This method uses climbing-string method to get an image close to the saddle 
point, then uses inexact newton method to accelerate the convergence.
The final state is used as the inpute of the initial guessed saddle point for
climbing-string.