Sensitivity analysis in high-dimensional space
Abstract: Calculating the atomic charge is one of the challenging parts in the molecular simulation process which is an important goal in computational chemistry. In this thesis, we used the Alexandria Charge Model to generate charges. The calculation was achieved by the Simulated annealing algorithm while the Markov Chain Monte Carlo method was implemented to investigate the uncertainty of model parameters. We presented different convergence criterions, with the convergence information, the influence of temperature and annealing start time on the calculation results were analyzed. For the high-dimensional parameter set in the atomic charge model, we proposed a process to extract highly sensitive and uncorrelated parameter sets. The correlation between the parameters were measured by different correlation coefficients. We introduced a variety of methods to analyze the sensitivity around the local minimum point of the parameter sand the whole parameter space. A variance-based approach was introduced to estimate global sensitivity indices for the charge model, which also considered the interactions between parameters. According to theexperiment results, the methods are shown to be efficient.
AT THIS PAGE YOU CAN DOWNLOAD THE WHOLE ESSAY. (follow the link to the next page)