Module oc_solverreg_mod

1.

Purpose: implements a regularized least-squares gradient solver

$$\fbox {$ \displaystyle \left [\mathcal W\mathcal L\atop\epsi... ...ht ] \bold m\approx \left [\mathcal W\bold D\atop0\right ] $}$$

		 $\left [\Rd\atop\Rm\right ] = \left [\mathcal W\mathcal L\atop\epsilon\mathcal A\right ] \bold m_0- \left [\mathcal W\bold D\atop0\right ]$

iterate { 
		 		  $\bold g= \left [\mathcal L^*\mathcal W^*\quad\epsilon\mathcal A^*\right ] \left [\Rd\atop\Rm\right ]$
		 		  $\left [\Gd\atop\Gm\right ] = \left [\mathcal W\mathcal L\atop\epsilon\mathcal A\right ] \bold g$
		 		 $\left (\bold m,\left [\Rd\atop\Rm\right ] \right )\longleftarrow$step$\left (\bold m,\left [\Rd\atop\Rm\right ] ,\bold g,\left [\Gd\atop\Gm\right ] \right )$
		 		 }

2.

Functions and subroutines

(a)

subroutine oc_solverreg_init(niter,eps,maxmem,verb,mmovie,dmovie,resstop,rescale)

Purpose: initialize the regularized solver

• niter: iterations number
• eps: scaling factor
• verb: verbose flag (optional)
• mmovie: model movie output flag (optional)
• dmovie: data movie output flag (optional)
• resstop: stop iterations at this residual power (optional)
• rescale: rescale model (optional)

(b)

subroutine oc_solverreg( L,A,S, x_,yy_, nreg, W ,k_,x0_,res_,op1 ...op9)

Purpose: regularized solver

• L: out-of-core linear operator
• A: out-of-core regularization operator
• S: gradient step
• nreg: dimension of the regularization output
• W: out-of-core weighting operator
• k_: data mask tag
• x0_: starting model tag
• res_: residual tag