Exploring Multi-Anion Chemistry in Yttrium Oxyhydrides: Solid-State NMR Studies and DFT Calculations

Rare earth oxyhydrides REOxH(3–2x), with RE = Y, Sc, or Gd and a cationic FCC lattice, are reversibly photochromic in nature. It is known that structural details and anion (O2–:H–) composition dictate the efficiency of the photochromic behavior. The mechanism behind the photochromism is, however, not yet understood. In this study, we use 1H, 2H, 17O, and 89Y solid-state NMR spectroscopy and density functional theory (DFT) calculations to study the various yttrium, hydrogen, and oxygen local environments, anion oxidation states, and hydride ion dynamics. DFT models of YOxH(3–2x) with both anion-ordered and anion-disordered sublattices are constructed for a range of compositions and show a good correlation with the experimental NMR parameters. Two-dimensional 17O–1H and 89Y–1H NMR correlation experiments reveal heterogeneities in the samples, which appear to consist of hydride-rich (x ≈ 0.25) and hydride-poor domains (x ≈ 1) rather than a single composition with homogeneous anion mixing. The compositional variation (as indicated by the different x values in YOxH(3–2x)) is determined by comparing static 1H NMR line widths with calculated 1H–1H dipolar couplings of yttrium oxyhydride models. The 1D 17O MAS spectrum demonstrates the presence of a small percentage of hydroxide (OH–) ions. DFT modeling indicates a reaction between the protons of hydroxides and hydrides to form molecular hydrogen (H+ + H– → H2). 1H MAS NMR indicates the presence of a mobile component that, based on this finding, is attributed to trapped molecular H2 in the lattice.


■ INTRODUCTION
Rare-earth oxyhydrides are an emerging class of multianion compounds that show a gamut of interesting optical and magnetic properties. 1−6 The heteroanionic sublattice, consisting of oxide (O 2− ) and hydride (H − ) ions, and specifically, the ionic ordering on this lattice control their properties. 4 The anion arrangement has a significant influence on the hydride ion dynamics as well. 7 Metal hydrides are well-known for their hydrogen storage properties, 8,9 whereas rare earth oxyhydrides are utilized as a photocatalyst for ammonia formation 2,6,10 and as H − conductors. 5,11 A recent study on cubic lanthanumbased oxyhydrides attributes this conductivity to the soft and polarizable nature of hydrides that facilitates ionic mobility. 11 The low ionic mass, ample availability, and high polarizability thus make the hydride ion a very versatile anion.
The Sc, Y, and Gd oxyhydrides show color-neutral, reversible photochromism, 12 the mechanism of which is still not well-known. The photochromism is induced by the photoexcitation of electrons of these semiconductors and is characterized by an unusually wide optical range. 13 They are synthesized as thin films that are up to 1 μm thick. On exposure to UV/visible light, these materials change from a yellowish, translucent state to a dark, opaque state. They exhibit a considerable drop in optical transmission after UV illumination, as was first noted by Mongstad et al. 14 The empirical formula for these oxyhydrides follows a specific compositional trend that is in between the trihydrides and oxides with the formula REO x H (3−2x) 13 (assuming a full oxidation of the RE and oxygen and hydrogen to be in the 2− and 1− oxidation state, respectively). They are distinctly different from the hydroxides in terms of band gap, lattice structure, and photochromic properties. 15 These materials are semiconductors, as is confirmed from optical transmission studies ( Figure S1) whose band gap can be tuned by varying the O 2− :H − ratio during deposition. 16 The REO x H (3−2x) oxyhydrides have a FCC cation lattice 17 with the XRD patterns best matching the Fm3m space group of REH 2 . 16,18,19 The anions occupy the tetrahedral and octahedral sites in the lattice (Figure 1), with a preference for the tetrahedral site, mainly because of the favorable lattice energy as compared to the octahedral sites. 18 An EXAFS study has shown that oxide ions have a stronger preference for the tetrahedral sites than hydride ions and, hence, substitute the tetrahedral hydrides during air oxidation (Figure 1b). 18 This is a consequence of a stronger ionic bond of the cation with the oxide, which is further rationalized by calculating the Madelung energy. 18 The precise arrangement of the anions in the lattice remains, however, unknown. The photochromic efficiency is observed to change with the O 2− :H − ratio. The soft and polarizable nature of the hydride ion is in stark contrast with the hard and highly electronegative nature of the oxide ion. It has been observed that with such a contrasting nature of the anions, the local arrangement of the anionsublattice plays a major role in establishing the properties of the material. 20−22 Other relevant structural characteristics are coordination defects and the distribution of vacancies. Hence, understanding the local arrangement and dynamics of the anions in the rare earth oxyhydride lattice is important to comprehend the underlying mechanism of the photochromic effect.
In this work, we use 1 H, 2 H, 17 O, and 89 Y solid-state NMR and density functional theory (DFT) calculations to investigate yttrium oxyhydride thin film structures because NMR spectroscopy is particularly sensitive to the local structural environments of the nuclei. Solid-state NMR studies on various hydrogen storage materials, including binary, tertiary, and complex metal hydrides, have provided information about the local structure, arrangement, and dynamics of the hydrides. 8,23−25 In this work, the different cation and anion environments, their oxidation states, the anion sublattice arrangement, and hydride ion dynamics are investigated. Quantitative 1 H NMR studies provide information on the O 2− :H − ratio on the anion sublattice, which can be useful for correlating the anion composition to the photochromic properties. The NMR findings are further compared with DFT calculations of model yttrium oxyhydride structures, which are constructed for various O 2− :H − compositions (ranging from x = 0.25 to 1.25), including both ordered and disordered anionic arrangements. Our previous NMR study of thin film oxyhydrides 26 indicated the presence of a very mobile hydride component that disappeared upon UV irradiation. More in-depth studies are needed, however, to gain insight into the structural effects leading to the photochromic behavior of these materials. In this study we perform a detailed computational and experimental investigation of these materials in a transparent (as-prepared) state. ■ EXPERIMENTAL SECTION Materials and Synthesis. Yttrium oxyhydride powders were made by first depositing a thin "delamination" layer of Au (∼10 nm) onto a 3 in. glass wafer by magnetron sputtering of a gold target at 25 W and under a flow of Ar (20 sccm) at a deposition pressure of 0.3 Pa. Next, a roughly 1 μm thick hydride was deposited by reactive magnetron sputtering of a Y target (MaTecK, 99.9%) in an atmosphere of Ar/H 2 at a flow ratio of 7:1. The deposition pressure was kept at 0.5 Pa. The input power was 100 W. 19 The as-deposited film is YH 1.9+x , but upon air exposure, the film oxidizes to the oxyhydride phase. As this happens, due to presence of the Au layer, the film delaminates as flakes and can be mechanically scraped from the glass wafer and studied in the NMR as powders.
Thin films of yttrium oxyhydrides were sputtered onto flexible FEP foils (3 in. diameter) by reactive magnetron sputtering of a Y target (MaTecK, 99.9%) and an input power of 100 W. Both YH x O y powders (with the gold flakes) and thin films (with FEP foils) deposited at 0.5 Pa deposition pressure were investigated in this study. The 1 H spectra for these two kinds of samples were measured and compared to verify their compositional similarities which showed good agreement. The aforementioned samples will be referred to as the powder samples and the thin film samples, respectively, in the rest of the text.
For both hydrogenated and deuterated films, the Y metal target was sputtered in the presence of Ar and either H 2 or D 2 (99.9 at. %) gas with a flow ratio of 7:1, and a controlled deposition pressure (0.5 Pa). While some hydrogenated films were oxidized in air, others were oxidized by either dry O 2 or isotopically enriched 17 O 2 (99.9 atom %). To achieve this, the sample was moved from the vacuum chamber to an oxidation cell in the glovebox without exposure to air. After sealing in the cell, >20 mbar of the appropriate gas was introduced, and the sample was left to oxidize for several hours. Yttrium dihydride was sputtered as a reference material in the same manner as the oxyhydrides, except with an input power of 200 W during reactive sputtering, and a deposition pressure of 0.3 Pa. It has been shown that at this pressure, the as-deposited film is too dense to allow oxidation to the oxyhydride phase. 16,18 The H 2 (Ar) deposition pressure determines the concentration of the incorporated O 2 in the lattice, with higher deposition pressures allowing more oxygen to be absorbed by the YH x . Thus, the samples deposited at higher H 2 (Ar) deposition pressures are expected to have a more oxide-rich overall composition. However, the deposition pressure was kept constant at 0.  HXY triple channel Varian probe was used for 1 H MAS and 89 Y CP-MAS 27 experiments. Empty rotor spectra were acquired to subtract proton background signals from the 1.6 mm Varian MAS probe and rotor. The probe was equipped with a low gamma tuning box to tune the 89 Y channel. 1 H single pulse excitation (SPE) and 1 H spin echo spectra were acquired. 2D 1 H− 1 H exchange spectroscopy (EXSY) was performed to study spin diffusion between 1 H sites, using the same equipment as mentioned above. Nuclei with a low gyromagnetic ratio (Y, Rh, W), which have a poor sensitivity 28 can be studied by combining MAS and proton detection. 29 Inversely detected (via the 1 H channel) heteronuclear correlation (HETCOR) experiments were performed to study the 1 H− 89 Y interactions. In some HETCOR experiments, Lee Goldberg Cross-Polarization (LGCP) 30 was performed to suppress 1 H spin diffusion to selectively probe interactions between yttrium and hydrogen that are in close proximity.
Static 1 H NMR experiments were performed on a 300 MHz Varian VNMRS spectrometer using a home-built, singlechannel, static, proton-free probe with a RF coil diameter of 1.6 mm. O-labeled L-tyrosine [S2 (Supporting Information (SI))] was used to optimize the experiments. 34−36 All experiments were performed in a dry nitrogen atmosphere. The solid-state NMR spectra were processed and fitted using the ssNake software package. 37 Detailed information about the acquisition of the NMR experiments can be found in S19 (SI).
DFT Calculations. Electronic structure and chemical shielding calculations were carried out with the Vienna ab initio simulation package (VASP) 38,39 using the projectoraugmented wave (PAW) 40 The gauge-including PAW (GIPAW) 44,45 method was used for the shielding calculations.
The empty 4d orbitals of Y are typically placed at too low energies by semilocal DFT. This can lead to an overestimation of the covalency of the Y−O bond and, therefore, substantial errors in calculated shieldings. This error can be repaired with the DFT+U method. 46 We checked the effect of a reasonable Hubbard U of 2.8 eV and observed only minor effects on the calculated chemical shifts [<3%, see S26 (SI)]. Hence, for our purposes, we do not need DFT+U and do not apply it here.
The Kohn−Sham orbitals were expanded in plane waves with a kinetic energy cutoff of 600 eV. The Brillouin zones were sampled with 8 × 8 × 8, 2 × 2 × 2, and 1 × 1 × 1 Γ-point centered k-point grids for structural optimization of the ordered structures, small cell, and large cell disordered structures, respectively (vide infra). Increasing the supercell volume results in a smaller Brillouin zone. Therefore, a larger cell requires fewer k-points to achieve the same k-point mesh density. A test was done for the YHO (x = 1 for YO x H (3−2x) ) disordered structure with 6 × 6 × 6 k-points to ensure the convergence of the lattice and NMR parameters.
Standard VASP PAW potentials were used for yttrium, oxygen, and hydrogen (these have default cutoff energies of 203, 400, and 250 eV, respectively). The Y and O potentials have [Ar]3d 10 and [He] frozen cores.
An yttrium FCC backbone with different heteroanionic ratios and arrangements was constructed. Previous studies have shown a clear preference of oxygen tetrahedral sites. 18 Hence, we first placed all oxygen atoms in the tetrahedral sites and filled the remaining tetrahedral sites with hydrogen atoms. The remaining hydrogens occupy the octahedral sites.
Both ordered and disordered anion sublattice models were used. We introduced most of these in ref 50, where complete details can be found. Below we briefly summarize the structural models. Because here we use a different kinetic energy cutoff and Brillouin zone sampling than in ref 50, for consistency, we reoptimized the structural parameters, giving only very small differences.
The anion ordered structures were constructed to investigate the nature of ordering of heteroanions in a CaF 2 type FCC unit cell (containing 4 Y atoms). The lattice constants obtained after cell optimization ranged from 5.23 to 5.34 Å depending on the composition, as the lattice constant increases with an increasing number of oxygen atoms in the structure. The formula YO x H (3−2x) was utilized for constructing the unit cell, with varying the x values as 0.25, 0.5, 0.75, 1, and 1. 25. Several space groups were tested for each composition to obtain different ordering schemes of the oxides and the hydrides. In the end, the lowest lattice energy structures [after structural relaxation; S3 (SI)] were selected.
The DFT calculations for the disordered anion lattice were divided into two groups. The first group consists of special quasi-random structures (SQS). 47 These have a FCC lattice of yttrium containing 32 atoms, i.e., a supercell consisting of 2 × 2 × 2 conventional FCC cells and a lattice constant of 10.65 Å. The anions occupy the tetrahedral and octahedral sites and observe the constraints outlined above; i.e., all oxygens are on tetrahedral sites, all tetrahedral sites are occupied, and any remaining hydrogen is put at the octahedral positions that are not all filled. The total number of anions depends on x, which was varied as 0.25, 0.5, 0.75, 1, and 1.25. SQS structures are ordered structures that closely reproduce a perfectly random arrangement of the anion sublattice for the first few shells Here, the anions are put randomly on their sublattice while still observing the constraints outlined above. These supercells are introduced for the first time in this work and only for x = 0.25 and 1, which correspond to hydride-rich and hydride-poor compositions, respectively. These two compositions were best matched to the compositional variation obtained via NMR experiments (vide infra). In these cells, the disorder extends to larger distances than in the smaller SQS models. The large cells have NMR parameters in close correspondence with those of the smaller SQS cells [S20 (SI)], but offer improved statistical quality for, in particular, the Y coordinations. The structures were relaxed using an electronic convergence threshold of 10 −6 eV. The convergence criterion for structural optimization was 0.001 eV. First, the atomic positions, shape, and size of the YH (3−2x) O x cells were relaxed. Subsequently, these well-converged structures were used for NMR shielding calculations.
The references used for the Y, O, and H nuclei are yttrium hydroxide (Y(OH) 3 ), yttrium oxide (Y 2 O 3 ), and adamantane, respectively. The calculated shieldings, σ cal , are converted to chemical shifts, δ cal , by using the following equation: The second moments were calculated using the Van Vleck equations 48,49 and multiplied with √(8 ln (2)) to obtain the theoretical line widths for a rigid yttrium hydride (YH 2 and YH 3 (Table 1) and are denoted peaks I, II, and III, respectively. The resonance (II) at 4.6 ppm is very narrow, meaning that the dipolar interactions are extremely small. Therefore, this resonance is due to either very isolated protons, which is unlikely in these materials, or a highly mobile H species. Moreover, peak II has a very long spin−spin relaxation time (T 2 ) compared to the broad peaks (S4), which indicates that this peak has a weak dipolar interaction with the neighboring spins. Consequently, peak II can be attributed to a highly mobile species for which the dipolar interactions to neigboring spins are averaged. The 1 H chemical shift information is insufficient to assign this peak to an exact chemical species. However, subsequent 17 O NMR spectroscopy and DFT modeling indicate that this resonance could be due to molecular hydrogen trapped in the structure (vide infra).
The lines at 2.8 ppm (I) and 4.8 ppm (III) have a substantial line width. Experiments as a function of external field strength [S5 (SI)] show that this line width is to a large extent   50 Therefore, we rule out the presence of ordered structures. The ideal anion disordered sublattice is expected to contain two distinctly different hydrides at the tetrahedral and octahedral sites. However, DFT calculations on anion disordered models suggest that the hydrides are considerably displaced from their respective tetrahedral and octahedral positions after structural relaxation, thus attaining positions at intermediate locations between the ideal octahedral and tetrahedral lattice sites. The proximity of the octahedral and tetrahedral hydrides results in this displacement that is described in further detail in ref 50. As a result, the 1 H chemical shifts of the originally tetrahedral and octahedral hydrides strongly overlap after lattice relaxation [S6 (SI)]. Therefore, the observed 1 H peaks in the spectrum cannot be assigned to separate tetrahedral and octahedral 1 H sites. Previous NMR studies on metal hydrides point to a direct proportionality between 1 H chemical shift and metal to hydride distance. 3 Therefore, the distribution in 1 H chemical shifts obtained for components I and III reflects the distortion of the disordered anion sublattice. Hence, each of the components I and III contain contributions from hydrides displaced from both tetrahedral and octahedral sites. This is reflected in the chemical shift calculations for different compositions represented in Figure 3. 2 H MAS NMR was employed to overcome the effect of residual 1 H dipolar couplings and to obtain well-resolved hydrogen chemical shifts 51−53 as 2 H NMR does not suffer from substantial dipolar broadening because of its low gyromagnetic ratio (compared to 1 H). 2 H NMR studies of YD x systems (1.98< x < 2.08) have shown enhanced hydride ion mobility with increasing temperature and exchange between tetrahedral and octahedral sites. 54 Figure 4a,b displays the 2 H MAS NMR spectrum of a 2 H isotope-labeled yttrium oxyhydride thin film. The deconvolution of the spectrum included the spinning sideband manifold reflecting the first-order quadrupolar interaction (Figure 4a). The spectrum corresponds to three overlapping resonances (Figure 4b) at positions closely corresponding to the 1 H MAS spectrum ( Table 1). The appearance of the narrow component (peak 2) in the 2 H spectrum confirms the presence of a mobile component as a part of the structure. Component 2 has no detectable quadrupolar interaction that again hints at averaging of anisotropic (now quadrupolar) interactions due to mobility. The MAS line widths for both peaks in ppm are comparable (both ranging from 1 to 10 ppm), which confirms that they are dominated by a distribution in chemical shift.
To summarize, based on the line width of the 1 H and 2 H MAS NMR spectra, which originate from chemical shift distributions, we conclude that there is substantial disorder in the distribution of the anions in the lattice. This is corroborated by DFT calculations of anion-disordered oxyhydrides. The lattice relaxation of these model structures further increases the disorder by displacing the hydride ions from their ideal octahedral and tetrahedral positions. The 2D EXSY spectrum clearly shows that there are three 1 H chemical  environments that are not in close contact with each other. Except for peak II, peaks I and III are dominated by a distribution of chemical environments because of the disorder. The DFT calculations (Figure 3) indicate that each composition of the oxyhydrides displays a distinct chemical shift distribution, showing a deshielding trend as x (in YO x H (3−2x) ) increases. These distributions overlap when comparing different compositions due to their large width. Therefore, the spatially separated 1 H chemical environments observed experimentally could be due to a compositional variation in the anionic arrangement, i.e., each of the 1 H peaks might correspond to a domain in the sample with a specific x value. These results exclude a uniform composition with a single x-value with random arrangement of the hydrides and oxides, in which case a single close to Gaussian distribution would be observed ( Figure 3) and this is evidently far from our experimental findings. The cross-sectional SEM studies of a 270 nm yttrium oxyhydride thin film reported previously by Nafezarefi Figure 5 shows the presence of a narrow Gaussian component and two broad Gaussian components, as is shown in the deconvolution in Figure 5 (bottom). For the deconvolution, the integral ratios were restrained to those obtained from the 1 H MAS spectrum ( Table 1). The three Gaussian contributions are observed to have distinct line widths of 42.1, 23.5, and 2.9 kHz, which are associated with peaks III, I and II of the 1 H spectrum ( Figure  5) respectively. Below we will correlate this to the calculated second moments obtained from our DFT model yttriumoxyhydride systems using the van Vleck equations for dipolar line widths 48,49 (see Table 2). The mobile narrow fraction (peak II), which has been noted in the previous static 1 H NMR experiments, 26 accounts for nearly 10% of the total peak integral. Static line widths of peaks I and III in the absence of MAS, can be assumed to be predominantly determined by 1 H− 1 H dipolar couplings, if the lattice is rigid. In that case, there is no effect of mobility that could partly average the dipolar interactions (and thus reduce the line widths).
Comparing the DFT calculated 2 H quadrupolar coupling constants C Q [S16 (SI)] with experiments (Table 3), the 2 H quadrupolar coupling constant data show no evidence of mobility for the two broad components, as the experimental values agree well with the calculations for static lattice models. This is further confirmed by low temperature 2 H measurements [S7 (SI)] in which the line widths of the resonances remain independent of temperature. Therefore, the experimental static 1 H line widths can be interpreted to be dominated by the 1 H− 1 H dipolar interactions in a rigid lattice. The contribution from chemical shift distributions, as observed in 1 H MAS spectra (vide supra), is minor compared to the size of the dipolar couplings. Table 2 shows the calculated 1 H line widths of the model structures for a range of x values. We observe that the 42.1 kHz Gaussian peak closely matches the calculated line width for the model structure with x = 0.25, whereas the 23.5 kHz peak compares well to the x = 1 model structure, both having a disordered anion arrangement. Therefore, we attribute the two broad Gaussian resonances to a hydride-rich (x ∼ 0.25) and a hydride-poor (x ∼ 1) part of the anion sublattice. For the 1 H MAS spectrum, this implies that the 4.8 ppm peak (III) corresponds to a hydride-rich domain and the 2.8 ppm peak (I) corresponds to a hydride poor domain. This is, however, contradictory to the trend that is observed for the calculated 1 H chemical shift distribution (Figure 3), where the hydride-poor structures (x ≥ 1) show higher chemical shift values compared to the hydride-rich    , giving an average composition of x ∼ 0.55, which comes close to number obtained from the quantitative 1 H NMR experiments. So we conclude that the sample consists of hydrogen-rich and hydrogen poor domains in a ratio of 3:2 and that its average composition is approximately YO 0.5 H 2 17 O Solid State NMR and DFT Calculations. 17 O solidstate NMR has been extensively employed in recent years, specifically because of its wide chemical shift range and the high sensitivity of its quadrupolar coupling constant to the local symmetry. 55−59 Here we use 17 O NMR to obtain a deeper understanding of the structure of the anionic sublattice. The 17 O NMR spectrum for the isotope labeled yttriumoxyhydride thin film (Figure 6a) shows the central transition (CT, expanded in Figure 6b) and the spinning sideband manifold of the satellite transitions due to the quadrupolar interaction. The CT resonances cover a chemical shift range from 300 to 475 ppm, which is in the same range as the 17 O spectra of yttrium oxide, 60 yttrium stannate 61 and yttrium titanate. 61 The peaks can therefore be assigned to oxides in the anion lattice. The calculated 17 O chemical shifts for anion disordered SQS model structures with oxygen in tetrahedral sites (S9) indicate a similar chemical shift range. To confirm the assignment of the oxide resonances to tetrahedral sites, a model structure containing oxide ions in octahedral sites was simulated (S10). The calculated 17 O chemical shifts for these octahedral oxides are significantly lower (248 ppm) and hence are distinctly different from those of the tetrahedral sites. Moreover, it is observed in the DFT calculations and also noted in our previous theoretical study 50 that, unlike the hydrides, the oxides are not displaced significantly from their tetrahedral sites after lattice relaxation and therefore resonate in a different chemical shift range than the octahedral oxides. The spectrum in Figure 6a is deconvoluted, including the spinning sideband manifold of the satellite transitions (full deconvolution shown in S15) and assuming a Czjzek distribution of the quadrupolar parameters. The resulting fit shows three Gaussian lines centered at 355, 375, and 402 ppm for the central transition (Figure 6b). All of the resonances are observed to have a relatively small quadrupolar coupling constant C Q ( Table 3). The small C Q observed for the oxides hint at a rather symmetrical local environment, as expected for (near) tetrahedral coordinations. As a result, the second-order quadrupolar line broadening of the CT is negligible. This was verified by measuring at a different magnetic field strength (14.09 T) and comparing the line widths. The second order quadrupolar line broadening scales inversely with external magnetic field. In this case, however, the two spectra, as shown by the spectral overlay in S11, have nearly identical linebroadening (in ppm) showing a chemical shift distribution rather than quadrupolar broadening. 62−64 This chemical shift distribution thus originates from the variation in the outer coordination shells which is due to both a disordered anionic sublattice and displacement of the hydrides from their ideal positions.
To correlate these experimental findings with the DFT models, a simulated 17 O spectrum was constructed using large, anion disordered cells for the compositions x = 0.25 and 1, with a domain size ratio of 3:2 (vide supra; S22), as obtained from the 1 H spectra ( Table 1). The comparison shows a satisfactory correspondence between the experimental and simulated spectra. However, the calculated 17 O chemical shift distributions are slightly deshielded compared to the experimental values, possibly due to small deviations from the predicted compositions (x = 0.25 and 1). Moreover, DFT Interestingly, an additional resonance is observed at ∼110 ppm, which is most clearly visible in the spectrum acquired at a spinning speed of 15 kHz (Figure 6a, top). DFT calculations suggest (S12) that this peak belongs to a hydroxide oxygen. Earlier NMR studies have noted that a metal hydroxide oxygen appears to be more shielded than the corresponding oxygen in a metal oxide. 57,65 Apparently, this happens for the oxides in the oxyhydride lattice as well. The hydroxide peak contribution can be determined quantitatively from the integral by selectively exciting the central transitions 63 and was found to be nearly 7.2% of the total integral value, as obtained using the quadrupolar fitting tool in ssNake 37 (S21).
Based on these results, DFT models with OH − groups in the anion sublattice were constructed for both the hydride-poor and hydride-rich structures (as detailed in S17). In the case of the hydride-poor structure (x ∼ 1), all the hydroxide groups were retained after lattice relaxation (Figure 7b), whereas for the hydride-rich structures (x ∼ 0.25), the lattice relaxation led mostly to the formation of H 2 molecules, apart from a few remaining hydroxides. The H 2 formation mainly occurs in the hydride-rich lattice (x = 0.25) due to the presence of nearby octahedral hydrides (Figure 7a), which are absent in the hydride-poor structures (x ∼ 1). The presence of OH − and H 2 has a significant deshielding effect on the 1 H chemical shift distribution for x ∼ 0.25 (Figure 8), which results in a strong overlap of the simulated 1 H resonances of x = 0.25 and 1 compositions. Hence, it substantially resolves the initial contradiction between the calculated and experimentally observed trends for the 1 H shift as a function of x ( Figure  3). However, introduction of OH − groups in the lattice might have effects on the overall composition of the oxyhydrides, making it more oxide rich (due to remaining oxides after H 2 formation). Therefore, we cannot exclude the presence of intermediate compositions, especially in the hydride-rich domain. The hydride-rich domains in the presence of trapped H 2 (originating from OH − only) might possibly have a composition in the range of 0.25 < x < 0.50, while still maintaining a hydride-rich composition. There could, however, be other sources for the formation of molecular H 2 in the lattice, such that the overall composition does not change. For example, H + from absorbed moisture could combine with H − in the anion lattice to form H 2 . The quantitative prediction of  The Journal of Physical Chemistry C pubs.acs.org/JPCC Article the origin of the observed chemical species is beyond the scope of our current studies. The effect of the OH − incorporation does not have a large impact on the yttrium and remaining oxide chemical shifts (S12 and S13). The H 2 formation in the model structures matches well with the observation of a narrow peak at 4.6 ppm in the 1 H NMR spectra (Table-2) and is therefore assigned to trapped H 2 molecules in the lattice. Indeed, several NMR studies show that trapped H 2 in metal hydrides is extremely mobile and in many systems occurs at chemical shift values from 4.3 to 5.0 ppm. 8,66−69 Nevertheless, we discuss other possible assignments of the narrow component (II) in the 1 H NMR spectrum. As we deduced from the 17 O spectrum, traces of hydroxide (OH − ) groups are present; therefore, they should also appear in the 1 H spectrum. However, their intensity is expected to be very low. The OH − groups have a relatively integrated intensity of 7.2% in the oxygen spectrum. Using the formula YO x H (3−2x) and a 3:2 ratio of domains of composition x ∼ 0.25 and 1.0, we can calculate the corresponding percentage of OH − in the 1 H spectrum to be approximately 2% (S21). This is much less than the relative intensity of 10% of the narrow peak (II) in the proton spectrum. Hence, we can discard the option that this resonance comes from the OH − groups. Moreover, the protons in the hydroxyl groups are expected to be less dynamic, as shown in previous studies, 70−72 and are therefore less likely to give a very narrow line.
Another possible assignment for peak II is H 2 − formation, as it has been found to be thermodynamically stable in particular cases. 73 There is insufficient data to identify the properties of such species, however, and hence, its presence remains ambiguous. Therefore, current results for H 2 formation in the anion lattice are the most likely assignment of peak II.
To gain insight into the proximity of different oxygen and hydride species, 17 O− 1 H PRESTO recoupling experiments were carried out. This pulse sequence is affected by the socalled dipolar truncation effect, 74 meaning that polarization transfer from 1 H to distant 17 O nuclei is attenuated and one therefore mainly probes close proximities, i.e., the 17 O resonances that are relatively near to 1 H show a faster buildup of signal intensity. Figure 9 shows the spectra of the 17 O ( 1 H) PRESTO for three different recoupling times (200, 400, and 800 μs). As expected, the hydroxyl oxygen signal (at ∼110 ppm) is very strong already at short recoupling times because of the very close proximity of the hydroxide (OH − ) proton. For the oxides (O 2− , between 300 to 475 ppm), we observe that the (convoluted) resonance moves to higher chemical shift values (deshielding) as the recoupling time is increased. This indicates that the more deshielded 17 O resonances (peak iii, Figure 6b) have a weaker correlation with hydrides than the more shielded ones (e.g., peak i, Figure 6b). In other words, the 17 O resonances are shifted to higher field (more shielded) as the hydride concentration in their immediate surroundings is larger. This difference in signal build-up for the oxygen resonances substantiates our interpretation that the samples consist of hydride-rich and hydride-poor domains. For a completely disordered anion lattice with a fixed composition, all of the oxides would have the same average dipolar coupling with the hydrides, as the hydride concentrations around them would be similar. Moreover, spin diffusion would be active to distribute the 1 H polarization uniformly due to strong 1 H− 1 H homonuclear dipolar couplings, rendering the couplings to 17 O indistinguishable. This is clearly not the case. The net effect is that if x increases, the entire spectrum moves toward lower ppm values (higher shielding). Unfortunately, due to these various effects, it is also not possible to assign specific peaks in the 89 Y spectrum to specific coordinations or domains with a different composition. The spectrum of different domains consists of various overlapping peaks, and domains with different composition have strongly overlapping spectra ( Figure  12a,b). However, it is clear that the yttrium resonances at lower ppm values (higher shielding) experience a weaker coupling to protons, indicating a major contribution from hydride-poor domains. This is apparent from Figure 12, which shows that an increased 89 Y intensity on the shielded side of the 89 Y projection over a spectral area corresponding to higher 1 H shielded area is reflected in DFT calculations for a higher x. The yttrium resonances at higher ppm values are correlated more prominently to the deshielded 1 H spectral area, which indicates that these resonances have a higher contribution from the hydride-rich domains; again, this is reflected in the DFT calculations as well.
We can also extract information about the existence of domains with different compositions by looking at the efficiency of the cross-polarization (CP) process between 89 Y and the protons in the sample. The evolution of the CP signal intensity with the contact time is characterized by a build-up that is determined by the strength of the dipolar interaction (i.e., distance and number of spin pairs involved) with a subsequent (T 1ρ ) decay that depends on dynamics in the kHz regime. 75 As shown in Figure 10b, the 89 Y CP signal builds up more slowly for the more shielded peaks (centered at 210 and 120 ppm) compared to the less-shielded peaks (centered at 350 and 275 ppm). The CP build-up is slower for regions containing less hydrides (i.e., a hydride-poor region; x ∼ 1) compared to hydride-rich domains (x ∼ 0.25). As described above the DFT calculations show that yttrium nuclei in the hydride-rich domains are overall more deshielded (higher chemical shift values), hence, the faster signal build-up in the deshielded region of the CP spectra. Despite the strong overlap of the spectra of different domains, the relative contributions are different on moving from a more shielded region to a less shielded region (see Figure 10b). This confirms the observations from the 2D HETCOR spectrum described above extracting 89 Y projections over the 1 H shift region from the hydride-poor and hydride-rich domains (Figures 12 and  S24).
To even further strengthen this analysis, 2D 1 H− 89 Y HETCOR spectra, using Lee Goldburg CP, were obtained at different (4 and 10 ms) LGCP contact times (Figure 11b). At shorter contact times, signals from strongly dipolar coupled nuclei dominate, whereas at longer contact times signals with a weaker dipolar coupling become more prominent. We see a clear shift from the deshielded to the shielded region for both the 89 Y and 1 H shifts, corresponding to the hydrogen rich and poor domains, respectively, in line with the prediction by the DFT calculations for 89 Y. The fact that the more shielded proton resonances correspond to the hydride-poor domains is contrary to the initial DFT calculations, but explainable when the effects of OH − incorporation were taken into account (vide supra).

■ CONCLUSION
In summary, the NMR experiments and DFT calculations provide new insights in the structure of yttrium oxyhydride, in particular, regarding the arrangement of the heteroanions. The presence of hydride-rich and -poor domains was established. The domain formation could be a result of unequal pore sizes in the yttrium hydride films, leading to different concentrations of oxygen incorporated in different regions of the thin films upon exposure to air/oxygen. Moreover, the thickness of the films (1 μm in this work) might contribute to the observed compositional variation, but a direct correlation cannot be established from the current studies. Although we modeled our experiments mostly as a binary system of compositions x ∼ 0.25 and 1, we cannot rule out the presence of intermediate compositions, especially for the hydride-rich domains. Further insight regarding the anion arrangement might be gained from knowledge of the precise domain sizes, which was not studied in detail here. Nevertheless, the presence of hydride-rich and -poor domains strongly indicates a lower propensity for  The Journal of Physical Chemistry C pubs.acs.org/JPCC Article complete mixing of the hydrides and the oxides, which could be due to the contrasting nature of the two anions in the multianion lattice. Note that since we deposit a dihydride, the formation of an x ∼ 0.25 compound requires the addition of hydrogen to the lattice during exposure to air. Alternatively, it could suggest phase segregation on oxygenation. Previously, Hans et al. 76 have reported the occurrence of dual phases in gadolinium oxyhydride thin films, where they noted the presence of Gd 2 O 3 and GdH 2 regions. Here we do not observe such drastic compositional variation, as the 89 Y and 2 H NMR do not show any evidence of presence of metallic YH 2 . 54,77 The combination of 17 O NMR and DFT modeling proves to be an important tool in understanding not only the environments and oxidation state of the oxides but also allowed extraction of the oxidation states and different chemical species of hydrogen. Quantitative studies on the presence of both neutral (H 2 ) and hydroxide (OH − ) hydrogen species help in the assignment of the remaining 1 H peaks as hydrides (H − ). The concentration of the trapped H 2 in the lattice may have significant influence on the dynamics and efficiency of the photochromic nature of the materials, as our previous study showed that a mobile fraction of the H atoms plays an important role in the photochromism. 26 The presence of hydroxyl groups and trapped hydrogen in the lattice was established and quantified by 1 H and 17 O NMR studies. The separation of the 17 O and 1 H spectra for the hydride-rich and hydride-poor domains could not be realized because of the substantial overlap of the chemical sites of the regions owing to the complexity of the anionic arrangement.
Although showing a lot of overlap, 89 Y spectra for both domains were extracted, and an agreement with the simulated 89 Y spectra was obtained. 1 H and 2 H NMR studies, along with DFT calculations, are combined to study hydride ion dynamics, which show minimal or no dynamics for the hydrides, whereas the trapped hydrogen displays very high mobility.
The mechanism for photochromism is far from settled. At its heart is the electron−hole pair generated by the bandgap excitation. 81 Presumably, the excited electron reduces the RE 3+ while the hole oxidizes the H − -ion, in particular the hydrogen ion at the octahedral vacancy seems prone to this. 82 Probably some (filamentary) clustering of the reduced rare earth metal takes place, which causes the optical absorption. 50,83 However, the nature and the role of the oxidized H − are still unclear. Solid-state NMR studies of more suitable metal oxyhydrides could be useful in getting more insight into this phenomenon. We are currently pursuing solid state NMR of scandium oxyhydride thin films for such studies. As 45 Sc is 100% abundant and has a higher gyromagnetic ratio than 89 Y, it is favored in terms of sensitivity.