Modification of TiO2 with metal chalcogenide nanoclusters for hydrogen evolution

Using density functional theory, corrected for on-site Coulomb interactions (DFT + U), we have investigated surface modification of TiO2 with metal chalcogenide nanoclusters for hydrogen evolution. The nanoclusters have composition M4X4 (M = Sn, Zn; X = S, Se) and are adsorbed at the rutile (110) surface. The nanoclusters adsorb exothermically, with adsorption energies in the range −2.8 eV to −2.5 eV. Computed density of states (DOS) plots show that cluster-derived states extend into the band-gap of the rutile support, which indicates that modification produces a redshift in light absorption. After modification, photoexcited electrons and holes are separated onto surface and cluster sites, respectively. The free energy of H adsorption is used to assess the performance of metal chalcogenide modified TiO2 as a catalyst for the hydrogen evolution reaction (HER). Adsorption of H at nanocluster (S, Se) and surface (O) sites is considered, together with the effect of H coverage. Adsorption free energies at cluster sites in the range −0.15 eV to 0.15 eV are considered to be favourable for HER. The results of this analysis indicate that the sulphide modifiers are more active towards HER than the selenide modifiers and exhibit hydrogen adsorption free energies in the active range, for most coverages. Conversely, the adsorption free energies at the selenide nanoclusters are only in the active range at low H coverages. Our results indicate that surface modification with small, dispersed nanoclusters of appropriately selected materials can enhance the photocatalytic activity of TiO2 for HER applications.


Introduction
Photocatalytic water splitting, wherein solar energy is converted to chemical energy, stored in the bonds of hydrogen gas (H 2 ), is a promising strategy for the production and storage of clean, renewable fuel. Solar driven water splitting proceeds at the surface of a semiconductor photocatalyst material as two half reactions, mediated by photoexcited electrons and holes. The half reactions are the oxygen evolution reaction (OER) and the hydrogen evolution reaction (HER), which are oxidation and reduction reactions, respectively. Each of these reactions requires catalyst architectures that are tailored for their promotion and technologies for overall water splitting will incorporate multifunctional components for optimal efficiency.
Among the most studied photocatalytic materials is titanium dioxide (TiO 2 ) [1][2][3][4][5][6][7]; TiO 2 is a cheap, non-toxic and abundant material, which is stable under photocatalytic operating conditions. However, the large band gap of TiO 2 (>3 eV) restricts its activity to the UV range of the solar spectrum. Thus, in order to maximise the ambient solar-driven photocatalytic capacity of titania-based materials, it is necessary to modify the electronic structure to induce a redshift in light absorption. Moreover, oxide-based photocatalysts are well studied for water oxidation [8,9]; that is, the OER half reaction. They are unsuitable for the HER without a co-catalyst. This is because H ions bind too strongly to the oxygen ions of the catalyst surface to form surface bound hydroxyls.
Typically, co-catalysts are employed to promote the HER and the most efficient co-catalyst materials are the platinum-group metals (PGMs) [8]. PGMs are critical materials, which are rare and expensive; this limits their implementation, and thus, the viability of water splitting technologies. This provides the impetus for the considerable scientific effort devoted to the discovery of new, cheap and earth-abundant materials for efficient HER catalysis. In this endeavour, chalcogenides (sulphides, selenides and tellurides) have been demonstrated to catalyse the HER with efficiencies approaching those of the PGMs [8,10]. In particular, layered transition metal dichalcogenides (TMDs), containing sulphur or selenium, have garnered interest due to high HER activities [11,12]. Through combinations of experiment and first-principles simulations, low-coordinated chalcogen sites at the edges of these layered TMDs have been identified as the active sites for HER [13][14][15].
There are two proposed pathways for the two-step HER mechanism: the Volmer-Tafel reaction and the Volmer-Heyrovsky reaction [8,16]. Here, we restrict ourselves to acidic media, as this is beneficial for the HER activity [17,18]. The Volmer step is common to both pathways and may be described as: Volmer: H + + e − → H ads .
In the Volmeer step, an electron is transferred to a proton at an adsorption site of the catalyst surface, resulting in a surface bound H species (H ads ). The HER reaction is completed with desorption of a H 2 molecule, which can proceed via a Tafel or Heyrovsky process: Tafel: H ads + H ads → H 2 . Heyrovsky: For optimal HER activity, there must be a balance between the competing adsorption and desorption steps. A widely accepted computational descriptor for this balance is the Gibbs free energy of adsorption of a H atom (∆G H ) at the catalyst surface [19]. This is termed the computational hydrogen electrode model and permits a rapid comparison and assessment of different catalysts and potential active sites for H adsorption.
DFT computations of the free energy of H adsorption at low-coordinated edge sites of MoS 2 revealed that this quantity, ∆G H , was close to thermoneutral (0 eV) and was comparable to that computed for Pt [20]. Conversely, ∆G H for sites in the MoS 2 basal plane were endothermic and so the basal plane is inactive for HER. Similar studies have been performed on a range of sulphides [15,[21][22][23], selenides [22,23] and phosphides [24][25][26] and show that ∆G H provides a useful descriptor for screening materials and sites for their HER activity. Values for ∆G H that are close to 0 eV are desirable, in accordance with the Sabatier principle. If the interaction is too strongly exothermic, the H ions will bind too strongly to the surface and if H adsorption is too uphill, or endothermic, no reaction will take place [20,24].
In this work, we examine modification of the TiO 2 rutile (110) surface with nanoclusters of composition M 4 X 4 (M = Sn, Zn; X = S, Se). Surface modification in this way can be performed using atomic layer deposition (ALD) [27], incipient wetness impregnation [28][29][30], or chemisorption-calcination cycles [7,31] and permits modulation of the light absorption properties of the titania substrate, promotes separation and stability of photoexcited electrons and holes, and provides low coordinated active sites for catalytic reactions. We have studied surface-modified TiO 2 for water oxidation [32] and CO 2 activation [33,34]. In the context of hydrogen evolution, this strategy enables us to combine the desirable properties of TiO 2 through modification with nanoclusters that display low-coordinated, active chalcogen sites, which promote the HER.
By computing the projected electronic density of states (PEDOSs), we analyse the impact of modification on the energy gap and light absorption. The modification with the chalcogen nanoclusters extends the valence band maximum (VBM) to higher energies, thereby inducing a redshift in light absorption compared to unmodified TiO 2 . We impose a triplet electronic state on the system to model photoexcitation and with this model, we compare the spatial separation and stability of photoexcited charges. Modification promotes the separation of electrons and holes and enhances their stability in the excited state.
Finally, we investigate the HER activity of the modified surface via computations of ∆G H . To be consistent with the literature and allow for errors in computed energies within the DFT set-up, we consider the range of ∆G H = (−0.15 eV to + 0.15 eV) to be relevant for assessment of the potential for HER. As metal oxide surfaces easily form surface bound hydroxyls [35], we first compute ∆G H for H adsorbed at sites of the titania support and consider coverages that range from 1H to enough hydrogen to saturate the available surface sites. For H adsorption at surface sites, we consider only the bridging oxygen sites (O br ) that have no interfacial bonds with the nanocluster modifier. For each surface H coverage, we then compute ∆G H for H adsorption at cluster sites. In these calculations, all chalcogen sites of the modifiers are considered and the most stable configurations that we find are discussed. We find that the sulphide modifiers exhibit free energies close to thermoneutral and within our desired range at most coverages. By contrast, for the selenide modifiers, the cluster sites are active only for low surface coverages of H. We rationalise these findings on the basis of the electronic structure of the chalcogen modifiers and propose metal sulphide-modified TiO 2 as a material for hydrogen evolution.

Methodology
All calculations are performed in a periodic plane wave basis set with the VASP5.4 [36,37] code. Calculations are spin polarised and use projector augmented wave potentials [38,39] to describe the core-valence electron interactions. The Perdew-Wang (PW91) approximation to the exchange correlation functional is used [40]. This functional was used in our earlier studies of nanocluster modified TiO 2 [33,41,42] and thus provides a consistent description of nanocluster-modified TiO 2 models, which facilitates comparison across these materials. Moreover, our models based on the PW91 functional have corroborated experimental observations in collaborative studies of doped TiO 2 and modified TiO 2 surfaces [32,43,44].
The plane-wave energy cut-off is 400 eV; Ti is described with four valence electrons; Sn with 14; Zn with 12; O, S and Se with six; and H with one. The convergence criteria for the energy and forces are 10 −4 eV and 0.02 eV Å −2 , respectively. The k-points are sampled using a (4 × 4 × 1) grid and aspherical gradient corrections are applied throughout. Gaussian smearing with σ = 0.1 eV is implemented in calculations of the DOS. Additional, single-point calculations, using PBE0, which has 25% exact exchange [45,46], were performed on the DFT + U relaxed geometries. For these calculations, a (2 × 2 × 1) k-point sampling grid was used.
A Hubbard U correction is applied to the Ti and Zn 3d states with U(Ti) = 4.5 eV and U(Zn) = 7.8 eV, consistent with previous work [47][48][49][50][51][52][53]. These corrections are necessary in describing the partially filled d-orbital and reduced Ti 3+ states [54,55] and to address shortcomings in the standard GGA description of Zn electronic states [53]. An additional +U correction is applied to the O 2p, S 3p, and Se 4p states (U = 5.5 eV) in the photo-excited state model to describe hole localisation [32,41,42,56]. We do not apply Where E NC@r110 , E r110 , and E NC are the energies of the composite surface, the bare surface, and the gas-phase NC, respectively. These energies can be thought of as describing the stability of the modifier on the TiO 2 substrate. Thus, a large ∆E ads signifies that the modifier would only desorb at high temperatures. We adopt the notation M 4 X 4 -r110 (M = Sn, Zn; X = S, Se) for these composite surfaces.
Regarding the size of these nanoclusters, these models have been previously used [53] and our older work with Tada and co-workers [7,[57][58][59] showed that smaller nanoclusters deposited onto TiO 2 showed higher activity and therefore small, sub-nm nanoclusters are likely to be the most active, while their deposition can be well controlled through ALD techniques.
The systems under study in the current work consist of metal chalcogenide nanoclusters in intimate contact with the rutile surface. The nanocluster and surface interact via the formation of interfacial bonds and this establishes new energy levels. The impact of modification on the light absorption properties will depend on the positions of these levels relative to the VBM and CBM of the support. Calculations of the energy gaps and band alignments of the isolated materials have been the subject of a number of papers [60,61], and this is useful in the rational selection of possible modifiers. However, due to the intimate nature of the interface, one cannot determine, a priori, the impact of surface modification based on these considerations. Thus, we compute the projected DOS for the M 4 X 4 -r110 and implement a model for photoexcitation, described below, to elucidate the light absorption properties of the composite surfaces.
Photoexcitation is modelled by imposing a triplet electronic state on the system. This forces an electron from the filled valence band to the empty conduction band, leaving a hole in the valence band. This computational approach has been used in previous work [32,33,41,42] and is described in detail in the supplementary material. We use this model to assess the impact of modification on the optical band gap, the stability and spatial separation of charges in the excited state. In the second set of calculations, shown in the second row, for each coverage of H at the TiO2 surface, one H is adsorbed at a site on the chalcogenide nanocluster. For those configurations for which the hydrogen adsorption free energy at a cluster site is in the active range, we proceed with the third set of calculations in which a second H is adsorbed at a chalcogen nanocluster site, as shown in the third row. In each image, the red H represents the current calculation.
Hydrogen adsorption is examined at both the surface and the nanocluster. We explore adsorption of H at twofold coordinated bridging O ions (O br ) of the rutile (110) surface and at chalcogen sites of the modifiers. We follow the workflow shown in the schematic in figure 1 and begin with H adsorption at O br surface sites. These calculations, represented by the top row of figure 1, are performed first and we identify the most stable configuration for each hydrogen coverage ranging from 0 H to saturation of the surface sites. The adsorption energy of the nth H atom at the most stable surface with an existing coverage of (n − 1)H atoms is computed via: Where E nH@surf , E (n−1)surf , and E H2 are the computed energies of the surface with nH atoms adsorbed, the surface with (n − 1)H atoms, and an isolated, gas phase H 2 molecule.
In the second set of calculations, represented by the second row of figure 1, for each surface coverage of hydrogen, we examine hydrogen adsorption at chalcogen sites. For a surface coverage of (n − 1)H, the adsorption energy of the n th H atom at a cluster site is computed via: Where E H@cluster is the energy of the system with 1H at a cluster site and (n − 1)H adsorbed at surface sites. These calculations elucidate the impact of surface hydrogen coverage on the strength of interaction between H and the cluster site. For those configurations with adsorption free energies within our range, we proceed with the third set of calculations, represented by the third row in figure 1, and examine adsorption of a second H atom at a cluster site to assess any trends in H coverage on the nanocluster modifier. From the chemisorption energies (∆E H ), we compute the free energy of adsorption using: Where ∆E ZPE is the difference in zero point energy (ZPE) between the H atom adsorbed at the surface and in the gas phase; and T∆S H accounts for the difference in entropy between the final and initial state. The zero point energies, for the gas phase H 2 molecule and for H adsorbed at O, S and Se sites, are derived from computations of the vibrational frequencies, according to the method prescribed in the work of Liao et al [62]. For adsorption, only the vibrations of the H ion and the surface/cluster site (O, S, Se) to which it is adsorbed are considered; the ZPE for the adsorption site is then subtracted from this value. As is typical in such studies, the entropic contributions for H adsorbed at the surface are omitted, and so the value for −T∆S H is taken as half the value for molecular H 2 . Thus, the ZPE and entropic contributions to the free energy are obtained from:

Atomic structure
The relaxed structures of the M 4 X 4 -r110 are shown in figure 2; the computed adsorption energies are included in the insets. The negative adsorption energies indicate that the modifier-surface interaction is favourable and the magnitudes of these energies suggest that the nanoclusters are strongly bound at the surface [32,33,41,42,56]. From these data, we can see that the Ti-S bonds are shorter than the Ti-Se bonds. This is expected as Se has a larger ionic radius than S [8,10]. Nanocluster metal-S bonds are also shorter than metal-Se bonds, both in the gas-phase and after adsorption at the rutile TiO 2 surface. Metal-S bonds are consistently shorter by 0.12-0.16 Å, than equivalent bonds in the selenide structures. These values are in agreement with the ionic radii of S 2− and Se 2− , which are 1.84 and 1.98 Å, respectively [65]. However, despite differences in the composition of the nanocluster modifiers, the adsorption energies are similar in all cases.

Density of states
The PEDOS plots, computed for the M 4 X 4 -r110 heterostructures, are shown in figure 3. The VBM of the titania support has been set to 0 eV and the band gap of the rutile (110) surface, from this computational set-up, is 2.20 eV. After modification, occupied cluster-derived electronic states, which are predominantly chalcogen p states, extend into the energy gap. Occupied states emerge at 1.18, 1.30, 1.33 and 1.57 eV above the titania VBM for modification with Sn 4 S 4 , Sn 4 Se 4 , Zn 4 S 4 and Zn 4 Se 4 , respectively. This is combined with an enhancement of the DOS near the VBM of the titania support due to modification. Moreover, for Sn 4 X 4 modification, Sn-derived states emerge in the energy gap. These states are due to the lone pair, as has been discussed in previous work on Sn chalcogenides [66,67].
To confirm these results, single-point PBE0 calculations, with 25% exact enchange, were performed on the DFT + U relaxed geometries. The resulting PEDOS plots are presented in figure S3 of the supporting information document. While there are quantitative differences in the description of the electronic structure and the extent to which cluster-derived states extend into the titania energy gap, both computational set-ups predict an appreciable redshift in the light abosprtion properties due to metal-chalcogenide modifications. Thus, we may conclude that the modified surfaces exhibit a redshift in the energy gap, with respect to the unmodified rutile (110) surface. Moreover, modifier-derived states near the Fermi level will have important consequences for the HER activity.

Photoexcitation model
The energies computed from the photoexcitation model are presented in table 1. First, the model applied to the bare rutile (110) surface yields a vertical energy, E vert , of 2.03 eV. This quantity is equivalent to the optical band gap and indicates that the underestimation of the energy gap, typical of approximate DFT methods, is present in the current computational setup. Nevertheless, this model lends itself to useful qualitative comparisons between the bare surface and the modified systems.
The values for E vert indicate that modification induces a redshift in the energy gap, as was also shown in analysis of the PEDOS plots. Optical gaps of 1.17, 1.11, 1.58 and 1.30 eV are computed for the rutile (110) surface modified with Sn 4 S 4 , Sn 4 Se 4 , Zn 4 S 4 and Zn 4 Se 4 , respectively. The excitation energy, E exc , is the energy difference between the singlet ground state and the fully relaxed triplet excited state. Each of the modified surfaces exhibit a reduction in this value, with respect to that computed for the bare rutile (110) surface. The computed relaxation energies, E rel , are larger for the modified systems. E rel represents the energy gained by the system after structural changes and relaxation in the excited state and is a measure of the stability of electron and hole localisation.
To take part in photocatalytic reactions, photoexcited charges must separate and migrate to active surface sites and this entails overcoming the electron-hole binding energy, which is not accounted for in this simple model for photoexcitation. However, low-coordinated surface sites can act as charge traps and suppress carrier recombination [68], and this is captured in our model. Moreover, the computed values E rel are useful as a measure of the stability of charge trapping and indicate that this is enhanced upon modification with the metal chalcogenide nanoclusters. Taken together, the values shown in table 1 indicate that modification of  rutile (110) with nanoclusters of composition M 4 X 4 induces a redshift in light absorption and enhances the stability of photoexcited charges. By examining the excess spin density plots, shown in figure 4, in combination with analysis of computed Bader charges and spin magnetisations, we identify at which ions the photoexcited charges localise. For each system, the electrons and holes localise at the surface and modifier, respectively, which promotes charge separation. The yellow isosurfaces show that the electrons localise at sub-surface Ti ions, resulting in a reduction from Ti 4+ to Ti 3+ . This is corroborated by an increase in Bader charge from 1.3 electrons for Ti 4+ to 1.6/1.7 electrons for Ti 3+ and a computed spin magnetisation of 0.94 µ B for the reduced Ti 3+ ion.
For Sn 4 S 4 -r110, the hole state localises predominantly on an S ion and a neighbouring Sn ion. For the S ion, the Bader charge decreases from 6.9 to 6.7 electrons and this ion has a spin magnetisation of 0.29 µ B . For the Sn ion, the Bader charge decreases from 12.8 to 12.6 electrons, indicating some hole localisation on Sn, consistent with the DOS analysis; the spin magnetisation for this Sn ion is 0.19 µ B . For Sn 4 Se 4 -r110, the hole localises on an Se ion, for which the Bader charge decreases from 6.8 to 6.4 electrons; the spin magnetisation for this ion is 0.41 µ B .
For both Zn 4 S 4 -r110 and Zn 4 Se 4 -r110, the hole state is distributed over two chalcogen ions; the S ions have computed spin magnetisations of 0.27 and 0.39 µ B , while for the Se ions the spin magnetisations are 0.32 and 0.35 µ B . For Zn 4 S 4 -r110, the Bader charges for these S ions decrease from 6.9 and 6.8 electrons to Table 2. Free energies (in eV) for H adsorption at surface sites (O) and cluster sites (S, Se) of Sn4S4-, Sn4Se4-, Zn4S4-, and Zn4Se4-r110. The data presented herein follows from the procedure described in the methodology section and summarized in the schematic in figure 1. For each system and coverage, adsorption at surface sites is more favourable than adsorption at cluster sites. For adsorption of H at cluster sites, adsorption free energies in the active range are highlighted in bold.

System
Site 6.7 and 6.5 electrons, respectively. Similarly, the Bader charges for the Se ions at which the hole localises in Zn 4 Se 4 -r110 decrease from 6.8 to 6.5 electrons and from 6.6 to 6.4 electrons. In summary, each of the modified surfaces exhibit a reduction in the optical gap and an enhanced stability of excited charges, with respect to bare rutile (110). Moreover, after modification, the photoexcited electrons and holes are spatially separated; the electrons localise at subsurface Ti sites and the holes localise at chalcogen ions of the supported modifiers.

Hydrogen adsorption
Next, we examine the free energy of H adsorption at the modified surfaces and, as mentioned in the introduction, we consider that these modified materials will promote HER should they exhibit H adsorption free energies close to thermoneutral, i.  N)H, we identify the most stable configuration for the n th H atom adsorbed at a surface site by computing ∆G H relative to a surface coverage of (n − 1)H atoms for each available site, using equation (2). The results of these computations are shown in the rows labelled 'O' for each surface in table 2. For surface coverages of between 0 and N hydrogens, we then examine hydrogen adsorption at all chalcogen sites and computed ∆G H for the most stable chalcogen sites are presented in table 2, in the rows labelled 'X' , (X = S, Se). The adsorption free energy of the n th H atom at a cluster site is calculated relative to the surface with (n − 1)H at surface sites, according to equation (3). Henceforth, we distinguish between 'coverage' and 'surface coverage'; the former refers to the total H coverage, including adsorption sites at the nanoclusters, whereas the latter refers only to H adsorbed on O sites of the rutile (110) support.
For each modified surface, H adsorption at TiO 2 bridging oxygen sites is exothermic for all surface coverages. As the surface coverage increases, approaching full occupation of O br sites, ∆G H decreases. However, for all coverages, H adsorption at surface sites is more favourable than adsorption at chalcogen sites. Thus, for the n th H atom, ∆G H for adsorption at a cluster site is computed relative to the system with a surface coverage of (n − 1)H.
For configurations with a H atom adsorbed at a cluster site with a free energy in the active range, we also investigate ∆G H for adsorption at a second cluster site. The results of these computations are included in the rows labelled 'X + X' (X = S, Se). All adsorption configurations for the data in table 2    In summary, the data presented in table 2 indicate that the sulphide-modified surfaces will be more active in HER, with respect to the selenide-modified surfaces. In general, H adsorption at S sites is more favourable than adsorption at Se sites; adsorption at Se sites is generally highly endothermic. This result is in agreement with a recent study on the stabilisation of hydrogen adsorption on Chevrel-Phase Mo 6 X 8 (X = S, Se, Te) electrocatalysts [69]. Through a combination of experiment and computation, the authors reported that H adsorption strength increased with the electronegativity of the chalcogen (S = 2.58; Se = 2.55) [70]. This result manifests in the electronic structure as a lower p-band centre. The p-band centre is defined as:

Conclusions
Metal chalcogenides have emerged as promising candidates for HER catalysis. Extensive experimental and computational studies of TMDs have revealed that low-coordinated chalcogen sites are active sites for HER. This is attributed to the near optimal adsorption free energy of H at chalcogen sites.
In this work, we have examined, via DFT + U computations, the surface modification of TiO 2 rutile (110) with nanoclusters of composition M 4 X 4 (M = Sn, Zn; X = S, Se). Surface modification strategies aim to combine the desirable properties of the substrate (TiO 2 ) with those of the nanocluster modifiers. In this instance, the M 4 X 4 modifiers provide low-coordinated chalcogen sites, which we have investigated for their HER activity via computations of the free energy of H adsorption.
Our results indicate that the M 4 X 4 modifiers bind to the rutile surface with the formation of interfacial M-O and Ti-X bonds. The modification induces a red shift in light absorption due to the emergence of occupied, nanocluster-derived (predominantly chalcogen 2p) states in the titania energy gap. In addition, modification with M 4 X 4 nanoclusters promotes the spatial separation of photoexcited charges and enhances their stability in the excited state.
Analysis of the free energies of H adsorption, ∆G H , reveals that the sulphide modifiers exhibit values close to thermoneutral, (−0.15 eV, +0.15 eV), for most H coverages, whereas this is only true for the selenide modifiers at low coverages. This is a widely accepted descriptor for HER activity and suggests that sulphide-modification will promote the HER to a greater extent than selenide modification. In general, H binding at sulphur sites is more favourable than at selenium sites; we attribute this to the higher electronegativity of S, with respect to Se. This effect is manifested in the computed p-band centres of the modified systems-the sulphide-modified systems exhibit lower p-band centres than the selenide-modified systems.
In conclusion, modification of titania with dispersed metal chalcogenide nanoclusters has the potential to enhance the HER activity of the titania support. However, despite their desirable photocatalytic properties, chalcogenides can suffer from poor chemical stability [71] and this must be accounted for in the design of practical photocatalysts. While metal chalcogenides can oxidise in aqueous environments, many of the chalcogenide catalysts presented in our recent review of non-metal oxide catalysts for HER exhibit stable performance under HER conditions [8]. Thus, one strategy for preserving the integrity of metal chalcogenide catalysts for water splitting applications will entail the design of a suitable photoelectrochemical cell, in which the OER and HER proceed in separate compartments. Another approach to promote the stability is to deposit very thin, protective layers via ALD or similar techniques [72][73][74].
Careful analysis is required to elucidate the impact of modification on the properties which govern the photocatalytic HER activity. However, rational selection of the composition of the modifiers, in combination with first principles computations of appropriate material descriptors can facilitate high-throughput screening of candidate materials.