Characterization of few-layer 1T-MoSe₂ and its superior performance in the visible-light induced hydrogen evolution reaction

Artificial photosynthesis has been recognized as a potential means of water splitting. Various strategies which can be employed for this purpose like dye sensitization of semiconductors or use of light-harvesting semiconducting nanostructures. TiO₂ was the first material to be used as the photocatalyst for water-splitting.¹ Since then several inorganic catalysts have been used for photocatalytic, photoelectrochemical, and electrocatalytic production of H₂ from water. Semiconducting oxide nanoparticles are one of the most common photocatalysts used for this purpose^2,3 and are preferred because they are chemically robust and stable against photocorrosion during water splitting. The intrinsic limitation of oxides is that they generally have a highly positive valence band (O 2p), making it difficult to find a material which has both sufficiently negative conduction band to reduce H₂O to H₂ along with a sufficiently small bandgap to absorb visible light.^4,5 Metal sulfides and selenides, on the other hand, have less positive valence bands making them visible light active. Majority of the metal sulfides however undergo photocorrosion during the hydrogen evolution reaction (HER).

Transition metal dichalcogenides of lamellar structure have gained attention recently because of their interesting electronic properties and easy availability.^6,7 Exfoliation of these materials into single or few-layers often brings about drastic changes in the electronic structure as compared to the bulk species. Dichalcogenides of MoS₂ and WS₂ generally occur in the 2H form with the trigonal prismatic arrangement of hexagonal S–M–S (M = Mo/W) triple layers are among the most studied of the layered metal chalcogenides. While the 2H forms of these metal dichalcogenides are semiconducting, the 1T forms are metallic.^8–10 MoS₂ has been widely used as a catalyst for electrochemical, photoelectrochemical, and photocatalytic H₂ generation from water^11–14 in consequence of having the conduction band minimum well above the H₂O reduction potential.^15–17 Nanoparticles of 2H-MoS₂ as well as composites of 2H-MoS₂ with graphene and other materials have been employed as catalysts yielding 0.05–10 mmol h⁻¹ g⁻¹ of H₂ with a turn over frequency anywhere between 0.2 and 6 h⁻¹.^13,18–21 It has been shown recently that the composite of MoS₂ with heavily nitrogen-doped graphene has an activity of 10.8 mmol h⁻¹ g⁻¹ and a turn over frequency of 2.9 h⁻¹ under a 100 W halogen lamp.²¹ This was further improved by using 1T-MoS₂ prepared by the exfoliation of bulk MoS₂ by Li-intercalation. The metallic nature of 1T MoS₂ is expected to be responsible for H₂ evolution.^21–23 2H-MoSe₂ with an indirect bandgap of 1.05 eV has its conduction band minimum 0.37 eV higher than 2H-MoS₂, and well above the water reduction potential,^15,16 thereby making it an ideal catalyst for H₂ evolution. The 1T form of MoSe₂ is also metallic and could be expected to be a better catalyst than its 2H analogue for water reduction. Based on these findings, we have exfoliated bulk 2H-MoSe₂ after Li-intercalation to obtain few-layer 1T-MoSe₂ with the octahedral co-ordination of Mo²² and employed it for HER by under visible light. We find that 1T-MoSe₂ is superior to 2H-MoSe₂ as well as 1T-MoS₂.

Bulk MoSe₂ in the 2H form was intercalated with lithium using n-butyl lithium and exfoliation carried out by reacting the intercalated product with water.^24–26 The high resolution transmission electron microscope image in Fig. 1(a) shows that the exfoliated sample corresponds to the 1T phase with an octahedral (O_h) or trigonal antiprismatic symmetry. The 1T form has the √3a × a arrangement which is related to its electronic structure.^25,27 The shifting of the atoms from their equilibrium positions, probably arises because of the Jahn-Teller instability, resulting in chain clusterization of the metal atoms with the formation of a superlattice.²⁸ This distorted phase is stable as a dispersion in water even after Li is removed, but restacks when dried,²¹ to transform into thermodynamically stable 2H phase. The Mo atoms in the 2H form of MoSe₂ have trigonal prismatic coordination as is evident from the high resolution TEM image in Fig. 1(b). The packing of atoms in 2H MoSe₂ is AbA type while in the 1T form it is AbC type. The point group of the trigonal prismatic 2H-MoSe₂ is D_3h while the 1T polytype belongs to the D_3d point group.²⁴ The 1T phases of both MoSe₂ and MoS₂ exhibit a Raman spectra which are distinctly different from those of the 2H-phases. In Fig. 2, we show the Raman spectra of the 1T phases of MoSe₂ and MoS₂ and compare them with the spectra of the 2H phases. We list the band positions of these phases in Table I.

Bulk (2H) MoSe₂ has d² electronic configuration and hexagonal (D_3h) symmetry which would induce splitting of the 4d orbitals into three orbitals of closely spaced energies (shown in Fig. 3(a) (i)). The Mo 4d_z² level is occupied and spin paired forming the valence band minimum (VBM), while the other four orbitals form the empty conduction band. During lithium intercalation, a structural re-orientation occurs to the stable half-filled d-orbital configuration in the octahedral geometry (D_3d) as shown in Fig. 3(a) (ii) which is responsible for the Mo atoms to go from the prismatic co-ordination to the anti-prismatic co-ordination in the 1T-form. Addition of water causes an exothermic reaction because of which the MoSe₂ sheets get separated and remain in this metastable 1T-form. The crystal-field splitting of Mo 4d under the octahedral O_h field generates two set of degenerate orbitals as shown in Fig. 3(a) (iii). The incompletely filled d_yx, d_zx, d_zy orbital gives rise to the metallic properties of 1T-MoSe₂. The Fermi level in 1T-MoSe₂ therefore lies in the Mo 4d making it metallic. Based on the electronic configuration of 1T and 2H phases of MoSe₂ it is clear that when an extra electron is added to 2H-MoSe₂, it resides in the degenerate d_yx, d_x²_-y² states and destabilizes the lattice, while in case of 1T-MoSe₂ the extra electron induces half-filled configuration of d_yx, d_zx, d_zy and increases the stability of the 1T phase.

The hydrogen evolution activity of 1T-MoSe₂ was studied using Eosin Y as the sensitizer and triethanolamine as the sacrificial electron donor. The reaction of dye-sensitized H₂ evolution over MoSe₂ involves photosensitization of Eosin Y followed by formation of Eosin Y anion (EY⁻). EY⁻ being highly reactive donates this electron to MoSe₂, which then catalyzes the reduction of proton to H₂ as shown in Fig. 3(b). Fig. 4 shows the time course of hydrogen evolution by 1T MoSe₂. The yields are in range of the 60–75 mmol g⁻¹ h⁻¹ and remains the same at least up to 5 cycles, i.e., 30 h, with 0.014 mM of dye being added after each cycle (see Fig. S2 of the supplementary material).²⁹ The turn over frequencies (TOF) are in the range 15–19 h⁻¹. The catalytic activity of the 1T form of MoSe₂ is nearly few hundred times higher than that of the 2H form (see inset of 4(a)). It is noteworthy that the yield of H₂ and TOF with 1T-MoSe₂ is superior even to those found with 1T MoS₂. The 2H form of MoSe₂ too shows better activity than that of 2H-MoS₂ (yield of 0.05 mmol g⁻¹ h⁻¹ and TOF of 0.008 h⁻¹). In Table II, we have compared the hydrogen evolution activity and TOF of different transition metal chalcogenides. 1T-MoSe₂ shows higher activity compared to these transition metal chalcogenides and their composites. 1T forms of other metallic transition metal sulfides like TaS₂ and TiS₂ have earlier been shown to be active co-catalysts for H₂ evolution.³⁰ However, the activity for H₂ evolution is much lower in these chalcogenides as compared to 1T-MoS₂²¹ and 1T-MoSe₂. Greater stability afforded by the extra electron to Mo 4d level by inducing a half-filled configuration of d_yx, d_zx, d_zy, as compared to Ta 5d (with 1 electron) and Ti 3d (with no electron) is probably the reason for higher activity of 1T-MoS₂ and 1T-MoSe₂.

It is worthwhile to note that MoSe₂ shows H₂ evolution on sensitization with dye. It may therefore not be considered to be a photocatalyst similar to TiO₂, where photo-excited electrons directly reduce H₂O. MoSe₂ is still different from co-catalysts like Pt which show H₂ evolution in the presence of another catalyst like TiO₂ and cannot evolve H₂ from water even on sensitization with a dye. Conduction Band Minimum of MoSe₂ is favourable for HER but cannot generate enough photoelectrons by absorption of visible light (the bandgap being very small). The dye donates an electron to MoSe₂ and induces the H₂ evolution reaction. We must point out that the exfoliated sample on restacking loses activity due to partial conversion to 2H-form. On annealing it transforms completely to the 2H form, thereby losing its hydrogen evolution activity (see Fig S3 of the supplementary material).²⁹ 

In order to understand the higher activity of 1T-MoSe₂ in comparison to 2H-MoSe₂ and the 1T and 2H forms of MoS₂, we have carried out first-principles calculations based on density functional theory as implemented in Quantum ESPRESSO package,³¹ in which the ionic and core-valence electron interactions are modeled with ultrasoft pseudopotentials.³² The exchange-correlation energy of electrons is treated within a Generalized Gradient Approximation (GGA) functional as parametrized by Perdew, Burke, and Ernzerhof.³³ We use an energy cutoff of 35 Ry to truncate the plane wave basis used in representing the Kohn-Sham wave functions, and an energy cutoff of 280 Ry to represent the charge density. Structures are relaxed till the Hellman-Feynman forces on each atom are less than 0.02 eV/Å. We have used a periodic supercell geometry to simulate a 2D sheet, including vacuum of 15 Å to separate the adjacent periodic images of the sheet. For self-consistent Kohn-Sham (scf) calculations, configurations of √3 × √3 and √3 × 1 supercells, the BZ integrations are sampled over uniform meshes of 7 × 7 × 1 and 12 × 7 × 1 k-points, respectively. Since KS-DFT typically underestimates electronic bandgaps (a known limitation), we employ hybrid functional based on Hartree-Fock-Exchange (HSE)³⁴ with screened Coulomb potential to estimate the bandgaps more accurately. The calculations were based on first-principles DFT using Projector Augmented Wave (PAW) method^35,36 as implemented in the VASP (Vienna Ab-initio Simulations Package).³⁷ We have studied two superstructures of 1T-MoX₂ (where X = S and Se), √3 × √3 and √3 × 1²⁶ (Fig. 5). Among the two superstructures, √3 × 1 is metallic and shows dimerization of Mo atoms, and √3 × √3 is semiconducting with trimerized Mo atoms. From phonon dispersion, we find that, both MoS₂ and MoSe₂ are stable in the √3 × √3 and √3 × 1 superstructures. However, MoS₂ is energetically more stable in the √3 × √3 compared to √3 × 1 by 27 meV/f.u., while the √3 × 1 super-structure of MoSe₂ is energetically more stable than the √3 × √3 super-structure by 33 meV/f.u.

Experimentally, MoSe₂ is found to be in the √3 × 1 super-structure, in agreement with our first-principles results. Henceforth, we shall consider the √3 × √3 superstructure for MoS₂ and √3 × 1 superstructure for MoSe₂. To determine the efficiency of MoX₂ in reducing a proton to hydrogen as observed in experiments, we have estimated their electron affinities (EA) and work function (φ). For metallic states, the relevant property here is the work function. The EA is estimated as the difference between the vacuum potential (E_vac) and the lowest energy conduction band (E_CB). Since DFT is a ground state theory, estimation of the bandgap and hence the E_CB is not accurate. Hence, we replace the E_CB with E_VB + E_g, where E_VB is the energy of the highest energy valance band and E_g is the bandgap. Since E_g is grossly underestimated in DFT calculations, we use the HSE corrections (using VASP) to determine E_g accurately. For the monolayered MoS₂, experimental value of bandgap (1.8 eV⁶) is available.

Comparison of the experimental bandgap with calculated bandgap for 2H-MoS₂ reveals that Kohn-Sham bandgap is underestimated by 7.2% and the HSE bandgap is overestimated by 17.7% (see Table III), in agreement with Ahuja et al.³⁸ It is thus clear that the HSE method overestimates the experimental bandgap, whereas the KS-DFT calculation (GGA) yields a better estimate. We use estimates of Eg obtained from KS-DFT calculations in this work. The work function for metals and semiconductors is calculated as φ = E_vac − E_F (where E_F = Fermi energy) and φ = E_vac − E_VB respectively. We find that (a) the 2H and 1T-polytypes of MoS₂ have a greater φ than that of the respective structure of MoSe₂ (refer to Table IV). This implies that it is easier to extract an electron from MoSe₂ compared to that of MoS₂ in both 1T and 2H polytypes. (b) The 1T polytype has a lower φ than that of 2H, which means that it is easier for the 1T to donate electron compared to the 2H-structure. This explains why the 1T-polytype of MoSe₂ produces hydrogen more efficiently than the 2H-polytype as observed in experiments. The electron affinities of both 1T and 2H polytypes indicate that MoS₂ has a stronger electron affinity (indicating a higher tendency to attract electrons) than that of MoSe₂ (refer to Table IV), and the work function is also larger for MoS₂. Thus, though MoS₂ more readily attracts/accepts electrons, it does not donate it that easily. Hence, MoSe₂ is efficient in hydrogen evolution as compared to that of MoS₂ as observed in experiments here.

To connect the hydrogen (H) binding energy to the catalytic activity of the MoX₂ compounds, we have determined the binding energy of hydrogen (H) to √3 × 1 1T superstructure of MoS₂ and MoSe₂. This superstructure is relevant to the experiments reported here, and has the lowest work function.

The hydrogen binding energy is calculated as

where n is the number of H atoms considered in the simulation.

Bulk MoS₂ and MoSe₂ do not absorb hydrogen (E_ads > 0). It has been reported that edges of these dichalcogenides are catalytically active in hydrogen adsorption.²² We have therefore simulated ribbons of MoX₂ with two different types of edges (Mo terminated edge and X terminated edge), and their interaction with H (with 100% H coverage at the edges). The hydrogen binding energies at Mo sites at the edges of MoS₂ and MoSe₂ are −33.8 meV/f.u. and −32.3 meV/f.u., respectively. The respective Mo–H bond lengths are 1.72 Å and 1.73 Å. The hydrogen binding energies at the S/Se edges of MoS₂ and MoSe₂ are −34.6 meV/f.u. and −13.1 meV/f.u. respectively. The corresponding X–H bond lengths are 1.35 Å and 1.48 Å. The binding energy of hydrogen at the metal edge is about the same in the two compounds but the Se edge shows weaker binding with hydrogen than the S edge. According to the volcano plot,^39,40 this suggests a higher exchange current for hydrogen evolution over MoSe₂ compared to MoS_2. This is consistent with our analysis based on the work functions. Since MoSe₂ has a lower work function than MoS₂, its Fermi energy (E_F) lies closer to the normal hydrogen electrode (E_NHE), which allows an easy exchange of an electron with MoSe₂ (H atom has a weaker binding at the Se edge) as compared to MoS₂. Thus, MoSe₂ is more efficient in facilitating the hydrogen evolution reaction.

In conclusion, metallic 1T-MoSe₂ prepared by Li intercalation followed by exfoliation of bulk 2H-MoSe₂ shows excellent H₂ evolution activity in comparison to few-layered semiconducting 2H-MoSe₂. Interestingly, 1T-MoSe₂ shows better H₂ evolution activity than 1T-MoS₂ as well. Our first-principles analysis reveals that MoSe₂ has a lower work function as compared to MoS₂, and that the 1T-structure exhibits lower work function than the 2H-structure for both MoX₂ (X = S, Se). This results in easy transfer of electron from the MoSe₂ for the reduction to hydrogen, and hence MoSe₂ is more efficient for hydrogen evolution reaction compared to MoS₂, and in agreement with the experimental results.