TEMPERATURE INDEPENDENT THERMAL EXPANSIVITIES OF SODIUM ALUMINOSILICATE MELTS BETWEEN 713 AND 1835 K

Abstract

The thermal expansivities of eight sodium aluminosilicate liquids were derived from the slope of new volume data at low temperatures (713-1072 K) combined with the high temperature (1300-1835 K) volume measurements of Stein et al. (1986) on the same liquids. Melt compositions range from 47-71 wt% SiO2, 0-31 wt% Al2O3, and 17-33 wt% Na2O; the volume of albite supercooled liquid at 1092 K was also determined. The low r temperature volumes were derived from measurements of the glass density of each sample at 298 K, followed by measurements of the glass thermal expansion coefficient from 298 K to the respective glass transition interval. This technique takes advantage of the fact that the volume of a glass is equal to the volume of the corresponding liquid at the limiting fictive temperature (T'f), and that T'f can be approximated as the onset of the rapid rise in thermal expansion at the glass transition in a heating curve (Moynihan, 1995). No assumptions were made regarding the equivalence of enthalpy and volume relaxation through the glass transition. The propagated error on the volume of each supercooled liquid at T'f is ~0.25%. Combination of these low temperature data with the high temperature measurements of Stein et al. (1986) allowed a constant thermal expansivity of each liquid to be derived over a wide temperature interval (763-1001 degrees) with a fitted 1σ error of 0.6-4.6%; in every case, no temperature dependence to dV/dTliq could be resolved. Calibration of a linear model equation leads to fitted values +/-1σ (units of cm3/mole) for VSiO2 (26.91 +/- .04),VAl2O3 (37.49 +/- .12), VNa2O (26.48 +/- .06) at 1373 K, and dVNa2O/dT (7.64 +/- .08 x 10-3 cm3/mole-K). The results indicate that neither SiO2 nor Al2O3 contribute to the thermal expansivity of the liquids, and thatdV/dTliq is independent of temperature between 713-1835 K over a wide range of liquid composition. Calculated volumes based on this model recover both low and high temperature measurements with a standard deviation <0.25%, whereas values of dV/dTliq can be predicted within 5.6%.