Corrosion Science 72 (2013) 10–19 Contents lists available at SciVerse ScienceDirect Corrosion Science journal homepage: Characterisation of complex alteration layers in medieval glasses Tiziana Lombardo a,⇑, Lucile Gentaz a,1, Aurélie Verney-Carron a, Anne Chabas a, Claudine Loisel b, Delphine Neff c, Eric Leroy d a LISA, Laboratoire Interuniversitaire des Systèmes Atmosphériques, UMR CNRS 7583, Université Paris-Est Créteil, Université Paris Diderot, 61, avenue du Général de Gaulle, 94010 Créteil Cedex, France b LRMH, Laboratoire de Recherche des Monuments Historiques, 29, rue de Paris, 77420 Champs sur Marne, France c SIS2M/LAPA, Laboratoire d’Archéométrie et Prévision de l’Altération, UMR 3299 CEA-CNRS, CEA-Saclay, 91191 Gif-sur-Yvette Cedex, France d ICMPE, Institut de Chimie des Matériaux Paris-Est, UMR CNRS 7182, Université Paris Est Créteil, 2-8 rue H. Dunant, 94430 Thiais, France a r t i c l e i n f o Article history: Received 8 November 2012 Accepted 22 February 2013 Available online 7 March 2013 Keywords: A. Glass B. SEM B. STEM B. TEM C. Amorphous structure C. Atmospheric corrosion a b s t r a c t Silica–potash–lime stained-glasses from medieval age in Northern Europe are found in a poor conservation state. Their mechanisms of atmospheric corrosion are still not fully understood and need deeper investigation. A multi-scale analysis of K–Ca rich silicate glasses (ancient and model glass), showed modified layers characterised by a multilayer sequence consisting in the repetition of nm-thick laminae with different compositions. Crystalline phases (sulphates, carbonates and phosphates) are observed. The formation of these sequences is due chiefly to interdiffusion, followed by local dissolution phenomena. Precipitation of secondary phases is enhanced by the presence of fractures, which facilitate fluid circulation. Ó 2013 Elsevier Ltd. All rights reserved. 1. Introduction Whatever the chemical composition or the surrounding environments (atmosphere, soil, sea water, etc.) are, glass materials undergo alteration processes leading to the modification of their structure and chemical composition. Mechanisms of glass alteration in contact with aqueous solution are common to all silicate glasses. Three mechanisms can be distinguished: interdiffusion, dissolution and secondary phase precipitation. Interdiffusion consists in anion exchange between glass alkalis and hydrogenated species in solution (H+, H3O+, and H2O) [1–3]. It leads to a selective leaching of glass elements and results in the formation of a dealkalinised hydrated glass layer at the glass surface. The second process consists in the dissolution of this hydrated glass network by hydrolysis of iono-covalent bonds (Si–O–Si, Si–O–Al) [4]. Local condensation reactions of insoluble species can occur to form a gel [5,6]. Interdiffusion and hydrolysis act simultaneously but with different kinetics that depend upon the glass composition, but also upon the temperature, pH, and ⇑ Corresponding author. Tel.: +33 1 45171677; fax: +33 1 45171564. E-mail addresses: (T. Lombardo), lucile.gentaz@ (L. Gentaz), (A. Verney-Carron), anne.chabas@ (A. Chabas), (C. Loisel), delphine.neff@ (D. Neff), (E. Leroy). 1 Present address: SIS2M/LAPA, Laboratoire d’Archéométrie et Prévision de l’Altération, UMR 3299 CEA-CNRS, CEA-Saclay, 91191 Gif-sur-Yvette Cedex, France. 0010-938X/$ - see front matter Ó 2013 Elsevier Ltd. All rights reserved. composition of the solution and thus upon the rate of solution renewal [7,8]. Finally, the formation of secondary phases results from the precipitation of released glass elements by the first two mechanisms, jointly or not with elements present in the environment. The structure of these phases varies from amorphous to crystalline. The formation and nature of secondary phases depend greatly upon the alteration environment as they form through the reaction between glass elements and exogenous compounds. Indeed, in the case of glass continually in contact with aqueous solution (immersed in sea water or buried in saturated soil) they are the result of either an intense hydrolysis–reprecipitation, in this case they consist principally in phyllosilicates [9–14], or of a selective leaching of alkaline and alkaline-earth elements reacting with exogenous anions such as phosphates, carbonates, and oxides [15–19]. In the case of discontinuous contact with aqueous media (condition which is typical of the atmosphere) secondary crystalline phases arise from the reaction between elements leached from the glass and atmospheric gases. In the latter case, salts, such as sulphates and carbonates, precipitate at the glass/atmosphere interface [15,20,21,22]. Very often at the micrometric scale, a laminated structure is observed in the alteration layers of archaeological glasses immersed in sea water [9,12,23,24] or buried in soil [11,13–19,23,25–33], as well as in basaltic glass [10,34] and in nuclear waste glass [35]. Such laminations are sometimes observed on weathered stained glass [10,36–38]; nevertheless, the authors never refer to T. Lombardo et al. / Corrosion Science 72 (2013) 10–19 them as laminated structure and no precise mechanisms of formation are evoked for atmospheric medium. Based on a multiscale characterisation of glass altered in natural conditions in both short term (model glass) and long term (stained glass windows), the present work intends to investigate the nature of the mechanisms involved in the formation of alteration layers and understand their temporal progression. 2. Materials and methods 11 (France) [39]. Its NBO/T is 1.43 very close to the value of the stained glasses of Rouen and Evreux (spanning between 1.08 and 1.32) [36]. In order to reproduce natural weathering on model glass, several samples were exposed for various durations (up to 4 years) in a test site, in the centre of Paris [40] in both sheltered and unsheltered from rain conditions. Results presented in this paper will focus on a sample exposed 36 months (MG36) to the direct impact of the rain which was submitted to an extensive multi-scale characterisation. The analysis of short term weathering will allow a better understanding of the temporal evolution of the alteration. 2.1. Sample description 2.2. Analytical methods To assess both long and short term weathering of glass, two kinds of samples have been studied. The first one is represented by medieval glasses dating from the 14th century coming from two French historical monuments: the Abbey of Saint-Ouen in Rouen (OU) and the Cathedral of NotreDame of Evreux (Ev). These are eight fragments of uncoloured glass presenting visible evidence of weathering at both indoor and outdoor surfaces. Nevertheless analyses presented in these paper only concerns the outdoor surface. Their chemical composition, quite complex, is characterised by high proportion of potassium (13–17 wt.% K2O) and calcium (12–14 wt.% CaO), explaining the relatively low silicon concentration (53–58 wt.% SiO2), and lower content in magnesium (7 wt.% MgO), phosphorus (4 wt.% P2O5), and sodium (2.5 wt.% Na2O) (for full description see [36]). The analysed glass were removed from the glass window during the 1960s and stored in controlled conditions. The second kind of sample is a Si–K–Ca model glass which was synthesised in the laboratory (at the Stazione Sperimentale del Vetro, Murano, Italy). Its simplified composition (50 wt.% SiO2, 2 wt.% Al2O3, 25 wt.% K2O, 18 wt.% CaO, 3 wt.% MgO, and 2 wt.% P2O5, [40]) was determined from the medieval glass of Saint Urbain of Troyes Samples were analysed by using Scanning and Transmission Electron Microscopy (SEM and TEM) in order to identify the various alteration patterns and the related glass structural modifications. In order to access the chemical and morphological changes occurring at the micrometric scale, sample cross sections were characterised first by SEM (Jeol JSM-6301 F) linked to an EDX detector (Link ISIS 300). Prior to analysis, samples were coated with a carbon layer. It should be noted that SEM analysis of medieval glass were related in a previous work [36] and are here reinterpreted using newly acquired data. To gain a finer understanding of nano-scale alteration processes samples were then studied with a TEM. Ultra-thin sections were previously prepared via Focused Ion Beam (FIB) technique, at the University Aix-Marseille III (CP2M), using a Philips FIB 200 TEM system. Firstly, a thin layer of platinum was deposited on the sample cross section. After the selection of the interested areas, a thin section was then obtained by excavating two trenches (4 lm deep) into the glass block using a 30 kV Ga+ beam operating at 20 nA. Up to this step, an automatic cutting procedure allowed obtaining Fig. 1. SEM images of the cross section of sample OU2 ((a) backscattering mode) and ((b) secondary electron mode), Ev1 ((c) backscattering mode) and MG36 ((d) backscattering mode). In (b) are indicated the location of the FIB thin sections. Sections FIB 1 was excavated in the altered layer, while FIB 2 was shaped at the interface between the pristine and the altered layer. Sections indicated as ‘‘bis’’ are two failed attempts of cutting into the brittle glass layer. 12 T. Lombardo et al. / Corrosion Science 72 (2013) 10–19 a 15 lm (large)  4 lm (high)  5 to 10 lm (thick) section. Sample was further slimmed down to the final thickness (<100 nm) by a manually driven operation, using a 300 pA current beam. Then, the bottom and the edges of the thin section were cut with the ion beam, in this case sample were 65° inclined. Finally, the thin section was transferred at room pressure with a micromanipulator on the membrane of a carbon-coated 200 mesh copper grid. For each studied glass, two thin sections have been realised: the first one inside the altered layer and the second at the limit between the altered and the pristine glass (Fig. 1b). The obtained thin sections were fully characterised by a TEM/ STEM (FEI Tecnai F20 ST) equipped with a Field Emission Gun (FEG) and a Super Twin objective lens, operating at 200 kV with a probe current >0.5 nA, for a 1 mm probe. In conjunction with the TEM image mode (bright and dark fields), TEM is equipped with an EDX Si(Li)EDAX detector (129 eV resolution) and an EELS spectrometer (Gatan GIF2001) allowing, on one hand, analysing elemental composition at a precise location (beam diameter down to 1 nm in STEM mode) and, on the other hand, establishing elemental maps (with resolution varying according to the analytical conditions). At last, TEM allows acquiring electron diffraction patterns (EDiff) of crystalline phases. 3. Results and discussion 3.1. Morphological characterisation of the alteration layers The alteration state of the eight ancient stained glass fragments (3 from Evreux and 5 from Rouen) has been investigated by using SEM. All glasses present an alteration layer whose morphology and thickness, as well as their chemical composition have been described and quantified (detailed results can be found in [36]). Alteration layer (AL) refers here to the chemically and physically modified layer, which is to say that it includes the modified glass as well as the crack network therein. Analyses carried out showed that thickness and the morphology of the AL vary greatly from one sample to another and, in a lesser extent, within the same sample. According to the morphology of the AL observed at the micrometric scale two groups of samples could be distinguished. A major difference between the two types consists in the glass integrity. Indeed in the first group (represented by sample OU2), the original surface is generally preserved, albeit losses due to the presence of localised pits were observed. Upon this surface, the presence of a grisaille (a Pb-based glass used as paint layer during the medieval age) is sometimes observed, as well as a surface deposit (Fig. 1a). To the opposite, in the second group (represented by sample Ev1), the original surface is completely lost as an intense crack network incises the alteration layer, which is thus extremely brittle and fragmented. Therefore, no depositional layer could be found on this latter sample (Fig. 1c). The same kind of analyses has been performed on the Si–K-Ca model glass exposed 36 months (MG36) to the real atmosphere. This sample presents a rather horizontally continuous AL whose thickness varies between 0.6 lm and 25.7 lm (14 lm on average). The layer is extremely fragmented: both horizontal and vertical (parallel and perpendicular to the glass surface) cracks are visible (Fig. 1d). Thus its variations in thickness are due to the significant loss of glass fragments. In order to investigate this heterogeneity at the nano-scale indepth, samples were further investigated by TEM. At this scale, the alteration layers (AL) of all samples (OU2, Ev1 and MG36) show a peculiar alteration pattern characterised by two sequences of discrete bands (Fig. 2). These features are not observed in the pristine glass; therefore they are the results of a profound chemical modification due to the alteration. The first bands correspond to very thin layers, here called laminae (Fig. 2). Their thickness spans from 20 to 50 nm for OU2 and MG36 and about 20 nm for Ev1. The number and frequency of laminae are extremely high in the case of OU2 and MG36 (Fig. 2a and c) while they are much limited for Ev1 (Fig. 2b). The second type of bands, called laminations, consists in clusters of equally thick laminae with the same general orientation, visible as a succession of bright and dark zones, whose contrast differences can correspond either to density variations or porosity heterogeneity. The thickness of the laminations is 0.5– 4 lm for OU2 and MG36, while it is tens of nm up to 1 lm for Ev1. The interface between adjacent laminations can be in discordance (drastic change in orientation) or V-shaped (slight change in orientation). This latter shape results from the merging of two laminations (Fig. 2a). Occasionally, the discordance can correspond to a crack rendering the laminations independent, thus the orientation of laminae within the two contiguous laminations would be the same (Fig. 2c). In conclusion, all the alterations layers were found to be built up by the juxtaposition of discrete alteration units, laminae and lamination. At first, laminae and laminations can be seen as the vestiges of previous alteration front. The extremely different spacing between these sub-units testifies of a non-linear evolution of the alteration. The longer the spacing, the more constant and continuous the phenomenon is. Inversely, a short spacing indicates the repetition of several shorter alteration cycles. In any case, the formation of laminae/lamination is a quite rapid phenomenon, taking place in the span of several months at most (maybe even weeks), as shows the copious presence of laminae in the short term alteration. Besides the separation in sub-layers (laminations and lamina) of the AL, the other common feature is the presence of an intense crack network at both the micrometre and nanometre scales. Assimilating cracks to plans, their thickness is commonly of few tens of nm and attains several hundreds of nm in rare cases. Cracks can be empty or filled with alteration products (see Section 3.3). Fig. 2. TEM images (bright field) and schematic pictures of alteration layers observed on samples OU2 (a), Ev1 (b) and MG36 (c). T. Lombardo et al. / Corrosion Science 72 (2013) 10–19 All samples present two series of cracks: perpendicular (secant) and parallel to the laminae/lamination plans. In the case of secant cracks, since the orientation of the laminae at the two opposite borders is strictly the same, it might be inferred that their formation took place after the formation of the laminations. Indeed if these cracks were pre-existent they would have provided enough fluids to induce the formation of laminae oriented parallel to the crack itself. Thus, this kind of cracks forms as a consequence of the weakening of previously existent AL, probably caused by the humid/drying cycles in the atmosphere. However secant cracks allow the migration of atmospheric fluids in deeper parts of the glass and help to move further the alteration front. This hypothesis is confirmed by the fact that in certain cases secant cracks present in the AL spread into the pristine glass. The second series of cracks are parallel to the laminae/laminations. They can be both concordant and discordant to the laminae and, as previously said, in some cases constitute the limit between two separate laminations. Contrary to the former ones, these cracks seem to be pre/synalteration, as testified by their orientation. Their formation can be also facilitated by the loss of cohesion taking place at a particularly fragile lamina/lamination plan within the AL or when the thickness of a group of laminae becomes too large. It seems indeed, that each crack might facilitate the separation of series of laminae and induce the formation of two distinct laminations. The change in orientation of separated laminations could be caused by textural or chemical heterogeneities witnessed in the original glass (see an example in Fig. 7c). 13 3.2. Chemical composition of the alteration layers The SEM analysis of both stained glasses showed that alteration layers were found to be highly dealkalinised, but the analysis also revealed a high degree of heterogeneity (inter and intra-sample). In particular, sample Ev1 showed a ‘‘classical’’ dealkalinisation profile, that is to say that alkali and alkali-earth concentrations decrease rather continuously moving from the sample surface toward the pristine glass. The opposite behaviour characterises the silicon concentration. Sample OU2 shows instead a more complicated chemical distribution. The dealkalinisation profile is here troubled by the presence of micron-size lumps or veins with anomalous composition compared to the immediate surroundings and with easily distinguishable brighter appearance, in backscattering mode (Fig. 1a). Due to their small size, their chemical composition was difficult to access by SEM, as the analysis would be constantly disturbed by the signal issued by the surrounding glass. Furthermore, in the case of lumps, a bias in the chemical composition could arise, influenced by the signal coming from the rear of these features. Indeed, extrapolating their 2D shape to the third dimension, one can speculate a quite shallow extension and could conclude that their volume would be smaller than the volume of Xray emission zone. Finally, SEM analyses showed that for both glass the boundary between the AL and the pristine glass (PG) is a very sharp limit, which separate two zones displaying a drastic electronic density contrast (Fig. 1a and c). Concerning sample MG36, the AL presents micrometric thick veins with brighter intensity Fig. 3. TEM image (bright field) of sample OU2 showing three laminations and the analysed zone (white rectangle) and elemental distribution maps of the analysed zone (resolution of maps: 1 pixel = 25 nm). Fig. 4. TEM image (bright field) of a lamination in sample OU2 and chemical composition (oxide wt.%) across the line in white. 14 T. Lombardo et al. / Corrosion Science 72 (2013) 10–19 Fig. 5. TEM image (bright field) of a lamination in sample MG36 and elemental distribution maps of the zone within the white rectangle (resolution of the map: 1 pixel = 40 nm). Fig. 6. TEM images (bright field) of part of the AL (a and b) on sample OU2, and the respective diffraction patterns (c). Image b is a blown up of a crack (within the square in a) filled with crystalline phases. (in backscattering mode). These latter are enriched in Ca, and in lesser extent in Mg, while silicon is almost absent. Similarly with the ancient glasses, the border between AL and PG is sharp (Fig. 1d). In order to understand the formation of laminated alteration layers, it is important to determine whether laminae result from differences of chemical composition or from local reorganisation within the alteration layer. A deeper investigation of the chemical composition of the AL of samples OU2 and MG36 has been undertaken by STEM–EDS technique. The analyses confirmed SEM investigation: the AL is highly dealkalinised. Indeed, K, Ca, Na and Mg, are highly depleted (Fig. 3), while Si and P are relatively enriched (Figs. 7 and 9). STEM analyses also showed that the ALs present a chemical variability at the lamination scale (sub-micron scale). For instance, the analysis of two contiguous laminations (Fig. 3) within sample OU2 has revealed that although alkaline and alkaline-earth elements display comparable concentrations, Si and in a lesser extent O, C, S and P contents differ. In this specific case the upper lamination (on the left in Fig. 3), the oldest, is enriched in exogenous elements and depleted in glass constituents such as silicon. Chemical analyses also revealed a chemical heterogeneity of laminae (Figs. 4 and 5) (at the nanometre scale). It has to be noted that due to the size of the electron beam, it was quite difficult to access the chemical composition of one very thin lamina; therefore analyses often corresponded to clusters of several thin laminae. In the case of OU2, contiguous laminae (within the same lamination) display similar proportions of Fe, K, Al, Na, Ca and Mg (Fig. 4). These elements are present in low concentrations mainly because they were extracted from the glass matrix by diffusion, with the exception of Fe and Al for which low concentrations are an original feature. Therefore, for major elements, it is not possible to distinguish significant composition differences between each lamina. Nonetheless, beside this background composition, analyses revealed some specific enrichment in P, up to five times more concentrated that in the pristine glass (several peaks between 0 and 0.2 lm, at 0.53 lm, 0.6–0.7 lm, 0.8–0.85 lm, 0.10–0.11 lm and around 1.55 lm in the line scan). With few exceptions, these P enrichments are accompanied by an increase in Ca (up to four times more concentrated than in the pristine glass) and sometimes in S concentrations (absent from the pristine glass). The large zone (corresponding to several laminae), located between 1 and 1.2 lm, is rich in crystalline phases containing Ca, P, S, and C (not included in the diagram). Moreover, some thick and dark laminae (at 0.65, 0.7, and 0.85 lm in the line scan) are rich in Pb and Mn, their concentrations can reach values as high as 38% while they are almost absent in the original glass. Further analyses indicate that these latter are crystalline phases. Because of the absence of Si in the bands rich in crystalline phases, they might be interpreted as previous cracks filled with precipitates (this aspect will be further developed in Section 3.3). In sample Ev1, similarly to OU2, P and S enrichments were detected in the laminae. Sample MG36 shows different patterns (Fig. 5). In this case, less dense laminae are depleted in Si, O, Al and Mg, as well as in P, while a systematic C enrichment is observed (Fig. 5). C is sometimes associated with T. Lombardo et al. / Corrosion Science 72 (2013) 10–19 15 Fig. 7. Sample OU2, TEM images (bright field) of the AL–PG transitional zone at different magnifications (a and b), elemental distribution maps obtained within the black rectangle (c) (resolution 1 pixel = 10 nm), electron diffraction patterns (d). Fig. 8. TEM image (bright field) of a crystalline zone on sample Ev1 (a) and associated diffractions patterns (b and c). Fig. 9. TEM image (dark field) of the transitional zone between the AL and the PG and chemical composition (oxides wt.%) analysed within the black rectangle for sample Ev1. 16 T. Lombardo et al. / Corrosion Science 72 (2013) 10–19 Ca (left side of the analysed area) and K. These last two elements seem to be concentrated in area around cracks as well as in adjacent laminae on each side of them. Apart from this specific area (Fig. 5), this trend is common to almost all the cracks encountered in this sample. Unfortunately, with the information acquired it is not possible to understand whether these Ca-enrichments are due to the presence of a diffusion barrier constituted by the crack filled with crystalline phases and/or to a backward motion of Ca that had previously diffused into the crack and formed the crystalline phases. In summary, the presence of laminae seems to result from local rearrangement of elements in the glass. Alteration layers are depleted in alkaline elements released by interdiffusion, which triggers the local precipitation of the most insoluble elements (Ca, P) and exogenous elements (C, S) and the modification of their texture (formation of laminae) but not of their structure as laminae are formed of amorphous materials (see Section 3.3). 3.3. Characterisation of crystalline phases The alteration layer is amorphous. However, crystalline phases have been largely detected inside the cracks present in the AL in all glasses. The detection of these phases was helped by the very characteristic sparkling aspect (Fig. 6b) they exhibit when tilting the section under the electron beam. Contrary to the conclusion of [15] stating that the crystalline phases were only present in cracks parallel to the glass surface, here, no preferential distribution were found as these phases were localised inside both parallel and secant cracks (Fig. 6a). Their presence in the AL itself is rarer (Fig. 8). Analyses showed that in general cracks in sample OU2 and MG36 were filled with crystalline phases, furthermore, in the case of OU2, all the cracks generally presented crystalline precipitates, while in the case of MG36 cracks close to the AL–PG interface were usually devoid of crystals. In the case of Ev1crystalline phases are rarer and localised exclusively in the upper part of the AL. No clear relationship was found between the nature of the crystalline phases and their localisation in secant or parallel cracks. In the case of the cracks with a thickness higher than several tens of nm and a high degree of crystallinity, EDiff images were taken. Very often, crystals are present as polycrystalline aggregates; the signal is therefore highly complex with a very high number of inter-reticular distances to take into account. Furthermore, due to the experimental setup, upon which sample has to be tilted to record the EDiff, it was complicated to perform co-located EDX and EDiff analysis, as the extent of crystalline precipitates was quite often very limited. Anyway, in a large number of cases both information were available, thus allowing formulating consistent interpretations in terms of mineralogical species. Here are reported a few cases to illustrate the most frequent phases and their location (Figs. 6–8). The whole list of crystalline phases found on all samples is reported in Table 1. For sample OU2, the area in Fig. 6 is used as an example to illustrate cracks which are totally filled with crystalline phases. In this zone, EDX analysis and EDiff have been performed in all the points indicated with a star in Fig. 6a. The area within the white rectangle (Fig. 6a and b) was found to be enriched in C, Ca, P, S, Pb, Mg, and Mn. A diffraction pattern (Fig. 6c) indicates the presence of complex phosphates, such as (Fe,Mg,Mn)3PO48H2O and Ca2P2O72H2O, Ca-sulphates (gypsum and/or anhydrite), carbonate (eitelite: Na2Mg(CO3)2) and metallic lead. In another investigated zone (triangle in Fig. 6a) metallic lead and a hydrated carbonate (ikaite: CaCO36H2O) were found. Calcite (Fig. 6a, triangle), phosphates (Ca2P2O72H2O) (Fig. 6a, triangle) and oxides (PbMnO4) (Fig. 6a, circle) were also detected. Although no sulfur was found in the EDX spectra, diffraction seems to indicate the unequivocal presence of a magnesium sulphate (Mg3(OH)4SO48H2O) (Fig. 6a, triangle). Table 1 List of crystalline secondary phases found on stained glass (SGs) samples (OU2 and Ev1) and on glass analogue (MG36). SGs MG36 Gypsum (CaSO42H2O) and/or Anhydrite (CaSO4) Syngenite (K2Ca(SO4)2.H2O) Arcanite (K2SO4) and/or Mercallite (KHSO4) Polyhalite (K2Ca2Mg(SO4)42H2O) X X X X X X X X Calcite (CaCO3) and/or Ikaite - CaCO36H2O K2CO3 Ankerite (Ca(Fe2+,Mg)(CO3)2) Eitelite (Na2Mg(CO3)2) Cerussite PbCO3 X X X X X X X Phosphates K6P6O123H2O K2CaP6O186H2O K3P3O9 (Fe,Mg,Mn)3PO48H2O Ca2P2O72H2O X X X X X X X Silicate Silica (Quartz and amorphous) Clay minerals (SG: Stevensite, GA: Palygorkite) X X Sulphates Carbonates Chlorides Sylvite (KCl) X Oxides metals Hematite, Fe2O3 X Mn2O3 Pb-permanganate (PbMnO4) Metallic Lead (Pb) X X X X X X Furthermore, the analysis of dark bands in Fig. 4a, confirmed the presence of cracks filled with crystalline phases (CaCO3, PbMnO4, metallic Pb, and Ca2P2O72H2O). A detailed analysis performed in another area, where cracks were partially empty, appeared to contain nanometre sized globular crystals (Fig. 7a and b). EDX analysis of these specific phases showed an enrichment in Ca, K, C, S, and Na, diffraction patterns identify them as CaSO4 (anhydrite, Fig. 7d1) and K5(HSO3)3S2O5 (imprecise name, Fig. 7d2). Besides EDiff obtained on these small crystals, it was impossible to acquire a consistent signal in the rest of the cracks. Nevertheless, elemental distribution maps (Fig. 7c) indicate high concentration in K, Ca, C, and S in both crystals and cracks. Therefore, it must be inferred that low crystallinity degree phases are present. Right under the cracks (white circle in Fig. 7b), chemical analysis showed the presence of Ca, C, S, F, P, and Pb (Fig. 7c), crystalline phases such as cerussite (PbCO3), ikaite, Pb permanganate (PbMnO4), sulphates and thiosulphate (Mg3(OH)4SO48H2O; Na2S2O3), meta-phosphates and phosphates (K3P3O9; Mg2P4O128H2O; (Fe,Mg,Mn)3PO48H2O)) were detected (Fig. 7d3 and d4). Finally, other carbonates (ankerite: Ca(Fe,Mg,Mn)(CO3)2; Na2Mg(CO3)2; K2CO3), hydrated sulphates (polyhalite: K2Ca2Mg(SO4)42H2O), and oxides (Mn2O3) were detected in additional zones within sample OU2. A clay mineral of the smectites group, the stevensite ((Ca, Na)xMg3 x(Si4O10)(OH)2), was detected in a localised area. As previously said, in the case of Ev1 crystalline phases are rare in the crack close to the AL and totally absent in the zone close to the AL–PG interface. Nevertheless, the few detected crystals were the same as the ones previously listed for OU2: mainly K and/or Ca rich-carbonates, sulphates, and phosphates. Although diffraction patterns could not be observed, EDX analyses showed a systematic enrichment in P and S within the cracks, indicating possible presence of low crystallinity phases. Moreover, on this sample, crystalline phases, mainly Ca-sulphate and carbonate and K–Ca phosphate, were found outside the cracks, in a zone with a lumpy aspect (Fig. 8) (a similar zone was also found on OU2). Although no further experimental evidence were found, these crystals might be localised within a highly porous zone in the glass. T. Lombardo et al. / Corrosion Science 72 (2013) 10–19 In the case of sample MG36, crystalline phases were largely detected inside the crack network, except in cracks close to the AL–PG interface. In all cracks, only polycrystalline mixtures have been discovered, they consisted mainly in Ca carbonates (calcite) and K carbonates; phosphates and sulphates were rare. As said for OU2, Ca enrichments, accompanied by C and O, are observed in laminae at the crack borders. These areas might testify of a very porous glass imbued with low crystallinity degree phases (since no signal was found for electron diffraction patterns). As for sample OU2, in one case, a clay mineral of the smectite family was found, the palygorskyte ((Mg,Al)5(Si,Al)8O20(OH)28H2O). To conclude, the composition of the crystalline phases is highly dependent on the glass composition as well as on the surrounding environment. Indeed, the greater part of them is salts composed by the cations of the glass (K and/or Ca, the principal constituents of the glass beside Si, Mg, and Na). Mn-rich crystals were also found, but their frequency was much lower than the one of the previous listed cations. Pb-rich species were detected only on OU2 and since this element is not likely to be found in large quantity in airborne aerosols, Pb has obviously not an atmospheric origin. Both medieval glasses contain very low amount of lead, moreover OU2 presents a grisaille layer at its surface. Therefore, Pb is likely to be originated from the dissolution/remobilisation of the grisaille in which it is highly concentrated. For the anionic part, sulphates, carbonates, and phosphates are the most encountered. Oxides and extremely rare silicates are also present (Table 1). Taking into account this chemical distribution, the formation of these phases might be summarised as follows: modifiers cations are extracted from the glass structure via alteration process, some of them react with the atmospheric gases (mainly CO2 and SO2) previously dissolved in aqueous solution transported inside the AL through the crack network. As for the phosphates, due to the scarce presence of P in the atmospheric medium, an endogenous origin is quite obvious. Phosphorus, extracted from the glass, re-concentrates at specific locations (laminae and lamination border) where it will react with glass cations to form secondary crystalline phases. Lastly, in the case of silicates endogenous origin is quite certain, since clay minerals, structurally similar to those found here and formed via the dissolution of glass matrix, are abundantly found in ALs of glass constantly in contact with water [9,10,34,41]. In the present case, as those silicates are very uncommon, it might be inferred that the interdiffusion is much more efficient than the glass dissolution. 3.4. Boundary between altered and pristine glass (AL–PG) In order to better understand the evolution of the alteration of the glass matrix and the formation of laminae/laminations system, FIB thin sections were realised at the border zone between the AL and the pristine glass (PG). Although at the micro-scale, all samples show a rather sharp limit separating two zones with a drastic chemical difference, observations at the nanometre scale still showed a sharp limit for Ev1 (Fig. 9), while it is more diffuse for OU2 (Fig. 7a) and MG36. For these two samples, images indicate also that the lamination neighbouring the boundary presents less noticeable laminae than those observed in the upper part of the AL (Fig. 7a). Proceeding to the chemical characterisation of these zones (Figs. 7c and 9), analyses indicate that the relative concentration in Si diminishes moving from the AL to the PG, while concentration in modifier elements (K, Ca, Mg, Na, Mn) increases. O, Al, and Fe display similar concentration in both zones. It is important to notice that Si concentration in the AL is lower than in the bulk glass. Instead, P concentration is higher in AL compared to PG (Fig. 7b). In many cases, P concentrations are found to increase considerably at lamination borders, within certain laminae and also at the boundary AL–PG. These P rich bands are also highly depleted in Si 17 (Fig. 7c). As for the exogenous elements, sample Ev1 shows important S concentration in the AL compared to PG (Fig. 9b), while for OU2 only the C is present in the AL and is usually associated with Ca. Looking carefully at the transitional zone localised in what seems to be the PG (at the micro-scale), analyses showed that in reality, for all samples, compositions gradually change to reach the concentrations of the original glass (Fig. 9). More precisely, concentrations in Mg and Ca reach values closer to the original ones and in a shorter distance as well, when compared to K concentrations. Indeed, potassium concentration remains much lower than the original one, and over a broader zone than the alkalineearth elements. Consequently, this intermediate zone can be seen as part of the alteration layer with a low alteration degree (LAL: low alteration layer). Hence, the visible interface is not the most advanced alteration front but just a chemical, and probably rheological, boundary within the AL. Taking into account these findings, it can be, thus, deducted that the extraction of alkaline elements is more effective and happens prior to that of alkaline-earth ones. Lastly, interdiffusion of glass cations and hydrated exogenous species is active in the LAL. Mobilisation of P and its consequent reconcentration at the AL–PG interface is also very significant. 3.5. Alteration scenario All the findings previously discussed on both stained and model glasses allow establishing a scenario of the alteration of K and Ca rich silicate glasses. The first step of alteration is the interdiffusion process between alkaline (K) first then followed by alkaline earth elements (Ca) and exogenous hydrogenated species (Fig. 10 step 1), as showed by the study of the transitional zone (LAL). At this stage, the necessary fluids diffuse directly from the glass surface and AL is devoid of any internal segregation. At some point the interdiffusion is limited (either because of the glass chemical heterogeneity or due to the important AL thickness) thus a discontinuity is created and isolates an alteration sub-unit: the lamination (Fig. 10, step 2). This latter then undergoes a subsequent division in nanometre thick laminae (Fig. 10, step 3). The segregation laminae/lamination is provoked by local reaction of hydrolysis of the glass matrix followed by local precipitation of secondary phases, composed by specific elements such as P and Ca. Indeed P, which structural role is not fully understood [42], is known to be insoluble and tends to induce the crystallisation of phosphate phases [43]. The calcium is present in high concentration in these glasses. Nevertheless, it is not possible to determine whether these are threshold or continuous processes. Parallel to these phenomena, the loss of glass elements and the subsequent structural modifications will lead to the weakening of the glass and the formation of cracks (Fig. 10, steps 3 and 4). In aqueous solution, formation of cracks is a direct consequence of the substitution of large glass cations by smaller species, inducing first contraction then cracking [44]. In atmospheric medium where water supply is discontinuous, cracks might also form as a consequence of the dehydration followed by shrinkage of the AL during the dry phases. These cracks would favour the transport of aqueous solutions toward the bottom part of the AL and help the development of alteration in the PG (Fig. 10, steps 3 and 4). Furthermore, after the creation of a discontinuity limiting a new lamination, a new series of laminae will form, their orientation can be different or similar to the previous one. In the first case, a structural discordance is created and jagged and undulating laminae are formed (Fig. 10, step 4). If this discordance is too strong (high rotation degree), a sub-division into two distinct lamination can take place (Fig. 10, step 5). Finally, laminae are chemical discontinuities which can be preferential zones for the formation of parallel cracks (Fig. 10, step 4). It has to be noticed that, since the LAL is usually 18 T. Lombardo et al. / Corrosion Science 72 (2013) 10–19 Fig. 10. Schematic evolution of the alteration layer (AL) with time. crack-free, the diffusion of fluids within the seemingly highly porous AL should not be excluded. Whilst the AL forms, via the repetition of several cycles of diffusion–dissolution/shrinkage, several processes take place within the cracks: first of all, the transport, by advection/diffusion, of aqueous solutions charged with exogenous compounds (S and C rich) to the deeper part of the glass; second, the precipitation of secondary phases when saturation is attained for specific species (Fig. 10, steps 4–6). The successive precipitation of these phases favours the mechanic separation and the loss of glass fragments. The crystallinity of these phases seems to be linked with the degree of alteration; it would increase in a more extensive way with the glass further weathering. Also, it seems that the filling up of cracks obeys to a temporal sequence, it starts first in superficial cracks (those present in strongly altered layers) and continue towards the deeper part. Once all cracks will be filled by neo-formed species, the circulation of fluid would significantly decrease and the alteration would slow down and eventually stop. The crystalline phases would, thus, act as a barrier to the diffusion of glass cations, as testified by the presence of Ca or P-rich zones bordering cracks (Fig. 10, step 6). The frequent presence of these zones in pluri-secular alteration layers, might justify the decrease in alteration with time. 4. Conclusion The multiscale analysis of alteration layers performed in the short and long term on samples of K–Ca rich silicate glasses showed that although at the micro-scale the alteration patterns found were different, they present a higher degree of similitude at the nanometre scale. Whatever the exposure duration and the glass composition are, alteration layers are formed by a complicated succession of concentric and radial features (laminae and lamination), deeply cut by an intense network of parallel and secant cracks. Although cracks should supply the necessary fluid to facilitate the progression of the alteration, it seems that alteration can also take place prior to their formation, via the diffusion of fluids within the AL which must be seen as highly porous. The presence of laminae and lamination indicates a local chemical and structural reorganisation resulting from the sequence of interdiffu- sion and dissolution processes. The former is the prevailing mechanism in atmospheric media, probably due to the discontinuous water supply. Lastly, with time, cracks are filled up by crystalline secondary phases formed by the interaction of glass elements and atmospheric exogenous species. Their presence might be a limiting factor to the progression of the alteration, and will eventually result in the slowing down of the alteration rate. This study highlights the importance of nano-scale heterogeneities in order to understand the temporal progression of the alteration. In the prospect of modelling the alteration kinetics on long term, it is therefore crucial to consider these complex mechanisms as they seem to influence the alteration rate evolution. Acknowledgements The authors would like to thank CP2M (Université Aix-Marseille III) for the realisation of the FIB thin sections and M. Verità for the manufacturing of model glass. The GdRChimarc and the French Ministry of Culture and Communication are acknowledged for their financial support. References [1] Z. Boksay, G. Bouquet, S. Dobos, The kinetics of the formation of leached layers on glass surface, Phys. Chem. Glasses 9 (2) (1968) 69–71. [2] R. Newton, S. Davison, Conservation of Glass, Butterworth-Heinemann Ltd., Oxford, 1989. [3] R.H. Doremus, Interdiffusion of hydrogen and alkali ions in a glass surface, J. Non-Cryst. Solids 19 (1975) 137–144. [4] B.C. Bunker, Molecular mechanisms for corrosion of silica and silicate glasses, J. Non-Cryst. Solids 179 (1994) 300–308. [5] N. Valle, A. Verney-Carron, J. Sterpenich, G. Libourel, E. Deloule, P. Jollivet, Elemental and isotopic (29Si and 18O) tracing of glass alteration mechanisms, Geochim. Cosmochim. Acta 74 (2010) 3412–3431. [6] L. Gentaz, T. Lombardo, A. Verney-Carron, A. Chabas, C. Loisel, D. Neff, S. Gin, E. Leroy, Ubiquitous presence of laminae in altered layers of glass artefacts, in: Proceeding of the Conference Integrated Approaches to the Study of Historical Glass, IASHG – SPIE, vol. 8422, Bruxelles, 2012, 116-123. [7] P. Frugier, S. Gin, Y. Minet, T. Chave, B. Bonin, N. Godon, J.E. Lartigue, P. Jollivet, A. Ayral, L. De Windt, G. Santarini, SON68 nuclear glass dissolution kinetics: current state of knowledge and basis of the new GRAAL model, J. Nucl. Mater. 380 (2008) 8–21. [8] A. Verney-Carron, S. Gin, P. Frugier, G. Libourel, Long-term modeling of alteration-transport coupling: application to a fractured Roman glass, Geochim. Cosmochim. Acta 74 (2010) 2291–2315. T. Lombardo et al. / Corrosion Science 72 (2013) 10–19 [9] A. Verney-Carron, S. Gin, G. Libourel, A fractured roman glass block altered for 1800 years in seawater: analogy with nuclear glass in a deep geological repository, Geochim. Cosmochim. Acta 72 (2008) 5372–5385. [10] J.L. Crovisier, T. Advocat, J.L. Dussaussoy, Nature and role of natural alteration gels formed on the surface of ancient volcanic glasses (natural analogs of waste containment glasses), J. Nucl. Mater. 321 (2003) 91–109. [11] C. Macquet, J.H. Thomassin, Archaeological glasses as modelling of the behavior of buried nuclear waste glass, Appl. Clay Sci. 7 (1992) 17–31. [12] G.A. Cox, B.A. Ford, The Corrosion of glass on the sea bed, J. Mater. Sci. 24 (1989) 3146–3153. [13] G.A. Cox, B.A. Ford, The long-term corrosion of glass by ground-water, J. Mater. Sci. 28 (1993) 5637–5647. [14] H. Römich, Studies of ancient glass and their application to nuclear-waste management, Mat. Res. Symp. Bull. (2003) 500–504. [15] J. Sterpenich, G. Libourel, Using stained glass windows to understand the durability of toxic waste matrices, Chem. Geol. 174 (2001) 181–193. [16] I.C. Freestone, N.D. Meeks, A.P. Middleton, Retention of phosphate in buried ceramics: an electron microbeam approach, Archaeometry 27 (1985) 161–167. [17] M. Gulmini, M. Pace, G. Ivaldi, M. Negro Ponzi, P. Mirti, Morphological and chemical characterization of weathering products on buried Sasanian glass from central Iraq, J. Non-Cryst. Solids 355 (2009) 1613–1621. [18] A. Silvestri, G. Molin, G. Salviulo, Archaeological glass alteration products in marine and land-based environments: morphological, chemical and microtextural characterization, J. Non-Cryst. Solids 351 (2005) 1338–1349. [19] M.T. Doménech-Carbó, A. Doménech-Carbó, L. Osete-Cortina, M.C. Saurí-Peris, A study on corrosion processes from the Valencian region (Spain) and its consolidation treatment, Microchim. Acta 154 (2006) 123–142. [20] L. Gentaz, T. Lombardo, A. Chabas, A. Verney-Carron, C. Loisel, Impact of neocrystallisations on the SiO2–K2O–CaO glass degradation in dry atmospheric conditions, Atmos. Environ. 55 (2012) 459–466. [21] I. Munier, R.A. Lefèvre, F. Geotti-Bianchini, M. Verità, Influence of polluted urban atmosphere on the weathering of low durability glasses, Glass. Technol. 43 (6) (2002) 225–237. [22] M. Melcher, M. Schreiner, Leaching studies on naturally weathered potashlime-silica glasses, J. Non-Cryst. Solids 352 (2006) 369–379. [23] B. Dal Bianco, R. Bertoncello, L. Milanese, S. Barison, Glass corrosion across the Alps: a surface study of chemical corrosion of glasses found in marine and ground environments, Archaeometry 47 (2005) 351–360. [24] A. Longinelli, A. Silvestri, G. Molin, G. Salviulo, 1.8 ka old glass from the Roman ship Julia Felix: glass-water oxygen isotope exchange, Chem. Geol. 211 (2004) 335–342. [25] R.H. Brill, H.P. Hood, A new method for dating ancient glass, Nature 189 (1961) 12. [26] R.H. Brill, The record of time in weathered glass, Archaeology 14 (1961) 18–22. [27] R.G. Newton, The enigma of the layered crusts on some weathered glasses, a chronological account of the investigations, Archaeology 13 (1971) 1–9. [28] G. Salviulo, A. Silvestri, G. Molin, R. Bertoncello, An archaeometric study of the bulk and surface weathering characteristics of early medieval (5th–7th century) glass from the Po valley, Northern Italy, J. Archaeol. Sci. 31 (2004) 295–306. 19 [29] P. Clifford, Patinated glass from Cossack, Western Australia, J. Royal. Soc. West Aust. 91 (2008) 275–291. [30] R. Prochazka, Natural corrosion of the uranium-colored historical glasses, J. Non-Cryst. Solids 353 (2007) 2052–2056. [31] A. Genga, M. Sicilian, L. Famà, E. Filippo, T. Siciliano, A. Mangone, A. Traini, C. Laganara, Characterization of surface layers formed under natural environmental conditions on medieval glass from Siponto (Southern Italy), Mat. Chem. Phys. 111 (2008) 480–485. [32] R. Procházka, V. Ettler, V. Goliáš, M. Klementová, M. Mihaljevicˇ, O. Šebek, L. Strnad, A comparison of natural and experimental long-term corrosion of uranium-coloured glass, J. Non-Cryst. Solids 355 (2009) 2134–2142. [33] P. Croveri, I. Fragalà, E. Ciliberto, Analysis of glass tesserae from the mosaics of the ‘Villa del Casale’ near Piazza Armerina (Enna, Italy), Appl. Phys. A 100 (3) (2010) 927–935. [34] N.A. Stroncik, H.U. Schmincke, Evolution of palagonite: crystallization, chemical changes, and element budget, Geochem. Geophys. Geosyst. 22 (2001) 680–697. [35] T. Geisler, A. Janssen, D. Scheiter, T. Stephan, J. Berndt, A. Putnis, Aqueous corrosion of borosilicate glass under acidic conditions: a new corrosion mechanism, J. Non Cryst. Solids 356 (2010) 1458–1465. [36] T. Lombardo, C. Loisel, L. Gentaz, A. Chabas, M. Verità, I. Pallot-Frossard, Long term assessment atmospheric decay of stained glass windows, Corr. Eng. Sci. Technol. 45 (5) (2010) 420–424. [37] N. Carmona, L. Laiz, J.M. Gonzalez, M. Garcia-Heras, M.A. Villegas, C. SaizJimenez, Biodeterioration of historic stained glasses from the Cartuja de Miraflores (Spain), Int. Biodeter. Biodegrad. 58 (2006) 155–161. [38] M. Verità, Paintwork in medieval stained-glass windows: composition, weathering, and conservation, in: The Art of Collaboration Stained-Glass Conservation in the Twenty-First Century, Miller H. Publishers, 2010, pp. 210– 216. [39] F. Geotti-Bianchini, C. Nicola, M. Preo, M. Verità, MicroIRRS and EPMA study of the weathering of potash-lime-silicate glasses, Riv. StazioneSper. Vetro 3 (2005) 49–61. [40] L. Gentaz, T. Lombardo, C. Loisel, A. Chabas, M. Vallotto, Early stage of weathering of medieval-like potash–lime model glass: evaluation of key factors, Environ. Sci. Poll. Res. 18 (2) (2011) 291–300. [41] P. Frugier, S. Gin, J.E. Lartigue, Deloule, SON68 Glass dissolution kinetics at high reaction progress; mechanisms accounting for the residual alteration rate, Mat. Res. Soc. Symp. Proc. 932 (2006) 305–312. [42] M.J. Toplis, B. Reynard, Temperature and time-dependent changes of structure in phosphorus containing aluminosilicate liquids and glasses: in situ Raman spectroscopy at high temperature, J. Non-Cryst. Solids 263 & 264 (2000) 123– 131. [43] S. Padilla, J. Román, A. Carenas, M. Vallet-Regí, The influence of the phosphorus content on the bioactivity of sol–gel glass ceramics, Biomaterials 26 (2005) 475–483. [44] A. Tournié, P. Ricciardi, Ph. Colomban, Glass corrosion mechanisms: a multiscale analysis, Solid State Ionics 179 (2008) 2142–2154.