• Aucun résultat trouvé

4 Les aluminosilicates de sodium vitreux et fondus

4.2 Effet de l’aluminium

4.2 Effet de l’aluminium

4.2.1 Role of Al3+ on rheology and nano-structural changes of sodium silicate and aluminosilicate glasses and melts.

Charles Le Losq, Daniel R. Neuville, Pierre Florian, Grant Henderson et Dominique Massiot Manuscript soumis à la revue Géochimica et Cosmochimica Acta

Résumé :

Les alumonisilicates d’alcalins fondus sont des matériaux importants pour les domaines de recherche en science de la Terre et pour l’industrie verrière. L’aluminium influence les propriétés des verres et liquides, et ses effets dépendent fortement de sa concentration. Cet article présente une étude d’alumi-nosilicates fondus et vitreux compris dans le système Na2O-Al2O3-SiO2, contenant 75 mol% SiO2 et différents rapports Al/(Al+Na). La spectroscopie RMN MAS 1D du 23Na et la spectroscopie Raman indiquent toutes les deux que le rôle du Na+ change en fonction de la proportion d’Al2O3. Dans les compositions pauvres en aluminium, les distances interatomiques Na-O sont plus courtes, et le Na, en coordinence ∼ 6, joue un rôle de modificateur de réseau. Lorsque l’on remplace Na par Al, la spectro-scopie RMN MAS 1D du23Na et du29Si ainsi que la spectroscopie Raman montrent que les espèces Q3 se transforment en espèces Q4, et que les distances Na-O augmentent, en accord avec une augmentation de la coordinence du Na+. Ce dernier devient alors compensateur de charge de l’Al3+, qui est incor-poré en espèces Q4 dans les domaines les plus polymérisés du réseau (espèces Q4

(4Si)). Ces changements structuraux augmentent fortement la viscosité des aluminosilicates fondus, qui atteint un maximum lorsque Al/(Al+Na) = 0.5 (composition albite). En passant le joint tectosilicaté, où Al/(Al+Na) = 0.5, les spectres RMN 1D MAS de l’27Al montrent que de l’Al[5] est présent dans les verres. Les T g et les fragilité plus élevées des aluminosilicates peralumineux fondus par rapport à celles des tectosili-cates fondus indiquent que, proche de la T g, l’Al[5] favorise la connectivité du réseau, alors qu’à hautes températures, c’est une espèce transitoire permettant la diffusion des NBOs et BOs dans les liquides. Lorsque la température augmente, la formation d’Al[5] peut aussi permettre d’expliquer la dépendance à la température des capacités calorifiques des tectosilicates et aluminosilicates peralumineux fondus.

Role of Al3+ on rheology and nano-structural changes of sodium silicate

1

and aluminosilicate glasses and melts.

2

3

Charles Le Losq1, Daniel R. Neuville1*, Pierre Florian2,3, Grant Henderson4 and Dominique Massiot2,3

4

5

1 Géochimie & Cosmochimie, CNRS-IPGP, Paris Sorbonne Cité, 1 rue Jussieu, 75005 Paris,

6

neuville@ipgp.fr

7

2 CNRS, UPR3079 CEMHTI, 1D avenue de la Recherche Scientifique, 45071 Orléans cedex2, France

8

3 Université d'Orléans, Faculté des Sciences, Avenue du Parc Floral, BP 6749, 45067 Orléans

9

cedex 2, France

10

4 Geology department, University of Toronto, 22 Russel Street, Toronto, ON, Canada.

11

12

13

14

Abstract:

15

Alkali aluminosilicate melts are material of prime importance for both the geology research

16

field and the glass making industry. Aluminium greatly influences glasses and melts properties, and

17

its effect strongly depends on its concentration. This paper present a study of Na2O-Al2O3-SiO2

18

glasses and melts, containing 75 mol% SiO2 and different Al/(Al+Na) ratios. 23Na 1D MAS NMR and

19

Raman spectroscopy both highlight the change of the role of Na+ with Al2O3 content. In

aluminium-20

poor composition, Na-O distances are short and Na is in coordinence ~6 and acts as a network

21

modifier. When Al replaces Na, 29Si and 23Na 1D MAS NMR and Raman spectroscopies show that Q3

22

species transform in Q4 species, and that Na-O distances increase, in agreement with an increase of

23

the Na coordination number. Then Na+ becomes charge compensators of Al3+. Interestingly, NMR

24

results show that the latter are incorporated in Q4 species in the most polymerized domains of the

25

glass network (Q4

(4Si) species). These structural changes increase drastically the melt viscosity,

26

which is maximum for Al/(Al+Na) = 0.5 (albite composition). When crossing the tectosilicate join,

27

Al/(Al+Na)=0.5, 27Al 1D MAS NMR spectra display that Al[5] is present in glasses. The higher Tg and

28

the lower fragility of peraluminous melts compared to the tectosilicate albite melt indicate that,

29

near the Tg, Al[5] enhances the network connectivity while at high temperature, it is a transient and

30

dynamic specie allowing the diffusion of NBO and BO in melts. Formation of Al[5] when increasing

31

the temperature can also account for the temperature dependent heat capacity of tectosilicate and

32

peraluminous melts.

33

34

35

1. Introduction

36

The structure and properties of glasses, melts and minerals in the ternary Na2O-Al2O3-SiO2

37

system are important for both geological and industrial processes. These three oxide components

38

constitute more than 80% of haplo-andesitic and -granitic magmatic systems, and are used in the

39

glass (window glass) and glass ceramic (analogue of LAS glass ceramic) industries. The Na2

O-40

Al2O3-SiO2 mineral phase diagram is well known (Shairer and Bowen, 1956), and the structure and

41

properties of sodium silicate and aluminosilicate glasses and melts have been previously

42

characterized using different tools: viscosity measurements (Riebling, 1966; Taylor and Rindone,

43

1970; Toplis et al. 1997a,b); heat capacity measurements (Richet and Bottinga, 1984; Navrotsky

44

et al. 1982; Roy and Navrotsky, 1984; Tangemann and Lange, 1998; Webb, 2008); NMR

45

spectroscopy (Maekawa et al., 1991a; George and Stebbins, 1996; Stebbins and Xue, 1997; Lee

46

and Stebbins, 2003, 2009); Raman spectroscopy (Seifert et al., 1982; Mysen et Frantz 1994a;

47

Neuville and Mysen, 1996); XANES at the Na K-edge (Neuville et al., 2004b) and X-ray radial

48

distribution analysis (Taylor and Brown, 1979a,b).

49

At high temperature and at constant SiO2 content, previous experiments have shown that

50

viscosities increase when Al substitutes for Na (for the ratio Al/(Al+Na) =

51

mol%Al203/(mol%Al203+mol%Na20) <0.5, in the peralkaline field), reach a maximum when

52

Al/(Al+Na) is around 0.5 and decrease for Al/(Al+Na) higher than 0.5 (in the peraluminous field;

53

Riebling, 1966; Toplis et al. 1997a,b). Near the glass transition temperature, viscosity variations

54

still show a large increase when Al substitutes for Na for Al/(Al+Na) < 0.5. However, when

55

Al/(Al+Na) ≥ 0.5 viscosities increase slightly (Taylor and Rindone, 1970). Several structural models

56

have been proposed to interpret these observations. Day and Rindone (1962a,b,c) postulated that

57

for Al/(Al+Na) ratios lower than – or equal to – 0.5, all Al2O3 ions are distributed in AlO4 species.

58

This has been confirmed by several studies (Day and Rindone, 1962c; Taylor and Brown, 1979b;

59

Seifert et al., 1982; McMillan et al., 1982; McKeown et al., 1984; Merzbacher and White, 1988; Zirl

60

and Garofalini, 1990; Neuville and Mysen, 1996; Lee, 2004; Lee et al., 2006). Day and Rindone

61

(1962a,b,c) also assumed that for a Al/(Al+Na) ratio > 0.5, i.e. with excess aluminum, Al3+ and

62

Na+ ions will go into a higher coordination state, probably 6 and 9 respectively. Recent XANES

63

measurements at the Na K-edge made on minerals and glasses, as well as 23Na NMR observations

64

have validated this statement (George and Stebbins, 1996; Neuville et al., 2004b). From other

65

viscosity and density measurements, Riebling (1966) reached the same conclusions. In order to

66

explain his observations, he suggested that some Al in five or six fold coordination may exist when

67

Al/(Al+Na) > 0.5. On the other hand, Lacy (1963) proposed a different model. He assumed that,

68

when Al/(Al+Na) > 0.5, all Al3+ ions are in a four-fold coordination state and three tetrahedra (2

69

SiO4 and 1 AlO4 tetrahedra) share only one oxygen atom to form “triclusters”. Toplis et al. (1997b)

70

followed Lacy’s idea in order to explain the viscosity maximum observed at high temperature near

71

the tectosilicate join. Unfortunately, oxygen “triclusters” have not been observed experimentally

72

except in calcium aluminates (Iuga et al., 2005).

73

In the present communication, we have studied the effect of the Al/(Al+Na) ratio, at constant

74

SiO2 concentration, on the structure and properties of glasses and melts using viscosity

75

measurements, NMR and Raman spectroscopy. We show that when Al substitutes for Na, it first

76

results in an increase in the melt’s viscosity and polymerization. Observed structural changes are

77

consistent with important thermodynamical and rheological changes depending on glass chemistry.

is added.

81

82

2) Experimental methods

83

84

2.1. Starting Materials

85

Six samples (Table 1) were made by melting a mixture of Al2O3, Na2CO3 and SiO2 powders

86

previously dried at 1100°C (SiO2 and Al2O3) and 350°C (Na2CO3), following the protocol already

87

described in Schairer and Bowen (1956) and Neuville (2006). Approximately100g of powders were

88

crushed in an agate mortar in ethanol for 1 hour. After that, decarbonation was performed in a

89

platinum crucible by slowly heating the sample up to 1100-1500°C in a electric muffle furnace.

90

When decarbonation was achieved, the melt was quenched by dipping the bottom of the crucible

91

into pure water. Four successive melting, quenching and grindings operations were made to

92

prepare homogeneous glasses. Finally, samples were maintained a few hours at high temperature

93

(1100-1500 °C depending on the viscosity of the sample) to obtain bubble-free samples required

94

for viscosity measurements. Densities of all samples have been measured with the Archimedes

95

method using toluene as the immersion liquid (Table 1). Chemical compositions have been

96

measured using a Cameca SX50 electron microprobe (Table 1), with a 30 nA current, U=30kV, and

97

5s of counting. The values reported in Table 1 are the statistical mean of 10-20 individual

98

measurements.

99

Glasses are named NA75.XX, with XX the concentration in mol% of Al2O3. They contain 75

100

mol% SiO2 and the Na2O concentration is given by 100 – (75+XX) mol%. NA75.15 data come from

101

the literature (see Table 1 caption).

102

103

104

105

2.2. High Viscosity Measurements

106

Viscosity measurements on melts near their glass transition temperature have been performed

107

in air using a creep apparatus (see Neuville et Richet, 1991 and Neuville, 2006 for further details).

108

Samples used for measurements are small cylinders of approximately 5 mm diameter x 10 mm

109

length. Their approximate weight is 0.5 g. The temperature gradient is the most critical part of

110

such experiment and needs to be minimized. To limit these gradients, a silver cylinder was placed

111

around the sample, creating a small chamber where the temperature is homogeneous.

112

Furthermore, thermal gradients along the sample were checked using two Pt-PtRh10 thermocouples

113

(ITS90 type S thermocouples). The temperature difference between the top and the bottom of

114

samples was always less than 0.2 K during viscosity measurements. To measure samples viscosity

115

at one temperature, we performed 20 to 40 measurements at different stresses (between 6.6 and

116

8.9 log Nm-2) to be sure that no non-Newtonian behaviour appeared; the reported viscosity value

117

at a single temperature is the statistical mean of these measurements. Measurements carried out

118

on the NBS 717 glass show that the viscosity uncertainty and reproducibility is less than 0.03 log

119

units with this technique (see Neuville, 2006).

120

121

2.3 Raman Spectroscopy at Room Temperature

122

Raman spectra were recorded using a T64000 Jobin-Yvon® triple Raman spectrometer equipped

123

with a confocal system, a 1024 CCD (Charge-Couple Detector) cooled by liquid nitrogen and an

124

Olympus® microscope. The optimal spatial resolution allowed by the confocal system is 1-2 μm2

125

with a x100 Olympus® objective, and the spectral resolution is 0.7 cm-1. A Coherent® laser 70-C5

126

Ar+, having a wavelength of 514.532 nm, was used as the excitation line. Samples were excited

127

with a laser power of 200 to 250 mW. No laser-induced damage was observed.

128

Spectra were acquired on freshly exposed surfaces (fresh breaks) between 15-20 and 1500 cm

-129

1. Acquisition conditions such as time and repetition were adjusted to the signal emitted by the

130

sample area. Typical values are 180 to 300 seconds and 3 repetitions. The analysed volume was

131

adjusted close to the surface in the Raman optimum region, i.e on the first 10 µm of depth. All

132

reported spectra are unpolarized.

133

Before deconvolution, spectra were corrected for temperature and excitation line effects using a

134

correction factor introduced by Long (1977)and given by Neuville and Mysen (1996):

135

I = Iobs .[o3 [1-exp(-hc/kT)]  / (o-)4], (1)

136

with h the Planck constant, h=6.626038.10-34 J.s, k the Boltzmann constant; k=1.38066.10-23

137

J.K-1, c the speed of light, c=2.9979.1010 m.s-1, T is the absolute temperature, o the wavenumber

138

of the incident laser light (in the present study the 514.532nm 70-C5Ar+ line, o =19435.1cm-1),

139

and  the measured Raman shift in cm-1.

140

In order to perform deconvolution of the 850-1300 cm-1 spectral range, representative of the Qn

141

species distribution (see below), this spectral range was normalized to its intensity maximum.

142

Spectral deconvolution was performed using the quasi-Newton algorithm given by Tarantola (2005)

143

implemented in Matlab® (Le Losq and Neuville, 2012). This iterative algorithm is based on the

144

method of non-linear least square minimization. Inputs for the software are:

145

– Raman spectra, i.e raman shifts, intensities and their associated errors. Data errors were

146

estimated as sqrt(n) on raw spectra with n the counts per second, and corrected also by the Long

147

equation. We have then input the total error in the algorithm, which is the sum of the corrected

148

counting error and of the standard deviation between spectra data points and an ideal line. This

149

last contribution to the total error has been determined on a linear spectral region, devoid of any

150

features, in corrected and normalized spectra (in our case between 1300 and 1400 cm-1). We

151

assume that by doing so, all errors from data collection and processing are taken into account.

152

Before deconvolution of the 850-1300 cm-1 spectra region, intensities and their associated errors

153

were normalized in order to obtained an intensity maximum of 100;

154

– The model, which is a simple sum of Gaussian bands, and the estimations of the initial

155

parameters values and of their uncertainties. Initial values of Raman shifts, FWMH and intensity of

156

bands, and their associated uncertainties were the same for all spectra. The input uncertainties

157

were: i) = 15 cm-1 for the Raman shifts; ii) = 20 cm-1 for FWMH (depending of the initial peak

158

FWMH) and iii) 100 for intensity, in order to have no constrains on this parameter.

159

The quasi-Newton algorithm takes into account parameter uncertainties and data errors in order

160

to minimize the least square criterion and to converge to an optimal point. In order to check if the

161

output solutions were the global minimum (best solution), and not just a local minimum

162

(intermediate solution), several sensitivity tests of input parameters were performed. Initial

other, allowing the highest degree of reproducibility to be achieved.

166

167

2.4 Nuclear Magnetic Resonance Spectroscopy at Room Temperature

168

27Al and 23Na NMR experiments were performed at the CEMHTI-CNRS Orléans on a high field

169

NMR Bruker Avance III 750 (17.6 T) spectrometer working at a 27Al frequency of 195.5MHz and a

170

23Na frequency of 198.4 MHz. Chemical shifts for the 27Al and 23Na are referenced to a 1M aqueous

171

Al(NO3)3 or NaCl solution respectively. Magic Angle Spinning was performed at a speed of 30 kHz in

172

aluminium free zirconia rotors of 2.5mm diameter. A small pulse angle (less than /18 for 27Al and

173

 /12 for 23Na) was used with a radio-frequency of 50 kHz to ensure the quantitative character of

174

the spectra (Lippmaa et al., 1986). Four thousand (resp. 2000) scans were accumulated for 27Al

175

(resp. 23Na) with a recycling time of 0.5s (estimated spin-lattice relaxation times T1 of less than

176

100ms), using a spectral window of 1 MHz. The decomposition of the 1D spectra where obtained

177

using the "dmfit" program (Massiot et al., 2002) which allows for retrieval of the mean isotropic

178

chemical shift iso (which does not coincide with the position of the peak maximum), the distribution

179

of isotropic chemical shift iso and the mean quadrupolar coupling constant, CQ within the

180

framework of the Gaussian Isotropic Model (also called “Czjzek”, Le Caer et al., 1998).

181

29Si NMR was performed on a Bruker Avance I 200 spectrometer operating at a frequency of

182

39.7 MHz. Samples were packed in 7mm diameter zirconia rotor and spun at a speed of 5 kHz. To

183

avoid dead-time problems, a synchronized Hahn Echo sequence was used to record the spectra at

184

a radio-frequency field of 25 kHz and an echo shift of 3 rotor periods. The spin-lattice relaxation

185

time T1 being estimated (using a saturation-recovery experiment) between 250s for alumina-poor

186

and 150s for alumina-rich samples, the recycle delay has been adjusted accordingly between 250s

187

and 150s. Between 800 and 1100 scans were accumulated for each composition (i.e. 3 days of

188

acquisition time).

189

190

3) Results

191

3.1 Viscosity

192

The measurements performed on supercooled melts and in the liquid state are listed in Table 2

193

and plotted in figure 1a. They are in good agreement with previous measurements performed

194

using the fibre elongation method (Taylor and Rindone, 1970) or micropenetration (Toplis et al.

195

1997a). Figure 1a shows that at a given temperature, the viscosity of the NA75.00 melt is the

196

lowest compared to other compositions. When Al substitutes for Na, and for Al/(Al+Na) < 0.36

197

(NA75.00 to NA75.09 compositions), viscosity increases slightly while for Al/(Al+Na) between 0.36

198

and 0.5 (NA75.09 and NA75.12 compositions), viscosity increases drastically. For Al/(Al+Na)

199

higher than 0.5, in the peraluminous field, the viscosity increases very slightly. This behaviour is

200

also observed in the dependence of the glass transition temperature (Tg, which correspond to a

201

viscosity of 12 log Pa.s) on the Al/(Al+Na) ratio (Fig. 1b). Tg increases non-linearly, with a clear

202

transition occurring when Al/(Al+Na) is between 0.36 and 0.5 (NA75.09 and NA75.12 melts). In

203

this range, Tg increases by 240° in less than 3 mol% of substituted Na. The Tg of the glass of

204

Albite composition (NA75.12) is of 1085.7K, in good agreement with the calorimetric

measurements of Richet and Bottinga (1984). When Al/(Al+Na) > 0.5, Tg increases very slowly, 206

reaching 1102.4 K for the NA75.16 glass. 207

The viscosity variations observed at low temperature are also visible at high temperature when 208

looking at the data of Riebling (1966) and Toplis et al. (1997a,b; Fig. 2a). However, they are less 209

pronounced in the high-temperature / low-viscosity range. Furthermore, the NA75.00 viscosity 210

versus 1/T curve shows strongly non-arrhenian behaviour, while the viscosity curves of the 211

tectosilicate (NA75.12) and peraluminous (NA75.15 and 16) melts display arrhenian behaviour (Fig 212

2a). The different non-arrhenian behaviours are linked to melt fragility: NA75.00, NA75.02, 213

NA75.06 and NA75.09 melts are more fragile than tectosilicate NA75.12 and peraluminous 214

NA75.15/NA75.16 melts (Fig. 2b, Angell, 1991). The NA75.00 is the most fragile. The NA75.12 215

melt is stronger than other peralkaline (Al/(Al+Na) < 0.5) and peraluminous (Al/(Al+Na) > 0.5) 216

melts (Fig. 2b). 217

Near the glass transition (viscosities between 109 and 1014 Pa.s), the relationship between 218

log η and 1/T is not linear and can be fitted using a Tamman-Vogel-Flucher (TVF) equation: 219

log 𝜂 =  𝐴 +   !

(!!!!) , (2)

220

where A, B, T1 are adjustable parameters given in Table 3, T the temperature in K and η the 221

viscosity in Pa.s. This equation was used in order to interpolate viscosity data to low temperature. 222

However, it is an empirical fit of viscosity data and does not provide pieces of information about 223

the thermodynamic state of the studied melt (Neuville and Richet, 1991). In order to model the 224

whole range of temperature and to be able to extract thermodynamic information, one can use the 225

Adam and Gibbs equation: 226

log 𝜂 =  𝐴 +   !"

!!!"#$(!) , (3)

227

where Ae is a pre-exponential term, Be a constant proportional to the potential barrier 228

opposed to the cooperative rearrangement of the liquid structure, Sconf(T) the melt configurational

229

entropy, η its viscosity in Pa.s and T its temperature in K (Richet, 1984; Neuville and Richet,

230

1991). This model is based on the Adam and Gibbs’ theory of relaxation processes (Adam and

231

Gibbs, 1965). It assumes that matter transport in a viscous melt implies the cooperative 232

rearrangement of subsystems having a configurational entropy Sc* and a size z*(T) at a 233

temperature T, separated by a Gibbs free-energy barrier Δµ opposed to their movements. 234

Considering one mole of particles in a melt, we can write the Be and Sconf(T) parameters of

235 equation 3 as: 236 𝐵𝑒 =  !!  !! !! , (4) 237 and 238 𝑆!"#$ 𝑇 =   𝒩! !∗ (!")𝑆!+ (𝒩! !∗ ! 𝒩! !∗ !")𝑆!, (5) 239

with kB the Boltzmann constant and NA the Avogadro constant (Adam and Gibbs, 1965; Bottinga

240

and Richet, 1996). We deliberately wrote equation 5 in a developed form because it is linked from 241

a macroscopic point of view to the following equation (Richet, 1984): 242 𝑆!"#$(𝑇) = 𝑆!"#$(𝑇𝑔) + ! 𝐶𝑝!"#$ !" /𝑇𝑑𝑡 , (6) 243 with 244 𝐶𝑝!"#$ 𝑇 = 𝐶𝑝! 𝑇 − 𝐶𝑝!(𝑇𝑔) . (7) 245

Cpg(Tg) is the heat capacity of the glass at Tg. We thus directly see comparing equation 5 and 6

248

that Sconf(Tg) is linked to Sc* and z*(Tg), and to the variation of z* depending on

249

temperature.

250

Be in equation 3 is not temperature dependent, implying through equation 4 that bothand

251

Sc* have a negligible temperature dependence. This implies that the microscopic mechanism

252

controlling viscous flow in a melt does not change with the temperature (Toplis, 1998), at least on

253

the applicable viscosity range of Eq. 3 (10-1012 Pa.s, Bottinga et al., 1995). Only the melt

254

composition strongly influences non-linearly Be, hence andSc*. Since Sc* influences Sconf(Tg) as

255

shown by equation 5, this explains why viscosity variations near Tg strongly depend on melt

256

chemistry. At high temperature, the z*(T) influence on Sconf(T) surpasses that of Sc*, and the

257

configurational heat capacity of the melt will play a fundamental role on Sconf(T) and hence

258

viscosity.

259

From equations 5 and 6, it appears that the melt configurational entropy at Tg and hence the

260

residual entropy of glass, reflects the configurational entropy of melt subsystems. Sconf(Tg) has two

261

contributions: i) a topological part, which is mainly due to the various distributions of bond angles,

262

interatomic distances and coordination numbers, and ii) a chemical part, induced by the mixing of

263

different elements (Neuville and Richet, 1991). As a result, Sconf(Tg) reflects the melt

264

configurational state, providing an image of its atomic disorder, and variations of Sconf(Tg)

265

dependent on melt chemistry bring insights about the influence of chemical and structural changes

266

on thermodynamic properties.

267

By combining equations 3 and 6, one can calculate the Sconf(Tg), Ae and Be parameters if

268

viscosity data and heat capacities values are available. Estimation of heat capacities are of great

269

importance because small errors on Cpg and Cpl estimations can lead to strong deviations of the

270

calculated Be and Sconf(Tg) parameters. The Cpg of the studied glasses were computed according to

271

the model of Richet (1987), that considers Cpg as additive functions of the composition.

272

For melts with Al/(Na+Al) < 0.5, liquid heat capacity, Cpl, is an additive function of the