Abstract:
Some physical properties of ideal solutions, e.g. the molar volume and the molar refraction,
vary linearly with composition. Others can be expressed, either as ratios or as products of
two other properties which vary with composition in this way. It is postulated that the nonideal
behaviour of real solutions can be adequately modelled by substituting these linear
functions with higher order Scheffé polynomials. A suite of such models is presented for
which the parameters are fully determined by knowledge of pure component properties and
binary behaviour. Their binary data representation ability, and capacity to predict ternary
properties, was tested using density and refractive index data for the acetic acid–ethanolwater
ternary system as well as fourteen additional ternary data sets. Model performance
was ranked on the basis of the Akaike Information criterion. With respect to predicting
ternary density and refractive index behaviour from knowledge of binary data, it was found
that lower-order models outperformed higher order models.
