An adaptive distance-based group contribution method for thermodynamic property prediction.
Tanjin He,Shuang Li,Yawei Chi,Hong-Bo Zhang,Zhi Wang,Bin Yang,Xin He,Xiaoqing You
DOI: https://doi.org/10.1039/c6cp02929a
IF: 3.3
2016-01-01
Physical Chemistry Chemical Physics
Abstract:In the search for an accurate yet inexpensive method to predict thermodynamic properties of large hydrocarbon molecules, we have developed an automatic and adaptive distance-based group contribution (DBGC) method. The method characterizes the group interaction within a molecule with an exponential decay function of the group-to-group distance, defined as the number of bonds between the groups. A database containing the molecular bonding information and the standard enthalpy of formation (H-f,H- 298K) for alkanes, alkenes, and their radicals at the M06-2X/def2-TZVP//B3LYP/6-31G(d) level of theory was constructed. Multiple linear regression (MLR) and artificial neural network (ANN) fitting were used to obtain the contributions from individual groups and group interactions for further predictions. Compared with the conventional group additivity (GA) method, the DBGC method predicts H-f,H- 298K for alkanes more accurately using the same training sets. Particularly for some highly branched large hydrocarbons, the discrepancy with the literature data is smaller for the DBGC method than the conventional GA method. When extended to other molecular classes, including alkenes and radicals, the overall accuracy level of this new method is still satisfactory.