Abstract
The predictive error relative to analytic values of entire-curve Wagner constants is studied for the reduced vapor pressures predicted by Wagner constants that are parameterized from a limited data interval. An algebraic solution for the fully-determined case based on only four data points is used to estimate the limited-data Wagner constants. First, seventy-two species are used to assess the impact of the location of the two interior points and the location and width of the limited-data interval upon the error in predicted Pv,r due to data imprecision. Hydrogen, helium, R152a, and water are used to assess error due to Wagner imperfection and compare predictive capability of the algebraic fully-determined and regressed over-determined approaches. Second, the repeatability/reproducibility of VLE data in the literature is studied by comparing reduced pressures calculated from Antoine constants applicable to a limited temperature interval with the entire-curve Wagner analytic values over the same limited-data interval. The entire-curve Wagner analytics are treated as “true” or “best” values and the Antoine analytic values as surrogate experimental data. Wagner constants for fifty-five species are subsequently estimated from the Antoine analytics for the fully-determined case, from which reduced vapor pressures below and above the interval are predicted and compared with the entire-curve Wagner analytics to estimate the ability of limited VLE data to be used to accurately represent the entire two-phase curve. The predictive capability of such limited-data Wagner constants is compared with that of the semi-theoretical Riedel and the empirical Ambrose-Walton equations. Lastly, reduced vapor pressures predicted from the standard and modified forms of the Riedel and Ambrose-Walton equations are used to parameterize the Wagner equation, again using the algebraic, fully-determined solution. The predictive power of such Wagner constants is compared to that of the underlying source correlations themselves. This is the first time the error of limited-data Wagner constants is segmented by interval location and width, between that due to data imprecision vs. equation imperfection, and fully- vs. over-determined solutions. Neither using a four-point, fully-determined solution rather than over-determination, nor using predictive correlations to supply “data” to a Wagner parameterization are instinctive approaches, and hence their novelty.