i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 3 7 ( 2 0 1 2 ) 8 0 3 3 e8 0 4 2 Available online at www.sciencedirect.com journal homepage: www.elsevier.com/locate/he Hydrogen energetics and charge transfer in the Ni/LaNbO4 interface from DFT calculations K. Hadidi a,*, T. Norby b, O.M. Løvvik a,c, A.E. Gunnæs a a Department of Physics, University of Oslo, FERMiO, Gaustadalleen 21, NO-0349 Oslo, Norway Department of Chemistry, University of Oslo, FERMiO, Gaustadalleen 21, NO-0349 Oslo, Norway c SINTEF Materials and Chemistry, Forskningsveien 1, NO-0314 Oslo, Norway b article info abstract Article history: Calculations based on density functional theory have been used to simulate interface Received 30 July 2011 structures between the tetragonal and monoclinic phases of LaNbO4 (LN) and Ni. Schottky Received in revised form barrier heights were calculated using the interface electronic structure; they were 3.0 and 7 October 2011 1.8 eV for p- and n-type barriers. The hydrogen interstitials were found to be significantly Accepted 4 November 2011 higher stable in the LN part of the interface than in bulk LN. Also, the potential energy Available online 14 December 2011 curve of hydrogen diffusion from Ni into LN exhibited a deep well of around 2 eV, located in the gap region between two components. The stability of H atom in the gap region and Keywords: interfacial layers of LN is explained by metal-induced gap states and indicates that there LaNbO4 will be an accumulation of hydrogen in this area. It was shown that hydrogen is ionized Ni when enters from Ni to the LN interfacial layer, approaching to the O atoms and that the DFT electron lost from hydrogen resides in the interface states, located in the band gap of LN. Hydrogen Copyright ª 2011, Hydrogen Energy Publications, LLC. Published by Elsevier Ltd. All rights reserved. Fuel cell Interface 1. Introduction Proton conducting fuel cells (PCFCs) have received much attention related to their potential utilization in solid oxide fuel cell (SOFC) technology [1]. In PCFC operation, details of the physical phenomena of the pathway of hydrogen (proton), starting with a H2 molecule at the anode side and ending with protons in the electrolyte and water at the cathode side, are still largely unknown. In this regard, characterization of the metaleoxide interface structure as a part of the hydrogen (proton) pathway appears essential. The application of lanthanum niobate (LaNbO4) based ceramics as high temperature proton conducting electrolyte in PCFCs has recently been discussed in the literature [2]. Acceptor doped LaNbO4 (LN) displays a proton conductivity of nearly 0.001 S cm1 at 900 C [3]. The preference for this compound to other state-of-the-art proton conductors with higher conductivity, such as Ba- and Sr-based perovskites is due to its chemical stability under realistic operational conditions. LN transforms from the monoclinic fergusonite phase (m-LN) to a tetragonal schelite structure (t-LN) within the temperature range of 490 Ce525 C [4]. In a LN based PCFC, Ni is often chosen as anode, where it will assumingly do the job as current collector and catalyze dissociative adsorption of H2 since the most stable LN surface is not receptive toward H atoms [5]. In the present work, Ni and the two polymorphs of LN were employed as components to simulate two different interface structures; tetragonal (t-) and monoclinic (m-) interface models. The interfacial region (IR) can be considered as * Corresponding author. E-mail address: kianoosh.hadidi@fys.uio.no (K. Hadidi). 0360-3199/$ e see front matter Copyright ª 2011, Hydrogen Energy Publications, LLC. Published by Elsevier Ltd. All rights reserved. doi:10.1016/j.ijhydene.2011.11.032 8034 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 3 7 ( 2 0 1 2 ) 8 0 3 3 e8 0 4 2 a transition region, which begins from the gap region and can extend to a few layers of each component. The characteristics of the interfacial region are neither purely metallic nor oxide like. In our work, the whole super cell represents the interfacial region. The fundamental properties of the interfacial region, such as interface dipole or Schottky barrier originate from its chemical bonds [6,7], which depend on the type of surfaces that meet, as well as the lattice mismatch between the constituents. As shown subsequently, we do not expect the binding configuration at the m-interface to be different from that in the t-interface. Hence, more comprehensive analyses, including Schottky barrier estimation, hydrogen interstitials energetics, and charge transfer across the IR were performed for the m-interface only. In some cases calculations were compared to the conventional bulk models of LN and Ni (referred to as Ni or LN bulk in the text). Various options can be applied to simulate an interface structure, related to all the possible surface terminations of two materials being accommodated with each other. Many of the physical properties can be affected by the choice of surfaces. In this study, the LN (010) and Ni (100) surfaces were chosen to generate interface structures, due to the low surface energies of these surfaces and the possibility to create relatively small atomistic models. The quantitative results clearly depend on this choice, so any numbers presented in this study should be interpreted with care. Nevertheless, many of the results are qualitative of nature, and should give considerable insight into the detailed mechanisms governing hydrogen transport and membrane performance. Also, some of the results should be applicable for metaleoxide interfaces in general. 2. Methodology 2.1. Computational details Band structure density functional theory (DFT) through the Vienna ab initio simulation package (VASP) [8,9] was implemented to perform the present calculations. The generalized gradient approximation within the PerdeweBurkee Ernzerhofscheme (PBE-GGA) [9,10] was utilized to account for the exchange-correlation corrections. We used soft potentials, which describe the valence electronic orbitals as 5s25p66s25d1 for La, 4p65s24d3 for Nb, 2s22p4 for O and 4s23d8 for Ni. The projected augmented wave (PAW) [11,12] method was employed to consider the core shell electrons. Cut-off energy of 500 eV was found to be sufficient to achieve numerical convergence of relative electronic energies within 0.001 eV. A similar accuracy was obtained by using a maximum distance of 0.25 Å1 between the k points; this was achieved with a grid of 5 2 5 points for the interface structure. The experimental covalent radii of 1.69, 1.37, 0.73 and 1.21 Å were applied for La, Nb, O and Ni, respectively, to calculate the local density of states (LDOS). For ionic optimization, a conjugate gradient algorithm [13] was utilized in most of the cases unless the primary configuration of atomic geometries was close to the local minimum. In that case we used the residual minimization scheme with direct inversion in the iterative subspace (RMMDIIS) [14]. The residual forces were less than 0.05 eV Å1 when all geometries were relaxed. All calculations were performed with spin polarization. We applied Bader analysis [15] to understand the charge transfer between different species in the interface structure. In this method, the space of the molecular system is divided into atomic volumes according to their charge densities, such that the gradient of the electron density vanishes at every point on the dividing surfaces. One of the challenging issues when designing the atomistic interface model is to satisfy the periodic boundary conditions of two materials with different lattice parameters at the same time. A variety of methods have been discussed in the literature [16]: (1) incoherent interfacial region; to keep the lattice parameters of both materials in their equilibrium state. The misfit can be adjusted when extending the interface super cell along the interfacial plane. Since the size of the super cell needed for this adjustment is directly dependent on the misfit ratio of two lattice parameters, there is in some cases no rational limitation for the super cell extension. Due to the system size limitation of DFT based methods; this is not feasible in our case. (2) Coherent interfacial region; to accommodate the lattice parameters of two materials, straining the smaller lattice into ‘tensile coherent boundary conditions’ or compressing the larger lattice into ‘compressive coherent boundary conditions’. It is also possible to change both lattice parameters to achieve the coherency. (3) Semicoherent interfacial region; local coherency and misfit dislocations alternating along the interfacial plane. The coherent method (2) offers the possibility of using smaller unit cell for calculations and the tensile coherent method is superior to the compressive coherent one from its smaller overestimation of the work of adhesion and interfacial energy [16]. In the present work, we implemented the tensile coherent method to construct the atomistic interface model between Ni and LN. Therefore, the lattice parameters of Ni parallel to the (010) surface were strained to be adjusted to those of LN. It was kept fixed at the bulk value (as relaxed by DFT) along the normal of the interfacial plane. Subsequently, ionic relaxation of the interface structure was used to minimize the induced strain.There is a possibility that some strain remained perpendicular to the interface plane due to the change in volume of the Ni phase, but this was not investigated further. 2.2. Construction methodology It is important to select appropriate surface terminations of the two interface components. This selection will be based on the following considerations: which are the most abundant surfaces in reality, how large will be the calculation expenses, and how can the lattice mismatch be minimized. The (010) surface is the most stable among low index surfaces of LN [5], and should thus be the most abundant under real conditions. Furthermore, the surface unit cell of LN (010) is the smallest available of this compound, giving the opportunity to create an atomistic model with fewer atoms. Fig. 1 is an illustration of a 2 2 1 super cell of Ni along with the (010) surface of t- and m-LN. The PBE-GGA optimized lattice parameters of Ni and LN bulk unit cells are tabulated in Table 1. As shown in Fig. 1, the diagonal of the Ni unit cell face (specified at the (100) surface) with a value of 4.98 Å is comparable to the (010) surface dimensions (5.56 and 5.20 Å in m-LN i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 3 7 ( 2 0 1 2 ) 8 0 3 3 e8 0 4 2 8035 Fig. 1 e (a) The 2 3 2 3 1 unit cell of fcc Ni. The lattice distances applied to create the Ni/LN interface are specified by darker balls. (b) The (010) monoclinic LN surface and (c) the (010) surface of tetragonal LN seen from above. and 5.45 Å in t-LN). Hence, the Ni (100) surface can be aligned to the (010) surface of both the LN phases. The lattice misfit can be obtained by: d¼ aLN pffiffiffi 2aNið100Þ aLN (1) where a designates the PBE-GGA relaxed lattice parameters. In m-LN, the two lattice constants parallel to the interfacial plane are slightly different. The lattice mismatch will then be 10% and 5% along the [001] and [100] directions of the interfacial plane, respectively. For the tetragonal phase of LN, the lattice misfit is 7.6%. These values are rather large, and add considerable uncertainty to our calculations. As mentioned previously, however, this gives the opportunity to achieve qualitative results with a relatively small unit cell size. In order to accommodate Ni atoms properly at the LN surface, the most stable sites for Ni adsorption at the LN (010) surface were calculated. Adsorption energy was defined for this purpose as Eadsorp ¼ ENi=LNslab ELNslab ENi (2) Here, ELN-slab and ENi are total electronic energies of the clean LN slab and bulk Ni, respectively, and ENi=LNslab is the Table 1 e The experimental [32,33] and calculated lattice parameters of fcc Ni and LaNbO4. a, b and g are structural angles. This work Parameters a t-LN m-LN Ni b c a Experiment b LN slab energy including a Ni ad-atom. The calculated H adsorption energies for the sites specified in Fig. 2a are plotted in Fig. 2b. The most stable Ni atoms are found between surface Nb and La atoms (the sites labeled “O2O3” and “Nb1Nb3” in Fig. 2b). This insight was used to position the Ni component on LN when creating the interface models. Although these calculations were only performed for the monoclinic structure, it is reasonable to believe that the results are valid also for t-LN. The interatomic distances are not very different in the two LN phases, so the stability of Ni binding should not rely significantly on the LN phase. The thickness of the interface models is another parameter to determine. We used a 5 layers Ni slab in combination with a 3 layers LN slab; one layer of LN is here defined to consist of a stoichiometric amount of La, Nb, and O, and is thus a larger entity than the single-atom layers of Ni. This gave a thickness of 7.04 and 7.08 Å for the Ni and LN components, so that the unrelaxed interface models were in total 18.00 Å thick, including gap region. a g b c b 5.45 5.45 11.76 90 90 90 5.40 5.40 11.67 90 5.58 11.67 5.26 90 90.65 90 5.56 11.52 5.20 94.1 3.53 3.53 3.53 90 90 90 3.52 3.52 3.52 e 3. Results 3.1. Relaxed geometric structures As shown in Fig. 3, the geometric structures of the m- and tinterface are very similar. The relaxed structures exhibit a significant distortion of the Ni part induced by the lattice mismatch. The relaxed interatomic distances are tabulated in Table 2 and compared to the corresponding experimental interatomic distances [17]. The range of the interatomic distance within the first coordination shell is a measure of local distortions of the lattice. We see from Table 2 that this range is significantly more extended for the NieNi and NieO interatomic distances than for any other pairs of atoms; this 8036 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 3 7 ( 2 0 1 2 ) 8 0 3 3 e8 0 4 2 Fig. 2 e (a) The most stable Ni adsorption sites at the (010) surface of monoclinic LN. Surface atoms are specified with labels. (b) Calculated adsorption energies as a function of height above the surface for a Ni atom at different sites on the 2 3 2 (010) LN surface. In order to specify the initial Ni interstitial sites, legend of each plot shows the surface atoms, which surround a Ni ad-atom. is due to the tensile distortion of the Ni component. It is also clear that the bond length distributions in the Ni component for both m- and t-interface structures are quite similar. Compared to the bulks of m- and t-LN, all optimized minimum interatomic distances in the LN component of the interface structures have been increased. The minimum variation is 0.01 Å for LaeNb in the m-interface, and the maximum change is 0.17 Å, which belongs to the NbeNb interatomic distance of the t-interface. It is interesting to see that the interatomic distances between Ni and LN (NieO, NieLa, NieNb) of both m- and t-interface models are very similar to corresponding experimental values; within approximately 0.1 Å. This suggests that proper binding between Ni and LN has been obtained, and gives support to the physical relevance of our atomistic models. It is important to note that, even if the distortions of the Ni component are too large to be of any relevance to the bulk part of a physical system, it is not unreasonable to assume that such distortions may occur locally at the interfacial region in real systems. Even if a semi-coherent situation (with dislocations in addition to coherent regions) or even an amorphous Ni component is much more likely to occur in reality, a local distribution of NieNi and NieO distances similar to that of our interface model is likely to exist. The real excess of energy, related to the formation of an interface is defined according to the chemical potentials of constituents [18]: g ¼ Ginter X mi N i (3) i where G is the total Gibbs free energy of the interface model, mi the chemical potential and Ni the number of atoms of the components. This energy comprises both chemical and elastic contributions g ¼ ðEe1 þ Ech Þ. From ab initio calculations in T ¼ 0 K the chemical part of interfacial energy is defined as: g¼ ENi=LN nNi EBNi nLN EBLN 2A (4) Here, A is the surface unit cell area and ENi/LN is the total electronic energy of the interface structure (Ginter) from DFT calculations, and EBX is the bulk energy of compound X. The latter is multiplied by the number of layers of this compound in the interface structure, nx. The chemical interfacial energies for the monoclinic and tetragonal Fig. 3 e Side view of a part of the Ni/LN interface models: (a) unrelaxed tetragonal, (b) relaxed tetragonal, (c) unrelaxed monoclinic, and (d) relaxed monoclinic (Ni atoms are specified with white color). 8037 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 3 7 ( 2 0 1 2 ) 8 0 3 3 e8 0 4 2 Table 2 e Interatomic distances (Å) for the monoclinic and tetragonal interface models within the first coordinate shell, from DFT calculations compared to the experimental values [17]. The maximum and minimum values are reported when there is a distribution of this quantity. Atoms Interatomic distances monoclinic interface La-Ni Nb-Ni O-Ni Ni-Ni Nb-O Nb-La La-O Nb-Nb tetragonal interface Unrelaxed Relaxed Unrelaxed Relaxed 2.72 2.88 2.42e2.96 2.55e2.79 1.77e1.93 3.59 2.27 3.37 3.05 2.85 2.14e2.9 2.37e2.87 1.86e1.96 3.7 3 2.36 3.58 2.12 2.2 1.94e2.96 2.61,2.72 1.84e1.94 3.85 2.36 3.92 2.96 2.63 2.09e2.98 2.38e2.91 1.84e1.94 3.84 2.38 3.83 interface super cell have been calculated as 2.85 and 2.82 J/ m2, respectively. The difference between the interfacial energies can be attributed to the increase of lattice mismatch in the monoclinic structure compared to the tetragonal one. 3.2. Electronic structure Fig. 4 illustrates the total (DOS) and local density of states (LDOS) of the m-interface model, compared to the DOS of bulk Ni and LN. In metalesemiconductor (MS) interfaces (although LN with an experimental band gap of 4.8 eV is an insulator, it can also be regarded as a wide band gap semiconductor), the Fermi level of oxide is generally aligned with metal. When the DOS in bulk LN and Ni are compared with that of the interface model, the interface electronic structure is seen to be dominated by metallic states; as an example, there is no band gap in the interface structure. Likewise, the LDOS projected on LN atoms at the interfacial layer (the neighbor layer of Ni) exhibit spin polarization; this is clearly visible in the Nb p orbital, shown in the inset of Fig. 4. The valence bands further exhibit hybridization between the s and p orbitals of the LN elements and Ni. There is also a small overlap between the d orbitals of Ni, La and Nb in the valence band region. The orbital overlap between LN elements and Ni presented in the LDOS of interface model confirms that chemical bonds have been formed between Ni and LN components. Similar bonds are seen at the t-interface (not shown here). The electronic states appearing in the band gap of the interface model can be interpreted as metal-induced gap states (MIGS). These states are also demonstrated in LDOS projected on the LN elements. MIGS are due to the relaxation of metal electronic wave functions to the semiconductor at the interface [19,20]. These energy states decay exponentially in deeper layers of the semiconductor. Recent studies have suggested that the Schottky barrier height (SBH) [21] does not strongly depend on the metal work-function since the Fermi level may be pinned by the MIGS [22]. The SBH is a potential barrier at the metalesemiconductor interface, serving as a rectifying factor for charge transport across the IR. The p- and n-type SBHs are defined as: Experimental 2.92 in all La-Ni compositions 2.69-2.79 (Ni6Nb6O) 2.08-2.95 (La2(NiO4 2.48 (fcc Ni) 1.86 (m-LN), 1.89 (t-LN) 3.74 (m-LN), 3.85 (t-LN) 2.48 (m-LN), 2.52 (t-LN) 3.60 (m-LN), 4 (t-LN) Fp ¼ Ef VBT (5) Fn ¼ CBM Ef ¼ Eg Fp (6) Here, Fp and Fn are the SBHs for holes and electrons, respectively, Ef is the Fermi level of the interface structure, CBM and VBT are the conduction band minimum and valence band top of the LN component, and Eg is the LN band gap. To find the LN component VBT, the valence band minimum (VBM) of LN bulk was adjusted with the VBM of the interface structure. Since DFT at the GGA level tends to underestimate band gaps, the experimental LN band gap value of 4.8 eV was used to calculate the n-type SBH. We obtained Fn ¼ 1.8 eV and Fp ¼ 3 eV for our interface model (the calculated band gap from DFT calculation is w3.6 eV [5,23]). This means that electrons in the LN conduction band, at the interfacial region will tend to move toward and into Ni. 3.3. Stability of hydrogen in the interface model It is interesting to study the potential energy of a hydrogen atom when it moves through the interface structure from the Ni component to the LN component. Dissociative hydrogen adsorption at the surfaces of Ni and its diffusion pathway through the bulk have been thoroughly discussed in the literature previously [24,25], and are therefore not the topics of this study. We have placed H at interstitial sites according to Fig. 5a, which should provide a representative selection of sites through the interfacial region. The relaxed H positions are shown in Fig. 5b. We report here the binding energy Ebin of H in the interface model, which is defined as Ebin ¼ EInterfaceþH EInterface 1=2EðH2 Þ (7) The formation energy of H2 from H atoms was found to be E(H2) ¼ 4.52 eV, which agrees reasonably with experimental values [26]. Similar calculations were also performed for bulk Ni and LN for comparison with the interface, and to compare with previously reported results in the literature. As shown in Table 3, our results for the binding energy of H in bulk Ni compare very well with those previously reported by Wimmer et al. [27]. In the Ni component of interface model, H was placed at tetrahedral sites (Nos. 3 and 5) and octahedral sites (Nos. 4 and 8038 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 3 7 ( 2 0 1 2 ) 8 0 3 3 e8 0 4 2 Fig. 4 e DOS and LDOS of the m-interface structure. LDOS are projected on atoms in the interfacial layers of LN and Ni at the IR. Contributions are shown for O (wine dashes), La (solid black lines), Ni (green dots), and Nb (solid, bright gray, filled areas). The orbital is specified by a label at each plot. The region of the LN band gap in the interface electronic structure, which is filled by the metal-induced gap states is determined by two parallel arrows in the LDOS (p and s orbitals). To clarify the spin polarization of (refer to the text) energy states, the p orbitals of La and Nb are also shown separately in an inset. The DOSs of bulk LN and Ni are shown in the lowest panel. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) 6). Sites 3 and 4 can be considered as subsurface sites, while sites 5 and 6 placed in the middle of Ni component should be similar to bulk sites (see Fig. 5). The calculated binding energies should, however, not be expected to be the same as the corresponding energies in bulk Ni due to the large tensile distortion of the Ni component. The interatomic NieH distances show unique values in bulk Ni: 1.8 Å for the octahedral site and 1.6 Å for the tetrahedral site. The H positions are then symmetrical, at least in the low-density regime. In contrast, these distances exhibit a range of values in the interface models (Table 3). This is due to the loss of symmetry induced by the tensile stress. Hence, the calculated Ebin also exhibits a range of values in the interface model. When comparing with the previously reported results in literature, Ebin is higher (less stable) in the Ni component than in subsurface Ni, while it is lower (more stable) than in bulk Ni. 8039 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 3 7 ( 2 0 1 2 ) 8 0 3 3 e8 0 4 2 Fig. 5 e Positions of hydrogen interstitial sites in the monoclinic interface model before (a) and after (b) relaxation. The starting points for the hydrogen sites in the LN part were between La and Nb (sites 10 and 13), La and La (site 12), and Nb and Nb (site 11). Sites 2, 7, 8, and 14 were between LN and Ni (gap region). In the Ni component, octahedral sites (4 and 6) and tetrahedral sites (3 and 5) were chosen. In (b), the relaxed H positions are shown together with the unrelaxed positions of the other interface atoms, to clearly exhibit how H atoms have moved during relaxation. In (c), the corresponding relaxed H positions in the bulk LN model are presented. The initial H sites are not shown since H atoms are not significantly displaced due to the relaxation in this case. This is most pronounced for the subsurface tetrahedral site, which in the interface structure has stability rather like that of the octahedral site than of the subsurface tetrahedral site in pure Ni. This is due to the stressed Ni lattice close to the interface, which has led to tetrahedral sites with NieH interatomic distances, approaching those of the octahedral site (Table 3). The calculated H binding energies and the interatomic distances between H and its nearest neighbors in both the LN component and bulk are similarly listed in Table 4. The initial and relaxed H sites are shown in Fig. 5. From this table, we see that the relaxed interatomic distance between hydrogen and oxygen atoms is between 0.98 and 1.00 Å in the LN component. This is equal to the distance found for adsorbed H atom at the (010) LN surface [5] and is the typical OeH bond length in solid compositions when H atom has a formal charge of þ1 [28]. The same OeH bond length can be seen in the two most stable H binding sites (10 and 13) of bulk LN as well. The distance between H and La is slightly larger in the LN component than in bulk LN. This is also the case for the NbeH distances when only the most stable sites are regarded. The explanation of this must be a larger freedom of ionic displacements in the interface models close to the interface plane (the plane between LN and Ni). This is illustrated in Fig. 6, where the relaxed interface model is compared to the situation when H is placed at site 11. In the H-included model, the minimum interatomic distance for LaeNb is increased by up to 0.7 Ǻ, while the LaeLa and NbeNb interatomic distances are increased by 1.0 Ǻ. The NbeNi distance is similarly extended by 0.33 Ǻ, while La and Ni become 0.7 Ǻ closer. All Habsorbed interface models exhibit similar geometric relaxations. The large relaxations are accompanied by significantly lower H binding energy (more stability) in the LN component compared to the LN bulk. The binding energies are listed in Table 4 and schematically presented in Fig. 7 across the Table 3 e Interatomic distances (Å) and binding energies of hydrogen (eV/atom) in the Ni component of the interface model in comparison with corresponding calculated and experimental values for Ni bulk. For the Ni component, the H sites are numbered according to Fig. 5. Ni component Ni Bulk Literature[24,27] Subsurface Bulk Subsurface Site 3-tetra 4-octa 5-tetra 6-octa Ni-H Eb 1.63e1.7 0.197 1.72e1.9 0.2 1.48e1.58 0.349 1.58e1.98 0.179 This work Bulk Bulk Tetra Octa Tetra Octa Tetra Octa e 0.51 e 0.2 1.58 0.33 1.77 0.075 1.6 0.34 1.8 0.07 8040 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 3 7 ( 2 0 1 2 ) 8 0 3 3 e8 0 4 2 Table 4 e Interatomicdistances (Å) between the relaxed H interstitial and its nearest neighbors in the first coordination shell as well as H binding energies (Eb) in eV/atom, at the LN component, gap region and LN bulk. The specified H sites have been described in Fig. 5. yini and yre represent the y coordinates of H before and after ionic relaxation, according to the coordinate system of interface model in Fig. 5. Site LaeH (Å) NbeH (Å) NieH (Å) OeH (Å) yini (Å) yre (Å) Eb LN component 12 13 11 10 2.7 2.66 e 1.00 5.22 5.98 0.01 2.7 2.66 e 0.98 5.4 6.02 0.19 2.68 2.55 e 0.98 5.16 6.26 0.38 2.6 2.63 e 0.98 6.36 6.72 0.41 LN bulk 8 e 2.64 2.38 1.00 7.66 8.32 0.38 7 14 1.58 2.68 2.68 1.44 e e e e 1.7 e e e 8.3 9.02 1.44 6.59 8.17 0.32 9.53 9.35 1.6 interface. The subsurface sites of the LN component (all sites in the LN component, shown in Fig. 5b), exhibit a H binding energy of around 1.8 eV more stable than in bulk LN (the leftmost points of Fig. 7). This is most readily explained by MIGS offering electronic states in the band gap, thus increasing the stability of H in the LN component. Furthermore, hydrogen atoms in the LN component relax toward the interfacial plane. This is quantified in Table 4, Fig. 6 e Relaxed interatomic distances between selected metal atoms in the LN component in (a) the perfect interface model and (b) the interface model when a H interstitial has been included in site 11. 2 12 13 11 10 15 2.31 1.24 e e e e 3.05 2.46 1.92 e 1.00 e e 2.16 e 2.12 e 1.22 e e 2.76 2.64 2.46 e 0.98 e e 1.77 2.03 e 1.27 e e 2.29 comparing the y coordinate (refer to Fig. 3) of the relaxed hydrogen positions with that of the initial positions. Such a directed relaxation of H interstitials occurs neither in bulk LN nor Ni component (Fig. 5). As mentioned above, hydrogen interstitials attain much lower energies and higher stability in the LN component than in bulk, which is related to the influence of MIGS. Since these gap states exponentially reduces through the oxide part of the interface structure; H atoms in the LN component are more stabilized close to the interfacial plane. The gap region between LN and Ni components is extended perpendicular to the interfacial plane as 2.44 Å, when considering the nearest Ni and O atoms. In this region, we find that the NieH distances in sites 2, 7, and 14 (1.44e1.7 Å) are Fig. 7 e Potential energy curve of hydrogen at various interstitial positions in the interfacial region (red diamonds) compared to corresponding energies in bulk LN from the present work (green crosses) and in the Ni subsurface and bulk [24,27] (green squares). The H binding energy is defined in Section 3.3. Solid lines specify the borders of Ni and LN interfacial surfaces, considering sites 2 and 8. Sites 10 and 3 are at the interfacial layers of LN and Ni components. (For interpretation of the references to color in this figure legend, the reader is referred to the web version of this article.) 8041 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 3 7 ( 2 0 1 2 ) 8 0 3 3 e8 0 4 2 Table 5 e The calculated charge of the hydrogen atom when placed at different sites (refer to the Fig. 5 for definition of the sites) as well as the charge of the Ni and LN interfacial layers when including H atom in these sites. The nearest atoms to the H interstitial are listed in the “Neighbors” row. Site H-Neighbors H charge Ni charge LN charge 12 11 10 14 8 7 2 4 3 6 Nb,La,O þ1 0 1 Nb,O,La þ1.0 þ0.1 1.1 La,Nb,O þ1.00 þ0.01 1.01 La,Ni,Nb 0.39 þ0.72 0.33 Ni,Nb,O þ1.0 0.1 0.9 Ni 0.3 þ0.7 0.4 Ni 0.31 þ0.70 0.40 Ni 0.28 þ0.70 0.42 Ni 0.37 þ0.80 0.52 Ni 0.28 þ0.72 0.43 comparable with what is found for H adsorption at the (100) surface of pure Ni (1.84 Å for the most stable site at the (100) Ni surface [24]). Likewise, the OeH and NbeH distances at site 8 are very close to the corresponding distances at the pure LN (010) surface [5]. In this respect, these sites can be considered to be on the interfacial LN and Ni surfaces, respectively. The stability of hydrogen atoms still increases, when moving to the gap region in the interface model. At the interfacial surface of LN (site 8), the binding energy is 0.38 eV; this is around 1.5 eV more stable than the adsorption energy of H on the pure LN (010) surface [5]. At the most stable site on the interfacial surface of Ni (site 2), there is still a significant effect of the MIGS stabilization; here, the binding energy is 1.6 eV, which is 1.1 eV below the adsorption energy on the pure Ni (100) surface (approximately 0.5 eV [29]). The stabilization rapidly disappears when we enter Ni; the subsurface site is stabilized by 0.3 eV compared to pure Ni, while the remaining sites are virtually unchanged. A deep potential well can thus be seen, localized between the Ni and LN components, bonded to the Ni interfacial surface (Fig. 7). The consequence of this is that any available hydrogen atoms will be trapped near the interfacial plane (Fig. 7), where the potential energy of hydrogen is very low. In addition, hydrogen will be accumulated in the region of LN close to the interface plane, due to the significant stabilization of Fig. 8 e The sum of Bader charges for each layer (each layer is specified by black dots for Ni and black squares for LN) of the Ni and LN components, relative to the atomic states. The interfacial layers for LN and Ni (on both sides of each component) were adjusted at the first and last points of the plots, to present the charge transfer between LN and Ni interfacial layers which shown to be from Ni to LN. This makes the interfacial dipole which is demonstrated by arrows. hydrogen relative to bulk LN. We believe that this is a general phenomenon in all metal/oxide interfaces, since it relies on the existence of MIGS which are commonly found in this kind of interfaces. 3.4. Charge transfer Charge transfer to and from H atoms through the interface structure was investigated through the Bader analysis. For these calculations, we used the same relaxed H positions as specified in the previous section (Fig. 5b). The results of these calculations are summarized in Table 5. When H is in contact with Ni, charge transfer is from Ni to H in all cases, where the NieH distance is less than 2 Ǻ. This can be justified by the electronegativity of Ni being slightly lower than that of H. When H is within or at the LN component, it is ionized into a proton bonded to an oxide ion with a bond length of 0.98 Å, as expected [28]. Fig. 8 presents a plot of the integrated charge per layer across the LN and Ni components in the perfect interface model. The total average charge of the Ni layers displays a periodic behavior, due to the charge screening effects on the different metal layers. The Ni and LN interfacial layers (leftmost and rightmost in the plot) exhibit an average positive and negative charge respectively, which indicates a charge transfer from Ni to the LN component at the interface. This is due to the chemical bonds between metal and semiconductor; thus the binding configuration is responsible for the formation of the interface dipole [7,30,31]. The existence of chemical bonds between the Ni and LN components was affirmed in the previous sections. Due to the incoherency between components at the IR and a variety of surface terminations meeting each other, the bond configuration will certainly vary in polycrystalline materials. In this respect, there will be several different local dipole values at the IR and only an effective average dipole can be measured for the whole structure. We have shown in Section 3.3 that the significant stability of H atom in the IR is grounded in the presence of MIGS. The electron of ionized H will thus be placed in these gap states close to the Fermi level since the O valence bands are already filled. In the operation of a fuel cell, these electrons can be delocalized and enter to the conduction band to feed an external electronic current in the circuit connected to the Ni electrode. 4. Conclusions We have constructed two atomistic interface models between the Ni (100) surface and the (010) surface of LaNbO4 (LN) in its 8042 i n t e r n a t i o n a l j o u r n a l o f h y d r o g e n e n e r g y 3 7 ( 2 0 1 2 ) 8 0 3 3 e8 0 4 2 tetragonal and monoclinic phases. Local density of states confirmed the formation of chemical bonds between Ni and LN elements. In the interfacial region (IR), the calculated ntype Schottky barrier height (SBH) was found to be 1.8 eV, indicating that at the IR, the LN conduction band electrons will migrate to the Ni part. The potential energy of hydrogen across the interfacial region displayed a deep potential well of around 2 eV between the LN and Ni components, with H primarily bonded to Ni. It is induced by metal-induced gap states (MIGS), stabilizing H at a site resembling adsorption sites on the pure Ni surface. These very stable interstitial sites will lead to hydrogen trapped at the interfacial plane. In addition, stabilization of hydrogen or hydrogen ionized to protons in a larger region within LN close to the interface plane will lead to further accumulation in the interfacial region. Bader charge analysis demonstrated that a dipole is formed at the interface, due to charge transfer from Ni to the LN component; this is a result of chemical binding between the two components. Furthermore, it was shown that H donates its electron at the LN interfacial layer, and thus becomes a proton bonded with an oxide ion. The electrons lost from H atoms reside in the MIGS, placed in the band gap of LN close to the Fermi level. Regarding the significant value of the n-type SBH, these electrons will migrate to the Ni part where they, for instance, make up the current flowing when the Ni part is connected as an electrode in a fuel cell. Acknowledgment The authors would like to thank the NOTUR consortium for computational resources and the Research Council of Norway for funding (NANOMAT project NANIONET 182065/S10). references [1] Norby T. Proton conductivity in perovskite oxides. In: Ishihara T, editor. Perovskite oxide for solid oxide fuel cells. US: Springer; 2009. p. 217e41. [2] Haugsrud R, Norby T. Proton conduction in rare-earth orthoniobates and ortho-tantalates. Nat Mater 2006;5:193e6. [3] Haugsrud R, Norby T. High-temperature proton conductivity in acceptor-doped LaNbO4. Solid State Ionics 2006;177: 1129e35. [4] Jian L, Wayman CM. Monoclinic-to-tetragonal phase transformation in a ceramic rare-earth orthoniobate, LaNbO4. J Am Ceram Soc 1997;80:803e6. [5] Hadidi K, Hancke R, Norby T, Gunnæs AE, Løvvik OM. Atomistic study of LaNbO4; surface properties and hydrogen adsorption, submitted for publication. [6] Andrews JM, Phillips JC. Chemical bonding and structure of metalesemiconductor interfaces. Phys Rev Lett 1975;35:56. [7] Dandrea RG, Duke CB. Interfacial atomic composition and Schottky-barrier heights at the AL/GAAS(001) interface. J Vac Sci Technol B 1993;11:1553e8. [8] Hohenberg P, Kohn W. Inhomogeneous electron gas. Phys Rev 1964;136:B864. [9] Kohn W, Sham LJ. Self-consistent equations including exchange and correlation effects. Phys Rev 1965;140:A1133. [10] Kresse G, Furthmüller J. Efficient iterative schemes for ab initio total-energy calculations using a plane-wave basis set. Phys Rev B 1996;54:11169. [11] Blöchl PE. Projector augmented-wave method. Phys Rev B 1994;50:17953. [12] Kresse G, Joubert D. From ultrasoft pseudopotentials to the projector augmented-wave method. Phys Rev B 1999;59: 1758e75. [13] White SDM. In: Presse WH, Flannery BP, Teukolsky SA, Vetterling WT, editors. Numerical recipes e the art of scientific computing, vol. 1; 1986. 23 p. [14] Pulay P. Convergence acceleration of iterative sequences e the case of SCF iteration. Chem Phys Lett 1980;73:393e8. [15] Bader RFW. Atoms in molecules e a quantum theory. Oxford, UK: Oxford University Press; 1990. [16] Schnitker J, Srolovitz DJ. Misfit effects in adhesion calculations. Model Simul Mater Sci Eng 1998;6:153e64. [17] Inorganic crystal structure database (ICSD), http://icsd.fizkarlsruhe.de/. [18] Sutton AP, Ballffi RW. Interface in chrystalline materials. Oxford University Press; 1996. [19] Heine V. Theory of surface states. Phys Rev 1965;138:A1689. [20] Louie SG, Chelikowsky JR, Cohen ML. Ionicity and the theory of Schottky barriers. Phys Rev B 1977;15:2154. [21] Tung RT. Schottky-barrier height e do we really understand what we measure. J Vac Sci Technol B 1993;11:1546e52. [22] Mead CA, Spitzer WG. Fermi level position at metalesemiconductor interfaces. Phys Rev 1964;134:A713. [23] Arai M, Wang YX, Kohiki S, Matsuo M, Shimooka H, Shishido T, et al. Dielectric property and electronic structure of LaNbO(4). Jpn J Appl Phys, Part 1 2005;44:6596e9. [24] Bhatia B, Sholl DS. Chemisorption and diffusion of hydrogen on surface and subsurface sites of flat and stepped nickel surfaces. J Chem Phys; 2005:122. [25] Kresse G, Hafner J. First-principles study of the adsorption of atomic H on Ni(111), (100) and (110). Surf Sci 2000;459: 287e302. [26] Huber KP, Herzberg G. Molecular structure and molecular spectra IV. Constants of diatomic molecules. New York: Van Rostrand-Reinhold; 1979. [27] Wimmer E, Wolf W, Sticht J, Saxe P, Geller CB, Najafabadi R, et al. Temperature-dependent diffusion coefficients from ab initio computations: hydrogen, deuterium, and tritium in nickel. Phys Rev B; 2008:77. [28] Schuster P, Zundel G, Sandorfy C. The hydrogen bond. Amsterdam: North-Holand; 1976. [29] Nørskov JK, Bligaard T, Logadottir A, Kitchin JR, Chen JG, Pandelov S. Trends in the exchange current for hydrogen evolution. J Electrochem Soc 2005;152:J23e6. [30] Fujitani H, Asano S. Schottky barriers at NISI2/SI(111) interfaces. Phys Rev B 1990;42:1696e704. [31] Ludeke R, Talebibrahimi A, Jezequel G. Delocalization of defects e a new model for the Schottky-barrier. Appl Surf Sci 1989;41-2:151e8. [32] Alexander SA, Coldwell RL. Vibrational-rotational energies of all H-2 isotopomers using Monte Carlo methods. Int J Quant Chem 2006;106:1820e6. [33] Jette ER, Foote F. Precision determination of lattice constants. AIP; 1935.