Simulation-based material design system using data mining

近年来，GPGPU或多核计算机作为小规模的中型计算环境已经非常熟悉。许多计算化学和计算物理的包程序也是完全成熟的。另一方面，这类程序的结果是大量的数值数据，因此我们需要对数据进行自动处理，以利用许多计算结果。本主题研究了一个基本的性能评估，以讨论合适的系统规模。此外，本课题还尝试利用数据挖掘技术对计算结果中的分子结构进行分类。结果，我们发现我们可以用不到10个碳原子运行大量的计算作业。此外，分子结构的自动分类可以找到与人类相同的规则。

Recently, GPGPU or multi-core computers are very familiar as the small-scale of middle-scale computing environment. Many package programs for computational chemistry and computational physics are also full-grown. On the other hand, a result of such package programs is the mass of numerical data so that we require auto-processing of the data for utilizing many computational results. This theme examined a fundamental performance evaluation in order to discuss suitable system size. Furthermore, this theme tried classifying molecular structure in a computational result via data mining technology. As the result, we found that we can run massively many computational jobs for less than ten carbon atoms. Moreover, auto-classification for molecular structure can find out as the same rule as the human do it.