Aqueous corrosion of borosilicate glass under acidic conditions: A new corrosion mechanism

Journal of Non-Crystalline Solids 356 (2010) 1458–1465

Contents lists available at ScienceDirect

Journal of Non-Crystalline Solids j o u r n a l h o m e p a g e : w w w. e l s ev i e r. c o m / l o c a t e / j n o n c r y s o l

Aqueous corrosion of borosilicate glass under acidic conditions: A new corrosion mechanism Thorsten Geisler a,b,⁎, Arne Janssen a, Daniel Scheiter a, Thomas Stephan c,1, Jasper Berndt a, Andrew Putnis a a b c

Institut für Mineralogie, Westfälische Wilhelms-Universität Münster, Corrensstrasse 24, 48149 Münster, Germany Mineralogie, Department für Geowissenschaften, University of Hamburg, Grindelallee 48, 20146 Hamburg, Germany Institut für Planetologie, Westfälische Wilhelms-Universität Münster, Wilhelm-Klemm-Str. 10, 48149 Münster, Germany

a r t i c l e

i n f o

Article history: Received 29 September 2009 Received in revised form 19 April 2010 Available online 25 May 2010 Keywords: Glasses G205; Chemical durability C160; Corrosion C275; Non-linear dynamics N195

a b s t r a c t One important application of borosilicate glass is its use as a nuclear waste form to immobilize high-level nuclear waste. Understanding the corrosion mechanism of borosilicate glasses in aqueous solutions is essential to reliably predict their long-term behavior in the worst-case scenario of glass-groundwater contact in a geologic repository. Traditional models evaluate the long-term corrosion process on the basis of diffusion-controlled hydration and ion exchange reactions that are followed by solid-state reconstruction of the hydrolyzed glass network. Here we report textural, chemical, and 18O and 26Mg isotope tracer results from corrosion experiments with a borosilicate glass in an acidic aqueous solution (initial pH ≈0, T = 150 °C, 6 to 336 h) that contradict such a paradigm. We propose a new mechanistic model for glass corrosion under acidic conditions that is based on congruent (stoichiometric) dissolution of the glass that is spatially and temporally coupled to the precipitation of amorphous silica at an inward moving reaction front. The model potentially provides a novel framework to understand apparently contradictory observations made under more moderate conditions and to evaluate the long-term aqueous durability of silicate glasses. © 2010 Elsevier B.V. All rights reserved.

1. Introduction Borosilicate glass is currently the internationally preferred material for the immobilization of high-level nuclear waste, including excess plutonium from dismantled nuclear weapons and highly radioactive liquid/solid waste resulting from the reprocessing of spent fuel, for permanent disposal in a geological repository [1]. Due to the refractory properties and physical strength of borosilicate glass, it is also used for a number of medical and industrial applications for which knowledge of its corrosion in aqueous environments is needed (e.g., for the evaluation of the performance of tableware glasses in dish washers, or of the biopersistence, in relation to suspected cancerogenic potency, of inhaled man-made glass fibers). In a geological repository, the nuclear waste containing borosilicate glass is embedded in a multi-barrier confinement system that should prevent both the access of water to the glass and the release of radioactivity. However, the corrosion of the glass by groundwater cannot be ruled out for long-term storage over geological time scales and experimental glass corrosion studies, in particular those dealing

⁎ Corresponding author. Institut für Mineralogie, Westfälische Wilhelms-Universität Münster, Corrensstrasse 24, 48149 Münster, Germany. E-mail addresses: [email protected], [email protected] (T. Geisler). 1 Present address: Department of the Geophysical Sciences, University of Chicago, 5734 South Ellis Avenue, Chicago, IL 60637, USA. 0022-3093/$ – see front matter © 2010 Elsevier B.V. All rights reserved. doi:10.1016/j.jnoncrysol.2010.04.033

with the corrosion mechanism, are thus critical for the evaluation of the long-term performance of a nuclear waste glass in the worst-case scenario of glass-groundwater contact. Numerous experimental investigations on glass corrosion in the last 40 years have indicated that once water comes in contact with the glass, protons/hydroxonium ions diffuse into the glass network and ion exchange occurs between protons and alkali network modifiers such as Na, producing silanol (Si–OH) groups [1–24]. During the ion exchange reaction (Stage I) of the form −

þ

þ

−

þ

þ

≡Si–O Na þ H → ≡ Si–O H þ Na

the silicate network becomes hydrolyzed and – as long as the concentration of elements lost to the solution is low and the hydrolysis reactions are faster – the glass dissolves congruently (stoichiometrically) with an initial rate r0. Under static conditions, as would be expected in a natural repository environment, Stage I is followed by an intermediate phase during which the release of elements to the solution is slowed down due to the accumulation of dissolved glass constituents in the solution, resulting in a decrease of the chemical potential between glass and solution (Stage II). Once silica reaches saturation in the solution with a quasi-steady-state Si concentration, the rate of elemental release usually drops further by several orders of magnitude below r0 (Stage III). Under these conditions a “gel” layer forms around the pristine glass. The “gel” layer has been interpreted to consist of a hydrolyzed, residual skeleton

T. Geisler et al. / Journal of Non-Crystalline Solids 356 (2010) 1458–1465

of the pristine glass that eventually recondensates in the solid state as described by following reaction:

1459

2. Materials and methods 2.1. Scanning electron microscope

≡Si–OH þ HO–Si ≡ → ≡ Si–O –Si≡ þ H2 O

ð1Þ

This repolymerization reaction was observed regardless of the nature of the solution, while the gel morphology was different in acidic and basic solutions [2,4]. The “gel” layer was found (i) to be porous with a typical pores size in the nanometer range [9–11], (ii) to show a well-defined phase boundary towards the unaltered glass [12,13], and (iii) to contain molecular water [14,15]. Secondary phases such as, for instance, phyllosilicates or zeolites may also precipitate at the “gel” layer/solution interface [16]. A sketch of the conceptual model as proposed in Ref. [5] is shown in Fig. 1. Currently, there is an ongoing and passionate debate [1,12,17,18] about whether the kinetics of glass corrosion in silica saturated solution (Stage III) is controlled by (i) a protective effect of the “gel” layer [8– 11,17,18], by (ii) the degree of silica saturation in solution as assumed by classic kinetic laws that are based on the notions of “chemical reaction affinity” and “deviation from equilibrium” [1,5,15,19–21], or by (iii) persisting alkali-proton exchange reactions [22]. Other authors have shown that a mathematical combination of the affinity-driven dissolution (first-order) rate law with water diffusion, forming the hydrated glass, can explain the observed reaction kinetics in silica saturated solution [20]. Under such conditions, the steady state between the glass dissolution and ion exchange is believed to be eliminated, and water diffusion becomes rate limiting. The current discussion highlights an inherent problem of the interpretation of short-term laboratory solution data, namely that they can usually be fitted equally well by different kinetic laws. It is thus questionable whether reliable information about the mechanism of the corrosion reactions can be deduced by modeling the solution composition of short-term laboratory experiments. Nevertheless, despite the intensive discussion about the rate-limiting reaction step during glass corrosion, there seems to be a general consensus that the formation of a “gel” layer is the result of diffusion-controlled ion exchange and hydration reactions, leaving behind a partly hydrolyzed glass network that eventually undergoes solid-state re-condensation reactions. Here we report new results from an aqueous corrosion study with a 18 26 borosilicate glass under acidic conditions using O and Mg as tracers that contradict such a view. During such isotope tracer experiments, atoms of the tracers mix with material of normal isotopic composition that has been released into the solution from the reactants, and are simultaneously taken up by product phases. By monitoring the isotopic composition of such elements in the product phase, it is possible to assess the reaction mechanism. Based on the results of the tracer experiments as well as textural observations we propose a new mechanistic model for glass corrosion in acidic solutions. However, this model potentially provides a novel framework to evaluate the longterm aqueous durability of silicate glasses in general.

Fig. 1. Schematic drawing of the surface region of an altered glass (after [5]).

Backscatter electron (BSE) images were obtained using a JEOL TM 840 scanning electron microscope. The acceleration voltage was set to 20 kV at a beam current of 6 × 10− 9 to 3 × 10−8 A. Chemical analyses were performed using an OxfordTM INCA EDX (energy dispersive Xray) detector. Scanning electron (SE) imaging of the surface of the experimentally altered glass cuboids was carried out using a JEOL 6300F field emission scanning electron microscope at 1 kV and a beam current of 10−10 A. 2.2. Inductively-coupled plasma optical emission spectrometry (ICP-OES) The experimental solution was analyzed for the elements Na, Ca, Al, Mg, Si, B, Li, Ti, and Ce with a Thermo Jarrell AshTM Atom Scan 25. Before the measurements the experimental solution was centrifuged to remove any solid particles and then diluted with p.a. grade 2% nitric acid using a dilution factor of 1:9. The internal precision for all analyzed elements was determined from repeated analyses to be better than ±2%. 2.3. Reflectance infrared (IR) spectroscopy Unpolarized reflectance infrared spectra of the pristine and altered glass were recorded with a Bruker Equinox 55 spectrometer on polished specimen. The measurements were carried out on an A590 microscope by narrowing the beam with a 30 µm aperture. Spectra represent the average of 256 scans with a wavelength resolution of 4 cm− 1. 2.4. Electron microprobe analyses The synthesized glass and the corrosion products have been quantitatively analyzed for Si (Kα), Al (Kα), Ce(Lα), Ca(Kα), Mg(Kα), Na(Kα), and Ti(Kα) by a Cameca SX100 electron microprobe using an acceleration voltage of 15 kV at 10 nA beam current. Total counting times for peak and background were 10 s for the Si–Ka line and 30 s for all other elements. Natural and synthetic oxides and silicates were used as standards and the phi–rho correction procedure was applied to correct for matrix effects. To reduce any Na loss during the measurements, an electron beam spot size of 20 µm was chosen and Na was analyzed at the beginning of each spot analysis. However, the analyses on the silica corrosion rim should be considered as ballpark figures, since the silica is highly porous and unstable under the electron beam. 2.5. Laser ablation inductively-coupled plasma mass spectrometry (LAICP-MS) Sample ablation for Mg isotopic ratio analyses of the glass, altered 26 in Mg-enriched solution, has been carried out with a pulsed 193 nm ArF excimer laser (UP193HE, New Wave Research). A repetition rate of 5 Hz and a laser energy of ∼9 J/cm2 were used. The beam spot diameter was 8 µm. Isotope analyses were carried out with an Element 2 mass spectrometer (ThermoFisher). Forward power was 1320 W and reflected power was b2 W, gas flow rates were 0.78 l/m for He (carrier gas of ablated material), and 1.0 l/m and 1.3 l/m for the Ar-auxiliary and sample gas, respectively. Cooling gas flow rate was set to 16 l/min. Before starting an analysis, the system has been tuned 139 (torch position, lenses, gas flows) on a NIST 612 glass measuring La, 232 232 16 139 Th and Th O to get stable signals and high sensitivity on La and 232 232 16 232 Th peaks, as well as low oxide rates ( Th O/ Th ∼0.1%) during ablation. The NIST 612 glass was used to correct for mass bias and fractionation. Dwell time was 100 ms on each isotope, while measuring 12 samples per peak. Overall time of a single analysis was 60 s (20 s for background, 40 s for peak after switching laser on).

1460

T. Geisler et al. / Journal of Non-Crystalline Solids 356 (2010) 1458–1465

2.6. Time-of-flight secondary ion mass spectrometry (TOF-SIMS) 18

During the TOF-SIMS analysis the glass sample, altered in the Oeniched solution, was bombarded with a pulsed 25 keV Ga + primary ion beam (pulse width 1.5 ns). Prior to the analysis the sample was sputter-cleaned with 3 keV argon ions. A fine-focused gallium beam with a diameter of ∼300 nm was subsequently rastered over the two sample areas of about 400 × 400 µm2 and 45 × 20 µm (256 × 256 pixel), respectively. For each pixel, entire mass spectra were determined for 18 16 18 both polarities in two consecutive measurements. The O/( O + O) ratio could be quantified directly from the number of counts because only a single detector is used and matrix effects have a little effect on isotope ratios. More details of the TOF-SIMS technique are described elsewhere [25].

2.7. Glass preparation For our experimental study we synthesized a borosilicate glass that has a composition similar to the German waste glass used in experi-

ments at the reprocessing plant in Karlsruhe, Germany. The composition of the glass was (in wt.%): 54.0SiO2 · 14.8B2O3 · 2.4Al2O3 × 1.1TiO2 · 1.8MgO · 4.8CaO · 2.9Li2O · 7.2Na2O · 10.0CeO2, as determined by the weighed-in quantity (B and Li) and 250 electron microprobe measurements, respectively. For the synthesis of the glass following oxides or carbonates were used: Li2O (Alfa Aesar 99.95%), Na2CO3 (Aldrich 99.5%), CaO (Alfa Aesar 99.95%), MgO (Alfa Aesar), Al2O3 (Alfa Aesar 99.95%), B2O3 (Alfa Aesar 99%), CeO2 (Alfa Aesar 99.5%), SiO2 (Alfa Aesar), TiO2 (Alfa Aesar 99.9%). All oxide-powders were kept in a cabinet desiccator at 60 °C for several days in order to remove possible humidity. Additionally, CaO and MgO were both dried for 3 h at 750 °C shortly before weighing to avoid the formation of MgCO3 and CaCO3, respectively. After weighing, the powders were homogenized in ethanol for 1 h using an agate mortar and subsequently dried under ruby light. The final mixture was melted in a platinum crucible at 1200 °C for 3 h and then quenched in air. The glass was triturated again using to repeat the melting/quenching procedure for better homogenization. The color of the synthesized glass was brownyellowish (Fig. 2a), resulting from Ce that was added as a surrogate for Pu. The glass density was 2.715 ± 0.004 g/cm3. X-ray diffraction

Fig. 2. Textural and chemical characteristics of experimentally altered borosilicate glass cuboids and naturally altered silicate glass. (a, b) Optical images of (a) an untreated and (b) a borosilicate glass cuboid treated for 96 h at 150 °C in a 1 M HCl solution. (c) SE images of the surface of a borosilicate glass cuboid shown in (b). Note the occurrence of silica spherules (inset diagram) and rutile (TiO2) crystals with typical (301) rutile twinning at the surface. (d) BSE image of a cross-section of the cuboid shown in (b). (e) BSE image of the corrosion zone. Note the occurrence of an outer plain zone (pz) and an inner zone that shows an oscillating banding pattern (ptz) and a sharp interface towards the pristine glass (p). The white line marks the position of the EDX line profile that is shown below the image. The given average elemental concentrations were obtained by quantitative electron microprobe measurements. The spot locations are marked by white circles in the BSE image. (f) BSE image of a corrosion rim formed after 6 h of treatment in a 1 M HCl solution at 150 °C. (g) BSE image of a 2500 years old, naturally altered glass from an archeological site in Turkey (Dülük Baba Tepesi near Doliche) also showing non-equilibrium banding patterns inside the corrosion zone (unpubl. image).

T. Geisler et al. / Journal of Non-Crystalline Solids 356 (2010) 1458–1465

1461

measurements and optical investigations revealed that the glass was totally amorphous. To observe any morphological changes during experimental alteration, polished cuboids with edge length of about 2.5 mm were prepared for the experiments. The cuboids were cleaned in distilled water and ethanol, weighed, and scanned with a flat bed scanner with a resolution of 1600 dpi on all six sides to determine the geometrical surface area. The error of the total surface was estimated to be less than 1% by repeated analyses, partly carried out by different operators. The geometrical surface area was used to calculate the surface-to-volume ratio (S/V), which is needed to normalize the concentration of an element i in solution (NLi) according to   −2 −2 = 10 NLi g

Ci ; Xi S = V

where Ci is the concentration (µg/g) of element i in solution and Xi is the mass fraction of element i in the glass. 2.8. Hydrothermal experiments The experiments were carried out with the glass cuboids and a 1 M HCl solution (initial pH ≈0) at a temperature of 150 ± 1 °C for 6 to 336 h in home-made, 3 ml Teflon© reactors under static conditions, resulting in a surface-area-to-solution-volume ratio (S/V) of about 0.189 cm− 1 (Table 1). The HCl used in the experiments was distilled from p.a. grade, resulting in sufficiently low blanks (ng/g range) for the elements of interest. All Teflon© vials were cleaned in HCl and HNO3, bombed with HF-HNO3 before usage. Two additional experiments were performed for 18 26 96 h with a 1 M HCl solution that was enriched with O and Mg. The 18 26 O- and Mg-enriched solutions were prepared by mixing 2 M HCl with 18 18 H2 O (95 at.% O, IconIsotope) in equal proportions and by dissolving 26 26 10.77 mg of MgO (99 at.% Mg, Trace Sciences) in the 1 M HCl solution 26 (∼3300 µg/g of Mg in solution), respectively. Before and after the experiments, the closed reactors were weighted to test for any fluid loss. The experimentally treated glass cuboids have been washed several times in distilled water and then dried at 55 °C for about 24 h before they have been embedded in epoxy resin, cut in half, and polished for further analytical work. 3. Results 3.1. Glass corrosion kinetics Fig. 3 shows the normalized concentration (NLi) of all glass constituents in the experimental solution as a function of the duration of the experiments (Table 1). The most significant observation is that the glass continued to release all network modifiers and boron in close-to-stoichiometric proportions even after the silica concentration has reached a maximum after 24 h (CSi = 220.7 ± 0.6 µg/g, NLSi = 49.9 g/m2) and a quasi-steady-state concentration (CSi = 198.6 ± 0.5 µg/g, NLSi = 44.4 g/m2) after about 48 h. The observed Si concentrations in solution agree well with the solubility limit of

Fig. 3. The normalized concentration (NLi) of all glass constituents in the solution as a function of the duration of the experiments performed in a 1 M HCl solution at 150 °C. The inset diagram shows the evolution of the Si concentration in solution with time in comparison with the calculated solubility limit of amorphous silica in a 1 M HCl solution at 150 °C (thick gray curve).

amorphous silica in a 1 M HCl solution at 150 °C as calculated using the geochemical PHREEQC code [26] (Fig. 3), suggesting that the overall corrosion reaction is independent of the silica saturation in solution. Although the normalized concentrations in solution are about 10 to 100 times higher in our experiments than measured in numerous static glass corrosion studies that were performed with borosilicate glasses under weakly basic or acidic conditions and temperatures below 100 °C, the overall reaction trends are very similar (ignoring the behavior of Ti for the moment). The reaction kinetics reflected by our solution data can thus qualitatively be linked to the three major stages of borosilicate glass corrosion described in Ref. [18]. In our experiments the initial corrosion rate, as given by the B release rate, is ∼370 g− 2d− 1 (Stage I: 0–6 h), which reduces to ∼135 g− 2d− 1 during the intermediate stage (Stage II: 6–48 h) and finally to ∼ 41 g− 2d− 1 after the Si concentration has reached the steady-state concentration (Stage III: 48–336 h). The same pattern of corrosion suggests that our results can be compared with those obtained under moderate physico-chemical conditions, despite the more extreme conditions in our experiments. 3.2. Chemical, structural, and textural characteristics of the corrosion product Optical investigations revealed that the treated glass cuboids are covered with a thin corrosion layer that has a milky-white color (Fig. 2b), resulting from scattering of light by numerous pores, whereas their external shape and dimension are preserved (Fig. 2d); an observation that has been called isovolumetric corrosion [23]. Scanning electron microscope investigations showed that the milky-

Table 1 Normalized mass losses of all glass elements i (NLi) calculated for all experiments carried out in a 1 M HCl solution at 150 °C. t (h)

Weight (g)

S/Va (cm− 1)

NLCa (g/m2)

NLSi (g/m2)

NLTi (g/m2)

NLCe (g/m2)

NLB (g/m2)

NLLi (g/m2)

NLMg (g/m2)

NLNa (g/m2)

NLAl (g/m2)

db (µm)

6 24 48 96 168 336

0.04248 0.04227 0.04249 0.04217 0.04238 0.04214

0.1892 0.1877 0.1898 0.1892 0.1895 0.1890

117.2 190.3 410.6 539.6 706.0 1064.2

39.2 49.9 47.6 46.5 44.5 44.4

88.5 108.4 215.8 191.3 137.7 61.0

101.1 155.7 360.2 459.7 569.2 910.1

93.5 149.7 331.9 453.5 586.0 825.3

106.4 168.9 368.1 491.3 658.8 1089.8

97.8 159.7 367.5 460.9 684.6 957.9

95.4 161.2 353.6 463.1 668.3 1030.1

77.2 177.1 364.3 454.5 675.0 884.7

30 50 80 300 450 900

a b

Cuboid surface-to-solution-volume ratio. Estimated 2-sigma standard errors of NLi are in all cases smaller than ± 2%. Variations of the corrosion rim thickness d are in the order of ±20% for thicknesses below 100 µm and ±10% for thicker rims.

1462

T. Geisler et al. / Journal of Non-Crystalline Solids 356 (2010) 1458–1465

white corrosion layer consists of nm-sized spherules and a high porosity (Fig. 2c). Furthermore, 1 to 2 µm large, prismatic rutile (TiO2) crystals, often with typical (301) rutile twinning, are visible at the surface of those cuboids that have been treated for more than 48 h (Fig. 2c), explaining the observed decrease of the Ti concentration in solution with increasing duration of the experiments (Fig. 3). BSE images from cross-sections of the altered glass cuboids show up to ∼ 300 µm thick corrosion rims after 96 h (and up to ∼ 900 µm after 336 h; Table 1), which are characterized by an, on average, lower BSE intensity than the pristine glass (Fig. 2d, e, and f). Such a corrosion thickness is within the order of magnitude of the thickness of alteration crusts observed around 2500 years old archeological glasses that were in contact with groundwater (Fig. 2g). A closer look reveals that the reaction rims consist of two texturally distinct zones (Fig. 2e and f): (1) an inner zone with complex oscillatory banding structures with bifurcations, composed of BSE-bright and dark bands, which indicate variations in the mean atomic number (labeled ptz in Fig. 2e and f), and (2) a plain outer rim (labeled pz in Fig. 2e and f). Whereas the thickness of the patterned zone increases with increasing duration of the experiment (Table 1), the plain outer zone has a relative constant thickness between about 10 and 30 µm. An important observation is that the interface between the patterned zone and the unaltered pristine glass is marked by free space or a gap, and is thus sharp on the atomic scale. Due to this gap the corrosion layer easily separates from the unaltered glass (Fig. 2b). The reaction rim is truncated by numerous cracks that run approximately perpendicular to the reaction interface where they converge with the free space, indicating that they were formed after the experiments during drying of the sample as a result of shrinking [23]. Reflectance micro-IR spectra of both the patterned and plain area reveal that both zones consist of porous silica (Fig. 4). Chemically, the outer plain silica rim pz consists almost entirely of SiO2 with only trace amounts of other elements, as revealed by electron microprobe measurements (Fig. 2e). However, the patterned areas retained, on average, significant concentrations of Ce (∼2.4 wt.%), Ti (∼1.1 wt.%), and Al (∼0.4 wt.%), whereas all other elements occur only in trace concentrations (b0.07 wt.%). Whereas the maximal Ce and Al concentrations correspond to the bright bands (A bands) seen in the BSE image (Fig. 2e and f), Ti is relatively enriched within the dark-BSE bands (B bands) that also have slightly higher Si concentrations, i.e., there is a

Fig. 4. Representative reflectance micro-IR spectra from the pristine borosilicate glass 16 18 and the plain and patterned zone of corrosion rims formed in a O and O-enriched solution (96 h, 150 °C, initial pH = 0). Note that the IR spectra from the corrosion rim resemble those obtained from porous silica [38].

clear phase shift between the Ti and Si oscillations and those of Ce and Al, respectively (Fig. 2e). The Na and Ca concentrations drop dramatically at the reaction interface to a trace level, with no apparent diffusion profile, neither inside the reaction rim nor inside the pristine glass (Figs. 2e and 5). The TOF-SIMS images of B, Li, and Mg also indicate a sharp drop at the reaction interface (Fig. 5). However, high-Ti areas in the TOF-SIMS image indicate the existence of TiO2 (rutile) crystals also inside the silica corrosion zone. 18 18 The O/O TOF-SIMS image from a corrosion zone formed in the O18 enriched solution clearly shows that O is strongly enriched in the reaction zone with a sharp (on a µm-scale) interface towards the pristine glass and no apparent diffusion profiles, neither in the corrosion zone nor across the reaction interface into the pristine glass 18 (Fig. 5). Aside from an obvious enrichment of O along the cracks, 18 which could be related to adsorbed water, the highest O concentrations occur in the plain outer zone and in a ∼100 µm wide zone that borders with the reaction interface. The TOF-SIMS measurements, 18 however, do not allow distinguishing between O that is located within the tetrahedral silica network and those that originates from any water species in the reaction rim. However, a distinction is possible using infrared spectroscopy due to the atomic mass effect on 18 16 the frequency of molecular vibrations. If O replaces O in the SiO4 tetrahedra of the silica network, one would expect a shift to lower frequencies of those [SiO4] vibrational modes that involve the motion of oxygen due to the different masses of both isotopes. Fig. 4 shows representative IR spectra from the patterned and plain areas formed 16 in a solution of a natural oxygen isotope composition (∼ 99.8 at.% O) 18 and in the O-enriched solution. Indeed, the Si–O–Si stretching mode 18 frequency in the reaction rim formed in the O-enriched solution is −1 shifted by ∼ 20 cm towards lower frequencies when compared to the frequency of this mode in the reaction rims formed in the “isotopically normal” solution. This unambiguously demonstrates that 18 O is incorporated into the tetrahedral silica network rather than being associated with a water species. It is noteworthy that our experimental findings are in line with oxygen isotope measurements of corrosion zones around basaltic glass samples from the Gordo Rise (Pacific Ocean), which indicate that all the oxygen atoms in the altered glass were replaced by oxygen from seawater [27]. Fig. 6 shows the result of LA-ICP-MS measurements made across 26 the reaction rim that formed in the experiment with the Mgenriched solution, which was used as a tracer to test whether the glass network modifiers also exchange with the solution. The measured 26 24 Mg/ Mg values of the reaction rim are 10 to 50 times higher than the natural isotope ratio of 0.1394, demonstrating a strong enrichment of 26 26 24 Mg in the rim. The measured Mg/ Mg values across the rim do not describe a diffusion profile, but show three distinct maxima, one in the plain outer zone and two inside the patterned areas. From the measured total count rates of all three stable Mg isotopes obtained from the corrosion zone and the pristine glass (1.09 wt.% Mg), we obtain Mg concentrations between about 60 and 240 µg/g (Fig. 5), which are significantly below the initial and final concentrations of Mg in the solution (∼3300 µg/g and ∼3400 µg/g, respectively). Such an observation contradicts the classical diffusion-based models for glass corrosion, as the chemical potential of Mg in solution (μ solution ) Mg after the experiment is higher than the chemical potential of Mg in solution solution the silica rim (μsilica − μ silica /msilica Mg ), i.e., μ Mg Mg ≈ RTln (m Mg Mg ) N 0, where R, T, and mMg are the gas constant, temperature, and the molality of Mg in silica and solution, respectively. This would imply that Mg diffused against a chemical potential after a certain amount of glass has been transformed. It is further noteworthy that the Mg con26 24 centrations are inversely correlated with the Mg/ Mg values, 26 indicating that the Mg enrichment cannot be the result of the 26 26 precipitation of MgCl2 or other Mg salts from relict pore solution during drying of the sample, because in such a case a positive correlation would be expected. Note also that the glass cuboids have thoroughly been washed in distilled water before drying.

T. Geisler et al. / Journal of Non-Crystalline Solids 356 (2010) 1458–1465

1463

18

Fig. 5. Representative TOF-SIMS images from two areas of a corrosion zone that formed around a borosilicate glass cuboid after 96 h in a O-enriched solution of an initial pH of 0 at 150 °C. The upper left image shows the corresponding BSE images. Abbreviations: p pristine glass, ptz patterned zone, and pz plain zone.

4. Discussion 4.1. A new model for borosilicate glass corrosion under acidic conditions Several observations made in this study such as (i) the occurrence 18 26 of chemical oscillations, (ii) the enrichment of O and Mg in the silica rim without observable diffusion profiles, (iii) the sharp phase boundary of the reaction rim towards the pristine glass, (iv) the high porosity in the silica, and (v) the occurrence of silica spherules at the surface are not at all compatible with classical theories about the formation of the “gel layer” that are based on diffusion-controlled hydration and ion exchange reactions and subsequent solid-state re18 condensation of the hydrolyzed glass network. In contrast, the O and

26

24

Fig. 6. Mg/ Mg values and total count Mg concentrations, measured by LA-ICPMS, across a corrosion zone that formed around a borosilicate glass cuboid after 96 h in a 26 Mg-enriched solution of an initial pH of 0 at 150 °C. Abbreviations: p pristine glass, ptz patterned zone, and pz plain zone.

26

the Mg enrichments are strong evidence that the amorphous silica directly precipitated from the solution. As the overall dimension and shape of the glass cuboids has been retained during the reaction (Fig. 2d) – an observation that has also been made in previous in situ glass alteration studies [23] – the rates of glass dissolution and silica precipitation have to be coupled. Such a reaction has recently been described as an interface-coupled dissolution–reprecipitation reaction, where both dissolution and instantaneous precipitation occur within a solution boundary layer that moves into the parent phase as the reaction proceeds [28–31]. At the beginning of the reaction, the glass dissolves congruently until the supersaturation of amorphous silica in the solution boundary layer is reached. As the mass transport rate of silica through the boundary layer is likely to be very low, because orthosilicic acid polymerizes to form silica colloids, silica supersaturation inside boundary layer may quickly be reached. At this stage the precipitation of amorphous silica occurs by aggregation of the silica colloids, which reduces the silica concentration in the boundary layer in the immediate vicinity of the precipitated silica so that locally further glass dissolution can take place. The existence of such a boundary layer at the glass–solution interface is supported by the observation that a silica corrosion zone has already been formed within the first 6 h of our experiments (Fig. 2f), although the solution reservoir was not saturated with respect to silica (Fig. 3). Note that a peculiarity of aqueous silica chemistry is the fact that when a solution becomes supersaturated with respect to amorphous silica, quartz does not crystallize, because of a slow reaction rate. The equilibrium between solid amorphous silica and dissolved silica is thus a metastable equilibrium. The observation of rutile crystals at the surface of the amorphous silica layer (Fig. 2c) either suggests that the solution boundary layer was supersaturated with respect to silica before it was supersaturated with respect to TiO2 or that there exists a kinetic barrier towards nucleation of TiO2, resulting in a high supersaturation of Ti in solution. The evolution of the Ti concentration in solution with time

1464

T. Geisler et al. / Journal of Non-Crystalline Solids 356 (2010) 1458–1465

(Fig. 3) as well as a comparison of the observed Ti concentration in solution (6.9–24.6 µg/g Ti) with results of rutile solubility experiments, which yielded a TiO2 solubility of less than 0.5 µg/g Ti in a solution of pH= 0 at 150 °C [32], indicate that the latter interpretation is more likely. The free energy difference needed to drive an interface-coupled dissolution–reprecipitation reaction is provided by the difference of the solubility of the parent and product phase in solution, which in turn may result from textural and/or chemical energy differences. Because both reactions take place at the glass–silica interface, only a small amount of material needs to be in solution in the boundary layer at the interface at any one time. This implies that the relative solubility of the two solids in the solution is more important than their absolute solubility. Such statement is supported by the observation that even solids having an extremely low solubility in aqueous solutions, such as, e.g., pyrochlore-group solids (A2B2O7) [29], often show intensive replacement reactions [28–31]. Once silica saturation in the surrounding solution reservoir has been reached, the silica rim is in chemical equilibrium with the solution and the reaction should stop. However, since the silica is highly porous, which is a general feature of the silica or “gel” layer observed under a wide range of pH conditions [9,10], mass transport through the silica corrosion zone to and from the reaction interface is possible. Therefore, the reaction can proceed, driven by the solubility difference between the pristine glass and the silica reaction product, even if the surrounding solution reservoir is saturated with silica, because the pristine glass is not in equilibrium with the interfacial solution. With increasing thickness of the silica reaction rim, however, the reaction becomes more and more controlled by the transport of material through the pore space of the silica rim. The nature of the porosity (e.g., pore size, connectivity) in the silica rim and its evolution with time thus controls the longterm reaction kinetics. Furthermore, experimental and theoretical studies have shown that the porosity in the corrosion zone is transient due to a pore-ripening process during which large pores appear at the expense of small ones to minimize the surface energy [10,33] — an effect that has also been observed in situ during an investigation of the replacement of KBr by KCl in an aqueous solution [29]. Note, however, that the solution data shown in Fig. 3 indicate that the reaction would have proceeded after 336 h even though the silica rim is already as thick as 900 µm. It is important to emphasize that the porosity is a direct function of the difference between the solubility of the borosilicate glass and the silica product in the solution as well as the potential difference of their molar volumes, since the silica covers the same volume as the original glass, i.e., the shape and dimension of the original glass cuboid are retained. In the present case, however, the molar volume of the silica product (Vm ≈ 27.3 cm3/mol, assuming a density of 2.2 g/cm3) roughly equals the molar volume of the borosilicate glass (Vm = 27.2(2) cm3/mol) and the porosity must thus stem mainly from the solubility difference between silica and borosilicate glass in the acidic solution. The solubility difference can be understood by considering the borosilicate glass as a complex solid solution of oxides and metasilicates. According to the thermodynamics of solid solution– aqueous solution systems, an aqueous solution, which is saturated with respect to a solid solution due to its congruent dissolution, is in equilibrium with a solid solution that is composed of those end member(s) that has(have) the lowest solubility in the solution [34]. In our experiments, amorphous SiO2 and TiO2 (rutile) are the end members with the lowest solubility of all end members in solution. It follows that the degree of porosity in the silica rim largely depends on the glass chemistry (determining the solubility difference between the glass and the silica rim), which can explain why different glasses alter with different rates even under identical physico-chemical conditions [4,24]. It is interesting to note here that Munier and coworkers [21] modeled the composition of the experimental corrosion layer that formed around two borosilicate glasses in pure water at 90 °C by

assuming that it formed by the precipitation of an ideal solid solution at equilibrium with respect to the solution after congruent glass dissolution and found a very good agreement between model and experiment. However, the authors still considered the silica layer, in accordance with the common view, as being “a residual reorganized hydrated glass”. 4.2. Origin of chemical oscillations in the silica corrosion product Many naturally altered, archeological silicate glass artifacts show similar non-equilibrium banding patterns to those observed in the experiments presented here (Fig. 2g) [35,36], indicating that our results obtained from experiments with a strongly acidic solution at a relative high temperature may also be relevant to conditions prevailing on the Earth surface. In any case, a mechanistic model adopted to explain silicate glass corrosion must also be able to explain the observed chemical oscillations (Figs. 2e, f, and 4). However, the occurrence of chemical self-organization in naturally altered glass has completely been ignored in previous development of glass corrosion models. Most models are based on the assumption of equilibrium between the glass and solution [3] or between the “gel” layer and solution [19]. This is somewhat surprising because observations on naturally altered glasses have been considered as being most significant for the evaluation of the long-term stability of waste glasses in a geological repository [37]. Although a detailed analysis of the intriguing periodic pattern is beyond the scope of this contribution, a reasonable, although in the first instance qualitative, explanation can be given. The key observation is the existence of an outer silica rim that does not show chemical oscillations (an observation that is further evidence against a model that is based on in situ reorganization/recondensation of a hydrated glass network, since such a mechanism is expected to produce, if at all, oscillations throughout the whole corrosion zone). Once this silica rim has reached a critical thickness (∼20 µm in our experiments), the gain of glass constituents to the interfacial solution by congruent glass dissolution at the interface becomes faster than the loss of elements to the surrounding solution by diffusion through the porous corrosion layer. This leads, in the present experiments, to the supersaturation of a silica phase in the interfacial solution that is relatively enriched in Ce and Al, both which have a lower solubility in acidic solutions than Na, Ca, Li, B, or Mg. The subsequent precipitation of such chemically enriched silica reduces the interfacial solution in Ce and Al, which again stabilizes chemically almost pure silica. The repetition of such a process generates alternating bands of different composition along an inward moving dissolution–reprecipitation front. The critical thickness should depend on the composition of the glass as well as on the aqueous environment (e.g., pH, temperature etc.), as these factors determine the solubility and molar volume difference between the glass and the silica product and thus the permeability/porosity of the corrosion zone. The bifurcations seen in Fig. 2e can be explained by fluctuations of the reaction rates along the reaction front. The observation that the network modifiers Li+, Na+, Ca2+, Mg2+ as well as the network former B3+, all of which are highly soluble in most aqueous solutions, have the lowest concentration in the silica product phase suggests that the solubility of different cations (or end members) in the interfacial solution determines the partitioning of the cations inside the precipitating silica, as predicted by the thermodynamics of solid solution–aqueous solution systems, rather than differences in the diffusivity of the various cations through the corrosion zone. 5. Concluding remarks Although many observations made in this study at an initial pH of 0 were also made in experimental studies with glasses under different temperature and pH conditions, as discussed in previous sections, we do not yet have experimental evidence that the corrosion of silicate glasses in neutral and basic solutions also proceeds via the interface-coupled

T. Geisler et al. / Journal of Non-Crystalline Solids 356 (2010) 1458–1465

dissolution–reprecipitation mechanism that is proposed here. In this respect it is important to note that glass dissolution, i.e. the release of orthosilicic acid into solution, depends on the nucleophilic character of water and is thus preferred in basic media. This means that an interfacecoupled dissolution–reprecipitation mechanism should be rather favored under more basic conditions. Therefore, the new model potentially offers a consistent thermodynamic and mechanistic framework to understand a number of observations made in numerous glass corrosion studies under a wide range of physico-chemical conditions, such as the occurrence of (i) a sharp structural and chemical interface between glass and the “gel” layer, including the frequently observed gap at the glass–silica gel interface, (ii) porosity in the corrosion product, (iii) non-equilibrium textures in naturally altered glasses, as well as the observation of (iv) constant and relative fast long-term corrosion rates even under silica saturated conditions [7,18], and of (v) different alteration rates of chemically different glasses [4,24]. The latter observation reflects the fact that the degree of porosity, which has a strong control on the reaction kinetics as also pointed out by other authors [8–11,17], is governed by the solubility difference between the glass and the corrosion product, which, in turn, is largely determined by the composition of the glass. Knowledge about the actual corrosion mechanism is the most important ingredient for computer simulation of glass corrosion [32]. With recent advances of first-principles or semi-empirical computer codes, computer simulations will likely become an important tool for predicting the experimentally non-accessible long-term corrosion of glasses. A mechanistic model, however, that is based on the coupling between congruent dissolution and reprecipitation will most likely yield different long-term predictions than current models that describe borosilicate corrosion in terms of diffusion-controlled parallel or subsequent reaction steps, even though the models may equally well fit a given set of short-term experimental data. Based on the fundamental difference between a diffusion-controlled and an interface-controlled corrosion process, it seems to be important to investigate whether the proposed corrosion model can be transferred to glass corrosion under more moderate physico-chemical conditions prevailing at the Earth's surface. This question appears to be critical for the prediction of the long-term performance of nuclear waste glasses in a nuclear repository. Acknowledgement We would like to thank V. Rapelius and B. Mihailova for their help with the ICP-OES and IR analyses, respectively. We are further grateful to U. Bröcker for preparing the numerous glass cuboids, E. Winter for

1465

giving the permission to use the BSE image of the archaeological glass shown in Fig. 2g, and an anonymous reviewer for helpful suggestions. TG further acknowledges financial support from the German Research Society in form of a Heisenberg Scholarship (GE1094/12-1).

References [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] [18] [19] [20] [21] [22]

[23] [24] [25] [26] [27] [28] [29] [30] [31] [32] [33] [34] [35] [36] [37] [38]

B. Grambow, Elements 2 (2006) 357. P. Frugier, et al., J. Nucl. Mater. 388 (2008) 8. R.H. Doremus, J. Non-Cryst. Solids 19 (1975) 137. A. Paul, J. Mater. Sci. 12 (1977) 2246. B. Grambow, Mater. Res. Soc. Symp. Proc. 44 (1985) 15. W.H. Casey, B. Bunker, in: M.F. Hochella, A.F. White (Eds.), Rev. Mineral., 23, 1990, p. 397. P. Frugier, S. Gin, J.E. Lartigue, E. Deloule, Mater. Res. Soc. Symp. Proc. 932 (2005) 305. C. Jégou, S. Gin, F. Larché, J. Nucl. Mater. 280 (2000) 216. O. Deruelle, O. Spalla, P. Barboux, J. Lambard, J. Non-Cryst. Solids 2261 (2000) 237. D. Rebiscoul, et al., J. Nucl. Mater. 326 (2004) 9. C. Cailleteau, et al., Nat. Mater. 7 (2008) 978. R. Prochazka, J. Non-Cryst. Solids 352 (2007) 2052. S. Gin, P. Jollivet, J.P. Mestre, M. Jullien, C. Pozo, Appl. Geochem. 16 (2001) 861. J.-C. Dran, G. Della Mea, A. Paccengella, J.-C. Petit, L. Trotignon, Phys. Chem. Glasses 29 (1988) 249. K. Ferrand, A. Abdelouas, B. Grambow, J. Nucl. Mater. 355 (2006) 54. J.-L. Crovisier, T. Advocat, J.-L. Dussossoy, J. Nucl. Mater. 321 (2003) 91. S. Gin, C. Jégou, P. Frugier, Y. Minet, Chem. Geol. 255 (2008) 14. E. Vernaz, S. Gin, C. Jégou, I. Ribet, J. Nucl. Mater. 298 (2001) 27. W.L. Bourcier, D.W. Pfeiffer, K.G. Knauss, K.D. McKeegan, D.K. Smith, Mater. Res. Soc. Symp. Proc. 176 (1990) 209. B. Grambow, R. Müller, J. Nucl. Mater. 298 (2001) 112. I. Munier, J.-L. Crovisier, B. Grambow, B. Fritz, A. Clément, J. Nucl. Mater. 324 (2004) 97. B.P. McGrail, J.P. Icenhower, D.K. Shuh, P. Liu, J.G. Darab, D.R. Baer, S. Thevuthasen, V. Shutthanandan, M.H. Engelhard, C.H. Booth, P. Nachimuthu, J. Non-Cryst. Solids 296 (2001) 10. N. Donzel, S. Gin, F. Augereau, M. Ramondam, J. Nucl. Mater. 317 (2003) 83. C. Cailleteau, C. Weigel, A. Ledieu, P. Barboux, F. Devreux, J. Non-Cryst. Solids 354 (2008) 117. T. Stephan, Planet. Space Sci. 49 (2001) 859. D.L. Parkhurst, C.A.J. Appelo, U.S. Geological Survey Water-Resources Investigations Report, 99–4259, 1999, p. 310. G.D. Garlick, J.R. Dymond, Geol. Soc. Amer. Bull. 81 (1970) 2142. A. Putnis, Mineral. Mag. 66 (2002) 689. C.V. Putnis, K. Tsukamoto, Y. Nishimura, Amer. Mineral. 90 (2005) 1909. T. Geisler, P. Pöml, T. Stephan, A. Janssen, A. Putnis, Amer. Mineral. 90 (2005) 1683. A. Putnis, C.V. Putnis, J. Solid State Chem. 180 (2007) 1783. K.G. Knauss, M.J. Dibley, W.L. Bourcier, H.F. Shaw, Appl. Geochem. 16 (2001) 1115. A. Ledieu, F. Devreux, P. Barboux, J. Non-Cryst. Solids 345/346 (2004) 715. F. Lippmann, N. Jahrb. Mineral., Abh. 139 (1980) 1. A. Silvestri, G. Molin, G. Salviulo, J. Non-Cryst. Solids 351 (2005) 1338. J. Sterpenich, G. Libourel, Chem. Geol. 174 (2001) 181. P. Ball, Nature 421 (2003) 783. T.A. Guiton, C.G.I. Pantano, Colloids Surf. A74 (1993) 33.