Intrinsic Electrocatalytic Activity for Oxygen Evolution of Crystalline 3d‐Transition Metal Layered Double Hydroxides

Abstract Layered double hydroxides (LDHs) are among the most active and studied catalysts for the oxygen evolution reaction (OER) in alkaline electrolytes. However, previous studies have generally either focused on a small number of LDHs, applied synthetic routes with limited structural control, or used non‐intrinsic activity metrics, thus hampering the construction of consistent structure–activity‐relations. Herein, by employing new individually developed synthesis strategies with atomic structural control, we obtained a broad series of crystalline α‐MA(II)MB(III) LDH and β‐MA(OH)2 electrocatalysts (MA=Ni, Co, and MB=Co, Fe, Mn). We further derived their intrinsic activity through electrochemical active surface area normalization, yielding the trend NiFe LDH > CoFe LDH > Fe‐free Co‐containing catalysts > Fe‐Co‐free Ni‐based catalysts. Our theoretical reactivity analysis revealed that these intrinsic activity trends originate from the dual‐metal‐site nature of the reaction centers, which lead to composition‐dependent synergies and diverse scaling relationships that may be used to design catalysts with improved performance.


Introduction
Thee lectrochemical generation of hydrogen from the splitting of water is an established process to convert electricity into the chemical energy of hydrogen, af uel with high gravimetric energy density.W ater splitting electrolyzers are commercially available,and in these devices the hydrogen evolution reaction (HER) at the cathode is accompanied by the oxygen evolution reaction (OER) at the anode.O ne of the major conversion efficiency losses in electrolyzers is associated with the catalysis at the anode,w here protoncoupled multi-electron transfers and multiple reaction intermediates result in slower kinetics than in the mechanistically simpler HER. In the acidic environment of polymer electrolyte membrane (PEM) electrolyzers,I r-based oxides represent the state of the art OER catalysts. [1] In contrast, in neutral [2] or in alkaline environments non-precious metalbased oxide catalysts,often based on 3d transition metals such as Mn, Fe,Co, and Ni, are the catalysts of choice. [3] Systematic comparative studies with variable compositions and atomic structures,a sw ell as the development of systematic benchmarking protocols,have,inturn, provided useful strategies to sift through the vast number of OER catalysts in non-acid electrolytes.
There have been several such systematic studies,b oth experimentally and theoretically, [4] since the pioneering work on OER catalysis by Rüetschi and Delahay in 1955, [5] Bockris and Otagawa, [6] and Tr asatti [7] in the 80s.Inalkaline environments,t he focus of comparative studies has been primarily placed on electrodeposited thin film oxide catalysts.O ne OER benchmarking study was conducted by Jaramillo and co-workers,comparing selected electrodeposited Ni-and Cobased metal oxide catalysts and an IrO x standard. [8] Other comparative studies focused on transition metal hydroxides and oxyhydroxides (in the following generally indicated as (oxy)hydroxides), [9] which often form by surface reconstruction or amorphization on the surface of metal oxide catalysts in alkaline electrolytes. [10] Boettcher and co-workers investigated as eries of electrodeposited (oxy)hydroxides in alkaline electrolyte (1 MK OH), where purification of the electrolyte to eliminate trace amounts of Fe impurities was performed for non-Feb ased catalysts. [11] Their reported activity trend was: T hese results highlighted the critical role of Fe in enhancing Ni and Co based (oxy)hydroxide catalyst activity, in agreement with early work by Corrigan. [12] More recently, Chung et al. suggested that Fe dissolution and electrochemical re-deposition over 3d transition-metal hydr(oxy)oxide clusters yields dynamically stable Fe active sites,f urther underlining the importance of Fe for highly OER active and stable oxyhydroxide catalysts in alkaline environments. [13] While the established reactivity trends of electrodeposited metal oxide catalysts in alkaline environments are compelling,afirm correlation of reactivity with their local or long range atomic structure has remained elusive.T his is because electrodeposited metal hydroxide OER catalysts tend to possess low crystallinity,which is why they are referred to as XRD-amorphous,a nd why advanced X-ray characterization techniques are necessary to clarify their individual local atomic structures and correlate them with their activity. [14] Well defined long-range ordered metal (oxy)hydroxides phases can be prepared by solvothermal homogeneous precipitation methods, [15] and their formation can be validated by X-ray diffractometry.However,for the purpose of acomparative study,the synthesis procedure of aparticular desired crystalline phase cannot be easily generalized and requires individual case-by-case adaption for all elemental combinations of the metal hydroxide ensembles.T his is why previous studies focused either on aNi-or aCo-based hydroxide series, where synthetic protocols are better transferable. [16] In contrast, high throughput screening studies [17] investigating oxides and hydroxides beyond binary metal compositions were also reported. [18] While high throughput methods rapidly collect data across al arge pool of elemental compositions, they rely on standardized routine synthesis routes,a nd therefore often fail to achieve synthetic control over the crystal phase and structure of the catalysts.
In combination with experiments on structurally-selected catalysts,D FT calculations permit understanding of their reactivity trends.O nt he one hand, advanced designing and screening methods have been developed to identify OER catalysts with superior activity.T hese methods include scaling-based and scaling-free optimization, [19] the electrochemical step symmetry index, [20] overpotential dependent volcano plots [21] and overpotential-dependent reaction free energies, [22] among others. [23] On the other hand, the active phases and catalytic mechanisms are still under debate even for classic OER catalysts such as IrO x . [24] Fora lkaline OER on transition metal layered double hydroxides (LDH), apart from the promoting effects of Fe in NiFea nd CoFe LDHs, [15,25] ag eneral understanding has not been reached. Forexample,itisunclear whether the OER trends that were established on those two most active catalysts can be extended upon introduction of ad ifferent element into the same hosts.I ti sa lso unknown whether those advanced designing and screening methods,w hich were developed on the basis of diverse model systems,a re applicable to more specific and more realistic OER catalysts.Such afundamental understanding is not only important for providing scientific insights to rationalize experimental trends,but is also crucial for establishing amore complete picture of OER mechanisms that can be used to calibrate the above designing methods and theories.These achievements would permit feasible design of OER catalysts with performance beyond the state-of-the-art catalysts.
As previously discussed, current understanding of reactivity trends of metal (oxy)hydroxide OER catalysts in alkaline environments considers the presence of Fe as an ecessary,y et not sufficient, criterion for high OER catalytic activities.I np articular,N iFea nd CoFe( oxy)hydr-oxides are generally reported to be the most active catalysts in the various activity trends reported in the literature,r egardless of the strategy of normalization of the activity metrics (see as ummary in Supplementary Table 1). [8,11, 13, 16b] This is because they have much higher intrinsic activity than the other compounds.H owever,a part from these top catalysts, significant differences can be found in the order of the remaining catalysts;f or example,c ontradictory reactivity orders were observed for NiMn and NiCo (oxy)hydroxides/ LDHs versus Ni and Co (oxy)hydroxides (see as ummary in Supplementary Table 1). [8, 11,13, 16b] These and similar discrepancies are most likely caused by three factors:1 )d ifferent crystal phases and catalytically active centers between the catalysts,d espite similar overall elemental compositions,2 ) use of non-intrinsic activity metrics or different approximations to the intrinsic activity,w hich may depend on experimental protocols and methodology,a nd 3) presence of unintentional impurities altering the surface kinetics,a s previously pointed out in the case of Fe.W hile broad trends studies have been reported for XRD-amorphous materials, those for well-defined (oxy)hydroxide crystalline phases with long-range order are scarce.A lso,t he existing activity trend studies have largely considered application-relevant activity metrics,such as OER overpotentials at fixed current densities or, for instance,c atalytic lower-limit turn-over-frequencies based on total metal content rather than on surface area.
In this work, we seek to fill some of the above knowledge gaps by combining well-defined experiments with rigorous calculations to study the intrinsic OER activities of welldefined crystalline layered double hydroxides (LDH) across the group of late 3d transition metals in alkaline conditions.In comparison to previous studies, [11] we synthesized crystalline LDH nanoplatelets,i nstead of XRD-amorphous (oxy)hydroxide thin films,u sing new individually developed synthesis strategies with atomic structural control based on co-and homogeneous precipitation instead of cathodic electrodeposition. In addition, we derived their intrinsic activity through electrochemical surface area (ECSA) normalization, instead of total metal-based turnover frequencies.I nc omparison to our previous publication, [15] which identified and elucidated the catalytically active phase,t he reaction center,a nd the OER mechanism of the most active NiFeL DH and CoFe LDH catalysts,t his work systematically studies the intrinsic OER activity of aseries of crystalline M A M B LDHs,providing af undamental understanding of the synergy (both positive and negative) between the dual metal atom sites that form the reaction centers,a ss uggested by our DFT calculations.O ur study is hence characterized by:1 )u nique preparation protocols with individually adjusted synthesis recipes that offered unprecedented control over the crystallinity and long range order of the desired alpha phase for both the Ni-and the Co-based LDH series;2 )thec omparison of surfacenormalized intrinsic activity based on av ery recently proposed advanced evaluation method of the electrochemical surface area of oxyhydroxide surfaces [26] (this ECSA normalization is important, since ah igher surface area can provide ah igher number of OER active surface sites,r esulting in ahigher activity,despite alower intrinsic activity of the sites); 3) electrolyte purification prior to the electrochemical experi-ments;a nd 4) theoretical insights that point to ad esign space of catalysts with OER performance beyond what is inferred from idealized scaling relationships. [4a, 19, 23, 27]

Results
Preparation and physico-chemical characterization NiFe, NiCo,N iMn, CoFe, Co II Co III , CoMn layered double hydroxides (LDH) and b-Ni(OH) 2 and b-Co(OH) 2 were synthesized by specific wet chemistry methods that led to the desired crystallized phases,h ydrotalcite-like for the LDHs and brucite-like for the two reference b-hydroxides.T he LDH catalysts can be grouped into two series,aNi series and aC os eries,b yd escribing the LDH formation as the result of the incorporation of as econd metal with oxidation state + 3i nN i II (OH) 2  ,and water molecules also occupy the interlayer region. In contrast, the brucite-like phases do not have species intercalated between the layers.T he XRD patterns,s hown in Figure 1a,c onfirmed that the targeted crystal phases were obtained. To achieve this,s ynthesis conditions were optimized specifically for each catalyst (see Supplementary Figure 1,2 and related discussion in supporting information for details). While the presence of < 1-2 nm nanoparticles sized below coherently scattering domains could not be observed using microscopic techniques, it cannot be generally excluded. Nonetheless,t he dominant LDH XRD peak reflections evidence the incorporation of the second metal (Fe, Co and Mn) in the Ni(OH) 2 or Co(OH) 2 hosts for all samples.T ov erify the LDH formation further,t he metal oxidation states were estimated by the analysis of the X-ray absorption near-edge structure (XANES) (Supplementary Figure 3a nd Supplementary Table 2), and found fully consistent with the expected M A II M B III LDH chemical formula. Theoxidation states for Co and Mn in CoMn LDH, 2.6 and 3.4 respectively,a ppeared slightly higher than expected, which is attributed to an excessive oxidation and deprotonation due to the hydrogen peroxide used in the synthesis.T he selected metal compositions included in the present trend study are listed in Table 1 and determined by inductively coupled optical emission spectroscopy (ICP-OES). Forh etero-metals LDHs,a3:1 ratio of the host metal to the dopant metal was targeted, which falls in the range where maximum activity is observed Note that carbonate is chosen as the representative anion intercalated within the LDHs in the model, though Co II Co III LDH actually incorporates Br À .The reference patterns for hydrotalcite (red, PDF# 00-035-0965)a nd brucite-like b-Ni(OH) 2 (blue, PDF# 00-014-0117) are also shown. [a] Targeted elemental ratios are given in brackets. The higher metal weight %inthe two brucite-like catalysts is due to the absence of intercalated water and anions.
for NiFeand CoFeoxyhydroxides. [28] Theonly small variation was necessary for NiCo LDH, with aratio of 2:1, for the phase purity requirement explained above.T he morphology was investigated by transmission electron microscopy (TEM) and consistent with the expected nanoplate shape for these layered materials (Figure 1b). Then anoplates showed large roughened terrace planes and thin edges,w ith dimensions that were found to vary strongly between the different catalysts (Figure 1b and Supplementary Table 3).

Surface redox chemistry and OER overpotentials
Each member of the prepared array of OER electrocatalysts was subsequently tested using rotating disk electrodes (RDEs) in 0.1 MK OH, depositing the same catalyst loading of 0.1 mg cm À2 on glassy carbon cylinders.T he electrolyte was purified of Fe traces according to published methods,u sing sacrificial Ni(OH) 2 and Co(OH) 2 for the Ni II M B and Co II M B LDH series,respectively. [28a, 29] An activation treatment consisting of cycling the potential from the resting potentials ( 1V RHE )t oO ER relevant potentials is first applied. During this treatment, the peaks in the cyclic voltammetry associated with the surface metal redox chemistry grew in intensity,w hich is attributed to more sites becoming electrochemically accessible.T he potential cycling also causes structural and electronic transformations from hydroxides to oxyhydroxides,w hich can be reversible,f or NiFeLDH, or only partially reversible,for CoFeLDH. [15] In Figure 2a and b, the 10 th cycles of each catalyst are compared. These catalysts showed the rich surface redox chemistry of the component 3d transition metals.T he characteristic Ni II oxidation peak and the corresponding reduction peak are shown in Figure 2a at potential lower than the OER onset. TheN io xidation peak is shifted more anodically for NiFe LDH, more cathodically for NiCo LDH, and remains roughly unshifted for NiMn LDH in respect to Ni(OH) 2 .T hese shifts indicate that an electronic effect is affecting the stability of Ni centers in the + 2oxidation state in the various LDHs,which is consistent with incorporation of the second metal. Thearea of the peaks also changes significantly,with Ni(OH) 2 showing the highest peak. This increase may be mainly caused by the fact that it has the highest surface area among all samples synthesized in the current work, as quantified in the next section, and partially related to alarger amount of Ni atoms in Ni(OH) 2 than in the other LDHs.F igure 2b shows the cyclic Figure 2. Cyclic voltammetric curves for the Ni series (a) and the Co series (b). The 10th cycle of the cyclic voltammetric activation treatment is shown. Scan rate:5 0mVs À1 .L inear sweep voltammetry curves recorded at 1mVs À1 (c) and corresponding OER overpotentials at 10 mA cm À2 (d). In (d), red is for the Co series and blue for the Ni series. Error bars represent the standard deviations of the averaged values of multiple samples. Curves in (c) for NiFeL DH, CoFe LDH, Ni(OH) 2 and Co(OH) 2 and corresponding data in (d) have been reproduced from Ref. [15]. voltammetric curves for the Co series,where double oxidation peaks and corresponding reduction peaks are observed, which is typical of Co-based (oxy)hydroxides and attributed to Co II and Co III oxidation. Potential shifts of the oxidation peaks were small in comparison to the Ni series and difficult to evaluate,p artly due to the broadness of the peaks.T he less anodic redox couples,assigned to oxidation/reduction of Co II , appear considerably suppressed and slightly anodic shifted for CoMn LDH in respect to the other catalysts.The same peaks in the CV of CoFeLDH do not show appreciable anodic shifts with respect to Co(OH) 2 ,a se lsewhere reported for electrodeposited CoFeo xyhydroxides,w here the shift became particularly large at above 30 Fe at %. [28a] This might be due to relatively low Fe at %inthe CoFeLDH. Co(OH) 2 showed the smallest peak areas,which is most likely related to the low surface area, as will be discussed further below.
After the cyclic voltammetric activation, linear sweep voltammetry (LSV) at the slow scan rate of 1mVs À1 was performed to evaluate the OER activity (Figure 2c). The OER overpotentials evaluated at 10 mA cm À2 are shown in Figure 2d.F e-containing NiFea nd CoFeL DHs lead the activity trend and are followed by the other Co containing catalysts.N i(OH) 2 and NiMn LDH occupy the lowest positions in the trend with the highest overpotentials.W hile the positive synergy between M-Fei sc lear in Figure 2d and well documented in the literature,the synergy between M-Mn is ambiguous in both Figure 2d and in literature.F or NiMn and NiCo,weobserved aneutral or slightly negative synergy based on overpotential comparison, which is consistent with the results reported by Markovic, Boettcher and co-workers, [11,13] but in contrast to the positive synergy reported by Jaramillo,C alle-Vallejo and co-workers. [8, 16b] ForC o-Mn, we observed as urprising result, with ap ositive synergy in comparison with Co(OH) 2 ,c onsistent with results reported by Hu and co-workers, [30] but as lightly negative synergy in comparison with CoCo LDH. Such aconclusion can be more clearly established at al ower current density,1 mA cm À2 (Supplementary Figure 4,5). Below,wewill use ECSA-based intrinsic activity to reconcile these apparent discrepancies.

ECSA determination and OER intrinsic activity
Theo verpotential at as pecified current density is ac onvenient metric for comparing the OER activity of the investigated catalysts,since it is related to the potential losses during operation, which is important for the commercial application of the catalysts in electrolyzers.H owever,t his metric convolutes several parameters,a nd al ower overpotential might be obtained by an increase in the number of active sites or by the presence of adifferent specific site with higher reactivity.T urn over frequencies (TOF) are calculated to estimate the intrinsic activity of catalysts. [31] However,f or LDHs catalysts,t he nature of the active sites is often unknown, and their surface concentration is also difficult to calculate.I no ur case,t he roughening of the edges of the nanoplates (Figure 1b-i)a nd the possible partial exfoliations [32] of part of the terraces of the nanoplates,t ypical of layered materials,c omplicate the application of geometric models to estimate the surface concentration of active sites. Therefore,"bulk" TOF, considering all the metal centers,are often calculated and indicated as al ower limit of the TOF (Supplementary Figure 6). However, this approach limits their effectiveness in providing real intrinsic activity trends.
Specific activities obtained by normalizing the current to the electrochemically active surface area (ECSA) can be used as alternative methods if the TOFs cannot be accurately calculated. ForL DHs,t he evaluation of the ECSA is not straightforward, due to 1) potential dependent changes of the electrical conductivity of LDHs, [33] 2) anarrow or non-existing potential window that is free of faradaic current in the conductive regime,3 )m etal oxidation peaks in cyclic voltammograms that overlap with the OER faradaic current, and 4) difficulties in obtaining model catalysts with smooth planar surfaces for the conversion of the calculated values, that is,c apacitances,i nt he unit of an area. [34] Among the various methods to estimate the ECSA, we used the capacitance of the adsorbed OER intermediates (C a )t hat was calculated by electrochemical impedance spectroscopy (EIS) at 1.6 V RHE .T his potential is more anodic than the oxidation peak associated to the oxidation of Ni II and Co II ,in which NiM and CoM LDH become conductive. [11] The equivalent circuit which was used to fit the impedance data was adapted from Watzele and Bandarenka, [26] and previously introduced and discussed by Lyons and Brandon. [35] Thevalue C a was calculated from the associated constant phase element and the parallel resistance (Figure 3a). Thespecific unit area capacitance (C S )o f0 .3 mF cm À2 that was obtained from as mooth Ni(OH) 2 surface by Watzele and Bandarenka was used to convert the C a capacitances into real surface areas. [26] Ideally,f or each material, as pecific capacitance obtained by asmooth unitary surface of the same material should be used. However,b ecause of the difficulty of obtaining smooth surfaces for metal hydroxide catalysts,s pecific capacitances for surfaces of equal roughness factor and similar (few nanometer) height variation as for Ni(OH) 2 are not currently available.Itisexpected that deviations of specific capacitance among these catalysts would be small, however,d ue to the similarity in their crystal structure.T hus,t he value from Ni(OH) 2 was used for all catalysts,a sw as assumed for the specific double layer capacitance in metal oxides. [8] Under this assumption, this method for determining the ECSA for transition metal LDH electrocatalysts solves and overcomes all the mentioned issues. Figure 3a shows the ECSA of the catalysts.I nterestingly, NiFeLDH and Co(OH) 2 ,which have the largest lateral sizes (Figure 1b- Figure 3b). This behavior supports the validity of the calculated ECSAs.W ewill show that while those differences do not influence the order of the most and least active catalysts,t hey have as ubstantial influence on those with intermediate activities,t hat is,F e-free Co-containing LDHs. Such an influence has aprofound impact on the reconciliation of discrepancies in literature and on the development of af undamental understanding of synergy between M A -M B . TheE CSA-normalized current densities,a nd the derived trend in overpotentials,i ss hown in Figure 3c,d (the specific current density trend at the overpotential of 350 mV is also shown in Supplementary Figure 7). NiFeL DH is the most active catalyst, followed by CoFeL DH. Ni(OH) 2 and NiMn LDH are the least active.T his trend implies an eutral or slightly negative synergy between Ni-Mn, which reconciles the discrepancies in literature. [8,11, 16b] Interestingly,due to the large differences in ECSA, the activity of CoCo LDH and Co(OH) 2 are now comparable.T his suggests that similar active sites are present in both catalysts during OER and that the presence of Br À anions in the as prepared CoCo LDH has anegligible effect (within error bar) on the activity.However, operando structural studies are necessary for definitive conclusions.F urthermore,t he specific activity trend showed that the intrinsic activity of these two Co catalysts is superior with respect to NiCo LDH, which showed one of the highest ECSAs.T his conclusion is in contrast to the other activity metrics that do not account for the active surface area. Thus, for Fe-free Co-containing species,t he orders are changed from CoCo LDH, CoMn LDH, Co(OH) 2 ,N iCo LDH in the geometric-area based overpotential trend to Co(OH) 2 ,CoCo LDH, CoMn LDH, NiCo LDH in the ECSA-normalized trend. Them ost notable change is that there is an egative synergy between Ni-Co and Co-Mn based on the intrinsic activity,while aneutral or positive synergy was derived based on the overpotentials in the current work and in some literature. [8,30] Thus,t or econcile those discrepancies in literature and achieve af undamental understanding of synergy between M A -M B ,n ormalization using ECSA is essential. Otherwise,contradictory conclusions may be drawn for acatalyst with the same composition but different ECSA due to different synthesis methods.T his may be also the reason for some other discrepancies in literature.B elow,w e will further study these synergies (both positive and negative) between M A -M B through DFT calculations.
Thep roposed intrinsic activity trend for crystalline transition metal LDHs is also important for the discussion about the effect of disorder on the activity for NiFeo xyhydroxides and, in general, for this type of catalysts. [36] While at rend for XRD-amorphous transition metal (oxy)hydroxides was provided by Boettcher and co-workers, [11] there was no comparable intrinsic activity trend to date for the corresponding crystalline LDH catalysts.F urthermore,i nc ontrast to an all metals-based TOF, which provides reasonable estimate for intrinsic activity only for very thin films, [11] the ECSA-based method is well suited for both amorphous and crystalline catalysts,a llowing the comparison of their activity using the same method in future works.

Revealing compositiondependent synergies through DFT calculations
Thea tomic-scale structures of the as-prepared phase and the active phase of NiFea nd CoFeL DHs,t he a-phase and the g-phase with water molecules and ions intercalated between layers,h ave been identified in our previous work. [15] Here,u sing those structures as the starting point, we studied the stability of M A M B LDHs synthesized in the current work through DFT calculations (Supplementary Table 4);a dditional details of the DFT calculations are given in the computational methods section in the supporting information. Theb ulk phase diagrams indicate that the phase transitions from a-NiM LDH (Figure 4a (Figure 2). These results suggest that g-phases (Figure 4c)a re the active phases for OER. Their geometric structures and electronic structures,i ncluding magnetic moments,a re summarized in Supplementary Figure 9a nd Table 5-9. To facilitate comparison with our experimental results,we focus on the NiM series which has alarger span of the activity than that of CoM series,and we used (01-10) surfaces that are exposed at the edge of the g-NiM LDH sheets to study their OER performance.The electrolyte was implicitly included in the calculations through the solvation corrections (Supplementary Table 10), which were evaluated in our previous work through ab initio molecular dynamics simulations with explicit liquid water filling of the vacuum region between the model and the image. [15] We first evaluated catalysts steadystate surface structure through surface free energy diagrams (Figure 4d,e). We found that, under OER conditions,t he surface bridge oxygen (O bri )s ites are saturated with H adsorption (H ad )b yf orming bridge OH* ("*" indicates acoordinatively unsaturated surface site or vacancy), and the coordinatively unsaturated metal sites (CUSs) are saturated by OH adsorption (OH*). Thus,OER on these surfaces starts from the deprotonation of the surface OH* species (OH* ! O*), the so-called Mars van Krevelen mechanism. We considered the following four consecutives steps for the OER: Thec orresponding reaction free energies are denoted as DG 1 (U), DG 2 (U), DG 3 (U)and DG 4 (U), respectively. U is the electrode potential in the reversible hydrogen electrode (RHE) scale.W en ote that this four-step reaction pathway has been corroborated by the spectroscopic identification of the OOH* intermediates on Au-a nd Co-based catalysts. [37] Thec alculated reaction free energies indicate that, consistently with our previous work, the oxidation of two-metal coordinated bridge OH* moieties is more favorable than that of one-metal coordinated atop OH*. Effectively  0.1 mA cm À2 from ECSA-normalized current densities,e xcept for NiMn (Figure 3d).
For g-NiMn LDH, the calculated overpotential implies as lightly negative synergy between Ni and Mn. As ac onsequence,the measured activity may stem primarily from Ni-Ni, instead of Ni-Mn reaction centers,d ue to their higher activity.F or g-NiCo,t hough the calculated overpotential is slightly smaller than that of g-NiNi, it is larger than that of the Co-Co reaction centers analyzed in our previous work (e.g. 0.6 V). [15] Thus,w hile there is ap ositive synergy between Ni and Fe, and Co and Fe, [15] our study suggests an egative synergy between Ni and Co,a nd Ni and Mn. Below,w ew ill demonstrate that these composition-dependent synergies originate from the unique geometric structures and electronic structures of the dual-metal-site reaction centers.T hose unique features have important implications for breaking scaling relationships and enlarging the design space of oxyhydroxide catalysts with OER activity beyond the state-of-the-art catalysts.
Revealing diverse OH-O scaling relationships at dual-metal-site reaction centers Figure 6s hows the OH-OOH and OH-O scaling relationships (the corresponding values are reported in Supplementary Table 12). Fort he OH-OOH scaling relationship,w e obtained as lope of one and an intercept of 2.94 (Figure 6a). It is worth noting that, for the intercept, the values of 3.0 AE 0.2 eV were reported in the literature,w ith ad ependence on the functional used in the calculations. [4a, 38] Thei ntercept in the current work is nearly identical to those obtained with the other dispersion corrected functionals. [38] However,t he OH-O scaling relationship has am uch more complex behavior than that of OH-OOH. While NiNi, NiFeand CoFefollow the ideal slope of 2, [27] NiCo and NiMn significantly deviate from this trend, resulting in actual slopes that generally vary from 0.4 to 2 ( Figure 6b). There are also cases of negative slopes (CoFe-NiCo and CoFe-NiMn), which break the ideal scaling relationship entirely.W en ote that due to the limited set of data points used for the calculation of the slopes and their small differences in energy, for example,a bout 50 meV for some cases,o ur work demonstrates the plausible existence of diverse slopes,rather than the specific values of these slopes.The diversity of slopes is inherently rooted in the unique thermodynamics of each individual element, which follow the general trends only when included with aset of elements with large energy span, on the scale of 5eV. [4a, 27, 39] In the literature,t hose deviations and possibilities of diverse slopes were usually treated as part of the error hidden behind the general trends.A st he absolute errors are usually on the order of 0.2 eV,and some individual  (b) and navy for Co in (c)) and the reaction intermediates (yellow is used instead of white for hydrogen and rose instead of red for oxygen). A dashed rose circle indicates the formation of asurface Ov acancy.The reaction free energies at 0Vare in black, 1.23 Vred, and the potentialw hen the potential-determining steps become downhill in blue. Ideal steps of 1.23 eV are shown with dashed gray lines. The potential limiting steps are highlighted with thick lines. The numbers on the atoms at the reaction centers are their magnetic moments in the Bohr magneton (m B ). The oxidation state, which is deduced based on the intrinsic magnetic moments (see Supporting Information Table S5-S9), are given in the reaction free energy diagram. deviations are 0.5 eV or larger, [27, 38a, 39] disentangling the errors and deviations may significantly enlarge the design space for catalysts,a swepropose in this work.
Thes caling relationships and their implications for catalyst design are illustrated in the plot of the 2D activity volcano (Figure 6c). Thetop of the volcano is determined by the OH*!O*!OOH* steps and so is limited by the OH-OOH scaling relationship.I nc ontrast, the above predicted overpotentials for the g-LDHs are determined either by the OH*!O* step or by the OOH*! O 2 + *s tep.T his is also shown in Figure 6c,w here the data points are either in the zone of OH*!O* or OOH*!O 2 + *asthe potential limiting step.Itisworth noting that, for the cases with OOH*!O 2 + * as the potential limiting step,the enrichment of OOH* on the surface makes its experimental measurement feasible.T his is likely the reason for the experimental observation of OOH intermediates on Au-based and Co-based catalysts. [37] As the top of the volcano is at the intersection of the OH*!O* step and the O*!OOH* step,n ot of any other two steps,t he identification of the OH*!O* and OOH*!O 2 + *s teps as the potential-limiting steps implies that the minimum overpotential that is determined by the intersection of these two steps is away from the top of the volcano.T herefore,e ven if the systems follow the ideal scaling relationships (e.g.O H-Os caling relationship with slope of 2), there would be no catalysts that could even approach the optimal activity dictated by the OH-OOH scaling relationship.T his is visualized in Figure 6c,w here the line representing the ideal OH-O scaling relation does not cross the top of the volcano.B elow, we will demonstrate that, while OH-OOH scaling relationships determine the tops of volcano curves,t he OH-O scaling relationships determine whether it is possible to approach them. We will show that the dual-site nature of the reaction centers in the LDHs studied in the current work provides the possibility of breaking the OH-O scaling relationship and enlarges the design space of oxyhydroxide catalysts with OER activity beyond the state-of-theart catalysts.
To understand the origin of diverse scaling relationships and its implications,w ep erformed electronic structure analyses of the reaction centers under steady state conditions and in response to the adsorption of OER intermediates (Figure 5a nd Figure 6d). Under OER conditions,f or g-NiOOH (g-NiNi LDH), the atoms at the reaction center are O-bridged Ni 3+ -Ni 4+ (the oxidation states are determined on the basis of the intrinsic magnetic moment of each element, along with considerations of the charge balance-see Supplementary Table 5-9 and computational methods). ForN iM LDHs,F ea nd Mn atoms occupy Ni 4+ sites by forming Obridged Ni 3+ -Fe 4+ and Ni 3+ -Mn 4+ reaction centers,w hile Co atoms occupy Ni 3+ sites by forming an O-bridged Co 3+ -Ni 4+ reaction center. Based on electronic structure analyses (Figure 5and Figure 6d), three major features can be derived for the redox of the reaction centers during the OER.
First, although ar eaction center is formed by two metal atoms,o nly one of them plays am ajor role in each reaction step of the OER. Fore xample,f or g-NiOOH with an Obridged Ni 3+ -Ni 4+ reaction center,3+ sites play amajor role on the OH*!O* step,a se videnced by the accompanying Ni 3+ !Ni 4+ and the intact oxidation state of Ni 4+ site ( Figure 5 and Figure 6d). On the other hand, the 4 + sites of Ni 3+ -Ni 4+ reaction center play amajor role in the OOH*!O 2 + *step, as evidenced by the accompanying Ni 4+ !Ni 3+ and the intact oxidation state of Ni 3+ sites (Figure 5and 6d). ForNiCo LDH with the Co 3+ -Ni 4+ reaction center, similar to that of gamma-NiOOH, 3 + sites play am ajor role on the OH*!O* step, and 4 + sites play am ajor role on the OOH*!O 2 + *s tep (Figure 5a nd Figure 6d).
Second, despite of the above similarity between Obridged Ni 3+ -Ni 4+ and Co 3+ -Ni 4+ reaction centers,metal sites that are primarily involved in each reaction step cannot be deduced ap riori. Fore xample,f or NiFew ith the Ni 3+ -Fe 4+ reaction center, 4 + sites,instead of 3 + sites,play amajor role on the OH*!O* step,a se videnced by the accompanying Fe 4+ !Fe 5+ ,w hile 3 + sites,i nstead of 4 + sites,p lay am ajor role on the OOH*!O 2 + *s tep,a se videnced by the accompanying Ni 3+ !Ni 2+ ( Figure 5). We note that, as OH*!O* is the potential limiting step,t he formation of transient Fe 5+ from the oxidation of Fe 4+ occurring during the OER might not be directly observable in conventional experiments.O nt he other hand, the Fe 4+ oxidation state, which is present under steady state conditions,h as been observed through operando Mçssbauer spectroscopy. [40] Third, in addition to dual metal atoms that form the reaction centers and are in contact with the reaction intermediates,t hird atoms that are not in direct contact with the reaction intermediates also can be involved in the reaction. Fore xample,f or NiMn LDH with the Ni 3+ -Mn 4+ reaction center, subsurface Ni 4+ sites,instead of either surface Ni 3+ or surface Mn 4+ ,play amajor role on the OOH*!O 2 + * step,a se videnced by the accompanying Ni sub 4+ !Ni sub 3+ transition ( Figure 5).
Thus,i ti st he dual-metal-site feature of the reaction centers,the composition-dependentinvolvement of the metal sites,a nd the possible involvement of at hird site that allow OER on LDHs to deviate from the ideal OH-O scaling relationship.T of urther understand the OH-O scaling relationship in LDHs,F igure 6d summarizes the features of OH and Oadsorption on the reaction centers.The OH adsorption on the vacancies (V) of the Ni 3+ -V-Ni 3+ ,N i 3+ -V-Mn 4+ -(Ni 3+ sub ), Ni 3+ -V-Co 3+ and Ni 2+ -V-Fe 4+ reaction centers leads to the oxidation of one surface or subsurface site (i.e.N i 3+ , Ni 3+ sub ,N i 3+ and Ni 2+ ,r espectively), as well as the formation of Ni 3+ -Ni 4+ ,N i 3+ -Mn 4+ ,C o 3+ -Ni 4+ and Ni 3+ -Fe 4+ reaction centers (Figure 5a nd Figure 6d). TheOadsorption leads to the oxidation of two sites,e ither two surface sites or one surface site and one subsurface site (Figure 5and Figure 6d). Due to the different chemical nature of those two sites,their average oxidation energy likely deviates from the oxidation energy of each individual site,w hich leads to the deviation from the ideal scaling relationship.F or Oa dsorption, which can be considered as two consecutive steps,O Ha dsorption and OH oxidation, that occur on two different sites,t he deviation from the ideal scaling relationship implies that it may be possible to disentangle and tailor the energetics of each individual step.Such atunability provides adesign space of catalysts with OER performance (Figure 6c and d) beyond what is constrained by the ideal scaling relationships.
Fore xample,b ased on the nearly ideal OH-O scaling relationship of NiNi-NiFe-CoFe, the OER activity can be improved only by slightly decreasing the binding energy of the OH* intermediate and the reaction energy of the OH*!O* step,i nc omparison with that on NiFe. This is because decreasing the binding energy of the OH* intermediate leads to,onthe one hand, adecreased overpotential for the OH*! O* step,b ut on the other hand, it causes an increased overpotential for the OOH*!O 2 + *s tep.W hen the overpotentials of those two steps become identical, further decreasing the binding energy of the OH* intermediate would make OOH*!O 2 + *t he potential limiting step, resulting in an increased overpotential, as with the case of CoFeL DH. In contrast, the OH-O scaling relationship of NiFe-NiCo-NiMn with aslope of 0.4 could lead to adecreased overpotential on both the OH*!O* step and the OOH*! O 2 + *step,reaching the minimum dictated by the OH-OOH scaling relationship by weakening the OH adsorption energy by 0.25 eV in comparison with that on NiFe( see Figure 6bd). Thecorresponding reduction of the overpotential implies an improvement in the OER activity of over three orders of magnitude.
As OH* and OH*!O* are two independent descriptors that uniquely determine the overpotential in the 2D volcano, Figure 6d also provides af acile way to predict the OER overpotential, and potentially sheds light on possible combinations of reaction centers with further improved activity.W e note that other strategies and descriptors have been proposed for the design of OER catalysts with significantly improved performance,s uch as the scaling-based and scaling-free optimization, [19] the electrochemical step symmetry index, [20] the overpotential dependent volcano plot, [21] and the overpotential-dependent reaction free energy, [22] among others. [23] To design new catalysts with the activity that is up to three orders of magnitude higher than that of the state-of-the-start catalysts,h owever, we propose that ap ossible route is to break the OH-O scaling relationship by forming binary metal oxyhydroxides with dual metal sites at the reaction centers,or by introducing athird element into NiFeorCoFeL DHs.

Conclusion
By combining well-defined experiments with rigorous calculations,w ec onducted as ystematic analysis of the intrinsic OER activity and composition-dependent synergy of acomprehensive set of crystalline LDH catalysts prepared from late 3d transition metals (Ni, Co,F e, Mn). Au nique synthesis protocol was developed for each and every compositional combination to ensure the presence of awell-defined atomic crystal structure,the alpha crystal phase,atthe outset of the kinetic tests.A ll catalysts showed the characteristic nanoplatelet morphology.T he intrinsic OER activity was obtained by normalization using the electrochemical surface area that was determined through the analysis of the adsorbed OER intermediate capacitance.S ynergies (both positive and negative) arising from each specific binary metal LDH combination resulted in characteristic variations in the surface redox electrochemistry and the catalytic OER activity.
Thea ctivity trend in terms of the overpotentials at fixed geometric electrode area-based OER current densities revealed that Fe-containing LDHs invariably displayed the highest kinetic OER activities,followed by Fe-free Co-based LDH catalysts,a nd finally by LDHs without Fe nor Co. Tr ends in intrinsic OER activities obtained by ECSA normalization confirmed this trend, which in its most generalized form reads:N iFeL DH @ CoFeL DH > Co(OH) 2 ,C oCo LDH > NiCo LDH, CoMn LDH, @ Ni(OH) 2 ,N iMn LDH. We found that the adopted ECSA normalization is critical to eliminate the effects of varying real surface areas on the activity and reconciles the discrepancies of overpotential based activity in literature.A lso,u nlike overpotential and total metal mass-based activities,r eal surface-area normalized OER activities have direct relevance for theoretical reactivity modelling. Our DFT calculations suggest that the above activity and the composition-dependents ynergy originates from the dual-metal site feature of the reaction centers. While this feature does not influence OH-OOH scaling relationship,i tl eads to diverse OH-O scaling relationships, including those with near-zero slopes and the negative slopes. This diversity provides aroute to approach the top of volcano curve that is dictated by the OH-OOH scaling relationship. Thus,catalysts with dual active sites provides adesign space of catalysts with OER performance beyond what is constrained by the ideal scaling relationships.