O-coordinated W-Mo dual-atom catalyst for pH-universal electrocatalytic hydrogen evolution

See allHide authors and affiliations

Science Advances  05 Jun 2020:
Vol. 6, no. 23, eaba6586
DOI: 10.1126/sciadv.aba6586


Single-atom catalysts (SACs) maximize the utility efficiency of metal atoms and offer great potential for hydrogen evolution reaction (HER). Bimetal atom catalysts are an appealing strategy in virtue of the synergistic interaction of neighboring metal atoms, which can further improve the intrinsic HER activity beyond SACs. However, the rational design of these systems remains conceptually challenging and requires in-depth research both experimentally and theoretically. Here, we develop a dual-atom catalyst (DAC) consisting of O-coordinated W-Mo heterodimer embedded in N-doped graphene (W1Mo1-NG), which is synthesized by controllable self-assembly and nitridation processes. In W1Mo1-NG, the O-bridged W-Mo atoms are anchored in NG vacancies through oxygen atoms with W─O─Mo─O─C configuration, resulting in stable and finely distribution. The W1Mo1-NG DAC enables Pt-like activity and ultrahigh stability for HER in pH-universal electrolyte. The electron delocalization of W─O─Mo─O─C configuration provides optimal adsorption strength of H and boosts the HER kinetics, thereby notably promoting the intrinsic activity.


Electrochemical hydrogen evolution reaction (HER) provides a cost-efficient and sustainable method to generate H2 effectively, which is an important component of developing green energy technologies (1). To date, Pt-based nanomaterials are widely considered as the state-of-the-art electrocatalysts for HER. However, the high cost and natural scarcity severely hamper their large-scale applications. Therefore, substantial efforts have been devoted to exploring cheap, efficient, and durable alternatives of Pt for hydrogen production. Recently, nonprecious metal-based catalysts have been studied extensively for HER such as carbides (2), nitrides (3), dichalcogenides (4), phosphides (5), and oxides (6); yet still, their HER activities and stabilities are far from satisfactory. In addition, to meet different applications, catalysts that function well over a wide pH range are highly regarded. For instance, water electrolyzers based on proton exchange membrane technology require catalysts operation in acidic solution; microbial electrolysis cells need catalysts that function well under neutral condition, and alkaline water electrolysis demands catalysts that operate in strongly basic media (7). Hence, designing and developing efficient, durable, and scalable HER catalysts working well in a pH-universal electrolysis system become important yet challenging.

The single-atom catalysts (SACs) maximize the efficiency of metal atoms utilization, enabling reasonable use of metal resources and achieving atomic economy, which is highly desirable for electrocatalysis (8). Unfortunately, the increased surface free energy renders SACs with only one kind of single metal site that tend to agglomerate, leading to a significant decline in performance. Compared to the single-atom components, the multiatom catalysts with tunable electronic environments could further improve the intrinsic activity and stability. In multiatom catalytic system, the metal cluster centers will offer a range of unique and often unexpected catalytic properties (9); the strong chemical interactions between neighboring atom-atom can efficiently stabilize the individual species and prevent agglomeration, thereby creating highly stable active sites. Within this context, dispersing metal atoms on support into a minimal cluster or dinuclear category could modulate the electronic structure by adjusting the ligand atom (10), coordination number (CN) (11), and structural distortion (12). In particular, the heteronuclear metal atom catalysis would optimize activity, stability, and selectivity through tuning metal active centers, which are not observable in their individual monometallic systems (13). Thus, it is very reasonable to expect that the dual-atom catalysts (DACs) with interacted biatomic metal cores could maximize the atom utilization and greatly improve the catalytic activity. However, synthesis and characterization of binuclear metal clusters remain a huge challenge, owing to the lack of atomic-scale control techniques under the harsh synthesis condition and the difficulty in identifying the exact noncrystallographic structures and active sites. Therefore, it is urgent and challenging to controllably synthesize DACs, as well as provide in-depth insight into the synthetic mechanism of heteronuclear species and the local environment associated with catalytic performance.

Polyoxometalates (POMs), a special class of metal oxide anion nanoclusters, feature stability, solubility, compatibility with nearly all other materials (i.e., supports, additives, and solvents), and diverse structural topologies, ensuring them attractive in broad fields of science. Moreover, the variable CN and geometry of POMs with separated units can inhibit agglomeration of metal atoms during the pyrolysis process. While Mo7Ox clusters derived from POMs have been reported (14), they have not yet been explored to construct heteronuclear diatomic catalysts. Here, we report bimetallic DAC consisting of O-coordinated W-Mo heterodimer anchored in N-doped graphene (W1Mo1-NG); the W─O─Mo─O─C configuration of which is distinct from the reported dual-metal sites using traditional methods. The W- and Mo-based POMs are used as precursors for producing W-Mo DAC, which is prepared through a hydrothermal reaction followed by chemical vapor deposition (CVD) process. In W1Mo1-NG, heteronuclear W-Mo dual atoms are immobilized in the NG vacancies and stabilized through W─O and Mo─O bonds. The W and Mo atoms are respectively located in oxygen-bridged [WO4] tetrahedron and distorted [MoO6] octahedron units; W atom is bridged with Mo atom via O atom, functioning as a bimetallic dimer environment. The distinctive W─O─Mo─O─C configuration with strong covalent interactions works as stable and excellent catalyst toward HER in pH-universal electrolysis. The electron delocalization of O-coordinated W-Mo heterodimer results in a favorable adsorption behavior of H and improved HER kinetics, ultimately enhancing the intrinsic activity.



W1Mo1-NG electrocatalyst was synthesized through a three-step procedure, as schematically illustrated in Fig. 1A. A precursor solution was first prepared by sonicating tungstic acid, molybdate, and graphene oxide (GO) in water. The Na2WO4·2H2O and (NH4)6Mo7O24·4H2O were selected as W and Mo sources, respectively. The well-mixed precursor solution was then subjected to a hydrothermal treatment, in which paired W-Mo species were formed and anchored into partially reduced GO (p-RGO). Subsequently, the homogeneous mixture was freeze-dried to minimize restacking of p-RGO sheets. Last, the W1Mo1-NG DAC was obtained with CVD treatment in the NH3/Ar gas at 800°C. For comparison, a series of DACs with various CVD times and molar ratios of W/Mo (table S1), together with homonuclear diatomic Mo and W supported on NG (denoted as Mo2-NG and W2-NG, respectively), were also synthesized. Further details of the experiments were provided in Materials and Methods.

Fig. 1 Synthesis of W1Mo1-NG DAC.

(A) Schematic illustration of the synthetic procedure. (B) The proposed chemical mechanism. The pairing of σ (*) and π (•) localized electrons at the zigzag sites is indicated.

The formation mechanism of W1Mo1-NG DAC was based on POMs self-assembly chemistry. Specifically, Na2WO4·2H2O and (NH4)6Mo7O24·4H2O were added to GO suspension (Fig. 1B, 1) in an ultrasonic bath (pH 6.1 to pH 6.3). Afterward, in the hydrothermal process, protonation of carboxyl anion occurred (Fig. 1B, 2), followed by the removal of protonated carbonyl and epoxy on GO (Fig. 1B, 3); the delocalized π-electron system and H+ underwent protonation to obtain positively charged p-RGO (Fig. 1B, 4) (15). In a mild acidic solution, protonation promoted Mo(VI) oxo in MoO42− fragment transformed to Mo hydroxo and lastly to Mo aquo ligands ([MoO4(H2O)2]2−) (16), while the hydrogentungstate anion ([WO3(OH)]) from WO42− fragment retained the tetrahedral coordination (17). Hence, heteronuclear W-Mo species could be obtained through self-assembly of [WO3(OH)] and [MoO4(H2O)2]2− with dehydration condensation reaction; before aggregation and rearrangement, the heteronuclear W-Mo species could be captured and stabilized with protonated p-RGO sheets as counterion (Fig. 1B, 5) (18). Because of the intensive coupling of negatively charged W-Mo dimer anions and positively charged protonated p-RGO sheets, the O-coordinated W-Mo heterodimers could be electrostatically attracted onto the p-RGO sheets without any additives. As a result, the hydrothermal self-assembly process created a tremendous opportunity for heteronuclear W-Mo dual atoms to anchor in p-RGO through strong covalent interactions. Where small amount of low-nuclear MM′Ox, MMOx, and M′M′Ox (M, M′ = Mo, W) clusters could also formed. Last, uniform W1Mo1-NG DAC (Fig. 1B, 6) was prepared by ammonia annealing. During CVD process, further reduction of p-RGO and N doping occurred simultaneously. The successful synthesis of W1Mo1-NG DAC is contributed to the structural asymmetry between the protonated p-RGO sheets and metal precursors in mild acidic solution, in which protonated p-RGO sheets working as counterion inhibit WO42− and MoO42− fragments aggregating into block materials (19). The general synthetic strategy of heteronuclear DAC can also be extended to other POM systems, such as vanadium, niobium, and tantalum, which provides an alternative way to develop atomically dispersed catalysts with higher complexity.

Structural characterization of W1Mo1-NG DAC

The morphology and structure of W1Mo1-NG were initially investigated by scanning electron microscopy (SEM), transmission electron microscopy (TEM), and aberration-corrected high-angle annular dark-field scanning TEM (AC HAADF-STEM). As revealed in Fig. 2A and fig. S1, W1Mo1-NG and RGO have similar morphology with abundant wrinkles and ripples, which serve as anchoring sites to stabilize the metallic species. No nanoparticles or clusters are observed in the brightfield TEM image (Fig. 2B). The AC HAADF-STEM image (Fig. 2C) presents high density of small bright dots, validating that the W-Mo species are atomically distributed on NG. Furthermore, the magnified AC HAADF-STEM image clearly demonstrates the existence of a large proportion of isolated heteronuclear W-Mo atoms (marked with circles, Fig. 2D). Apart from the dominant amount of W-Mo dual atoms, a few small clusters are also observed (fig. S2). The statistical analysis of 100 pairs of heteronuclear W-Mo atoms shows that the W-Mo distance is below 3.6 Å (Fig. 2E), which is conspicuously longer than the Mo─Mo/W─Mo/W─W bonds (ca. 3.0 Å) in metals bulk, indicating that the W and Mo atoms are bridged via extra atoms. In W1Mo1-NG, the coaxial line of W and Mo atoms may not be parallel to the NG plane due to the wrinkles in NG, resulting in variable W-Mo diatomic distances in HAADF images. Figure 2F is a simplified and intuitive depiction of the dizygotic W-Mo atoms anchoring in NG and exhibits the W-Mo distance in regions 1 to 3 of Fig. 2D. The heteronuclear W-Mo atoms can be clearly distinguished by intensity; the distances shown in the intensity profiles are consistent with the projection spacing of W-Mo atoms on the visual plane. The energy-dispersive x-ray spectroscopy (EDS) mapping analysis reveals the homogeneous distribution of C, N, O, Mo, and W elements (Fig. 2G and fig. S3). Simultaneously, the HAADF images of Mo2-NG and W2-NG also clearly reveal the homonuclear Mo2 and W2 atoms anchored in NG sheets, respectively (figs. S4 and S5). The above results demonstrate that the hydrothermal and CVD methods effectively create high-density heteronuclear W-Mo atoms distributed on pleated NG sheets.

Fig. 2 Morphology characterization of W1Mo1-NG.

(A) SEM image. (B) TEM image. (C and D) AC HAADF-STEM images. The image clearly displays paired W-Mo atoms (white circles). (E) Statistical W-Mo diatomic distance in AC HAADF-STEM images. (F) Intensity profiles obtained in areas 1 to 3 and the corresponding schematics show different W-Mo projection distances on the visual plane. For top, middle, and bottom schematics, tilt angles of NG plane relative to the visual plane are 0°, 25°, and 90°, respectively. Atom colors: royal blue, W; cyan, Mo; dark gray, C; blue, N; red, O. (G) HAADF-STEM image and corresponding EDS elemental mapping images.

To gain insights into the chemical composition and the valence states of W1Mo1-NG DAC, we carried out Raman spectrum, x-ray diffraction (XRD), and x-ray photoelectron spectroscopy (XPS) measurements. As exhibited in Raman spectra (Fig. 3A), the intensity ratio of D band to G band (ID/IG) for W1Mo1-NG (1.16) is higher than that of NG (1.05), which indicates that structural defects are introduced into W1Mo1-NG. Furthermore, the Brunauer-Emmett-Teller surface areas of W1Mo1-NG and NG are 566 and 268 m2 g−1, respectively (Fig. 3B), and the main pore size is centered at 3.5 nm (Fig. 3B, inset). These results suggest that the interacted W and Mo atoms can effectively inhibit the agglomeration of NG sheets and generate a considerable amount of voids (fig. S6), which provides nucleation sites for W-Mo heterodimers. The defective W1Mo1-NG with porous structure and large surface area can effectively promote the electrolyte permeation and charge transfer. In XRD patterns, no diffraction peaks of metal-based crystalline phases in DACs, such as unary W- or Mo-based carbides/nitrides/oxides, are detected (Fig. 3C). However, XPS survey spectrum confirms that W1Mo1-NG consists of W, Mo, N, O, and C elements (Fig. 3D), consistent with the EDS elemental mapping (Fig. 2G). The atomic concentrations of W and Mo in W1Mo1-NG determined by XPS are 0.38 and 0.36 atomic %, respectively, which are close to the inductively coupled plasma mass spectrometry (ICP-MS) results [5.01 and 2.55 weight % (wt %), respectively] (table S2). Notably, all these DACs have relatively high metal loading of 7.5 wt %, which could not be achieved with conventional deposition-reduction method (13). The Mo 3d XPS of W1Mo1-NG shows two peaks at 235.6 and 232.4 eV (Fig. 3E), corresponding to Mo 3d3/2 and Mo 3d5/2 core energy levels, respectively, validating the presence of Mo(VI) (20). For W 4f XPS (Fig. 3F), W ions have two well-defined peaks at 37.5 and 35.3 eV in W1Mo1-NG ascribed to the W 4f5/2 and W 4f7/2, respectively, being consistent with high oxidation state of W(VI) (21). The Mo 3d in W1Mo1-NG has some shift toward low binding energy compared to Mo2-NG, whereas the W 4f in W1Mo1-NG shifts positively relative to W2-NG. These microdisplacements indicate the altering of local electronic structure for heteronuclear W1Mo1-NG system, which could be attributed to the structural perturbation of W-Mo heterodimers. Moreover, the control samples with various ratios of W/Mo and CVD times were also subjected to XPS measurements. For Mo 3d and W 4f spectra, the varying metal ratio triggers an electron transfer at a certain CVD time (fig. S7), while adjusting the CVD time with a fixed metal ratio does not introduce any changes (fig. S8). As a result, the local electronic structure of W-Mo heterodimers can be effectively modulated via adjusting the ratio of W/Mo. As revealed in Fig. 3G, the deconvoluted peaks around 530.2 and 530.8 eV are assigned to Mo─O and W─O bonds (22, 23), respectively, further confirming that W-Mo heterodimers are O-coordinated within W1Mo1-NG; the peak around 532.0 eV derives from oxygen vacancy (Ov) (24). Notably, peak areas of Ov are different; W1Mo1-NG has higher Ov concentration than single metal–doped Mo2-NG and W2-NG, in accordance with the electron paramagnetic resonance spectra (fig. S9). Because the oxygen-containing functional groups are easily detached from fluffy NG during CVD process, the concentration of Ov may be positively correlated with the specific surface area. Meanwhile, the N 1s spectra are deconvoluted into three characteristic peaks of pyridinic-N (398.0 eV), pyrrolic-N (399.2 eV), and graphitic-N (401.4 eV) species (Fig. 3H), demonstrating the NG, while W─N bond only appears in W2-NG, suggesting W atoms coordinating with N atoms. Metal carbides are absent in the C 1s spectra of DACs (fig. S10). Distinctly, the metal centers in W1Mo1-NG and Mo2-NG are O-coordinated, while the W(VI) in W2-NG are co-coordinated with O and N atoms.

Fig. 3 Compositional characterization.

(A) Raman spectra of W1Mo1-NG and NG. The two-dimensional (2D) peak is obviously enhanced in W1Mo1-NG, demonstrating that the number of graphene layers is reduced compared to raw NG. The appearance of D + G band at 2922 cm−1 declares the increased structure disorder. a.u., arbitrary units. (B) N2 adsorption-desorption isotherm; inset is pore size distribution diagram. (C) XRD patterns. The weak diffraction peaks at 14.6° and 43.1°, respectively, demonstrate the presence of oxygen-containing functional groups and a short-range order in stacked graphene layers. (D) XPS survey spectra. The W is absent in Mo2-NG, and Mo is absent in W2-NG. Inset is the percentages of W, Mo, N, O, and C in W1Mo1-NG measured by XPS and ICP-MS. High-resolution (E) Mo 3d and (F) W 4f XPS spectra. (G) O 1s spectra of Mo2-NG, W1Mo1-NG, and W2-NG. The fitted peak at 533.2 eV can be assigned to C─O bond. (H) High-resolution N 1s spectra of W1Mo1-NG, Mo2-NG, and W2-NG.

To probe the local coordination chemistry of DACs, we conducted x-ray absorption near-edge structure (XANES) and extended x-ray absorption fine structure (EXAFS). In the Mo K-edge XANES (Fig. 4A), W1Mo1-NG and Mo2-NG exhibit a shoulder peak (■) in the pre-edge region, suggesting the formation of distorted [MoO6] octahedron with Mo═O bond (25). Moreover, according to the linear relationship between the Mo oxidation state and the energy position (♦) (26), the average oxidation state of Mo ions in W1Mo1-NG and Mo2-NG is close to 6, which matches well with the XPS results (Fig. 3E). Figure 4B shows the k3-weighted Fourier transform (FT)–EXAFS spectra at Mo K-edge. As expected, W1Mo1-NG and Mo2-NG have two strong peaks at 1.25 and 1.88 Å, which could be attributed to Mo═O and Mo─O bonds, respectively. Comparing with Mo foil, no apparent peaks (2.43 and 2.96 Å) for Mo─Mo bonds are detected in both W1Mo1-NG and Mo2-NG, confirming the absence of Mo nanoparticles, in line with the STEM observation. For the W L3-edge XANES spectra, the oxidation state of W ions in W1Mo1-NG and W2-NG is higher than W powder but lower than WO3 (Fig. 4C). The primary peak around 10,210 eV is attributed to the electronic transition from 2p3/2 to unoccupied 5d orbital (27). In comparison with W2-NG, W1Mo1-NG exhibits higher intensity of white line, which demonstrates the electron deficiency at W atoms by introducing Mo atoms, suggesting strong interactions between heteronuclear W-Mo atoms and NG support. Furthermore, the main peak of W1Mo1-NG (10,210.12 eV) moves toward low energy compared to W2-NG (10,210.71 eV) and WO3 (10,210.74 eV), revealing the low CN environment of W atoms in W1Mo1-NG (28). As displayed in the FT-EXAFS spectra at W L3-edge (Fig. 4D), W1Mo1-NG and W2-NG have only one prominent peak located at 1.36 Å assigned to the W─O coordination. The weak peaks at 2.86 and 3.20 Å correspond to the W─Mo and W─W contributions for W1Mo1-NG and W2-NG, respectively, indicating that W atoms within W1Mo1-NG and W2-NG are atomically distributed. To confirm the adjacent properties of the metal atoms, we conducted wavelet transform (WT)–EXAFS. As shown in Fig. 4E, the WT signal related to Mo─Mo bond in Mo foil is not detected in both W1Mo1-NG and Mo2-NG, further confirming the absence of Mo-containing nanoparticles. Compared to MoO3, W1Mo1-NG and Mo2-NG show two intensity maxima at 3.9 and 9.3 Å−1, which are respectively associated with the Mo─O and Mo─O─W/Mo paths, consistent with the Mo K-edge FT-EXAFS results. For WT-EXAFS at W L3-edge (Fig. 4F), W1Mo1-NG and W2-NG display the intensity maxima at 4.1 and 3.9 Å−1 for W─O and W─O/N paths, respectively, which are close to the W─O bond (4.2 Å−1) in WO3 reference, but distinct from the W─W bond (7.3 Å−1) of W powder. Moreover, WT signal at 9.2 Å−1 of W1Mo1-NG and W2-NG should be derived from the W─O─Mo and W─N/O─W contributions, respectively. Unambiguously, by combining the WT results at Mo K-edge and W L3-edge, the metal atoms are confirmed to be bridged via extra O or N atoms in the W1Mo1-NG, Mo2-NG, and W2-NG. To determine the configuration of these DACs, we used density functional theory (DFT) calculations to deduce the bimetal models based on the FT-EXAFS experimental spectra and fitted curves. The EXAFS fitting results exhibit that the Mo atom has four coordinating interactions of Mo═O (1.79 Å), Mo─O (1.99 Å), Mo─C/N (2.66 Å), and Mo─W (3.29 Å) and the corresponding CN values are 1.9, 2.2, 1.1, and 0.9, respectively (table S3). Meanwhile, the four coordination paths of W atom at 1.83, 2.01, 2.67, and 3.29 Å are respectively assigned to W═O, W─O, W─C/N, and W─Mo with corresponding CN of 2.1, 1.7, 1.2, and 1.1 (table S4). The EXAFS fitting curves at R-space and k-space are consistent well with experiment spectra of W1Mo1-NG (Fig. 4, G and H). Then, possible W-Mo heterodimer models were constructed and optimized with DFT calculations based on the coordination environment (fig. S11). For the O-coordinated W1Mo1-NG model (Fig. 4G, inset), the W═O, Mo═O, and O-bridged W─Mo paths length are 1.75, 1.71, and 3.54 Å (fig. S12), respectively, in good agreement with EXAFS fitting data (1.79, 1.79, and 3.29 Å, respectively). As a result, the geometry of heteronuclear W-Mo atoms anchored in NG vacancy by four O atoms and two C atoms is the most possible actual configuration. Wherein one W atom binding with O atom to form W─O motif, which further binds with a neighboring Mo atom to form a W─O─Mo─O─C configuration, while the W and Mo atoms respectively exist in the [WO4] tetrahedron and distorted [MoO6] octahedron units. Quantitative EXAFS fitting results for homonuclear systems are also summarized in Fig. 4, I and J and fig. S13. Note that the local coordination of Mo atom in Mo2-NG resembles that of W1Mo1-NG, suggesting similar local environment around the Mo center, while for W2-NG, the CN of the W─O/N path (second shell) is 2.3, which is higher than that of W1Mo1-NG (1.7). According to the EXAFS fitting and DFT calculation results, the homonuclear Mo2 and W2 dual atoms are determined to be inlaid in NG with seven O and N atoms, respectively (fig. S14). Other coordination configurations are excluded by comparing the EXAFS fitting and XPS results (fig. S15). Compared to homonuclear Mo2-NG and W2-NG, the coaxial line of W-Mo atoms is not parallel to the NG plane, which gives rise to structural deformation. The structural perturbation of W-Mo heterodimers can effectively modulate the electronic structure of the W, Mo, and O atoms in the defective NG, ensuring high intrinsic activity. Obviously, the paired metal atoms in W1Mo1-NG (W─O─Mo─O─C) and Mo2-NG (Mo─O─Mo─O─C) are anchored in NG by O atoms, while paired W atoms in W2-NG (W─N─W─N─C) are anchored in NG by N atoms. Usually, the metal atoms are dispersed on N-doped carbon support in SACs with form of M─N─C (M refers to transition metals). Unambiguously, the bulk oxides of W and Mo can be converted to the corresponding nitrides in an NH3 atmosphere at high temperature (29). In this work, the energy provided during the CVD process probably fails to reach the dissociation energy of M─O bonds (M = W, Mo) in W-Mo and Mo2 dimers obtained via hydrothermal process, thus keeping M─O coordination. For the tungstate system, the structural stability of isolated dinuclear tungsten oxide is extremely poor (30), so that the weaker W─O bond cannot survive in the high temperature CVD process compared to W─N bond. Consequently, we have prepared DAC with W─O─Mo─O─C configuration through facile self-assembly of W- and Mo-based POMs, followed by CVD procedure. In W1Mo1-NG, the O-coordinated W-Mo heterodimer with distorted structure provides the possibility to modulate the electronic structure around the W, Mo, and O atoms, thereby improving the electrocatalytic activity.

Fig. 4 X-ray absorption characterization.

(A) XANES and (B) k3-weight FT-EXAFS spectra of W1Mo1-NG, Mo2-NG, and the reference samples at Mo K-edge. (C) XANES and (D) FT-EXAFS spectra of W1Mo1-NG, W2-NG, W powder, and WO3 at W L3-edge. The W ions in WO3 are in a [WO6] octahedron environment. WT-EXAFS of W1Mo1-NG, Mo2-NG, W2-NG, and the references at (E) Mo K-edge and (F) W L3-edge. (G and H) The corresponding Mo K-edge and W L3-edge EXAFS fitting curves for W1Mo1-NG at R-space and k-space, respectively. (I) The corresponding Mo K-edge EXAFS fitting curve for Mo2-NG at R-space. (J) The corresponding W L3-edge EXAFS fitting curve for W2-NG at R-space. The insets of (G), (I), and (J) are the schematic models of W1Mo1-NG, Mo2-NG, and W2-NG, respectively. Atom colors: royal blue, W; cyan, Mo; dark gray, C; blue, N; red, O.

Electrocatalytic HER performance

The HER performance of DACs was investigated in 0.5 M H2SO4 and 1.0 M KOH solutions using a standard three-electrode setup. All potentials were referenced to a reversible hydrogen electrode (RHE) and without iR compensation. The commercial Pt/C and raw NG were tested for comparison. As illustrated in Fig. 5A, W1Mo1-NG exhibits excellent catalytic activity toward HER, giving a near-zero onset potential (Uonset) in acidic electrolyte. As a consequence of polarization, W1Mo1-NG produces cathodic geometric current density (j) of 10 mA cm−2 at an overpotential of 24 mV (η10 = 24 mV), which is much lower than those of Mo2-NG (145 mV), W2-NG (156 mV), raw NG (200 mV), and other counterparts (fig. S16). Impressively, W1Mo1-NG delivers a high cathode current density at overpotential (η) above 50 mV, which is even superior to that of commercial Pt/C catalyst. Distinctly, this overpotential (η10 = 24 mV) is the lowest among the reported earth-abundant HER catalysts, such as CoP/NiCoP nanotadpoles (125 mV) (31), Ni2P/Ni@C (149 mV) (32), and Fe3C-Co/NC (298 mV) (2), as well as other catalysts (table S5). Specifically, W1Mo1-NG affords a small Tafel slope of 30 mV dec−1, which is much lower than those of Mo2-NG (175 mV dec−1), W2-NG (107 mV dec−1), and NG (173 mV dec−1) (Fig. 5B), reflecting that W1Mo1-NG has highest HER kinetic process through the Volmer-Tafel mechanism and the surface recombination step is the rate-limiting step (33). The exchange current density (j0) of W1Mo1-NG is calculated to be 1.78 mA cm−2, which is much higher than those of Mo2-NG (1.50 mA cm−2), W2-NG (0.33 mA cm−2), and NG (0.67 mA cm−2) (fig. S17). Distinctly, W1Mo1-NG has such a lower Tafel slope and larger j0 value than those of Mo2-NG, W2-NG, and raw NG, suggesting the superior HER kinetics and enhancing the intrinsic activity of W1Mo1-NG. In addition, in alkaline electrolyte (Fig. 5, C and D), W1Mo1-NG exhibits a small Uonset of −26 mV, upon which the cathodic current density elevates substantially at more negative potentials; W1Mo1-NG delivers η10 of 67 mV, which is much lower than those of Mo2-NG (247 mV), W2-NG (223 mV), NG (281 mV), other counterparts, and reported transition metal–based HER catalysts (fig. S18 and table S6). The Tafel slope of W1Mo1-NG is 45 mV dec−1, which is lower than those of Mo2-NG (124 mV dec−1), W2-NG (94 mV dec−1), and NG (128 mV dec−1). Similarly, the j0 of W1Mo1-NG is calculated to be 0.26 mA cm−2, which is much higher than those of Mo2-NG (0.19 mA cm−2), W2-NG (0.04 mA cm−2), and NG (0.08 mA cm−2) (fig. S19). The results reveal that W1Mo1-NG has more favorable HER kinetics than Mo2-NG, W2-NG, and NG under both acidic and alkaline media. We further compared the HER activity with recently reported tungsten and molybdenum electrocatalysts (Fig. 5E). It is evident that W1Mo1-NG DAC requires the lowest η10 (24 and 67 mV, respectively) in 0.5 M H2SO4 and 1.0 M KOH, outperforming almost all W- and Mo-based HER catalysts. In addition, the electrochemical double-layer capacitance (Cdl) was measured by performing cyclic voltammetry (CV) at various scan rates (Fig. 5F). W1Mo1-NG has the Cdl values of 7.20 and 6.18 mF cm−2, which are better than those of Mo2-NG (5.26 and 4.15 mF cm−2), W2-NG (6.01 and 4.72 mF cm−2), and NG (2.06 and 1.80 mF cm−2) in 0.5 M H2SO4 and 1.0 M KOH solutions, respectively (figs. S20 and S21), suggesting that the heteronuclear W1Mo1-NG system can subjoin active surface area (22). Furthermore, geometric current density has been widely used to evaluate the activity of electrocatalysts. However, there is considerable dispute on its effectiveness especially in the case of highly porous electrocatalysts (5). To exclude the effect of surface area and compare the intrinsic HER activity, geometric current density was normalized by electrochemical active surface area (ECSA; jECSA) and to further identify the catalysts’s intrinsic activity. As shown in fig. S22, W1Mo1-NG delivers the jECSA values of 1100.49 and 165.00 μA cm−2 at η = 100 mV, which are markedly higher than those of Mo2-NG (32.28 and 11.50 μA cm−2), W2-NG (34.40 and 3.63 μA cm−2), NG (46.60 and 11.83 μA cm−2), and Pt/C (262.82 and 69.64 μA cm−2) in 0.5 M H2SO4 and 1.0 M KOH solutions, respectively. These reveal that W1Mo1-NG has better intrinsic activity than Mo2-NG, W2-NG, and NG, even outperforming benchmark Pt/C in pH-universal electrolyte (8).

Fig. 5 HER performance.

The polarization curves and the corresponding Tafel plots of Mo2-NG, W1Mo1-NG, W2-NG, NG, and Pt/C in (A and B) 0.5 M H2SO4 and (C and D) 1.0 M KOH, respectively. (E) Comparison of the η10 in 0.5 M H2SO4 and 1.0 M KOH solutions for W1Mo1-NG and reported W- and Mo-based HER electrocatalysts. (F) Cyclic voltammetry (CV) cycles for W1Mo1-NG with varying scan rates from 20 to 300 mV s−1; the inset is Cdl measurement in 1.0 M KOH. (G) Turnover frequency (TOF) values of Mo2-NG, W1Mo1-NG, and W2-NG in 0.5 M H2SO4 and 1.0 M KOH solutions, respectively. (H) Polarization curves recorded from W1Mo1-NG catalyst before and after 50,000 potential cycles in 0.5 M H2SO4. (I) Galvanostatic responses recorded on W1Mo1-NG and Pt/C at a cathodic current density of 10 mA cm−2 under acidic and alkaline solutions.

In addition, to gain further insight into the intrinsic catalytic activity, we measured the turnover frequency (TOF) per active site. Thus, the number of active sites was titrated from integrated charge of anodic CV cycles at phosphate buffer solution (pH 7) (fig. S23) (34). Evidently, W1Mo1-NG affords the highest TOF values of 2.36 and 0.42 H2 s−1 at η = 100 mV, which are about 59, 42, and 26, 21 times higher than those of Mo2-NG (0.04 and 0.01 H2 s−1) and W2-NG (0.09 and 0.02 H2 s−1) in 0.5 M H2SO4 and 1.0 M KOH, respectively (Fig. 5G). The results indicate that heteronuclear W1Mo1-NG system delivers prominent catalytic hydrogen production capacity compared to homonuclear Mo2-NG and W2-NG. This superior catalytic performance is related to the distinctive W─O─Mo─O─C configuration, which provides a unique electronic structure and improves electrical conductivity, notably affecting on the electron transfer from the catalyst surface to adsorbed species (35). Besides, electrochemical impedance spectroscopy (EIS) technique was also performed. Among them, W1Mo1-NG DAC yields small charge-transfer resistance (Rct) (1 and 2 ohms) in 0.5 M H2SO4 and 1.0 M KOH, which are much lower than those of Mo2-NG (5 and 5 ohms), W2-NG (4 and 10 ohms), and NG (40 and 20 ohms), respectively (fig. S24). The Mo2-NG and W2-NG both exhibit lower Rct than that of raw NG, indicating that the foreign metal atoms doping facilitate electron transport at the electrode/electrolyte interface. Notably, the HER performance of W1Mo1-NG catalyst is extremely superior to Mo2-NG and W2-NG, which can be attributed to the atomic W─O─Mo─O─C moiety having favorable local electronic structure. Here, the highest performance of W1Mo1-NG is achieved with feed molar ratio for W/Mo of 7:13 at CVD processing time of 3 hours, indicating that the synergy of W and Mo species decreases the reaction kinetic barrier for HER. Since the self-assembly of POMs strongly depended on the pH and the concentration of metal precursors, W1Mo1-NG heterocatalysts with W/Mo molar ratio of 1:1 are successfully synthesized with feed molar ratio for W/Mo of 7:13, based on XPS, ICP-MS, x-ray absorption, and DFT results. When the feed molar ratio of W/Mo precursors is 1:1, the loadings of W and Mo atoms in the catalyst are 6.82 and 1.02 wt %, respectively (table S2). Under this condition, the paired W-Mo atoms are not dominant amount, which weakens the synergistic effect of W and Mo atoms and reduces HER activity. As a result, the optimal performance in HER is achieved on W1Mo1-NG with electron delocalization of paired W-Mo atoms.

The stability and durability of W1Mo1-NG were further analyzed. As illustrated in Fig. 5H, the polarization curve of W1Mo1-NG exhibits negligible differences compared with the initial curve after 10,000 even 50,000 CV cycles in 0.5 M H2SO4. Meanwhile, the polarization curves before and after 50,000 HER cycles in 1.0 M KOH also remain virtually unchanged (fig. S25). Furthermore, we performed aggressive long-term stability test on W1Mo1-NG by galvanostatic measurement at a current density of −10 mA cm−2 at room temperature (25°C) (Fig. 5I). After 100,000 s, W1Mo1-NG DAC still retains 98% of the initial overpotential in 0.5 M H2SO4 and 95% in 1.0 M KOH solutions, respectively, higher than those of Pt/C with 93 and 82%. Again, to further study the effect of temperature on stability, W1Mo1-NG electrocatalyst was also verified by chronopotentiometric curves in 0.5 M H2SO4 and 1.0 M KOH solutions below (5°, 10°, 15°, and 20°C) and above (40°, 60°, 80°, and 90°C) room temperature. As exhibited in fig. S26, W1Mo1-NG exhibits substantially constant overpotentials over the temperature range 5° to 90°C for 100,000 s, highlighting the excellent stability over a wide temperature range. Besides, the XRD characterization reveals no evidence phase transition for W1Mo1-NG after cycling test (fig. S27). The chemical valence states of W and Mo ions remain almost unchanged, further signifying the robustness of the heteronuclear W1Mo1-NG catalyst (fig. S28). In particular, the HER activity of W1Mo1-NG DAC remains undegraded even storing for more than 2 years, confirming the remarkable chemical stability (fig. S29). The excellent physicochemical stability can be attributed to the robust W─O─Mo─O─C configuration with strong W─O and Mo─O bonds inherited from the W- and Mo-based POMs, respectively.

DFT calculation

To understand that the O-coordinated W1Mo1-NG system promotes the catalytic reaction over wide range of pH 0 to pH 14, we performed the correlative theoretical calculations based on DFT studies. In general, the Gibbs free energy of hydrogen adsorption (ΔGH) is a widely used descriptor to assess the activity of a catalyst for HER in both acidic and alkaline electrolytes (36). Here, a large amount of possible W-Mo, Mo2, and W2 dimer models were judiciously constructed by switching their coordinated O and N atoms while maintaining the metal centers unchanged. On the basis of EXAFS fitting results, the DFT geometry optimization indicates that the O-bridged W-Mo atoms are anchored in NG vacancies through oxygen atoms with W─O─Mo─O─C configuration in heteronuclear model of W1Mo1-NG, as shown in Fig. 6A, including all possible active sites, namely, O1, O2, C3, O4, O5, O6, O7, W8, and Mo9. While, for the homonuclear Mo2-NG and W2-NG models, atomic H adsorbs on terminal O atoms linked with metal centers (Fig. 6, B and C). We calculated ΔGH on the abovementioned sites for DACs, and the results are summarized in Fig. 6D and fig. S30. For heteronuclear W1Mo1-NG system, according to the Sabatier principle, we focused on six nonequivalent active sites: Mo-coordinated terminal O, O1; W-coordinated terminal O, O2; bridging oxygen in Mo─O─C motif, O4 and O5; bridging oxygen in Mo─O─W motif, O6; Mo-coordinated C, C3. Specifically, the bridged oxo in POMs is preferred for proton adsorption; thus, bridging oxygen in W─O─Mo─O─C configuration inherited from POMs shows optimal ΔGH in electrocatalytic HER. The ΔGH of these six nonmetal sites locates far optimal value than those of homonuclear Mo2-NG and W2-NG, implying that the heteronuclear W1Mo1-NG with diverse binding sites is more active than the homonuclear catalysts. Obviously, W1Mo1-NG system with six nonmetal sites guarantees optimal ΔGH throughout the entire pH range, indicating that W─O─Mo─O─C configuration could redistribute the electronic structure, which affords an improved electron environment for HER. In particular, the O1 site provides a low ΔGH of −0.065 eV, which is close to the optimal condition and even lower than that of commercial Pt catalyst (−0.10 eV). For O-coordinated W1Mo1-NG DAC, ΔGH can reach the optimal value (marked with dash line) at any pH, which is consistent with the electrochemical test results, suggesting that the unique O-coordinated W-Mo heterodimer is the active ingredient for HER under given pH condition. Moreover, when W-Mo heterodimers are coordinated with N atoms and tuned O/N atoms, the coordination environment of metal atoms is inconsistent with the EXAFS fitting and XPS results, and the corresponding W─O─Mo─N─C, W─N─Mo─N─C and W─O─Mo─(N, O)─C models show positive ΔGH from 1.760 to 2.347 eV with no H adsorption (fig. S31), further demonstrating the validity of the W─O─Mo─O─C model. In contrast, the experimentally determined O-coordinated Mo2-NG and N-coordinated W2-NG systems exhibit the ΔGH of −3.172 and −1.250 eV, respectively, indicating overbinding of H, thereby leading to the sites blocking. Furthermore, the other N/O-coordinated Mo2 and W2 homodimer configurations display ΔGH ranging from 1.09 to 2.278 eV (figs. S32 and S33), where the coordination of Mo and W atoms actually contradicts with the EXAFS fitting and XPS results, confirming the structural rationality of O-coordinated Mo2-NG and N-coordinated W2-NG models. Thus, according to the DFT calculation results, the O-coordinated W1Mo1-NG system has a near-optimum synergistic effect and gives more favorable ΔGH for HER than O-coordinated Mo2-NG and N-coordinated W2-NG systems as well as other Mo atom– and W atom–nucleated structures.

Fig. 6 HER mechanism.

Optimized geometries and possible active sites for H adsorption on (A) W1Mo1-NG, (B) Mo2-NG, and (C) W2-NG systems. Atom colors: cyan, Mo; royal blue, W; dark gray, C; blue, N; red, O; orange, H. (D) ΔGH diagrams of W1Mo1-NG, Mo2-NG, and W2-NG. The differential charge density maps of heteronuclear W1Mo1-NG system from (E) (100) and (F) (010) sections. The white dash line is the position of NG plane. The differential charge density maps of homonuclear (G) Mo2-NG and (H) W2-NG. The orange and aquamarine regions indicate electron depletion and accumulation, respectively. (I) Projected density of states (PDOS) of metal and O atoms in W1Mo1-NG, Mo2-NG, and W2-NG, the zero energy corresponds to the vacuum level. (J) pCOHP of the H adsorption on terminal O atoms for the W1Mo1-NG, Mo2-NG, and W2-NG models.

To investigate the electronic structure of W1Mo1-NG DAC with improved HER activity, we perform the Bader charge analysis to quantitatively chemical analyze the differences among the DACs (see Materials and Methods). The W1Mo1-NG, Mo2-NG and W2-NG systems give almost identical Bader charge value for O atom with 6.83 e. However, the O-coordinated W1Mo1-NG delivers a Bader volume of 291 pm3, which is much larger than those of Mo2-NG (206.6 pm3) and W2-NG (149.8 pm3). In addition, as shown in Fig. 6, E and F, electrons are delocalized in W1Mo1-NG system, well corresponding to the much larger Bader volume, indicating that the O-coordinated W-Mo heterodimer has strong covalent character with weak ionic property. Compared to heteronuclear W1Mo1-NG, the homonuclear Mo2-NG and W2-NG exhibit growing ionic nature because the electrons are strongly localized around coordination atoms (Fig. 6, G and H), via which the electrons from the metal centers are partially depleted to the coordinated O sites, leading to strong overbinding of H. Besides, the projected density of states (PDOS) on Mo, W, and O atoms was further performed to detect the electronic band structure of W1Mo1-NG. As illustrated in Fig. 6I, for three DAC systems, the Mo, W, and O atoms exhibit unoccupied states, deriving from the covalent bonding between metal ions and O atoms; the density of states (DOS) around the Fermi level (EF) is dominated by O 2p orbital. Notably, heteronuclear W1Mo1-NG exhibits an increased DOS for the occupied states from 0.30 eV to the EF (the shadowed part in Fig. 6I) due to the delocalized electrons (37), corresponding to the favorable adsorption behavior of H. Meanwhile, the delocalized electrons for W1Mo1-NG system can reduce the adsorption energy between the active sites and water in alkaline solution (37), which could lower the energetic barrier for the reorganization of the interfacial water network, thereby enhancing the hydrogen evolution rate (38). Besides, the results also suggest that W1Mo1-NG with delocalized electrons offers enhanced electrical conductivity compared to Mo2-NG and W2-NG, in accordance with EIS results (fig. S24). As a result, the distinctive W─O─Mo─O─C configuration with delocalized electrons ensures the optimal ΔGH and enhances reaction kinetics, thereby improving the HER activity in pH-universal electrolyte. In W1Mo1-NG system, the overlap of the electron clouds between W site and O site deforms the electron cloud of Mo site, and vice versa, which could cause distortion of the structure and electron redistribution, resulting in the electron delocalization of the W─O─Mo─O─C configuration (39). In addition, since the terminal O atom is the most promising active site, we extracted the bonding and antibonding states of H adsorbed on terminal O atom by projected crystal orbital Hamilton population (pCOHP) analysis. Fig. 6J displays the −pCOHP curves of the H─O coupling in the adsorption configurations. In O-coordinated W1Mo1-NG and Mo2-NG systems, O 2s and O 2pz valence orbitals are actively participating in the hybridization with H. The antibonding states of Mo2-NG system (1.67 eV) move to higher energy level compared to W1Mo1-NG system (0.50 eV), indicating strong interaction with hydrogen (40), which corresponds to the excessive adsorption strength of Mo2-NG (ΔGH = −3.172 eV). Whereas, in N-coordinated W2-NG system, the O 2px and O 2pz valence orbitals are participating in the hybridization with H, which causes an antibonding state (0.40 eV) above the EF, ensuring strong interaction with hydrogen (ΔGH = −1.25 eV).


In summary, we have developed a facile two-step method to prepare the W1Mo1-NG DAC via a hydrothermal self-assembly of pH-dependent POMs and GO, followed by CVD process. In W1Mo1-NG, O-coordinated W-Mo heterodimers are anchored in NG; W and Mo atoms are respectively located in oxygen-bridged [WO4] tetrahedron and distorted [MoO6] octahedron. The distinctive W─O─Mo─O─C configuration is the active site for HER, which guarantees excellent catalytic activity and high stability. DFT results reveal that electron delocalization of W1Mo1-NG system affords the desirable ΔGH and good HER kinetics, thereby promoting HER activity. These results suggest that the self-assembly of POMs can be a versatile method for preparing atomic-scale catalysts with uniform distributed active ingredients.


Materials synthesis

Na2WO4·2H2O and (NH4)6Mo7O24·4H2O precursors were added to aqueous suspension of GO to form a homogeneous solution. Subsequently, the resulting mixture was transferred to a Teflon-lined autoclave and heated at 190°C for 12 hours. The as-prepared product was freeze-dried to produce a spongy column. Sponge column of p-RGO with O-coordinated W-Mo heterodimers was obtained by freeze drying of the product after hydrothermal synthesis. Last, the nanocomposite was reacted with NH3/Ar at 800°C in a CVD apparatus to obtain Mo2-NG, W1Mo1-NG, W2-NG, and NG catalysts. The details are shown in the Supplementary Materials.

Materials characterization

A JEOL-JSM-7001F SEM and JEOL 2100 field emission gun TEM were used to examine the morphology. AC HAADF-STEM images and EDS elemental mappings were carried out by a Cs-corrected FEI Titan G2 60-300 equipped with a Super-X EDS detector and operated at 300 kV. Chemical compositions and elemental oxidation states were checked by XPS spectra (PHI-5702). X-ray absorption spectra (XANES and EXAFS) of W L3- and Mo K-edges were measured on VESPERS beamline at the Canadian Light Source, by scanning a double-crystal Si(111) monochromator and collecting emitted x-ray fluorescence.

Computational details

The first-principles calculations based on the DFT were performed within spin-polarized generalized gradient approximation. Core electron states were represented by the projector augmented-wave method (41) as implemented in the Vienna ab initio simulation package (42). The Perdew-Burke-Ernzerhof (43) exchange correlation functional and a plane wave representation for the wave function with a cutoff energy of 450 eV were used. Lattice parameters and geometry of crystal were fully optimized on the basis of the criteria of a maximum atomic force of 0.01 eV Å−1 and energy convergence criteria of 0.1 meV before the electron structure and total energy calculation. We adopted a single layer–doped graphene with 7 × 7 supercell model and chose the K-point mesh of 2 × 2 × 1. The spin polarization was also switched on to account its effect, and 15Å vacuum layer was added to separate the graphene layers. The Bader charge analysis of the electron distribution was based on the usage of quantum theory of atoms in molecules method provided in VTST (44). The free energy of the H adsorption was calculated asGH=E(H*)+ZPE(H*)TS(H*)E(*)12[E(H2)+ZPE(H2)TS(H2)](1)where asterisk means adsorption on the graphene substrate. The COHP analyses were performed by LOBSTER program [@Maintz16JCC37] (45).


Supplementary material for this article is available at

This is an open-access article distributed under the terms of the Creative Commons Attribution-NonCommercial license, which permits use, distribution, and reproduction in any medium, so long as the resultant use is not for commercial advantage and provided the original work is properly cited.


Acknowledgments: We thank B. Chen from Rice University for assistance with XPS spectroscopy. We also acknowledge H. Fei at Hunan University and J. M. Tour from Rice University for valuable scientific discussion. We would also like to acknowledge J. Zhang from the Scientific Instrument Center at Shanxi University for her help with ICP-MS measurement. Funding: We acknowledge the National Natural Science Foundation of China (no. 21603129), Foundation of State Key Laboratory of Coal Conversion (grant no. J18-19-903), the Fund for Shanxi “1331” Project Key Innovative Research Team (1331KIRT, no. TD201704) the Program for the Outstanding Innovative Teams of Higher Learning Institutions of Shanxi (OIT), the Fund for Shanxi “1331” Project for Featured Discipline of Chemistry in SXNU for finance support of this research. The authors thank the support from Analytical Instrumentation Center (no. SPST-AIC10112914), SPST, ShanghaiTech University. Y.L. acknowledges the support by the National Science Foundation (no. 1900039) and the Welch Foundation (no. F-1959-20180324). Author contributions: Y.Y., Y.Q., and X.F. conceived the experiment and scientific discussions. Y.Y. carried out the syntheses and electrocatalysis measurements. H.L., Z.Z., B.Z., Juncai Dong, W.Y., Jing Dong, and L.F. carried out the characterizations. R.F., J.Z., D.D., and P.Z. carried out the characterizations and provided the analysis. Y.L. and Y.M. carried out the computational investigation and provided the theoretical analysis. G.Y., Y.L., and X.Z. participated in the preparation of the manuscript. All authors discussed the results and revised the paper. Competing interests: The authors declare that they have no competing interests. Data and materials availability: All data needed to evaluate the conclusions in the paper are present in the paper and/or the Supplementary Materials. Additional data related to this paper may be requested from the authors.

Stay Connected to Science Advances

Navigate This Article