrogen of Ni 2 Mo 3 N has been observed to be less reactive than that in Co 3 Mo 3 N, with the ternary phase CoNiMo 3 N being comparable to Ni 2 Mo 3 N [172]. By judicious choice of preparation route, it is possible to prepare Ni 2 Mo 3 N to become comparable in performance to Co 3 Mo 3 N, which is a significant finding in the context that the nickel-containing material can be more conveniently prepared by treatment of its oxide precursor with ammonia synthesis gas mixture at 700 o C whereas the cobalt material requires an ammonolysis step [172]. The problematic nature of ammonolysis on the large scale is related to heat transfer, which has been documented in the production of binary molybdenum nitrides [173]. A further consideration in relation to the preparation of Ni 2 Mo 3 N is the difficulty in obtaining a pure phase [172,174]. It is also important to note that misidentification of Ni 2 Mo 3 N as Ni 3 Mo 3 N (which has yet to be prepared, although there is precedent of the related Ni 3 Mo 3 C phase in the literature) occurs fairly frequently in the literature [175] and still continues.
Recent studies have reported hydride and electrides to be new classes of material to serve as a component of ammonia synthesis catalysts. Both TiH 2 and BaTiO 2.5 H 0.5 have been demonstrated to be as active as the supported Ru systems [176]. In this context, the application of BaTiO 2.5 H 0.5 as a support for Ru, Fe and Co catalysts has been reported to enhance ammonia synthesis rates to the extent that Fe and Co supported on this hydride are more active than Ru/MgO [177]. The hydridic supports are much more active than their perovskite counterparts (Co/BaTiO 3 being 400 times less active than Co/Ba-TiO 2.37 H 0.63 , and Fe/BaTiO 3 70 times less active than Fe/BaTiO 2.35 H 0.65 at 400°C and 5 MPa). Other studies have demonstrated high activities to be associated with the presence of hydrides [178,179].
There is a series of publications outlining the application of electrides, such as Ca 2 N:e − and [Ca 24 Al 28 O 64 ] 4+ (e − ) 4 , as supports for catalytically active metals [180][181][182][183][184][185][186][187][188]. The electride supports can enhance the performance of Ru significantly, and the surface reactions of *N and *H become rate determining as opposed to N 2 dissociation. In some electrides, hydride formation has been documented and this results in reduction of hydrogen poisoning of the ammonia synthesis kinetics rather than the generation of an active species. To avoid the activity decline associated with sintering of metal nanoparticles supported on electrides, the interelectride phase LaRuSiH has been prepared and shown to exhibit high activity [188]. In this work, a model based on the following reaction steps and in which hydride ions play a pivotal role was proposed: (i) reversible exchange of H¯with e − , (ii) N 2 adsorption and dissociation via the hot atom mechanism, (iii) desorption of H facilitated by the heat generated by nitrogen adsorption, and (iv) formation of NH 3 for which its desorption via weakening of the Ru-N bond and the regeneration of the H¯occurs upon further hydrogenation. In the detailed discussion of the reaction mechanism for the LaRuSiH system, the authors stated that the data were also consistent with the Mars-van Krevelen-based associative mechanism which had been previously reported for Co 3 Mo 3 N [167]. Clearly, further work is necessary to definitively establish the mechanism in operation. Fig. 12 suggests that Co-based systems would not be expected to be very effective catalysts for ammonia synthesis due to the sub-optimal adsorption energy of nitrogen. However, over the years, a number of studies have documented rather high activities for supported Co and other Co-based materials, such as those presented in Table 2 [60], which include the combination of Co with hydrides and those supported on 12CaO·7AlO-based electrides [189]. For the latter samples, as for the proposals made for Ru, the enhanced ammonia synthesis activity has been ascribed to more facile N 2 activation as a consequence of facilitated electron injection into the supported Co nanoparticles. Related to this suggestion, LaCoSi in which Co was reportedly negatively charged enhanced N 2 activation and shifted the rate-determining step of the reaction to the formation of NH x species [187].
In addition to activity, a practical catalyst also requires stability. In this context, although Co/CeO 2 was reported to be more active than CoMo/CeO 2 for the first ∼25 h on stream, perhaps contrary to the expectation arising from Fig. 12, it exhibited significant deactivation whereas the CoMo system was relatively stable for > 100 h [160]. It has been suggested that the deactivation was due to sintering, which might be mitigated to some extent using a synthetic route employing dopamine [190].
Whilst not considered from the viewpoint of Co, but rather the stabilization of an unconfirmed enhanced stability rhenium nitride phase, CoRe catalysts are very active for ammonia synthesis [191,192]. Comparable materials comprising FeRe and NiRe were not as effective and, based upon the comparison with rhenium nitride, ammonolysis was employed to prepare materials from their oxidic precursors. The optimum catalyst corresponded to the composition CoRe 4 [192]. Subsequently, ammonolysis was replaced by pretreatment with ammonia synthesis gas and again active catalysts were formed [193]. Substituting N 2 /H 2 for Ar/H 2 during preparation led to an observed induction period which was also mirrored in a modified homomolecular nitrogen isotopic exchange pattern [193]. Whilst it is tempting to invoke the necessity of a (surface) nitride phase, in situ XAS studies demonstrated the pretreatment gas to influence the extent of Co-Re mixing [194]. This illustrates that pretreatment where the formation of active phases through processes such as segregation etc. can play a strong role. This is an aspect which is seldom studied in detail. Cesium-promoted aluminasupported rhenium catalysts have also been reported to be active for ammonia synthesis with some rhenium nitride being present, which is more active than rhenium metal [195]. Whilst cobalt rhenium systems exhibited activities comparable with the more active ammonia synthesis catalysts, they possessed very low surface area (frequently < 0.5 m 2 g −1 ) and hence had very significant surface area-normalized activity. On this basis, it certainly seems worthwhile to explore the performance of materials comprising highly dispersed CoRe phases.

Opportunities and challenges
Methods to improve an ammonia synthesis catalyst can be based on the conventional mechanistic understanding of the need to balance the activation of N 2 to form *N and the reactivity of this *N atom being limited by the scaling relationship shown in Fig. 12. In this context, attempts to exploit the synergistic combination of alkali and alkaline earth hydrides with transition metals [178] or nitrides [179] have demonstrated success. In such a combination, the hypothesis is that nitrogen activated by the metal/metal nitride is reduced by hydride ions and the resulting species undergo further hydrogenation forming the product and regenerating the active phase. Potentially related to this catalytic cycle is the observation of ammonia synthesis activity for amide-based catalytic materials [197]. The reports of using hydrides and electrides as supports are interesting and worthy of further investigation, especially to improve our understanding of the cause of their effects and the implication on the scaling relationship. Exploring different preparation methods to improve the exposed surfaced area of the active phase is also a recognized direction of research. It should be noted that whilst a number of interesting materials have been reported in the literature, little attention is directed towards prolonged lifetime/ accelerated deactivation testing and the stability of performance of catalysts when fed with feedstocks containing impurities, and even less to the practicality of scaling up the preparation and handling of the  [196] * Rate corresponding to 1.0 MPa reaction pressure. materials produced in exploratory tests. The recent publications of the relevance of the Mars-van Krevelen mechanism to ammonia synthesis [176] offer a new direction that could circumvent scaling limitations and serve as the key to the development of heterogeneous catalysts active under milder reaction conditions. In this context, there is interest in systems based on chemical looping. This approach involves the treatment of a nitride or related material such that ammonia is generated in a stoichiometric reaction involving the nitride following which the nitrogen-depleted material is regenerated such that the cycle can be repeated. In the literature, both metal nitrides [198][199][200][201][202][203][204] and imides [205] have been investigated, and ammonia production via hydrogenation of a nitride [206,207] or via hydrolysis of the nitride [198][199][200][201][202][203][204] has been reported. In such approaches, the reactivity of lattice nitrogen is an important factor, and theory has been employed to explore the modification of lattice nitrogen reactivity and enhanced ammonia production [166]. Most notably, nitrogen-doped perovskites [208], binary nitrides of magnesium, aluminum, calcium, chromium, manganese, zinc, and molybdenum [201], manganese nitrides [203] and calcium nitride [201] have been investigated for this purpose, as has the hydrolysis of lithium nitride which is of interest due to the nitridation of lithium metal directly with dinitrogen [209]. Whilst it is not strictly conventional catalysis, this approach affords the possibility of producing ammonia under conditions of ambient pressure avoiding potentially limiting scaling relationships. Together with the more conventional heterogeneous catalysis approach, it is an area of increasing interest and is certainly an area worthy of additional attention. The route based upon AlN has recently been discussed from a process orientated viewpoint where a high system efficiency close to 70 % has been reported [210].
It is worthy to note that the Mars-van Krevelen mechanism is increasingly postulated for electrocatalytic ammonia synthesis as well, as discussed in the next section. Likewise, there is an increasing interest to explore associative mechanism of nitrogen activation, akin to nitrogenase routes with electrons taking the place of ATP. At present, the attained productivity is low and there are false results due to contaminations and other artefacts. Nonetheless, they possess rich potential.
The drive to develop more active catalysts to facilitate smaller scale and localized ammonia production is based upon the increasing availability of low cost, sustainably-derived, fossil fuel-free, hydrogen production. Aside from the ever-increasing demand for ammonia for global food production, ammonia increasingly attracts attention as an alternative fuel or energy vector. Such considerations might also drive an expansion of highly centralized large-scale production facilities into areas where none exists currently, such that new processes, not dependent upon the optimized integration currently in place, are desirable. This increased recent interest in ammonia synthesis is reflected in the academic literature and a number of interesting catalytic materials and mechanistic approaches towards ammonia synthesis have been reported recently.
Whilst the impression is sometimes conveyed in some studies that the Haber-Bosch Process is inefficient, it has to be recognized that the process is highly integrated and optimized and that the catalyst employed which, when treated appropriately, is operational for a very long time (> 10 years) and is based on an earth-abundant, non-toxic and cheap element. Furthermore, whilst sometimes desirable from the fundamental point of view when the goal is to generate information on, e.g. lattice component reactivity or the influence of phase interconversion, the applicability of data generated from catalytic tests operated under conditions of ambient pressure to industrial practice needs to be carefully considered. For example, the equilibrium limited ammonia yield at ambient pressure and 400°C employing 3:1 H 2 :N 2 , a commonly applied testing regime, is only 0.4 % (Fig. 11). It is an unrealistic reaction regime for genuine application. A further related concern that is seldom addressed is that in industrial practice, reaction mixtures are recycled and contain some ammonia which potentially adversely affects catalytic performance. Aside from these considerations, recent years have witnessed a significant resurgence of interest in the area of ammonia synthesis. A number of exciting developments have been reported and it can be anticipated that further excitement lies ahead. Computational modelling is assuming a more central role [211] and there is increasing recognition of the mutual benefit to be gained through the combination of different approaches to nitrogen activation and the extent to which learning can be transferred between different approaches (e.g. between heterogeneous catalysis, electrocatalysis, photocatalysis and enzymatic catalysis) [212].

Electrochemical and photocatalytic processes
The production of NH 3 by electrochemical synthesis and other methods has been summarized in several recent reviews [23,138,213,214]. The associated electrochemical methods may be categorized depending on the type of electrolyte, operational temperature, and whether N 2 is directly reduced to NH 3 on the cathode or H 2 is initially generated and then fed to a thermal catalytic process for the conversion of H 2 and N 2 to NH 3 . Generally, in electrochemical synthesis, N 2 molecules either are reduced to N 3ˉi ons or react with H atoms at the cathode while electrons for the cathodic reactions are extracted from reactants at the anode. The electrochemical synthesis of NH 3 using an aqueous electrolyte solution has rarely been reported because the solubility of N 2 is low and the reaction temperature must be below the boiling point of water. The use of a single-compartment cell supplied with a mixture of N 2 and H 2 or the coexistence of H 2 in the cathode chamber has been shown to possibly increase the resistance of the cathode to oxidation due to O 2 impurity in a feed gas [215]. However, it is preferable to supply pure N 2 and H 2 separately so as to minimize the competitive adsorption of H 2 (or H 2 poisoning) while maximizing the adsorption of inert N 2 on the cathode catalysts. For these reasons, the most common electrochemical cell has a two-compartment design in which the compartments are separated by an electrolyte, such as a proton-conducting polymer membrane, molten salt or solid-state ceramic, depending on the operational temperature (Fig. 14).
In such systems, proton-conducting polymers, molten salts and ceramic membranes may be employed at low (T < 100°C), intermediate (100°C < T < 500°C) or high (T > 500°C) temperatures, respectively [23,138,213,214]. In the case of a solid proton-conducting electrolyte, H 2 is oxidized at the anode to give protons, and the resulting electrons are transferred to the cathode through an external circuit, while the protons migrate through the electrolyte to the cathode chamber. Ideally, at the cathode, N 2 reacts with the protons and electrons provided through the electrolyte and electric circuit, respectively. However, it is also possible for the protons to be reduced to H 2 without reacting with N 2 , which would reduce the current (or Faradaic) efficiency for NH 3 synthesis to below unity.
Other ionic species can also migrate between the anode and cathode, including nitride ions (N 3− ) in molten salts containing Li 3 N, hydroxide ions (OH-) in molten hydroxides and oxide ions (O 2-) in solid ceramic electrolytes [138]. In a system employing a Li 3 N-containing molten salt, N 3− ions generated by N 2 reduction at the cathode are released into the molten salt electrolyte and migrates to the anode to react with H 2 to produce NH 3 . In this case, oxidation of N 3− ions to N 2 is an undesirable side reaction. When molten hydroxides and solid ceramic electrolytes are used, N 2 and water are simultaneously fed into the cathode chamber, where N 2 is reduced to NH 3 and ionic species such as OH-and O 2are transferred to the anode chamber and oxidized to O 2 . In this scheme, the resulting NH 3 contains water owing to the presence of unreacted feed water. In any case, the rapid migration of ionic species is essential for efficient electrochemical NH 3 production, and higher temperatures are preferable to ensure efficient ionic conduction and kinetic activation for the NH 3 synthesis reaction. However, because the synthesis of NH 3 from H 2 and N 2 is an exothermic reaction, it is also desirable to lower the reaction temperature to obtain a more favorable equilibrium conversion of reactants to NH 3 . Therefore, there is a challenging trade-off between kinetics and thermodynamics in electrochemical NH 3 synthesis, similar to the Haber-Bosch process.
Kubota et al. reported dry NH 3 synthesis using electrochemical membrane reactors at a moderate temperature of 250 ⁰C that balanced the thermodynamics and kinetics of the reaction [216]. The reaction cell was composed of a catalyst that generates ammonia [217], a hydrogen-permeable alloy membrane and a water electrolyzer based on a phosphate electrolyte (Fig. 15). In this system, water is oxidized to O 2 and the resulting protons migrate through a phosphate electrolyte to a Pd-Ag alloy membrane cathode where H 2 is produced. This H 2 is subsequently absorbed by the Pd-Ag alloy membrane and transported to an adjacent NH 3 synthesis catalyst based on Ru/Cs + /MgO. This process involves catalytic reduction of N 2 by H 2 rather than an electrochemical process. Nevertheless, the mechanism is equivalent to electrochemical NH 3 synthesis because it involves electrochemical water splitting and NH 3 synthesis in thermal equilibrium. In an initial study of this system using N 2 and H 2 O feed gases at atmospheric pressure, the NH 3 production rates were in the range of 9.0-7.7 × 10 −10 mol s -1 cm -2 , with a current efficiency of 2.6-3.5 % [216,218]. H 2 evolution was found to be a major side reaction that lowered the current efficiency. However, a supply of excess N 2 evidently favored NH 3 synthesis by suppressing the competitive adsorption of H 2 on the Ru catalyst that otherwise prevented the cleavage of the N 2 triple bond. In fact, the reaction order with respect to H 2 was typically negative, although the order was dependent on the catalyst support employed and the activation states during the reaction [28,217]. More recently, the NH 3 formation rate and the current efficiency were improved to 1.24 × 10 -8 mol cm -2 s -1 and 12 %, respectively, by increasing the reaction pressure to 0.7 MPa [219].
Instead of direct reduction of nitrogen molecules, an electrochemical lithium cycling process at intermediate temperatures and ambient pressure was suggested in 2017 ( Fig. 16) [220]. LiOH is electrolyzed in a molten bath at 400-450°C, where ionic Li + is reduced into metallic Li on a cathode and OH − ions are oxidized into water and O 2 on an anode. These electrodes are separated with a porous alumina diffusion barrier to avoid side reactions of reactants and products. Metallic Li is collected and subjected to reaction with molecular N 2 to form Li 3 N, which is then hydrolyzed to LiOH to produce NH 3 . The net reaction is 2N 2 + 6H 2 O → 4NH 3 + 3O 2 . This process excels in the current efficiency (88.5 %) because the N 2 reduction reaction is physically and temporally separated from the NH 3 production reaction, which involves H atoms and therefore the opportunity for the competing, nonproductive H 2 evolution reaction. However, an operation voltage greater than 3 V is needed to drive electrolysis of LiOH. This is a major cause of the low energy efficiency (approximately 40 %). In comparison, the energy efficiency of conventional Haber-Bosch processes is estimated to be 56-70 %, depending on the basis used in the calculation [23,211]. A preliminary techno-economic analysis suggests that the cost of an electrochemical lithium cycling process is higher than the existing Haber-Bosch process using an average industrial electricity price ($0.071 kW h -1 ) [220]. Thus, very low-cost electricity ($0.02≈ kWh -1 or less) is needed to make this process competitive.
The photocatalytic production of NH 3 from H 2 O and N 2 has also been reported and recently reviewed [214,221]. This method was first reported in 1977, using both pristine and Fe-doped TiO 2 photocatalysts under UV irradiation [222]. More recently, a surface-reduced Bi 2 MoO 6 photocatalyst was reported that reduced molecular N 2 , supplied by bubbling air through water free of sacrificial reagents, to form NH 3 with apparent quantum yields of 0.73 % using 500 nm radiation and 0.25 % for 600 nm radiation [223]. Very recently, Cu-doped TiO 2 nanosheets were found to be effective in fixing N 2 in water under irradiation at wavelengths up to 700 nm, forming NH 3 [224]. In both cases, it was proposed that oxygen vacancies promoted the chemisorption and activation of molecular N 2 . In general, there is insufficient information on the overall reaction mechanism because of the lack of analysis of the oxidation products in most experiments. Likewise, the physical and chemical properties of defects such as oxygen vacancies are not fully understood, especially their evolution under reaction conditions, making it difficult to design and control them.

Opportunities and challenges
For electrochemical production of NH 3 , it has been estimated that a Faradaic efficiency > 50 % and an NH 3 production rate > 4-9 × 10 −7 mol s -1 cm -2 are required for industrial-scale operations [213]. To date, the reported rates of electrochemical NH 3 synthesis have been mostly on the order of 10 -8 mol cm -2 s -1 at best [138,213]. In addition, the Faradaic efficiency tends to decrease with increasing current density in a given system [213], suggesting that the H 2 evolution reaction is responsible for the decline in performance.
For the photocatalytic process, based on past work related to the development of photocatalysts for overall water splitting, the most important requirement is to reduce the density of defects so that active photoexcited electrons and holes can diffuse from the bulk of photocatalytic materials to the surface reaction sites [130]. Defect states tend to trap excited carriers while enhancing charge recombination, which are detrimental to effecting thermodynamic uphill reactions. For ammonia synthesis, the product, NH 3 , is also highly soluble in water, whereas the reactant, N 2 , is not, which is highly unfavorable on a thermodynamic basis. Consequently, it is a concern that NH 3 may not be able to be accumulated at sufficiently high concentrations to be practical because of its oxidation by photoexcited holes. In fact, it is debatable whether photocatalytic systems capable of efficiently evolving NH 3 can ever be constructed.
Low-temperature activation of the triple bond of molecular N 2 is an interesting but challenging research goal, and a variety of materials and reaction systems have been proposed for this purpose. For electrochemical and photocatalytic processes, it is important to consider the origin of NH 3 during experimental trials, particularly when the NH 3 evolution rate or concentration are very low. NH 3 is the primary pollutant emitted from animal production facilities and from field-applied NH 3 -based fertilizers [225]. Thus, the background concentration of NH 3 in the environment varies seasonally and by region, and it has increased over time historically. NH 3 is even found in human breath and can be generated from impurities in the N 2 stream employed in the synthesis system [225]. Furthermore, because it is both polar and basic, NH 3 is readily adsorbed onto, and dissolved into components of the experimental apparatus and test solution, introducing additional variations in the background NH 3 concentration. It is therefore very important to measure the ambient NH 3 concentrations in any aqueous samples to avoid false readings that occur rather frequently [135]. In electrochemical experiments, procedures using open-circuit conditions and employing Ar instead of N 2 should be applied when estimating the background NH 3 concentration, because electrochemical NH 3 production via N 2 reduction does not occur under these circumstances [226]. In addition, it is strongly advisable to perform isotope-labeling experiments to verify the origin of nitrogen atoms in the NH 3 product.

Concluding remarks
The two important chemicals focused in this article, NH 3 and H 2 O 2 , are uniquely interesting in that they can be produced readily by thermal, electrochemical, and photocatalytic techniques starting from the same set of reactants. Thus, they offer the rich possibility of mutually beneficial exchange of expertise and knowledge among the various approaches. To date, there is very limited cross-fertilization of knowledge and ideas across the disciplines. Undoubtedly there are common features among the approaches. For example, for H 2 O 2 production, the need to activate O 2 molecule without dissociation into O atoms, the ability to avoid decomposition of the H 2 O 2 product and prevent its further oxidation to H 2 O and O 2 or hydrogenation are independent of the approach. Likewise, stability of the catalyst and electrocatalyst against leaching is a common concern, and coating appears to be an approach applicable to all processes. What is not yet established is whether these issues share the same or different dependence on catalytic/electrode material properties for the different approaches. In other words, what knowledge is directly transferable? For those that are not, what is the appropriate modification necessary? Is there knowledge that should not be transferred because they are based on different physical principles?
Because less is known about electrochemical or photocatalytic synthesis of NH 3 , it is not known if the mechanisms developed for the thermal catalytic approach apply directly, especially requirements of the active sites. This leads to the same question as above: are there general guiding principles that would inform whether certain surface reaction intermediates and their reactions would be common to the different approaches and which would not. Under what circumstances would the reaction step of the highest potential barrier be the same or different? Are the critical steps determining selectivity different and is it possible to predict this a priori? Fig. 16. A scheme demonstrating the concept of sustainable ammonia synthesis based on an electrochemical lithium cycling process. Reprinted with permission from Ref. [220]. © 2017, The Royal Society of Chemistry.
For the specific reaction of electrochemical and photocatalytic H 2 O 2 and NH 3 synthesis, reactions have all been conducted under well-controlled conditions in laboratory experiments, primarily with the aim of increasing the production rate. In some cases, only the detection of intended products is presented. However, at the proof-of-concept stage, the mass balance of reduction and oxidation products should be determined whenever possible, because this information is essential to rule out the involvement of unexpected impurities or unrecognized reaction pathways. The supply and conversion of reactants are also important considerations. These information are necessary to determine the feasibility of scaling up novel production processes and whether such processes are sufficiently energy-and cost-effective to replace current industrial technologies.
Even if H 2 O 2 and NH 3 can be accumulated at moderate concentrations, separation, purification and concentration of these products will require additional energy inputs. Therefore, reactors and reaction processes including separation and recycling will be essential for practical applications, similar to the design of facilities for solar hydrogen production by water splitting [130]. Water electrolysis powered by photovoltaics is one of the most efficient means of producing renewable H 2 , but this process is thought to be too expensive to supply H 2 at a price acceptable by the market [227]. The supply of H 2 by solar thermochemical reactors using present-day technology [199] will also produce NH 3 , but at a cost above the acceptable market price (currently $0.5 kg −1 ) [214]. For all these reasons, the development of technology that will radically change the existing chemical industry will require input from the field of chemical engineering.

Declaration of Competing Interest
All authors have declared no financial interest associated with this manuscript.