Study of gas–liquid partitioning of alkane solutes in several organic solvents by using principal component analysis and linear solvation energy relationships

C. CASTELLS; M. RETA

ANALYTICA CHIMICA ACTA

ELSEVIER SCIENCE BV

Lugar: Amsterdam; Año: 2003 vol. 488 p. 107 - 122

0003-2670

Principal component analysis (PCA) was used to extract the number of factors which can describe the 737 gas–liquid partition coefficients of five linear, four branched, and two cyclic alkanes in 67 common solvents. Based on the reconstruction of partition coefficient data matrix, we concluded that the experimental dataset could readily be reduced to two relevant factors. Using only these two factors, there were no errors larger than 3%, 7 cases had errors larger than 2%, and in 34 cases, errors were between 1 and 2%. n-Hexane and ethylcyclohexane were chosen as the test factors, and all other partition coefficients were expressed in terms of these two test factors. Prediction of the logarithmic partition coefficient of these alkanes in seven chemically different solvents, which were originally excluded from the data matrix, was excellent: the root mean square error was 0.064, only in 11 cases the errors were larger than 1%, and only 3 had errors larger than 4%. Linear solvation energy relationships (LSERs) using both theoretical and empirical solvent parameters were used to explain the molecular interactions responsible for partition. Several combinations of parameters were tried but the standard deviations were not less than 0.31. This could be attributed to the model itself, imprecisions in the data matrix or in some of the LSER parameters. Solvent cohesive parameters and surface tension in combination with polarity–polarizability or dispersion parameters perform the best. Finally, the two principal component factors were rotated onto the most relevant physicochemical parameters that control the gas–liquid partitioning phenomena.