Introduction

The physico-chemical properties of silicate oxide glasses depend strongly on the nature and the relative proportions of the cations in their composition. The effects of these different elements on the chemical durability of glasses have been widely studied in various domains ranging from bioglasses with medical applications,1 materials used to regulate the CO2 cycle,2 lead crystal glass found in drinking vessels3 or vitreous matrices fabricated to confine radioactive waste.4 Understanding and modeling the behavior of nuclear waste packages in contact with groundwater is essential in the context of a deep geological disposal of vitreous matrices.5 The resulting glass composition can vary widely depending on the waste composition; in this context the effects of the major elements of nuclear glasses on the various alteration regimes have been investigated in depth (see for example Gin et al.6 and the references therein). Their exact roles can however be difficult to ascertain since comparing different glasses involves both structural and chemical composition changes that are difficult to separate. To consider only structural changes, the chemical composition has to be kept constant. This can be achieved by modifying only the thermal history of the glass. Variations in the geometric and chemical disorder can then be evaluated as a function of the glass quenching rate.7

This approach may be particularly relevant to better understanding the structural damages induced by self-irradiation in nuclear glasses due to alpha decay of minor actinides. The displacement cascades induced by the recoil nuclei generated by alpha decay may induce the same kind of structural reorganizations in borosilicate glasses as do a very rapid thermal quench, leading to a new structural state at a fictive temperature higher than that of the pristine glass.8,9,10 When the whole volume is damaged, the accumulation of additional displacements in the glass structure does not modify it significantly anymore. Some properties, such as mechanical properties or density, reach then a plateau.11 However, glasses submitted to external heavy-ion irradiation, simulating the effect of the recoil nuclei, show more significant effects than those seen in 244Cm doped glasses.12,13 This behavior was attributed to the local annealing induced by the alpha particle’s loss of electronic energy that partially repairs the damages generated by the recoil nuclei.14 Structural changes induced by self-irradiation could in particular affect the alteration behavior of glass matrices. Rapidly quenched glass may thus constitute a representative sample to investigate the chemical durability of these glasses in a laboratory setting. The hyperquenched glasses, as fibers, are not however easy to handle and reliable direct measurement of their fictive temperatures by the classical calorimetric method is challenging.15 To overcome this difficulty, we proposed here to use the Raman and Brillouin spectral signature of the cooling rate. Due to this new approach, an initial calibration of the spectroscopic evolution with the cooling rate was conducted on one bulk sample.

It has been shown that hydrogen diffusivity increases in soda-lime glasses during alteration at pH 1 when the fictive temperature increases.16 The lower density of these glasses may thus offer a more free volume that facilitates interdiffusion. A recent work was performed on dissolution in pure water of annealed and melt-quenched binary and ternary borosilicate glasses.17 It was found that glass compositions having the most significant structural variations after the thermal treatment, especially through boron coordinence, results in higher dissolution changes. We investigated here more complex glass composition through fibers, rapidly quenched during the drawing process and then annealed for reference (called in the paper “as-received fibers” and “annealed fibers”, respectively). The International Simple Glass (ISG) was considered, a six-oxide reference glass widely used to examine the alteration in aqueous media of nuclear glasses.18 Brillouin and Raman spectroscopies are respectively sensitive to acoustic (providing volume information) and optical (sensitive to local atomic vibrations) phonons. These techniques were used here to link glass structural and volume changes to cooling rates and the corresponding fictive temperatures. Structural changes between annealed and as-received fibers, and then between the pristine and altered fibers were characterized by multinuclear solid-state NMR spectroscopy, which is sensitive to both geometric and chemical disorder. Alteration experiments were performed at various stages of the reaction and different pHs. The forward dissolution rate, driven by hydrolysis mechanisms, was measured in media dilute enough to avoid any effect from the elements in solution on the alteration rate. The residual alteration rate, reflecting its long-term behavior, was obtained in initially silicon-saturated media (avoiding hydrolysis mechanisms19). Fibers ensured that the samples used in the dissolution experiments all had the same surface state, thus limiting surface-related uncertainties20 in the measured rates.

Results

Thermal history and thermodynamic properties of glass samples

Raman and Brillouin spectroscopies were used to evaluate the cooling temperature and fictive temperature of the glass fibers, in order to reveal their thermal history and quantify its effect. First of all, Brillouin and Raman calibration curves were measured in situ for bulk samples cooled at different rates. These data were then used to extract the quenching parameters of the as-received fibers. Differential scanning calorimetry (DSC) was used to address the thermodynamic properties of fibers.

Figure 1a shows the average position of the Brillouin shift for samples heated at 5 K min−1 between 150 and 620 °C after being subjected to various cooling rates between 0.5 K min−1 and 300 K min−1. At the beginning of the heating process, the Brillouin shift is very sensitive to the initial cooling rate of the glass. This is consistent with the fact that the density and elastic properties of the glass decrease with the increase in quenching rate.21,22 Between 150 and 450 °C, the Brillouin curves for the samples are all parallel, with a shallow slope. This evolution corresponds to thermal expansion, which is weakly sensitive to the initial cooling rate. Above the glass transition temperature (Tg = 575 °C), the Brillouin curves merge and decrease sharply, reflecting the complete relaxation of the glass structure. Some differences in behavior are observed between 450 °C and Tg. In the samples cooled more slowly, the glass structure remains into the equilibrated state during heating at 5 K min−1 and the transition to higher temperatures is smooth. For the samples cooled more rapidly on the other hand, at 75 K min−1 and above, heating at 5 K min−1 partially relaxes the glass before it becomes liquid like at higher temperatures.

Fig. 1
figure 1

a Brillouin shift as a function of temperature of ISG fibers cooled at different rates. b Brillouin shift extrapolated to room temperature as a function of the cooling rate of the fibers. The experimental error for confidence level of 95% was determined as ±0.08% of the mean value by realizing a map in an homogeneous sample (corresponding here to ±0.03 GHz which is the size of the square). The cooling rate predicted for the as-received fibers from its Brillouin shift is indicated in red. c Fictive temperature of ISG fibers as a function of the cooling rate. The fictive temperature error bars were estimated to be the deviation of the linear fits in a. The fictive temperature obtained for the as-received fibers from its cooling rate is indicated in red

Extrapolating the Brillouin curves to room temperature reveals the Brillouin shift corresponding to each of the quenching rates and yields the linear relationship shown in Fig. 1b. The Brillouin shift is greater, the slower the cooling rate is, in agreement with previous measurement on soda-lime glasses.21 Extrapolating these shifts to the value measured for the as-received fibers (quenched during the drawing process) gives a quenching rate of 0.7 × 106 K min−1 (Fig. 1b). The intersections between the glass linear extrapolation and the liquid one in the Brillouin curves (Fig. 1a), which occur at the transition between the supercooled liquid and the frozen glass, give the Brillouin fictive temperature of the fibers. Figure 1c shows that the fictive temperature increases linearly with the cooling rate. The fictive temperature deduced for the as-received fibers (from the previous determined quenching rate of 0.7 × 106 K min−1) is 623 °C. The same approach gives a fictive temperature for the annealed fibers of 558 °C, i.e., 65 °C lower.

Complementary glass local structural information can be obtained for specific structural groups by using the same method with Raman spectroscopy. We have shown previously on borosilicate glasses that the vibrations of SiO4 and BO3 groups are sensitive to variations of the glass fictive temperature.7 Recent molecular dynamics simulations clearly highlighted these effects in ternary sodium borosilicate glasses in relation with the cooling rate.23 Figure 2a compares the Raman spectra of as-received and annealed fibers and of the ISG bulk glass cooled at an intermediate rate (10 K min−1). The inset shows contribution of the Gaussian components typically assigned to SiO4 and BO3 species whose variations were determined as a function of the cooling rate in Fig. 2b, c. The band around 1450 cm−1, which consist of two Gaussian components, is assigned to stretching vibrations in BO3 units.24,25 We observed previously that the increase in intensity of this band was well correlated with the increase of the signal assigned to BO3 units in 11B MAS NMR spectra.26 The narrow band around 630 cm−1, assigned to vibrations from danburite-type B2O7–Si2O7 rings,27 is slightly weaker in the spectrum of the as-received fibers. The ratio of the intensities of the peaks at 1360 cm−1 and 1445 cm−1 is plotted in Fig. 2b for cooling rates ranging from 0.5 to 300 K min−1. By using least-square linear regression analysis, a cooling rate of about 105 K min−1 was obtained from the extrapolation of the data for the as-received fibers. The same approach was used for gaussian components at lower frequencies, in the large band between 830 cm−1 and 1275 cm−1 associated with the symmetric stretching modes in SiO4 tetrahedra.28,29 Plotting the ratio of the intensities of the peaks at 970 cm−1 and 1070 cm−1 as a function of the quenching rate yields a cooling rate for the as-received fibers of the same order as the one obtained from the boron vibrations (Fig. 2c). These values are about one order of magnitude lower than those calculated from the Brillouin data (close to 106 K min−1). Another notable difference between these two datasets is that while the evolution of the Brillouin shift shows complex relaxation process between 450 °C and Tg, no equivalent perturbation was observed in the Raman data (peak intensity ratio vs temperature, Fig. 3).

Fig. 2
figure 2

a Raman spectra of the as-received and annealed ISG fibers, and of ISG cooled at 10 K min−1. The inset shows the fit of the bands around 1000 cm−1 and 1400 cm−1. b, c Intensity ratios as a function of cooling rate of the peaks (b) at 1360 cm−1 over the one at 1445 cm−1 and (c) at 970 cm−1 over the one at 1070 cm−1. The error bars were taken as the standard deviation for the Gaussian fitting of the related peaks

Fig. 3
figure 3

Evolution of the intensity ratio of the fitted Raman band at 970 cm−1 over one at 1070 cm−1 as a function of temperature for ISG fibers cooled at different rates

Using the onset of the glass transition30 for the sample cooled at 10 K min−1, the glass transition temperature Tg was measured at 573 ± 1 °C. This value is in good agreement with previous works on ISG glass reporting 575 °C.22 The difference in molar heat capacity near Tg, ΔCP(Tg), between the annealed and as-received fibers was measured to be 28.9 ± 0.3 J mol−1 K−1 which is comparable to the value of 28.1 ± 0.2 J mol−1 K−1 given in ref. 31 Using the respective Brillouin fictive temperature of the fibers (Tf1 = 558 °C for the annealed ones and Tf2 = 623 °C as-received) and assuming that the heat capacity does not vary much with the fictive temperature and that ΔCP remains constant over the fictive temperature interval considered, this difference in heat capacity can be used to calculate the variations in thermodynamic properties. The difference of enthalpies of vitrification,

$$\Delta H_{v2} - \Delta H_{v1} = {\int}_{T_{f1}}^{T_{f2}} {\left( {C^ \circ _{P_l} - C^ \circ _{Pg}} \right){\rm{d}}T \approx \Delta C_P\left( {T_g} \right) \times \left( {T_{f2} - T_{f1}} \right)}$$
(1)

in which \(C^\circ _{P_l}\) and \(C^\circ _{Pg}\) are respectively the heat capacities of the liquid and the glass, is thus 1.88 kJ mol−1. Likewise, the difference in configurational entropy, calculated using

$$S^\circ \left( 0 \right)_2 - S^\circ \left( 0 \right)_1 = \mathop {\int }\nolimits_{T_{f1}}^{T_{f2}} \frac{{C^\circ _{P_l} - C^\circ _{Pg}}}{T}{\rm{d}}T \approx \Delta C_P\left( {T_g} \right) \times {\rm{ln}}\left( {\frac{{T_{f2}}}{{T_{f1}}}} \right)$$
(2)

is found to be 2.18 J mol−1 K−1. This gives a significant variation of Gibbs energy of formation,

$${\mathrm{\Delta }}G^ \circ _f\left( {298} \right)_2 - {\mathrm{\Delta }}G^ \circ _f\left( {298} \right)_1 = \left( {\Delta H_{v2} - \Delta H_{v1}} \right) - 298\left( {S^ \circ \left( 0 \right)_2 - S^ \circ \left( 0 \right)_1} \right)$$
(3)

of 1.23 kJ mol−1 between the two samples.

Pristine glass structure

The NMR parameters for the different nuclei are listed in Table 1. As expected, the annealed fibers are slightly denser (about 1%) than the as-received fibers (Table 2).

Table 1 Parameters used to fit the NMR spectra of ISG fibers
Table 2 Density and alteration rates for ISG fibers

Significant changes are observed between the 11B MAS NMR spectra of the as-received and annealed fibers (Fig. 4a). The best fits obtained for the two spectra using methods described elsewhere,26 put the proportion of tetrahedral boron in the annealed fibers at 50.0% vs 41.1% in the as-received ones (Fig. 5a, the fitting parameters are listed in Table 1). Rapid quenching is thus associated with a decrease of 8.9% in the proportion of BO4 groups. The chemical shift of the BO4 line is also more negative in the spectrum of the as-received fibers. This peak comes from boron atoms connected only to silicon atoms (4B(4Si)) and from boron atoms connected to three silicon and one boron atoms (4B(3Si,1B)).26,32 Since the latter has more positive chemical shifts than the former, the observed shift in the BO4 peak indicates that there is a larger proportion of 4B(4Si) sites in the as-received fibers.

Fig. 4
figure 4

a 11B and b 29Si MAS NMR spectra of as-received and annealed ISG fibers, either pristine (solid lines) or after 100 days of alteration at pH 7 and 90 °C (dashed lines). The spectra are scaled to the same height

Fig. 5
figure 5

Measured and simulated 11B MAS NMR spectra of the as-received and annealed ISG a pristine fibers and b altered fibers, using the same peak positions and linewidths for all the spectra

The 11B MQMAS NMR spectra shown in Fig. 6a provide more accurate information on the distribution of boron sites in the fibers. The distributions of isotropic NMR shifts are broader in the as-received fibers than in the annealed ones, for both BO3 and BO4 groups, as highlighted by the projections onto the isotropic NMR shift axis (Fig. 6b).

Fig. 6
figure 6

a 11B MQMAS NMR spectra of as-received and annealed ISG fibers (upper panels, BO3 region; lower panels, BO4 region) (*: spinning sidebands). b Projections onto the isotropic NMR shift axis of the spectra shown in part a

Figure 4b shows that there is no significant change between the 29Si MAS NMR spectra of the as-received and annealed fibers. This does not necessarily mean that the Qn (n indicating the number of bridging oxygens per tetrahedron) speciation of the SiO4 groups is identical in the two glasses. The 29Si chemical shift of silicon groups is sensitive to the nature of their next-nearest neighbors. In previous works7,33 we showed that the effects of depolymerization in the 29Si NMR spectra were masked by an increased presence of boron as a next-nearest neighbor. It is generally expected that substitution of one bonded silicon for one aluminum impact the isotropic chemical shift of the central silicon by an increase of about 4–5 ppm.34 Similar increase has been reported for Si–O–B linkages,35,36 but shown to depend on the boron speciation33 (no change is observed for Si–O–BIII in contrast to Si–O–BIV). Because of the glass composition complexity, these data are therefore inconclusive as to possible differences in the degree of polymerization in the as-received and annealed fibers.

The single broad line in the 23Na MAS NMR spectrum of the as-received fibers is slightly wider than the corresponding line for the annealed fibers (Fig. 7a). The interpretation of the width of 23Na MAS NMR lines only, indicative of high substantial disorder, is tricky. 23Na spin-echo experiments measuring 23Na–23Na dipolar couplings37 were performed to investigate the spatial distribution of sodium. Indeed, the spin echo experiment refocuses most of the NMR interactions (giving the echo), except the homonuclear dipolar interactions which are only averaged out by MAS (to first order). Decay of the spin echo is therefore dominated by the effects of the 23Na–23Na dipolar interactions. Figure 7b shows that the 23Na spin-echo signal from the as-received fibers decreases more rapidly, indicating that the 23Na–23Na dipolar couplings are much stronger than in the annealed fibers. This suggests that sodium is distributed less homogeneously in the as-received fibers.

Fig. 7
figure 7

a 23Na MAS NMR spectra. b 23Na spin-echo intensity as a function of the spin-echo delay. c 27Al MAS NMR spectra, and d 23Na{27Al} rotational-echo double-resonance (REDOR) intensity as a function the REDOR time for as-received and annealed ISG fibers

The 27Al MAS NMR line from the as-received fibers is broader than the one from the annealed ones (Fig. 7c), but no difference in the coordination of the aluminum atoms (tetrahedral in both cases, as confirmed by 27Al MQMAS spectra given in Fig. 8) was observed. Differences in the chemical environment of aluminum atoms, particularly concerning their charge compensation, can be evaluated using the dipolar couplings between 23Na and 27Al nuclei, measured by heteronuclear rotational-echo double-resonance (REDOR) experiments.38 The principle of REDOR, based on the spin echo experiment, is to collect two spin echo decay curves: the first denoted S0(t) using the standard pulse sequence 90°-τ-180°-τ; the second S(t) applying on the second channel a train of rotor-synchronized refocusing 180° pulses (one at each rotor period and one in the middle of each rotor period), so that heteronuclear dipolar interactions are selectively reintroduced. The REDOR signal is obtained as ΔS(t) = (S0(t)−S(t))/S0(t) and only incorporates the effects of the dipolar heteronuclear interactions. Figure 7d clearly shows that the 23Na{27Al} REDOR signal from the as-received fibers is weaker, indicating that there are fewer aluminum atoms in the vicinity of sodium nuclei.

Fig. 8
figure 8

27Al MQMAS NMR of the as-received ISG fibers

Glass alteration far and close to saturation

Two types of experiments were performed on the as-received and annealed fibers to investigate their initial and long-term alteration behavior, namely when the alteration process in aqueous solutions is governed by hydrolysis of the silicate network, and after by the formation of a passivating layer.19,39,40,41

The first series of experiments were performed under forward dissolution rate conditions at 90 °C and pHs of 6, 7.5, and 9. The dissolution rates obtained from the measured silicon release rates are summarized in Table 2 and compared in Fig. 9a. Measurements of the aluminum concentration showed that the normalized mass losses were very close to that of silicon (data not shown). The pH dependence coefficient of the forward dissolution rates for both types of fibers is 0.4, in agreement to the value determined for SON68 glass.42 The forward dissolution rates of the as-received fibers are higher under all three pH conditions than those of the annealed fibers, by a factor of 1.4–1.8.

Fig. 9
figure 9

a Forward dissolution rates as a function of pH (mean of measurements from triplicated experiments) and b equivalent thickness of the altered layer (measured from the normalized B and Na mass losses) as a function of leaching time at 90 °C for as-received and annealed ISG fibers. The uncertainty in the dissolution rates and equivalent thicknesses is 10%

Residual alteration rates were measured in solutions at pH 9 initially saturated with respect to amorphous silica, to investigate the effect of the alteration layer that forms on ISG fibers on its subsequent alteration.19 No secondary phases precipitate on the surface of ISG as it becomes altered, such that there is a direct link between the alteration rate of the glass and the passivating properties of the alteration layer.6,43 The residual rate calculated according to the first data points (both from the Na and B release in solution) for the as-received fibers is about 2.4 times higher than the value obtained for the annealed fibers. This difference diminishes with time at about 1.4 considering the data points between 72 and 244 days (Table 2). Figure 9b shows the equivalent thicknesses of the altered layers at the end of the experiment that are about 2.2 times higher for as-received fibers. Analyses of the solutions indicate that the compositions of the alteration layers are almost identical (Table 3).

Table 3 Chemical composition of altered layer deduced from solution analyses for ISG fibers altered for 224 days at 90 °C and pH 9

To investigate the structure of the alteration layer by NMR, a final experiment was performed under conditions chosen to maximize the proportion of the altered glass. Keeping constant the solution at pH 7 yielded a fraction of altered glass of 91% for the as-received fibers and 60% for the annealed fibers (calculated from solution analyses from boron release) after 100 days at 90 °C. This difference is consistent with the measurements performed at pH 9. As for the pristine samples, there is no observable change between the 29Si MAS NMR spectra of the two types of fibers after alteration. Figure 4b shows however that the single broad peak observed in each case appears at a more negative chemical shift for the altered than the pristine glasses, highlighting the structural rearrangements occurring in the altered layer. The 29Si chemical shift of silicate groups becomes more negative as the number of bridging oxygens increases (i.e., as the connectivity of the glass network increases); as shown above however, this variation could also be influenced by next-nearest neighbor effects. On the other hand, comparison of the 11B MAS NMR spectra in Fig. 4a reveals that the proportion of trigonal boron is lower in the altered samples, mainly for the as-received fibers, indicating that BO3 groups are released preferentially over BO4 groups. Considering that 91% of the quenched fiber was altered and subtracting the pristine glass contribution, the resulting BO3 proportion in the alteration layer is about 56.5% (58.9% in the as-received fibers). The proportion of boron species for altered fibers is given in Fig. 5b.

Discussion

The Raman, Brillouin, and NMR spectroscopy data collected in this study provide complementary information on the structural changes induced by very rapid quenching of soda-lime alumino-borosilicate glass and the resulting effects on its chemical durability. Studying quenched samples reveals the effect of the thermal history, and thus of the fabrication process, on the properties of the glass. This approach can also be relevant to investigate the radiation damages in nuclear glasses, whose effects are similar to those of quenching.8

The Raman spectroscopy study might suggest that increasing the cooling rate should increase the average number of bridging oxygens per SiO4 group since the peak contribution at 1070 cm−1, usually assigned to high Qn species, increases compared to the contribution at 970 cm−1, usually assigned to lower Qn species44 (Fig. 2a). This observation is in agreement with in situ Raman data obtained on ISG glass with the increase of temperature.22 An opposite trend was noted for sodium borosilicate glasses (without Zr and Al), whose degree of polymerization, as expected, was found to be lower the higher the fictive temperature was.7,8 However, it is worth noting that the Si-O stretching region of the Qn species is strongly affected by the presence of the T second-neighbors of Si (T = Zr, Al, B). Then, the decrease of the band at 970 cm−1 may be more likely related to the decrease of the Si atoms connected to two other second-neighbors than Si, Q4(2T) units, overlapping the Q2 units.13 As we noticed for the boron with an increase of 4B(4Si) instead of 4B(3Si,1B) sites, quenched glasses could then limit mixing between network formers around Si atoms. The slight shift in the Si–O–Si bending mode from 496 cm−1 in the annealed sample to 500 cm−1 in the as-received fibers, suggests that the average Si–O–Si angle decreases with the increase of glass fictive temperature.45 It has been reported that Si–O–Si bond angles tend to be smaller with the increase of Na in the vicinity of the Si groups,46 which is in agreement here with the additional Na near Si in the as-received fibers coming from charge compensation of BO4 groups in the annealed fibers.

The break in the overall trend observed in the Brillouin data between 450 °C and Tg for the glasses cooled most rapidly (Fig. 1a) was not observed in the Raman data (Fig. 3). This may be related to the lower relaxation energy of the groups probed by Raman spectroscopy which characterize the short-range order in the glass through iono-covalent bonds. At this length scale, the liquid to solid transition is similar at all quenching rates. The fact that the quenching rate determined from the Raman data for the as-received fibers (105 K min−1) is close to a factor of ten lower than the value extrapolated from the Brillouin data (0.7 × 106 K min−1) may also be due to the different length scales probed by the two techniques (short for the former and long for the latter). Fictive temperatures are traditionally determined by calorimetry, a bulk technique that is therefore presumably closer to the Brillouin than to the Raman data. The relatively good agreement between DSC and Brillouin is here confirmed by the fact that the Brillouin fictive temperature for a glass with a slow cooling rate of 10 K min−1 is of 570 ± 2 °C, within the error bar of the Tg at 573 ± 1 °C determined by DSC. Furthermore, the larger dispersion in the Raman measurements makes the extrapolations to determine the quenching rate of the as-received fibers more uncertain. The values obtained from the Brillouin data for the quenching rate (0.7 × 106 K min−1) and the fictive temperature (623 °C) therefore seem more reliable.

Multinuclear NMR first provides information on the structural disorder induced by rapid quenching. The 11B MQMAS spectra (Fig. 6) show that the isotropic chemical shift distribution (δiso) is broader in the as-received fibers than once they are annealed. Since it was shown that 11B δiso is sensitive to the B–O–(B, Si, or Al) bond angles in the glass,33 these data suggest that the bond angle distribution of the as-received fibers is broader, thus reflective of an increased geometrical disorder. Regarding the cations in the glass, the 23Na spin-echo experiments highlight that sodium is distributed more heterogeneously in the as-received fibers (the distance to the neighboring sodium atoms is shorter on average), possibly starting to form some alkaline clusters or local enrichment. In glasses with compositions similar to that of ISG, sodium acts mainly as a charge compensator, first for AlO4 and then, for the residual Na available, for BO4 groups.47 The 23Na–27Al REDOR experiments clearly show that the proportion of sodium atoms in the vicinity of aluminum groups is lower in the as-received fibers. Since the coordination of aluminum does not change, part of aluminum groups must be charge compensated by other cations than sodium, here calcium. As a consequence, an increase of the quadrupolar splitting and thus the linewidth of the 27Al MAS spectra is expected,47,48 as actually observed here. Therefore, rather than compensate the charges of BO4 and AlO4 groups, part of the sodium ions in the as-received fibers acts as additional network modifiers, probably forming non-bridging oxygen bonds with silicon groups and adopting a less homogeneous distribution. Some of AlO4 groups are Ca-compensated, while a part of BO4 is transformed into BO3 units.

The non-negligible difference in thermodynamic properties also reveals the effect of structural disorder. The variation of Gibbs energy of formation between as-received and annealed fibers (1.23 kJ mol−1) is similar to the one between cristobalite and quartz (1.7 kJ mol−1).49 This difference, which is substantial with respect to the intrinsic stability of the glasses, suggests that their properties also differ. The difference of Gibbs energy could here be related to a variation of mean bond energy and an increase of disorder. As for SiO2 polymorphs, the sample with the higher Gibbs energy will have a higher solubility.50 It could also reflect the impact on the observed dissolution rates.

Leaching experiments clearly point out correlations between the thermally induced structural disorder of ISG and its chemical durability in solution at different stages of alteration. Under forward dissolution rate conditions at various pH near neutrality, between 6 and 9, the silicate network of the as-received fibers was hydrolyzed 1.4 to 1.8 times faster than that of the annealed fibers. Hydrolysis of the silicate glass network, which is the main mechanism governing the forward dissolution rate of the glass, is affected by the structural disorder. A higher NBO and BO3 content and possibly a lower proportion of Si second-neighbors such as Zr or Al in the as-received fibers may also facilitate the hydrolysis reactions.39,51,52 Same trends were noticed in a ternary sodium borosilicate glass with similar B2O3 content of about 18 mol%; a variation of Tf of 46 °C increased the BO3 of 7.9%, resulting in a dissolution increase of about 1.5.17 Glass compositions with higher boron concentration were subjected to lower boron coordinence variations between quenched and annealed samples. The difference was then also less pronounced concerning the dissolution behavior.17

The experiments performed under silicon saturation conditions show that the annealed glass dissolves faster than the quenched one. The difference between the two rates is maximal at the beginning of the experiment, which corresponds to the case of a more direct glass contact with the fluid since no passivating layer has yet formed. Because the composition of the onset solutions is the same, the difference of reactivity noticed at the beginning origins from structural differences between the two pristine glasses. Next, during alteration under silica saturated conditions, structural changes occurred within alteration layer. The AlO4 et ZrO6 groups mostly charge-balanced by sodium in the pristine glass become compensated by calcium in the altered layer, as has been reported previously for other silicate glasses47 and more recently for ISG.53 While as has been shown elsewhere,19,43 boron is mostly absent from the altered layer, but trigonal boron groups are dissolved preferentially over tetrahedral units in the as-received fiber. Possible higher BO3 grouping in the rapid quenched glass may enhance the release of mobile species in the residual rate regime. The faster residual alteration rate of the as-received fibers can also be explained by their more disordered structure (higher bond angle distribution, less-distributed alkalis and network formers around Si).

During aging of the alteration layer at higher reaction progress in the residual rate regime, the rate drops quickly because of the passivation layer growth.19,43,54 Recent studies have shown that the alteration layer undergoes major reorganization owing to the high reactivity of the silicate network.53,54,55 It results in the formation of a microporous, hydrated, amorphous material with a large majority of non-connected pores, strongly reducing water diffusion. Moreover, as the alteration layer is an evolving material undergoing pore ripening, due to incessant hydrolysis and condensation of siloxane bonds,54 transport of aqueous species through the growing layer becomes rate-limiting. The present study points out that the residual rate of the two glasses becomes closer at longer duration. It can then be suggested that the passivating properties of the alteration layer, which are strongly linked to its reorganization,39,56 will eventually dominate glass alteration after a first step of a control by the pristine glass structure. To confirm this interpretation, more research is needed to determine water dynamics in the alteration layers formed on the two glasses.

The structural differences between the as-received and annealed fibers can be compared directly with those induced by irradiation. The decrease of 8.9% in the proportion of BO4 groups in the rapidly quenched glass is closer to the one observed in radioactive ISG doped with 244Cm (7%),12 than to the decrease induced in ISG by heavy-ions (Au) irradiation (13%).13 Furthermore, whereas a change in aluminum coordination and depolymerization was observed for the irradiated glass,12 no such changes were observed either for 244Cm doped ISG13 or for the quenched fibers in this study. These results therefore support the use of rapid quenching to mimic the effects of self-irradiation in nuclear glasses. This analogy could be used to study the mechanisms governing the chemical durability of high-activity nuclear glasses without having to handle radioactive samples.

Methods

Sample preparation

The ISG glass (60.1SiO2–16.0B2O3–12.6Na2O–3.8Al2O3–5.7CaO–1.7ZrO2) used in this study was prepared by MoSci Corporation (Rolla, MO, USA) and drawn into fibers having mean diameter of 12 μm by PPG industries, Inc. (Pittsburgh, PA, USA) (called “as-received fibers”). Some of the samples were annealed for 10 h at 575 °C to erase their thermal history and serve as a reference (called “annealed fibers”).

For the alteration rate experiments, the fibers were cut into pieces either 10–40 μm or 63–250 μm in length (Fig. 10). Fine particles were then removed by successive washes in acetone and absolute ethanol by sedimentation. The particles in the 10–40 μm size had a geometric surface area of 1653 cm2 g−1 but a surface area measured using the BET method of 3475 ± 200 cm2 g−1. This larger than expected surface area is due to the cutting process, which often left irregularities at the extremities of the cylinders. The geometric and BET surface areas are much closer for the larger particles, respectively 1384 cm2 g−1 and 1440 ± 80 cm2 g−1. All the alteration rates were calculated from the BET values. Densities were measured by helium pycnometry (AccuPyc II1340, Micromeritics).

Fig. 10
figure 10

Scanning electron micrographs of the chopped fibers, a 10–40 μm and b 63–250 μm in length. c Enlarged view of one of the fibers

NMR

NMR data were collected on a Bruker Avance II 500WB spectrometer operating at a magnetic field of 11.72 T, using a Bruker 4 mm (outer diameter of the rotor) cross-polarization MAS XH probe at a spinning rate of 12.5 kHz. 11B, 23Na, and 27Al MAS NMR spectra were acquired using short 1 μs pulses (corresponding to a tip angle of about π/12) to ensure a homogeneous excitation of all sites regardless of their quadrupolar coupling constant and thus obtain quantitative spectra. The recycle delays used were 1 s for the 23Na and 27Al and 2 s for the 11B measurements. Typically, 1024 scans were accumulated. Two types of 11B multiple-quantum MAS experiments were carried out in order to optimize the signal obtained from BO3 and BO4 units, as described previously.26 29Si MAS NMR spectra were acquired using a saturation-recovery pulse sequence with a recovery delay of 20 s (increasing the recycling delay did not alter the lineshape). For spin-echo and REDOR experiments,57,58 selective 90° and 180° pulses (frequency, 10–20 kHz; 180° pulse duration of 8–10 μs) were applied on the central transition (1/2↔1/2). The 23Na–27Al REDOR experiments were performed using a REDOR Box purchased from NMR Service GmbH, which splits the X-channel of the probe into a double-resonance circuit for close resonance frequencies (here \(\nu _0^{\rm{Na}} - \nu _0^{\rm{Al}} \approx 2\) MHz).

The 11B, 23Na, 27Al, and 29Si NMR spectra were referenced externally respectively to 1 M aqueous boric acid (19.6 ppm relative to boron trifluoride etherate), 1 M aqueous NaCl (0 ppm), 1.0 M aqueous Al(NO3)3 (0 ppm), and tetrakis(trimethylsilyl)silane powder (two lines at −9.9 ppm and −35.3 ppm with respect to tetramethylsilane). A correction was applied for the 11B MAS NMR spectra to account for the contribution of spinning sidebands to the BO4 centerband, as described by Hopf et al.51 The 11B, 23Na, and 27Al MAS NMR spectra were fitted using NMR parameter distributions, as detailed elsewhere.26,51,59 A single distribution was sufficient for the 23Na and 27Al spectra whereas at least four lines were required for the 11B spectra, two for BO3 units and two for BO4 ones.

Raman, Brillouin spectroscopies, and DSC

Raman spectra were recorded using a Horiba iHR320 spectrometer with 1800 g mm−1 gratings, powered by a blue “Sapphire SF” 488 nm laser. Data were collected from 150 to 1700 cm−1. The shorter laser wavelength is favorable for recording back scattered signals because of increased Rayleigh scattering, and allows data to be recorded at high temperatures without any overlap with black body radiation. The spectra were calibrated using a calcium carbonate reference. In order to compare spectra recorded at different temperatures, they were corrected for the wavelength and temperature Raman scattering dependence, described previously.60 Brillouin spectra were recorded using a JRS TFP 2 HC tandem multi-pass Fabry–Perot interferometer. The different heat treatments were carried out in a PerkinElmer 8500 differential scanning calorimeter that was modified to allow in situ Raman and Brillouin measurements simultaneously. The approach used, dubbed ARABICA, has been described in detail by Veber et al.61 Before each set of measurements the temperature was calibrated using α–β quartz and RbNO3 phase transitions which took place at 573 and 166 °C, respectively. All the measurements were carried out under a nitrogen gas flow of 30 mL min−1

The calibration experiments were done using double side polished plane parallel plate of the ISG glass (m = 28 mg). The spectra were recorded during the heating up cycle after a period of cooling at a set rate. The cooling rates used were 300, 150, 75, 30, 15, 10, 5, 2, 1, and 0.5 K min−1. For the in situ experiments the sample was placed in a platinum crucible covered with a pierced lid. The heating rate used was 5 K min−1 to leave enough time to properly collect Raman and Brillouin spectra. These were collected over the entire temperature range with an overall exposure time of 1.5 min (6 exposure cycles of 15 s each), such that a spectrum was recorded every 7.5 °C (1.5 min × 5 K min−1). For the heat capacity measurements a completely covered crucible was used. The heat flow measured by the calorimeter was calibrated using a sapphire disk.

Glass alteration experiments

All the alteration tests were performed at 90 ± 1 °C. The forward dissolution rate r0 was measured in highly stirred static mode using perfluoroalkoxy reactors at a low surface area to volume (SA/V) ratio of about 5 m−1. About 30 mg of glass fiber particles (63–250 µm in size) was placed in about 500 mL of solution. The forward dissolution rate (r0) rate was measured in triplicate at pH90° 6 (adjusted with nitric acid), pH90° 7.5 (initially deionized water with a resistivity of 18.2 MΩ cm) and pH90° 9 (adjusted with lithium hydroxide). The solution was regularly sampled and the Si concentration was determined by colorimetry with a Merck Spectroquant kit and a Cary Varian UV–visible spectrophotometer. ICP-OES analyses were performed to check the release of other elements. The r0 values were calculated from the normalized silicon mass loss, NL(Si), and correspond to the slope of NL(Si) as a function of time. The uncertainty in r0 was estimated to be about 10%, taking into account errors in the measured surface area of the glass, in the silicon concentration, and the repeatability of the measurements.20 The residual rates (rr) were calculated from the normalized boron and sodium mass losses between 72 and 224 days. The associated uncertainty was also 10%. Variations in the surface area of the fibers were taken into account using a shrinking core model for cylinders.62 The normalized mass losses, NL(i), (g m−2) were calculated allowing for variations in the volume of the leaching solution (when samples were extracted) using the following equations:

$${\rm{NL}}\left( i \right)_{n = 1} = \frac{1}{{x_i \cdot {\rm{SA}}}} \cdot V_0 \cdot C_{j = 1}$$
(4)
$${\rm{NL}}\left( i \right)_{n > 1} = \frac{1}{{x_i \cdot {\rm{SA}}}} \cdot \left[ {\left( {V_0 - \mathop {\sum}\nolimits_{j = 1}^{j = n - 1} {V_j} } \right) \cdot C_{j = n} + \mathop {\sum}\nolimits_{j = 1}^{j = n - 1} {V_j \cdot C_j} } \right]$$
(5)

where Cj is the concentration of element i in the jth sample of the solution (g m−3), xi is the mass fraction of element i in the glass fiber, SA is the surface area (m2) of the glass fibers in contact with the solution, V0 is the initial volume (m3) of the solution in the reactor, and Vj is the volume (m3) of the jth sample.

Residual rate (rr) measurements and the alteration experiments for NMR studies were carried out in non-stirred static mode using perfluoroalkoxy reactors. The altered glass studied by NMR was prepared by placing 300 mg of glass fiber particles (10–40 µm in size) in a silicon saturated solution ([Si] = 160 mg L−1) at pH90° 7 ± 0.2. The SA/V ratio was 120 m−1 and the pH was kept constant throughout the experiment using nitric acid. For the residual rate measurements, 350 mg of glass fiber particles (10–40 µm in size) was altered at a SA/V ratio of 150 m−1 in a silicon saturated solution ([Si] = 260 mg L−1) at pH90° 9 ± 0.1 (adjusted using lithium hydroxide). The pH of the solution remained constant throughout the experiment without any additional LiOH being required.