论文部分内容阅读
Geometrical optimization and electrostatic potential calculations have been per- formed for a series of halogenated hydrocarbons at the HF/ Gen-6d level. A number of electrostatic potentials and the statistically based structural descriptors derived from these electrostatic potentials have been obtained. Multiple linear regression analysis and artificial neural network are employed simultaneously in this paper. The result shows that the parameters derived from electrostatic potentials σ 2tot, V s and ∑ V s+, together with the molecular volume (Vmc) can be used to ex- press the quantitative structure-infinite dilution activity coefficients (γ∞) relationship of halogenated hydrocarbons in water. The result also demonstrates that the model obtained by using BFGS quasi- Newton neural network method has much better predictive capability than that from multiple linear regression. The goodness of the model has been validated through exploring the predictive power for the external test set. The model obtained via neural network may be applied to predict γ∞ of other halogenated hydrocarbons not present in the data set.
Geometrical optimization and electrostatic potential calculations have been per-formed for a series of halogenated hydrocarbons at the HF / Gen-6d level. A number of electrostatic potentials and the substantial based structural descriptors derived from these electrostatic potentials have been obtained. Multiple linear regression analysis and artificial neural network are employed simultaneously in this paper. The result shows that the parameters derived from electrostatic potentials σ 2tot, V s and Σ V s +, together with the molecular volume (Vmc) can be used to ex- press the quantitative structure- infinite dilution activity coefficients (γ∞) relationship of halogenated hydrocarbons in water. The result also demonstrates that the model obtained by using BFGS quasi- Newton neural network method has much better predictive capability than that from multiple linear regression. The goodness of the model has been validated through exploring the predictive power for the external test s et. The model obtained via neural network may be applied to predict γ∞ of other halogenated hydrocarbons not present in the data set.