In this paper, we analyse techniques of computing a search direction b
y minimizing the approximate quadratic model in the 2 dimensional subs
pace spanned by the current gradient and the last search direction. Th
e classical conjugate gradient methods are only the special cases wher
e the objective function is quadratic and line searches are exact. Bas
ed on our analyses on the case where line searches are not exact, we c
onstruct new conjugate direction type algorithms.