Introduction

Oxide glass components are basically classified into network formers, network modifiers and intermediates by Zachariasen1 and Sun2. Typical network formers satisfy Zachariasen’s rules1, but V2O5 is classified as a network former1 or intermediate3. This is because it is hard for this oxide to form glass on its own because the oxygen coordination number is five in a crystalline phase, which is larger than the coordination number of typical network formers of three and four. V2O5 based glasses have been widely studied for decades because V2O5 glass is a typical semiconducting glass originating from hopping conduction. Indeed, not only fundamental research on electrical properties4,5,6,7,8 but also applied research on cathode materials for lithium, sodium and magnesium ion batteries have been reported9,10,11,12. This glass also has a low glass transition temperature and relatively low thermal expansion4. These are quite attractive properties for low-melting glass used for sealing. In this field, lead borate glass with a high percentage of PbO has been applied to sealing below 400 °C in electronic devices, such as IC ceramic packages, crystal oscillators and micro-electro-mechanical systems. Since lead components are a hazardous substance for human health and the environment, it is necessary to avoid the use of PbO, and hence, V2O5 based glass is a promising material for overcoming this problem13,14,15,16,17,18,19. However, there are two critical problems that face the practical application of V2O5 based glass: poor water durability and low thermal stability.

Several components, i.e., P2O5, TeO2, GeO2, BaO and PbO, can vitrify V2O5 in binary systems4. Above all, the most common vitrification component is P2O5 because the V2O5-P2O5 system has the widest range of vitrification in binary systems. Moreover, the properties of this binary system change drastically with the amount of P2O5 compared with other components20,21. For example, the glass transition temperature increases about 200 °C, the electrical resistivity increases to the double-digit range, and water resistance improves from water soluble to nearly insoluble in the composition range from 80V2O5-20P2O5 to 50V2O5-50P2O520,21. Conversely, we can greatly design the properties of V2O5 based glass by controlling the amount of P2O5 content. To design compatible properties for various applications, understanding the glass structure is crucial. Structures of V2O5 glass and (100-x)V2O5-xP2O5 (VPx) glasses have been widely studied by using spectrometry techniques22,23 and diffraction techniques24,25,26. The results for VPx glasses obtained by using a Fourier transform infrared spectrometer (FT-IR) indicate a structural change from V2O5- to β-VOPO4-like structures around x = ~2522. The valence states and structure model of x = 40–70 glass obtained from X-ray photoelectron spectroscopy (XPS) data suggested that this model consisted of a mixture of vanadate phosphate phases, including V2O5, VOPO4, (VO)2P2O7, VO(PO3) and V(PO3)323. X-ray and neutron diffraction measurements conclude that V-O structural units undergo complex changes as the amount of V2O5 content decreases25. According to an analysis, short-range structural units change from a VO4+1 trigonal pyramid for x = 0, VO4+1 trigonal bipyramid and VO5+1 square pyramid for x = 27 and VO5+1 square pyramid for x = 27–50 to a VO5+1 square pyramid of β-VOPO4 and VO6 distorted octahedron of (VO)2P2O7 for x = 50. In addition, the average V-O coordination number (NV-O) for VPx glasses changes from 4.5 (x = 10) to 5.1 (x = 50) as the amount of V2O5 content decreases25.

Nevertheless, the structure-property relationship in VPx glass is still unclear; in particular, the water durability, thermal stability and hardness of this glass are important properties for practical application. One reason is that the coordination environment and valence of vanadium atoms are still unknown. The other is that the nature of network formation in the glass still has not been well revealed. In this article, we report on the properties and atomic structures of VPx glasses characterized by reverse Monte Carlo (RMC) modelling based on neutron/synchrotron X-ray diffraction data and extended X-ray absorption fine structure (EXAFS) data measured at the V-K absorption edge. In addition, we used Raman and 51V magic angle spinning (MAS) nuclear magnetic resonance (NMR) spectroscopy to determine the connectivity of PO4 tetrahedra and the coordination number of pentavalent VOx polyhedra, respectively. On the basis of state-of-the art experimental data aided by a data-driven modelling technique, we discuss the relationship between several properties and glass structures with a special focus on the coordination environment and valence of vanadium atoms.

Results

Glass properties

Glass compositions, the ratio of V4+/Vtotal, densities and melting temperatures of VPx glass samples are summarized in Table 1. The V4+/Vtotal ratio of the prepared samples increased as the amount of P2O5 increased. This trend is supported by the shift in the energy of the absorption edge of the X-ray absorption near-edge spectroscopy (XANES) spectra at the V-K edge (Fig. S1) and is consistent with the results of a previous study20. However, the values of the previous study differed from those of this study22 because the results were affected greatly by the melting conditions.

Table 1 Composition and properties of VPx glasses.
Full size table

The glass transition temperature Tg and crystallization temperature Tc determined from differential thermal analysis (DTA) curves of these samples are shown in Fig. 1a,b and are summarized in Table 1. It is demonstrated from Fig. 1b that these characteristic temperatures and thermal stability ∆T = Tc − Tg increased as the P2O5 amount increased, that P2O5 is an element for improving thermal stability in the V2O5 glass system. In particular, no crystallization peak was observed in the VP44 glass, and the glass softened and flowed to the melting temperature without crystallization. The normalized weight loss in water and Vickers microhardness of these samples, shown in Fig. 1c, suggest that the water durability improved drastically, whilst the glass hardened as the P2O5 amount increased. The atomic number densities and apparent molar volume of O ions for these samples are shown in Fig. 1d. It was confirmed that the atomic number density increased, whereas the apparent molar volume of O ions decreased as the amount of P2O5 increased. Both indicate that the packing density of atoms increased with the addition of P2O5. Indeed, the atomic number density and apparent molar volume of O ions for crystalline V2O5 were 0.0779 Å−3 and 10.83 cm3 mol−1, respectively, suggesting that the packing densities of these glass samples were smaller than that of crystalline V2O5.

Figure 1
figure 1

Characterization of VPx glasses. (a) Differential thermal analysis curves. (b) Glass transition temperatures and crystallization temperatures. (c) Normalized weight loss in water and Vickers microhardness. (d) Atomic number densities and apparent molar volume of oxygen ion. In VP44 glass, no crystallization peak was observed until glass melted.

Full size image

Diffraction and EXAFS data

Data on X-ray structure factors, SX(Q), neutron total structure factors, SN(Q) and EXAFS k3χ(k) measured at the V-K edge for a series of VPx glasses are summarized in Fig. 2a–c, respectively. It is confirmed that all samples are homogeneous, because we cannot observe any small angle scattering at Q < 1 Å−1 in Fig. 2a,b. Three-peak structure was observed for the neutron SN(Q) for VP100 [glassy (g)-P2O5]26 glass, for which we can assign the peak at Q = 1.25 Å−1 and that at Q = 2.1 Å−1 to a split first sharp diffraction peak (FSDP)27, and the peak at Q = 2.95 Å−1 to the principal peak (PP)27. A split FSDP was not observed for glassy (g)-SiO2 nor glassy (g)-GeO228, but it was observed for the VP100 glass due to the formation of a Q3 network, in which a phosphorous atom has three bridging oxygen atoms and one bridging oxygen atom. In contrast, we observed only a two-peak structure (FSDP at Q = 1.4 Å−1 and PP at Q = 2.1 Å−1) in the X-ray SX(Q) for the VP100 glass29 because a PP was not observed in the X-ray data due to small weighting factors of oxygen-related correlations for X-rays. Intriguingly, this complicated peak structure disappeared in the other VPx glasses, and it showed a sharp PP at Q  ~2.7 Å−1 in the neutron SN(Q) and a relatively sharp FSDP at Q  ~1.8 Å−1 in the X-ray SX(Q). The EXAFS data measured at the V-K edge differed substantially, suggesting that the local environment of V atoms changed with the glass composition.

Figure 2
figure 2

X-ray/neutron diffraction and EXAFS data in reciprocal space for VPx glasses. (a) X-ray total structure factors, SX(Q). (b) Neutron total structure factors, SN(Q). (c) EXAFS k3χ(k) measured at the V-K edge.

Full size image

The X-ray total correlation functions, TX(r) and neutron total correlation functions, TN(r), for a series of VPx glasses are shown in Fig. 3a,b, respectively, together with the data for crystalline (c-)V2O5. We can see an excellent contrast between the X-ray and neutron diffraction data because X-rays are sensitive to vanadium, whilst oxygen can be easily detected with neutrons, and it is very difficult to detect vanadium with neutrons. In addition, it is possible to detect both P-O and V-O correlations with X-rays, as can be seen in the Fig. 3a, where a tiny negative V-O correlation peak can be observed at r = 1.7 Å in the neutron T(r). Indeed, only oxygen atoms can be detected with neutrons, and this allows us to observe both corner-sharing and edge-sharing O-O correlations in c-V2O5. The most striking feature in the X-ray T(r) is the big difference in V-O correlation peaks between the VP0 glass [glassy (g)-V2O5] and c-V2O5. In the case of c-V2O5, both V-O and V = O bonds were clearly distinguished, whilst such a doublet feature was not observed in the VP0 glass. Moreover, it was found that the bond length of V-O increased, whereas V-V correlations and O-O correlations in the corner sharing VOx polyhedra, and the P-O distance decreased as the P2O5 fraction increased. These results are consistent with the results reported by Hoppe et al. for the VP27 and VP50 glasses25, indicating that the structural units of VOx and PO4 polyhedra and the connectivity of VOx-VOx and VOx-PO4 in each glass change with the glass composition. Figure 3c shows Fourier-transformed (FT) EXAFS spectra for the VPx glasses, in which significant composition-dependent modification of the local environment around the V atoms was observed in real space, too.

Figure 3
figure 3

X-ray/neutron diffraction and EXAFS data in real space for VPx glasses together with c-V2O5 data. (a) X-ray total correlation functions, TX(r). (b) Neutron total correlation functions, TN(r). (c) Fourier-transformed (FT) EXAFS spectra for VPx glasses. Black, c-V2O5; Purple, VP0 glass30; Red, VP10 glass; Green, VP28 glass; Blue, VP44 glass; Cyan, VP100 glass26,29.

Full size image

Structural model of the glasses

We performed RMC modelling on the basis of experimental data to uncover the relationship between glass structures and properties. As can be seen in Fig. S2, RMC-modelled X-ray SX(Q), neutron SN(Q) and V-K edge EXAFS k3χ(k) data for the VP0 (g-V2O5)30, VP10, VP28, VP44, and VP100 (g-P2O5)31 glasses agreed well with experimental data. Figure 4 compares the partial structure factors, Sij(Q), of a series of VPx glasses calculated from the RMC models. First of all, we address partial structures in the VP0 (g-V2O5) and VP100 (g-P2O5) glasses. It is well known that the FSDP of network formers, e.g., g-SiO2 and g-GeO2, shows up a positive peak in each Sij(Q)28. However, the FSDP of the SPP(Q) and SPO(Q) was a doublet31, and both the X-ray and neutron S(Q) did not exhibit a well-defined FSDP for VP100 glass. The reason for this is due to the different length scale that arose from both the P-O bonds (1.58 Å) and P = O bonds (1.43 Å) in the glass (see Fig. 3b), as discussed by Hoppe26. In comparison, a sharp positive FSDP observed at Q  ~1.8 Å−1 for three Sij(Q) in the VP0 glass, which is related to a V-O correlation peak that is relatively symmetrical in comparison with the VP0 glass. Other interesting behaviour is the very sharp negative and positive peaks observed at Q = 1.38 Å−1 and Q = 2.60 Å−1 in the SP-V(Q) of the VP44 glass, suggesting that the network was formed by the interplay between PO4 tetrahedra and VOx polyhedra.

Figure 4
figure 4

RMC-generated partial structure factors, Sij(Q), for VPx glasses. Purple, VP0 glass; Red, VP10 glass; Green, VP28 glass; Blue, VP44 glass; Cyan, VP100 glass31.

Full size image

Partial pair-distribution functions, gij(r), of a series of VPx glasses calculated from the RMC models are shown in Fig. 5. The first P-O correlation peaks observed at r = 1.6 Å for the VP10, VP28 and VP44 glasses were more symmetrical in comparison with the VP100 glass, suggesting that electrons were more delocalized in the VP10, VP28 and VP44 glasses. A significant composition-dependent change was observed for gO-O(r), gV-O(r) and gV-V(r), and it was confirmed that this reflected the changes in the experimental T(r). The first P-V correlation peak observed at 3.2 Å became sharp as the P2O5 fraction increased, which was in line with the behaviour in Q space, suggesting the interplay between PO4 tetrahedra and VOx polyhedra in forming the network.

Figure 5
figure 5

Short-range structural analysis on RMC-generated models for VPx glasses. Partial pair-distribution functions, gij(r). Purple, VP0 glass; Red, VP10 glass; Green, VP28 glass; Blue, VP44 glass; Cyan, VP100 glass31.

Full size image

The average coordination numbers calculated up to 2.5 Å and the ratio of edge-sharing VOx polyhedra calculated from the RMC models are summarized in Table 2. The oxygen-cation coordination number NO-M (M = V, P), the V-O coordination number NV-O and the ratio of edge-sharing VOx polyhedra increased as the amount of P2O5 increased. It is worth mentioning that 15% of VOx polyhedra were edge-shared in the VP44 glass, which is outside of Zachariasen’s rule1. These behaviours are consistent with the fact that the packing density of atoms in glass increased as the P2O5 fraction increased. The atomic configurations together with the cavity volume of a series of VPx glasses are shown in Fig. 6. It is noted that the cavity volume increased as the amount of P2O5 increased, whilst packing density increased.

Table 2 Coordination numbers of VPx glasses obtained by RMC models.
Full size table
Figure 6
figure 6

Cavity analysis. (a–e) RMC-generated atomic configurations with cavities for (a) VP0, (b) VP10, (c) VP28, (d) VP44 and (e) VP100 glasses. (f) Cavity volume as a function of P2O5 content.

Full size image

The Qn distribution of V and P species calculated from the RMC models are shown in Fig. 7a,b. Although the Qn distribution of V species shifted to a smaller number of n as the amount of P2O5 increased, the V species had a high proportion at n ≥ 2 (Fig. 7a). In comparison, the P species were almost isolated PO4 (Q0) tetrahedra not only in the VP10 glass but also in the VP44 glass, as shown in Fig. 7b. These behaviours are supported by the Raman spectra as shown in Fig. S3, in which a peak assigned to Q0 was observed for the VP10 and VP44 glasses. The ring statistics of VPx glasses are shown in Fig. 7c. It was demonstrated that the normalized number of V-O rings decreased as the amount of P2O5 increased, whereas P-O rings did not form in the range of glasses VP10 to VP44. In contrast, the normalized number of M-O rings increased as the amount of P2O5 increased. These results strongly indicate that the network former of the VPx glasses changed from VOx-VOx networks to the interplay between VOx and PO4 for the VP28 and VP44 glasses. These changes in the network structure and edge-shared VOx polyhedra were clearly visible in the three-dimensional atomic configurations obtained from the RMC models, as shown in Fig. 7d,e.

Figure 7
figure 7

Connectivity and ring statistics of VOx and PO4 polyhedra. (a,b) Qn distribution of V and P species obtained by RMC modelling. (c) Normalized numbers of -V-O-rings, -P-O-rings and -M-O-rings (M = P, V) in VPx glasses. Purple, VP0 glass; Red, VP10 glass; Green, VP28 glass; Blue, VP44 glass; Cyan, VP100 glass. (d,e) Atomic configurations of (d) VP10 and (e) VP44 glasses are obtained from the RMC models. Green and red-coloured bond represents V-O bond and P-O bond, respectively.

Full size image

The fraction of VOx polyhedra derived from the RMC models is shown in Fig. 8a. The figure clearly shows that the majority of short-range structural units in the VPx glasses consisted of VO4 and VO5 units. It was also found that the fraction of VO4 units decreased, whilst the fraction of VO5 units increased systematically as the amount of P2O5 increased. Taking the vanadium valence into account, VOx polyhedra were divided into six units, V4+O4-6 and V5+O4–6. To obtain information on the local structure around the V5+ in the VPx glasses, NMR spectroscopic measurements were performed. 51V NMR spectroscopy can selectively detect the diamagnetic V5+ state because the V4+ state cannot be analysed directly due to its paramagnetism. The 51V MAS-NMR spectra of the VPx glasses are shown in Fig. S4. The peaks observed at around −750, −540, −490 and −300 ppm can be assigned to symmetric-V5+O4 (s-VO4), distorted-V5+O4 (d-VO4), V5+O6 (VO6) and V5+O5 units (VO5), respectively32,33,34. The fractions of these units, as calculated from the area of each peak, are summarized in Table S1. It is suggested for the V5+ state in the VPx glasses that the fraction of V5+O4 units decreased as the amount of P2O5 increased, whereas the fraction of V5+O5 units increased. The fractions of V5+O4-6 and V4+O4-6 units calculated from the RMC models, 51V NMR spectroscopy and the fraction of V4+/Vtotal are shown in Fig. 8b. The result clearly indicates that the unusually small coordination number of V4+O4 units existed in the VPx glasses and the fraction increased as the amount of P2O5 increased, although it has been suggested for the V4+ state in glass that only V4+O5 or V4+O6 units are taken the same as crystals due to the larger ionic radius of V4+ 35,36,37,38. Such unusually small coordination numbers in oxide glass are also reported for MgO-SiO239, CaO-Al2O340 and ZnO-P2O541 glass systems, and, hence, it is suggested that this may be a characteristic feature in some oxide glasses with atypical network formers.

Figure 8
figure 8

Distribution of oxygen coordination number and valence of vanadium. (a) Total fraction of VO4, VO5 and VO6 units obtained by RMC modelling. (b) Fraction of V5+O4-6 and V4+O4-6 units calculated by RMC modelling and 51V NMR spectroscopy.

Full size image

Discussion

We set out to model the glass structure to understand the water durability, thermal stability and hardness, which are important in practical applications. The water durability of VPx glasses improves as the amount of P2O5 increases, even though VP0 (g-V2O5) and VP100 (g-P2O5) glasses readily dissolve into water31,42,43. Gin et al. reported on the water durability of glass divided into several stages44. This improvement of VPx glasses in terms of water durability was determined to be due to the suppression of the hydrolysis process, which is affected by the structure of VOx and PO4 polyhedra and the valence of vanadium. The cavity volume in the glass did not seem to significantly affect the water durability because, when the cavity volume increased, the water durability improved. Regarding the structure of VOx polyhedra and valence of vanadium, Nabavi et al. suggested that the V5+O4 units in amorphous V2O5 are highly reactive towards water, whereas the V5+O5 units in crystalline V2O5 are not. Moreover, they also suggested that the water durability increased when the number of V5+O4 units decreased, which implies that water durability is affected by V-O coordination35. Our results also show that there is a clear correlation between normalized weight loss in water and the fraction of V5+O4 units in all cations (see Fig. S5). In other words, we suggest that V4+O4 units are stable against water, whilst V5+O4 units are highly reactive towards water. We conclude that this might be affected by the polarizability of VOx polyhedra, as in the case of PO4 polyhedra. Therefore, the increase in edge-sharing VOx polyhedra might improve the water durability. Regarding the structure of PO4 polyhedra, several papers reported that phosphate glass that has PO4 tetrahedra with Q0 or Q1 structures shows good water durability, suggesting that the low polarizability in PO4 tetrahedral units makes it hard to attract polar water molecules45,46,47. Therefore, the reason that the water durability improved in the range of glasses VP0 to VP44 is because the structures of the PO4 tetrahedral units were only Q0 and Q1 structures, unlike the Q3 structures in the VP100 glass. Hence, we conclude that the mechanism for improving water durability by mixing materials with poor water durability is the change in the structure of VOx and PO4 units toward units with low polarizability that have good water durability. As reported by Feltz et al., if the amount of P2O5 is further increased in VPx glasses, the water durability decreases because the connectivities of PO4 tetrahedra might change into Q2 and/or Q3 structures20. In other words, our results imply that the water durability of VPx glasses can be improved by changing the interconnection of VOx and PO4 polyhedra and the valence of vanadium, for example, by adding additives. This presumption is supported by the fact that the water durability of VPx glasses improves with the addition of Sb2O3, which causes an increase in the ratio of V4+/Vtotal21,48.

The thermal stability, an indirect indicator of glass forming ability, of VPx glasses also improves as the amount of P2O5 increases, which indicates that the driving force of crystallization is reduced. Since a driving force arises from the difference in chemical potential between the glass state and crystalline state, our result indicates that the difference becomes smaller as the amount of P2O5 increases. We conclude that there are two reasons for this. One reason is that the atomic structure of the glasses is very far away from the conventional oxide glasses classified by Zachariasen1 and Sun2, and the similarity with a crystalline structure increases as the amount of P2O5 increases. The similarity between glass and crystal is often discussed for the fast phase-change materials used for DVD/Blu-ray media, where the topology in terms of the ring-size distribution between glass and crystal is similar, whilst coordination in the glassy phase is significantly lower than that in the crystal phase. Therefore, it is concluded that this similarity is the crucial reason for the rapid phase change between two phases49. Indeed, the increased packing density and large fraction of edge-sharing in the glassy state are a remarkable signature of P2O5 rich glass. Accordingly, these behaviours demonstrate that the similarity between glass and crystal beyond the nearest neighbour distance (for example, polyhedral connection and atomic packing density) is rather important for the glass properties because the first V-O coordination number and the valence of vanadium of VPx glasses are different from those of crystal. In other words, this indicates that the similarity beyond the nearest neighbour distance is important for decreasing the difference in chemical potential between the glassy state and crystalline state. The other reason is the high ratio of V4+/Vtotal in the VPx glasses. Sakurai et al. also reported that the thermal stability of VPx glasses improves as the amount of P2O5 increases, but that of their VPx glasses is significantly lower than that of our VPx glasses22. Although the packing densities of their VPx glasses were not mentioned in ref. 22, the ratio of V4+/Vtotal in their VPx glasses was lower than that of our VPx glasses. Accordingly, we suggest that the high ratio of V4+/Vtotal is an important factor for improving thermal stability, too. However, it is still not certain whether V4+ states themselves reduce the chemical potential between glass and crystal or that they increase the atomic packing density and edge-sharing VOx polyhedra.

The hardness of the VPx glasses also improves as the amount of P2O5 increases. We conclude that this should be explained by an increased packing density associated with a higher V-O/O-M coordination number and increase in edge-sharing VOx polyhedra. Indeed, Rosales-Sosa et al. reported that a high atomic packing density and dissociation energy per unit volume of components increases the values of hardness and the elastic modulus50. We calculated the atomic packing density in Fig. S6. It was found that the increase in P2O5 content caused the glass structure to have a higher packing density with increased V-O/O-M coordination numbers and increase in edge-shared VOx polyhedra. The dissociation energy per unit volume of components should increase with the number of bonds per unit volume of components. Furthermore, the reduction in vanadium ions might affect the dissociation energy, whilst the cavity volume in glass should be independent of hardness because it does not affect the packing density and dissociation energy.

In addition, we have previously reported that the thermal expansion coefficients of VPx glasses did not change significantly despite an increased glass transition temperature as the amount of P2O5 increased21. Generally, the thermal expansion coefficient of glass materials increased as the glass transition temperature decreased. Such an anomalous coefficient was also identified with the network configuration in ZnO-P2O5 glass systems41. We calculated the fractions of M-O ring distributions as shown in Fig. S7. The distribution remained the same in the range of glasses VP10 to VP44. Hence, our results imply that this anomalous thermal expansion coefficient was related to the fractions of the distributions only because of a structural feature that does not change in the range of glasses VP10 to VP44. These results for the structure-property relationship demonstrate that the properties of glass are strongly affected by the structure and valence of network forming cations.

Conclusion

In this article, we discussed the relationship between the properties and atomic structure of V2O5-P2O5 glass as characterized by RMC modelling on the basis of neutron and synchrotron X-ray data. The present findings indicate that the structure and valence of network formers is important for designing their properties. Adding P2O5 causes the packing density of atoms to increase and the amount of vanadium ions to decrease in V2O5-rich glass, resulting in a glass structure that is associated with an increase in edge-sharing VOx polyhedra. We find that these are important for improving the water durability, thermal stability and hardness. In particular, the valence and structural change of vanadium affect the change in the water durability and thermal stability. We are confident that the unusually small coordination number of V4+ is especially important for water durability and that the similarity between glass and crystal beyond the nearest neighbour distance is important for thermal stability. The results presented in this study are a significant advance in understanding the fundamental properties of glass materials. Furthermore, this work paves the way towards glass sealing materials being completely lead-free and cathode materials for secondary batteries being improved.

Materials and Methods

Sample preparation

Glass samples with a nominal molar composition of (100 − x)V2O5-xP2O5 were synthesized by melt quenching of V2O5 and P2O5 powders (Kojundo Chemical Laboratory Co., Ltd.). 100 g of the mixture was melted in a platinum crucible and kept at melting temperature for one hour. The molten glass was cast onto a stainless-steel plate at 100 °C. The prepared glass samples were annealed at 10 °C higher than the glass transition temperature and slowly cooled at 1 °C min−1 for 40 min to relieve residual internal stress. V2O5 glass was synthesized by using a twin roller method. The oxide powder was melted at 800 °C for 5 min in a platinum crucible and then rapidly quenched by rotating rollers.

Characterization

The cation compositions of the resulting material were determined by wavelength dispersive X-Ray fluorescence spectrometers (Rigaku, ZSX Primus II). The resulting material was fully amorphous, and this was confirmed by X-ray diffraction (XRD) using a diffractometer system equipped with a monochromatic Cu Kα radiation source (Rigaku, RINT-2000). The fraction of the reduced amount of V ion ([V4+]/[Vtotal]) was estimated by measuring the quantities of pentavalent vanadium ions [V5+] and total vanadium ions [Vtotal] by oxidation-reduction titration, assuming that the V ions in the glass consisted of V5+ and V4+. The densities of the glass samples were measured by using a dry pycnometer (Micromeritics, AccuPyc II 1340). The atomic number densities of these samples were calculated from the density and V4+/Vtotal ratio. The apparent molar volumes of O ions were also calculated from the density and V4+/Vtotal ratio by using the formula described by Drake et al.51. The glass transition temperature (Tg) and crystallization temperature (Tc) were measured by using differential thermal analysis (DTA) (Advance Riko, DT-1500) at a heat rate of 5 °C min−1. The weight loss in water at 70 °C for 30 min for the glass sample plates was measured to determine the water resistance. Vickers microhardness measurements were made on the sample surfaces at room temperature by using an auto hardness test system (Matsuzawa, AMT-X7FS) with an accuracy of ±0.04 GPa. A load was applied for 0.98 N, 15 s.

Structural analysis of glass

High-energy X-ray diffraction experiments (HEXRD) were carried out at room temperature by using the BL04B2 beamline of SPring-852. The incident X-ray energy was 61.6 keV as obtained from a Si(220) crystal monochromator. Diffraction patterns of the samples, placed in a silica glass tube, were measured in a transmission geometry. The collected data were corrected by using a standard program52. X-ray absorption fine structure (XAFS) experiments were carried out by using the BL14B2 beamline of SPring-853. The XAFS samples were ground with boron nitride and made into pellets. V-K edge spectra were measured by using a Si(111) double-crystal monochromator in transmission mode. Ionization chambers were used to measure the intensity of the incident and transmitted X-rays, and the quick scan technique (QXAFS) was used for this measurement. These spectra were normalized and analysed by using Athena54. Time-of-flight neutron diffraction (ND) experiments were conducted by using the total scattering spectrometer NOVA at the BL21 beamline of the Materials and Life Science Experimental Facility (MLF), Japan Proton Accelerator Research Complex (J-PARC)55. The samples were placed in a cylindrical vanadium cell (6 mm in diameter). The observed scattering intensities for the samples were corrected for instrumental background, absorption of samples and cell56, and multiple57 and incoherent scatterings and then normalized by the incident beam profile, which obtained from the scattering intensity for a vanadium rod. 51V spectra were measured at 11.7 T (JEOL, ECA-500 FT-NMR). A NaVO3 aqueous solution (0.16 mol dm−3) was taken as a reference (δ = −574.28 ppm). The pulse delay was 1 s. Raman spectra were recorded by using a LabRam spectrometer (Jobin-Yvon). A laser with an emission wavelength of 514.5 nm and a power of 0.2 mW was used.

Structural modelling

The RMC modelling was performed on an ensemble of 3500, 4138, 3989 and 3818 particles for VP0, VP10, VP28 and VP44 glass, respectively. The starting configurations were generated by hard-sphere Monte Carlo simulations with constraints applied to avoid physically unrealistic structures. The constraints on the P-O connectivity were that all phosphorus atoms were coordinated for four oxygen atoms for atomic distances of up to 1.7 Å. X-ray SX(Q), neutron SN(Q) and k3χ(k) EXAFS data measured at the V-K edge were fitted simultaneously by using RMC + + code58. EXAFS back scattering tables were obtained from FEFF calculations59. The ring statistics were calculated by primitive rings analysis60,61,62 using R.I.N.G.S. code63. Cavity analysis was carried out by employing pyMolDyn code64. The cut-off distance rc for cavity calculation was 2.3 Å.