The conjugate gradient method is performed for some number of iterations, say $LaTeX: k (< n),$ then it is restarted. (If $LaTeX: k=0,$ this is the special case of Cauchy's steepest descent.)