Phys Chem Minerals (1996) 23:157-172

9 Springer-Verlag 1996

P. R i c h e t 9 B . O . M y s e n 9 D . A n d r a u l t

Melting and premelting of silicates: Raman spectroscopy and X-ray diffraction of Li2SiO3 and Na2SiO3

Received May 9, 1995/Revised, accepted October 20, 1995

The isostructural lithium (Li2SiO3) and sodium (NazSiO3) metasilicates have been investigated from room temperature up to the melting point by single-crystal Raman spectroscopy and energy-dispersive X-ray powder diffraction. The unit-cell parameters and Raman frequencies of LizSiO3 vary regularly with temperature up to the melting point, which is consistent with the lack of premelting effects in calorimetric measurements. In contrast, Na2SiO 3 undergoes a transition at about 850 K from orthorhombic Cmc 21 symmetry, to a lower symmetry (possibly Pmc20, and shows near 1200 K changes in the Raman spectra that correlate well with the premelting effects as determined from calorimetry observations. In both compounds, a high alkali mobility likely sets in several hundreds of degrees below the melting point. Premelting in Na2SiO 3 is associated with extensive deformation of the silicate chains as evidenced near the melting point by similarities in the Raman spectra of the crystalline and liquid phases. Abstract

Introduction Melting is of course the most dramatic phase transition experienced by crystals, yet its nature is still understood poorly. Over often wide temperature intervals below the solidus, melting is generally announced by anomalous variations of physical properties of crystals (e.g., Ubbelhode 1978). For silicates, interest has been paid recently to these premelting effects which have been observed mainly through calorimetry experiments (e.g., Richet and Fiquet 1991). Rapid increases of the isobaric heat Pascal Richet (~) 9Denis Andrault Laboratoire des G6omat6riaux, Institut de Physiquedu Globe, URA 734 CNRS, 4 place Jussieu, 75252 Paris cedex 05, France Bjorn O. Mysen Geophysical Laboratory,Carnegie Institution of Washington, 5251 Broad Branch Rd, Washington,D.C., 20015, U.S.A., and NSF-sponsored Center for High-Pressure Research (CHiPR)

capacity (Cp) have been observed, beginning in some instances more than two hundred degrees below the melting point and accounting for up to 20% of the reported enthalpies of fusion of minerals. Premelting of silicates represents rapid, reversible, temperature-induced configurational changes within phases remaining crystalline up to their temperature of fusion (Richet et al. 1994). From structural data near the melting point, one might thus draw information on the premonitory steps through which long-range order of crystals eventually disappears. In this respect, silicates could be more amenable to structural studies than other kinds of crystals because their melting is generally sluggish and the energetics of their premelting has a large magnitude. Until now, however, such studies have been very scarce because of the difficulties of making structural measurements at the high melting temperatures of silicates. By taking advantage of recently designed heating cells, we have used high-temperature X-ray diffraction and Raman spectroscopy for studying premelting and melting of silicates. In this report, two simple model compounds, namely lithium (Li2SiO3) and sodium (Na2SiO3) metasilicates, have been investigated with these techniques which probe structures at different length scales. Both of these metasilicates have an orthorhombic symmetry at room temperature (space group Cmc 21) and are made up of Si206 chains along the c axis linked by either NaO5 trigonal bipyramids or by LiO 4 tetrahedra (McDonald and Cruickshank 1967; Hesse 1977). These compounds are isostructural, yet LizSiO3 displays almost no premelting (Stebbins et al. 1984; T6qui et al. 1992) whereas NazSiO 3 exhibits a wide premelting range of 160 K (Naylor 1944; Richet et al. 1984). In preliminary measurements, we showed that the contrasting premelting behavior of lithium and sodium metasilicates is accompanied by differences in the temperature dependence of the frequencies and linewidths of some Raman-active vibrational modes (Richet et al. 1994). In this paper, we describe X-ray diffraction exper-

158 i m e n t s m a d e u p to t h e c o n g r u e n t m e l t i n g p o i n t s o f N a 2 S i O 3 and L i 2 S i O 3 at 1 3 6 2 a n d 1 4 7 4 K, r e s p e c t i v e l y ( K r a c e k 1 9 3 0 a, b). W e a l s o r e r e c o r d e d t h e R a m a n s p e c t r a o f t h e s e c o m p o u n d s at h i g h t e m p e r a t u r e s to a c h i e v e a b e t t e r a c c u r a c y t h a n in o u r p r e l i m i n a r y study. T h a n k s to a h i g h e r r e s o l u t i o n , e s p e c i a l l y at l o w f r e q u e n c i e s , a n d to systematic curve fitting of the spectra, we could track a n u m b e r o f p e a k s u p to t h e m e l t i n g p o i n t o f t h e c r y s t a l and even investigate the spectral changes which take place on melting.

we actually determined only the variations of the lattice parameters relative to the room-temperature values. For convenience, however, we report absolute values, the room-temperature data representing the best fit of our result with respect to the singlecrystal unit-cell data. As already noted, the first diffraction pattern was recorded near 340 K and not at room temperature in the two series of measurements for sodium metasilicate. To take into account the slight expansion above 298 K, we simply extrapolated down to 298 K the relative changes in lattice parameters determined at higher temperatures.

Raman spectroscopy

Experimental methods Heating cells and materials For Raman spectroscopy and X-ray diffraction, the materials were heated in air with the high-temperature cells described in detail by Mysen and Frantz (1992) and Richet et al. (1993), respectively. With both cells, the sample is enclosed in the small hole drilled in the Pt-Rh alloy of a wire which acts as a heating element, and whose temperature is determined from calibrations made from the known melting points of various salts and silicates. The reported temperatures are accurate to about 10 K. Radial temperature gradients across the sample in the heating wire are simply given by the thermocouple temperature difference between beginning of melting, around the edge, and end of melting at the center of the cell. This gradient amounts to a few degrees at around 1000 K and increases to about 10-15 degrees at 1800 K. The compounds investigated are the synthetic crystals whose relative enthalpies were measured by Richet et al. (1984) and Tdqui et al. (1992), to whom we refer for detail on sample preparation. No other phase than Li2SiO 3 or Na2SiO 3 was apparent in the Raman spectra. Likewise, with the exception of two reflections of the platinum wire observed in a few X-ray patterns, all the observed diffraction peaks could be assigned to the metasilicate samples.

X-ray diffraction The X-ray diffraction experiments were made in an energy dispersive configuration on the wiggler line of the DCI storage ring of LURE (Orsay, France). Non-hygroscopic materials such as Li2SiO 3 are finely ground under alcohol in an agate mortar. After a few minutes, the upper part of the suspension is pipetted, the alcohol is evaporated and the resulting powder is then loaded in the heating cell. The powder obtained in this way is extremely fine so that orientation effects are avoided when recording the diffraction patterns. This procedure could not be followed for the hygroscopic Na2SiO 3 which was ground in a dry air glove box without an immersing liquid. After rapidly loading the sample in the cell, diffraction patterns were recorded from a temperature of 340 K which was high enough to prevent partial hydration of the sample. For all diffraction experiments, the X-ray beam was collimated to 200• gm 2. The diffracted beam was analyzed with a Canberra planar germanium detector at 20 angles of about 12 degrees for energies comprised between 10 and 50 keV. Good quality spectra were recorded in about ten minutes. For cell-parameter refinements, we used only reflections for which both the peak energy and linewidth could be determined reliably from room temperature up to temperatures close to the melting point. This procedure excluded most reflections at energies higher than 35 keV whose intensity was too small as a result of the rapid fall-off in the intensity of the X-ray beam which begins at energies of about 25 keV. The diffraction angle was determined from comparisons of the observed reflections at room-temperature with those reported in single-crystal studies (McDonald and Cruickshank 1967; Hesse 1977). Hence,

As described by Mysen and Frantz (1992), the cell was coupled through an inverted microscope to a Dilor XY confocal microraman spectrometer equipped with a CCD multichannel detector. The 488 or 514-nm lines of an argon ion laser were used as the excitating radiation. The laser power was set at 150 and 500 mW for the measurements on Na2SiO 3 and Li2SiO3, respectively. This laser power was low enough to prevent significant laser heating or degradation of the samples, particularly of sodium metasilicate which vaporized readily under the laser beam at higher powers. Crystalline phases examined a few tens of gm away from the coexisting liquids, i.e., at temperatures a few degrees below the melting points, did not present evidence of partial melting. These observations also indicate that laser heating was not significant, even at the highest temperatures investigated. An 1800 grating/mm was used, providing a resolution of about 2 c m - ~for a spectral range of about 700 cm -1. To speed up data acquisition, the spectra were obtained separately between 100 and 800 cm ~ and between 700 and 1400 cm-J in two different series of experiments. Slit widths and the confocal hole diameter were left unchanged throughout all the experiments. Accumulation times were kept constant at 120 and 60 s for Na2SiO 3 and Li2SiO 3, respectively. The Raman spectra were recorded on single crystals grown directly in the heating cell from slightly supercooled melts. Goodquality Raman spectra were obtained at once from any part of the crystals. This configuration constitutes a considerable advantage over experiments made on polycrystals or powders which have proven difficult either in commercially available heating stages or with a heating-wire technique. With polycrystals, for instance, spectra could be obtained only on some tiny fragments that had to be found at random, and the quality of the spectra generally degraded considerably with increasing temperatures in part due to enhanced black-body radiation from grain surfaces. Experience showed instead that similar spectra were obtained from various parts of the same single crystal or from different samples grown with a priori different orientations. Good agreement was also found for both metasilicates with previous measurements made at room temperature on polycrystals (Devarajan and Shurvell 1977; Brawer and White 1975). Because we are interested in the main spectral changes, which were reproducible, and not in recording all Raman-active vibrational modes, our conclusions should not be affected by minor orientation effects. All spectra were corrected for temperature- and frequency-dependent scattering intensities (Long 1977; Mysen et al. 1982 b). No attempt was made at merging the ends of the low- and high-frequency parts of the spectra since these were usually not recorded at the same temperatures. The linewidth of the Raman bands generally increases with temperature and a number of peaks merge to form shoulders and asymmetric bands. These effects can result in spurious temperature dependences of the vibrational frequencies if the resulting bands are not properly accounted for in terms of their individual components, The spectra recorded at higher temperatures were thus fitted with the bands first observed at room temperature which could then be tracked continuously. The curvefitting procedure described by Mysen et al. ( 1982 b) was followed, whereby individual modes were dropped out if they did no longer lead to improvements in Z2 and residuals. The lineshape of all Raman bands was clearly Lorentzian at room temperature. At higher temperatures, the spectra could not

159 generally be fit in a satisfactory manner with purely Lorentzian lineshapes because the tails of the Lorentzian bands yielded background levels inconsistent with the observed spectra. This progressive departure from Lorentzian lineshapes is the strongest for the low-frequency modes (see Appendix 1). It could at least partly result from the contribution of second-order spectra, but this feature cannot be regorously accounted for when structural changes takes place in materials like the investigated metasilicates. Empirically, we thus used above room temperature a combination of Lorentzian and Gaussian lineshapes to fit the experimental lineshapes. The proportion of Gaussian component was arbitrarily allowed to increase by 10% increments until a satisfactory fit was achieved, starting from the percentage used for the experiment performed at the previous lower temperature. In order not to introduce too many adjustable parameters, the low-frequency bands were grouped in two intervals below and above about 460 cm -~. Satisfactory fits could be obtained with a constant Gaussian character at a given temperature for all bands within the same interval. During the curve-fitting process, we also checked that the reported data were not biased by possible trade-offs between frequencies and linewidths.

X-ray diffraction Lithium metasilicate The cell parameters and volumes of lithium metasilicate are reported in Table 1. At room temperature, the 12 most intense diffraction peaks represented a total of 25 possible reflections. Owing to anisotropy in thermal expansion, four of these peaks could be progressively resolved into two separate components at higher temperatures. Above 600 K, 16 different peaks were thus followed, with which the quality of refinement remained essentially the same up to the melting point of Li2SiO 3. For the last experiment, just below melting, only 12 peaks could be resolved with a lower precision, whence the greater error margins of the cell parameters at the highest temperature (Table 1). The main result of these measurements is a smooth increase of the cell parameters with temperature up to the melting point (Fig. 1), without any evidence for symmetry changes from the room-temperature Cmc 21 space group. The linear and volume thermal expansion coefficients derived from the data of Table 1 are listed in Table 2. As c o m m o n for chain silicates, thermal expansion is smallest along the c axis, especially below 1000 K, before increasing at higher temperatures. Expansion is about twice as great and nearly the same along the a and b axes where it proceeds at a constant rate up to the melting point.

Sodium metasilicate In contrast to Li2SiO3, sodium metasilicate shows definite changes in the diffraction patterns with increasing temperatures in the two different series of experiments which were made. The main features are 6 new peaks (Fig. 2) which progressively develops, beginning at 850 K, and whose relative intensities increase with increasing temperature. These additional reflections are

Table 1 Unit-cell parameters (A) and volume (~k3) of lithium and sodium metasilicates a T (K) Li2SiO3 302 365 476 665 808 909 963 1062 1184 1284 1389 1404 1413 1474 Na2SiO3, 340 431 568 712 832 955 1051 1158 1228 1290 1338

a

9.381 (6) 9.385(6) 9.412(4) 9.460 (4) 9.487 (5) 9.510 (4) 9.522 (5) 9.544 (4) 9.564(5) 9.595(6) 9.616(6) 9.610 (4) 9.618(7) 9.623(14) 1st series 10.506(2) 10.522(2) 10.543(5) 10.566(5) 10.579(10) 10.578(5) 10.571(5) 10.586(7) 10.592(12) 10.599(4) 10.605(4) Na2SiO3, 2nd series 341 10.502(1) 834 10.573(5) 1155 10.591(9) 1223 10.593(5) 1272 10.584(9) 1309 10.596(6) 1336 10.619(10) 1348 10.621(14) 1373b 10.624(17)

b

c

V

5.399 (3) 5.411(3) 5.421(3) 5.445 (2) 5.463 (3) 5.475 (2) 5.484 (2) 5.496 (2) 5.509(3) 5.520(3) 5.535(3) 5.543 (2) 5.541(3) 5.553(7)

4.667 (3) 4.675(3) 4.673(5) 4.677 (2) 4.689 (3) 4.692 (2) 4.696 (2) 4.704 (2) 4.712(3) 4.720(3) 4.733(3) 4.737 (2) 4.739(3) 4.747(7)

236.4 (4) 237.4(5) 238.4(5) 240.9 (3) 243.0 (4) 244.3 (3) 245.2 (4) 246.7 (3) 248.3(4) 250.0(4) 251.9(4) 252.4 (3) 252.6(4) 253.6(10)

6.065(1) 6.073(1) 6.088(3) 6.102(3) 6.105(5) 6.109(3) 6.105(3) 6.112(4) 6.113(7) 6.118(2) 6.121(3)

4.821(1) 4.833(1) 4.845(3) 4.862(3) 4.878(6) 4.894(3) 4.914(3) 4.928(4) 4.940(8) 4.942(3) 4.951(3)

307.2(2) 308.8(1) 311.0(4) 313.5(5) 315.1(9) 316.3(5) 317.2(5) 318.9(6) 319.9(12) 320.4(4) 321.4(4)

6.063(1) 6.103(3) 6.110(5) 6.115(3) 6.110(5) 6.116(4) 6.127(6) 6.125(8) 6.125(10)

4.822(1) 4.885(4) 4.928(7) 4.934(4) 4.938(6) 4.939(4) 4.950(7) 4.953(10) 4.958(12)

307.2(1) 315.5(5) 319.1(10) 319.8(5) 319.6(9) 320.3(6) 322.3(10) 322.4(15) 322.8(18)

a Parameters actually referring to room-temperature unit-cell volumes of 236.15 and 306.6 ~3 for Li2SiO3 and Na2SiO3, respectively, with an assumed volume expansion of 2.0%~ for Na2SiO3 between 298 and 340 K (see text) u Nominal temperature higher than the reported melting point of 1362 K because of experimental errors

consistent with an orthorhombic unit-cell as obtained from the other reflections followed from lower temperatures, but not with the initial Cmc21 space group for! which they are all forbidden. A rigorous determination of! a new space group is not possible from our energy-dis-! persive diffraction patterns. We have thus assumed al Pmc 21 space group because the loss of the centered posi- I tion in the a, b plane represents the smallest symmetryl change consistent with non-zero factor structures for I most additional reflections, namely 300, 102 (or 410), I 500 (or 212 and 321), 012 and 300. That the symmetry] could be lower than orthorhombic is suggested by the i presence in both series of measurements of a weak 3011 reflection, which is forbidden in the Pmc 21 space group These observations in fact resemble those of West (1977)! on mixed (Na, Li) metasilicates where satellite reflec-!

t60 1.035

1.6

1.030

1.4.

x. n

o.~

1.005

0.2 0500

700

900 T(K)

1100

1300

1500

Fig. 1 Relative changes in the cell parameters (p) of lithium metasilicate up to the melting point at 1474 K. For clarity, the data for the a axis have been displaced upward by 0.01 unit

Table 2 Coefficients of linear and volume thermal expansion equations c{=C{o+Cq T+c~2T 2 (K i)

b c V

i

0.6

1.010

NazSiO3 a

L

d >,0.8

1.015

Li2SiO3 a b c V

3

1.o !

1.020

1.000 3O0

SiO

1.2

1.025 g

NQ2

106{20

109{Xi

25.32 23.19 3.21 51.72

- 2.97

27.49 4.61 34.57 5.90 21.76 83.82 32.27

-23.07

- 2.97

-35.43 5.84 -52.66 5.84

10120{2

T

15

i

i

[

20

p

i

p

r

I

~

f

q r

25 Energy (keV)

I

30

I

35

Fig. 2 Comparison between the X-ray diffraction patterns of sodium metasilicate at the nominal temperatures of 340 K (top) and 1373 K (bottom). The additional reflections are shown by the arrows, those of the heating wire by Pt 1.035 1.030

NezSi03~ /

1.025 1.020

& 1.o15

11.70 11.70

1.010

(T (T (T (T

830 K) 830 K)

1.005 1.000 0.995 300

(T 830 K)

tions indicate complex symmetry changes when substituting an alkali element by another. The cell parameters reported in Table 2 and plotted in Fig. 3 were calculated from the 10 most intense different peaks representing a total of 19 different reflections. For reasons of consistency, only those reflections used in cell-parameter refinements at lower temperatures were also used up to the melting point. For the extra reflections, the interplanar spacings do not agree as well in the two series of measurements as for the main reflections used in cell parameter refinements. In addition, their spacings as calculated from the obtained cell parameters tend to be systematically lower than the experimental values. In terms of X-ray energies, however, these differences amount to less than 80 eV, which is likely insignificant in view of the low intensity of the extra reflections and the fact that a monoclinic unit cell could be required to account for them properly. Thermal expansion coefficients are listed in Table 2 where two temperature ranges have been delimited to keep simple analytical expressions. At the lowest tern-

, 00

T(K}

Fig. 3 Relative changes in the cell parameters (p) of sodium metasilicate up to the melting point at 1362 K. No differences are observed between the first and second series of measurements. For clarity, the data for the a axis have been displaced downward by 0.005 unit

peratures, thermal expansion is similar along the three crystallographical axes. Then expansion along the c axis continues to increase smoothly whereas it begins to level off at around 700 K along the a and b axes, just before the first additional reflections appear. Expansion along the a axis could increase strongly in the immediate vicinity of the melting point (Fig. 3), but more accurate measurements are needed to confirm this effect.

Raman spectroscopy Spectroscopic data The general features of the room-temperature Raman spectrum of Li2SiO3 are shown in Fig. 4. To within experimental errors, the wavenumbers included in Fig. 4

161 Table 3 Raman-activevibrational modes of lithium metasilicate (data in cm-~ and K)

Mode VI V2

is the highest temperature up to which a linear temperature dependence of the wavenumber is observed, Tm~~ that at which this mode could be observed, For modes having a nonlinear temperature dependence (nl), the listed ~ 6o/~T refers to room temperature b Data of Devarajan and Shurvell (1977), included for comparison co2=231-4.35 10-3T -1.22 10-sT2 up to 1468 K ~6~ =589-3"58 10-~T -1.6910 57eupto1130K a Zlin

V3

v4 v5 v6 v7 v8 v9 v~~ v,j /]12

v, 3 v~4 v~5 Vl6 v~7 v~8

Assignment

0)(298 K)

103ac0/OT

rlin a

Zrnax a

Li-O str Li-O str, O - S i - O bend O - S i - O bend, Li-O str Li-O str Li-O str

207 232 294 326 348 362 398 411 465 498 527 587 613 728 843 917 977 1029

-22.4+3.0 -29.4-+0.7 -51.1-+5.0 -25.2 -36.9-+ 1.8 -14.5-+1.8 -59.3-+5.3 -18.3-+0.2

1130 nl~ 815 500 1453 1130 658 1468

-50.0 -20.1 -+0.3 -13.7-+5.9 -18.0-+0.4

500 1468 nld 1468

-19.0_+0.6 -18.9-+7.9 -21.5-+0,7 -23.4-+2.8

1473 1096 1473 926

1130 1468 1130 500 1468 1468 815 1468 298 500 1468 1130 1468 298 1473 1096 1473 926

Li-O str, O - S i - O bend O - S i - O bend Li-O str, O - S i - O bend O - S i - O bend, Li-O str Si-O str, S i - O - S i bend O - S i - O bend, Si-O str S i - O - S i bend, Si-O str Si-O str Si-O str Si-O str, O - S i - O bend Si-O str Si-O str, O - S i - O bend

I.O

120 lOO

B0 NozSiO3

C

\ O')~

20

i

li r 200

4.00

'

/

600 800 Wovenurnber (cm-'~)

/

Lt 2 S i O 3

1000

1200

Fig. 4 Room-temperature Raman spectra of lithium and sodium metasilicates, plotted without temperature corrections to show more clearly the low-intensity, low-frequency modes. Note that these spectra have been recorded in another series of measurements than those reported at higher temperatures, whence the differences in the relative intensities of the bands and the presence of a few additional hands. To avoid confusion, the frequencies included are those listed in Tables 3-4, with which the actual frequencies agree to within the resolution of our measurements

agree with the data reported previously for Li2SiO 3 by Devarajan and Shurvell (1977) who actually reported 29 Raman bands instead of the 19 modes shown in Fig. 4 for Li2SiO 3 . We did not make any attempt at tracking the 10 extra bands, however, because their very low intensity precluded their observation at higher temperatures. The mode assignments listed in Table 3 have been made from the normal mode calculations of Devarajan and Shurvell (1977) for Li2SiO 3 and the calculations of vibrational frequencies of Si206 chains by Furakawa et al. (1981).

~0b 210 234 291 325 345 398 410 496 520 587 610 731 852 945 983 1034

The room-temperature Raman spectrum of Na2SiO 3 (Fig. 4) is consistent with previous observations by Brawer and White (1975) and Konijnendijk and Stevels (1976) whose spectra extend down to 300 cm -1 only. As expected, the Raman spectrum of Na2SiO 3 is closely related to that of its isostructural litium counterpart (Fig. 4). Most of the modes of Na2SiO 3 could be assigned in a straightforward way because vibrational frequencies are systematically lower for Na2SiO 3 than for Li2SiO 3. The shifts are of the order of 10-15 mc - I for modes involving essentially S i - O bending and stretching. They are higher for the low-frequency modes involving alkalioxygen stretching whose wavenumbers range between 520 and 210 cm -1 for LiaSiO 3 and between 510 and 170 cm -1 for Na2SiO 3. We list in Appendix 1 the wavenumbers and full widths at half height (hereafter denoted more simply linewidths) of the bands observed above room temperature, along with the various Gaussian-Lorentzian percentages relevant for each of the three frequency ranges delineated for the curve fitting. For space reasons, we do not plot the wavenumbers as a function of temperature. We just give in Tables 3 - 4 the temperature derivatives of the wavenumbers (3 co/a T)p, calculated from the data of Appendix 1, for the temperature intervals where they are constant. For both metasilicates, the low-frequency modes are more sensitive to temperature than the highfrequency modes. Hence, the distinction between internal and lattice modes is more clear cut than for minerals like the olivines and garnets investigated recently at elevated temperatures (Gillet et al. 1991, 1992). The observed (3 in ~oi/~ T)p are about - 2 10 -5 K -1 for the internal modes, and can be as low as - 1 5 10 .5 K -1 for the lattice modes. Even though the pressure dependence of the wave numbers would be required for a rigorous evaluation of anharmonic parameters ai=(Oln ojOT)v (see Gillet et al. 1991), the modes with the lowest (aln coi/ T)p values are suggestive of an unusually high intrinsic anharmonic character near room temperature.

162 Fig. 5 Raman spectra of lithium metasilicate at the temperatures indicated for the three spectral ranges used for the curve fitting. The numbers refer to the labeling of the modes given in Table 3. The solid lines through the experimental data represent the spectra obtained from the fitted peaks also shown

Li2Si03

1287 K

973 K

1096 K

658 K

658 K

926 K

500 K

500 K

574 K

13

8

298 K

3 2

5

300

298 K

1

15

,

t

200

K

298

6

17

~oo

i

I,

500

J

1

600

i

I

i

700

Wavenumber (cm "1)

800

900

1000

163 Fig. 6 Raman spectra of sodium metasilicate at the temperatures indicated for the three spectral ranges used for the curve fitting. The numbers refer to the labeling of the modes given in Table 4. The solid lines through the experimental data represent the spectra obtained from the fitted peaks also shown. Note the asymmetric shape of the Vl2 band at the highest temperatures

Na2SiO3

1348 K

1348 K

\

1028 K

\

799 K

1028 K

965 K

f L 799 K

774 K

\

2d_

576 K

12

8

1

17

298K

t7 K

298K

10 200

300

400

500

12 ,_.~..~ 600

700

Wavenumber ( crn-1 ]

500

, 700

151~ 900

18

164 Lithium metasilicate For most modes, the wavenumbers decrease linearly with temperature for the entire temperature ranges where the Raman bands could be observed (Table 3). The main exception is the v2 mode which is well resolved up the melting point and has a nonlinear (but smooth) temperature dependence (Table 3). Likewise, the linewidths of the bands increase regularly up to the melting point at 1474 K, typically by a factor of three (see Appendix 1). Exceptions to these smooth variations concern bands in the frequency range 300-450 cm -~, as apparent in the spectra plotted in Fig. 5 for a few representative temperatures. Between 500 and 650 K, the intensity of the v3 band decreases strongly and its linewidth increases. In addition, the 127 band disappears and the v4, v5 and 126 bands merge to form a broad peak. Once these changes have taken place, the width of the resulting band increases with further temperature rises, and this broad feature eventually overlaps on either sides the v3 and v8 bands. As a result, the frequency of the v 3 band increases anomalously from 770 K and its linewidth increases by a factor of six between 300 and 1130 K, the highest temperature at which this band was identified. In view of the difficulties of fitting the 300-400 cm-~ band with its individual components, the anomalous values of (0 o)/8 7)p that would be obtained at higher temperatures from the data of Appendix 1 are thus likely biased by the extensive band merging. The sudden increase of the linewidth of the v5 band when the v3 band disappears above 1130 K (Table A 1) is another example of this bias.

Sodium metasilicate The frequencies and linewidths of most Raman bands of Na2SiO3 also vary quasi-linearly with temperature (see Appendix 1). Relative to the strong S i - O stretching bands, the low-frequency bands are less intense for sodium than for lithium metasilicate (Figs. 5-6). As a result, a much smaller fraction of the bands could be followed as a function of temperature (compare Tables 3 and 4). Even though comparisons between the two series of spectra could be affected by this difference, spectral changes resembling those described in the preceding section for LizSiO 3 appear to take place also in Na2SiO 3, albeit in a more gradual and profound manner (Fig. 6). Below the strong vl2 S i - O - S i bending band at 589 cm -~ (at room temperature), the weakest N a - O vibrational bands disappear rapidly and only the three most intense bands can still be followed above 800 K. Then the linewidths of these bands continue to increase markedly to the point that, with the exception of the distinct vj and v7 bands near 170 and 380 cm -1, respectively, an almost featureless low-frequency spectrum is observed near the melting temperature of 1362 K. For the v7 mode, an unusual increase of the wavenumber with temperature is found. This temperature effect seems real because this band re-

1.02

'

'

'

I

'

'

'

I

'

'

'

I

'

'

'

I

'

o

1.01

'

'

~ 2\

'

'

V12o

Vt7 a

1.00

I

o

o~

o

o

o

-.. 0.99 ;:> " m.

0.98

'r

o

lit

V17 9

"Q. - EL

QO0 0

q

V17

0.97

- ~q~4:~ V13

0.96

a00

,

i

,

I

600

,

,

i

I

800

,

,

i

I

1000 T(K}

,

,

~

I

1200

r

r

,

I

1400

,

,

' 600

Fig. 7 Variations with temperature of the wavenumbers of analogous bands of Li2SiO3 (open symbols) and Na2SiO 3 (solid and dotted symbols). The linear shift of the v~3 Si- O-Si bending band of Li2SiO3 contrasts with the nonlinear variations of the two components of the v12 band of Na2SiO3. Data from Appendix 1

mains rather well separated from the neighboring modes at high temperatures. The other important temperature-dependent changes apparent in the spectra of Na2SiO 3 (Fig. 6) are observed for the two intense v12 and v17 modes at 589 and 966 cm -1 representing S i - O - S i bending and S i - O stretching, respectively. For the v12 mode, a strongly nonlinear variation of the wavenumber is observed with temperature (Fig. 7) and both Raman bands become increasingly asymmetric with a rapidly growing high-frequency tail as the temperature is increased (Fig. 6). Separation of the v17 band into two components, regardless of their interpretation, gives a convenient measure of the observed peak asymmetry. In this way, the high frequency shoulder observed at higher temperatures is readily accounted for by an additional v17 a band whose frequency is close to that of the v18 mode observed at lower temperatures. Note however that this shoulder unlikely represents this v 18 mode because it develops after a definite temperature interval where the Va8 mode is barely present. Accounting for the asymmetry of the v12 mode is less straightforward. Clearly apparent at lower temperatures (Fig. 6), the weak v~l mode can be separated from the much more intense v~2 mode only when a small Lorentzian character is assumed in the curve fitting. The problem is that one obtains in this way two largely overlapping bands whose relative intensities depend markedly on the assumed Lorentzian character. With a high Lorentzian character, in contrast, the main peak is more satisfactorily decomposed into a strong v~2 band and a less strong v12 a sideband whose "frequency" anomalously tends to increase with temperature. But in this case the quality of fit is not as good as it should be on the low-energy side because the weak Vmi mode is poorly accounted for. This could suggest that the v12 peak actually represents a single mode that becomes more and more asymmetric with increasing temperatures and cannot be fit with standard

165 symmetric lineshapes. In any case, the results of the latter procedure gives an objective measure of the relative importance of the high-frequency shoulder of the v12 band, as described above for the v17 band.

Discussion

Expansion at moderate temperatures Lithium and sodium metasilicates are made up of similar silicate chains whose Si206 repeats have room-temperature S i - O tetrahedral distances of 1.68 and 1.59 A for bridging and nonbridging oxygens, respectively. The main structural differences between these compounds affect the S i - O - S i angle, at 124 and 134 ~ for Li2SiO 3 and NaaSiO3, respectively, and the crystallographical site of the alkali element (Fig. 8). In Na2SiO3, sodium is bonded to four oxygens forming a distorted tetrahedron, whose N a - O distances range from 2.28 to 2.40 A, and it achieves a five-fold coordination through an additional bonding with an other oxygen at 2.55 ]~ (McDonald and Cruickshak 1967). In Li2SiO 3 (Hesse 1977), the analogous trigonal bipyramid has shorter tetrahedral L i - O distances in the range 1.94-2.17 A and a fifth L i O distance at 2.76 A, which is so much greater than the other L i - O distances that lithium is usually considered in tetrahedral coordination. Because the mechanisms of thermal expansion of the alkali metasilicates cannot be determined rigorously from X-ray powder diffraction experiments, we will use the existing data for related chain silicates as a starting point for discussing our results. For a variety of monoclinic pyroxenes, single-crystal diffraction experiments at high temperature have shown that the S i - O bond lengths remain practically constant, the non silicate polyhedra MO 6 or MO 8 ( M = C a , Mg, A1, Na, etc.) expand markedly, and the resulting mismatch between the S i O 4 and MO~ polyhedra is accommodated by rotation of the tetrahedra around the nonbridging oxygens and outof-plane tilting of the tetrahedra (Cameron et al. 1973). As a matter of fact, MO distances expand at a rate comparable to that of the unit-cell parameters (Cameron et al. 1973). In view of the weak nature of alkali oxygen bonds, thermal expansion of Li2SiO 3 and Na2SiO 3 should stern mainly from expansion of the L i Q and NaO5 polyhedra and concomitant rotation of the S i O 4 tetrahedra. For lithium metasilicate, the X-ray diffraction data do not reveal significant structural changes with temperature. The expansivity is almost constant and at lower temperatures is the greatest along the a and b axes (Fig. 1). Above 1000 K, expansion along the c axis increases markedly (Fig. 1) and results in a slightly enhanced volume expansivity (Fig. 9). By comparison, expansion along the a axis is consistently lower than along the b axis for all the pyroxenes investigated by Cameron et al. (1973). This expansion contrast is associated with differences in the linkage of Si206 chains. These chains are linked in the a,

C e n t e r e d position

No2Si03

t

Li2SiO 3

Fig. 8 Interconnection of the Si206 chains (grey tetrahedra) with' the alkali polyhedra (in white) as seen in the 001 plane and direc-i tions of preferential dilations of the Na05 and LiO5 bipyramids as represented in the 100 plane 1.100'

' ' i , , ,

1.080 Li2 Si03 >o

i~

1.060

>

i

1.0Z,0

i ~

1.020

m

Na2SJO3

I (~

J 1.000 =~ , u , I, 300 500

D ,,

1, 700

000 T(K)

1100

1300

1500

Fig. 9 Relative increases of the volume of lithium and sodium metasilicates as a function of temperature

b plane by M O 6 octahedra in pyroxenes, and by a doublq chain of interconnected LiO 4 tetrahedra in lithium metasilicate. ] For sodium metasilicate, expansion along the a and 0 axes is also the same, but the unexpected feature is th6 almost zero expansivity along these axes above 850 I~ which contrasts with a unusually high expansion along the c axis (Fig. 3). The unit-cell ratio a/b of the tw~

166 metasilicates is almost the same (1.73). The expansion difference between Li2SiO 3 and Na2SiO3 correlates with differing c/a ratios, which are 0.50 and 0.46, respectively, and result in the aforementioned coordination difference between Li and Na. For Na2SiO3, the dilation stems from expansion of the polar axis of NaO 5 bipyramids. For Li2SiO3, where polar elongation of the LiO 5 bipyraraids is high enough that Li is viewed tetrahedrally coordinated (Fig. 8), expansion takes place through the equatorial plane of the LiO 5 polyhedra. The strong expansion of Si206 chains indicated by the X-ray data is associated with the anisotropy in the expansion of NaO 5 bipyraraids. This eventually results in a symmetry change due to the loss of an atomic position centered in the a, b plane, pointing to a change in the linkage of the Si206 chains by NaO 5 polyhedra. Not surprisingly, it is the v12 S i - O - S i bending mode which is affected significantly by this symmetry loss, the temperature dependence of this mode showing a marked break at about 850 K (Fig. 7). Discontinuities are not apparent in the variations with temperature of the main interplanar spacings, of the Raman frequencies and also of the enthalpy (Naylor 1946; Richet et al. 1984). Thus the symmetry change indicated by the X-ray diffraction patterns seems associated with a slight X-type phase transition, and not with a marked first-order transition. In any case, the unusual feature of this transition is a temperature-induced lowering of the symmetry of the crystal.

Alkali mobility at lower temperatures The Raman spectroscopic data show losses of detail at low frequencies, especially for sodium metasilicate, as a result of the increasing linewidths with temperature. Below 500 cm -1, most of the observed Raman bands represent alkali-oxygen stretching and a few bands O - S i - O bending. At higher temperatures, only two bands are unambiguously assigned in that part of the Raman spectra for both metasilicates. These are two most intense modes mainly associated with O - S i - O bending, namely v2 and v8 for LiaSiO 3 and v I and v7 for Na2SiO 3. The extensive merging due to band broadening thus affects essentially the L i - O and N a - O stretching modes. Since cation disordering is usually evidenced by large increases of linewidths (see White 1974), the spectral changes visible at low frequencies could reflect extensive disordering of alkali elements or averaging of their crystallographical sites which could remain undetectable in the X-ray powder diffraction patterns. This interpretation is borne out by NMR measurements which indicate a tremendous increase from about 720 K of 7Li spin-lattice relaxation times in Li2SiO 3 (Matsuo et al. 1983). In the same way, 29Si NMR experiments on a Li2SiOg-Li4NiO4 mixture have shown a reduced anisotropy of the chemical shift due to Li motion at temperatures as low as 670 K and just a "hint" of anisotropy at 1270 K in view of the short lifetime of S i - O bonds (Farnan and Stebbins 1990).

250

i i i J [ r l l l

ol

I

I

200 O

-5

150

NS

E

o._ 1 0 0 i (._)

L.."

o

oo~

~176176176

ooOOooo~ 1 7 6 oo

I I

I

f..."

I

50

--'-Fl~ 7i-~ ~-F"1-Vi~

300

600

900

I i I i f i

1200

1500

T(K) Fig. 10 H e a t c a p a c i t i e s o f s o d i u m and l i t h i u m m e t a s i l i c a t e s . E x p e r i m e n t a l data (symbols) from T6qui etal. (1992) and Richet etal.

(1994), and heat capacity differences between NaaSiO3 and Li2SiO3 as given by the experimental data (dashed curve) and as calculted from the room-temperatureRaman spectra (solid curve). As discussed in Appendix 2, the difference between these two curves seems to represent an excess heat capacity related to sodium mobility in Na2SiO3

As discussed in Appendix 2, the heat capacities of Li2SiO 3 calculated from the room-temperature vibrational frequencies of Table 3 are consistent with the calorimetrically determined values in the range 3001000 K plotted in Fig. 10 (when using guestimated bulk moduli and sound velocities). From a calorimetric standpoint, the changes apparent in the low-frequency part of the Raman spectra or in NMR spectra of lithium silicates thus do not seem to have a heat capacity counterpart. In the same manner, the high mobility of Li in pure Li20 (Matsuo et al. 1983) is not associated with anomalous heat capacity changes (see Dworkin and Bredig 1968). This does not appear to hold true for Na2SiO 3, however, for which an excess Cp of about 7 J/mol K is likely found below 1000 K (Fig. 10). Consistent with the more pronounced temperature effects on the Raman spectra, this calorimetric anomaly indicates that the changes affecting the alkali element are still more pervasive in Na2SiO 3 than in Li2SiO 3. Below 700 K the spectral differences between Li2SiO 3 and Na~SiO3, already apparent at room temperature, thus unlikely stem from lower Raman scattering cross sections for N a - O compared with the L i - O stretching modes.

Premelting Calorimetric measurements indicate that premelting effects are limited to a few degrees at most for LizSiO3 (T6qui et al. 1992; Stebbins et al. 1984). As noted above, the structural changes due to lithium motion at high temperatures are not associated with thermal expansion or calorimetric anomalies. For Na2SiO3, thermal expansion

167 Table 4 Raman-active vibrational modes of sodium metasilicate (data in cm-~ and K)

Mode V1 V2 V3

is the highest temperature up to which a linear temperature dependence of the wavenumber is observed, T,~ax that at which this mode could be observed. For modes having a nonlinear temperature dependence (nl), the listed ~c0/0 T refers to room temperature b Data of Brawer and White (1975) included for comparison c t%=172.5-14.0 10-3T -9.45 10-6T 2 up to 1282 K ~ ~o5=232.5+31.0 10-3T -81.0 10-6T2 up to 799 K e 6om=508.5+3.47 10-3T -20.8 10-6T2 up to 1028 K

v4 v5

a rlin

0.70

,

i

J i

i

t

J

i

,

i

i

v7

V8 V9

v~~ v~2 vI 3d

rlin a

rnnax a

Na-O str, O - S i - O bend Na-O str

168 184 200 215 235 264 301 395 433 510 549 588 607 716 750 865 914 965 1014 1063

-19.7+_ 1.9 -21.6+_ 1.4

nlc 1236

1356 1282

167 185

-17.2_+9.4 -44.3+_2.1 - 9.4_+1.3 -27.7+-0.5

nld

1028 799 I356

799 1290 1356 1356

235 263 303 397

- 8.9+_1.5 -13.6_+0.9 - 8.0+_2.1 - 17.5 +_8.7

nle 1028 1028 643

1356 1315 1356 643

507

-23.9_+5.8

1223

1223

880

-20.6+-1.2 -10.1+_2.9

1223 1113

1223 1198

973 1018 1083

Si-O str

VI 4

vH '

Si-O Si-O Si-O Si-O Si-O

vI5 v16 ]]17

1218 V~9

,

103 ~O)/~T

Si-Ostr, Si-O-Sibend O - S i - O bend, Si-O str S i - O - S i bend, Si-O str

Vll

,

~o(298 K)

Na-O str Na-O str O - S i - O bend, Na-O str Na-Ostr O - S i - O bend

V6

i

Assignment a

,

i

i

i

i

i

str str, O - S i - O bend str str, O - S i - O bend str

~ i

i

i

,

,

0.60 0.50 .o_ -5 0.40

-

NS

E~

2 o.3o 0.20 0.10 0 , 200

,

,

I

,

400

,

,

I

,

600

,

,

q

,

,

,

I~;--,~q~

,

,o,~,

,

800 1 0 0 0 1 2 0 0 1400 1600 T(K)

Fig. 11 Area ratios of the main v~2 and v~7 bands and sidebands of sodium metasilicate as a function of temperature (solid symbols). For comparison (open symbols), we have included the results obtained when fitting in the same manner the analogous v~3 and vw bands of LiaSiO3 with a high-frequency sideband. Except at the highest temperature for the v17band (see Fig. 12), the areas of such sidebands are barely significant

first changes markedly along the a and b axes and a symmetry element is lost in the a, b plane at about 800 K, above which N a - O stretching modes can no longer be followed in the Raman spectra (Table 4). In the high-frequency part of these spectra, then one observes that the areas of the v12 a and v17 a sidebands increase rapidly at around 1200 K relative to those of the most intense S i O - S i (v12) and S i - O (v~7) bands (Fig. 11). In addition, the temperature dependence of the v12 band changes at about the same temperature, an effect that is not apparent for the analogous v13 band of Li2SiO 3 (Fig. 7). The onset of premelting being observed at this temperature of 1200 K in calorimetric measurements (Naylor 1944; Richet et al. 1984), the changes in the v12 and v17 bands apparent in Figs. 8 and 10 constitute Raman spectro-

0~b

587

scopic signatures of premelting in N a 2 S i Q . The rest of this discussion will thus be devoted to this compound. The Raman spectroscopy observations indicate that the onset of premelting is associated with local structural modifications affecting the silicate framework, which could be made possible through the changes in the chain interlinkage observed by X-ray diffraction in the a, b plane at around 800 K. The changes in the high-temperature Raman spectra of Na2SiO 3 lend themselves to a simple interpretation according to available calculations of the vibrational modes of Si206 chains (Brawer 1975; Furukawa et al. 1981). The frequency of the intense v17 S i O stretching band at around 950 cm-1 is nearly independent of the S i - O - S i angle between adjacent tetrahedra and varies only through intratetrahedral distortion. At around 580 cm 1, the other intense v12 band represents a combination of S i - O - S i bending and S i - O stretching and its frequency increases markedly when the S i - O - Si angle decreases. Within this scheme, the broadening of the vls band points to strong distortions of the SiO 4 tetrahedra in the Si20 6 chains. As indicated by the asymmetric broadening of the v12 band, the distortion is associated with a broader distribution of S i - O - S i angles that is skewed toward smaller values of this angle. Hence, the increase of the mean S i - O - Si angle responsible for the high expansivity along the c axis (Fig. 3) is clearly not uniform. Faced with a general lack of structural data at high temperatures, Richet et al. (1994) suggested that cation disordering over the various crystallographical sites of the structure could generally account for the calorimetric effects of premelting. Based on the example of anorthite in which Si, A1 disordering near the melting point has long been recognized, this hypothesis cannot be tested by our X-ray diffraction results. It is not contradicted either by the Raman spectra which are almost devoid of details in the spectral region of N a - O stretching vibrations (Fig. 6), even though this loss of structure in the spectra

168 100-- Na2SiO 3ccystat

100 -- Li2SiO3 crystal

75--

75--

50

~ 5O

25 --

25 --

o

~ I

~

o-

I

I

I

I

i

100 -- No 25iO3 melt

75

i

I

I

I

1~176 I

Liz SiO3 mett

- -

i

~

so I

25--

50

- -

/

25

0L, _

0

I

I

I

I

I

I

I

I

I

486

647

809

970

1131

486

647

809

970

Wavenumber (cm-1)

Wovenumber

_1 1131

(cm-1}

Fig. 12 Comparison between the Raman spectra of the crystals right before melting with those of the liquids right after melting

peratures where the relevant modes have indeed a high intrinsically anharmonic character.

begins at temperatures much lower than the premelting range. From the extensive disordering affecting the SilO 6 chains, one can nevertheless assume that sodium atoms must be loosely bound to the silicate framework. That Li2SiO 3 still shows a broad, but well-defined envelope in the low-frequency region up to melting (Fig. 5) corroborates significant differences in the alkali environments between the two metasilicates. Richet et al. (1994) also suggested that such temperature-induced cation disordering could be at the root of intrinsic anharmonicity in oxide and silicate compounds. This link is consistent with the fact that the spectral changes associated with alkali vibrations are considerably enhanced in the premelting range. In this respect, we note that these changes already manifest themselves at much lower tern-

Melting A great deal of work has been devoted to determining silicate glass structures by vibrational spectroscopy (cf. Mysen 1988). Most efforts have been made to account for the high-frequency part of the Raman spectra in terms of the distribution of bridging and nonbridging oxygens among the S i O 4 tetrahedra which defines the so-called Q~ species, where n denotes the number of bridging oxygens linked to a given Si atom. Recently, the variation with temperature of the abundance of the Q species has been determined for alkali silicate melts from Raman spectroscopy measurements (Mysen and Frantz 1992, 1993, 1994). In this section, we will confront these observations with those made on crystals immediately be-

169 160

Fig. 13 Comparison between the Raman spectra of Li2SiO3 and Na2SiO3 glasses. The lowfrequency parts are enlarged in the insets

I

I

[

I

i

[I

%0

/

120

/

//

jr

///"

lOO

/t~

Li2Si03

J o/2

2o0 ' 360 ' ~6 9-'r c

a0

t 6O ,e- M \ , ,

~0 s

]

,v-

Na2Si03

20

200

300

01~--200

low melting and we will also discuss briefly the changes affecting the low-frequency region that has been hitherto little studied. At high frequencies, there is a rather close correspondence between the Raman spectra of crystalline and liquid Na2SiO 3 at temperatures near the melting point (Fig. 12). The most intense band is near 950 cm -1 for both phases. A strong shoulder at about 1020 cm-1 in the crystal has its counterpart at around 1030 cm -1 in the liquid, There is another correspondence between the 850 and 826 cm-1 shoulders in the crystal and liquid, respectively. Excluding an analogous correspondence between the 950 cm -1 bands, such similarities are less obvious for Li2SiO 3 crystal and liquid. The 820 cm -~ band observed for the crystal has its counterpart at 840 cm-1 in the liquid, but the large high-frequency shoulder of Li2SiO3 liquid at 1050 cm -1 has an almost insignificant analogue in the crystal. Regardless of specific interpretations, the changes observed for Na2SiO 3 and the contrasting behavior of the two metasilicates demonstrate that the spectral changes associated with premelting are premonitory effects of the structural changes that take place on actual melting. As illustrated in Fig. 12, the wide band found at highfrequencies in the Raman spectrum of alkalisilicate liquids can be deconvoluted into several bands. Up to 33mo1% Na20, bands at around 950, l l 0 0 and l l 5 0 c m -1 are observed. They are assigned to S i - O stretching vibrations in Q2, Q3 and Q4 species, respectively, whose relative abundances can be determined from the areas of the bands (e.g., Mysen et al. 1982 a; Mysen and Frantz 1992, 1993). In crystalline metasilicates, all the Si atoms are in Qa species. That the most

400

/-,-00

500

600 800 Wovenumber" (crn-~)

1000

1200

intense band in both metasilicate liquids is found near 950 cm 1 is thus expected from the stoichiometry of the melt. Likewise, the correspondence between the large S i - O - S i band at about 600 cm 1 is not a surprise as S i - O - Si bending modes of Si206 chains are expected at these frequencies (Furukawa et al. 1981). The correspondence between the other high-frequency bands in melt and crystal is more intriguing and suggests that local distortion of the silicate chains of the crystal near the melting point could result in the existence of short-lived silicate entities resembling those characteristic of the liquid state. Unfortunately, assignments of these other bands are less straightforward because melts with more! than 33 mol% Na20 have not been extensively investigated spectroscopically. We will thus refrain from interpreting these bands and just note that, since Brawer's (1975) original calculations, the width of the S i - O stretching envelope has been used as a measure of the silicate disorder in glasses and melts. The width of the high frequency peak is greater for Li2SiO 3 than for Na2SiO 3, as also observed at lower alkali contents (e.g., Brawer and White l 1975). This greater disorder thus contrasts with the i premelting difference between the metasilicates. In fact, the rather close correspondence between the I Raman spectra of crystalline and liquid Na2SiO 3 could bej misleading. Calorimetric measurements indicate that i premeling represents only 12% of the enthalpy and en-i tropy differences between the liquid and the "unper- I turbed" solid phase. In other words, the Raman spectral, changes do not scale with the major thermochemical! changes which accompany the loss of long-range orderl detected at the congruent melting point through X-ray! diffraction. Raman spectroscopy is thus a much morel

170 sensitive probe of premelting than of melting. One could nevertheless assume that some information regarding the melting process might be found at low frequencies. For instance, the broad L i - O feature of Li2SiO 3 indeed disappears on melting. This change is not due to increased background in the spectra of the liquid, for rapidly quenched Li2SiO 3 glass shows at r o o m temperature only very weak peaks between the silicate bands which are similar to those observed in quenched Na2SiO 3 glass (Fig. 13). As discussed above, however, the loss of detail in the spectra of Na2SiO 3 is associated with a minor therm o c h e m i c a l anomaly as c o m p a r e d to that taking place in the premelting range (cf. Fig. 10). The main change observed at low frequencies on melting could be the disappearance of the m o d e at around 200 c m -1 which is a combination of alkali-oxygen stretching and O - S i - O bending, and is not observed in the glass spectra.

Conclusions Richet et al. (1994) suggested that the anomalous increases in heat capacity observed in the premelting range should also obtain for other second-order t h e r m o d y n a m ic properties such as the thermal expansion or the compressibility. The volume data for Na2SiO 3 indicate that this is not necessarily the case, barring a surprisingly high influence of defects on the molar volume in the premelting range which could be detected by dilatometry but not by X-ray diffraction experiments. In this respect, premelting could thus differ markedly f r o m X-type transitions where volume and enthalpy effects are generally correlated. If it is beyond the scope of this paper to attempt quantitative models of premelting and melting, it nevertheless appears that correlations between thermodynamic and structural properties are possible at temperatures up to the melting point of silicates. The differences between the isostructural compounds Li2SiO 3 and Na2SiO 3 also show that premelting depends sensitivity on bonding energies. As discussed by Dworkin and Bredig (1968), high-temperature calorimetric data reveal a similar difference for Na20 and Li20 near their estimated melting points.

Table A I T(K)

Wavenumbers

%L a v 1

and linewidths of the low-frequency

va

298 100 206 8 232 500 90 204 8 230 658 70 195 8 228 815 70 196 6 227 973 60 191 8 224 1130 60 188 9 221 1287 50 217 1406 50 213 1445 40 212 1468 30 211

v3

5 5 6 7 9 11 18 23 22 20

293 285 273 268 272 280

v4

v5

13 326 6 347 7 22 321 8 343 16 36 338 37 77 324 26 105 317 23 119 315 25 312 69 308 68 305 65 313 77

As emphasized previously (Richet et al. 1994), the structural changes experienced by silicates in the premelting range are significant, rapid, reversible, and essentially unquenchable. Our results thus point again to the need for structural measurements near the melting point. The example of the alkali metasilicates shows that just below melting the high-frequency part of the R a m a n spectrum of a solid material can bear strong resemblance with that of the melt. As investigated by R a m a n spectroscopy, the structure of the silicate f r a m e w o r k can thus be similar in a crystal and in a melt at scale of a few interatomic distances. Even for a compound with strong premelting effects like Na2SiO3, the local deformation and distortion of this f r a m e w o r k in the premelting range are not detectable by powder X-ray diffraction. This technique records essentially the loss of long-range order an actual melting, which is as abrupt for silicates as for other classes of compounds. A more detailed picture of premelting through element specific probes could however provide strong constraints on the nature of the cooperative changes responsible for melting which involve both the silicate f r a m e w o r k and the network modifying cations.

Acknowledgements We thank J.R Iti6 for his work in setting up the X-ray diffraction line at LURE; A. Pierre helped with heating cells, D. Neuville with software; A. Polian provided us with his deconvolution program of diffraction patterns. We also thank I. Farnan and J. Ingrin for helpful discussions; and Y. Bottinga, D. Neuville and anonymous reviewers for comments. Work supported by NSF grant EAR-9218890 and the CNRS-CIW program PICS n~ 192.

Appendix 1 We list in Tables A l - A 4 the wavenumbers and full widths at half height (linewidths) of the R a m a n bands. Modes which were observed at r o o m tempeature only have not been included. For space reasons, we list less than half of the recorded values (a complete list can be obtained from the first author). All figures have been rounded to the nearest integer, even though smooth variations of the first digit after the decimal point were gener-

vibrational modes of lithium metasilicate I]6

362 360 356 353 351 351 355 354 354 358

Percentage of Lorentzian component in the fitted lineshapes

v7

5 6 10 27 38 47 40 38 39 17

397 387 376 377

v8

15 23 34 28

410 407 404 401 398 395 393 390 390 389

%L

rio

vii

1112

v13

5 100 498 15 526 5 586 8 611 9 6 100 488 40 524 7 584 8 610 12 7 100 521 9 580 9 607 17 9 90 517 11 575 10 604 20 11 80 514 12 568 11 601 23 13 80 511 15 565 11 597 27 16 70 508 18 595 30 17 60 505 18 592 32 17 60 504 19 592 33 18 60 502 32 593 34

171 Table A2 Wavenumbers and linewidths of the high-frequency vibrational modes of lithium metasilicate

Table A4 Wavenumbers and linewidths of the high-frequency vibrational modes of sodium metasilicate

T(K)

%L a

v15

T(K)

%L a

vls

298 574 752 926 1096 1263 1344 1425 1473

100 100 100 100 80 80 80 80 80

841 838 835 832 828 824 823 821 820

298 495 774 965 1217 1259 1291 1316 1332 1348

100 100 100 100 100 100 100 100 100 100

868 862 850 856 848 848 849 846 847

11 12 17 24 33 28 36 40 43

vl6

vlv

921 10 908 9 905 10 904 5 906 5

975 972 968 965 961 956 954 952 952

v18 13 16 19 22 26 26 28 30 32

1028 1025 1015 1015

11 12 61 52

vj7 5 11 2 8 11 5 12 15 8

965 962 958 955 950 949 948 948 947 947

vjTab 6 10 15 19 27 29 30 32 33 39

vls 1015 1012

1011 1009 1003 1009 1005 1007 1012 1013

12 14

84 94 80 124 111 133 135 86

a Percentage of Lorentzian component in the fitted lineshapes a Percentage of Lorentzian component in the fitted lineshapes

b Band introduced as a measure of the asymmetry of the main Vl7 band, see text Table A3 Wavenumbers and linewidths of the low frequency vibrational modes of sodium metasilicate T(K)

%L a

vl

298 485 643 799 923 1028 1198 1223 1290 1315 1339 1356

100 90 80 70 50 50 40 30 30 20 0 0

167 164 160 155 151 147 142 141 140 140 141 139

v2 4 5 6 9 10 13 15 17 22 24 33 31

183 180 179 172 169 167 164 165

5 12 10 15 20 22 15 12

v5

v6

234 7 229 10 218 12 206 15

263 257 248 240 236 232 232 231 228

v7 12 15 23 23 30 25 18 18 21

302 299 298 298 305 301 304 302 300 302 301 305

v8 14 24 33 36 47 47 32 31 25 25 21 21

394 391 387 383 378 374 370 370 368 368 368 368

15 18 23 28 28 29 29 30 31 31 29 30

%L

VIO

100 100 90 80 80 60 50 50 40 40 30 30

508 505 502 498 494 490 486 485 484 484 483 480

•11

8 9 10 12 12 15 15 15 16 15 13 18

549 547 545 543 541 539 532 533 531 531

Vl2ab

V12

4 5 4 8 10 18 13 15 16 11

589 587 585 583 584 587 587 587 586 586 585 585

10 13 19 24 29 32 34 35 35 35 35 35

637 642 645 641 640 629 631

26 28 28 30 36 48 42

a Percentage of Lorentzian component in the fitted lineshapes b Band introduced as a measure of the asymmetry of the main v~2 band, see text

ally o b s e r v e d for the w a v e n u m b e r s as a f u n c t i o n o f temperature. T h e (Oc0/~ T)p values listed in Tables 3 - 4 have thus b e e n c a l c u l a t e d w i t h the n o n - r o u n d e d n u m b e r s for the whole data set.

and Li2SiO 3 (see Fig. 19). H e n c e , in this t e m p e r a t u r e interval the data plotted in Fig. 10 suggest that Na2SiO 3 has an excess Cp of a b o u t 7 J / m o l K over Li2SiO 3 w h i c h is not a c c o u n t e d for b y d i f f e r e n c e s in v i b r a t i o n a l f r e q u e n cies.

Appendix 2 References I n Fig. 10, the d i f f e r e n c e b e t w e e n the isochoric heat cap a c i t y (Cv) of s o d i u m and l i t h i u m m e t a s i l i c a t e s has b e e n c a l c u l a t e d with K i e f f e r ' s (1982) m o d e l f r o m the R a m a n v i b r a t i o n a l f r e q u e n c i e s o b s e r v e d at r o o m t e m p e r a t u r e . F o r the o p t i c a l m o d e s , we have used five discrete highf r e q u e n c y E i n s t e i n oscillators and one c o n t i n u u m (ext e n d i n g f r o m 168 to 589 c m - ~for NaaSiO3, and f r o m 207 to 613 c m -1 for Li2SiO3). We have n e g l e c t e d the acoustic m o d e s b e c a u s e acoustic velocities s e e m u n k n o w n for both c o m p o u n d s . T h e acoustic f r e q u e n c i e s should be in fact lower for Na2SiO 3 t h a n for Li2SiO 3. This could exp l a i n at least p a r t of the o b s e r v e d Cv d i f f e r e n c e , b u t only below r o o m t e m p e r a t u r e w h e r e Cp-Cvis u s u a l l y n e g l i g i ble. B e t w e e n r o o m t e m p e r a t u r e and 800 K, c a l c u l a t e d Cv and Cp d i f f e r e n c e s should n o t differ greatly b e c a u s e o f the s i m i l a r t h e r m a l e x p a n s i o n c o e f f i c i e n t s of Na2SiO 3

Brawer SA (1975) Theory of the vibrational spectra of some network and molecular glasses. Phys Rev B l 1:3173-3194 Brawer SA, White WB (1975) Raman spectroscopic investigation: of the structure of silicate glasses. I. The binary alkali silicates. J Chem Phys 63:2421-2432 Cameron M, Sueno S, Prewitt CT, Papike JJ (1973) High-tempera- : ture crystal chemistry of acmite, diopside, hedenbergite, i jadeite, spodumene, and ureyite. Am Mineral 58:594-618 i Devarajan V, Shurvell HF (1977) Vibrational spectra and normal l coordinate analysis of crystalline lithium metasilicate. Can J! Chem 55:2559-2563 Dworkin AS, Bredig MA (1968) Diffuse transition and melting in! fluorite and anti-fluorite type of compounds; heat content ofl potassium sulfide from 298 to 1260 ~ J Phys Chem 72: 12771281 Farnan I, Stebbins JF (1990) High-temperature e9si NMR investi-: gation of solid and molten silicates. J Am Chem Soc 112: 32-! 39

172 Furukawa T, Fox KE, White WB (1981) Raman spectroscopic investigation of the structure of silicate glasses. III. Raman intensities and structural units in sodium silicate glasses. J Chem Phys 75:3226-3237 Gillet Ph, Richet P, Guyot F, Fiquet G (1991) Anharmonicity and high-temperature thermodynamic properties of forsterite. J Geophys Res 96:11 805-11 816 Gillet Ph, Fiquet G, Mal6zieux JM, Geiger C (1992) High-pressure and high-temperature Raman spectroscopy of end-member garnets: pyrope, grossular and andradite. Eur J Mineral 4:651664 Hatton WE, Hildebrand DL, Sinke GC, Stull DR (1959) Thermodynamic properties of KOH, K2CO3, K2Si205, Li2SiO3, Li2Si205, and Ca(OH)2. Thermal Lab, Dow Chemical USA Rep. SSR 249-954 Hesse KF (1977) Refinement of the crystal structure of lithium polysilicate. Acta Cryst B 33:901-902 Kelley KK (1939) The specific heats at low temperatures of crystalline ortho, meta, and disilicates of sodium. J Am Chem Soc 61: 471-473 Kieffer SW (1982) Thermodynamics and lattice vibrations of minerals: 5. Applications to phase equilibria, isotopic fractionation, and high-pressure thermodynamic properties. Rev Geophys Space Phys 20:827-849 Konijnendijk WL, Stevels JM (1976) Raman scattering measurements of silicate glasses and compounds. J Non-Cryst Solids 21:447-453 Kracek FC (1930 a) The system sodium oxide-silica. J Phys Chem 34:1583-1598 Kracek FC (1930b) The binary system L i 2 0 - S i O 2. J Phys Chem 34:2641-2650 Long DA (1977) Raman spectroscopy. McGraw Hill, New York, 277 pp McDonald WS, Cruickshank DWJ (1967) A reinvestigation of the structure of sodium metasilicate. Acta Cryst 22:37-43 Matsuo T, Ohno H, Noda K, Konishi S, Yoshida H, Watanabe H (1983) Nuclear magnetic resonance investigation of lithium diffusion in Li20, Li2SiO3 and LiA10~. J Chem Soc Farad Trans 2, 79:1205-1216 Mysen BO (1988) Structure and properties of silicate melts. Elsevier, 354 pp

Mysen BO, Frantz JD (1992) Raman spectroscopy of silicate melts at magmatic temperatures: Na20-Si2, and binary compositions in the temperature range 25-1475 ~ Chem Geol 96:321-332 Mysen BO, Frantz JD (1993) Structure and properties of alkali silicate melts at magmatic temperatures. Eur J Mineral 5: 393407 Mysen BO, Frantz JD (1994) Alkali silicate glass and melt structure in the temperature range 25-1651 ~ at atmospheric pressure and implications for the mixing behavior of structural units. Contrib Min Petr 117:1-14 Mysen BO, Virgo D, Seifert FA (1982 a) The structure of silicate melts: implications for chemical and physical properties of natural magma. Rev Geophys 20:353-383 Mysen BO, Finger LW, Seifert FA, Virgo D (1982b) Curve-fitting of Raman spectra of amorphous materials. Am Mineral 67:686-696 Naylor BF (1946) High-temperature heat contents of sodium metasilicate and sodium disilicate. J Am Chem Soc 67: 466467 Richer P, Fiquet G (1991) High-temperature heat capacity and premelting of minerals in the system M g O - C a O - A 1 2 0 3 SiO 2. J Geophys Res 96:445-456 Richet P, Bottinga Y, T6qui C (1984) Heat capacity of sodium silicate liquids. J Am Ceram Soc 67:C 6-C 8 Richet R Ingrin J, Mysen BO, Courtial R Gillet Ph (1994) Premelting effects in minerals: an experimental study. Earth Planet Sci Lett 121:589-600 Stebbins JF, Carmichael ISE, Moret LK (1984) Heat capacities and entropies of silicate liquids and glasses. Contrib Min Petr 86:131-148 T6qui C, Grinspan R Richet P (1992) Thermodynamic properties of alkali silicates: heat capacity of Li2SiO3 and lithium-bearing melts. J Am Ceram Soc 75:2601-2604 Ubbelhode AR (1978) The molten state of matter. John Wiley and Sons, New York West AR (1977) Long-period superstructure in sodum-lithium metasilicates. Acta Cryst A 33:408-411 White WB (1974) Order-disorder effects. In: The infrared spectra of minerals. Farmer VC (ed.), Mineralogical Society, London, p 87-109