In the lack of natural borosilicate glasses, well-dated medieval stained glasses provide a good proxy for understanding the processes and kinetics of glass alteration (Macquet and Thomassin, 1992; Sterpenich, 1998; Sterpenich and Libourel, 2001, 2006). Manufactured from natural materials (e.g. washed siliceous sand, limestone, plant ashes, etc.) and colored by transition elements, stained glasses offer a large variety of compositions (i.e., from Na- and Si-rich to K-rich and Si-poor glasses). It is possible to assess how chemical composition controls glass durability and the behavior of minor/trace elements, under natural alteration conditions and over 10²–10³ yrs. Depending on their exposure (Fig. 3), stained glasses can also provide a unique way to characterize the effects of weathering conditions on glass alteration (Libourel et al., 1994): while samples from archaeological sites afford information on alteration by groundwater, stained glass windows are subject to meteoric weathering or to alteration driven by moisture, for outside or inside surfaces, respectively.

Whatever the kind of alteration, stained glasses present traces of corrosion on their surface, including a weathered glass layer (gel) and precipitates. Optical or scanning electron microprobe observations reveal that the structure and thickness of the altered zone depend on the type of weathering (Sterpenich, 1998; Sterpenich and Libourel, 2001). For a fixed glass composition, alteration driven by moisture produces only scattered dissolution pits, whereas atmospheric weathering produces a continuous altered and micro-fractured layer up to 200 μm thick (Fig. 3a). Buried in soils, stained glasses show a more pervasive weathering with alteration of up to 1500–2000 μm (Fig. 3b) and a characteristic laminated structure. Dissolution rates for given weathering conditions depend on glass composition. For instance, under burial conditions in the archaeological site of Rouen (Normandy, France, 9th century), the mean thickness of the altered zone for soda-rich glasses is around 30 μm, while for K-rich samples it may exceed 1000 μm (Fig. 3b and d). Similarly, potassic glass windows are always darkened by alteration layers while Na-rich glasses on the same windows preserve their color and have alteration zones less than 2 μm thick, as in the blue stained glass of the Chartres cathedral (France, 13th century).

Due to the good mechanical cohesion between weathered and unweathered glass, mean corrosion rates can be estimated by dividing the mean thickness of the alteration crust by the duration of weathering. Mean corrosion rates for stained glasses in natural conditions over the last 1100 yrs, are ranged approximately 0.005–1.5 μm.yr⁻¹ and are dependent on glass composition and on alteration conditions. Mean corrosion rates being correlated with the bulk degree of polymerization of the glass, the NBO/T parameter (NBO/T denotes bulk melt non-bridging oxygen per tetrahedral cation, Mysen, 1988) allows to evaluate the specific role of glass composition and alteration conditions (Fig. 3c). For K-rich depolymerized (silica-poor) glasses, mean corrosion rates are almost two orders of magnitude greater than for polymerized Na-rich glasses. However, for the same composition, glass dissolution rates on the outside of windows are only 6–7 times slower than for glasses buried in soils. This suggests that variations in glass composition lead to variations in glass dissolution rates an order of magnitude more pronounced than those caused by different weathering conditions. Mean elemental dissolution rates can be also estimated for weathering durations over more than 10³ yrs (Sterpenich and Libourel, 2001). Alkalis (K, Na, Rb and Cs) and alkaline-earth elements (Mg, Ca, Sr and Ba) are the most leachable elements, with leaching rates ranging from 10⁻² to 10⁻³ g.m⁻².day⁻¹, while network forming elements (Al, Fe, Zr, Ti) are less depleted ≈10⁻⁴ g.m⁻².day⁻¹, Si being intermediate. Lanthanides and actinides, with the noticeable exception of U and Eu, are almost insensitive to alteration with the lowest dissolution rates at around 10⁻⁵ g.m⁻².day⁻¹.

The alteration of medieval stained glasses is thus strongly dependent on glass composition and on the weathering conditions with highly variable elemental dissolution rates. While alteration zones may play the role of a barrier limiting the release of several elements, stained glass studies reveal that a special care has to be taken in nuclear waste management strategy for the glass formulation and the weathering condition and their modifications in the repository during very long timescales (> 10⁵ years).

Nuclear glass packages are fractured due to a rapid cooling of the melt in the stainless steel canister (Fig. 4; Laude et al., 1982; Minet and Godon, 2003). Due to the release of strain upon glass cooling, fractures increase the reactive surface area and influence the alteration kinetics. Because of differences in the glass surface/water volume ratio or in the transport processes between the external border and the inside part of the cracks (Fig. 4c and d), surfaces can evolve with different alteration kinetics. In the following, it is shown how fractured archeological glasses can be used to assess the long-term alteration kinetics in the cracks and address the impact of fractures on the glass alteration.

Archaeological glass blocks (0.5–10 kg) were recovered from a shipwreck in the Mediterranean Sea near the island of Embiez (Var, France; Fig. 4a). They have been altered for 1800 yrs in seawater at a depth of 56 m and a fixed temperature of 15 °C (Fontaine and Foy, 2007; Foy et al., 2005). The glass was produced in the eastern Mediterranean region. The materials were melted in a large furnace and cooled in air, which caused the fracturing. Glass slabs were cut into blocks to be shipped to Europe, where they would have been manufactured (Fontaine and Foy, 2007; Foy et al., 2005). The composition is typically that of a Roman soda-lime glass (∼ 70 wt% SiO₂, 20 wt% Na₂O, 5 wt% CaO, and 2 wt% Al₂O₃).

The altered glass (Verney-Carron et al., 2008) reveals, in addition to a small pristine core, an external zone, in contact with seawater, where the cracks have a large alteration thickness of 400 − 500 μm (Fig. 4c), and an internal zone, where the cracks become denser but are less altered, 5–30 μm thick (Fig. 4d). All cracks are filled with alteration products that were characterized by electron microprobe and μ-X-ray diffraction. The cracks in the external zone contain Mg- and Fe-rich smectites, whereas internal cracks have a large (relative to the total thickness) hydrated glass layer at the interface with the pristine glass and only a few μm thick of Mg-rich smectites in the center of the crack (Fig. 4). This shows that the kinetics of the alteration mechanisms (ion exchange and dissolution) differs according to crack location. These differences are well explained by a coupling between alteration kinetics and element transport in solution (Verney-Carron et al., 2010a). The continuous renewal of the seawater in contact with external cracks maintained an “initial rate” for 1800 yrs and leads to thick neoformed smectite layers. In the internal cracks, the transport is assumed to be diffusive and the slow renewal of the solution favored saturation effects and low alteration rates. Moreover, crack sealing by alteration products slowed down the diffusive transport (Verney-Carron et al., 2008, 2010a).

In the present French models of R7T7 alteration rates, used in safety calculations, the increase of the reactive surface area is taken into account via two fracturing ratios (τ): τ₀ for the external cracks and τ_r for the internal cracks that are assumed to be altered at the “residual rate” (Gin et al., 2005). The study on Embiez archaeological validates that this hypothesis remains conservative (Verney-Carron et al., 2010b).

To gain insight in the long-term behavior of nuclear waste packages at the glass–metal–clay interfaces in their geological repository conditions, we show in the following how slags from ancient iron making sites or primitive meteorites can be used as valuable proxies (Fig. 5). The stainless steel canister of nuclear waste is expected to corrode in anoxic conditions. Both analogues provide unique information about long-term corrosion under anoxic conditions of metal iron, which is rare on Earth.

Ancient slags: The question of the interaction between glass and iron is particularly important because nuclear glasses are poured in a stainless steel canister and the resulting package is expected to be surrounded by a thick carbon steel overpack (Andra, 2005). Some laboratory experiments under anoxic conditions have shown that the presence of iron and, more generally, iron corrosion products sustain glass alteration at a rate higher than the glass would undergo in the presence of site water only. This enhancement of glass alteration is due to the strong affinity of iron rich phases with respect to silicon, the main component of the glass (De Combarieu et al., 2011; Jollivet et al., 2000). It has been asserted that the formation of Fe-Si containing phase delay the formation of the protective gel layer responsible for the drop of alteration rate observed in glasses altered by water under static or low flow rate conditions (De Combarieu et al., 2011). That is why understanding and modeling interactions between glass, iron and clay is a key preliminary step in order to predict long-term behavior of nuclear glass packages. Consequently, studies of natural or archaeological analogues give precious observational insight for this issue. For example, the site of Glinet dated from the 16^th century (Neff et al., 2010), has been taken as a reference site for studying glass/iron interactions. Indeed, the by-product slags produced by the blast furnace present partly glassy matrix and contain metallic particles of micrometric to millimetric sizes. The blocks present numerous cracks that were immediately saturated with water as they were buried at depths continuously under the water level. It has been verified by on-field measurements and laboratory tests that at the sample location, water was carbonated and anoxic. Moreover, the crack network connects the metallic zones of the slags to the external solution of the soil and is entirely filled by corrosion products of iron such as those found in desaerated media (Fig. 5a). Microstructural characterization of the network embedded with the corrosion products by μRaman spectroscopy, reveal the presence of siderite (FeCO₃ – Fig. 5b). This siderite contains several % of calcium partially substituting the Fe atoms. This substitution was also observed on iron artefacts corroded in the same medium and could come from the groundwater but also, in the present case, from the glass matrix, that contain this element. Glasses identified in Glinet are calcium silicate glass enriched in iron and aluminum (MgO: 0.7–2.2 wt%, Al₂O₃: 5.1–8.8 wt%, SiO₂: 61.9–76.5 wt%, K₂O: 1.8–2.3 wt%, CaO: 15.6–25.1 wt%, MnO: 0.2–2.9 wt%, FeO: 1.2–9.1 wt%). This range of compositions for the ternary CaO-A₂O₃-SiO₂ defines an area known for its immiscibility. SEM-FEG observations confirm a non-homogeneous nanostructure characterized by the presence of spheroids of several hundreds of nanometers distributed in a glass matrix. This specific structure of the glass implies different alteration kinetics for each of the phases observed by TEM (Fig. 5c). The spheroids, mainly constituted of SiO₂, are less altered than the embedding matrix containing higher concentration the other elements.

At the interface between glass and iron corrosion products, different layers have been observed, linked to the variation of composition (Fig. 5c and d). EDS analyses coupled to STEM indicate that the altered glass is depleted in Ca but not in Al. Interestingly, an external festooned layer between altered glass and siderite, made of amorphous iron silicates is observed (Fig. 5c). The study of this layer can give insight into the influence of iron on the glass corrosion. Moreover, iron coming from the metal corrosion products has penetrated into the altered layers of the glass. Laboratory experiments simulating a crack have been conducted for 120 days at 50 °C on synthetised glass samples with the structure and the chemical composition of the slag glass matrix. The chemical profiles of the glass alteration layers are very comparable to the archaeological ones and are currently under study. The results are expected to help discriminating between the controlling parameters of glass alteration: transport vs chemical equilibria.

Meteorites: CR chondrites collected immediately after their fall, are primitive meteorites that have experienced aqueous alteration of metal and glass in their parent bodies. Constituted of mixtures of ferro-magnesian minerals and alumino-silicate glasses in chondrules (Fig. 5e), iron-nickel metal and sulfides blebs seated in phyllosilicate-rich matrices, CR chondrites show a range of low-temperature aqueous alteration stages, from pristine, unaltered (type 3) to completely hydrated (type 1) material (Weisberg et al., 1993). They provide an analogue for alteration and corrosion processes at the glass–metal–clay interfaces over a longer time frame than is possible in laboratory. In addition, they allow the investigation of all steps in the expected corrosion of nuclear waste, given that the physico-chemical environment of the parent body during the alteration (e.g., Zolensky et al., 1993), i.e. temperature (50–150 °C), water/rock ratio (0.4 to 1.1), low oxygen fugacity and mineralogy, is similar to that expected in a disposal facility (Andra, 2005).

Fig. 5e depicts the corrosion products after partial alteration at the metal/glass interface in a chondrule from the Al Rais CR2 chondrite. This chondrule contains several metal grains at different alteration stages. SEM images show that Fe-Ni metal grains are surrounded by several alteration layers, with intrusions of neoformed phases into metal grains (Fig. 5e and 5f). These corrosion products have a complex mineralogy, with magnetite-rich layers alternating with layers enriched in sulfides (Fig. 5f). A genetic sequence for metal/glassy mesostasis, confirmed by elemental mapping, would be metal/magnetite ± sulfide/phosphate? ± carbonate? ± sulphate?/(altered) mesostasis. The alteration layers shown in Fig. 5e and f are enriched relative to the metal in S, Si, Mg, Ca and Mn. They also have significant contents in Ni, Cr, Co, P, which are enriched in some parts compared to the metal. In the case of metal/matrix interfaces, the sequence of corrosion products is: metal/magnetite ± sulfide/magnetite/phosphate + matrix. The alteration rims become more complex with increasing alteration, with steel components moving into the surrounding environment.

In addition to document corrosion processes at metal/glass/clay interfaces, such extraterrestrial analogues provide information on the elemental mobility during alteration in anoxic environments, even if the time-frame of parent body alteration is not well constrained (10⁴–10⁶ yrs or more). Chemical distribution maps show that most of the “dissolved” elements, including iron, does not go very far. Even for the most altered metal grains, the phases precipitate at distances of several 100 μm. This happens in the form of sulfides, phosphates, sometimes carbonates, and Fe-oxides, which tend to form characteristic rings in the near vicinity of the corroded grains. At the metal/glass interface, element migration is even smaller even less, in the order of tens of μm.

This methodology was applied to the Embiez glass blocks, altered for 1800 yrs in seawater at a constant temperature of 15 °C (Verney-Carron et al., 2008, 2010a). The stability of these alteration conditions allows long-term modeling. Verney-Carron et al. (2010a, b) demonstrated that a model parameterized in the laboratory is able to simulate the alteration of fractured historical glass blocks. Moreover, the coupling of alteration chemistry and transport in solution explains that external cracks in direct contact with renewed seawater were altered at a high dissolution rate (400 − 500 μm), whereas internal cracks are less altered (by 1 or 2 orders of magnitude) because the low renewal of interstitial water and the sealing of the cracks by alteration products favored low alteration kinetics.

A key issue to make a reliable and robust prediction of the long-term behavior of nuclear glasses is to determine realistic values of residual rate. Basaltic glasses, with a silica content close to nuclear glasses, altered in different environments (seawater, groundwater, meteoric water) can help in this direction, with the advantage of long timescales but large uncertainties on the alteration history. This ‘residual rate’ kinetic regime is expected to be the longest one in geological disposal conditions and explains most of glass alteration (Techer et al., 2001). The mechanisms involved at this stage of reaction progress are water diffusion through the passivating gel layer and precipitation of secondary crystalline phases that consume elements from the gel (Frugier et al., 2008). It has been shown that the residual rate depends on glass and water composition, the latter being influenced by chemical reactions in the near-field. Even if residual rates are accurately measured in laboratory, natural analogues may help to determine how this rate evolves as a function of water composition, renewal rate and time. Studying basalt alteration therefore provides a new challenge for geochemists.