Accurate Calculations Of Physico-Chemical Properties Of Compounds By The Combination Of DFT And Statistical Correction Methods

Quantum chemistry is an important and useful theory for the calculation the properties of the organic compounds by solving of Schrodinger equation. The solving of the equation can often be restricted by size of the studied system and availability of the computing resources. Quantitative structure-activity/property relationship (QSAR/QSPR) is also a useful method for the property calculation with the advantages of less time consuming and low demand of the computing resource. So, in this thesis, the quantum chemistry calculation was combined with the QSAR/QSPR methods. And we found that the new method we used can predict the properties of the compounds more accurately and spent less time than quantum chemistry calculation. In this thesis, the following studies were carried out:Chapter 1:A brief description of the principal of quantum chemistry, QSAR/QSPR, and the combined use of quantum chemistry and QSAR/QSPR methods in property calculation were introduced.Chapter 2:The enthalpy of formation of 927 small- to medium- sized organic molecules were calculated using a density functional theory B3LYP/6-311+G(d,p) method. Multi-linear regression (MLR) and local lazy regression (LLR) algorithm were then used to correction the calculation results of B3LYP method. The root mean square error (RMSE) of the data corrected by MLR approach reduced from 126.6kJ/mol to 17.9kJ/mol. The corresponding errors of the LLR model decreased from 126.6kJ/mol to 16.1kJ/mol. These calculation results indicate that the correction result using LLR to calculate enthalpy of formation is more accurate than using MLR algorithm. And this correction method can be extended to predicting other thermodynamic properties.Chapter 3:The bond dissociation energies (BDE) of 148 compounds which contain Chlorine atom were calculated using B3LYP/6-311+G(d,p) method. Genetic algorithm (GA) was used to choose the descriptors which affect the result of the BDE. And use multi-linear regression (MLR) and least squares-support vector machine (LS-SVM) method to correct the DFT calculated data. We found that the root mean square error (RMSE) of the data corrected by MLR approach reduced from 6.13kcal/mol to 2.66kcal/mol. The corresponding errors decreased from 6.13kcal/mol to 2.47kcal/mol by using LS-SVM calibration approach. These calculation results indicate that using LS-SVM to correct the DFT calculated C-Cl bond dissociation energy is more accurate than using MLR method. And this correction method can be extended to predict the bond dissociation energies other compounds.Chapter 4:In this section, the ionization potential (IP) of 159 organic compounds were calculated by B3LYP/6-311+G(d,p) method. Genetic algorithm (GA) was used to choose the descriptors which affect the result of the ionization potential (IP). And then use multi-linear regression (MLR) and least squares-support vector machine (LS-SVM) method to correct the DFT calculated data, and found that the root mean square error (RMSE) of the data corrected by MLR approach reduced from 0.52eV to 0.42eV, the data decreased from 0.52eV to 20.29eV by using LS-SVM calibration approach. These calculation results indicate that using LS-SVM to correct the DFT calculated ionization potential is more accurate than using MLR algorithm. And this correction method can be extended to predicting ionization potential of other kind of organic compounds.