Geometry optimization has become a powerful tool to study mechanisms of chemical reactions. However, it requires some guesses about the mechanisms. To study highly-complicated multistep mechanisms, a fully-automated method which can find not only expected pathways but also unexpected ones is desired. We have worked on development of such theoretical approaches. We have applied our own methods to various systems toward systematic understanding of reaction mechanisms, selectivity, optical responses (including nonadiabatic pathways through the conical intersection), etc.