xref: /petsc/src/mat/utils/zerodiag.c (revision c8a8475e04bcaa43590892a5c3e60c6f87bc31f7)
1 /*$Id: zerodiag.c,v 1.44 2001/08/06 21:16:10 bsmith Exp $*/
2 
3 /*
4     This file contains routines to reorder a matrix so that the diagonal
5     elements are nonzero.
6  */
7 
8 #include "src/mat/matimpl.h"       /*I  "petscmat.h"  I*/
9 
10 #define SWAP(a,b) {int _t; _t = a; a = b; b = _t; }
11 
12 #undef __FUNCT__
13 #define __FUNCT__ "MatReorderForNonzeroDiagonal"
14 /*@
15     MatReorderForNonzeroDiagonal - Changes matrix ordering to remove
16     zeros from diagonal. This may help in the LU factorization to
17     prevent a zero pivot.
18 
19     Collective on Mat
20 
21     Input Parameters:
22 +   mat  - matrix to reorder
23 -   rmap,cmap - row and column permutations.  Usually obtained from
24                MatGetOrdering().
25 
26     Level: intermediate
27 
28     Notes:
29     This is not intended as a replacement for pivoting for matrices that
30     have ``bad'' structure. It is only a stop-gap measure. Should be called
31     after a call to MatGetOrdering(), this routine changes the column
32     ordering defined in cis.
33 
34     Options Database Keys (When using SLES):
35 +      -pc_ilu_nonzeros_along_diagonal
36 -      -pc_lu_nonzeros_along_diagonal
37 
38     Algorithm Notes:
39     Column pivoting is used.
40 
41     1) Choice of column is made by looking at the
42        non-zero elements in the troublesome row for columns that are not yet
43        included (moving from left to right).
44 
45     2) If (1) fails we check all the columns to the left of the current row
46        and see if one of them has could be swapped. It can be swapped if
47        its corresponding row has a non-zero in the column it is being
48        swapped with; to make sure the previous nonzero diagonal remains
49        nonzero
50 
51 
52 @*/
53 int MatReorderForNonzeroDiagonal(Mat mat,PetscReal atol,IS ris,IS cis)
54 {
55   int      ierr,prow,k,nz,n,repl,*j,*col,*row,m,*icol,nnz,*jj,kk;
56   PetscScalar   *v,*vv;
57   PetscReal   repla;
58   IS       icis;
59 
60   PetscFunctionBegin;
61   PetscValidHeaderSpecific(mat,MAT_COOKIE);
62   PetscValidHeaderSpecific(ris,IS_COOKIE);
63   PetscValidHeaderSpecific(cis,IS_COOKIE);
64 
65   ierr = ISGetIndices(ris,&row);CHKERRQ(ierr);
66   ierr = ISGetIndices(cis,&col);CHKERRQ(ierr);
67   ierr = ISInvertPermutation(cis,PETSC_DECIDE,&icis);CHKERRQ(ierr);
68   ierr = ISGetIndices(icis,&icol);CHKERRQ(ierr);
69   ierr = MatGetSize(mat,&m,&n);CHKERRQ(ierr);
70 
71   for (prow=0; prow<n; prow++) {
72     ierr = MatGetRow(mat,row[prow],&nz,&j,&v);CHKERRQ(ierr);
73     for (k=0; k<nz; k++) {if (icol[j[k]] == prow) break;}
74     if (k >= nz || PetscAbsScalar(v[k]) <= atol) {
75       /* Element too small or zero; find the best candidate */
76       repla = (k >= nz) ? 0.0 : PetscAbsScalar(v[k]);
77       /*
78           Look for a later column we can swap with this one
79       */
80       for (k=0; k<nz; k++) {
81 	if (icol[j[k]] > prow && PetscAbsScalar(v[k]) > repla) {
82           /* found a suitable later column */
83 	  repl  = icol[j[k]];
84           SWAP(icol[col[prow]],icol[col[repl]]);
85           SWAP(col[prow],col[repl]);
86           goto found;
87         }
88       }
89       /*
90            Did not find a suitable later column so look for an earlier column
91 	   We need to be sure that we don't introduce a zero in a previous
92 	   diagonal
93       */
94       for (k=0; k<nz; k++) {
95         if (icol[j[k]] < prow && PetscAbsScalar(v[k]) > repla) {
96           /* See if this one will work */
97           repl  = icol[j[k]];
98           ierr = MatGetRow(mat,row[repl],&nnz,&jj,&vv);CHKERRQ(ierr);
99           for (kk=0; kk<nnz; kk++) {
100             if (icol[jj[kk]] == prow && PetscAbsScalar(vv[kk]) > atol) {
101               ierr = MatRestoreRow(mat,row[repl],&nnz,&jj,&vv);CHKERRQ(ierr);
102               SWAP(icol[col[prow]],icol[col[repl]]);
103               SWAP(col[prow],col[repl]);
104               goto found;
105 	    }
106           }
107           ierr = MatRestoreRow(mat,row[repl],&nnz,&jj,&vv);CHKERRQ(ierr);
108         }
109       }
110       /*
111           No column  suitable; instead check all future rows
112           Note: this will be very slow
113       */
114       for (k=prow+1; k<n; k++) {
115         ierr = MatGetRow(mat,row[k],&nnz,&jj,&vv);CHKERRQ(ierr);
116         for (kk=0; kk<nnz; kk++) {
117           if (icol[jj[kk]] == prow && PetscAbsScalar(vv[kk]) > atol) {
118             /* found a row */
119             SWAP(row[prow],row[k]);
120             goto found;
121           }
122         }
123         ierr = MatRestoreRow(mat,row[k],&nnz,&jj,&vv);CHKERRQ(ierr);
124       }
125 
126       found:;
127     }
128     ierr = MatRestoreRow(mat,row[prow],&nz,&j,&v);CHKERRQ(ierr);
129   }
130   ierr = ISRestoreIndices(ris,&row);CHKERRQ(ierr);
131   ierr = ISRestoreIndices(cis,&col);CHKERRQ(ierr);
132   ierr = ISRestoreIndices(icis,&icol);CHKERRQ(ierr);
133   ierr = ISDestroy(icis);CHKERRQ(ierr);
134   PetscFunctionReturn(0);
135 }
136 
137 
138 
139