Welcome to BCLS!

Go to the source code of this file.


Functions
static void	aprod_free_lsqr (const int mode, const int mSubProb, const int nFree, double dxFree[], double y[], void *UsrWrk)
	Mat-vec routine called by LSQR.
int	bcls_newton_step_lsqr (BCLS ls, int m, int nFree, int ix[], double damp, int itnLim, double tol, double dxFree[], double x[], double c[], double r[], int itns, double *opt)
	Compute a Newton step using LSQR.

Detailed Description

Function Documentation

static void aprod_free_lsqr	(	const int	mode,
		const int	mSubProb,
		const int	nFree,
		double	dxFree[],
		double	y[],
		void *	UsrWrk
	)			`[static]`

Mat-vec routine called by LSQR.

LSQR calls this routine, which in turn calls bcls_aprod: This routine is declared "static" so that it won't be confused with the user's own Aprod routine.

If mode = 1,
- y(1:m) <- y(1:m) + A(:,ix) * dxFree.
- y(m+1:) <- y(m+1:) + damp * dxFree

If mode = 2,
- dxFree <- dxFree + A'* y(1:m) + damp y(m+1:).

Note that mSubProb is the number of rows in the subproblem. This may or may not be equal to m, which is the number of rows in the original problem. If the subproblem is damped (because the user has either provided a linear term or an explicit damping parameter), then mSubProb = m + nFree. Otherwise, mSubProb = m.

Parameters:

`[in]`	mode	Determines which producte with A is required.
`[in]`	mSubProb	Number of rows in the matrix seen by LSQR. Also: length of y.
`[in]`	nFree	Number of columns in A(:,ix). Also: length of ix length of dxFree.
`[in,out]`	dxFree	Primal variables
`[in,out]`	y	Dual variables.
`[in,out]`	UsrWrk	Transit pointer to the BCLS problem context.

Definition at line 79 of file bclsqr.c.

References bcls_aprod(), BCLS_PRECON_U, BCLS_PRECON_Ut, BCLS_PROD_A, BCLS_PROD_At, bcls_usolve(), cblas_daxpy(), cblas_dcopy(), BCLS::damp_actual, BCLS::dx, BCLS::ix, BCLS::m, BCLS::Usolve, and BCLS::wrk_u.

Referenced by bcls_newton_step_lsqr().

Here is the call graph for this function:

int bcls_newton_step_lsqr	(	BCLS *	ls,
		int	m,
		int	nFree,
		int	ix[],
		double	damp,
		int	itnLim,
		double	tol,
		double	dxFree[],
		double	x[],
		double	c[],
		double	r[],
		int *	itns,
		double *	opt
	)

Compute a Newton step using LSQR.

See also:: bcls_newton_step_cgls.

Parameters:

`[in,out]`	ls	BCLS problem context.
`[in]`	m	Number of rows in A.
`[in]`	nFree	Number of columns in A(:,ix). Also: length of ix and dxFree
`[in]`	ix	Index of free variables.
`[in]`	damp	Regularization parameter.
`[in]`	itnLim	Iteration limit on current LSQR call.
`[in]`	tol	LSQR's atol and btol.
`[in,out]`	dxFree	Search direction on free variables.
`[in,out]`	x	Current point.
`[in]`	c	Linear term.
`[in,out]`	r	Residual. Used as RHS for LSQR.
`[out]`	itns	Number of LSQR iterations on current subproblem.
`[out]`	opt	Optimality achieved by LSQR on current subproblem.

Returns:

0: Required accurace was achieved.
1: The iteration limit (itnLim) was reached.
2: A(:,ix) is excessively ill-conditioned.

Definition at line 168 of file bclsqr.c.

References aprod_free_lsqr(), bcls_timer(), BCLS_TIMER_LSQR, BCLS_TIMER_START, BCLS_TIMER_STOP, cblas_dscal(), BCLS::damp_actual, BCLS::damp_min, lsqr(), BCLS::stopwatch, and BCLS::wrk_v.

Referenced by bcls_newton_step().

Here is the call graph for this function:

bclsqr.c File Reference

Functions

Detailed Description

Function Documentation