Does Water Cleaning Mitigate Atmospheric Degradation of Unstable Heritage Glass? An Experimental Study on Glass Models

Abstract

Glass curators often question how their treatments affect the long-term stability of historical glass. While damp cotton swabs are commonly used to remove surface salts and dust, the use of water remains controversial, particularly for heavily altered glass, due to concerns about worsening hydration. This study investigates the effect of water rinsing on an unstable soda-lime glass altered for six months (monoliths) and fifteen months (powders) at 35 °C and 85% relative humidity. Samples were then rinsed with Milli-Q water at 20 °C or 50 °C, and the monolithic glass was subsequently subjected to an additional 15 months of alteration under the same conditions. The glass surface was characterized by optical and scanning electron microscopy (SEM) as well as Raman spectroscopy to identify the nature of the salts. The evolution of the hydrated layer was assessed using transmission FTIR, Raman and solid-state NMR spectroscopies, ToF-SIMS, and thermogravimetric analysis (TGA). The results show that rinsing effectively removes surface salts—primarily sodium carbonate—and induces structural changes in the hydrated layer, promoting silicate network polymerization. Upon resuming alteration, rinsed monolithic samples exhibit no further degradation after the additional 15 months of alteration. These findings offer promising insights for conservation practices and may help curators refining their treatment strategies for altered glass.

1. Introduction

Because silicate glass bears relatively high concentrations of alkalis used as fluxes, this material is chemically sensitive to water. Depending on the glass composition and environmental conditions, glass objects of cultural heritage may degrade over time in atmosphere and require restorative interventions [1,2,3,4,5]. Depending on the museum collections and experimental tools to detect alteration, about 5% to 60% of the glass objects are assessed in active state of degradation [4,6,7,8]. The visible, macroscopic signs of atmospheric degradation of glass are the presence of white alkali-bearing salts (alkali- formates, sulfates, acetates, carbonates, etc.) in the first stage and the fracturing of the glass surface in the second stage. The latter phenomenon has been called “crizzling” since the 16th century and is caused by the hydration and drying of a surface layer. The compositions the most affected by these phenomena are low-lime, high-alkali glasses, many of them having been produced during the 16th–19th centuries. These compositions are often qualified as unstable or unbalanced [1,5,9,10].

The primary strategy to fight this glass degradation in museums is preventive: it relies on the control of the environmental conditions in terms of temperature, relative humidity (RH), and pollutants [4,11,12,13]. Air movement in the showcases to destroy microclimates is highly recommended. In addition to these general rules, it is advised to wash “all glasses that can be safely washed, including glasses that have already been identified as crizzling” [4]. The washing allows to restore the original appearance of the object when degradation has not reached the crizzling stage. Moreover, removing the alkali salts and other hygroscopic dirts reduces the amount of water adsorbed on the glass surface, and it reduces the local pH. For cultural heritage as well as commercial glass conservation, these measures are logically considered as favorable to slow down the degradation because the silicate network is better protected in these conditions [14,15]. However, no detailed case study of the impact of washing over time has been reported in the literature to the best of our knowledge.

Refs [5,16,17,18,19,20,21,22] and their quoted references depict a detailed historical overview of the glass treatment diversity, of which first attempts to conserve glass started between the 19th and 20th centuries. However, there is no universal method for cleaning glass objects, and it can differ depending on the curator’s practice. Different solvents have been used for the cleaning of the altered glass surface: water, water/ethanol, ethanol, and acidified water [4,5,11,12,20,23,24,25,26]. Among these treatments, water is indeed a simple, eco-friendly, and non-hazardous treatment for human health.

The aim of the present study is to complete the knowledge about the effect of water rinsing on an altered glass at a molecular scale. To this purpose, a model glass was altered in the laboratory in accelerated but close-to-ambient conditions, and the impact of the water rinsing was investigated using complementary techniques. The chosen glass is a sodium silicate glass poor in alkaline earths (~74 wt% SiO₂, ~19 wt% Na₂O, and ~6 wt% in CaO + MgO) and therefore with an unstable character against atmospheric alteration. This composition is close to an original Venetian glass artefact from the Victoria and Albert Museum (~72 wt% SiO₂, ~18 wt% Na₂O, and ~4 wt% in CaO + MgO) [8].

The glass was prepared as polished glass monoliths and calibrated powders for this study. After three months of accelerated aging at 35 °C and 85% relative humidity (RH), the surface of the glass monoliths was covered with numerous carbonate salts. After six months of aging, in addition to the surface salts, the fracturing of the surface was observed at the microscopic scale, with a delaminating hydrated layer of about 2.5 µm thickness. A simple protocol of rinsing the altered surface with pure water was designed and applied to the weathered glass monoliths and powders. The efficiency of water cleaning to remove the surface salts was investigated using optical microscopy (OM), scanning electronic microscopy (SEM), Raman spectroscopy, and X-ray diffraction (XRD). The impact of the cleaning on the hydrated silicate layer composition and structure was studied by time-of-flight secondary ion mass spectroscopy (ToF-SIMS) depth profiling, Raman, and Fourier-transform infrared spectroscopy in transmission mode (TR-FTIR). The calibrated powders were altered for 15 months to achieve complete alteration in the bulk of the glass particles. Their study by thermal gravimetry analysis (TGA) and solid-state nuclear magnetic resonance (¹H, ²³Na, and ²⁹Si MAS NMR) provided additional clues on the changes induced by water cleaning. The rinsing waters of the glass monoliths and powders were collected and analyzed by ICP-AES (not presented). In parallel, altered glass monoliths, either not cleaned as reference samples or cleaned with water, were put again in accelerated aging conditions for up to 15 months. The effect of the water cleaning on the continuation of the atmospheric degradation could be assessed and was spectacular: the alteration was considerably slowed down in the cleaned samples. Herein, we discuss these results considering the changes induced by the rinsing of the glass surface and their interest in the conservation of cultural heritage glasses.

2. Materials and Methods

2.1. Glass Preparation

The CaO-poor soda-lime silicate glass was produced by Saint Gobain Research in Aubervilliers, France, by melting 1 kg of reagent-grade raw materials (SiO₂, Al₂O₃, Na₂CO₃, and CaCO₃) at 1450–1550 °C for 3 h, followed by a quench of the melt on a metallic plate and an annealing step at 560 °C for one night. The glass composition was determined by ICP-AES analysis in Saint Gobain Research (Courbevoie, France) and by SEM-EDX in the laboratory and is given in

Table 1. Average compositions of the glass measured by ICP-AES and SEM-EDX. Glass transition temperature (Tg) measured by DTA and glass density measured by the Archimedes method.

Sample	SiO2	Al2O3	MgO	CaO	Na2O	Fe2O3	Tg (±4 °C)	Density (±0.002 g·cm−3)
wt% ICP-AES	74.2	0.87	0.8	5.29	18.8	0.03		2.465
wt% SEM-EDX	73.8	0.65	0.79	5.37	19.4	-	518
mol% ICP-AES	74.4	0.5	1.3	5.7	18.3	-

. The glass transition temperature was measured by differential thermal analysis (DTA), and the glass density was measured by the Archimedes method with water as the immersion liquid. The values are shown in the last two columns of Table 1.

Two kinds of glass samples were prepared: polished glass monoliths (noted GM) and calibrated glass powder (noted GP). Glass monoliths of dimensions 13 × 13 × 3 mm³ were cut from the cast and annealed glass block using a diamond saw and ground on a diamond plate with water as lubricant. Then, the pieces were polished on satin woven polishing cloth with a mixture of oil and ethanol as lubricant and diamond abrasive to a 1 µm finish. The glass powder, with a grain size between 20 and 63 µm, was washed with acetone in an ultrasonic bath several times to remove fine particles. These preparations were done by external companies.

Prior to aging, all the glass monoliths were washed to remove surface dust and organic contamination. First, the samples were immersed for approximately 20 s in a saturated sodium dodecyl sulfate (SDS) solution before being abundantly rinsed with about 10 mL of deionized water. Next, they were immersed in an ultrasonic bath containing absolute ethanol for 10 min before being vertically dried on pure cellulose paper. Any remaining droplets were gently blotted with the same paper. The surface that was not examined was marked with an engraved cross. Finally, the samples were stored in a compartmentalized polystyrene box and labeled on the box’s surface. Both before and after aging, all samples, whether monoliths or powders, were kept in a storage cabinet with silica gel. An RH sensor inside the cabinet allowed to check that the RH value never exceeded 50%.

2.2. Accelerated Aging Experiments

Table 2. Sample names, types and parameters of the aging experiment.

Name	Sample Type	Aging Parameters
		Duration	Temperature	RH
GM-3m	Glass monolith	3 months	35 °C	85%
GM-6m	Glass monolith	6 months	35 °C	85%
GP-15m	Glass powder	15 months	35 °C	85%

reports the sample names along with their type and aging parameters. Two accelerated aging tests were applied to the glass monoliths and one to the glass powder. All the tests were conducted at 35 °C and 85% relative humidity (RH) in an airtight PMMA box containing a KCl-saturated solution and placed in an oven. According to [27], the KCl saturated solution buffers the RH close to 83% at 35 °C. Practically, sensors placed inside the airtight box indicated stable conditions around 35 ± 0.7 °C and 85 ± 4 RH%. The duration of the experiment was 3, 6, or 15 months and is indicated next to the GM or GP letters in the sample name. For every duration, two or three glass monoliths were aged in the same conditions to increase the statistics and allow for different characterization techniques.

Two samples altered for 3 months (GM-3m) or 6 months (GM-6m) were engaged in the experimental study of the effects of water cleaning on the pursuit of the alteration. One sample was cleaned with water, and the other was the control sample. For these four samples, aging was resumed for 2, 5, 7, and 15 months in the same aging conditions (35 °C and 85 RH%). After every duration, the samples were taken out of the airtight box, characterized, and placed back in the box.

2.3. Water Cleaning Protocol

In this paper, water cleaning treatment is alternatively referred to as rinsing. No mechanical cleaning, such as using cotton wool soaked in water, was employed; instead, water was either poured onto the glass surface, or the powders were immersed in water. Figure 1 illustrates the protocol for both types of samples. Milli-Q water (18 MΩ) heated to 20 °C or 50 °C was used for rinsing, and the rinsing water was kept for subsequent ICP-AES analyses.

For glass monoliths, 20 mL of water was poured onto the surface using a 25 mL syringe, after which the samples were dried on pure cellulose paper in a vertical position to allow any remaining water to be absorbed by the paper. Then, the samples were stored in a compartmentalized box while awaiting characterizations. For glass powders, 750 mg to 1 g of powder was placed in a 15 mL centrifuge tube, and 10 mL of milliQ water was added. The mixture was then centrifuged at 9000 rpm for 10 min to separate the powder from the water. The rinsing water was collected for ICP-AES analysis, and the wet powder was air-dried next to silica gel to absorb the residual humidity.

2.4. Characterization Methods

The following paragraphs describe the tools and experimental conditions used to characterize the samples. Because the glass composition is unstable, the storage duration between the end of an aging experiment and the characterization time is an additional parameter that may influence the characterization results. We decided to study thoroughly the impact of water cleaning on the glass surface composition and microstructure only after we discovered the striking effect of rinsing on slowing down the alteration kinetics. Therefore, some cleaning steps and comparative observations before/after cleaning were carried out after about 12 months of storage of the altered samples. This is the case for SEM-EDX examinations of glass surface and Raman spectroscopy study.

2.4.1. Morphological Studies

The surfaces of the glass samples were systematically observed before and after aging experiments using a Keyence VHX—5000 digital microscope (Keyence Corporation, Itasca, IL, USA) equipped with a CMOS camera featuring a 1/1.8-inch CMOS image sensor. Images were taken using the Z20 (×20 to ×200 for the magnification) and the Z500 (×500 to ×2000 for the magnification) objectives.

Furthermore, observations were conducted with a scanning electron microscope, either a FEG-SEM JEOL JSM-7800F (JEOL Ltd., Taiwan, China, operating under 10⁻⁴ Pa vacuum) or a W-SEM Hitachi (operating at 30 Pa). Typical acceleration tensions and working distances were used (5–15 kV and 5–10 mm). To observe the edge of the aged glass samples and evaluate the hydrated layer thicknesses, samples were cut into two pieces. The cutting was guided by a line engraved by using a diamond pen.

For SEM observations using the FEG-SEM JEOL instrument (JEOL Ltd., Taiwan, China), a minimum of 1 nm of platinum coating was deposited on the surface to prevent charge accumulation. Most of the images shown in this paper were obtained with the Hitachi electron microscope to avoid this metallization step.

2.4.2. Salts Identification

To identify the alteration salts, Raman mapping was performed on the glass monoliths with a Renishaw Invia Spectrometer (Renishaw, Wotton-under-Edge, UK) coupled with a Leica optical microscope. An Nd-YAG green laser (ALPHALAS GmbH, Göttingen, Germany) (532 nm, 100 mW power) and a 100× objective were used, with data acquisition managed by WIRE software (version 4.2). The following acquisition parameters were used: mapsize 25 × 30 µm², step 1 × 1 µm², confocal aperture and slit width 60 µm (standard mode), grating 1800 lines/mm, laser power about 100 mW, and 10 s acquisition time per point. With this grating and slit width, the spectrometer configuration gives a spectral resolution of about 3 cm⁻¹, and the spectral range on the CCD detector was 100–1800 cm⁻¹. Two mappings were conducted per sample. The Empty Modelling algorithm of WIRE was employed to explore the variance of the dataset as to not miss any spectral component present in the map area. When all spectral features were identified, colored maps were computed by integrating the intensity to the baseline of the main peaks. The color code was defined between 5% (black) and 95% (saturated color) of the maximum peak-to-baseline area.

X-ray diffraction (XRD) was carried out for powders, using a Panalytical X’Pert Pro equipment (Thermo Fisher Scientific, Waltham, MA, USA) with a Cu X-ray source and a 10–70° range in 2θ. The 2θ step was 0.023°, and the total acquisition time was 3 h. The peaks’ attribution was carried out by consulting the ICDD database with the help of Highscore software, version 5.2. XRD analyses of the glass monoliths surfaces were attempted, but due to the small quantity of salts, XRD was not sensitive enough to identify them.

2.4.3. Structural Studies

The structure of the hydrated layer of glass monoliths was characterized by Raman spectroscopy. In this case, the spectra were acquired in static mode, with a spectral range of 150–1500 cm⁻¹ corresponding to the vibrations of the silicate network or 2500–3800 cm⁻¹ corresponding to the stretching vibration of hydroxyl bonds. The measurement conditions were as follows: grating 1800 lines/mm, confocal aperture of 20 µm (high confocality mode), laser power of about 50 mW, and 20 s of acquisition, repeated 10 times. The two ranges were successively measured to stay in the same position on the surface. Three measurements at three different locations were collected on each sample. To reduce noise, each spectrum is the average of the three measurements. The resulting spectrum was leveled to a zero absorbance at 1300 cm⁻¹ before being normalized to the intensity of the 550 cm⁻¹ band.

Moreover, FTIR spectroscopy in transmission mode was performed using a Bruker Tensor27 FTIR spectrometer (Bruker, Billerica, MA, USA), with a 4 cm⁻¹ resolution and 64 scans for both samples and background. Depending on the hole diameter of the sample holder used, 1 or 2 measurements of each sample were done. All the spectra were leveled to a zero absorbance at 4000 cm⁻¹.

Pristine, aged, and aged-rinsed glass powders were studied by ¹H, ²³Na and ²⁹Si MAS NMR spectroscopy. MAS spectra were collected with Bruker Avance II (Bruker, Billerica, MA, USA) and Avance Neo IV spectrometers (Bruker, Billerica, MA, USA) operating at magnetic fields of 7.02 T (300 WB) for ²⁹Si NMR and 11.72 T (500 WB) for ¹H and ²³Na NMR, respectively. The powder was packed in 4 mm outer diameter ZrO₂ rotor and spun from 10 kHz (7.02 T) to 12.5 kHz (11.72 T) using Bruker CPMAS probes (Bruker, Billerica, MA, USA). As explained later, after a few NMR measurements, the powders were heat-treated in the opened rotor at 150 °C for 5 to 20 min to remove physically sorbed water. The ¹H MAS NMR spectra were acquired with a recovery delay of 4 s (after a pre-saturation made of a train of twenty 90° pulses with an interpulse delay of few ms), which was the optimal value maximizing signal intensity (no change in line shape was observed for longer recycle delays). The ²³Na MAS NMR spectra were acquired using a single short non-selective pulse (1 μs, tip angle of about 20°) to obtain quantitative spectra, with a recycle delay of typically 1 s. The ²⁹Si MAS NMR spectra were measured with a recovery delay of 8 s after a pre-saturation (no change in line shape was observed for longer values). The ¹H chemical shifts were referenced with respect to an external tetrakis(trimethylsilyl)silane (TKS) sample at +0.2 ppm relative to tetramethylsilane (TMS). ²³Na and ²⁹Si chemical shifts were referenced to a 1 M NaCl solution and to an external TKS sample (−9.9 ppm for the most intense line relative to TMS), respectively.

2.4.4. Chemical Analysis of the Alteration Layer

When the hydrated layer was too thin to be measured by SEM, its thickness was estimated with time-of-flight secondary ion mass spectrometry (ToF-SIMS) according to the method described in ref. [28]. Also, these ToF-SIMS depth profiles allowed to assess the composition of the hydrated layer. Analysis was carried out with a ToF-SIMS V instrument (Ion-ToF company, Münster, Germany), using an O²⁺ ion beam (2 keV, 100 nA) for in-depth abrasion, a Bi+ ion beam (25 keV, 0.3 pA) for the analysis, and an unfocused electron beam of some eV for the charge neutralization. The crater was 300 × 300 µm² in size, and at its center, the analysis zone was 100 × 100 µm². The sputtering rate was determined by measuring the crater depth with mechanical profilometry and was considered constant over the whole analyzed depth. It was 1.4 ± 0.04 nm/s and 0.9 ± 0.04 nm/s for the not-rinsed GM-6m and rinsed GM-6m samples, respectively.

Thermogravimetric analysis (TGA) was carried out on the totally altered glass powders (GP 15m) using a NETZCH STA449 F3 Jupiter equipment (NETZSCH, Selb, Germany), to measure the weight loss during the heating. About 80–100 mg of powder was put in an alumina crucible and heated from 20 °C to 1000 °C with a rate of 10 °C/min under nitrogen flux. The reference was an empty alumina crucible. The TGA measurements were corrected from a baseline separately measured using two empty alumina crucibles.

3. Results

In a first paragraph, the impact of water cleaning on the altered glass surfaces is described based on the comparison between the unrinsed and the rinsed monoliths. In a second paragraph, the consequence of the cleaning on the qualitative kinetics of degradation is examined thanks to the experiments resuming the accelerated aging after the cleaning step.

3.1. Modifications Induced by Water Cleaning on the Altered Glass Surface

3.1.1. Microscopic Changes of the Surface State and Distribution of Salts

• Monolith samples

After 3 months of aging, salts appeared on the surface of the glass monoliths. Two main morphological groups were observed by OM: very small crystals with needle morphology and bigger crystals with flat and spherical shapes. After 6 months of aging, salt accumulation was more significant on the surface, and a network of cracks appeared (crizzling) (Figure 2a). The first rinsing step was carried out with milliQ water at 20 °C. This rinsing treatment was partially effective in removing salts from GM-3m and GM-6m samples. However, some residual salts remained undissolved on the surface. Therefore, a second rinsing was deliberately performed using milliQ water heated to 50 °C to enhance salt removal. OM observations showed that this warmer rinsing step did not remove more salts than the initial 20 °C rinsing. Thus, the remaining salts on the surface are not soluble in water.

These GM-3m and GM-6m (rinsed and not rinsed) were then aged in the same atmospheric conditions at 35 °C and 85% RH (Figure 1). The corresponding results are presented below in Section 3.2.

The following section presents only the results of the GM-6m samples obtained from a single 50 °C rinsing procedure, as presented in Figure 2, to assess the modification induced by the water cleaning on the GM-6m glass.

The delimitated rectangular area outlined in black and blue in the OM images depicts the area examined with SEM (Figure 2a,b,d,e). Figure 2c,f were taken from a different area of this glass monolith not visible in the OM images. These images show the surface state of the glass beneath the flakes before and after rinsing. Because the sample was mechanically cut into two pieces, some flakes detached from the surface and revealed the glass surface below them. On the top of flakes, the two morphologies of salts are present, unlike below the flakes, where only tiny needle-shaped ones are present (Figure 2b,c). Because the cracking and the flakes were produced during the drying and cooling down of the samples at the end of the accelerated aging step, these tiny salts below the flakes likely formed during the storage that lasted about 12 months. After being water-cleaned at 50 °C, only the little needle-shaped salts were removed from the surface. Figure 2f suggests that some print of these salts remained on the surface. The bigger flat salts could not be dissolved in water even at 50 °C.

The salts were identified using Raman spectroscopy (Figure 3). Two Raman maps per sample were observed to confirm the results. Only one is presented here for the GM-6m samples.

Before rinsing (Figure 3a), different types of salts could be identified. Calcite (red color, with major peaks at 1084 cm⁻¹, 711 cm⁻¹, and 279 cm⁻¹ [29]) is preponderant on the surface. The intense peaks at 1060 cm⁻¹ and 1069 cm⁻¹ were assigned to the symmetric stretching of the carbonate group in trona (Na₃(HCO₃)(CO₃)₂·2H₂O) (green) and pirssonite (Na₂Ca(CO₃)₂·2H₂O) (pink), respectively, based on the RRUFF database (R050228 for trona and R070192 for pirssonite) [30,31] and on their detection in the XRD diagrams of the altered powder (Figure 4 and Figure 5). Calcium sulfates were also found (1025 cm⁻¹, assigned to CaSO₄ anhydrous III [32]) with their band always linked with trona bands. Another minor species with its main peak at 1000 cm⁻¹ suggests another sulfate (yellow). Because the Raman analysis was carried out 12 months after the end of the aging, sulfate compounds had time to form through dissolution of atmospheric SO₂ in the water film adsorbed on the altered glass surface and through reaction with the basic species (Na⁺, Ca²⁺, and OH⁻) in equilibrium with the basic carbonates like pirssonite. These sulfate compounds could accumulate over time at the expense of the carbonates because they are more insoluble than the carbonates. This is observed in the “dry” weathering of stained glass [33]. The association of sulfates with trona is attributed to the more acid character of the sulfate surrounding region (sulfate and trona are more acidic salts than thermonatrite Na₂CO₃·H₂O, or pirssonite). After 50 °C water rinsing, only calcite remained on the surface.

Note that XRD analyses were also performed on this altered glass monolith before rinsing. However, due to the low quantity of salts on the surface, the different peaks’ intensities were too low to identify the salts’ nature.

• Glass powders

The effects of rinsing the altered powder GP-15m were evaluated by SEM-BSE (Figure 4). After 15 months of aging, cracks and salts appeared on the particles’ surface. The impact of the rinsing with 50 °C water is barely visible in these images. As already seen with the glass monoliths, only the tiny salts were removed from the surface. Note that the GP-15m powder was stored only for one month between the end of the aging and the SEM and XRD study of the water cleaning.

The XRD diagrams of the GP-15m powder, not rinsed and rinsed, are displayed in Figure 5. Calcite and pirssonite were identified unambiguously in the powder before rinsing. After rinsing, the pirssonite disappeared, the calcite remained, and aragonite (CaCO₃) appeared.

3.1.2. Evolution of the Hydrated Silicate Network

Monolith samples GM-6m samples were characterized by IR spectroscopy in transmission mode, both before and after rinsing, just before being put again in an airtight box to resume the accelerated aging. The spectrum of the pristine glass is added for comparison. These spectra are shown in Figure 6.

The pristine glass shows two bands at 3500 cm⁻¹ and 2780 cm⁻¹, corresponding to the stretching vibration mode of O-H groups involved in a distribution of hydrogen bond strengths with neighboring oxygens, from weakly H-bound at 3500 cm⁻¹ to tightly H-bound at 2780 cm⁻¹ [34,35,36]. The extinction coefficient of this IR absorption band increases considerably with the strength of the hydrogen bond so that the absorbance profile is not related to the H-bonds distribution profile in the glass [36,37]. For GM-6m (black line), the absorbance of these two bands increases, and the absorbance in the 3400–3300 cm⁻¹ domain grows, with additional distinctive contributions at 3580 cm⁻¹ and 3360 cm⁻¹. According to [38], the 3400–3300 cm⁻¹ absorbance is typical of “liquid-like” adsorbed water on the surface or within the hydrated layer. In their study of hydrous sodium tetrasilicate glass, Zotov et al. observed a narrow band at 3580 cm⁻¹ that they proposed to assign to H₂O solvating Na⁺ ions because a similar band occurs in analcime and pollucite, which are two zeolite compounds bearing solvated alkalis [36]. In our experiments, this band disappeared with the rinsing, which tends to confirm this attribution. The narrow band in the 3400–3330 cm⁻¹ range possibly relates to structural water in the hydrated carbonates such as pirssonite (band at 3460 cm⁻¹ in [39]) and possibly trona (band at 3470 with shoulder at 3210 cm⁻¹ in [30]). The rinsing with water at 20 °C (purple line) decreased the absorbance a bit without changing the profile significantly. This glass monolith was additionally rinsed with 50 °C water (orange line). The 3580 cm⁻¹ band disappeared, and the absorbance in the 3400–3330 cm⁻¹ region decreased. A band at 3420 cm⁻¹ became more visible, which may be assigned to liquid water possibly trapped by capillarity in the rugosity and around calcite salts of the hydrated layer surface. Indeed, there were no bands exactly at this position in IR studies of similar hydrous silicate glasses [35,36,40,41], and the calcite salts were not hydrated.

The impact of water cleaning on the silicate network structure of the hydrated layer is studied by Raman spectroscopy in Figure 7. Figure 7c shows the 150–1500 cm⁻¹ region, and Figure 7d shows the 2500–3800 cm⁻¹ region. The optical microscopy image in Figure 7a shows a lot of salts on the altered glass surface, mainly calcite, as previously determined (Figure 2). Moreover, this glass monolith also presents some detached and removed flakes, showing the glass below, but these are not visible on the optical microscopy in Figure 7a. The presented point analysis was carried out on areas with no salts and where the altered glass surface was visible, as in the black-circled area in Figure 7a. The analysis was also performed on the glass below the flake, and the spectrum (not shown) was identical to the Raman spectrum of the pristine glass (black dotted line). After the rinsing step, Raman analysis was carried out on the top of one flake (blue circle) and on the glass below (red circle). The Raman spectrum of the glass below the flakes (red line) was also identical to the Raman spectrum of the pristine glass, which tells us that the flakes, both before and after the rinsing step, represent all the hydrated layer and cover the pristine glass.

In Figure 7c, the Raman spectrum of pristine glass is characteristic of a sodium silicate glass close to the Na₂O-3SiO₂ = NS3 composition [42,43,44,45]. Note that according to the composition of the glass (Table 1), the non-bridging oxygen-over-Si ratio, i.e., NBO/Si, equals 0.67, as in NS3. This ratio corresponds to a silicate network composed of two-thirds Q³ units and one-third Q⁴ units if we assume a homogeneous distribution of the NBO over the SiO₄ units. The narrow and intense signal at 550 cm⁻¹ is attributed to symmetric stretching-bending motions of Si-O-Si in depolymerized regions (enriched in Q³ units). Weaker intensity between 300 and 500 cm⁻¹ is attributed to the same motions in polymerized regions (enriched in Q⁴ units). The depolymerized regions are indeed characterized by a lower mean Si-O-Si angle than the more polymerized ones, and there is an inverse relationship between the mean Si-O-Si angle and the bending frequency [44]. The signal at 790 cm⁻¹ is an internal bending mode of the SiO₄ tetrahedron, which is barely sensitive to the glass composition but is observed to shift at a slightly higher frequency with the porosity in hydrated silicate [46]. Then, the high-frequency envelope in the 900–1200 cm⁻¹ range corresponds to the Si-O stretching in the Qⁿ units. The less intense signal at 950 cm⁻¹ corresponds to Q² connected to Q¹, noted as Q^2-1 [45,47]. This band indicates that the distribution of NBO is not perfectly homogeneous, as expected in a glass, and moreover, the presence of divalent cations as alkaline earths in the studied glass promotes the formation of these Q² and Q¹ units. Finally, the broad and intense band at 1100 cm⁻¹ corresponds to the Si-O stretching modes of Q³ units linked to Q³ or Q⁴ units and is noted as Q^3-3 in Figure 7c.

After the 6-month aging, the spectrum of the GM-6m sample showed an increase in intensity in the 300–500 cm⁻¹ region, suggesting an increase in the polymerized components of the network. On the high-frequency range, a decrease in the 1100 cm⁻¹ band indicates that the Q³ units’ concentration decreases, whereas a 1030 cm⁻¹ band grows. This new band could be associated to Q^3-2 connections [45,47]. The initial 950 cm⁻¹ band seems to be shifted to 930 cm⁻¹, which may be related to more Q^2-1 with respect to Q^2-2 connections. According to [48], this band could also be assigned to symmetric stretching of HO-Si-ONa. The Si-O stretching mode of silanol bonds was expected at about 970 cm⁻¹ [46]. Its presence is not obvious here. At last, the narrow peak at 1350 cm⁻¹ was assigned to sodium formate salts. This peak grew during the 12-month storage period after the end of the aging. The formation of formates during this period is attributed to a contamination of the storage box with formic acid. After rinsing, this formate-associated peak disappeared. Overall, the 6-month aging induced a slight polymerization of the silicate network, mostly at the expense of the Q³ units. This is due to the migration of modifier ions (mostly Na⁺) to the surface, leaving behind silanol bonds that condensed to form bridging Si-O-Si bonds, repolymerizing the network. These aging-induced modifications have been well described in [8,46,49].

After water cleaning (blue line), the intensity in the 300–500 cm⁻¹ region further increased, while the intensity of the 1100 cm⁻¹ Q^3-3 band decreased. A distinct band at 970 cm⁻¹ appeared that can be attributed to the stretching of silanol bonds and/or the formation of Q^2-2 connections. The rinsing with water increased the polymerization of the network, with depolymerized areas becoming enriched in Q² units and silanols.

Moving to the 2500–3800 cm⁻¹ region (Figure 7d), several distinct bands are visible at 2750 cm⁻¹, 2840 cm⁻¹, 2940 cm⁻¹, 3290 cm⁻¹, 3430 cm⁻¹, and 3590 cm⁻¹. The bands at 2750 cm⁻¹, 2820 cm⁻¹, and 2940 cm⁻¹ may be attributed to stretching of C-H from Na-formate [8] among other organic contamination species. They were still present after rinsing and became larger. The band at 3590 cm⁻¹ was noticed in hydrous glass with a high water content in [43,50]. Its attribution is unclear, and we suggest that it could be related to the IR absorption band at 3580 cm⁻¹ assigned to Na⁺-solvating water molecules. The main intensity occurred around 3400 cm⁻¹, corresponding to O-H with an intermediate hydrogen bonding strength like liquid water. Note that the intensity profile of the Raman OH stretching band better reflects the distribution of H-bond strengths compared to IR absorption because the Raman scattering section of this mode is less sensitive to the H-bond strength. After rinsing, the intensity around 3400 cm⁻¹ increased a bit, and the 3590 cm⁻¹ band slightly decreased.

The ToF-SIMS depth profiles of the pristine glass and GM-3m samples before and after rinsing at 50 °C are displayed in Figure 8.

The GM-3m samples were selected for this analysis because their hydrated layer thickness is less than 1 µm, and thus, they can be entirely and accurately profiled using ToF-SIMS. The glass composition does not bear any K₂O, but traces of K⁺ ions penetrated the hydrated layer during the aging, probably due to the KCl-saturated solution used to buffer the RH. The affinity of K⁺ ions for the hydrated layer has been observed elsewhere [51]. Because K⁺ ions are not present in the alteration salts nor in the glass, their profile is useful to measure the thickness of the hydrated layer. By taking the half value of the maximum count in K⁺, the thickness of the hydrated layer is estimated at about 830 ± 20 nm in GM-3 sample and about 630 ± 20 nm after water cleaning.

The two reference depth profiles of the pristine glass are marked by very similar Si⁺/Si⁺_bulk, Al⁺/Si⁺_bulk, Ca⁺/Si⁺_bulk, and Mg⁺/Si⁺_bulk intensity ratios as expected, where Si⁺_bulk is the Si⁺ signal measured in the bulk at the very end of each profile. However, the Na⁺/Si⁺_bulk ratio is dissimilar by around 40%. It is well known that the Na⁺ ion extraction yield is naturally very high under the ionic beam. In the glass composition of this study, the Na₂O content was elevated (about 18 mol%). Therefore, the Na⁺ signal was saturated, so the Na⁺/Si⁺_bulk ratio is not reliable, probably explaining the discrepancy observed only for this ratio.

The normalized Na⁺, Ca⁺, and Mg⁺ depth profiles of the not-rinsed GM-3m sample were strongly modified with respect to the reference due to the migration of part of these ions to the surface, forming carbonate salts. As the analysis covered an area of 100 × 100 µm², some salts certainly contributed to the profile, although the area was chosen as much as possible outside the salts. The profiles put into evidence an accumulation of Ca²⁺ cations at the surface, probably partly in calcite crystals, and a similar accumulation of Mg²⁺ with two maxima: one at the surface and the other below the surface. This suggests the distribution of Mg²⁺ ions in surface salts and in a silicate layer just below the salts. Such a phenomenon has already been observed by authors (unpublished results), and for instance in reference [52]. The Na⁺ ions slightly decreased from the bulk to the surface, with this profile resulting from the addition of two contributions: Na⁺ in salts and Na⁺ in the hydrated layer. The Si⁺ and Al⁺ profiles are very similar, with a bump due to the screening effect of salts at the surface and the densification of the silicate network at the pristine glass interface due to the departure of Ca²⁺ and Mg²⁺.

After rinsing, Na⁺ was still present, with a pronounced step around 730 nm. Given that Na-bearing salts are efficiently removed by the water cleaning, we consider that the Na⁺ ions depicted in this profile are incorporated into the hydrated layer until ~730 nm and then into the pristine glass. Moreover, Ca⁺ was detected at the same count ratio values, which is probably due to the remaining calcite not removed by the rinsing with water. On the contrary, the other sputtered ions, Si⁺, Al⁺, and Mg⁺, had enhanced ionic intensity ratios after the rinsing. The Mg²⁺ cations were supposed to be concentrated in insoluble surface crystals (first peak at about 25 nm) and in the hydrated silicate layer just below (bump at 100–200 nm).

The rinsing did not modify this distribution, except for the enhancement effect. Because these three ions stemmed from the hydrated layer, the intensity enhancement could mean that the silicate network densified with the rinsing step. This hypothesis is coherent with the rinsing-induced shrinkage of the layer from about 800 to 600 nm. It is further supported by the sputtering rate as deduced from the measurement of the crater depth: a sputtering rate of 1.4 nm/s was calculated for the non-rinsed sample compared to 0.9 nm/s for the rinsed one.

Finally, SiOH⁺ and H⁺ ions depth profiles are plotted in Figure 8e,h. Their absolute intensity cannot be compared because both measurements were performed at different moments, with different levels of residual hydrated species in the ToF-SIMS chamber. Their profile did not change much before and after rinsing.

Considering the glass powder, after 15 months of aging at 35 °C and 85% RH, it was altered in the bulk, as revealed by NMR spectroscopy. This powder was rinsed with water at 20 °C or at 50 °C and was analyzed by TGA and solid-state NMR spectroscopy for comparison with the not-rinsed powder.

The three TGA curves of GP-15m powder in Figure 9 put into evidence three different zones for weight loss: 50 to 150 °C corresponds to the loss of physically adsorbed water or dehydration; 150 °C to about 600 °C is the temperature domain of both desorption of solvation water bound in the hydrated layer and dehydroxylation of silanols [53] (both are added in the term “bound water”); then, from about 600 °C to 800 °C, we observe the decarbonation of carbonate salts: pirssonite, and calcite for GP-15m or calcite only for GP-15m rinsed at 20 °C or 50 °C [39,54]. A little distinctive loss visible at 170 °C only for the GP-15m curve is attributed to the pirssonite dehydration [39,55]. After rinsing at 20 °C or 50 °C, pirssonite was removed from the grain surface, which is consistent with the XRD in Figure 5. Weight losses in percentage are presented in

Table 3. Weight loss contributions of free water, bound water (solvation water + silanols), and CO₂ from carbonates in wt%.

	% Free Water	% Bound Water	% CO2
GP-15m	13.6	12.6	4.6
GP-15m + 20 °C H2O	14.4	9.0	3.2
GP-15m + 50 °C H2O	13.5	7.2	3.2

. Each weight loss is calculated according to the following equations:

$${\%}_{\mathrm{free}\mathrm{water}}={\displaystyle \frac{100 -{\mathsf{\Delta }\mathrm{m}}_{(150 °\mathrm{C})}}{{\mathrm{m}}_{\mathrm{glass}(800 °\mathrm{C})}}}\times 100$$

$${\%}_{\mathrm{bound}\mathrm{water}}={\displaystyle \frac{{\mathsf{\Delta }\mathrm{m}}_{(150 °\mathrm{C}) }- \mathsf{\Delta }{\mathrm{m}}_{(600 °\mathrm{C})}}{{\mathrm{m}}_{\mathrm{glass}(800 °\mathrm{C})}}}\times 100$$

$${\%}_{{\mathrm{CO}}_2}={\displaystyle \frac{{\mathsf{\Delta }\mathrm{m}}_{(600 °\mathrm{C})}-{\mathsf{\Delta }\mathrm{m}}_{(800 °\mathrm{C})}}{{\mathrm{m}}_{\mathrm{glass}(800 °\mathrm{C})}}}\times 100$$

where $\mathsf{\Delta }\mathrm{m}$ is the %weight loss measured at the given temperature, and m_glass is the %weight left at 800 °C. The amount of physically adsorbed water, i.e., free water, is equal in the three powders as expected because this is related to the water film in equilibrium with the humidity in the ambient atmosphere of the laboratory. After alteration, the %bound water was 12.6%, close to values observed for other bulk altered glass powders in author’s previous studies [10,54] and close to the value reported for sodium silicate water glass after it was dehydrated by heating at 150 °C [53]. This weight fraction decreased to 9.0% and 7.2% after the rinsing step at 20 °C or 50 °C, respectively. This decrease has probably two contributions: (i) the rinsing step removed some Na⁺ from the hydrated layer, together with their solvating water shell; thus, bound water molecules disappeared by this way, and (ii) the rinsing step induced condensation of silanols that produced molecular water in free state, which was removed by the drying or 150 °C heating. Note that the decrease in bound water content is consistent with the transmission IR spectra of the glass monolith before and after rinsing (Figure 6), as the global absorbance of the OH stretching band decreased with the rinsing. On the other hand, as expected, the CO₂ content of the powder decreased with the rinsing, with no effect of the water temperature. The rinsing removed about 30% of the carbonates, which seems consistent with the qualitative observations of the monoliths surfaces by OM and SEM.

The GP-15m powder was studied by solid-state NMR before and after rinsing with pure water at 20 °C. Every powder sample was heat-treated at 150 °C for 5 min to 20 min before measuring the spectra to remove the free water signal and achieve a better description of the hydrogenated species chemically bound in the hydrated glass. The ²⁹Si NMR spectra were measured before and after this heat treatment to study the effect of the 150 °C drying step on the network structure. All the spectra are shown in Figure 10.

The ²⁹Si and ²³Na MAS NMR spectra of the pristine glass are typical of sodium silicate glasses. Indeed, the black ²⁹Si MAS NMR spectrum contains two broad components at −105 ppm and −92 ppm that can be assigned to Q⁴ and Q³ species, respectively [56], and the black ²³Na MAS NMR spectrum comprises a wide, featureless peak broadened by both the distribution of chemical shifts and the second-order quadrupolar broadening [57,58]. Note that in Figure 10, the black ²³Na MAS NMR spectrum is scaled with respect to the spectra of GP-15m powders to put into evidence the remaining pristine glass contribution in these latter spectra. The very low level of this contribution demonstrates that the GP-15m powder was altered almost completely.

In the not-rinsed GP-15m powder (red spectrum, continuous line), the two components Q⁴ and Q³ of the ²⁹Si MAS NMR spectrum still dominated, with similar relative proportions, and a third contribution appeared at a lower field, which can be assigned to Q² species. These contributions are better resolved than in the pristine glass spectrum, showing that the distribution of chemical environments for the tetrahedral species generates narrower NMR shift distributions. Moreover, the chemical shifts of the components shifted to more negative values at −108 ppm, −98 ppm, and −88 ppm for the Q⁴, Q³, and Q² species, respectively. These band positions are reported in hydration studies of crystalline sodium silicates [59] or in gels produced by acidic leaching of sodium aluminosilicate glasses [60]. According to first-principle calculations of NMR spectra of sodium silicate glass structural models [61], the decrease in the ²⁹Si chemical shifts is linearly correlated with the increase in the average Si-O-Si bond angle, which in turn reflects the polymerization of the silicate network. Thus, ²⁹Si MAS NMR spectral modifications indicate that a slight concentration of NBOs in the silicate network (appearance of Q² units) together with the enhancement of polymerized areas (negative shift of the positions) is induced by aging, in good agreement with the changes of the Raman spectra.

With the water cleaning at 20 °C, these structural features of the hydrated silicate network remain the same, although the relative band area of the Q⁴ units increased, and the positions shifted to slightly more negative values, suggesting further polymerization of the network. Again, this evolution is consistent with the Raman spectrum of the GM-6m rinsed sample (Figure 7).

The interpretation of the ²³Na and ¹H MAS NMR spectra is less straightforward for studying the changes induced by aging and by water cleaning because a heating treatment step at 150 °C was carried out, which modified the hydrated glass network structure to some extent. Notably, considering the comparison of the ²⁹Si MAS NMR spectra before and after the heating at 150 °C in Figure 10c, the heating induced a broadening of the Qⁿ bands (mostly Q² and Q³), meaning that the distribution of the Si-O-Si bond angle widened, and the chemical disorder increased. However, these spectra were not considerably modified by the rinsing treatment, which is meaningful.

The ²³Na MAS NMR spectrum of the aged, not-rinsed glass bears a narrow contribution visible through the sharp feature at −30 ppm, which disappeared with the rinsing. This contribution relates to an ordered environment, probably the Na⁺ ions in carbonates (as pirssonite) [54] because it disappears with the rinsing. Furthermore, this spectrum bears a main contribution corresponding to Na⁺ ions in the hydrated glass. In this study, we did not simulate the ²³Na MAS NMR spectra to extract values of δ_iso and C_Q. The transformation of the spectrum from the pristine glass (black line) to the aged, hydrated glass (red main line) is very similar to the results of a previous study on a mixed alkali silicate glass [28]. The δ_iso and C_Q values both decreased with hydration, revealing a less distorted coordination environment with slightly greater Na-O bond distances for Na⁺ ions in the aged glass. These changes are induced by the arrival of water molecules and the rearrangement of the Na⁺ coordination sphere, as described in the detailed study of the hydration of layered Na₂Si₂O₅ crystalline silicates [62]. After rinsing with water at 20 °C, the area of this main contribution only slightly decreased, which is a hint that the rinsing efficiently removes the Na carbonates, but it is less efficient for removing the Na⁺ ions embedded in the hydrated layer.

The ¹H MAS NMR spectra of the GP-15m powders are complicated due to numerous contributions, of which two are dominant. Similar spectra have been observed and described in detail in previous studies of atmospheric degradation of silicate glass [28,54]. The narrow peaks at about 1.5 and 3 ppm are assigned to isolated silanols and H-bonded silanols in silica gel-type regions [63]. The 150 °C heat treatment removed the physically sorbed water molecules whose ¹H MAS NMR signal is at 4–5 ppm. The remaining hydrogenated species were solvation water molecules (i.e., molecules of the Na⁺ solvation sphere or in strong electrostatic interaction with the silicate network) and H-bonded silanols. According to ref. [53], solvation water may be retained in the hydrated glass until temperatures as high as 450 °C. The ¹H chemical shift distribution divides into two broad components: one at 3–8 ppm and the other at 10–16 ppm. The former relates to water and silanols with “liquid water-like” H-bond strengths, while the latter reveals very strong H-bonds of solvation water molecules and silanols due to the negative charge of nearby NBOs [53,58,59]. This high chemical shift component is proof of the retention of Na⁺ ions that are still bound to NBOs in the hydrated silicate network (these Na⁺ form the main ²³Na MAS NMR peak of Figure 10b). This was further confirmed by our ¹H → ²³Na REDOR measurements (not shown), indicating stronger dipolar coupling, and thus shorter Na-H distances, between Na⁺ ions and the ¹H species contributing at +15 ppm.

With the water rinsing, additional narrow peaks at +6 and +8 ppm appeared, likely due to partial crystallization during the 150 °C heat treatment; otherwise, the two broad contributions were essentially maintained. The contribution at low field/high chemical shift related to strong H-bonds to NBO slightly decreased with respect to the other, which reveals an expected loss in Na⁺ and NBO in the rinsed hydrated glass, but this loss seems moderate. This finding is consistent with the ²³Na NMR spectra in Figure 10b, showing a limited decrease in the ²³Na main peak with rinsing and with the TOF-SIMS depth profiles in Figure 8f showing the retention of some Na⁺ amount in the hydrated layer after the rinsing.

To summarize, the Raman spectra of the hydrated glass flakes and the NMR and TGA study of the aged glass powder before and after water cleaning are together consistent. These three techniques indicate that water cleaning induces partial recondensation of the hydrated silicate network, which becomes slightly more polymerized and with less bound water (as silanols and Na⁺ solvation water). From these experiments, it is not clear to what extent the rinsing removes Na⁺ out of the hydrated glass. According to ²³Na MAS NMR spectra (Figure 10b) and to ToF-SIMS Na⁺ profiles (Figure 8f), some Na⁺ ions were still bound to the hydrated network in the same environment as before water cleaning. As a matter of fact, these changes induced by the water cleaning had a major impact on the continuation of atmospheric alteration, as described in the next paragraph.

3.2. Impact of Water Cleaning on Further Atmospheric Degradation of Altered Glass

3.2.1. Evolution of the Surface State

Figure 11 illustrates the evolution of the surface state of GM-6m samples during the 15 months of alteration, comparing those rinsed with water to those that were not. Similar effects are observed for GM-3m samples (not shown). After six months of initial aging, the GM-6m surface was covered with salts and displayed a crack network (see also Figure 1), representing the baseline condition for the resumed alteration experiments (with or without rinsing). In the not-rinsed samples, substantial changes arose with the further alteration in the 35 °C, 85%RH atmosphere. After five months, large flakes measuring several tens of micrometers appeared and detached from the surface (Figure 11a). On the contrary, the rinsed samples did not present any significant modifications visible at this scale, even after fifteen months (Figure 11b).

SEM-BSE images confirm these findings (Figure 12). Not-rinsed samples, after fifteen months, were severely degraded, with the presence of many stacks of flat crystalline salts lying on the surface (assigned to calcite according to their shape) and a dense network of deep cracks, forming flakes up to 100 µm length, some of which were detached. In comparison, rinsed samples showed a different morphology, with only smaller salt deposits and the initial formation of a crack network, without any detached flakes.

3.2.2. Evolution of the Alteration Layer Thickness

Cross-sectional SEM-BSE analysis revealed a hydrated layer distinct from the bulk material (Figure 13). Prior to resumed alteration, the hydrated layer of the GM-6m samples measured approximately 2.5 ± 0.5 µm (Figure 13a). After 15 months or resumed alteration, the not-rinsed sample exhibited an increased layer thickness of about 8.5 ± 0.5 µm (Figure 13b), with a noticeable salt deposit on its surface, consistent with the top-view observation (Figure 11). This indicates that the hydrated layer tripled in thickness. In contrast, the rinsed sample maintained a hydrated layer thickness of 2.5 ± 0.5 µm even after 15 months (Figure 13c), with only a minor salt deposit visible on top.

Figure 14 presents the FTIR transmission spectra of the GM-6m sample (both not-rinsed and rinsed) over 15 months of resumed alteration. For the not-rinsed samples, the spectra showed increasing absorbance at 5, 7, and 15 months (Figure 14a). Based on the assignments provided in Section 3.1.2, this increase reflects the progress of hydration involving additional water solvating Na⁺ (3580 cm⁻¹), structural water within hydrated carbonates (3360 cm⁻¹), and water and silanols with a distribution of H-bond strengths (3400 to 2880 cm⁻¹). From 7 months onward, the signal reached the maximum absorbance detection limit that prevents the quantification of the hydration increase.

In contrast, the absorbance of the rinsed sample remained almost constant, at the level of the rinsed sample at 0 months (blue curve, corresponding to the GM-6m sample just after the rinsing step) (Figure 14b). This observation indicates that water cleaning effectively limits further hydration during aging.

4. Discussion

Nowadays, most research on glass conservation focuses on environmental parameters and the optimization of preservation conditions. Curators routinely use water, water–alcohol, or mild acidic solutions to remove salts or dust from the surface [10,11,12,23,64,65]. However, the choice of treatment depends on the degree of alteration. In cases of significant crizzling, curators often avoid water, fearing it may worsen hydration. Moreover, they often express frustration over the lack of collaboration with scientists and call for more interdisciplinary efforts [18]. In this context, our study offers new insights into the effects of rinsing on glass conservation.

The alteration of the glass under accelerated ambient atmospheric conditions (here, 35 °C and 85%RH) produces an alteration layer made of a hydrated silicate layer and salts on the top. The two reactions in play are the acid–base reaction of NBO with hydrogen species (R.1 written with water as hydrogen species) and the silicate network hydrolysis (R.2) (

Table 4. Reactions occurring in the water film between the glass and the atmosphere. The continuous ingress of water into the glass due to reactions R.1 and R.2 measures the glass alteration progress. All the reactions of this table contribute to fixing the pH of the water film, which in turn determines the advancement and kinetics of reactions R.1 and R.2. The water film (H₂O_(w) and CO_2(w)) is constantly renewed due to the infinite reservoir formed by the atmosphere.

Reactions of the water with the glass	Si − O − Na + H2O(w) ↔ Si − OH + Na+ + OH− Si − O − (½ Ca) + H2O(w) ↔ Si − OH + ½ Ca2+ + OH− Si − O − Si + H2O(w) ↔ 2 Si − OH	(R.1Na) (R.1Ca) (R.2)
Reactions of the water with the atmosphere	H2O(g) ↔ H2O(w) CO2(g) ↔ CO2(w) CO2(w) + H2O ↔ H2CO3 H2CO3 + OH− ↔ HCO3− + H2O HCO3− + OH− ↔ CO32− + H2O	(R.3) (R.4) (R.5)
Precipitations in the water film	Ca2+ + 2 Na+ + 2 CO32− + 2 H2O ↔ Na2Ca(CO3)2·2H2O Ca2+ + CO32− ↔ CaCO3	(R.6) (R.7)

).

In the water film adsorbed on the glass surface, dissolved carbonate species are in equilibrium with atmospheric CO₂ (reactions (R.3) to (R.5)). Due to reactions (R.1Na) and (R.1Ca), the pH of the water film is basic, and thus, CO₃²⁻ anions are abundant (guidance values are pKa HCO₃⁻/CO₃²⁻ = 10.33 and pKa Si(OH)₄/Si(OH)₃O⁻ = 9.8 at 25 °C), precipitating carbonate salts as pirssonite and calcite. Because they come from a weak acid (H₂CO₃), these salts buffer the pH at slightly basic values: for instance, pH is about 8.3 at 25 °C for a film equilibrated with calcite and atmospheric CO₂, and it is higher for sodium carbonate salts because they are more soluble than calcite. The salts appear in the first stage of the alteration because the reactions (R.1Na) and (R.1Ca) are rapid. However, the kinetics of these reactions are limited by diffusion. For the water to further ingress into the glass, hydrolysis reaction (R.2) is necessary. This reaction is slow, but it is catalyzed by OH⁻ present in the film or in the hydrated layer that forms. As a case result, the glass of this study became hydrated over a thickness of 2.5 µm in 6 months and about 8.5 µm over 6 + 15 = 21 months, which corresponds to a linear hydration rate of about 0.4 µm/month in these alteration conditions. The linear rate is expected because the reaction controlling the kinetics of the water ingress is (R.2), which is a thermally activated surface reaction, of which the kinetics are linear with time.

In these alteration experiments, ToF-SIMS analyses and NMR data (¹H and ²³Na) showed that Ca²⁺ ions were more displaced to the surface (forming calcite) than Na⁺ ions, of which a significant amount remained embedded in the hydrated silicate layer. In other words, (R.1Ca) is more displaced to the right than (R.1Na). This behavior was already observed in our previous studies and attributed to the very low solubility of calcite (compared to sodium carbonates) and to the far more negative solvation enthalpy of Ca²⁺ (compared to Na⁺) [10]. Therefore, solvated Ca²⁺ ions diffuse to the surface, driven by the very negative chemical potential gradient imposed by the formation of calcite salts. On the contrary, part of the Na⁺ stays in the hydrated layer, connected to NBO or to OH⁻ (although this latter species has not been directly observed by the employed techniques).

The Raman and the ²⁹Si MAS NMR spectra both indicate that the hydrated silicate layer formed by the alteration is slightly more polymerized than the initial glass. It is well known in silicate sol–gel chemistry and in glass alteration studies in immersion conditions that the structure of the hydrated silicate layer changes with time, temperature, and pH conditions and also depending on its composition [51,66,67,68]. Indeed, according to the current understanding of glass alteration, this layer results from hydrolysis–condensation reactions within the hydrated silicate network or coupled dissolution–precipitation at the interface with the hydrated glass [69,70] so that Schalm proposed to call it the “transformed layer” to avoid confusion with the interdiffusion layer/hydrated glass [68]. In the present study, the dissolution of gaseous CO₂ followed by the precipitation of pirssonite and calcite salts induced an acidification of the water film that remained, however, slightly basic. These chemical conditions explain the slightly more polymerized state of the transformed layer, with respect to the initial glass.

Now, we discuss the effects of water cleaning. According to the rinsing experiments on the altered glass monoliths and powders, pirssonite (Na₂Ca(CO₃)₂·2(H₂O)) was completely dissolved, while most CaCO₃ remained on the surface. NMR data of the powders show that many of the Na⁺ embedded in the transformed silicate layer were maintained. Careful ICP-AES analyses of the rinsing solutions would be necessary to quantify the exact proportion of Na⁺ and Ca²⁺ remaining on the glass surface and the proportion dissolved. Our attempted analyses (not shown) could not accurately quantify the dissolved fraction of these ions, but they quantified aqueous Si and put into evidence no dissolution of Si at 20 °C and a very tiny dissolution at 50 °C. Globally, these observations comply well with the expected solubility ranking of the compounds: pirssonite > calcite ~ hydrated polymerized silicate (transformed layer).

Moreover, the rinsing triggered a structural evolution of the transformed silicate layer towards a further increase in the degree of polymerization. This was clearly shown by Raman on the glass monoliths and NMR on the glass powders as well as TGA (slightly less silanols in the rinsed powders). The probable shrinkage put into evidence in the ToF-SIMS depth profiles suggests that it is accompanied by a slight densification of the layer. This structural evolution is attributed to two factors. First, the rinsing saturated the transformed layer with water, which probably provided the conditions for the hydrolysis–condensation or dissolution–precipitation reactions to occur more rapidly and completely. Second, the pH of the contacting water decreased during the rinsing, both due to the removal of the basic salts and the addition of liquid water in excess (dilution effect). This pH decrease may have induced the observed slight enhancement of the polymerization degree through in situ condensation (reverse R.2) and/or dissolution–precipitation of silicate species.

Note that the detailed microstructure of the transformed layer was not investigated in this study. SEM-FEG observations with high spatial resolution would be valuable in studying the impact of the rinsing at the 10 nm to 1 µm scale and belong to the perspectives of this work.

As a matter of fact, the changes induced by the cleaning (notably the removal of basic salts and slight repolymerization of the hydrated layer, clearly observed in this study) are responsible for a considerable decrease in the alteration rate in atmospheric conditions (35 °C, 85%RH), as evidenced by the alteration resumption experiments. The slowing down was so pronounced that it may be claimed that water rinsing passivated the glass. The pH decrease in the surface water film certainly stabilized the silicate network against hydrolysis, slowing down the water ingress. Furthermore, the structural changes and densification may have reduced the diffusion coefficients of water and solubilized species through the hydrated silicate layer.

To our knowledge, the effect of water cleaning of altered glass, followed by resumed atmospheric alteration, had not been previously investigated. A somewhat similar case was reported for geopolymers, in which rinsing was used to improve the specific surface area [71,72]. Regarding glass, Fearn et al. used ToF-SIMS to study the cleaning effect using a surfactant (Synperonic N) diluted in de-ionized water on an alkali lime silicate glass sample altered for one week at room temperature and 40% RH. Before being analyzed by ToF-SIMS, all the samples were cleaned with dry ethanol to remove any corrosion products. The main results showed that the surfactant did not cause more glass deterioration [73]. Regarding water rinsing, a preliminary study by Alloteau et al. [10] included an experiment addressing the effect of rinsing on samples altered at 80 °C for a few hours or at 40 °C for a few days. They obtained FTIR spectra comparable to ours. After rinsing and resuming alteration for 24 h at 80 °C and 85% RH or for 22 days at 40 °C and 85%RH, the rinsed samples showed no further alteration, unlike the control samples that exhibited a significant increase in the OH band.

Verhaar et al. used ionic chromatography to identify the salts removed after cleaning by using a moistened cotton swab on about 300 glass objects of the Rijksmuseum glass collection [74]. Among the deteriorated glass objects, alkali formate, acetate, or carbonate salts were detected. It is very interesting to notice that in this extensive study, alkali carbonate salts were only detected when formates and acetates were in very low concentration or were not detected at all. Moreover, the detection of carbonate salts was associated with the worst cases of glass deterioration. From this, we infer that the presence of alkali carbonate salts on the glass surface is highly detrimental to glass conservation, probably because the water film in equilibrium with these carbonates is still pH-basic, contrary to the other salts (formates and acetates) that are more pH-neutral. Removing these alkali carbonates with water will surely have a considerable beneficial effect in slowing down the degradation, as shown in this laboratory study.

In current conservation practice, curators do not pour water over glass like we did in the present study. Instead, they use soft, water-soaked cotton and gently wipe the surface. This coupled chemical and mechanical action helps remove residual salts from the surface. Yet, this practice raises the following question: can it also modify the silicate network, as we observed experimentally? Could it decrease the basicity as efficiently and thereby mitigate hydrolysis processes? If so, this would warrant further investigation.

Additionally, curators often use alternative solvents such as ethanol or ethanol–water mixtures. Their effect on the altered glass surfaces would merit further investigation. Our study suggests that water alone may be the most efficient because it is the most polar solvent able to efficiently dissolve the ionic alkali salts. On the other hand, Kunicki reported that slightly acidic water could be used to facilitate the calcite removal since its solubility increases with lower pH values [5,75]. However, rinsing with acidic solutions may pose risks for glasses containing heavy metals, as suggested by studies on geopolymers [72].

From a future perspective, it would be valuable to explore how routine conservation treatments affect the glass at the structural level by working with different glass compositions. Finally, these experimental results should be tested under real conditions—such as in museum displays—within long-term monitoring frameworks.

5. Conclusions

In this study, the effects of water rinsing on six months altered soda-lime glass, both in monolithic and powdered form and altered at 35 °C and 85%RH, were investigated. The results demonstrate that water rinsing effectively removes soluble surface salts—particularly sodium-bearing carbonates—and promotes structural reorganization of the hydrated layer, notably favoring its polymerization. Upon resuming alteration, the rinsed monolith samples exhibited no further degradation, highlighting the potential of this simple treatment to improve the long-term durability of altered soda-lime glass. From a heritage glass conservation perspective, these results encourage the use of water-based cleaning practices for curators, especially when alkali carbonates are detected on their glass objects’ surfaces.