The idea is to evaluate the function at a point close to one of the edges and then probe again in bigger gap on the other side. The most efficient way to pick the relative distance of the new probing point to the edge would be according to the golden ratio. However since we cannot assume that the function is unimodal we want to trade effciency for robustness in finding the minimum and hence choose the relative distance of the new probe to the edge freely.
The idea is to evaluate the function at a point close to one of the edges and then probe again in bigger gap on the other side. The most efficient way to pick the relative distance of the new probing point to the edge would be according to the golden ratio. However since we cannot assume that the function is unimodal we want to trade effciency for robustness in finding the minimum and hence choose the relative distance of the new probe to the edge freely.