Composition–structure–property relationships of transparent Ca–Al–Si–O–N oxynitride glasses: The roles of nitrogen and aluminum

We explore the formation and composition–structure–property correlations of transparent Ca–Al–Si–O–N glasses, which were prepared by a standard melt-quenching technique using AlN as the nitrogen source and incorporating up to 8 at.% of N. Their measured physical properties of density, molar volume, compactness, refractive index, and hardness—along with the Young, shear, and bulk elastic moduli—depended roughly linearly on the N content. These effects are attributed primarily to the improved glass-network cross-linking from N compared to O, rather than the formation of higher-coordination AlO 5 and AlO 6 groups, where 27 Al magic-angle-spinning nuclear magnetic resonance experimentation revealed that aluminum is predominately present in tetrahedral coordination as AlO 4 units. Yet, several physical properties, such as the refractive index along with the bulk, shear, and Young’s elastic moduli, increase concomitantly with the Al content of the glass. We discuss the incompletely understood mechanical–property boosting role of Al as observed both herein and in previous reports on oxynitride glasses, moreover suggesting glass-composition domains that are likely to offer optimal mechanical properties.

9][20][21] Including Al in the melt also favors the preparation of more transparent oxynitride glasses. 4,18,21ost of previous studies on silicate-based oxynitride systems focused on their glass forming region and composition-property relations, as reviewed in Refs.[3-5].These studies have provided abundant evidence that many physical and mechanical properties-such as the glass transition and crystallization temperatures, viscosity, hardness, and elastic moduli of an M-(Al)-Si-O-N glassenhance concurrently with its N content.4][5] As discussed further in Section 3.2, the Al speciation of oxide/oxynitride glass networks may also involve variable amounts of AlO 5 and AlO 6 polyhedra, that is, higher-coordination Al [5] and Al [6] species, respectively.
The present article examines a plethora of physical properties of transparent Ca-Al-Si-O-N glasses with variable N contents and Ca:Al:Si molar ratios; this encompasses data on the glass density (ρ), molar volume (V m ), compactness (C; "atom-packing density"), refractive index (RI), nano-indentation-derived hardness (H) and reduced elastic modulus (E r ), along with the Young (E), shear (G), and bulk (K) elastic moduli.We discuss the established composition-property relations along with their composition-structure counterparts, as assessed by 27 Al magic-angle-spinning (MAS) nuclear magnetic resonance (NMR) experiments.The latter also complements the hitherto relatively sparse solid-state 27 Al NMR reports on M-Al-Si-O-N glasses, 15,27,38,48,49 where we are not aware of similar studies on Ca-based oxynitride glasses.
The Ca-Al-Si-O-N glasses considered herein are compiled in Table 1 and conform to either of two series, I and II, the respective members of which are labeled by "I-n" and "II-n", where n denotes the at.% of N in the glass.The stoichiometric amount of element E is denoted by  Ē but, for reasons of convenience, normalized such that the sum over all glass constituents is 100 mol, such that the  Ē value becomes equal to the at.% of E; see Table 1.The glasses of each series feature near-constant n Al /n Ca and n Ca /(n Al + n Si ) molar ratios.The members of series I exhibit n Al /n Ca ratios of 1.90-1.98along with n Ca /(n Al + n Si ) = {0.22,0.24}, whereas the corresponding ratios of series II are 1.82-1.85and ≈0.33 (Table 1).The other main distinction between the two glass branches concerns their relative Al and Si contents, where series I and II feature n Al /n Si < 1 and n Al /n Si > 1, respectively.Note that besides the progressively increasing N content within each glass branch, the n Al /n Si ratio also varies among its members (Table 1): The n Al /n Si values of series II are uncorrelated with the corresponding N contents, whereas the I-n glasses with n < 6.5 reveal concurrently increased n Al /n Ca and n N parameters.

Glass preparation
All Ca-Al-Si-O-N glasses were prepared from 600 g ballmilled mixtures of high-purity (99  precursor mixture was placed in a sealed quartz crucible and heated sequentially in air up to the final temperature of 1650 • C. The melt was held for 90 min and poured into a pre-heated graphite mold (200 • C) to relieve internal stresses.Each glass was subsequently annealed for 2 h at 700 • C, which is well below the glass transition temperature.
The amorphous nature of each glass specimen was verified by powder X-ray diffraction, using a Panalytical X'pert Pro MPD diffractometer and Cu K α radiation (λ = 154.1 pm).The glass compositions were obtained by an electron microprobe analysis (EMPA) with a JEOL 8500F instrument operating at 12 kV and 30 nA.Each analyzed glass composition listed in Table 1 was obtained as an average over 10 separately analyzed points with an uncertainty within 1 at.% of each element.The cation compositions were obtained using standards of α-Al 2 O 3 for Al 3+ and CaSiO 3 for Si 4+ /Ca 2+ .The N contents were measured using an LDE1-type diffracting crystal as detector for EMPA, which was calibrated with a CVD Si 3 N 4 standard.The glass compositions listed in Table 1 were chargebalanced by scaling the as-measured anion contents to match the net positive charge of the cations.

Solid-state NMR experiments
All 27 Al (spin 5/2) solid-state NMR experiments were performed with a Bruker Avance-III spectrometer at a magnetic field of 14.1 T, which corresponded to a −156.4MHz 27 Al Larmor frequency.Glass powders were filled in 3.2 mm zirconia rotors that were spun at 24.00 kHz.
The 3QMAS NMR spectra were used for determining the average values of the isotropic chemical shift, δ[] iso , and quadrupolar product, C[]  , of each detected 27 Al [p] (p = {4, 5}) species according to standard procedures. 23The  in the structure: where C Q = e 2 qQ/h, and η j are the quadrupolar coupling constant and asymmetry parameter of the electric field gradient tensor of Al []  , respectively. 23The 3QMAS NMR spectra were also employed for estimating the relative Al [4] and Al [5] populations by integrating their respective 2D NMR peak volumes.We note that because the latter depend on the (average) quadrupolar products, systematic errors result for 27 Al sites with widely differing quadrupolar products, as reviewed in more detail by Edén. 23This dependence of the 3QC reconversion process on the quadrupolar parameters is pronounced for the earliest 3QMAS incarnation that relied on a single strong RF pulse, which is known to yield progressively underestimated relative site populations for increasing quadrupolar products.However, for the FAM technique 52

Physical property measurements
Densities were measured by the Archimedes method on glass monoliths in water at 22 • C with ρ(H 2 O) = 0.998 g/cm 3 .Each of three glass monoliths was measured five times, from which the average density and its standard deviation (±0.01 g/cm 3 ) were calculated.
From the density data and the glass compositions, the molar volume  m = ∕ (uncertainty ±0.02 cm 3 /mol) and the glass compactness (atom packing density; ±0.001),  =  m [calc]∕ m , were calculated from the average molar mass of the glass,  = ∑      , and where   =   ∕( Ca +  Al +  Si +  O +  N ) is the atomic fraction of element E,  A is Avogadro's number, and   = 4 3  /3 is the ion volume for the radius  3  .We used the Shannon-Prewitt ionic radii 53 for the following (average) coordination numbers { Ca , Al ,  Si ,  O ,  N } = {7, 4, 4, 2, 4} throughout.
The (nano)hardness and the reduced modulus of the Ca-Al-Si-O-N glasses were measured by nanoindentation, using a NanoTest Vantage instrument (Micro Materials, UK).A standard Berkovich diamond tip at 20 mN load was employed, which was calibrated using fused silica.Each glass specimen was measured 12 times, and the average values of H and E r and their standard deviations were calculated (Table 2) according to the procedure of Oliver and Pharr, 54 using the unloading elastic part of the load-displacement curve, along with the expressions  =  max /A and  r = (dP∕dh)∕(2) − 1.Here, P max and A are the maximum indentation load and the corresponding contact area, respectively, ∕ℎ is the slope of the indentation unloading curve, and  = 1.096 is a constant associated with the Berkovich diamond indenter. 54,55he Young modulus along with the shear (G), and bulk (K) moduli, as well as Poisson's ratio (ν), were determined by ultrasonic echography on polished monoliths at room temperature, and using the following expressions 44,56 : The elastic moduli were obtained by using an ultrasonic echography piezoelectric transducer (10 MHz) driven by shock excitation and a pulse-receiver (Model PR 35, JSR Ultrasonics, USA).Silicon oil was used as a coupling agent between the specimen, the delay rod, and the transducer.The thickness of each specimen (L) was determined using a digital micrometer with an accuracy of ±2 μm.The ultrasonic velocities of longitudinal (V l ) and transverse (V t ) waves were calculated from the thickness and transit time values.The time-interval separating two successive echoes (τ) is related to the propagation velocity V t by τ = 2L/V t where L is the thickness of the sample.

Glass formation
The photographs of the Ca-Al-Si-O-N glasses from series I and II shown in Figure 1 reveal that all glasses comprising  up to ≈7 at.%N are optically transparent, although their transparency tend to decrease for increasing N content.
In contrast, glasses with N contents >7 at.% were optically translucent (Figure 1), where we observed that the quartz crucible reacted with the melt whenever the AlN precursor (used as the N source) content increased beyond 8 at.%.Previous reports on Ca-(Al)-Si-O-N 30,39,42,44 and other M-(Al)-Si-O-N 31,32,35,36,38,57 glasses demonstrated that high amounts of N and glass-network modifier contents can be incorporated by using fine powders of either metals or metal hydrides.Unfortunately, such preparation procedures lead to impurities of metallic silicides and/or elemental Si particles that presumably accounted for the gray/black and non-transparent glasses. 30,38,469][60] Yet, most of the previously reported N-rich oxynitride glasses were prepared in crucibles of either metal, alumina, boron nitride, or graphite, all of which could react with the melt and thereby result in translucent glasses, as noted in Refs.[11, 20, 47, 61].Support for this proposal is given by recent RF magnetron sputtering-derived transparent Mg-Si-O-N and Ca-Si-O-N thin films, notwithstanding their very high N contents up to 40 at.%. [62][63][64] Altogether, these observations  27 Al nuclear magnetic resonance (NMR) spectra recorded at 14.1 T and 24.00 kHz magic-angle-spinning (MAS) from Ca-Al-Si-O-N glasses from series I (left panel) and series II (right panel) with increasing N content from top to bottom.Each red number specifies the 27 Al shift at the peak-maximum, whereas the two dotted lines in each panel mark the highest and lowest peak-maximum shifts observed across the set of glass samples.suggest that melt/crucible reactions may cause impurities or micrometer-scale heterogeneities that are responsible for the diminished transparency of oxynitride glasses, rather than their degree of N incorporation per se.

3.2
Glass structure probed by solid state 27 Al NMR

MAS NMR results on Al coordinations
Figure 2 presents 27 Al MAS NMR spectra recorded from oxynitride glasses of series I and II.As expected from the spin-5/2 nature of 27 Al, all spectra manifest a broad NMR peak shape that reflects a convolution of a Gaussian shape from the chemical-shift distribution with a "tail" extending toward lower shifts that stem from the distribution of quadrupolar products. 23,24The absence of significant NMR signal intensities in the 25-35 ppm spectral region evidence that the amount of Al [5] species remain low throughout all glasses (Figure 2).Hence, all 27 Al resonances originate predominantly from 27 AlO 4 tetrahedra in the structure, as evidenced from the narrow span between 55 and 63 ppm of the shift at each peak maximum (δ max ) in Figure 2, which is typical for such groups. 23,24e minute but narrow 27 Al NMR peak observed at ≈15 ppm, which feature variable intensities in the 27 Al MAS NMR spectra of Figure 2, is attribted to small amounts of unreacted α-Al 2 O 3 , the resonance of which is expected around that 27 Al shift on the basis of its isotropic chemical shift of 16 ppm and small quadrupolar coupling constant C Q = 2.4 MHz. 65The low C Q value is also consistent with the absence of the 15 ppm peak in the 3QMAS NMR spectra discussed in section 3.2.2, the 3QC excitation and conversion events of which were optimized for the much larger quadrupolar interactions of the 27 Al sites in the oxynitride glass network (Section 2.2).The minor NMR signal from the minute crystalline alumina content was conservatively estimated to be <1 wt.% throughout all glasses.The α-Al 2 O 3 impurity is onwards disregarded.

3.2.2
Composition-structure relationships The 27 Al MAS NMR spectra of Figure 2 evidence that AlO 4 tetrahedra dominate the network {AlO p } speciation of the two (essentially) N-free glasses, I-1.0 and II-0.
Owing to the modest N contents throughout (Table 1) and the higher preference for Si-N relative to Al-N bond formation 5,38,66,67 discussed further in the follow-ing, the AlO 4 dominance remains for the N-bearing glass networks, encompassing the N-richest ones.For each Al-O→Al-N bond replacement, the 27 Al [4] chemical shift is increased by 7-12 ppm. 23,27,67,68Hence, the minor but consistent trend of increasing δ max values across both glass series I and II for increasing N content (and also considering the overall stronger δ max -dependence on the n Al /n Si ratio; see below) suggests a minor but growing mixedanion AlO 3 N populations coexisting with the prevalent AlO 4 groups.The present 27 Al shift-alterations of the NMR spectra from Ca-Al-Si-O-N oxynitride glasses with increasing N content (Figure 2) accord well with those observed previously for members of the La-Al-Si-O-N system, 38 which only revealed significant fractions of AlO 3 N and AlO 2 N 2 groups for markedly N-richer glasses than the present ones (>30% out of all O and N atoms).The herein observed extents of ≈3 ppm 27 Al [4] deshielding (i.e., increased chemical shift) are also similar to those reported by Jin et al. 48for a Y-Al-Si-O-N glass with 5.8 at.% of N relative to its N-free counterpart with the same n Al /n Si ≈0.86 ratio.Notably, the authors concluded the absence of Al-N bonds on the basis of different neutron diffractograms between the oxide and oxynitride glasses. 48However, we note that the signal-tonoise ratio of the neutron-diffraction difference data could, at best, only exclude a substantial Al-N bond formation (i.e., comparable to that of Si-N), which is not expected for the low N contents of the glasses herein or in Ref. [48].In contrast, an 27 Al MAS NMR analysis by Deckwerth et al. 27 of Mg-Ca-Al-Si-O-N glasses with increasing N contents up to 7.2 wt.% N concluded substantial AlO 3 N and AlO 2 N 2 contributions of ≈35% and ≈20% out of all AlO 4−n N n groups, respectively, thereby implying a markedly stronger extent of Al-N formation than those inferred herein or in Refs.[38, 69] for similar N contents.We conclude that although all hitherto presented 27 Al and 29 Si MAS NMR results of Si/Al-based oxynitride glasses suggest a much lower propensity for Al-N bond formation relative to that of Si-N, 5,38,66,67,70 more work is required to (dis)prove this inference, where conclusive data would, for instance, be offered by heteronuclear 27 Al{ 15 N} double-resonance MAS NMR experiments on 15 N-enriched M-Al-Si-O-N glasses.
Notably, the requirement of charge compensation of the prevailing tetrahedrally coordinated [AlO 4 ] − groups implies that nearly all Ca 2+ cations act as charge compensators in the present glasses.Consequently, their NBO, N [2] , and (particularly) N [1] populations must remain low throughout, in particular, in all glass members with n Al /n Ca ≈ 2 (series I; Table 1).In view of the discussion earlier about a limited number of Al-N bonds in our glasses, we may conclude that the Ca-Al-Si-O-N glass networks are predominantly built by SiO 4 , SiO 3 N, TA B L E 3 27 Al nuclear magnetic resonance (NMR) parameters obtained from 3QMAS NMR a
The Al [p] speciations in the oxynitride glasses were examined further by the 3QMAS NMR technique 51 that offers enhanced spectral resolution along the vertical dimension of the resulting 2D NMR spectrum by the suppression of anisotropic second-order quadrupolar broadenings 23,50,51 ; see Figure 3. Yet, the dispersions of both isotropic chemical and second-order quadrupolar shifts from the structural disorder remain intact; they manifest as broad "ridges" extending along both spectral dimensions, where the projection along the horizontal "MAS dimension" is nearly identical to the corresponding 1D MAS 27 Al NMR spectrum of Figure 2, as expected. 23he most striking feature of the enhanced-resolution 3QMAS 27 Al NMR spectra of Figure 3 concerns the unambiguous detection of very small Al [5] populations in all glasses examined, which were selected to comprise both low and high N contents as well as variable n Al /n Si ratios.Table 3 lists the 3QMAS-derived relative Al [4] and Al [5] contributions, confirming overall very low Al [5] populations, with the highest value of 4.2% observed for the Al-rich II-5.7 glass.Overall, the fraction of Al [5] environments correlates with the n Al /n Si ratio of Table 1: Although the variations are not strong, the Al-richer glasses of series II reveal consistently higher Al [5] contributions.However, those rather marginal responses most likely reflect significant deviations from the "Loewenstein Al [4] avoidance rule" 71 of the Ca-Al-Si-O-N glass networks.71,72 Yet, although the networks of ordered aluminosilicate phases (e.g., minerals and zeolites) strictly obey the rule and strive to minimize the local negative charge accumulation of [AlO 4 ] − -[AlO 4 ] − motifs in the aluminosilicate networks, 24 27 Al nuclear magnetic resonance (NMR) spectra recorded at 14.1 T and 24.00 kHz magic-angle-spinning (MAS) from the (A) I-1.0, (B) II-0, (C) I-6.7, and (D) II-5.7 glasses.The 2D NMR spectra are shown together with their projections along the vertical "isotropic" and "MAS" dimensions at the right and top, respectively.The vertical dotted lines in each spectrum mark the peak maxima of the signals from the 27 Al [4] and 27 Al [5] sites along the MAS dimension.No significant 2D NMR intensity from 27 Al [6] coordinations was observed in any glass.
SiO 2 glasses, that is, n Al /n Si ≈ 1 and n Al /n Ca = 2. 73,74 For molar ratios n Al /n Si > 1-such as for all series IIn members-the Loewenstein Al [4] avoidance cannot be fulfilled with solely SiO 4 and AlO 4 tetrahedra in the structure. 23,2525]72 Notably, however, Figure 3 and Table 3 evidence that all present Ca-Al-Si-O-N glasses incorporate negligible Al [6] and very low Al [5] populations (with the Al [6] amounts consistently remaining below the detection level of ≈0.5% out of all {Al [p] } sites).This common feature of both series I and II glasses suggests that the Loewenstein avoidance is far from strict, and that Al [4] -O-Al [4] linkages are present in all glass networks, encompassing those of series I for which n Al /n Si < 1 (Table 1).
Our inference of a limited relevance of the Loewenstein Al [4] avoidance rule of the present oxynitride glasses is also reflected in the increased 27 Al [4] shift for increasing n Al /n Si ratio of the glass (Figure 2), as is most evident from the concurrent increase in δ max observed for the 27 Al MAS NMR spectra shown in Figure 4 for glasses with similar N contents (5-7 at.%) but elevating n Al /n Si ratios.Notably, all hitherto discussed trends in the 27 Al NMR-peak maxima, each carrying contributions from both the isotropic chemical and second-order chemical shifts, 23,24 match well the alterations of the 3QMAS-derived isotropic chemical shifts listed in Table 3.Hence, all 27 Al shift trends discussed herein apply to both sets of {δ max } and { δ [4]  iso } values, where F I G U R E 4 27 Al magic-angle-spinning (MAS) nuclear magnetic resonance (NMR) spectra recorded from oxynitride glasses with similar N contents of 5-7 at.% but increasing n Al /n Si molar ratio.Note the trend of increasing 27 Al NMR  max for growing n Al /n Si , which is attributed to a progressively increased number of Al [4] -O-Al [4] linkages in the oxynitride glass structure at the expense of Al [4] -O-Si.
an increase in either parameter for increasing n Al /n Si ratio reflects progressive Al [4] -O-Si→Al [4] -O-Al atom substitutions in the second coordination sphere of Al [4] , where each such Al-for-Si replacement is known to increase the chemical shift by 3-6 ppm. 23,24,75Notably, these trends also underscore that Al and Si are closely integrated within the same network, meaning that these glasses cannot exhibit any significant chemical heterogeneities beyond a ∼1 nm scale.
We conclude by stressing the following inferences: (i) The correlation between the 27 Al [4] (chemical) shifts and the Al content holds for glasses of both series I and II, where the former shifts would be expected to manifest lower alterations if the Loewenstein Al avoidance would operate strictly.Nonetheless, we cannot exclude that the herein (indeed minor) 27 Al deshielding could instead stem from a slight reduction in the average Al [4] -O-T bond angle that is expected when T=Si is replaced by T=Al [4] . 23,24,75(ii) The herein observed trend of increasing 27 Al shift for elevating n Al /n Si ratio mirrors that documented for N-free CaO-Al 2 O 3 -SiO 2 glasses, 24,[76][77][78][79] as well as for other M-Al-Si-O systems-such as those based on RE cations, [80][81][82][83] for which significant deviations from the Loewenstein avoidance are well documented both by experiments and molecular dynamics simulations. 23,25,82,84

Density, molar volume, and compactness
Table 2 lists each experimentally determined Ca-Al-Si-O-N glass density along with its accompanying molar volume and compactness values.Figure 5A plots ρ against the N content of the glass, revealing (i) concomitant increases of ρ and n N within each series I and II, as well as that (ii) the II-n glasses, which are richer in the heaviest glass constituent (Ca), exhibit consistently higher densities than their I-n counterparts of similar N content.Altogether, these glass-composition trends underlie the observed density range between 2.62 and 2.84 g/cm 3 for the respective I-1.1 and II-7.2 glass members (Table 2).
Our observed monotonic density increase against n N mirrors that reported previously for several oxynitride glass systems, 7,9,11,19,21,39 including Ca-based ones. 7,19,20,39owever, ρ depends strongly on the masses of the glass constituents, whereas both the molar volume and the glass compactness properties are also affected by the amounts of the elements involved in the oxynitride glass, 3,5,13,30,38,44,56,[85][86][87][88] as moreover highlighted for members of the parent Ca-Al-Si-O system. 89Hence, the density variations of our glasses cannot be attributed to their N contents alone.Indeed, the following fit against n N along with n Al  = 0.017(0.003)N + 0.019(0.003)Al +2.378(0.052)( 2 = 0.831), (7)   captured the density variations across the entire glass ensemble.Figure 5B depicts the correlation graph with the experimental densities plotted against the values ρ[calc] calculated from Equation (7).We next consider the molar volume and compactness of the Ca-Al-Si-O-N glasses.Figure 5C,D reveals roughly linear relationships between the N contents within each glass branch I and II: V m is decreased, whereas C is increased, for increasing N content in the glass network.The compactness ranges between 0.500 (glass I-1.1) and 0.548 (glass I-7.7), corroborating earlier reports on Ca-(Al)-Si-O-N glasses with similar compositions. 3,11,30,90lthough neither V m nor C offered statistically significant correlation against the pair of N and Al contents for the entire glass ensemble, both species assume structurecondensing roles.

Refractive index
The results of Figure 6A,B reflect the dual effects of O and Si atoms being partially replaced by the more polarizable N and Al atoms, 72,91,92 respectively.Hence, the refractive index of the glass is enhanced when an Al or N atom is introduced, with the equal best-fit coefficients of Equation ( 8) suggesting equal contributions from both species.The consistently higher RI values of the Al-richer II-n oxynitride glasses relative to their I-n counterparts of similar n values (Figure 6A,B) accords with previously reported trends of increasing refractive index when Si is replaced by Al. [93][94][95] Furthermore, Figure 6C verifies an excellent linear RI/ρ correlation across the entire glass ensemble (except for the two II-0.0 and II-1.9 specimens): The strong relationship between the density and refractive index of each glass is unsurprising when considering the results of Figures 5B and 6B and their corresponding best-fit expressions in Equations ( 7) and (8).Indeed, the refractive index of an oxide-based glass is normally governed by both the atom/ion polarizabilities and the packing density of the constituent atoms. 7,11,19,31,37,46,56,64,96

Hardness
Figure 7 illustrates the bearings on the nano-indentationderived hardness from the N content of each Ca-Al-Si-O-N glass, revealing that H is enhanced from 8.3 to 10.5 GPa when the amount of N grows from 1 to 8 at.%.The hardness of our oxynitride glasses offered a good linear correlation with the N content across the entire glass ensemble:  = 0.263(0.023)N + 8.13(0.12)( 2 = 0.905).
Such a monotonic trend accords well with previous findings. 14,15,19,31,32,37,42,97,98In contrast to most other physical properties considered herein, however, attempts of fitting the (nano)hardness to both n N and n Al gave no further improvements and resulted in a statistically insignificant Al coefficient of −0.001 ± 0.029.This finding must reflect that the hardness is foremost governed by the cross-linking Nano-indentation-derived hardness plotted against the amount of N in the glass.The green line represents the best fit result for all glasses; see Equation (10).
effects from N alone, which swamp any minor bearings from Al on the hardness.56,97 Several early studies have attributed the latter feature as dominantly dictating the mechanical and elastic properties of M z+ -based oxynitride glasses with (near) constant stoichiometric n Si /n Al and n O /n N ratios.Indeed, the hardness of oxide/oxynitride M-Al-Si-O-(N) glasses indeed increase concurrently with the M z+ cation field strength (CFS). 3,4,22,30,99ore recently, however, another explanation for the enhanced hardness of RE-based M-Al-Si-O glasses was proposed, 23,83,100 which is argued to most likely also hold for their oxynitride counterparts. 100Despite originating from the M z+ CFS, it is not the (potentially enhanced) network cross-linking from the M-O-Si/Al bond fragments themselves that boost the mechanical glass properties, but merely the presence of high-coordination Al [5] /Al [6] populations of the aluminosilicate network induced by the high-CFS network-modifier cations.Likewise, the comparatively stronger Al [5] -O/N and Al [6] -O/N bonds (relative to M-O/N) may well dictate the hardness of M-Al-Si-O-(N) glasses, as well as other glass-network-governed physical properties, where Becher et al. 3 have attributed variations of Young's modulus to partially stem from varying Al [5] and Al [6] populations.Nonetheless, such a structure-densifying effect must be insignificant for the present Ca-bearing oxynitride glasses, for which Al [4] accounts for essentially all Al sites (Table 3 and Section 3.2).Consequently, we can more firmly attribute the hardness enhancements of the present Ca-Al-Si-O-N glasses to the replacement of O [2] sites by N [3] .The hitherto unsettled role of Al for affecting the oxynitride glass properties is discussed further in Section 3.7.

Elastic properties
56,97 The elastic properties of both oxide and oxynitride glasses are also governed by several structural parameters across both shortand medium-range length scales, encompassing the cation-anion bond strengths/energies, the atom/bond densities (both encoded in C), and the glass-network connectivity, 11,21,22,44,56,96,101,102 which in turn dictates its network 2D/3D topology.However, although all elastic moduli are expected to increase concomitantly with C, neither of E nor G revealed any readily discernable general correlation with the glass compactness (Table 2): Merely, the E and G values of series I are essentially independent on C, whereas the glasses of series II manifest a weak but significantly scatted increase.Only the bulk modulus hinted a reasonably linear increase for the entire glass ensemble (R 2 = 0.589).Moreover, the I-1.1 specimen features significantly lower results of each modulus than any other glass.We have no explanation for these trends.
Although none of the elastic properties considered herein offered reasonable correlations with n N alone across the entire glass ensemble, they yielded decent fits within each glass series I and II (Figure 8), reflecting that the relative Ca, Al, and Si contents also influence the elastic moduli significantly.We first consider the reduced modulus, which ranges from 99 GPa for the I-1.1 glass to 121 GPa for I-7.7, moreover revealing a linear E r /n N relationship within each glass branch I and II (Figure 8A).The overall higher E r value of the Al-richer II-n glasses relative to their I-n counterparts of similar N contents evidences a significant contribution from Al, as given by the following best-fit relationship (Figure 8B): Figure 8C plots the ultrasonic echography-derived Young's modulus results of the Ca-Al-Si-O-N glasses against their N contents.E ranges from 94 to 122 GPa for I-1.1 and II-7.2, respectively.As for the reduced modulus, Young's modulus values of the entire glass ensemble gave a reasonable fit against both N and Al contents (Figure 8D):  = 1.955(0.41)N + 1.764(0.478)Al +75.340(8.36) Hence, both E r and E increase concomitantly with n N and n Al .A reasonable linear E/E r correlation was observed (R 2 = 0.663), altogether suggesting that our Ca-Al-Si-O-N glass specimens are homogeneous, at least over a nm scale.As revealed by Figure 8E,G, also the shear and bulk moduli increase linearly with the N content within each glass branch I and II, as well as confirming additional enhancements of the moduli for increasing amount of Al: For the bulk modulus, the proportionally larger best-fit coefficient for Al relative that of N suggests a higher impact from Al than for the shear modulus.Poisson's ratio relates to the rigidity of the glass network and its dimensionality, where a higher (lower) ν value normally correlates with a lower (higher) network connectivity/rigidity. 56,72Yet, oxynitride glasses generally exhibit higher Poisson's ratios than their N-free aluminosilicate-based counterparts. 30,44,99,103The present Ca-Al-Si-O-N glasses reveal a narrow span of Poisson's ratios between 0.260 and to 0.278 for the I-3.2 and II-7.2 glass specimens, respectively (Table 2).The minute variations of the Poisson's ratio did not suggest any clearly identifiable trend against n N or any other glass constituent.

Bearings from Al on physical properties
Here, we discuss the impact from Al on the physical properties of the present Ca-Al-Si-O-N glasses in relation to previous literature findings from related M-Al-Si-O-N systems.We also identify further research directions toward an improved insight into the potentially (very)  11)- (14)   significant role of Al for governing the mechanical (and thermal) properties of oxynitride glasses.
A key finding herein is the dependence of the E r /E/K/G elastic glass properties on both the N and Al contents of the Ca-Al-Si-O-N glass, in contrast to the hardness and compactness, for which only the N content had significant bearings.However, given that these properties normally correlate and manifest very similar composition-property relationships, [3][4][5] the apparent invariance of H of the glass on its n Al most likely reflects a lower dependence of the hardness on the Al content relative to those of its E r /E/G/K moduli.This inference is supported by the results of Ref. [15].Also notable is that essentially the same quality of correlation resulted if instead fitting the properties discussed earlier to n N together with the molar ratio n Al /n Si .
Introducing alumina to an M-Si-O-N melt will reduce its NBO content because the M z+ cations are consumed for charge-balancing the [AlO 4 ] − groups (Section 3.2).Consequently, the NBO content remains very low throughout each of the present Ca-Al-Si-O-N glasses (Table 1).Hence, complications from network-weakening effects from NBO species on the physical properties are largely absent.Moreover, the reduced molar volume but enhanced elastic properties for increasing n Al suggest a structurecondensing effect of Al, which must derive from the dominating AlO 4 groups (along with minor AlO 3 N moieties) of the Ca-based oxynitride glasses, whereas any bearings from higher coordination Al [5] /Al [6] species may safely be excluded (see Section 3.2).Yet, given that Al-O bonds are weaker than their Si-O and Si-N counterparts, 3,17 the precise network-strengthening mechanism from the AlO 4 groups remains unclear.
There are surprisingly few reports on the bearings from Al on hardness and elastic properties from oxynitride glasses with constant N content, all of which concerned RE-based M-Al-Si-O-N glasses, 14,15,22,97 and particularly members from the Y-Al-Si-O-N system. 14,15,97Given the difficulties of "isolating" the effects on physical properties from Al alone, the glass design ensured a constant amount of Al along with that of one cation of the set {RE 3+ , Al 3+ , Si 4+ }, whereas the contents of the other two species were varied such that one increased, whereas the other decreased. 14,15,22,97Yet, as commented further further below, the complex relationship between an RE-Al-Si-O-(N) glass composition in relation to its {Al [4] , Al [5] , Al [6] } and {O [p] , N [p] } speciations may lead to alternative rationalizations of the observed compositionproperty trends, particularly in the absence of (precise) experimental data on both Al and O speciations.Here, the use of glasses in (slightly) different Y-Al-Si-O-N composition-regimes may in part explain the somewhat contradictory results between earlier studies. 14,15,97Indeed, Hampshire et al. 97 reported a non-monotonic dependence of the hardness and elastic moduli for increasing n Al /n Y ratio (and increasing n Al /n Si ): Following an initial decrease in the property values, with a minimum found at ≈7 at.%Al, they all increased concurrently with n Al /n Y in the Al-richer regime.That finding accords well with our observation from the present Ca-Al-Si-O-N glasses, whose Al contents all exceed 7 at.%(Figure 8).
In contrast, the study of Ref. [14] did not find any significant influence from the Al content (i.e., on n Al /n Y ) on any mechanical property.Also, Lemercier et al. 15 observed very modest changes in the Vickers hardness (H V ) along with the shear and Young's moduli for increasing Al content-as probed for glasses with either constant Si content and decreasing Y amount, or at constant Y content and decreasing Si amount.Given that the very small variations of the hardness values were within their stated ±1σ uncertainties, we conclude that our observed invariant hardnesses accord qualitatively with the data of Ref. [15], as well as with their statistically unambiguous minor increase in Young's modulus for Y-Al-Si-O-N glasses (see Section 3.6 and Figure 8).Their finding also accords with our observed larger alterations in the elastic moduli as compared with the hardness (the effects of which become swamped by those from N; see Section 3.5).We also remark that the authors of Ref. [15] as well as subsequent review articles 3,16 have made somewhat different inferences about the data trends of Ref. [15] than we do.
We conclude that more investigations are required to better understand the effects of Al on the physical properties of oxynitride glasses, as well as for identifying the structural factors that underlie these trends.One weakness of our glass ensemble is that all elements vary simultaneously, and the changes in the cation contents among the glasses are small (Table 1).Besides noting that the very sparse reports on the (potential) impact from Al on mechanical properties were all published ∼25 years ago, we believe that future studies will benefit by targeting oxynitride glasses essentially free from NBO species (as those herein).For instance, the glass series design by Lemercier et al. 15 implied that the NBO amount of the glass decreased for increasing Al content.Hence, it is difficult to ascertain whether the physicalproperty changes stemmed from Al per se or merely reflected alterations in the BO/NBO speciation of the glass structure.
Moreover, the high-CFS RE cations produce significant AlO 5 (and to a lesser extent, AlO 6 ) populations in M-Al-Si-O-(N) glasses over large composition spaces, encompassing the domain of formally chargecompensated glasses with n Al /n M ≤ z. 23,25,80 The overall higher average Al coordination numbers partially account for the enhanced mechanical properties of RE-Al-Si-O-(N) glasses relative to those of alkali-and alkaline-earthbased counterparts, such as the present Ca-Al-Si-O-N glasses.These M + /M 2+ cation-based families adhere closer to the conventional glass structure model for aluminosilicate glasses, 23 thereby facilitating the identification of composition-structure-property relationships because glasses obeying n Al /n M ≈ z are essentially devoid of NBO species, whereas nearly all Al sites are four-coordinated by O/N (unless n Al /n Si ≫ 1); see Table 1 and Section 3.2.
Yet, although M-Al-Si-O-(N) glasses with M 2+ cations of moderate field strengths (i.e., all alkaline-earth metal ions but Mg 2+ are advantageous for better understanding the potential bearings from Al on the mechanical (and thermal) properties, the ultimate goal in this field is the design of oxynitride glasses with optimal properties.For the reasons given earlier, these are expected to be found for NBO-free glasses with the highest possible Al [5] /Al [6] contents, that is, for comparatively Al-rich RE-Al-Si-O-N glasses.

CONCLUSIONS
We have shown that transparent Ca-Al-Si-O-N glasses with up to ≈8 at.% of N (17% out of the anions) are readily prepared by employing quartz crucibles and AlN as a the sole N source.Along with previous findings for similar M-Al-Si-O-N glasses, [3][4][5] the physical properties of the present Ca-based oxynitride glasses enhance monotonically with the N content, encompassing the density, compactness, refractive index, nanohardness along with the reduced modulus and the shear, bulk, and Young's moduli.Yet, although only the hardness offered a good linear correlation against the N content alone throughout the glass ensemble, decent/good linear increases with n N resulted for each physical property within each glass series I and II, except for the molar volume that merely contracted, as expected.These observations reflect the impact on the physical properties from the precise Ca, Al, and Si contents, where each comparatively Al-richer glass of series II (n Al /n Si > 1) exhibited a higher value of the density, refractive index, as well as all elastic moduli than its counterpart from series I with a similar N content but lower n Al /n Si < 1 ratio.Indeed, except for the compactness and hardness, all properties examined herein gave decent or good fits against their pair of N and Al contents.We stress that very similar correlation coefficients (R 2 ) resulted if instead correlating the properties with n N and n Al /n Si .
The Ca-Al-Si-O-(N) glass constitution was probed from the viewpoint of its Al sites by 27 Al MAS NMR experiments, revealing that AlO 4 tetrahedra accounted for >95% of all AlO p moieties in all glass networks, and AlO 5 polyhedra for the remaining.In contrast, AlO 6 octahedra were absent throughout, where the very minor 27 Al [6] resonances detected in the NMR spectra were readily traced to minute remnants of unreacted α-Al 2 O 3 .The 27 Al MAS NMR results also gave circumstantial evidence for a partial violation of the Loewenstein Al avoidance rule throughout all Ca-Al-Si-O-N glass networks, meaning that they comprise non-negligible populations of Al [4] -O-Al [4] linkages, while also corroborating earlier inferences that Si-N bonds are preferred relative to Al-N in oxynitride glasses. 5,38,48,66,67The violation of the Loewenstein rule presumably also explains the near absence of highercoordination Al [5] and Al [6] species.
On the basis of our 27 Al (3Q)MAS NMR data together with inferences from previous 27 Al/ 29 Si NMR studies of related M-Al-Si-O-N glasses, 13,38,48,66,70,87 the present Ca-Al-Si-O-N glasses form 3D aluminosilicate oxynitride glass networks that are built primarily by interlinked SiO 4 , AlO 4 , and SiO 3 N tetrahedra, with (very) minor contributions from AlO 3/2 N 1/3 , and AlO 5 moieties.Moreover, essentially all O and N atoms coordinate 2 and 3 network-forming atoms, respectively, because nearly all Ca 2+ cations are consumed for balancing the negatively charged [AlO 4 ] − groups (notably so the glasses of series I; see Table 1), thereby precluding any significant formation of either NBO or N [1] /N [2] anions.
Notably, the near-sole presence of four-coordinated Al species in the Ca-Al-Si-O-N glasses allowed us to more firmly attribute their enhanced physical properties to the concomitantly increased glass-network connectivity resulting from the gradual N [3] -for-O [2] substitutions for increasing N content, whereas similar networkstrengthening cross-linking mechanisms by AlO 5 and AlO 6 polyhedra 23,83,100 may safely be discarded.This contrasts with the scenario of RE or Mg-based aluminosilicate oxynitride glasses, for which physical-property enhancements may partially stem from an increased network cross-linking by the presence of high-coordination Al [5] and Al [6] species. 3,4,15Hence, the herein observed concurrently enhanced glass properties with the Al and N contents of the glasses must originate from the AlO 4 tetrahedra, despite that Al-O bonds are expected to be weaker than those of Si-O/N. 3,17Our somewhat puzzling findings demand further research to shed light on the precise structural role of Al for governing mechanical and thermal properties of M-Al-Si-O-N glasses.We suggest that although such a road map may benefit from employing alkaline-earth-based systems, optimized mechanical properties are most likely offered by RE-Al-Si-O-N glasses rich in N and Al but devoid of NBO species.

A C K N O W L E D G M E N T S
We gratefully acknowledge Aleksander Jaworski and Renny Mathew for arranging the NMR data.We thank Mirva Eriksson for density measurements of a few samples, and John Mauro for discussions.

a
Physical properties and their experimental uncertainties (specified in the bottom row): density (ρ); molar volume (V m ); compactness (C; atom packing density); nano-indentation-derived hardness (H), reduced elastic modulus (E r ), Young's modulus (E); shear modulus (G); bulk modulus (K), Poisson's ration (ν), and refractive index (RI).F I G U R E 1 Photos of Ca-Al-Si-O-N glasses, with the transparency decreasing from left to right in both series I and II.Note that each glass label is placed behind the glass piece.

Figure
Figure 6A depicts the dependence of the refractive index on the N content of the Ca-Al-Si-O-N glass.Within

F I G U R E 5
Experimental values of the (A and B) density, (C) molar volume, and (D) compactness plotted against the N content (in at.%) of the Ca-Al-Si-O-N glasses, except for (B), which depicts the experimental density versus that calculated (ρ[calc]) from the best-fit expression to n N and n Al (Equation7).Black and red symbols represent data from glass series I and II, respectively; see the legend in (A).The lines in (A, C, D) represent best fit results.Here and in other figures: Open symbols were not included in the data fitting.F I G U R E 6 (A)Refractive index plotted against the N content of the oxynitride glass for each branch I and II; (B) correlation of the experimental refractive index against that calculated from Equation (8), RI[calc], which accounted for both the N and Al contents of all glasses; (C) refractive index plotted against the density, yielding a good linear correlation across the entire glass ensemble; see Equation(9) some data scatter, RI increases linearly against the N content within each of series I and II.Yet, as evidenced by Figure6B, a markedly better fit-unified for both series I/II glass branches-was obtained when both the N and Al contents were accounted for RI = 0.007(0.001)N + 0.007(0.001)Al +1.477(0.013)( 2 = 0.926).
Figure 8F,H plots the experimental G and K values against those calculated from the two best-fit expressions given above, G[calc] and K[calc], each obtained across the entire Ca-Al-Si-O-N glass ensemble.

F I G U R E 8
Left panel: Experimental values of the (A) reduced elastic modulus, along with (C) Young's, (E) bulk, and (G) shear moduli, plotted against the N content of the Ca-Al-Si-O-N glasses, where black and red symbols represent data from series I and II, respectively, as identified by the legend in (A).The lines represent best fit results with the as-indicated correlation coefficients.Right panel: Experimental values of (B) E r , (D) E, (F) K, and (H) G, plotted against those calculated by the corresponding best fit expressions of Equations (

n Al /n Si n Al /n Ca n Ca /(n Al + n Si ) n N /(n N + n O ) δ max (ppm) δ CG (ppm)
27alyzed oxynitride glass stoichiometries, molar ratios,27Al magic-angle-spinning (MAS) nuclear magnetic resonance (NMR) peak maxima (δ max ) and center-of-gravity shifts (δ CG ) a TA B L E 1a Each glass composition is normalized such that n Ca + n Al + n Si + n O + n N = 100 mol, that is, the value of n E becomes equal to the at.% of element E in the glass.