The Application Of High-Performance Computing In The Theoretical And Computational Chemistry

In recent years, the computational chemistry, as a new research direction of the high-performance computing, has made great progress. Many high-performance computing software, mainly Gaussian, has been helping theoretical chemistry researchers realize their research projects and inductions. The field of computational chemistry has a series of complicated problems, which has to rely on HPC clusters to be solved, but not on the conventional mini-computers as in the past decades. As development of cluster techniques, a variety of computational software on theoretical chemistry is successfully running in the computational cluster systems. The further progress of cluster techniques would certainly promote the rapid development of computational chemistry and its related fields.In this thesis, the parallel efficiency of Gaussian program package, the most popular theoretical chemistry computational software, was analyzed, and an appropriate scheme for the practical parallel computing was put forward. The proposed scheme takes advantage of cluster techniques, and could provide a systematic and effective compile script for users to submit their jobs. The content of the present research is as follows:Gaussian program package usually adopt the NFS file system to realize the synchronization of data and programs, and the performance of NFS depends on the network facility in the connection of data transformation. This demands the high-performance data exchange facility with all-speed-up and no-block. We analyze the factors that influence the network capability in the job running of Gaussian. The Linda parallel technique in Gaussian program package involves a large amount of system communications and synchronizations when the parallel communication is carried out, thus in practice the big bandwidth and low delay are needed.By analyzing the relation between the number of CPUs and speed-up ratio, we found that the practical ratio is linear in the in the single node, but the ratio is not so good in the inter nodes. When the 32 core of CPUs is considered, the speed-up is not as favorite as expected.Gaussian is the most widely used quantum chemistry computational software package in ab intio and semi-empirical quantum chemical calculations, which can be used to investigate the molecular energies and structures, transition state energies and structure, chemical bonding and reaction energies, molecular orbitals, dipolar moment and multi-polar moments, atomic charge and potential, vibrational frequency, infrared and Raman spectra, NMR, polarization and hyper-polarizations, thermodynamic properties, reaction pathway, and so on. Therefore, large memory is needed to support the properly running of Gaussian. It's reported that the parameter %mem can control the amount of the memory used during the course of the operation. Using this memory, the computing speed can be improved, and the parameter %men may improve the overall operation speed. It should be noted that the value of the parameter set is not better, when the parameter reaches a certain level, the operational performance parameters are no longer increases with the increase.Usually, we write some calculation contents into a file, then the file is be modified, and the final results is obtained. We can find that there is an IO peak at the start of the stage. Following, the IO maintain a smaller value.Based on the results mentioned above, a lot of computational works are running at the high-performance computing platform. In this thesis, the hERK1 enzyme is described in detail.