Adsorption and Decomposition of H2S on MgO(100), NiMgO(100

(b) Rodriguez, J. A.; Jirsak, T.; Chaturvedi, S.; Kuhn, M. Surf. .... Raza, H.; Harte, S. P.; Muryn, C. A.; Wincott, P. L.; Thorton, G.; Casanova, R.;...
0 downloads 0 Views 195KB Size
3630

J. Phys. Chem. B 2000, 104, 3630-3638

Adsorption and Decomposition of H2S on MgO(100), NiMgO(100), and ZnO(0001) Surfaces: A First-Principles Density Functional Study Jose A. Rodriguez* Department of Chemistry, BrookhaVen National Laboratory, Upton, New York 11973

Amitesh Maiti Molecular Simulations Inc., 9685 Scranton Road, San Diego, California 92121 ReceiVed: January 4, 2000; In Final Form: January 28, 2000

The adsorption and dissociation of H2S on MgO(100), Ni-doped MgO(100), and ZnO(0001) was studied using first-principles density-functional calculations (DFT-GGA) and periodic supercells. The bonding of H2S and its S-containing dissociated species (HS and S) is substantially stronger on ZnO(0001) than on MgO(100), making dissociation easier on zinc oxide. This behavior can be explained by the smaller ionicity in ZnO, which leads to a larger electron density around the Zn atoms and a larger reactivity toward S-containing molecules. Replacing some of the metal centers of MgO(100) with Ni atoms enhances the binding of S-containing species through new electronic states associated with the Ni 3d levels and located above the occupied {O 2p + Mg 3s} bands. In addition, structural defects, like steps, expose metal centers with lower coordination and larger reactivity than pentacoordinated Mg atoms in MgO(100). A simple model based on perturbation theory and band-orbital mixing is able to explain the differences in the reactivity of MgO(100) and ZnO(0001) and the behavior of other oxides (Al2O3, Cr2O3, Cr3O4, Cu2O) in the presence of sulfurcontaining molecules. The model predicts a negative correlation between the reactivity of the oxides and the size of the electronic band gap, with the chemical activity of an oxide depending mainly on how well its bands mix with the orbitals of H2S. The electrostatic interactions between the Madelung field of the oxide and the dipole moment of the molecule play only a secondary role in bonding.

I. Introduction Sulfur-containing molecules are common impurities in fossilderived fuels and chemical feedstocks.1,2 Today, these impurities constitute a major problem in our industrial society due to their negative effects in environmental pollution and chemicals processing.2-5 The sulfur impurities form sulfur oxides (SOx), which are major air pollutants that lead to acid rain,2 during the burning of fuels. In addition, sulfur impurities rapidly deactivate or poison most metal/oxide catalysts used in the chemical or petrochemical industries and in the control of CO and NOx emissions from automobile exhaust.3-6 Millions of dollars are lost every year as a consequence of sulfur poisoning.4-6 A fundamental understanding of the interaction of sulfurcontaining molecules with oxide surfaces is important for two reasons. First, most commercial catalysts poisoned by sulfur involve oxides, and there is a clear need to improve their lifetime.5,6 And, second, in many industrial operations, oxides are used as sorbents for the removal or destruction of sulfurcontaining molecules.2,7-10 On the surface of a metal oxide, sulfur can interact with a metal or an oxygen atom, producing species with very different electronic properties (“sulfide” versus “sulfate” formation). Experiments for the adsorption of S2, H2S, CH3SH, and thiophene on a series of oxides (Al2O3, ZnO, Cu2O, MoO2, Cr2O3, CeO2) reveal that sulfur species produced by the dissociation of these molecules mainly interact with the metal centers of the surface.3,8a,11,12 On the other hand, SO2 prefer* Corresponding author. FAX: 631-344-5815. E-mail: [email protected].

entially reacts with O centers, readily forming SO3 and SO4 species.8b,9,13,14 For practical reasons, it is very important to establish which types of oxides have a high reactivity toward sulfur-containing molecules. The metal elements form oxides that can adopt a large diversity of crystal structures.15,16 These oxides can behave as semiconductors or insulators, and in some cases even exhibit metallic properties,16 leading to the possibility of large variations in chemical reactivity.11,12 Figure 1 shows the sulfur uptake (HS + S) for the dissociative adsorption of hydrogen sulfide on a series of oxides at 300 K.11,12 Oxides that have a large degree of ionicity like Al2O3 17 and MgO 18 exhibit the lowest reactivities. A qualitative correlation is observed between the chemical activity of an oxide and the size of its band gap: the smaller the band gap in an oxide, the faster the rate of dissociation of hydrogen sulfide. This trend can be explained using a simple model based on the strength of the mixing of the frontier orbitals of the admolecule with the conduction and valence bands of the oxide surface.11,12 But additional factors might also contribute to the behavior seen in Figure 1. In this article, we use first-principles density functional theory (DFT) and periodic supercells to study the bonding interactions of H2S, HS, and S with an insulator, MgO(100), an insulator doped with a transition metal, NiMgO(100), and a semiconductor, ZnO(0001). In previous works the chemistry of hydrogen sulfide on MgO(100)12 and ZnO(0001)19,20 has been examined in detail using core and valence level photoemission. The experimental results show molecular adsorption of H2S on metal centers of MgO at 80 K. The molecule dissociates into HS upon

10.1021/jp000011e CCC: $19.00 © 2000 American Chemical Society Published on Web 03/18/2000

Adsorption and Decomposition of H2S on Oxides

Figure 1. Total coverage of sulfur (HS + S) as a function of H2S exposure to a series of oxides (Al2O3, MgO, Cr2O3, ZnO, Cu2O, Cr3O4) at 300 K.11,12 The numbers in parentheses denote the band gap of each oxide.

heating to 300 K, and further heating to 400 K leaves atomic S bonded to Mg on the oxide surface.12 Zinc oxide is able to dissociate hydrogen sulfide at very low temperatures.19,20 Adsorption of the molecule on this oxide at ∼100 K produces only Zn-bonded HS species. These species decompose into S adatoms at temperatures between 300 and 400 K. The reaction between hydrogen sulfide and zinc oxide at elevated temperatures, 400-500 K, produces water and zinc sulfide: H2S(gas) + ZnO(solid) f H2O(gas) + ZnS(solid).21

J. Phys. Chem. B, Vol. 104, No. 15, 2000 3631 were chosen high enough in order to ensure convergence of the computed structures and energetics. Since the DFT calculations were performed at the GGA level, one can expect that they did not overestimate the bonding energies of adsorbates on the oxide surfaces.32-35 Frequently, DFT-GGA calculations predict adsorption energies within an accuracy of 5 kcal/ mol.24,35,36 In any case, in this work our main interest is in qualitative trends in the energetics, not in absolute values. Due to the delocalized (plane wave) nature of its basis set, CASTEP yields useful electronic information (levels and band structure) only in the k-space. To investigate localized charges and localized electronic density of states around various atoms of interest, we used another DFT program from Molecular Simulations: DMol3.23 In contrast to CASTEP, DMol3 uses localized functions to describe the atomic orbitals. Our calculations employed numerical basis sets of double-ζ quality plus polarization functions to describe the valence orbitals of O, Mg, Ni, Zn, S, and H. DFT in DMol3 was performed within the GGA approximation using Becke-88 for exchange37 and Perdew-91 for correlation.38 The charge distributions were estimated using the approach proposed by Mulliken.39,40 They are useful in obtaining trends and providing simple interpretation of results, but the charges must not be interpreted in absolute quantitative terms because of the uncertainty in uniquely defining a charge-partitioning scheme.41 The Mulliken method has well-known shortcomings but is one of the most popular procedures for electron population analysis. To model the MgO(100), NiMgO(100), and ZnO(0001) surfaces, we used periodic supercells. First, the bulk MgO and ZnO crystals were geometrically optimized with CASTEP in order to determine the equilibrium atomic positions and the lattice constants. After the optimization of the bulk crystals, the appropriate surfaces (i.e. the (100) face of MgO and the (0001) face of ZnO) were cleaved, followed by the construction of a three-dimensionally periodic supercell with a vacuum of 12.5 Å on the top of the free surface. The slab models were 4-6 atomic layers in depth normal to the surface plane. The bottom 2-4 layers were frozen at the bulk crystalline spacing in order to mimic the presence of a semiinfinite crystalline material beneath the surface. The number of layers and the amount of vacuum chosen were expected to be sufficiently large to yield accurate geometries and bonding energies at the oxide surfaces.18,24,26,42

II. Theoretical Methods All the calculations reported in section III below were performed using commercial versions of the density functional programs CASTEP22 and DMol323 available from Molecular Simulations Inc. Geometries and bonding energies of sulfurcontaining species on MgO(100), NiMgO(100), and ZnO(0001) were estimated with the CASTEP code. A large body of existing work indicates that CASTEP is excellent for predicting structural geometries and energy changes associated with chemical transformations involving oxides.24-28 In this code, the wave functions of valence electrons are expanded in a basis set of plane waves with kinetic energy smaller than a specified cutoff energy Ecut. The presence of tightly bound core electrons is represented by nonlocal ultrasoft pseudopotentials.29 Reciprocalspace integration over the Brillouin zone is approximated through a careful sampling at a finite number of k-points using the Monkhorst-Pack scheme.30 The exchange-correlation contribution to the total electronic energy is treated in the generalized gradient corrected (GGA)31 form of the local density approximation (LDA). In all calculations, the kinetic energy cutoff Ecut and the density of the Monkhorst-Pack k-point mesh

III. Results III.1. Adsorption of H2S, HS, and S on MgO(100). The flat (100) face of MgO was modeled using a periodic slab containing four layers of Mg and O atoms arranged in the way shown in Figure 2. Previous works have shown that slabs of 3-4 layers provide a very good representation of the MgO(100) surface.18,26,42 The geometry optimization of bulk MgO gave a typical rock-salt structure with ao ) 4.26 Å. This value compares well to those derived from experimental measurements (4.22 Å)15,43 and other theoretical calculations.26,42 In the slab calculations, the structural geometry of the clean or adsorbate-covered MgO(100) was determined by relaxing the two layers near the surface and keeping the other two layers of the slab in the geometry of bulk MgO. The CASTEP calculations predict almost no reconstruction of the clean MgO(100) surface with respect to the (100) face of bulk MgO, in agreement with several experimental16 and theoretical42 studies. Bulk MgO is known to be a highly ionic compound.17,18 Table 1 lists the Mulliken charges calculated with DMol3 for magnesium sites in a MgO(100) surface and metal centers in

3632 J. Phys. Chem. B, Vol. 104, No. 15, 2000

Rodriguez and Maiti TABLE 2: DFT-GGA Results for the Adsorption of H2S, SH, and S on MgO(100) species

Figure 2. Schematic view of the four-layer slab used to model a flat MgO(100) surface. The Mg and O atoms are represented by dark and gray octagons, respectively. Each layer of the slab contains the same number of Mg and O atoms. The sulfur-containing molecules were adsorbed only on the top layer of the slab, and the geometry of the first two layers was optimized in the DFT calculations.

TABLE 1: Estimated Charges on Metal Sites metal centera

Mulliken chargeb

metal centera

Mulliken chargeb

Mg-6 in bulk MgO Mg-5 in MgO(100) Mg-3 defect in MgO(100)

1.22 1.14 0.86

Ni-5 in MgO(100) Zn-4 in bulk ZnO Zn-3 in ZnO(0001)

0.81 0.93 0.75

a The number close to the element name indicates its coordination number. b Here, our interest is in qualitative trends, no absolute values (see section II).

ads energy (kcal/mol)

bond length (Å) Mg-S S-Ha

On Flat Surface (Mg-5 Site) S (0.25 monolayer)b 34 2.46 S (0.50 monolayer) 31 2.48 HS (0.25 monolayer) 19 2.63 HS (0.50 monolayer) 15 2.64 H2S (0.25 monolayer) 9 2.71 H2S (0.50 monolayer) 8 2.73

1.36 1.36 1.35 1.35

On Step of Surface (Mg-3 Site) S (0.25 monolayer) 48 2.34 HS (0.25 monolayer) 27 2.49 H2S (0.25 monolayer) 13 2.65

1.36 1.36

a

For the free H2S and HS molecules the S-H distances were equal to 1.35 Å. b The numbers in parentheses denote the coverage of the adsorbate. One-quarter (0.25 monolayer) or half (0.5 monolayer) of the metal sites exposed in a MgO(100) surface are covered by the sulfurcontaining species.

and Mg26O25 clusters.12 The adsorption energies predicted for atomic sulfur (31-34 kcal/mol) are consistent with the behavior seen at an experimental level for the S/MgO(100) system, where the S adatoms remain on the surface at temperatures well above 500 K.12 As seen from Table 2, atomic S and HS form strong bonds on the perfect MgO(100) surface, but the bonding interactions of the H2S molecule are relatively weak (∼8-9 kcal/mol). For the dissociative chemisorption of H2S on two adjacent Mg and O sites of the flat oxide surface,

H2Sgas + Mgsurface + Osurface f HS-Mgsurface + H-Osurface (2)

Figure 3. Bonding geometries for H2S, HS, and S on the oxide surfaces. The sulfur-containing species were adsorbed on top of a Mg, Ni, or Zn atom (empty circle in figure) of the substrate. The molecular axes of H2S and SH were tilted with respect to the surface normal. The metal-S and S-H bond distances, plus the metal-S-H bond angles were optimized in the DFT calculations.

the bulk oxide. For the pentacoordinated Mg atoms in the surface (Mg-5 in our notation) the positive charges are close to those seen in hexacoordinated Mg atoms in the bulk (Mg-6 in our notation). A similar fact has been noted in previous theoretical studies that use periodic slabs to model the MgO(100) surface.18 H2S, HS, and S were adsorbed on top of metal centers of MgO(100) with the bonding configurations shown in Figure 3. These bonding configurations are typical for these species on metal oxides.11,12,44 Table 2 displays adsorption geometries and energies predicted by CASTEP. Following the usual convention,24 adsorption energies (Eads) were calculated according to the expression

Eads ) Eadsorbate + Eslab - E(adsorbate+slab)

(1)

where Eadsorbate is the energy of the isolated adsorbate in its equilibrium configuration, Eslab is the total energy of the bare slab, and E(adsorbate+slab) is the total energy of the adsorbate/slab system. Positive adsorption energies denote an exothermic adsorption process. In Table 2, the bonding energies of the adsorbates increase in the following sequence: H2S < SH < S. The listed values are similar to those obtained in DFT-GGA calculations for the adsorption of these species on large Mg18O17

we calculate that the energy released in the reaction is close to 14 kcal/mol (θH2S ) 0.25 monolayer). Thus, dissociation of the molecule is an energetically favorable process, but it may not be easy due to the lack of strong interactions between H2S and the flat oxide surface. The decomposition of H2S in reaction 2 is helped by the formation of O-H bonds on the oxide surface. For the reactions

H2Sgas + Mgsurface f HS-Mgsurface + 0.5H2,gas

(3)

H2Sgas + Mgsurface f S-Mgsurface + H2,gas

(4)

the CASTEP calculations predict ∆E values (θHS or θS ) 0.25 monolayer) of +4 and +19 kcal/mol, respectively, which mean that one has to put a lot of energy in the system to induce the dissociation of H2S in this way (i.e. no H-Osurface bond formation). Recent studies highlight the importance of defect sites on the chemistry of the MgO(100) surface.13,45,46 We studied the adsorption of the sulfur-containing species on an infinite slab of the type shown in Figure 4 that contains flat regions of the (100) face and steps with Mg atoms coordinated to only three oxygen atoms. The geometry of the top two layers of the slab was optimized during the CASTEP calculations. In the adsorbatefree slab, the tricoordinated magnesium atoms (Mg-3 in our notation) had two “short” Mg-O distances of 2.09 Å and one “long” Mg-O distance of 2.14 Å. In contrast, pentacoordinated magnesium in a flat MgO(100) surface exhibited Mg-O distances of 2.13 (oxygen underneath) and 2.14 Å (oxygen neighbors in the surface plane). For the Mg-3 sites, DMol3 predicts a positive charge that is substantially smaller than that of Mg-5 sites in perfect MgO(100)

Adsorption and Decomposition of H2S on Oxides

J. Phys. Chem. B, Vol. 104, No. 15, 2000 3633

Figure 4. Schematic view of the four-layer slab used to model step sites in a MgO(100) surface. The Mg and O atoms are represented by dark and gray octagons, respectively. In the first layer, the Mg atoms have only three nearest neighbors (two oxygen atoms in the same layer and one oxygen atom underneath).

Figure 5. Schematic view of the four-layer slab used to model a flat NiMgO(100) surface (θNi ) 0.25 monolayer). Only one Ni atom is shown in the figure. The Mg and O atoms are represented by dark and gray octagons, respectively. The sulfur-containing molecules were adsorbed only on top of the nickel atoms, and the geometry of the first two layers of the slab was optimized in the DFT calculations.

or Mg-6 sites in bulk MgO (see Table 1). The extra electron density on the Mg-3 sites probably leads to a higher chemical activity. For the interaction of the sulfur-containing species with the Mg-3 sites (a-top adsorption), CASTEP predicts bonding energies that are significantly larger than those for adsorption on Mg-5 sites (see Table 2). This is particularly important in the case of H2S, because in a kinetic process it will increase the residence time of the molecule on the surface and enhance the probability for dissociation. III.2. Adsorption of H2S, HS, and S on NiMgO(100). In this section we investigate the bonding of H2S, HS, and S to NiMgO(100) surfaces generated by replacing some of the magnesium atoms in a MgO(100) surface with nickel atoms (see Figure 5). By doing so one can transform magnesium oxide from an insulator into a semiconductor.16,47 Compounds of the NixMg1-xO type are stable,47 and it is known that transitionmetal atoms bond to magnesium vacancies present in MgO(100).16,48 A nickel atom in a matrix of magnesium oxide can be seen as a cation in NiO, a compound that also adopts a rock-salt structure like MgO.15,16 Our objective here is to compare the behavior of magnesium and a transition-metal when both elements are bonded to the same number of oxygen atoms and are in a similar chemical environment. The results of geometry optimization with CASTEP indicate that in NiMgO(100) surfaces with nickel coverages of 0.250.50 monolayer, the Ni atoms shift 0.1-0.15 Å upward with respect to the plane of the Mg and O atoms. This can be

Figure 6. Calculated electron-density plots for MgO(100), top, and NiMoO(100), bottom, surfaces. The plots were obtained using CASTEP, and for simplicity only a few metal and oxygen atoms in the fourlayer slabs are shown.

attributed to the fact that the Ni atoms have a smaller positive charge (i.e. larger electron density) and, therefore, effectively a bigger size than the Mg atoms. In Table 1, the results of DMol3 indicate that a Ni atom pentacoordinated in a matrix of MgO(100), Ni-5 in our notation (θNi ) 0.25 monolayer), has a positive charge clearly smaller than that of a Mg-5 atom in pure MgO(100). Figure 6 shows plots for the electron density around Mg-5 and Ni-5 atoms in MgO(100) and NiMgO(100) surfaces, respectively. In the case of pure MgO(100), the total electron density on the magnesium atoms is small and the electrondensity plot is dominated by maxima located on top of the oxygen atoms. This is not the case for NiMgO(100), where one can see a large electron density around the nickel atoms. As we will see below, this has an effect on the chemical reactivity of the metal centers. Figure 7 displays calculated density-of-states (DOS) plots for the systems in Figures 2 and 5. The graphs were obtained using CASTEP and include only occupied states. For MgO(100), the O 2s levels are located from -18 to -16 eV and states that contain a mixture of O 2p (main component) and Mg 3s character appear between -5 and -1.5 eV. These types of features are also seen in the DOS plot for NiMgO(100), and in addition two peaks appear near the top of the valence band that contain Ni 3d character. There is a splitting of the 3d orbitals of the nickel atoms due to interactions with the ligand field generated by the surrounding oxygen neighbors.16,49 The Ni 3d features appear at higher energy than any feature in the MgO(100) system. When comparing the properties of the Mg-5 and Ni-5 atoms in these systems, one finds that the latter have

3634 J. Phys. Chem. B, Vol. 104, No. 15, 2000

Rodriguez and Maiti

Figure 8. Section of the six-layer slab used to model the ZnO(0001) surface. The Zn and O atoms are denoted by dark and gray octagons, respectively. In the slab, the first, third, and fifth layers contained only zinc atoms. The second, fourth, and sixth were oxygen layers. The sulfur-containing species were adsorbed only on the top layer, and the geometry of the first two layers in the slab was relaxed during the DFT calculations.

Figure 7. Total DOS for the occupied bands of the MgO(100) and NiMgO(100) (θNi ) 0.5 monolayer) slabs. The results were obtained using CASTEP. The reference for the “0” of energy is not the vacuum level.22

TABLE 3: DFT-GGA Results for the Adsorption of H2S, SH, and S on NiMgO(100) species

ads energy (kcal/mol)

S (0.25 monolayer)b S (0.50 monolayer) HS (0.25 monolayer) HS (0.50 monolayer) H2S (0.25 monolayer) H2S (0.50 monolayer

59 55 33 32 17 15

bond length (Å) Ni-S S-Ha 2.20 2.23 2.38 2.37 2.52 2.55

1.40 1.39 1.39 1.39

a For the free H S and HS molecules the S-H distances were equal 2 to 1.35 Å. b The numbers in parentheses denote the coverage of the adsorbate and Ni in the NiMgO(100) system. One-quarter (0.25 monolayer) or half (0.50 monolayer) of the Mg atoms in a MgO(100) surface were replaced by Ni atoms.

a larger local density of states and that many of their states are less stable in energy than the states in the Mg-5 atoms. The electronic differences observed in Figures 6 and 7 suggest that a NiMgO(100) surface is better suited to respond to the presence of sulfur-containing molecules than a MgO(100) surface. Table 3 lists structural parameters and bonding energies predicted by CASTEP for the adsorption of H2S, HS, and S on top of Ni sites of NiMgO(100). Again atomic S and HS form strong bonds on the oxide surface, but now the bonding interactions of H2S become substantial (15-17 kcal/mol). The Ni-5 sites are more chemically active than the Mg-5 or Mg-3 sites in Table 2. On top of the transition metal, one sees not only stronger metal-adsorbate bonds but also a larger elongation of the H-S bonds. Both effects should enhance the rate of dissociation of H2S and HS with respect to that seen in the H2S/ MgO(100) system. III.3. Adsorption of H2S, HS, and S on ZnO(0001). As an industrial sorbent, zinc oxide is highly efficient for the trapping of H2S, SO2, disulfides, and mercaptans.6,7,8a,9,10 Previous theoretical studies have shown that the use of infinite slabs is a valid and reliable approach for studying adsorption reactions on surfaces of zinc oxide.28,50,51 In our DFT calculations the ZnO(0001) surface was modeled using an infinite slab contain-

ing three layers of zinc and three layers of oxygen atoms, arranged as shown in Figure 8. The top layer represents the (0001)-Zn face of zinc oxide. The geometry optimization for bulk ZnO gave a wurtzite hexagonal unit cell with a ) b ) 3.29 Å and c ) 5.25 Å (CASTEP results). These values are very close to those derived from experimental measurements (a ) b ) 3.25 Å and c ) 5.21 Å at 298 K)52 and other theoretical calculations (a ) b ) 3.29 Å and c ) 5.24 Å).50 In the slab calculations, the structural geometry of the two layers near the surface (one Zn layer and one O layer) was relaxed, while the other four layers of the slab were kept in the geometry of bulk ZnO. In the case of clean ZnO(0001), the Zn outermost layer moved closer to the first oxygen layer by 0.23 Å (with respect to the separation observed for the (0001) plane of bulk ZnO). A similar relaxation (0.20-0.25 Å) has also been detected in LEED I-V experiments.16,53 This relaxation was partially removed by adsorption of the sulfur-containing species. Table 1 shows the Mulliken charges calculated with DMol3 for tricoordinated zinc sites in a ZnO(0001) surface, Zn-3 in our notation, and tretracoordinated metal centers in bulk ZnO, Zn-4 in our notation. In DFT calculations for large zinc oxide clusters, Mulliken charges of 0.81-0.97e have been reported for the zinc cations.44a The results in Table 1 indicate that the degree of ionicity in ZnO is significantly smaller than in MgO. This is commonly accepted: MgO is highly ionic,17,18 whereas the bonds in ZnO involve a large degree of covalency.16,54 Figure 9 shows a plot for the electron density of a ZnO(0001) surface. Clear maxima are seen on top of the zinc centers. This is very different from the result displayed in Figure 6 for pure MgO(100), where the electron density on the metal centers is low. Thus, as in the case of NiMgO(100), ZnO(0001) is better suited to respond to the presence of sulfur-containing molecules than MgO(100). Table 4 lists the CASTEP results for the adsorption of H2S, HS, and S on ZnO(0001). The calculated adsorption energy for H2S is 18-21 kcal/mol. DFT calculations at the LDA level predict a bonding energy of 32 kcal/mol for H2S on Zn atoms of zinc oxide clusters.44b The discrepancy with respect to the GGA values in Table 4 can be attributed to the fact that LDA calculations are known to overestimate chemisorption energies.24,32,33,55 Calculations carried out with CASTEP at the LDA level give an adsorption energy of 34 kcal/mol for H2S on ZnO(0001) (θH2S ) 0.25 monolayer). The values shown in Table 4 for the bonding energy of atomic S (59-64 kcal/mol) are consistent with the behavior found at an experimental level for

Adsorption and Decomposition of H2S on Oxides

J. Phys. Chem. B, Vol. 104, No. 15, 2000 3635

Figure 9. Electron-density plot for the ZnO(0001) surface calculated with CASTEP. For simplicity only a few metal and oxygen atoms in the six-layer slab are shown.

TABLE 4: DFT-GGA Results for the Adsorption of H2S, SH, and S on ZnO(0001) species

ads energy (kcal/mol)

S (0.25 monolayer)b S (0.50 monolayer) HS (0.25 monolayer) HS (0.50 monolayer) H2S (0.25 monolayer) H2S (0.25 monolayer)

64 59 38 35 21 18

bond length (Å) Zn-S S-Ha 2.27 2.29 2.40 2.41 2.49 2.52

1.40 1.39 1.41 1.39

a For the free H2S and HS molecules the S-H distances were equal to 1.35 Å. b The numbers in parentheses denote the coverage of the adsorbate. One-quarter (0.25 monolayer) or half (0.50 monolayer) of the metal sites exposed in a ZnO(0001) surface are covered by the sulfur-containing species.

the S/ZnO20,56 and S/ZnO(0001)57 systems, in which the S adatoms desorb from the surface at temperatures above 700 K. For the reactions

H2Sgas + Znsurface f HS-Znsurface + 0.5H2,gas

(5)

H2Sgas + Znsurface f S-Znsurface + H2,gas

(6)

the CASTEP calculations give ∆E values (θHS or θS ) 0.25 monolayer) of -15 and -11 kcal/mol, respectively. Even more exothermic decomposition reactions can be expected if H-Osurface bonds are formed on the oxide surface. On a ZnO(0001) sample this will involve O at defect sites, but in systems such as ZnO(101h0) or polycrystalline ZnO the formation of H-Osurface bonds should be easy due to the large amount of oxygen atoms exposed on these surfaces. After comparing the adsorption energies in Tables 2 and 4 (see Figure 10), it is clear that ZnO(0001) should be a better sorbent for H2S than MgO(100). And indeed, the experimental results in Figure 1 show that at 300 K the rate of dissociation of H2S on zinc oxide is larger than on magnesium oxide. IV. Discussion The bonding mechanism of hydrogen sulfide on metallic surfaces involves a transfer of electrons from the H2S(5a1,2b1) orbitals into unoccupied orbitals of the metal (σ and π donation) and a charge transfer from the metal into the vacant H2S(3b2,6a1) orbitals (σ-back-donation).58 On metal centers of an oxide, the bonding is more complex.16 In addition to the mixing of the frontier orbitals of the adsorbate with the conduction and valence bands of the oxide, the bonding also contains contributions from the interactions between the dipole of the admolecule (sulfur

Figure 10. Calculated adsorption energies (CASTEP, DFT-GGA) for H2S and HS on MgO(100), white bars, and ZnO(0001), hatched bars. In the case of magnesium oxide, results are shown for adsorption on penta- and tri-coordinated metal centers (Mg-5 and Mg-3, respectively).

negative-end down, hydrogens positive-end up)58 and the electrostatic field generated by the charges in the oxide.16 The electrostatic interactions are expected to be particularly strong on oxides that have a large degree of ionicity such as Al2O3 17 and MgO.18 Interestingly, these oxides exhibit the lowest reactivities in Figure 1. For the systems in Figure 10, on the other hand, the smaller the positive charge on the metal center (Table 1), the larger its bonding ability (Tables 2 and 4). Thus, it is very difficult to explain the trends in Figures 1 and 10 in terms of electrostatic bonding between H2S and the oxides. Clearly, the chemical activity of an oxide probably depends on how well its bands mix with the orbitals of H2S, and the electrostatic interactions with the dipole moment of H2S play only a secondary role in bonding. Figure 11 shows the energy positions for the valence and conduction band of a series of bulk oxides (Al2O3, MgO, ZnO, Cu2O). To generate this figure, we used values reported in the literature for the electronic properties of the oxides,11,12,16,59 instead of results of DFT calculations. Note that for the CASTEP results in Figure 7, the “0” of energy is not the vacuum level.22 In addition, the DFT calculations do predict that the band gap in ZnO is smaller than in MgO but the magnitude of the band gaps is underestimated by 2-2.5 eV. This is a common problem in DFT calculations.60 For the oxides in Figure 11, when the band gap increases the valence band moves toward higher binding energy, while at the same time there is a decrease in the stability of the conduction band. Therefore, simple models based on band-orbital mixing61-63 would predict that the smaller the band gap in an oxide, the larger its chemical reactivity. For H2S, thiols (RSH) and thiophenes, the HOMOs appear at energies between -9 and -12 eV, with the LUMOs usually located in the range of -1 to +5 eV.8a,11,58,64 Following perturbation theory in combination with the Hu¨ckel and tightbinding methods,61-63 one can get an approximate expression for the bonding energy (Q) derived from the interaction between the HOMO of a sulfur-containing molecule and the conduction band of an oxide:

QHOMO-conduction ∝ (βHOMO-conduction)2/(Econduction - EHOMO) (7) where βHOMO-conduction is the resonance integral for the interacting levels, and Econduction and EHOMO are the energies for the centroid of the conduction band and HOMO, respectively. The corresponding expression for the bonding energy that arises from

3636 J. Phys. Chem. B, Vol. 104, No. 15, 2000

Figure 11. Energy positions for the valence and conduction bands of bulk Al2O3, MgO, ZnO, and Cu2O.11,12,16,59 The empty and occupied states are indicated by dotted and solid lines, respectively. For comparison we also include the energies for MO’s of H2S, 5a1 and 2b1 (HOMO), and other sulfur-containing molecules (thiols and thiophenes).8a,11,58,64 Not shown in the figure is the LUMO of H2S, 3b2 orbital, which is located at ∼4 eV.11,58 All energies are reported with respect to the vacuum level.

the hybridization of the LUMO of the sulfur-containing molecule and the valence band of the oxide is

QLUMO-valence ∝ (βLUMO-valence)2/(ELUMO - Evalence) (8) where the βLUMO-valence, ELUMO, and Evalence terms have a definition similar to that of the terms in eq 7. The magnitude of the resonance integrals (βHOMO-conduction, βLUMO-valence) depends on the overlapping of the orbitals of the adsorbate and adsorption site.61,62 The electron-density plots in Figures 6 and 9 indicate that the metal centers in ZnO(0001) should overlap better with the orbitals in an adsorbate than the metal centers in MgO(100). In MgO and Al2O3, the cations have a very large positive charge, which leads to orbitals that are more contracted (i.e. smaller) than the orbitals on cations of oxides that are not highly ionic like ZnO and Cu2O. Therefore, on MgO and Al2O3, the adsorbate-surface orbital overlap is poorer and the values of β are smaller. After application of eqs 7 and 8 to the systems in Figure 11, the differences in the band energies of the oxides predict that the strength of the adsorbateToxide bonding interactions should increase following the sequence: Al2O3 < MgO < ZnO < Cu2O, in agreement with the experimental trends seen in Figure 1. The qualitative correlation seen between the band-gap size and reactivity in an oxide also involves the ionicity of the system. For the oxides in Figure 11, the degree of ionicity and size of the band gap go together, and the highly ionic systems offer metal centers whose orbitals are relatively small and have energies that do not match well with those of the orbitals of the adsorbate. The simple model of eqs 7 and 8 can be applied not only to sulfur-containing molecules but also to other types of adsorbates in general.61-63 On oxide surfaces, a correlation between

Rodriguez and Maiti chemical activity and band-gap size is also observed for CO, NO2, and alkali metal chemisorption. The case of CO is particularly well-documented for adsorption of the molecule on MgO(100)18,65,66 and TiO2(110).24,55,67 CO interacts much stronger with metal centers of TiO2(110), band gap ∼2 eV 16 and not highly ionic,67 than with metal centers of MgO(100). In the CO/MgO(100) system the contributions to the chemisorption bond from band-orbital mixing are very weak,18,65a but become substantial in the CO/TiO2(110) system.24,67 For the interaction of NO2 with the cations of ZnO(0001),28 CASTEP predicts bonding energies that are larger than those found on the cations of MgO(100),57 a trend that reflects variations in the strength of band-orbital mixing, as was seen for the adsorption of the sulfur-containing molecules on the oxides.57 In the case of alkali metal adsorption, one finds very strong bonding interactions on oxides that have a narrow band gap (5 eV), the bonding energies are weak and controlled by interactions of the adsorbate with vacancies and defect sites of the oxide surface.68 According to eqs 7 and 8, one can increase the reactivity of an oxide surface by creating metal centers which have orbitals that are less contracted (higher values of β) or have a good match in energy with the orbitals of the adsorbate (no very stable occupied states or unstable vacant states on the metal centers). This can be accomplished by introducing oxygen vacancies on a surface and reducing the positive charge on the adjacent cations. Defect sites can exhibit substantial chemical activity, as the results in Table 2 (Mg-5 vs Mg-3 adsorption energies) show. Another approach to enhance reactivity consists of doping the oxide surface with a second metal that has the right electronic properties. Thus, for example, NixMg1-xO is more reactive than MgO (Table 3 and ref 47). In previous experimental studies,8a,20 we have found that cesium enhances the efficiency of zinc oxide as a sorbent of sulfur-containing molecules. The Cs adatoms have electronic states that are located within the band gap of ZnO and are very good for bonding interactions with adsorbates.8a V. Summary and Conclusions The adsorption of H2S, HS, and S on MgO(100), NiMgO(100), and ZnO(0001) was studied using DFT calculations and slab models. In general, the bonding energies of the adsorbates on a given surface increase in the following sequence: H2S < HS < S. For the surfaces the chemical activity increases in the order MgO < NiMgO < ZnO. Atomic S and HS interact strongly with the perfect MgO(100) surface with bonding energies of 34 and 19 kcal/mol, respectively, at a coverage of 0.25 monolayer. The dissociation of hydrogen sulfide (H2Sgas + Mgsurface + Osurface f HS-Mgsurface + H-Osurface) is an energetically favorable process (∆E) -14 kcal/mol at θH2S ) 0.25 monolayer), but it may not be easy due to the relatively weak interactions between the molecule and pentacoordinated Mg atoms of the flat oxide surface (H2S bonding energy of 9 kcal/mol). For tri-coordinated Mg atoms in step sites of MgO(100), the H2S bonding energy (13 kcal/ mol) is larger, enhancing the probability for dissociation of the molecule. When metal centers of MgO(100) are replaced with nickel atoms, new electronic states are created above the occupied {O 2p + Mg 3s} bands. In the NiMgO(100) system, there is a large electron density around the Ni atoms. Atomic S and HS form very strong bonds on this oxide surface, and in addition the bonding interactions of H2S are substantial (H2S adsorption

Adsorption and Decomposition of H2S on Oxides energy of 17 kcal/mol at θH2S ) 0.25 monolayer) favoring decomposition. DFT calculations also indicate that the degree of ionicity in ZnO is significantly smaller than in MgO and that the cations in ZnO(0001) are more chemically active than penta- or tricoordinated metal sites in magnesium oxide. On top of ZnO(0001), one sees not only stronger H2S-metal bonds (21 vs 9 or 13 kcal/mol) but also a larger elongation of the H-S bonds (0.05 vs 0.01 Å). These effects make easier the dissociation of H2S and HS on zinc oxide than on magnesium oxide. A simple model based on perturbation theory and bandorbital mixing is able to explain the differences in the reactivity of MgO(001) and ZnO(0001) and the behavior of other oxides in the presence of sulfur-containing molecules. The reactivity of an oxide mainly depends on how well its bands mix with the orbitals of H2S, and the electrostatic interactions with the dipole moment of H2S play only a secondary role in bonding. Acknowledgment. The authors thank J. Z. Larese for several thought-provoking conversations on the behavior of MgO. The research carried out at Brookhaven National Laboratory was supported by the Division of Chemical Sciences of the US Department of Energy (Contract DE-AC02-98CH10086). References and Notes (1) Speight, J. G. The Chemistry and Technology of Petroleum, 2nd ed.; Dekker: New York, 1991. (2) Stern, A. C.; Boubel, R. W.; Turner, D. B.; Fox, D. L. Fundamentals of Air Pollution, 2nd ed.; Academic: Orlando, FL, 1984. (3) Rodriguez, J. A.; Hrbek, J. Acc. Chem. Res. 1999, 32, 719. (4) Topsøe, H.; Clausen, B. S.; Massoth, F. E. Hydrotreating Catalysis; Springer-Verlag: New York, 1996. (5) DeactiVation and Poisoning of Catalysts; Oudar, J., Wise, H., Eds.; Dekker: New York, 1985. (6) Thomas, J. M.; Thomas, W. J. Principles and Practice of Heterogeneous Catalysis; VCH: New York, 1997. (7) Satterfield, C. N. Heterogeneous Catalysis in Practice; McGrawHill: New York, 1980. (8) (a) Jirsak, T.; Dvorak, J.; Rodriguez, J. A. J. Phys. Chem. B 1999, 103, 5550. (b) Rodriguez, J. A.; Jirsak, T.; Freitag, A.; Hanson, J. C.; Larese, J. Z.; Chaturvedi, S. Catal. Lett. 1999, 62, 113. (9) (a) Chaturvedi, S.; Rodriguez, J. A.; Jirsak, T.; Hrbek, J. J. Phys. Chem. B 1998, 102, 7033. (b) Rodriguez, J. A.; Jirsak, T.; Chaturvedi, S.; Kuhn, M. Surf. Sci. 1999, 442, 400. (10) Slack, A. V.; Holliden, G. A. Sulfur Dioxide RemoVal from Waste Gases, 2nd ed.; Noyes Data Corp.: Park Ridge, NJ, 1975 (11) Rodriguez, J. A.; Chaturvedi, S.; Kuhn, M.; Hrbek, J. J. Phys. Chem. B 1998, 102, 5511. (12) Rodriguez, J. A.; Jirsak, T.; Chaturvedi, S. J. Chem. Phys. 1999, 111, 8077. (13) Pacchioni, G.; Clotet, A.; Ricart, J. M. Surf. Sci. 1994, 315, 337. (14) (a) Raza, H.; Harte, S. P.; Muryn, C. A.; Wincott, P. L.; Thorton, G.; Casanova, R.; Rodriguez, A. Surf. Sci. 1996, 366, 519. (b) Kurtz, R. L.; Henrich, V. E. Phys. ReV. B 1987, 36, 3413. (15) Wycoff, R. W. G. Crystal Structures, 2nd ed.; Wiley: New York, 1964. (16) Henrich, V. E.; Cox, P. A. The Surface Science of Metal Oxides; Cambridge University Press: Cambridge, U. K., 1994. (17) Sousa, C.; Illas, F.; Bo, C.; Poblet, J. M. Chem. Phys. Lett. 1993, 215, 97. (18) Mejias, J. A.; Marquez, A. M.; Fernandez-Sanz, J.; FernandezGarcia, M.; Ricart, J. M.; Sousa, C.; Illas, F. Surf. Sci. 1995, 327, 59. (19) Lin, J.; May, J. A.; Didziulis, S. V.; Solomon, E. I. J. Am. Chem. Soc. 1992, 114, 4718. (20) Rodriguez, J. A.; Jirsak, T.; Chaturvedi, S.; Hrbek, J. Surf. Sci. 1998, 407, 171. (21) Evans, J.; Corker, J. M.; Hayter, C. E.; Oldman, R. J.; Williams, B. P. Chem. Commun. 1996, 1431. (22) Payne, M. C.; Allan, D. C.; Arias, T. A.; Johannopoulus, J. D. ReV. Mod. Phys. 1992, 64, 1045.

J. Phys. Chem. B, Vol. 104, No. 15, 2000 3637 (23) (a) Delley, B. J. Chem. Phys. 1990, 92, 508. (b) Delley, B.; Schefer, J.; Woike, T. J. Chem. Phys. 1997, 107, 10067. (c) Kessi, A.; Delley, B. Int. J. Quantum Chem. 1998, 68, 135. (24) Sorescu, D. C.; Yates, J. T. J. Phys. Chem. B 1998, 102, 4556. (25) (a) Lindan, P. J. D.; Harrison, N. M.; Holender, J. M.; Gillan, M. J.; Payne, M. C. Surf. Sci. 1996, 364, 431. (b) Lindan, P. J. D.; Harrison, N. M.; Holender, J. M.; Gillan, M. J. Chem. Phys. Lett. 1996, 261, 246. (26) Refson, K.; Wogelius, R. A.; Fraser, D. G.; Payne, M. C.; Lee, M. H.; Milman, V. Phys. ReV. B 1995, 52, 10823. (27) Rodriguez, J. A.; Hanson, J. C.; Chaturvedi, S.; Maiti, A.; Brito, J. L. J. Chem. Phys. 2000, 112, 935. (28) Rodriguez, J. A.; Jirsak, T.; Dvorak, J.; Sambasivan, S.; Fischer, D. J. Phys. Chem. B 2000, 104, 319. (29) Vandervilt, D. Phys. ReV. B 1990, 41, 7892 (30) Monkhorst, H. J.; Pack, J. D. Phys. ReV. B 1976, 13, 5188. (31) (a) Perdew, J. P.; Wang, Y. Phys. ReV. B 1992, 46, 6671. (b) White, J. A.; Bird, D. M. Phys. ReV. B 1994, 50, 4954 (32) Ziegler, T. Chem. ReV. 1991, 91, 651 (33) (a) White, J. A.; Bird, D. M.; Payne, M. C.; Stich, I. Phys. ReV. Lett. 1994, 73, 1404. (b) Hu, P.; King, D. A.; Crampin, S.; Lee, M. H.; Payne, M. C. Chem. Phys. Lett. 1994, 230, 501. (34) Whitten, J. L.; Yang, H. Surf. Sci. Rep. 1996, 24, 55. (35) van Santen, R. A.; Neurock, M.; Catal. ReV.sSci. Eng. 1995, 37, 557. (36) Lopez, N.; Illas, F.; Ro¨sh, N.; Pacchioni, G. J. Chem. Phys. 1999, 110, 4873. (37) Becke, A. D. Phys. ReV. A 1988, 38, 3098. (38) (a) Wang, Y.; Perdew, J. P. Phys. ReV. B 1991, 44, 13298. (b) Perdew, J. P. In Electronic Structure of Solids ‘91; Ziesche, P., Esching, H., Eds.; Akademie: Berlin, 1991. (c) Perdew, J. P.; Chevary, J. A.; Vosko, S. H.; Jackson, K. A.; Pederson, M. R.; Singh, D. J.; Fiolhais, C. Phys. ReV. B 1992, 46, 6671. (39) Mulliken, R. S. J. Chem. Phys. 1955, 23, 1841. (40) Szabo, A.; Ostlund, N. S. Modern Quantum Chemistry; McGrawHill: New York, 1989. (41) Wiberg, K. B.; Rablen, P. R. J. Comput. Chem. 1993, 14, 1504. (42) (a) Scamehorn, C. A.; Hess, A. C.; McCarthy, M. I. J. Chem. Phys. 1993, 99, 2786. (b) Ferro, Y.; Allouche, A.; Cora, F.; Pisani, C.; Girardet, C. Surf. Sci. 1995, 325, 139. (c) Chen, L.; Wu, R.; Kioussis, N.; Zhang, Q. Chem. Phys. Lett. 1998, 290, 255 (43) Deer, W. A.; Howie, R. A.; Zussman, J. An Introduction to the Rock Forming Minerals; Longman: Essex, U.K., 1966. (44) (a) Casarin, M.; Maccato, C.; Tondello, E.; Vittadini, A. Surf. Sci. 1995, 343, 115. (b) Casarin, M.; Maccato, C.; Vitaddini, A. Surf. Sci. 1997, 377-379, 587. (45) Scamehorn, C. A.; Harrison, N. M.; McCarthy, M. I. J. Chem. Phys. 1994, 101, 1547. (46) (a) Lopez, N.; Illas, F. J. Phys. Chem. B 1998, 102, 1430. (b) Anchell, J. L.; Hess, A. C. J. Phys. Chem. 1996, 100, 18317. (c) Johnson, M. A.; Stefanovich, E. V.; Truong, T. N.; Gu¨nster, J.; Goodman, D. W. J. Phys. Chem. B 1999, 103, 3391 (47) Tomishige, K.; Chen, Y.-G.; Fujimoto, K. J. Catal. 1999, 181, 91, and references cited therein. (48) He, J. W.; Møller, P. J. Surf. Sci. 1987, 180, 411. (49) Huheey, J. E. Inorganic Chemistry, 3rd ed.; Harper & Row: New York, 1983. (50) (a) Jaffe, J. E.; Harrison, N. M.; Hess, A. C. Phys. ReV. B 1994, 49, 11153. (b) Jaffe, J. E.; Hess, A. C. J. Chem. Phys. 1996, 104, 3348 (51) Baetzold, R. C. J. Phys. Chem. 1985, 89, 4150. (52) Abrahams, S. C.; Bernstein, J. L. Acta Crystallogr. B 1969, 25, 1233. (53) Duke, C. B.; Lubinsky, A. R. Surf. Sci. 1975, 50, 605. (54) (a) Rodriguez, J. A.; Campbell, C. T. J. Phys. Chem. 1987, 91, 6648. (b) According to the Phillips ionicity scale54c,d and Pauling’s electronegativity,54e zinc oxide is in the border of ionic and covalent compounds. (c) Phillips, J. C. Bonds and Bands in Semiconductors; Academic: New York, 1973; Chapter 2. (d) Phillips, J. C. ReV. Mod. Phys. 1970, 42, 317. (e) Pauling, L. The Nature of the Chemical Bond, 3rd ed.; Cornell University Press: New York, 1960; pp 93, 98, 246. (55) Casarin, M.; Maccato, C.; Vittadini, A. J. Phys. Chem. B 1998, 102, 10745. (56) Chaturvedi, S.; Rodriguez, J. A.; Hrbek, J. Phys. Chem. B 1997, 101, 10860. (57) Rodriguez, J. A.; Jirsak, T.; Sambasivan, S.; Fischer, D.; Maiti, A. J. Chem. Phys., in press. (58) Rodriguez, J. A. Surf. Sci. 1990, 234, 421, and references cited therein.

3638 J. Phys. Chem. B, Vol. 104, No. 15, 2000 (59) Rodriguez, J. A.; Chaturvedi, S.; Kuhn, M. Surf. Sci. 1998, 415, L1065. (60) Parr, R. G.; Yang, W. Density-Functional Theory of Atoms and Molecules; Oxford University Press: New York, 1989. (61) (a) Hoffmann, R. Solids and Surfaces: A Chemist’s View of Bonding in Extended Structures; VCH: New York, 1988. (b) Libit, L.; Hoffman, R. J. Am. Chem. Soc. 1974, 96, 1370. (62) Shustorovich, E.; Baetzold, R. C. Science 1985, 227, 876. (63) Hammer, B.; Morikawa, Y.; Nørskov, J. K. Phys. ReV. Lett. 1996, 76, 2141. (64) (a) Rodriguez, J. A. J. Phys. Chem. 1997, 101, 7524. (b) Rodriguez, J. A. Surf. Sci. 1992, 278, 326.

Rodriguez and Maiti (65) (a) Illas, F.; Pacchioni, G.; Pelmenschikov, A.; Pettersson, L. G. M.; Dovesi, R.; Pisani, C.; Neyman, K. M.; Ro¨sch, N. Chem. Phys. Lett. 1999, 306, 202. (b) Wu, R.; Zhang, Q. Chem. Phys. Lett. 1999, 306, 205. (66) (a) Wichtendahl, R.; Rodriguez-Rodrigo, M.; Ha¨rtel, U.; Kuhlenbeck, H.; Freund, H.-J. Surf. Sci. 1999, 423, 90. (b) Freund, H.-J.; Kulenbeck, H.; Staemmler, V. Rep. Prog. Phys. 1996, 59, 283. (67) Pacchioni, G.; Ferrari, A. M.; Bagus, P. S. Surf. Sci. 1996, 350, 159. (68) (a) Campbell, C. T. Surf. Sci. Rep. 1997, 27, 1. (b) Madey, T.; Yakshinky, B. V.; Ageev, V. N.; Johnson, R. E. J. Geophys. Res. 1998, 103, 5873.