Impact of alkali on the passivation of silicate glass

Amorphous silica-rich surface layers, also called gels, can passivate silicate glass and minerals depending on environmental conditions. However, several uncertainties remain on the mechanisms controlling the formation of these layers. In this paper, the influence of exogenous ions supplied by solutions is studied, both on the formation and on the properties of the gel formed on international simple glass (ISG). ISG was altered at 90 °C, pH90 °C 7, in silica-saturated solutions containing various alkaline cations separately (Li+, Na+, K+, and Cs+). The alteration kinetics observed with Li and Na in the solution is similar to that observed with no ions, while K and Cs in the solution tend to decrease glass alteration. Furthermore, for K or Cs ions, the kinetics decreases as the ionic strength of the solution increases. The passivation layer formed in these solutions shows a selectivity toward cations following the series K > Cs > Na >> Li. These alkalis replace Ca from pristine glass in the altered structures, leading to differences in [AlO4]− units charge compensation. Importantly, exchange between Ca and alkali also affects the total quantity of water inside each gel and this effect is well correlated with the observed drop in glass alteration. The presence of ions has been observed to affect the speed at which aqueous solutions alter the surface of a glass of nuclear interest. Borosilicate glasses are used to store highly radioactive waste in deep geological repositories, however, they are expected eventually to come into contact with ground water. Therefore, how they change when exposed to aqueous environments must be understood. Now, a team, led by Stéphane Gin at the CEA, France, has studied the effects that various ionic solutions have on the formation and properties of the passivating silica-rich layers that can form on the surface of a reference glass known as International Simple Glass. They see that K+ and Cs+ ions decrease glass alteration kinetics, while the kinetics observed with Li+ and Na+ present are similar to solutions with no ions.


INTRODUCTION
The current French approach for high-level waste storage is by incorporating radionuclides inside a borosilicate glass matrix. Subsequently, these glass canisters will be stored in a deep geological formation. 1 However, ground water will eventually reach the glass after corrosion or diffusion through the different barriers. It is well known that glass, when in contact with water, goes through major structural and chemical changes, which need to be thoroughly understood. [2][3][4][5] Glass alteration mechanisms are still heavily debated inside the scientific community, as no consensus has been reached yet for the way the passivation layer forms on the glass surface. [6][7][8][9] To better understand glass alteration at high reaction progress, Gin, et al. 10 and Collin, et al. 11 performed studies in Si-saturated conditions and slightly alkaline pH conditions, and characterized the passivating material. In these specific conditions, B, Na, and to a lesser extent Ca are leached out. Zr and Al are not released into the solution, and no exchange is observed between 29 Si from the solution and 28 Si of the network. This supports the idea that the remaining network formers are not completely hydrolyzed, and partly re-condense after B dissolution to form a porous gel layer.
However, differences arose between these studies, mainly concerning mobile element diffusion and altered glass quantities. The only significant variance between both studies is the concentration of K in the solution (>80 mmol L −1 for Gin, et al. 10 vs <1 mmol L −1 for Collin, et al. 11 ). This raises the question on the effect of alkalis on glass alteration in Si-saturated solutions. It is well known that alkalis affect the alteration of minerals [12][13][14] and glasses 15,16 in the first stage of corrosion (highly diluted conditions leading to network hydrolysis). The presence of alkalis in solution enhances both quartz and nuclear glass dissolution. [12][13][14][15] Jollivet, et al. 15 suggested that this could be due to interactions between cations and deprotonated hydroxyls (-X-O − where X = Si, Al or Zr). These interactions could weaken Si-O bonds of the network, which could in turn facilitate network dissolution. However, at higher reaction progress, i.e. in conditions close to saturation, network hydrolysis is not the rate-limiting mechanism anymore, 17,18 yet the effect of alkalis on glass alteration in those conditions has not been studied.
In the present study, the effect of Li, Na, K, and Cs on glass alteration in Si-saturated conditions, 90°C, pH 90°C 7, was studied. In these conditions, a significant drop in alteration kinetics occurred when K or Cs were introduced in the solution. However, the drop in the kinetics was gradual as the ionic strength was increased. To explain the differences in alteration kinetics depending on the alkali in the solution, altered materials, here called gels, were characterized. Their chemical compositions were determined, highlighting a higher selectivity of the gel for K and Cs than for Na and Li. Gel structures were then examined using solid-state nuclear magnetic resonance (NMR) spectroscopy. No significant structure variations were observed, except on [AlO 4 ] − units charge compensation. Finally, water quantity and speciation were determined in each gel by combining thermogravimetric analysis (TGA) and 1 H NMR, showing a decrease in water amount when the concentration of alkali increased. The drop in alteration rate could therefore be a consequence of the gel selectivity for larger cations that could reduce water accessibility to the pristine material. This raises interesting question regarding the upgrade of nuclear waste safe storage.

RESULTS
Alteration kinetics ISG glass powders of 20-40 µm in particle size were altered at 90°C , pH 90°C 7 in Si-saturated conditions with alkali introduced separately in the solution (LiCl, KCl and CsCl with concentrations ranging from 0 to 70 mmol L −1 and NaCl with a concentration of 20 mmol L −1 : as Na is present in pristine glass, only one concentration was studied to compare with other alkali). For all experiments, the Si concentration remained stable over time, while Zr and Al concentrations stayed under the detection limit (0.10 ppm). For a similar concentration in alkali (20 mmol L −1 ), Bused as glass alteration tracer-displays different behaviors (Fig.  1a) depending on the cation present in solution. While the glass alteration kinetics in the Li-and Na-solutions is quite similar to that observed in the no alkali-solution, a significant drop is observed for K-and Cs-solutions. Additionally, the equivalent thickness ET(B) is well correlated with t 1/2 in Li-, Na-and no alkali-solutions (Fig.  1b) up to 130 days. This could indicate that diffusion of some reactive species through the growing gel is rate-limiting during this first period. The glass alteration kinetics appears to leave this square root of time tendency after 130 days in the Li-and Nasolutions. However, the model used to calculate the ET(B) (known as the shrinking core model 19 ) tends to diverge over 85% of altered glass AG(%), because actual glass particles are not spherical.
The apparent diffusion coefficient D Fick B , calculated for B between 0 and 130 days, is derived from Fick's second law, The diffusion coefficients are similar ((3 ± 0.3)-(4 ± 0.4) × 10 −18 m 2 s −1 ) and indicate that the release of mobile species is limited by diffusion through the growing gel layer for these experiments, while another mechanism occurs before~20 days in the K-and Cssolutions, drastically reducing the release of mobile species. If Na and B releases are congruent for all experiments (refer to supplementary material Fig. S1 a-e for more data), Ca release is highly dependent on the alkali introduced in the solution (Fig. 1c): Ca dissolves almost congruently with B and Na in K-and Cssolutions, while it is more retained in gels formed in no alkali-, Li-, and Na-solutions.
The ionic strength of the solution also affects the glass alteration kinetics. While with Li this effect is not notable, K and Cs have a strong effect: the K concentration effect is gradual, while the Cs concentration effect, despite being also slightly progressive, is more pronounced (Fig. 1d-f and supplementary material Fig. S1 f). These differences in alteration kinetics are interestingly correlated to differences in Ca/B congruence ratio (Fig. 1c). The more altered glasses powders are those that also display a high retention of Ca in gels (refer to supplementary material Fig. S1 g-i for more data).

Gel composition
The gel composition was determined on 3.5-5.5 µm glass powders altered for durations ranging from 1 month to 1 year in Sisaturated conditions, at 90°C, pH 90°C 7, with alkali concentration of 20 mmol L −1 . Note that using smaller particles than in previous experiments does not change the glass dissolution kinetics but makes it possible to obtain fully altered samples. The detailed chemical composition of the gels can be found in Table 1.
The results are consistent with those obtained from the solution analysis: B and Na loses are almost complete in each case. A small quantity of K can be found in each gel due to electrode contamination when measuring the pH. The percentages of remaining Ca in gel powders are coherent with Ca quantities found in the solution (Fig. 2a). The retained Ca quantity is higher for glass powders altered in no alkali-and Li-solutions, lesser in Nasolution, and significantly lower in K-and Cs-solutions. This quantity is anticorrelated with that of alkali supplied by the solution and incorporated in the gel ( Fig. 2b and Table 1), strongly suggesting that Ca exchanges preferentially with alkali following the order K > Cs > Na >> Li. Interestingly, a rise in the concentration of K is observed in the sample altered for 1 year compared to that of the sample altered for 1 month, indicating that the material keeps on evolving even after complete B and Na loss. Knowing the concentration of alkalis in the bulk solution and the water content in the different gels, the amount of alkali confined in pore water was calculated (supplementary material part 2 and Table S2 c for more information). These quantities are found to represent less than 5% of the total amount of alkali in the gel structure, indicating that the amount of alkali inside the gel structure is not correlated to the water quantity entering the gel.
The electric charge of the structure was calculated for each gel composition 20 to assess whether there is an excess of alkali (which would be confirmed by a positive charge). An excess of alkali would lead to a depolymerization of the network with formation of non-bridging oxygen atoms (NBOs). Overall, the charge is found to be balanced within analytical error, proving the absence of excess alkali in each altered structure. Such an observation is highly dependent on pH. Indeed, at higher pH, excess of cationsmainly Ca-has been observed previously. 21 Due to the absence of an excess of alkali, the remaining NBOs are expected to be silanol groups. Ca ensures most of the charge compensation in No Table 1. alkali-, Li-, and Na-samples, but is supplanted by K and Cs in the Kand Cs-samples, respectively (Fig. 2b). ToF-SIMS elemental profiles were obtained on two monoliths: one altered 30 days in a 20 mmol L −1 Na-solution and one altered 134 days in a 20 mmol L −1 Cs-solution ( Fig. 2c-d). The results are consistent with solution data (supplementary material Fig S1 c and e for more information). For each sample, the drops in concentration observed for every mobile element (B, Na, Ca and H) are located at the same depth (~2.7 µm for the Na-sample and 1.1 µm for the Cs-sample). The widths of these interfacial gradients are found to be similar within measurement error (~25 nm for the Na-sample and~30 nm for the Cs-sample). The altered layer formed in the Na-solution is homogenous, which is similar to what was observed in a previous study on a sample altered in a no alkali-solution. 11 The mean retention factor for Na and Ca in the Na-sample is of 7 ± 1% and 44 ± 4%, respectively. Na retention is slightly underestimated compared to the solid analysis data which provides a retention of~15%, while Ca retention is coherent with the data presented in Fig. 2a. For the Cs-sample, the B and Ca retention in the gel is dependent on the depth, with values of~1-11% in the first 0.5 µm and~3-30% from 0.5 to 1.1 µm. Such a phenomenon was observed by Gin, et al. 10 on a sample altered in a solution highly concentrated in K (>80 mmol L −1 ). In this medium, the authors also reported that the B retention in the inner region of the gel increases with time.
Structural analysis of the pristine glass and gels All 3.5-5.5 µm gel powders were examined using 1 H, 29 Si, 27 Al, and 11 B solid-state NMR spectroscopy. No 11 B signal was detected, confirming that it was almost totally leached out. 29 Si magic-angle spinning (MAS) Carr-Purcell-Meiboom-Gill (CPMG) NMR spectra and 1 H-29 Si cross-polarization MAS (CP-MAS) NMR signal spectra, displayed in Fig. 3a, are similar within experimental noise. A hypothesis for such a similarity is that the presence and nature of the alkali in the solution do not affect the internal network reorganization taking place under Si-saturated conditions. 11 However, Ca tends to cause a shift of δ Si to a more negative chemical shift. 22 The effect of Ca concentration variation from one gel to another may counterbalance any variations in polymerization. We then cannot rule out the hypothesis that some gels are more polymerized than others.
The two samples altered for 1 month and for 1 year respectively in a K-solution were also analyzed to assess whether the structure continues to reorganize after the first month of alteration. Both CPMG and CP-MAS signals are identical (Fig. 3b). As their chemical compositions are relatively similar (Table 1), it can be deduced that while network reorganization still occurs, the degree of order and polymerization of the gel structure is not significantly modified over time. 27 Al MAS NMR spectra acquired before and after heat treatment (300°C) are depicted in Fig. 3c,d. As observed in previous studies, 11,23 27 Al MAS NMR peaks in gels are sharper before heat treatment, which is typical of a hydrated local environment of [AlO 4 ] − units that dehydrates at high temperature. After heat treatment at 300°C, the signals are devoid of most surrounding protons leading to a less symmetric environment. As a consequence, the local electric field gradient increases, yielding a broader spectrum. Variations in the spectra widths (Fig. 3d) are  24,25 For a glass altered in a no alkali-solution, it was demonstrated that the charge compensation is mostly ensured by Ca remaining in the structure. 11 Li-and Na-gels have a signal that is similar to that of the no alkali-gel. By contrast, the signals obtained on K-and Cs-gels are narrower. To help identify the Al charge compensator, several pristine ISG-based glasses with Li, K, and Cs replacing Na were prepared, and 27 Al MAS NMR signals were acquired for each of these glasses and compared to the gel signals (supplementary material part 3 for more information). The signal of the Li-gel is similar to that of pristine ISG-Li glass (supplementary material Fig. S3 b). However, the quantity of Li incorporated in the gel structure is very low and inferior to that of K (from the pH electrode contamination). Considering that point, and noticing that the 27 11 The free volumes generated by the dissolution of mobile B, Na, and Ca, and Ca/alkali exchange were then calculated considering pristine glass and gel compositions, following the method described by Collin, et al. 11 Although the uncertainties of this method are significant, a trend seems to emerge: slightly higher free volume values are found for no alkaliand Li-gels, while the value decreases gradually for Na-, K-, and Csgels (Table 2). This is consistent with the fact that Ca tends to be exchanged with cations with larger ionic radius (Na, K and Cs).
Water speciation analysis 1 H NMR signals were acquired for every 3.5-5.5 μm gel powders. Spectra are normalized to the same sample weight so that their areas are reflective of the proton content. The no alkali-, Li-, and Na-gels appear to be more hydrated than the K-and Cs-gels (Fig.  4a)  hydrated (Fig. 4b). All the signals display similar shapes (refer to supplementary material Fig. S5 a), suggesting a similar proton speciation population in all samples. TGA was used to quantify the total water content inside gels. This analytical technique was applied to the 3.5-5.5 µm gel powders; the resulting data are displayed in Fig. 4c. These data are strongly correlated with those obtained by 1 H MAS NMR (Fig. 4d), despite uncertainties at low water content leading to a calibration curve that does not run through the origin (possibly due to sample rehydration, see supplementary material Fig. S5 b). Such a curve will therefore only be used for a sample (K-gel altered 1 year) that is situated in the central zone of the graph, where the correlation is the best.
Using the method detailed by Collin, et al., 11 the population of water species was determined for every gel from 1 H MAS NMR spectra. Each proton species has a distinct contribution to the 1 H NMR signal. The positions of these contributions were found to be globally identical in each gel from Hahn Echo (HE) experiments (supplementary material Fig. S5 c to m). Molecular water (H 2 O mol ) contribution is centered at~4.7 ppm, while hydroxyl groups forming H bonds (-X-OH HB ) contribution is centered at~7.1 ppm. Hydroxyl species that do not form H bonds ("free" -X-OH) contribution is centered at 1.8 ppm. A deconvolution of the 1 H NMR signal for each altered sample was performed considering these contributions (supplementary material Fig. S5 d to n), providing the H atoms repartitions as presented in Table 2. As presumed from the shape of the 1 H NMR signals, a similar H repartition is found for every gel.
Combining the H repartition with TGA data and chemical analysis, the volumes occupied by the water species (H 2 O mol only and H 2 O mol + "free" -X-OH + -X-OH HB ) can be calculated (Table  2), as well as the O repartition. 11 Oxygens are found inside the altered sample as bridging-oxygen O BO and non-bridging-oxygen (hydroxyl species) O NBO , as well as water-oxygen O H2O . O repartition is dependent on total water quantity: the less hydrated samples therefore display a higher percentage of O BO (Table 2) and are more polymerized.

DISCUSSION
The effect of alkalis on ISG glass alteration under silica-saturated and neutral pH conditions was assessed in this study. The glass alteration rates, calculated between 25 and 80 days, r 45-80 , are similar for glass powders altered in no alkali-solution, Li-solution and Na-solution. They are however divided by 6 when K or Cs are introduced in the solution (supplementary material Table S6 a for more data). The absence of Li effect on glass alteration is confirmed by experiments using a large range of Li concentrations. By contrast, K influences the glass alteration gradually from low to high solution ionic strength, while Cs displays a striking effect regardless of its concentration in the solution (r 45-80 is reduced by a factor of 5.5 between 0 and 3 mmol L −1 of Cs, and only a factor of 2 between 3 and 70 mmol L −1 of Cs in solutions).
Both solution and solid analyses confirm that B and Na releases during glass alteration in all tested solutions are almost complete. However, the retention of Ca in the gel is highly dependent on the nature of the alkali added to the solution. A higher selectivity of the gel is observed for Na, K, and Cs rather than Li following the order

K > Cs > Na ) Li
Such an order is surprising at first, considering that previous works have shown a higher mobility of Li than the other cations, with an unexpected ability of Li to diffuse inside the pristine glass in aqueous 9 and non-aqueous media. 26,27 However, while the capacity of aqueous Li to diffuse in glass was observed, it was not quantified. The order observed here is close to the order of the Hofmeister series (also called lyotropic series). 28 This series, used Cs-gel Water speciation: proton repartition in every sample was obtained from 1 H NMR spectrum deconvolution. Oxygen repartition was determined for every sample from H repartition, TGA and chemical composition. 11 Gel porosity parameters: the specific surface area and the volume of the porosity are calculated by combining TGA and NMR data Impact of alkali M Collin et al.
first as a classification of the ability of ions to increase or decrease proteins solubility, has also been observed in clay selectivity for cations. 29 Cs and K are reversed in the lyotropic series, compared to what is observed here. However, several uncertainties (mainly analysis error and K contamination in the Cs-solution leading to a Cs-K competition inside the gel structure) exist such that we cannot exclude the K ≈ Cs > Na >> Li order.
Alkalis act as charge compensators for clays, similar to what they do inside the gel structure, justifying a comparison between both materials to propose a hypothesis explaining the same observed selectivities. Many studies conducted on clay selectivity proposed a hypothesis to explain why certain ions were more retained than others. The hydrated radius is often cited as a reason for selectivity on a clay surface 29 and on amorphous negatively charged silica surfaces. 30 However, this hypothesis is often rebutted when considering cations entering inside a clay structure, as it is in contradiction with many observed selectivities for various clay materials. 29 Another model, based on the Eisenman model for ion-specific glass electrodes, 31 explains cation selectivity by a competition between two electrostatic forces: the force of attraction between the cation and its hydration shell, and the force of attraction between the cation and the material. 29,32 According to this model, ions that are weakly hydrated such as K and Cs (see Table 3) have an attraction force to their hydration shell that is more likely to be lower than the attraction force between the cation and the material. This model is coherent with the Hofmeister series, and is also a likely hypothesis for our gel structure selectivity for cations. The more strongly hydrated cation (Li) is nearly not exchanged with Ca, while Na (a mildly hydrated cation) exchanges more. K and Cs, the less energetically hydrated cations, are almost totally exchanged.
Al changing of charge compensator from one gel to another (Table 1) can be seen as a consequence of the gel structure selectivity for some cations. When Ca is still present inside a gel, it  Beyond the local effect of alkalis supplied by the solution on the charge compensation of negatively charged units within the gel, the gel structure selectivity with respect to alkali also appears to have consequences on water quantity found inside the gel. In order to explain this trend, the free volume that is supposed to be found inside the gel after mobile species leaching and Ca/alkali exchange are compared to the volume occupied by water species (H 2 O mol + "free" -X-OH + -X-OH HB ). Indeed, as Ca is replaced by Na, K, or Cs, cations whose ionic radii are larger than that of Ca (Table 3), smaller free volumes are expected, which would be consistent with the decrease in water quantity observed above. However, the volume occupied by all water species (H 2 O mol + "free" -X-OH + -X-OH HB ) is greater than the free volume (Fig.  5a). This is coherent with the fact that hydroxyl species ("free" -X-OH + -X-OH HB ) can be considered as part of the network and not of the porosity. Yet, no clear correlation is observed between the free volume and the volume occupied by water molecules alone (H 2 O mol ) (Fig. 5a). This would indicate that the calculation of the free volume from steric considerations is too simple and should consider the network re-organization following the release of B as it was shown in Collin, et al. 11 However, we can see that the best correlations are obtained for the Li-and No alkali-sample, i.e. samples with low Ca/alkali exchange. Cations such as K and Cs could therefore have an unforeseen effect on the water quantity entering the gel layer. Our data do not allow us to conclude on this hypothesis, but molecular dynamic simulation of alkali effect on water quantity inside nanoporosity could give more information.
The water quantities found inside the gel are well correlated to the degree of glass alteration measured from boron release (Fig.  5b). This observation is also supported by ToF-SIMS profiles: those obtained for Cs-sample (Fig. 2d) and in K-sample (previously published by Gin, et al. 10 ) display a retention of B in the inner part of the gel, which is characteristic of a strong passivation effect, 10 while no retention is observed for the Na-sample (Fig. 2c) and No alkali-sample (previously published by Collin, et al. 11 ).
Another global effect is observed due to this decrease of water quantity: the percentage of NBOs-a good proxy of gel reorganization-is significantly diminished in K-and Cs-gels relative to the other gels (Table 2). In general, a high degree of reorganization favors pore closure leading to a lower glass dissolution rate. [33][34][35] To evaluate this effect of reorganization on gels properties, we have measured their dissolution rate (r 0 ). Tests were conducted in dilute conditions, i.e., far from equilibrium, to avoid any effect of dissolved gel species on the Gibbs free energy of reaction (supplementary material Table S6 d). Despite the fact that r 0 does not appear to be very sensitive to variation of the gels composition and structure, and that the uncertainty is approximately 25%, 36 a lower r 0 is measured for the Cs-rich gel than for the other gels. This indicates that Cs improves both the passivation properties and the resistance of the gel towards hydrolysis.
The gel has a dynamic structure, and reorganization dramatically affects its passivation properties. 11,35 This raises the question on the effect of aging of the gel structure. This question was studied here by comparing a sample altered in a K-solution for 1 month to one altered in similar conditions for 1 year. A slight enrichment of K is observed, indicating that the Ca-K exchange process is slightly slower than B and Na release, as it continues after the release of the other mobile species. Overall, the two gels are similar, with identical degree of polymerization. The potential evolution of the material formed during the first month of alteration appears to be slow enough to be undetectable. The main mechanisms leading to the differences in the alteration kinetics appear to happen during the first month depending on the alkali in the solution, meaning that the evolution over time of each gel might differ. It is thus difficult to extend the conclusion stated above to the other alkali-samples. This could be further investigated in future studies.
To conclude, this study demonstrated the ability of certain alkalis to decrease glass alteration kinetics when present in solutions. It is demonstrated that K and Cs can exchange with Ca and compensate negatively charged Al units in a gel. As a consequence, K-and Cs-gels are less hydrated than the others, and thus more passivating. One must however keep in mind that Ca mobility is highly pH-dependent. 21 Such exchange might be different at more alkaline pH.
It is also found that the Cs-rich gel is more resistant to hydrolysis than the others. These findings could be used in the future to improve glass formulation to come up with intrinsically more durable materials. However, replacing small alkali by bigger one often results in a higher viscosity of the molten glass, which is an issue for processing nuclear glass. Another idea could be the design of engineered barriers containing alkali salts (such as CsCl) surrounding glass canisters in future geological disposal facilities to optimize geochemical conditions.

Leaching experiments
All experiments were performed on an ISG batch prepared by MoSci Corporation (Rolla, MO, USA). 10 The glass was crushed into powder, which was sieved before washing in acetone and pure ethanol to remove fine particles. Leaching experiments were conducted on the fraction with particles of 20-40 µm in size for the alteration kinetic study, and on the 3.5-5.5 µm size fraction to obtain a totally altered glass powder for solid characterization and initial rate measurements. Refer to Collin, et al. 11 for more information on powder preparation.
All experiments were conducted in perfluoroalkoxy reactors using 18.2 MΩ cm deionized water initially saturated with respect to amorphous silica (C 0 (Si) = 143 ± 12 mg L −1 at pH 90°C 7). SiO 2 was introduced under agitation at 90°C until total dissolution occurred. Various alkali chloride salts (LiCl, NaCl, KCl, and CsCl) were then added to the solution with different concentration values.
The pH was maintained at 7.0 ± 0.5 during the experiment by adding small quantities of a 0.5 N ultrapure nitric acid solution. The solution was regularly sampled over time. The samples were filtered (0.45 µm cutoff) and analyzed by spectrophotometry (Cary® 50 Scan UV-Vis spectrophotometer for B and Si concentrations, with methods similar to DIN 38405-17 and ASTM D959-10, respectively) and inductively coupled plasma atomic emission spectroscopy (ICP-AES; Thermo Scientific iCAP™ 6000 Series).
Alteration kinetic study. Specimens of 200 mg of the 20-40 µm glass powder and two monoliths measuring 0.5 × 0.5 × 0.2 cm, polished to 1 µm on the two largest sides, were introduced in 0.5 L of silica-saturated solution at 90°C (S/V = 34 m −1 with S geo powder = 0.08 m 2 g −1 ). Alkali salts were then introduced in the solution with the following concentrations: • a reference without alkali (except K due to electrode contamination <1 mmol L −1 ). Note that the 20 mmol L −1 K experiment was reproduced three times, and the results were found to be consistent (supplementary information Fig. S1 j), validating the repeatability of the procedure. The first monoliths were withdrawn after 25 days and the second after 134 days. Various parameters were derived from the solution analysis data, such as the altered glass percentage and equivalent thickness (refer to Collin, et al. 11 for more information).
Preparation of fully altered glass samples. Samples of~500 mg of 3.5-5.5 µm glass powder were introduced in 1 L of silica-saturated solution at 90°C (S/V = 274 m −1 with S geo = 0.55 m 2 g −1 ) with alkali concentration of 20 mmol L −1 . The solution was regularly checked over time until total B release from the glass had occurred in each case (approximatively 1 month of alteration). Note that an additional experiment was performed with a solution containing K until 1-year duration. The solutions were then filtered using a Büchner funnel, and the resulting altered glass powders called gel powders were dried for 24 h at 50°C and stored in a desiccator.
Initial dissolution rate measurement. The initial dissolution rates of both pristine glass and gels were measured by static alteration measurements at pH 90°C 7 in deionized water. In contrast to the previous experiments, no silica was added in the onset solution. Duplicates were prepared on 3.5-5.5 µm pristine glass and gel powders (S/V = 0.18 cm −1 ) to ensure repeatability of the experiment. The Si concentration was monitored over time by spectrophotometry and ICP-AES. For more extensive details on the procedure, please refer to Fournier, et al. 36 Solid analysis Solid analyses were performed on the 3.5-5.5 µm gel powders and on monoliths altered in a 20 mmol L −1 alkali solution.
Chemical analysis. A small fraction of each altered 3.5-5.5 µm glass powder was dissolved in an acid solution (10 HCl + 5 HNO 3 + 4 HF). The resulting solutions were analyzed by ICP-AES (Thermo Scientific iCAP™ 6000 Series) to quantify Si, B, Al, Ca, Zr, Li, Na, K and Cs. Two analyses were conducted for repeatability. O in the gels was considered using the  (2) where i = Si, B, Na, Al, Ca, and Zr, O i pristine glass being therefore the number of O atoms associated with each element in the pristine glass.
Density calculation. The densities of both the pristine glass and the gels were assessed using a model developed by Fluegel 37 for calculating the room-temperature glass density using its composition: this model does not consider water within the sample. Another model, developed by Iacovino, 38 which considers water, was also used for the gel density calculation, but this could not be applied to the pristine glass as the model does not consider B.
Depth profiling analysis. The monoliths withdrawn from Na (after 25 days) and Cs (after 134 days) experiments were analyzed using time-of-flight secondary ion mass spectrometry (ToF-SIMS: IONTOF TOF 5). An O ion beam was set at 2 keV and 510 nA and used for surface abrasion (200 × 200 µm). Actual analysis was provided by a Bi ion sputtering beam (25 keV, 2 pA, 30 × 30 µm) followed by ToF investigation of secondary ions extracted from the monolith. The surface was neutralized during analysis by a low-energy electron flux (<20 eV). Measuring the depth of the final crater allowed an elemental profile depth calibration. Si, Al, B, Na, and Ca profiles were then normalized to that of Zr to avoid matrix effects. Zr was chosen as it was not leached in the solution during the experiment. 33 The resulting data were normalized to the mean value measured in pristine glass to extract quantitative information (Eq. (3)).
with i = Si, Al, B, Na, and Ca. The H profile was not normalized, and was thus only used to evaluate the position of the alteration front.
White-light vertical scanning interferometry (VSI). In order to assess the volume change during glass alteration in the Si-saturated solution, the surfaces of glass monoliths were partially coated using room-temperature vulcanization (RTV) silicone. 39 The RTV silicone held at pH 90°C 7 was checked prior to the experiment. The monolith was then altered for 15 days in conditions similar to those described above. The silicon RTV coating was removed and the sample surface was analyzed by VSI (FOGALE Nanotech, magnification 10). The difference between the mean height values of the coated and uncoated surfaces was measured. The uncertainty was determined from the height variations of both sections. 39 Thermogravimetric analysis. TGA analyses were performed on a Seratum TG-DSC 92-16.18, under a flow of Ar in an Al 2 O 3 crucible. Samples of 3.5-5.5 µm pristine glass and gel glass powders were analyzed (~70 mg for each sample). For the pristine glass, a typical heating ramp of 10°C min −1 was used between room temperature and 1200°C. For the altered samples, an increased rate (30°C min −1 ) was used to attain specific temperatures of 90, 150, 300, and 450°C, with 2 h plateaus at each temperature. A final heating ramp of 10°C min −1 was then performed to reach 1200°C.
NMR spectroscopy. 1 H and 29 Si magic-angle spinning (MAS) nuclear magnetic resonance (NMR) analyses were performed on a Bruker Avance WB 300 MHz spectrometer (magnetic field 7.05 T). A Bruker 4 mm (external diameter of ZrO 2 rotor) cross-polarization magic-angle spinning (CP-MAS) probe was used at a sample spinning frequency of 10 kHz. The sample mass was 90 mg in average. 29 Si/ 1 H CP-MAS NMR spectra were collected using contact times ranging from 2-8 ms. 29 Si MAS NMR spectra were acquired, using the Carr-Purcell-Meiboom-Gill (CPMG) pulse sequence 40 with a recycle delay ranging from 2 s for altered samples, which were found to relax faster owing to protons, and 20 s for pristine glasses. 1 H MAS NMR spectra were obtained by direct acquisition (DA) and using a Hahn Echo (HE) pulse sequence with varying rotor-synchronized echo delays ranging from 100 to 80000 µs. 1 H NMR spectra were deconvolved using the Dmfit program, 41 considering a mixed Gaussian/Lorentzian model. Positions of each contribution were determined using the HE data and the width and amplitude were left free. 11 B and 27 Al MAS NMR spectra were collected on a Bruker Avance II 500WB spectrometer (magnetic field 11.72 T), using a Bruker 4 mm CP-MAS probe at a spinning frequency of 12500 Hz. 11 B MAS NMR data were acquired for the pristine ISG, as no signal was detected from the altered sample. Impact of alkali M Collin et al.
Si, 27 Al, and 1 H were acquired before and after heat treatments at 90, 150, 300, and 450°C.

Data availability
The data that support the findings of this study are available from the corresponding author upon reasonable request.