Water oxidation on a mononuclear manganese heterogeneous catalyst

Articles https://doi.org/10.1038/s41929-018-0158-6

Water oxidation on a mononuclear manganese heterogeneous catalyst Jingqi Guan   1, Zhiyao Duan2, Fuxiang Zhang   1, Shelly D. Kelly3, Rui Si4, Michel Dupuis1,5, Qinge Huang1, John Qianjun Chen3, Chunhua Tang6 and Can Li   1* Water oxidation is the prerequisite for dioxygen evolution in natural or artificial photosynthesis. Although it has been demonstrated that multinuclear active sites are commonly necessary for water oxidation, as inspired by the natural oxygen-evolving centre CaMn4O5, a multinuclear manganese cluster, whether mononuclear manganese can also efficiently catalyse water oxidation has been a long-standing question. Herein, we found that a heterogeneous catalyst with mononuclear manganese embedded in nitrogen-doped graphene (Mn-NG) shows a turnover frequency as high as 214 s−1 for chemical water oxidation and an electrochemical overpotential as low as 337 mV at a current density of 10 mA cm−2. Structural characterization and density functional theory calculations reveal that the high activity of Mn-NG can be attributed to the mononuclear manganese ion coordinated with four nitrogen atoms embedded in the graphene matrix.

C

onverting solar energy to chemical energy through artificial photosynthesis is a promising approach to address energy and environmental issues1. Water oxidation is the key step for both natural and artificial photosynthesis2–5. However, the sluggish kinetics of the water oxidation reaction (WOR) greatly limits the efficiency of artificial photosynthesis. In nature, the oxygen-evolving complex (OEC, a CaMn4O5 complex) in photosystem II can efficiently oxidize water with a turnover frequency (TOF) of 100– 400 s−1 (ref. 6), which is much higher than most catalysts used in artificial photosynthesis systems4,5,7. Inspired by nature’s OEC, many multinuclear transition metal complexes have been synthesized and tested for water oxidation, and some of them did show high activity for water oxidation, such as [OH2(terpy)Mn(O)2Mn(terpy)OH2]3+ (refs 8,9), Mn3Ca cluster10, [Mn4O4L6] (refs 11–13), Mn4Ca cluster14, [Mn4V4O17(OAc)3] (refs 3−15), Mn12DH (ref. 16), the dinuclear Ru complex17–19, [Co4(H2O)2(PW9O34)2] (refs 10−20) and [Feii4Feiii(µ​3-O) (µ​-L)6]3+ (ref. 21). It has been claimed that multinuclear sites play the crucial role in accelerating the O−​O bond formation in the WOR22,23. Although some mononuclear complexes (for example, [Mnv(N) (CN)4] (refs 2−24), Fe(TAML-F2-Cl2)OH2 (ref. 25), Ru(bda)(isoq)2 (ref. 26) and Cp*Ir(ppy)Cl (ref. 27)) showed high activity for water oxidation, the high activity might be attributed to the formation of dimeric species and cooperative mechanisms22,23. Even though a mononuclear manganese complex [(Py2N(tBu)2)Mn(H2O)2]2+ was reported to be active in electrocatalytic water oxidation, the overpotential was relatively high (approximately 800 mV)28. Although enormous efforts have been dedicated to mimicking the CaMn4O5 cluster in recent decades, it remains unknown whether mononuclear manganese is capable of being highly active in the WOR. In this study, we prepared a highly active heterogeneous water oxidation catalyst with a mononuclear Mn site coordinated to four N atoms embedded in an N-doped graphene. This study demonstrates that a heterogeneous catalyst with a mononuclear Mn site

can also be highly active in the WOR when the manganese site is in an appropriate coordination environment.

Results

Synthesis of mononuclear manganese catalyst. Mononuclear manganese embedded in nitrogen-doped graphene (Mn-NG) was prepared by carefully calcinating a mixture precursor of manganese chloride and graphene oxide under ammonia atmosphere at different temperatures, followed by a leaching treatment with aqueous nitric acid to remove possible non-framework Mn-based nanoparticles. For comparison, a sample denoted Mn-G was also synthesized by calcinating the precursors under argon instead of ammonia atmosphere. A HCl leaching treatment was applied to Mn-G. All the chemical WORs were performed by using Mn-G and Mn-NG after acid treatment. Structural characterization of Mn-G and Mn-NG. Highly dispersed manganese in Mn-NG before acid treatment can be directly observed by atomic-resolution scanning transmission electron microscopy (STEM) performed in high-angle annular dark field (HAADF) mode (Fig. 1a). It can be seen that the Mn-NG before acid treatment is composed of a mixture of mononuclear Mn species and multinuclear Mn-based clusters. After acid treatment, only singlesite Mn ions are observed, and Mn-based clusters can no longer be found (Fig. 1b). In addition, Mn-based nanoparticles or clusters of Mn-G and Mn-NG samples are not observed in the transmission electron microscopy (TEM) (Supplementary Fig. 1) and high-resolution STEM (HR-STEM) images (Supplementary Figs. 2 and 3), and no obvious X-ray diffraction patterns (Supplementary Fig. 4) and Raman peaks (Supplementary Fig. 5) can be assigned to multinuclear Mn-related species. The above-mentioned characterization results clearly indicate that Mn ions in Mn-NG after acid treatment are of mononuclear form. In addition to the observation of

State Key Laboratory of Catalysis, Dalian Institute of Chemical Physics, Chinese Academy of Sciences, Dalian National Laboratory for Clean Energy, iChEM, Dalian, China. 2Department of Chemistry and Institute for Computational and Engineering Sciences, The University of Texas at Austin, Austin, TX, USA. 3Research and Development, Honeywell UOP, Des Plaines, IL, USA. 4Shanghai Institute of Applied Physics, Chinese Academy Sciences, Shanghai, China. 5Department of Chemical and Biological Engineering and Computation and Data-Enabled Science and Engineering Program, University at Buffalo, State University of New York, Buffalo, NY, USA. 6Department of Materials Science and Engineering, National University of Singapore, Singapore, Singapore. *e-mail: [email protected] 1

Nature Catalysis | www.nature.com/natcatal

Articles

Nature Catalysis

a

b

c Mn 2p

Intensity (a.u.)

Mn-NG

5 nm

5 nm

2p3/2 Mn-G

660

2p1/2

655

650

Sat.

645

640

635

Binding energy (eV)

e

Normalized absorption (a.u)

1.6

f

7

Mn–N/O

6 1.2

5

0.8

Mn foil MnO Mn2O3 Mn-G Mn-NG Mn-NG before acid treatment

0.4 0.0 6,540

6,560

6,580

∣χ(R)∣ (Å–4)

d

Mn–Mn

4 3

Mn-G Fit Mn-NG Fit Mn-NG before acid treatment

2 1 0 0

2

E (eV)

4

6 R (Å)

8

10

Mn

C

N

Fig. 1 | Structural characterizations of Mn-G and Mn-NG. a, HAADF-STEM image of Mn-NG before acid treatment. b, HAADF-STEM image of Mn-NG, showing single Mn ions (the bright dots in the circles) dispersed in the carbon matrix. c, High-resolution XPS Mn 2p of Mn-NG and Mn-NG. Sat, satellite peak. d, Normalized XANES spectra at Mn K-edge of Mn foil, MnO, Mn2O3, Mn-G, Mn-NG and Mn-NG before acid treatment. E, energy. e, FT-EXAFS spectra of Mn-G, Mn-NG and Mn-NG before acid treatment at the Mn K-edge, showing the surrounding atoms adjacent to Mn ions. f, Atomic model of MnN4-G cluster obtained from DFT simulation.

mononuclear manganese dispersion in the local HAADF image, the homogeneous dispersion of Mn, N and O elements in Mn-G and Mn-NG were also confirmed by elemental mapping in the much larger scale view (Supplementary Figs. 2 and 3). In the atomic-resolution STEM energy-dispersive spectroscopy spectra and elemental mapping images (Supplementary Figs. 6–8), Mn and N elements were intriguingly found to be always adjacently distributed, demonstrating the possible bonding between the two types of atoms. On the basis of inductively coupled plasma-atomic emission spectroscopy (ICP-AES) analysis and Brunauer–Emmett–Teller (BET) measurement, the Mn-NG and Mn-G samples after acid treatment were estimated to contain 0.026 wt% and 0.1 wt% Mn content and have specific surface areas of ~350 m2 g−1 and ~370 m2 g−1, respectively. The oxidation state of Mn ions in both Mn-NG and Mn-G were determined to be Mn2+ according to the binding energy of Mn 2p3/2 and Mn 2p1/2 centred at ca. 641.4 and 653.0 eV, respectively, as shown in the X-ray photoemission spectroscopy (XPS) spectra in Fig. 1c, Supplementary Figs. 9 and 10, and ref. 29. Based on the highresolution analysis of N 1s XPS spectrum (Supplementary Fig. 9), the N 1s peak of the Mn-NG sample can be fitted into four peaks located at 398.2, 400.0, 401.3 and 405.4 eV, which can be ascribed to pyridinic, pyrrolic, graphitic and N-oxide nitrogen, respectively30. The surface N content in Mn-NG was also determined by XPS to be ca. 7.4 at% (Supplementary Table 1). As expected, no obvious XPS peak assigned to N species can be observed for the Mn-G sample as no nitrogen source was introduced in the calcination or acid washing procedure (Supplementary Fig. 10). The oxidation state of Mn ions was further characterized by the X-ray absorption near-edge structure (XANES) as shown in Fig. 1d. The Mn K-edge of XANES in the Mn-G sample exhibits a very similar near-edge structure to MnO, demonstrating the same valence (Mn2+) in agreement with the above XPS analysis.

In comparison, the Mn K-edge for Mn-NG before acid treatment is shifted to higher energy compared with that of the Mn-G sample and shows more resemblance to the Mn2O3 reference, which may be due to the oxidized MnN clusters, for example Mn3N7O3 of higher Mn valency (Supplementary Figs. 11 and 12). After acid treatment, the K-edge for Mn-NG shifts to lower X-ray energy and shows more resemblance to the MnO reference, indicating that the valence state of Mn is 2+​and the oxidized clusters have been removed through acid treatment as indicated by the smoothing of the XANES spectrum. The valence state of Mn ions in Mn-NG after acid treatment was further characterized by electron paramagnetic resonance (Supplementary Fig. 13). The six-line resonance arises from Mn2+ (spin configuration S =​ 5/2) and is characteristic of hyperfine interactions (nuclear spin I =​  5/2)31–34. The local structure of Mn sites in Mn-G, Mn-NG and Mn-NG before acid treatment was characterized by extended X-ray fine structure (EXAFS). The Fourier-transformed (FT) EXAFS spectra for Mn-G, Mn-NG and Mn-NG before acid treatment are shown in Fig. 1e and the extracted structural parameters are listed in Supplementary Table 2. The average Mn atomic environment in Mn-NG before acid treatment consists of O/N in the first and Mn in the second coordination shells with average coordination numbers of 3.1 ±​ 0.3 and 1.9 ±​ 0.3, respectively, consistent with the HAADF-STEM observation of a distribution of Mn clusters including mononuclear Mn site and ultrasmall Mn clusters. To disentangle the atomic structures of Mn-containing species, we simulated EXAFS spectra of various Mn–N clusters using the first-principles molecular dynamics method to fit the experimental EXAFS data. It is found that MnN4-G mixed with Mn3N7O3-G clusters with a ratio of 3:7 could well reproduce the experimental EXAFS spectrum (Supplementary Fig. 12). The Mn–Mn coordination number of the theoretically predicted structure is 1.4, which is only slightly smaller Nature Catalysis | www.nature.com/natcatal

Articles

Nature Catalysis a

b

250

250 200

4,000

TOF (s–1)

200

Mn-NG TOF = 214 s–1 TOF (s–1)

O2 evolution (mol mol–1 Mn)

6,000

150

0 300 400 500 600 700 800 900 Calcination temperature (°C)

50 Mn-G 0

30

60

90

0

120

0.0

0.5

Time (s)

d

20

0.44 Mn-

0.36

Mn-NG

0.32

ec

Vd

39 m

G1

0.40

55 mV

–1

dec

0.28 0.3

0.6

0.9

1.2

log[J (mA cm–2)] 10

Mn-NG Mn-G

2.0

2.5

14 12 10 8 6 4 2

G

0

1.5

–1

Current density (mA cm–2)

30

0.48

1.0

Initial Mn content (wt %)

40 Overpotential (V)

Current density (mA cm–2)

100 50

100

2,000

0

c

150

0 1.4

1.5

1.6

1.7

Potential (V versus RHE)

0

20

40

60

80

100

120

Time (h)

Fig. 2 | Chemical and electrochemical water oxidation properties of Mn-G and Mn-NG. a, Time profile of O2 evolution containing 0.05 mg of catalyst in 3 ml Ce(NH4)2(NO3)6 (0.15 M) solution. b, Effect of initial Mn content and calcination temperature (inset) on the catalytic behaviour of Mn-NG after acid treatment. The error bars (±​11.0%) arise from the measurement of Mn content. c, LSV curves of graphene, Mn-G and Mn-NG in 1.0 M KOH at a scan rate of 5 mV s−1 without iR compensation. The inset shows the Tafel plots. d, Plot of current density versus time for Mn-NG at a constant anode voltage of 1.57 V versus RHE in 1.0 M KOH.

than the 1.6 (lower limit of the measured value of 1.9 ±​ 0.3). In comparison, after acid treatment, no strong Mn–Mn signal in the FT spectrum (Fig. 1d) was observed for Mn-G or Mn-NG, suggesting that the Mn clusters have been removed by the acid treatment leaving only mononuclear Mn sites. Detailed EXAFS analysis of the measured spectrum from Mn-NG is consistent with an average Mn coordination environment consisting of three shells of light atoms O, N and C (Supplementary Table 2 and Supplementary Figs. 14–16). The measured EXAFS spectrum from Mn-NG was interrogated for a small but measurable Mn signal consistent with small Mn clusters, but none were shown to improve the quality of the model. The lack of a Mn neighbouring signal in the Mn-NG spectrum provides strong evidence of isolated Mn atoms within the graphene structure. Based on the EXAFS results, the average Mn environment within Mn-NG is consistent with one O atom at a distance of 1.98 ±​ 0.02 Å, four N atoms at a distance of 2.19 ±​ 0.01 Å and two C atoms at a distance of 2.52 ±​ 0.02 Å (Supplementary Fig. 16). The O signal is likely from O2 adsorbed to Mn centre out of plane with the graphene sheet structure35,36. The signal from the four N atoms with a distance of 2.19 Å is evidence for the MnN4-G configuration shown in Fig. 1f. The C signal with a distance of 2.52 Å is also predicted by this MnN4-G structure but with about half the expected coordination number, which could be due to defects. Overall the average atomic environment of Mn in Mn-NG is coordinated by three shells of light atoms (O/N/C) consistent with isolated Mn atoms within the graphene sheet similar to the MnN4-G structure shown in Fig. 1f (O2 not shown). EXAFS results are for the average local atomic environment of Mn so it is not possible to determine Nature Catalysis | www.nature.com/natcatal

the precise distribution that makes up this average. This interpretation is consistent with the chemistry and other characterization and in particular density functional theory (DFT) calculations for the most active site. The Mn environment in the Mn-G sample prepared without N (Supplementary Table 2) is similar to Mn-NG but with O instead of N in the coordination shell at 2.19 Å, ultimately leading to lower oxygen evolution reaction (OER) activity as discussed below. Catalytic activity in water oxidation. Chemical water oxidation of Mn-G and Mn-NG was evaluated in Ceiv-containing acidic solution as commonly done in the literature24–27. Graphene or nitrogendoped graphene showed hardly any WOR (Supplementary Fig. 17 and Supplementary Table 3). Surprisingly, the Mn-NG sample exhibited extremely high water oxidation activity with a TOF value up to 214 s−1 (Fig. 2a and Supplementary Fig. 18), which is seven orders of magnitude higher than those for MnO, Mn2O3 or MnO2 (Supplementary Fig. 19) and is two orders of magnitude higher than the most efficient Mn-based catalysts reported to date (Supplementary Table 4). In contrast, the Mn-G sample (free of nitrogen incorporation) did not show obvious O2 evolution (Fig. 2a). These results clearly show that the mononuclear Mn coordinated with N ligands is responsible for the high activity in water oxidation. The fact that leaching of Mn ions from the Mn-NG sample into the Ceiv solution leads to deactivation further affirms that the mononuclear Mn is the active site for water oxidation (Supplementary Fig. 20). Figure 2b shows that the chemical water oxidation performance of Mn-NG varies with the initial Mn content in the preparation and with the calcination temperature (inset in Fig. 2b) and

Articles 6 +

–

O2(g) + * + 4(H + e )

0

c

Upds = 0.45 V

O*

OOH*

OH*

Upds = 0.55 V

O*

OOH*

*

OH*

O*

OOH*

2

2H2O(l) + *

0

Reaction coordinate

MnN3-G

d

OH*

*

Reaction coordinate

6

MnO3-G

4

O*

2

∆G (eV)

4

∆G (eV)

* Mn3N7-G

H2O(l) + OH* + (H+ + e–)

6

0

6

4

H2O(l) + O* + 2(H+ + e–) 2

b

OH*

*

OOH* + 3(H+ + e–)

∆G (eV)

4

MnN4-G

∆G (eV)

a

Nature Catalysis

OOH*

2

Upds = 1.2 V

0

Upds = 0.73 V

Reaction coordinate

Reaction coordinate Mn

C

N

O

H

Fig. 3 | Evaluation of catalytic activity by DFT simulations. a–d, Free energy diagram for water oxidation on MnN4-G (a), Mn3N7-G (b), MnN3-G (c) and MnO3-G (d). The dashed lines indicate the step-free energy for an ideal electrocatalyst. The surface structures with various reaction intermediates are shown alongside the free energy diagram. Upds, equilibrium potential for the potential determining step.

time (Supplementary Fig. 21). The low loading of Mn and high specific surface area of graphene favour high dispersion of manganese, while high temperature is responsible for the formation of Mn–N bonds. The high activity of Mn-NG for water oxidation was further confirmed by chemical water oxidation using another oxidant, sodium periodate (standard electrode potential Eθ(H5IO6/ IO3−) =​ 1.6 V), where Mn-NG showed a TOF of 0.9 s−1, while Mn-G and Mn2O3 MnO2 were all inactive (Supplementary Fig. 22). To further confirm the high activity of the Mn-NG catalyst for water oxidation, the electrochemical water oxidation of Mn-G and Mn-NG was performed in 1.0 M KOH alkaline electrolyte. Figure 2c shows the linear sweep voltammetry (LSV) curves of the Mn-NG as the electrocatalyst. The onset overpotential of Mn-NG is negatively shifted by ca. 90 mV compared with that of Mn-G, and the overpotential (η) at a current density (J) of 10 mA cm−2 is 337 and 459 mV for Mn-NG and Mn-G, respectively. The great down shift in overpotential from Mn-G to Mn-NG strongly suggests that the nitrogen coordination plays a crucial role in the formation of the highly active mononuclear Mn for water oxidation. The measured Tafel slope of Mn-NG (55 mV dec−1) is far smaller than that of Mn-G (139 mV dec−1) (inset in Fig. 2c). Combining the characterization results and electrochemical measurements, we believe that mononuclear Mn coordinated to N atoms is the catalytic active site for the WOR. Figure 2d shows the electrochemical stability of the Mn-NG catalyst in the alkaline electrolyte and the current density initially obtained at 1.57 V versus reversible hydrogen electrode (RHE) does not show an obvious decrease at least in 120 h. The Faradaic efficiency of Mn-NG was determined to be ~100% (Supplementary Fig. 23), confirming that the anode current is from the WOR. The Mn still remains atomically dispersed in Mn-NG after the 120 h test of continuous operation (Supplementary Fig. 24). The stability of Mn-NG in electrochemical water oxidation is due to the high barrier of oxidation of the pyridyl moiety (160.7 kJ mol−1), which

coordinates to the mononuclear Mn site and stabilizes the structure during the WOR37. The electrochemically active surface area of Mn-NG was measured to be 227.5 cm−2, which is greater than that of Mn-G (only 14.4 cm−2) (Supplementary Fig. 25). DFT calculations and water oxidation mechanism. To gain insights into the water oxidation mechanism of Mn-NG, the OER activities of MnN4-G and various proposed MnN-G and MnO-G sites were examined by DFT calculations using the computational hydrogen electrode model35,38–42. As seen in Fig. 3, the theoretical overpotential (450 mV) of MnN4-G is much lower than that of a trinuclear Mn3N7-G (550 mV). The difference in overpotential between MnN4-G and Mn3N7-G is assigned to the over-destabilization of OOH* intermediate on Mn3N7-G. As a result, the overpotential-determining step for Mn3N7-G is the O–O coupling step instead of the oxidation of OH* to O* for MnN4-G. We also investigated other models of active sites, such as MnN3-G (Fig. 3c) and MnO3-G (Fig. 3d), to further understand the excellent performance of MnN4-G. MnN3-G demonstrated that the change in the coordination environment (from N3 to N4) of the Mn site leads to dramatic decrease in overpotential due to MnN3-G binding the OH* intermediate too strongly. MnO3-G was considered as a possible active site in the Mn-G sample, and the overpotential for MnO3-G was calculated to be as high as 1,200 mV due to the excessively strong adsorption of OH*. In addition, we also compared the activity of MnN4-G to graphic-N, pyridinic-N, MnN3O-G, Mn phthalocyanine (MnPc), Mn4O4 cubane and Mn3O4 (001) surface (Supplementary Fig. 26 and Supplementary Note 1). All of these active site models exhibited higher overpotential than MnN4-G. These comparisons give strong support to the vital responsibility and necessity of the mononuclear Mn site coupled to four N coordination environments for the superior activity of MnN4-G. To understand how the four-electron transfer mechanism takes place during the O2 evolution from water oxidation on the mononuclear Mn of the Mn-NG catalyst, cyclic voltammogram (CV) Nature Catalysis | www.nature.com/natcatal

Articles

Nature Catalysis a

b

MnII/MnIII III

Mn /Mn

0.2

IV

H+ –e–

Current (mA)

–e

H+

– II

Mn

0.0

–0.2

MnIII

Mn C N O H

MnIII –e–

–0.4

H+ 0.4

0.6

0.8

1.0

1.2

1.4

–

–e

1.6

MnIV

H+

Potential (V versus RHE)

Fig. 4 | Proposed water oxidation mechanism. a, CVs of Mn-NG in Ar-saturated 1.0 M KOH at a scan rate of 50 mV s−1. b, Mechanism of O2 evolution from water catalysed by MnN4-G.

and further DFT simulations were performed. Figure 4a shows the CV of Mn-NG, in which two reversible oxidation waves at reversible half-wave potential E1/2 =​ 0.74 and 1.36 V can be assigned to two redox reactions involving Mniii/Mnii and Mniv/Mniii redox couples, respectively (also Supplementary Figs. 27 and 28). These redox couples can be further evidenced by the differential pulse voltammogram of Mn-NG where two reduction peaks are clearly visible in the range of 1.4 to 0.6 V and assigned to the reductions of Mniv →​  Mniii →​  Mnii, respectively (Supplementary Fig. 27c). For Mn-G, these processes occurred at relatively higher potentials than for Mn-NG (Supplementary Fig. 29). On the basis of the CV experiment and the DFT simulations on the oxidation states of the mononuclear Mn site at each OER step, a four-step reaction mechanism for water oxidation on the MnN4-G active site could be deduced (Fig. 4b and Supplementary Table 5). The first H2O molecule is dissociated on the Mnii site to form a Mniii–OH intermediate while releasing one proton and one electron. This step is followed by the oxidation of Mniii–OH to form a Mniv=​O site. The second water molecule then acts as a nucleophile and attacks the Mniv=​O to form a Mniii–OOH intermediate after coupling of the oxygen atoms. Finally, the oxidation of OOH and liberation of O2 from Mniii–OOH with low activation energy complete the catalytic cycle. Here the highest oxidation state (Mniv) of the Mn ion occurs on the formation of the O* intermediate. This finding is in contrast with the possible but more difficult formation of Mnv required for O* as found for other Mn-based water oxidation catalysts such as Mn3N7-G, Mn4O4 cubane and Mn3O4 (001)43–45. It should be noted that the WOR mechanism on such a mononuclear Mn site is different from that on multinuclear complexes, where the rate-determining step has been proposed to be the oxo–oxo coupling step for the O2 release22,46, rather than the electrochemical OH* oxidation step as demonstrated for the mononuclear manganese active site (MnN4-G) in Mn-NG catalyst.

Conclusions

We synthesized Mn-NG catalyst with a mononuclear manganese site embedded in nitrogen-doped graphene using a convenient calcination and leaching strategy, and found that the mononuclear manganese catalyst exhibits superior water oxidation activity with a TOF of over 200 s−1, which is close to that of nature’s catalyst CaMn4O5 in photosystem II and two orders of magnitude higher than those of reported Mn-based water oxidation catalysts to date. The mononuclear manganese site is stabilized via the coordination with four N atoms of the NG. The MnN4-G site is much more active than other Mn-based water oxidation catalysts, mainly owing to the Nature Catalysis | www.nature.com/natcatal

four N coordination to Mn in favour of the formation of Mniv–oxo species and the further nucleophilic attack of the second water molecule to Mniv–oxo to evolve O2 with low activation energy via the Mniii–OOH intermediate.

Methods

Synthesis of Mn-G and Mn-NG. In a typical synthesis, a precursor solution containing graphene oxide and MnCl2·4H2O (Mn/GO =​ 0.3 wt%) was sonicated for 1 h. Then, the water was removed by a rotary evaporation apparatus. The residual solid was calcined under a NH3 atmosphere for 2 h to obtain Mn-NG. The obtained catalysts were subjected to leaching using aqueous HNO3 (10 mol l−1), and then dried at 80 °C for 12 h. The Mn loading before and after HNO3 acid treatment was 0.5 wt% and 0.026 wt%, respectively, as analysed by ICP-AES. Besides Mn element, we did not detect any other transition metal element by ICP-AES. To distinguish, Mn-NG before and after acid treatment is denoted as Mn-NG before acid treatment and Mn-NG (or Mn-NG after acid treatment), respectively. All the chemical WORs were performed by using Mn-NG, not Mn-NG before acid treatment. Mn-G before acid treatment was prepared by the same procedure but changing the calcination atmosphere to argon. The obtained catalyst was subjected to leaching using aqueous HCl (10 mol l−1), and then dried at 80 °C for 12 h. The Mn loading in Mn-G before and after HCl acid treatment was 0.5 wt% and 0.1 wt%, respectively, as analysed by ICP-AES. Besides Mn element, we did not detect any other transition metal element by ICP-AES. To distinguish, Mn-G before and after acid treatment is denoted as Mn-G before acid treatment and Mn-G, respectively. Material characterizations. The as-prepared samples were characterized by X-ray powder diffraction (XRD) on a Rigaku D/Max-2500/PC powder diffractometer. The sample powder was scanned using Cu-Kα​radiation with an operating voltage of 40 kV and current of 200 mA. A scan rate of 5° min−1 was applied to record the patterns in the range of 10–80°. Raman spectra were recorded on a Raman spectrometer (Renishaw) using a 532 nm laser source. TEM images were observed by a Hitachi HT7700. High-resolution STEM images were recorded on a JEM2100 transmission electron microscope (Japan) at 200 kV. Scanning electron microscopy images were recorded on a LEO 1530 VP field emission scanning electron microscope (Germany) operated at 5 kV. HAADF imaging was performed with an aberration-corrected JEM-ARM 200F microscope. The loading amount of manganese in the catalysts was determined using ICP-AES on a Shimadzu ICPS-8100. Before ICP-AES measurement, 50 mg acid-leached catalyst was placed in a 50 ml beaker and calcined in an oven at 520 °C for 6 h to completely remove the carbon. The residue was dissolved in aqua regia and diluted with water to test the manganese content. The loading amount of manganese in Mn-NG was measured ten times by this method, which varied from 0.023 to 0.029 wt%. The average value was 0.026 wt%. The valence state of manganese was determined using XPS recorded on a Thermo ESCALAB 250Xi. The X-ray source selected was monochromatized Al Kα source (15 kV, 10.8 mA). Region scans were collected using a 20 eV pass energy. Peak positions were calibrated relative to C 1s peak position at 284.6 eV. X-ray absorption data collection, analysis and modelling. The XAFS spectra of Mn-NG before acid treatment at the Mn K-edge (photoelectron energy origin E0 =​ 6,539 eV) were performed at the BL14W1 beamline of

Articles Shanghai Synchrotron Radiation Facility operated at 3.5 GeV under ‘topup’ mode with a constant current of 240 mA. The XAFS data were recorded under fluorescence mode with a seven-element Ge solid-state detector. The energy was calibrated accordingly to the absorption edge of Mn foil. Athena and Artemis codes were used to extract the data and fit the profiles. For the XANES part, the experimental absorption coefficients as function of energies μ(E) were processed by background subtraction and normalization procedures, and reported as ‘normalized absorption’. For the EXAFS part, the FT data in r-space were analysed by applying first-shell approximate model for Mn–O and Mn–Mn contributions. The passive electron factor, S02, was determined by fitting the experimental data on Mn foil and fixing the coordination number of Mn–Mn for further analysis of the measured samples. The parameters describing the electronic properties (for example, correction to E0) and local structure environment including coordination number, bond distance (R) and Debye–Waller factor around the absorbing atoms were allowed to vary during the fit process. The fitted range for k-space was selected to be k =​  3−​10  Å−1 (k3-weighted). The XAFS spectra of Mn-G and Mn-NG after acid treatment at the Mn K-edge (E0 =​ 6,539 eV) were performed at the MRCAT47 beamline of the Advanced Photon Source at Argonne National Laboratory operated at 7 GeV. The XAFS data were recorded in fluorescence mode with a Stern–Heald detector48. The energy was calibrated accordingly to the absorption edge of Mn foil. Athena and Artemis49 codes were used to extract the EXAFS χ(k) data (Supplementary Fig. 14) data and model the spectra. For the XANES part, the experimental absorption coefficients as function of X-ray energy μ(E) were processed by standard normalization procedures, and reported as ‘normalized absorption’. The EXAFS spectra were modelled in r-space. The passive electron factor, S02 =​  1.05  ±​ 0.08 was determined by fitting the experimental data on MnO with fixed coordination number for Mn–O and Mn–Mn based on the crystal structure. The parameters describing the local structure environment of Mn in the Mn-NG spectrum including coordination number and bond distance (R) were determined during the optimization process. The disorder in the distances, σ2, was constrained using a Debye model with characteristic Debye temperature determined to be 489 ±​ 80 K. The k range and R range were selected to be k =​ 3–10.5 Å−1 and R =​ 1–3 Å with simultaneous optimization using k-weighting of k1, k2 and k3. The data and modelled ranges result in thirteen independent points. The model contains seven parameters; three coordination numbers and distances along with one Debye temperature. The EXAFS spectrum is shown in Supplementary Fig. 14. The magnitude of the FT-EXAFS spectra and model for Mn-NG and Mn-G are shown in Supplementary Fig. 15. As shown, the model reproduces the data well. The percent miss fit between the data and the model (r-factor) is 0.06% and the reduced chi-squared ( χγ2) is 131. The model consists of three shells of light atoms such as O, N and C. The contribution from each signal to the model is shown in Supplementary Fig. 16. Based on the scattering amplitude of the EXAFS signal alone, neighbouring atoms on the periodic table cannot be distinguished50. We have assigned the atom types based on the chemistry of our system, previously published EXAFS results from similar systems35,36 and comparison with DFT models. The measured Mn EXAFS spectra from Mn-NG were tested for many different combinations of possible neighbouring atoms. In particular, the split Mn–O and Mn–N shell was tested by using only for one Mn–O/N shell in the range of 2.0 to 2.2 Å instead of two. The model with only one Mn–O/N shell resulted in an increase in the reduced chisquared by a factor of 7. The large increase in reduced chi-squared (a factor of 2 is significant) indicates that the split Mn–O shell is required. The spectra were also tested for a small Mn signal that would indicate the presence of Mn clusters. The EXAFS-determined coordination numbers were consistent with zero indicating that the model was not improved by including this signal. MRCAT operations are supported by the Department of Energy and the MRCAT member institutions. This research used resources of the Advanced Photon Source, a US Department of Energy (DOE) Office of Science User Facility operated for the DOE Office of Science by Argonne National Laboratory under Contract No. DE-AC02-06CH11357. Computational methods. We employed DFT to determine the structure and the catalytic properties of the synthesized Mn-NG catalysts. Specifically, we used the Vienna ab-initio simulation package (VASP) code51,52 with electron correlation treated within the generalized gradient approximation using the Perdew–Burke– Ernzerhof (PBE) exchange-correlation functional53. The one-electron Kohn–Sham states were expanded by plane-wave basis with a kinetic cut-off energy of 400 eV, and the projector augmented wave (PAW) method54 was used to treat the effect of the inner cores on the valence states. Suitable Monkhorst–Pack meshes55 have been used to sample the reciprocal space for different models. The structural models are considered fully relaxed until forces on the relaxed atoms were smaller than 0.05 eV Å−1. All calculations were spin polarized. We investigated the theoretical overpotential for the OER by DFT calculations using the computational hydrogen electrode model38. The method has previously been successful in predicting trends in electrochemical activity on oxide surfaces39–41. Briefly, we used DFT calculations to evaluate the binding energies of OH*, O* and OOH* on the catalysts under consideration. Based on the calculated binding energies, the Gibbs free energy changes of electrochemical

Nature Catalysis elementary steps are calculated along the OER reaction coordinate of a four-step reaction mechanism: H 2O+* → OH*+H + + e− (1) OH* → O* + H + + e−

(2)

H 2O + O* → OOH*+H + + e−

(3)

OOH* → *+O2+H + + e−

(4)

We assume standard conditions in calculating the Gibbs free energy changes. With this approach, the theoretical overpotential (ηOER) at standard conditions is defined as: ηOER = (G OER / e) − 1.23 V

(5)

where GOER is the potential determining step defined as the highest free energy step in the process of OER and e is unit charge. We consider several Mn-containing systems including: Mn4O4 cubane, Mn3O4 (001) surface, MnO3 cluster embedded onto graphene (MnO3-G), Mn phthalocyanine (MnPc), and two Mn-NG structures determining from extensive structural characterization (MnN4 and Mn3N7O3). Furthermore, the activity of two N-doped graphene active sites without Mn ions, namely graphitic-N and pyridinic-N, were also examined. The employed Mn4O4 cubane model is similar to the Co4O4 cubane model in previous studies56–58. This model system is a representative system for Mn oxide with small nuclearity. Due to the molecular nature of the Mn cubane, a single gamma point is used. We modelled the Mn3O4(001) surface with slabs consisting of eight Mn layers in a (1 ×​  1) supercell. The k-point sampling consists of 4 ×​  4  ×​ 1 Monkhorst–Pack points. During structure relaxation, we allow relaxation of both the adsorbates involved in OER (O*, OH* and OOH*) and the top four Mn layers for Mn3O4 (001) slab model. The rest of the Mn layers of the Mn oxides were kept frozen at their bulk positions. The two Mn-NG structures are MnN4 and Mn3N7 clusters embedded in the graphene sheet. The k-point sampling consists of 2 ×​  2  ×​ 1 Monkhorst–Pack points. The solvation effects were also taken into account using an implicit solvation model implemented in VASP59. The relative permittivity for the continuum solvent was set to 80 to simulate an aqueous electrolyte. DFT-PBE in general has the tendency to delocalize unpaired electrons due to the self-interaction error, which is problematic to treat the electronic structure of Mn ions exhibiting magnetic properties that require electron localization. A way to overcome the problem is to apply the DFT +​  U method60, which employs a Hubbard model to penalize the double occupancy of electronic states. The penalty is parameterized by the U value. Localized states are more preferable with larger U value. In this study, a U value was applied to 3d orbitals of Mn ions to correct the on-site Coulomb interactions. We parameterized the U value by evaluating the performance of DFT +​  U method in reproducing the catalytic properties of the Mn4O4 cubane cluster generated by the Heyd–Scuseria–Ernzerhof (HSE) hybrid functional61. A hybrid functional provides a more universal way of correcting the self-interaction error and is considered more accurate than the empirical DFT +​  U method. The flavour of the HSE hybrid functional is the HSE0662, which is found to yield good results on thermochemistry, bandgaps and lattice constants of solids. Supplementary Fig. 11 compares the free energies of the OER elementary steps calculated by DFT +​  U method with various U values and HSE06. It could be seen that with plain PBE, the binding energies of O and OH intermediates are overestimated compared with the HSE06 calculations. With increasing U value, the discrepancy diminishes. With U =​ 5 eV, it gives the closest result to that of HSE06. Hence, we employed U =​ 5 eV in this work to study the catalytic properties of the interested materials. It should be noted that in doing these calculations, we always assume a high spin state for the magnetism of Mn ions as the state is favoured according to our tests. When two and more Mn ions are involved in the structure, we assume the high spin states of Mn ions are ferromagnetically coupled. The assumption could cause uncertainty in the calculations since antiferromagetic arrangement could be the most favourable state, but the energy difference between anti- and ferromagnetic states is expected to be as small as 0.1 eV (ref. 63). Similar treatments to the magnetism of Mn ions have already been successfully employed for study OER on binuclear Mn model system45,64. To simulate the EXAFS spectra for Mn-NG structures, for each proposed structure, a 10 ps molecular dynamics (MD) simulation was performed at 300 K to an ensemble of equilibrium structures at the temperature. The time integration was carried out under a canonical ensemble of constant number of atoms, constant volume, and temperature using the Nosé–Hoover algorithm. A time step of 1 fs was employed. The MD simulations were also performed with VASP and PBE functional as outlined above. Theoretical EXAFS signals were simulated using an approach similar to that reported previously65,66. The Mn K-edge EXAFS spectra were calculated from the MD trajectories by averaging the signal arising from each Mn ion in the interested structure. Each MD simulation was allowed to thermalize for at least 5 ps and the per-configuration EXAFS spectra were Nature Catalysis | www.nature.com/natcatal

Nature Catalysis calculated from snapshots of the trajectory at intervals of 50 fs for 5 ps giving 100 independent configurations in the canonical average. The multiple-scattering calculations were performed using FEFF6-lite67. All atoms up to 6.0 Å away from each photoabsorbing atom were included in the scattering calculations. Once the theoretical EXAFS signal was produced, the corrections to the energy origin (Δ​E0 =​ 7.1 eV) and to the amplitude reduction factor (S02 =​ 0.7) that were found for the experimental EXAFS by fitting to the Mn foil sample were applied to the theoretical data to align the experimental and theoretical data in k-space. The Fourier transformation of k3-weighted EXAFS spectrum in k-space was done with a k-range of 2.8 to 10.5 Å−1. The Hann window function was used with dk equal to 1 Å−1. Based on the experimental EXAFS spectrum, we attempted to reconstruct the atomic structures of the Mn-NG before acid treatment by using the first-principles DFT-based calculations. The method has been successfully applied in characterizing structures of metal nanoparticles65,66 and cobalt oxide nanoclusters68,69. To search for the atomic structures of Mn-NG before acid treatment, we constructed different Mn–N/O clusters as guided by the structural inferences from the experimental EXAFS analysis. The coordination number of Mn–N/O and Mn–Mn obtained from the EXAFS fitting are 3.1 ±​ 0.3 and 1.9 ±​ 0.3, respectively. Accordingly, we systematically examined monometric, dimeric and trimeric Mn clusters with each Mn ion coordinating with three or four N atoms. Hydroxyl and atomic oxygen adsorption on Mn ions were also considered. Dimeric Mn cluster have very short Mn–Mn distances, so we excluded them from consideration for EXAFS simulation. The EXAFS spectra for monometric and trimeric Mn clusters were simulated. No simulated EXAFS for the proposed monometric and trimeric Mn–O/N clusters matched the experimental data well. Considering that the Mn-NG before acid treatment may be heterogeneous and consisted of different Mn–O/N clusters, we then mixed EXAFS spectra of two atomic structures with various ratios in k-space. All possible combinations were considered and we found that the simulated EXAFS of 30% MnN4 plus 70% Mn3N7O3 is consistent with the experimental EXAFS as shown in Supplementary Fig. 12. Tests of chemical water oxidation. A solution of 3 ml 0.15 M Ce(NH4)2(NO3)6 or 0.1 M NaIO4 was kept at 25 °C for 5 min with stirring, and then 0.1 ml solution containing different contents (0.01–0.3 g l−1) of catalyst was injected into the Ceiv or NaIO4 solution. The generated oxygen was monitored by a Clark-type electrode (Strathkelvin SI130 UK)70,71. The TOF was calculated by using the initial constant O2 evolution rate and assuming all the manganese ions as active sites. Electrochemical measurements. All the electrochemical tests were performed in a conventional three-electrode system at an electrochemical station (CHI 440C), using saturated calomel (saturated KCl solution) electrode (SCE) as the reference electrode, Pt plate as the counter electrode, and Mn-NG or Mn-G on graphite plate electrode as the working electrode. LSV with scan rate of 5 mV s−1 was conducted in 1 M KOH once the system achieved equilibrium. Timedependent current under a constant potential measurement was performed to evaluate the long-term stability of the system. All potentials measured were calibrated to RHE using the following equation: E (versus RHE) =​  E (versus SCE) +​ 0.241 V +​ 0.0591 pH. Faradaic efficiency measurements. To measure the Faradaic efficiency of the Mn-NG catalyst, O2 production was performed in a closed pyrex glass reactor at a constant anodic current density of 10 mA cm−2. Continuous gas flow inside the whole reaction line was maintained by using a constant Ar airflow. Quantitative analysis of produced O2 was online measured by gas chromatography (Agilent 7890 A) using a thermal conductivity detector.

Data availability

The data that support the findings of this study are available from the corresponding author upon reasonable request.

Received: 23 January 2018; Accepted: 5 September 2018; Published: xx xx xxxx

References

1. Bard, A. J. & Fox, M. A. Artificial photosynthesis—solar splitting of water to hydrogen and oxygen. Acc. Chem. Res. 28, 141–145 (1995). 2. Yano, J. & Yachandra, V. Mn4Ca cluster in photosynthesis: where and how water is oxidized to dioxygen. Chem. Rev. 114, 4175–4205 (2014). 3. Ogata, K., Yuki, T., Hatakeyama, M., Uchida, W. & Nakamura, S. All-atom molecular dynamics simulation of photosystem II embedded in thylakoid membrane. J. Am. Chem. Soc. 135, 15670–15673 (2013). 4. Yang, J., Wang, D., Han, H. & Li, C. Roles of cocatalysts in photocatalysis and photoelectrocatalysis. Acc. Chem. Res. 46, 1900–1909 (2013). 5. Spoeri, C., Kwan, J. T. H., Bonakdarpour, A., Wilkinson, D. P. & Strasser, P. The stability challenges of oxygen evolving catalysts: towards a common fundamental understanding and mitigation of catalyst degradation. Angew. Chem. Int. Ed. 56, 5994–6021 (2017). Nature Catalysis | www.nature.com/natcatal

Articles 6. Dismukes, G. C. et al. Development of bioinspired Mn4O4-cubane water oxidation catalysts: lessons from photosynthesis. Acc. Chem. Res. 42, 1935–1943 (2009). 7. Garrido-Barros, P., Gimbert-Surinach, C., Matheu, R., Sala, X. & Llobet, A. How to make an efficient and robust molecular catalyst for water oxidation. Chem. Soc. Rev. 46, 6088–6098 (2017). 8. Limburg, J. et al. A functional model for O–O bond formation by the O2-evolving complex in photosystem II. Science 283, 1524–1527 (1999). 9. Yagi, M. & Narita, K. Catalytic O2 evolution from water induced by adsorption of [(OH2)(terpy)Mn(μ​-O)2Mn(terpy)(OH2)]3+ complex onto clay compounds. J. Am. Chem. Soc. 126, 8084–8085 (2004). 10. Kanady, J. S., Tsui, E. Y., Day, M. W. & Agapie, T. A synthetic model of the Mn3Ca subsite of the oxygen-evolving complex in photosystem II. Science 333, 733–736 (2011). 11. Hocking, R. K. et al. Water-oxidation catalysis by manganese in a geochemical-like cycle. Nat. Chem. 3, 461–466 (2011). 12. Chen, H. Y., Faller, J. W., Crabtree, R. H. & Brudvig, G. W. Dimer-of-dimers model for the oxygen-evolving complex of photosystem II. Synthesis and properties of [Mniv4O5)(terpy)4(H2O)2](ClO4)6. J. Am. Chem. Soc. 126, 7345–7349 (2004). 13. Ruettinger, W. F., Campana, C. & Dismukes, G. C. Synthesis and characterization of Mn4O4L6 complexes with cubane-like core structure: a new class of models of the active site of the photosynthetic water oxidase. J. Am. Chem. Soc. 119, 6670–6671 (1997). 14. Zhang, C. et al. A synthetic Mn4Ca-cluster mimicking the oxygen-evolving center of photosynthesis. Science 348, 690–693 (2015). 15. Schwarz, B. et al. Visible-light-driven water oxidation by a molecular manganese vanadium oxide cluster. Angew. Chem. Int. Ed. 55, 6329–6333 (2016). 16. Maayan, G., Gluz, N. & Christou, G. A bioinspired soluble manganese cluster as a water oxidation electrocatalyst with low overpotential. Nat. Catal. 1, 48–54 (2018). 17. Xu, Y. et al. Chemical and light-driven oxidation of water catalyzed by an efficient dinuclear ruthenium complex. Angew. Chem. Int. Ed. 49, 8934–8937 (2010). 18. Gersten, S. W., Samuels, G. J. & Meyer, T. J. Catalytic-oxidation of water by an oxo-bridged ruthenium dimer. J. Am. Chem. Soc. 104, 4029–4030 (1982). 19. Duan, L., Wang, L., Li, F., Li, F. & Sun, L. Highly efficient bioinspired molecular Ru water oxidation catalysts with negatively charged backbone ligands. Acc. Chem. Res. 48, 2084–2096 (2015). 20. Yin, Q. et al. A fast soluble carbon-free molecular water oxidation catalyst based on abundant metals. Science 328, 342–345 (2010). 21. Okamura, M. et al. A pentanuclear iron catalyst designed for water oxidation. Nature 530, 465–468 (2016). 22. Shaffer, D. W., Xie, Y. & Concepcion, J. J. O–O bond formation in rutheniumcatalyzed water oxidation: single-site nucleophilic attack vs. O–O radical coupling. Chem. Soc. Rev. 46, 6170–6193 (2017). 23. Meyer, T. J., Sheridan, M. V. & Sherman, B. D. Mechanisms of molecular water oxidation in solution and on oxide surfaces. Chem. Soc. Rev. 46, 6148–6169 (2017). 24. Ma, L. et al. Cerium(iv)-driven water oxidation catalyzed by a manganese(v)nitrido complex. Angew. Chem. Int. Ed. 54, 5246–5249 (2015). 25. Ellis, W. C., McDaniel, N. D., Bernhard, S. & Collins, T. J. Fast water oxidation using iron. J. Am. Chem. Soc. 132, 10990–10991 (2010). 26. Duan, L. et al. A molecular ruthenium catalyst with water-oxidation activity comparable to that of photosystem II. Nat. Chem. 4, 418–423 (2012). 27. Hull, J. F. et al. Highly active and robust Cp* iridium complexes for catalytic water oxidation. J. Am. Chem. Soc. 131, 8730–8731 (2009). 28. Lee, W.-T., Munoz, S. B., Dickie, D. A. & Smith, J. M. Ligand modification transforms a catalase mimic into a water oxidation catalyst. Angew. Chem. Int. Ed. 53, 9856–9859 (2014). 29. Sun, Y., Hu, X., Luo, W., Xia, F. & Huang, Y. Reconstruction of conformal nanoscale MnO on graphene as a high-capacity and long-life anode material for lithium ion batteries. Adv. Funct. Mater. 23, 2436–2444 (2013). 30. Xue, Y. et al. Low temperature growth of highly nitrogen-doped single crystal graphene arrays by chemical vapor deposition. J. Am. Chem. Soc. 134, 11060–11063 (2012). 31. Schnegg, A. et al. Probing the fate of Mn complexes in Nafion: a combined multifrequency EPR and XAS study. J. Phys. Chem. C 120, 853–861 (2016). 32. Colmer, H. E., Howcroft, A. W. & Jackson, T. A. Formation, characterization, and O–O bond activation of a peroxomanganese(iii) complex supported by a cross-clamped cyclam ligand. Inorg. Chem. 55, 2055–2069 (2016). 33. Gallagher, A. T. et al. A structurally-characterized peroxomanganese(iv) porphyrin from reversible O2 binding within a metal-organic framework. Chem. Sci. 9, 1596–1603 (2018). 34. He, M., Li, X., Liu, Y. & Li, J. Axial Mn–C–CN bonds of cyano manganese(ii) porphyrin complexes: flexible and weak? Inorg. Chem. 55, 5871–5879 (2016).

Articles 35. Fei, H. et al. General synthesis and definitive structural identification of MN4C4 single-atom catalysts with tunable electrocatalytic activities. Nat. Catal. 1, 63–72 (2018). 36. Zitolo, A. et al. Identification of catalytic sites for oxygen reduction in iron-and nitrogen-doped graphene materials. Nat. Mater. 14, 937–942 (2015). 37. Li, Y.-Y., Ye, K., Siegbahn, P. E. M. & Liao, R.-Z. Mechanism of water oxidation catalyzed by a mononuclear manganese complex. ChemSusChem 10, 903–911 (2017). 38. Norskov, J. K. et al. Origin of the overpotential for oxygen reduction at a fuel-cell cathode. J. Phys. Chem. B 108, 17886–17892 (2004). 39. Man, I. C. et al. Universality in oxygen evolution electrocatalysis on oxide surfaces. ChemCatChem 3, 1159–1165 (2011). 40. Garcia-Mota, M. et al. Importance of correlation in determining electrocatalytic oxygen evolution activity on cobalt oxides. J. Phys. Chem. C 116, 21077–21082 (2012). 41. Su, H.-Y. et al. Identifying active surface phases for metal oxide electrocatalysts: a study of manganese oxide bi-functional catalysts for oxygen reduction and water oxidation catalysis. Phys. Chem. Chem. Phys. 14, 14010–14022 (2012). 42. Xu, H., Cheng, D., Cao, D. & Zeng, X. C. A universal principle for a rational design of single-atom electrocatalysts. Nat. Catal. 1, 339–348 (2018). 43. Fernando, A. & Aikens, C. M. Theoretical investigation of water oxidation catalysis by a model manganese cubane complex. J. Phys. Chem. C 120, 21148–21161 (2016). 44. Britt, R. D., Suess, D. L. M. & Stich, T. A. An Mn(v)-oxo role in splitting water? Proc. Natl Acad. Sci. USA 112, 5265–5266 (2015). 45. Busch, M., Ahlberg, E. & Panas, I. Electrocatalytic oxygen evolution from water on a Mn(iii–v) dimer model catalyst—a DFT perspective. Phys. Chem. Chem. Phys. 13, 15069–15076 (2011). 46. Siegbahn, P. E. M. Water oxidation mechanism in photosystem II, including oxidations, proton release pathways, O–O bond formation and O2 release. Biochim. Biophys. Acta Bioenergy 1827, 1003–1019 (2013). 47. Segre, C. U. et al. The MRCAT insertion device beamline at the Advanced Photon Source. AIP Conf. Proc. 521, 419–422 (2000). 48. Stern, E. A., Elam, W. T., Bunker, B. A., Lu, K. Q. & Heald, S. M. Ion chambers for fluorescence and laboratory EXAFS detection. Nucl. Instrum. Methods Phys. Res. 195, 345–346 (1982). 49. Ravel, B. N. M. & Newville, M. Data analysis for X-ray absorption spectroscopy using IFEFFIT. J. Synchrotron Radiat. 12, 537–541 (2005). 50. Stern, E. A., Heald, S. M. & Koch, E. Handbook on Synchrotron Radiation (North-Holland, Amsterdam, 1983). 51. Kresse, G. & Hafner, J. Abinitio molecular-dynamics for liquid-metals. Phys. Rev. B 47, 558–561 (1993). 52. Kresse, G. & Furthmuller, J. Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set. Phys. Rev. B 54, 11169–11186 (1996). 53. Perdew, J. P. & Wang, Y. Accurate and simple analytic representation of the electron-gas correlation-energy. Phys. Rev. B 45, 13244–13249 (1992). 54. Blochl, P. E. Projector augmented-wave method. Phys. Rev. B 50, 17953–17979 (1994). 55. Monkhorst, H. J. & Pack, J. D. Special points for Brillouin-zone integrations. Phys. Rev. B 13, 5188–5192 (1976). 56. Kwapien, K., Piccinin, S. & Fabris, S. Energetics of water oxidation catalyzed by cobalt oxide nanoparticles: assessing the accuracy of DFT and DFT plus U approaches against coupled cluster methods. J. Phys. Chem. Lett. 4, 4223–4230 (2013). 57. Wang, L.-P. & Van Voorhis, T. Direct-coupling O2 bond forming a pathway in cobalt oxide water oxidation catalysts. J. Phys. Chem. Lett. 2, 2200–2204 (2011). 58. Li, X. & Siegbahn, P. E. Water oxidation mechanism for synthetic Co–oxides with small nuclearity. J. Am. Chem. Soc. 135, 13804–13813 (2013). 59. Mathew, K., Sundararaman, R., Letchworth-Weaver, K., Arias, T. & Hennig, R. G. Implicit solvation model for density-functional study of nanocrystal surfaces and reaction pathways. J. Chem. Phys. 140, 084106 (2014).

Nature Catalysis 60. Dudarev, S. L., Botton, G. A., Savrasov, S. Y., Humphreys, C. J. & Sutton, A. P. Electron-energy-loss spectra and the structural stability of nickel oxide: an LSDA +​U study. Phys. Rev. B 57, 1505–1509 (1998). 61. Heyd, J., Scuseria, G. E. & Ernzerhof, M. Hybrid functionals based on a screened Coulomb potential. J. Chem. Phys. 118, 8207–8215 (2003). 62. Krukau, A. V., Vydrov, O. A., Izmaylov, A. F. & Scuseria, G. E. Influence of the exchange screening parameter on the performance of screened hybrid functionals. J. Chem. Phys. 125, 224106 (2006). 63. Grant, R. W., Geller, S., Cape, J. A. & Espinosa, G. P. Magnetic and crystallographic transitions in α​-Mn2O3–Fe2O3 system. Phys. Rev. 175, 686–695 (1968). 64. Busch, M., Ahlberg, E. & Panas, I. Hydroxide oxidation and peroxide formation at embedded binuclear transition metal sites; TM =​Cr, Mn, Fe, Co. Phys. Chem. Chem. Phys. 13, 15062–15068 (2011). 65. Yancey, D. F. et al. A theoretical and experimental examination of systematic ligand-induced disorder in Au dendrimer-encapsulated nanoparticles. Chem. Sci. 4, 2912–2921 (2013). 66. Duan, Z. et al. A combined theoretical and experimental EXAFS study of the structure and dynamics of Au 147 nanoparticles. Catal. Sci. Technol. 6, 6879–6885 (2016). 67. Zabinsky, S., Rehr, J., Ankudinov, A., Albers, R. & Eller, M. Multiplescattering calculations of X-ray-absorption spectra. Phys. Rev. B 52, 2995–3009 (1995). 68. Hu, X. L., Piccinin, S., Laio, A. & Fabris, S. Atomistic structure of cobaltphosphate nanoparticles for catalytic water oxidation. ACS Nano 6, 10497–10504 (2012). 69. Chen, Z. et al. Amorphous cobalt oxide nanoparticles as active wateroxidation catalysts. ChemCatChem 9, 3641–3645 (2017). 70. Guan, J. et al. CoOx nanoparticle anchored on sulfonated-graphite as efficient water oxidation catalyst. Chem. Sci. 8, 6111–6116 (2017). 71. Guan, J. et al. Synthesis and demonstration of subnanometric iridium oxide as highly efficient and robust water oxidation catalyst. ACS Catal. 7, 5983–5986 (2017).

Acknowledgements

This work was supported by the National Natural Science Foundation of China (No. 21633010), the 973 National Basic Research Program of China (No. 2014CB239403), the Strategic Priority Research Program of Chinese Academy of Sciences (No. XDB17000000) and Honeywell UOP research cooperation (No. 13C0008). Special acknowledgement goes to the assistance of M. Charochak of UOP and J. Wright of IIT for assistance with data collection at MRCAT, and S. Pennycook of NUS for providing the HAADF-STEM device.

Author contributions

C.L. conceived the project. J.G., F.Z. and J.Q.C. designed the experiments. J.G. performed synthesis, characterization and catalytic reaction experiments. Q.H. performed some catalytic reaction experiments. Z.D. and M.D. performed the DFT calculations. S.D.K. and R.S. measured the XAFS spectra. C.T. measured the HAADF-STEM images of Mn-NG. J.G., C.L. and Z.D. analysed the data and co-wrote the manuscript. All authors discussed the results and commented on the manuscript.

Competing interests

The authors declare no competing interests.

Additional information

Supplementary information is available for this paper at https://doi.org/10.1038/ s41929-018-0158-6. Reprints and permissions information is available at www.nature.com/reprints. Correspondence and requests for materials should be addressed to C.L. Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations. © The Author(s), under exclusive licence to Springer Nature Limited 2018

Nature Catalysis | www.nature.com/natcatal