The conjugate gradient method also uses the first derivatives of the potential energy. But instead of local gradient for downhill as in the steepest descent method, the conjugate gradient technique define the new gradient direction for each iteration by using information from previous gradient directions to determine the optimum direction for the line search. It is considered to select a successive search direction in order to eliminate repeated minimization along the same direction. This made the method more efficient and gives smaller number of iterations to reach the convergence, comparing to the steepest descent method. The conjugate gradient method is displayed in Figure 3.4.
Generally, this method converges in approximately M steps for a quadratic function, where M is the number of degrees of freedom of the function. Note that several terms in the potential energy are quadratic. Nevertheless, the disadvantage is that the line minimizations need to be performed accurately in order to ensure that the conjugate direction is set up correctly and thus time consuming. In addition, the method can be unstable if conformation is so far away from a local minimum.