schurcompl Class Reference

#include <schurcompl.h>

Public Member Functions
void	gather_bound_vect (double lv, double gv, long *domproc)
void	globcnnum_pdd (gtopology top, long ltg, long domproc, FILE out)
void	initiate (partopjk ptop, FILE out)
void	initiate (partop ptop, selnodes selnodschur)
double	pss_gather_bound_vect (double lv1, double gv1, double lv2, double gv2, long *domproc)
double	pss_gather_bound_vect_old (double lv, double gv, long *domproc)
void	scatter_bound_vect (double lv, double gv, long *domproc)
	schurcompl (int np, int mr, long nd)
void	schurordering_new (gtopology top, long nodeidentif)
long	selected_norm_calculation (long cid, double err, double thresh, partopjk ptop, gtopology top, double lv1, double gv1, double rhs, double grhs, long domproc, long n, FILE out)
void	solve_red_sys (long domproc, double condmat, double condvect, long decomp, FILE out)
void	solve_red_sys_fin (long domproc, double condmat, double condvect, long decomp, FILE out)
void	solve_red_sys_iter (long domproc, double condmat, double condvect, FILE out)
void	solve_system (gtopology top, gmatrix gm, long domproc, double lhs, double rhs, FILE out)
double	unbalanced_values (double lv, long domproc, FILE *out)
	~schurcompl ()
Public Attributes
double	aerrcgsch
long	anicgsch
gmatrix *	arr
long	destr_nbdofmas
storagetype	dmstor
	type of storage of domain matrix
double	errcgsch
gtopology *	gtop
long	maxnbdof
int	myrank
	rank of processor
long	nbdof
	number of boundary degrees of freedom (unknowns) on subdomain
long *	nbdofmas
long	ndof
	number of degrees of freedom on subdomain
long	ndofcp
long	ndom
	number of subdomain
long	nicgsch
long	nidof
	number of internal degrees of freedom (unknowns) on subdomain
int	nproc
	number of processors
storagetype	rsmstor
	storage of reduced system matrix
slesolv *	ssle
	type of linear system solver
redsystsolver	trssol
	type of reduced system solver
double	zero

Detailed Description

class paral contains information about local-global ordering

PRIMAL DOMAIN DECOMPOSITION METHOD

nproc, myrank, ndom number of processors, myrank and number of domain are established after constructor execution

ndof, nidof total number of local unknowns and number of internal unknowns on subdomains are established after execution of function schurordering

ngdof (M) total number of global unknowns is established after execution of function globcnnum_pdd

maxnbn, maxnrdof, totmaxndofn maximum number of boundary nodes, maximum number of boundary unknowns on subdomain and maximum number of degrees of freedom on node are established after execution of globcnnum_pdd

gcn array of local to global code numbers correspondence is established after execution of function globcnnum_pdd

nbndom (M) array of numbers of boundary nodes on subdomains is established after execution of function globcnnum_pdd

nrdofdom (M) array of numbers of reduced unknowns on subdomains is established after execution of function globcnnum_pdd

masgcn (M) array of local to global code numbers of all subdomains is established after execution of function globcnnum_pdd

Definition at line 55 of file schurcompl.h.

Constructor & Destructor Documentation

schurcompl	(	int	np,
		int	mr,
		long	nd
	)

constructor

Parameters:

	np	- number of processors
	mr	- my rank (process id)
	nd	- number of subdomains

JK, revised 24.7.2008

Definition at line 19 of file schurcompl.cpp.

References aerrcgsch, anicgsch, arr, destr_nbdofmas, errcgsch, gtop, maxnbdof, myrank, nbdof, nbdofmas, ndof, ndom, nicgsch, nidof, nproc, rsmstor, ssle, trssol, and zero.

~schurcompl ( )

Definition at line 72 of file schurcompl.cpp.

References arr, destr_nbdofmas, gtop, myrank, nbdofmas, and ssle.

Member Function Documentation

void gather_bound_vect	(	double *	lv,
		double *	gv,
		long *	domproc
	)

function gathers boundary/interface contributions from subdomains into one global vector the global vector must be deleted outside of the subroutine (the global vector contains all components defined on the boundaries/interfaces) (the local vector contains all components defined on a subdomain)

Parameters:

	lv	- local vector
	gv	- global vector, it is allocated only on the master processor the array must be deleted outside of the subroutine
	domproc	- domain-processor correspondence

JK, 6.11.2004

Definition at line 1242 of file schurcompl.cpp.

References gtopology::give_code_numbers(), gtopology::give_ndofe(), gtop, locglob(), maxnbdof, MPI_ANY_SOURCE, MPI_ANY_TAG, MPI_Barrier(), MPI_COMM_WORLD, MPI_DOUBLE, MPI_Recv(), MPI_Send(), MPI_Status::MPI_TAG, myrank, ndof, nidof, nproc, and stat.

Referenced by unbalanced_values().

void globcnnum_pdd	(	gtopology *	top,
		long *	ltg,
		long *	domproc,
		FILE *	out
	)

void initiate	(	partopjk *	ptop,
		FILE *	out
	)

Definition at line 392 of file schurcompl.cpp.

References partopjk::bdofd, gtopology::cnstate, gtop, gtopology::initiate(), maxnbdof, partopjk::maxnbdofd, myrank, partopjk::nbdof, nbdof, partopjk::nbdofd, nbdofmas, partopjk::ndof, ndof, partopjk::ndofc, ndofcp, partopjk::nidof, nidof, and nproc.

void initiate	(	partop *	ptop,
		selnodes *	selnodschur
	)

Function search prescribed common code numbers on boundary nodes in the given subdomain.

Parameters:

	top	- pointer to the sequential general topology
	ltg	- local to global array
	idgnn	- array with node numbers of boundary nodes
	nbn	- number of boundary nodes
	gnnc	- array with boundary node numbers at positions of nodes with prescribed common code numbers
	id	- searched node id
	ccn	- searched common code number id function generates ordering of unknowns on subdomains
	top	- pointer to the sequential general topology
	ltg	- local to global array

JK, 7.8.2007 function orders unknowns on subdomains with respect of Schur complement method inner variables are ordered at the beginning boundary variables are ordered at he end

Parameters:

	top	- pointer to domain topology
	ptop	- pointer to topology for paralle computation

JK, 8.7.2005 JK, 7.8.2007

Definition at line 366 of file schurcompl.cpp.

References selnodes::cnmas, gtopology::cnstate, gtop, gtopology::initiate(), maxnbdof, selnodes::maxsnndof, myrank, partop::nbdof, nbdof, nbdofmas, partop::ndof, ndof, ndofcp, partop::nidof, nidof, nproc, selnodes::snndofmas, and selnodes::tndofsn.

double pss_gather_bound_vect	(	double *	lv1,
		double *	gv1,
		double *	lv2,
		double *	gv2,
		long *	domproc
	)

function computes dot product of two vectors which are split on several processors it gathers boundary/interface contributions from subdomains into global vectors it gathers results of dot product of internal components of the vectors the global vectors must be deleted outside of the subroutine

Parameters:

	lv1,lv2	- local vectors
	gv1,gv2	- global vectors, they are allocated only on the master processor the arrays must be deleted outside of the subroutine

JK, 6.11.2004, revised 2.6.2011

Definition at line 1307 of file schurcompl.cpp.

References gtopology::give_code_numbers(), gtopology::give_ndofe(), gtop, locglob(), maxnbdof, MPI_ANY_SOURCE, MPI_ANY_TAG, MPI_Barrier(), MPI_COMM_WORLD, MPI_DOUBLE, MPI_Recv(), MPI_Send(), MPI_Status::MPI_TAG, myrank, ndof, nidof, nproc, and stat.

double pss_gather_bound_vect_old	(	double *	lv,
		double *	gv,
		long *	domproc
	)

function gathers contributions from subdomains into one global vector the global vector must be deleted outside of the subroutine

Parameters:

	lv	- local vector
	gv	- global vector, it is allocated only on the master processor the array must be deleted outside of the subroutine

JK, 6.11.2004

Definition at line 1617 of file schurcompl.cpp.

References gtopology::give_code_numbers(), gtopology::give_ndofe(), gtop, locglob(), maxnbdof, MPI_ANY_SOURCE, MPI_ANY_TAG, MPI_Barrier(), MPI_COMM_WORLD, MPI_DOUBLE, MPI_Recv(), MPI_Send(), MPI_Status::MPI_TAG, myrank, ndof, nidof, nproc, and stat.

void scatter_bound_vect	(	double *	lv,
		double *	gv,
		long *	domproc
	)

function scatters contributions from one global vector into local vectors function rewrites part of the local vector by contributions from boundaries

Parameters:

	lv	- local vector
	gv	- global vector, it is allocated only on the master processor

JK, 6.11.2004

Definition at line 1699 of file schurcompl.cpp.

References gtopology::give_code_numbers(), gtopology::give_ndofe(), globloc(), gtop, maxnbdof, MPI_ANY_SOURCE, MPI_ANY_TAG, MPI_Barrier(), MPI_COMM_WORLD, MPI_DOUBLE, MPI_Recv(), MPI_Send(), myrank, nbdofmas, ndof, nidof, nproc, nullv(), and stat.

Referenced by unbalanced_values().

void schurordering_new	(	gtopology *	top,
		long *	nodeidentif
	)

long selected_norm_calculation	(	long	cid,
		double	err,
		double	thresh,
		partopjk *	ptop,
		gtopology *	top,
		double *	lv1,
		double *	gv1,
		double *	rhs,
		double *	grhs,
		long *	domproc,
		long	n,
		FILE *	out
	)

function computes norm of a vector where only selected components contribute the vector is split on several processors

function assumes that no boundary/interface node is equipped with the Dirichlet boundary condition function assumes that no boundary/interface node is coupled with any other node

Parameters:

	cid	- component id (id of required component)
	ntm	- number of transported media
	err	- required residual
	ptop	- pointer to the parallel topology
	top	- pointer to the subdomain topology
	lv1,rhs	- local vectors
	gv1,grhs	- global vectors, they are allocated only on the master processor the arrays must be deleted outside of the subroutine
	domproc	- domain-processor correspondence

if the vectors gv1 and grhs are not used outside of this function, they will be removed

JK, 13.7.2011

Definition at line 1405 of file schurcompl.cpp.

References partopjk::bnid, gtopology::give_code_numbers(), gtopology::give_dof(), gtopology::give_ndofe(), gtopology::give_ndofn(), gtop, partopjk::inid, locglob(), maxnbdof, MPI_ANY_SOURCE, MPI_ANY_TAG, MPI_Barrier(), MPI_COMM_WORLD, MPI_DOUBLE, MPI_LONG, MPI_Recv(), MPI_Send(), MPI_Status::MPI_TAG, myrank, partopjk::nbn, ndofcp, partopjk::nin, nproc, nullv(), and stat.

void solve_red_sys	(	long *	domproc,
		double *	condmat,
		double *	condvect,
		long	decomp,
		FILE *	out
	)

function solves reduced system of linear algebraic equations

Parameters:

	domproc	- domain-processor correspondence
	condmat	- array containing reduced matrices (Schur complements)
	condvect	- array containing reduced vectors
	decomp	- indicator for valid condensed matrix of reduced system used for elimination of reduced system only. (0=condensation have to be performed/1=only backward substitution is necessary)
	out	- output file (for auxiliary output)

JK

Definition at line 998 of file schurcompl.cpp.

References master_sol, myrank, par_print_err(), paral_sol, solve_red_sys_fin(), solve_red_sys_iter(), and trssol.

Referenced by solve_system().

void solve_red_sys_fin	(	long *	domproc,
		double *	condmat,
		double *	condvect,
		long	decomp,
		FILE *	out
	)

function solves reduced system of linear algebraic equations on the master processor by a direct method

Parameters:

	domproc	- domain-processor correspondence
	condmat	- array containing reduced matrices (Schur complements)
	condvect	- array containing reduced vectors
	decomp	- indicator for valid condensed matrix of reduced system. (0=condensation have to be performed/1=only backward substitution is necessary)
	out	- output file (for auxiliary output)

JK

Definition at line 828 of file schurcompl.cpp.

References gmatrix::alloc(), arr, gtopology::give_code_numbers(), gtopology::give_ndofe(), globloc(), gtop, gmatrix::initiate(), gmatrix::localized(), locglob(), maxnbdof, memset(), MPI_ANY_SOURCE, MPI_ANY_TAG, MPI_COMM_WORLD, MPI_DOUBLE, MPI_Recv(), MPI_Send(), MPI_Status::MPI_TAG, myrank, nbdof, nbdofmas, ndofcp, ndom, nproc, nullv(), gmatrix::prepmat(), rsmstor, gmatrix::setval(), slesolv::solve_system(), ssle, stat, and gmatrix::ts.

Referenced by solve_red_sys().

void solve_red_sys_iter	(	long *	domproc,
		double *	condmat,
		double *	condvect,
		FILE *	out
	)

function orders unknowns in coarse problem

function initializates object gtop of the class gtopology it represents subdomains as superelements

Parameters:

ptop

- pointer to parallel topology

JK, 8.7.2005 function solves reduced system of equations from primal domain decomposition by conjugate gradient method matrix of reduced system is not assembled iteration process is directed by the master processor

Parameters:

	domproc	- domain-processor correspondence
	condmat	- array containing reduced matrices (Schur complements)
	condvect	- array containing reduced vectors
	out	- output file (for auxiliary output)

JK, 2.12.2001

Definition at line 599 of file schurcompl.cpp.

References aerrcgsch, anicgsch, copyv(), errcgsch, gtopology::give_code_numbers(), gtopology::give_ndofe(), globloc(), gtop, locglob(), maxnbdof, memset(), MPI_ANY_SOURCE, MPI_ANY_TAG, MPI_COMM_WORLD, MPI_DOUBLE, MPI_Recv(), MPI_Send(), MPI_Status::MPI_TAG, mxv(), myrank, nbdof, nbdofmas, ndof, ndofcp, nicgsch, nidof, nproc, nullv(), p, ss(), stat, and zero.

Referenced by solve_red_sys().

void solve_system	(	gtopology *	top,
		gmatrix *	gm,
		long *	domproc,
		double *	lhs,
		double *	rhs,
		FILE *	out
	)

function solves system of linear algebraic equations by the Schur complment method

Parameters:

	top	- pointer to the sequential general topology
	gm	- pointer to the subdomain matrix stored in the form gmatrix
	domproc	- domain-processor correspondence
	lhs	- array containing solution
	rhs	- array containing right hand side
	out	- output file (for auxiliary output)

JK

Definition at line 1029 of file schurcompl.cpp.

References gmatrix::condense(), condmat(), cp, gmatrix::decomp(), maxnbdof, memset(), MPI_ANY_SOURCE, MPI_ANY_TAG, MPI_Barrier(), MPI_COMM_WORLD, MPI_LONG, MPI_Recv(), MPI_Send(), MPI_Status::MPI_TAG, myrank, nbdof, nproc, solve_red_sys(), stat, and time.

double unbalanced_values	(	double *	lv,
		long *	domproc,
		FILE *	out
	)

function computes norm of vector of unbalanced values function gathers vector of unbalanced values on boundaries function scatters correct values into subdomains function returns norm of vector of unbalanced values

Parameters: