The existing solution methods for unconstrained optimization concentrate on single extremal problems. Most of the methods usually converge to the nearest local optimal point from the initial point when applied to a multiple extremal problem. In this paper a radial method is developed which emits a bundle of search directions instead of one direction each time to gain a better chance of reaching the global minimum. Some empirical results show that the radial method usually converges to a better local optimum with a larger amount of computations than other methods. A typical multimodal problem in mechanism design is presented to illustrate how the radial method is able to find a better solution than several other methods.
All Science Journal Classification (ASJC) codes