next up previous print clean
Next: Module oc_cgstep_mod Up: REFERENCES Previous: Module oc_sd_mod

Module oc_cg_mod

1.
Purpose: conjugate-gradient descent step

		      CG($\bold m,\bold R,\bold g,\bold G$)  {
		 		 $\b= \frac{\parallel \bold g_k \parallel^2}{\parallel \bold g_{k-1} \parallel^2}$ 		$\left \{\begin{array}
{l} \bold s=\bold g+\b\bold s\\  \S=\bold G+\b\S \end{array}\right .$

$\alpha=-\frac{\parallel \bold g\parallel^2 }{\parallel \S \parallel^2} \quad$ $\left \{\begin{array}
{l} \bold m=\bold m+\alpha\bold s\\  \bold R=\bold R+\alpha\S \end{array}\right .$ }

2.
Functions and subroutines
(a)
integer function oc_cg(forget,x_,g_,rr_,gg_,s_,ss_,maxmem)

  • forget: re-initialize operator
  • x_: model file tag
  • >_: gradient file tag
  • rr_: residual file tag
  • >g_: conjugate gradient file tag
  • s_: previous step file tag
  • ss_: conjugate previous step file tag


Stanford Exploration Project
4/16/2001