xref: /petsc/src/snes/interface/snesj.c (revision 029af93f72d387caa45cf6909ac9aed2d04296ca)
1 
2 #ifndef lint
3 static char vcid[] = "$Id: snesj.c,v 1.41 1997/01/06 20:29:45 balay Exp bsmith $";
4 #endif
5 
6 #include "src/snes/snesimpl.h"    /*I  "snes.h"  I*/
7 
8 #undef __FUNC__
9 #define __FUNC__ "SNESDefaultComputeJacobian"
10 /*@C
11    SNESDefaultComputeJacobian - Computes the Jacobian using finite differences.
12 
13    Input Parameters:
14 .  x1 - compute Jacobian at this point
15 .  ctx - application's function context, as set with SNESSetFunction()
16 
17    Output Parameters:
18 .  J - Jacobian
19 .  B - preconditioner, same as Jacobian
20 .  flag - matrix flag
21 
22    Options Database Key:
23 $  -snes_fd
24 
25    Notes:
26    This routine is slow and expensive, and is not currently optimized
27    to take advantage of sparsity in the problem.  Although
28    SNESDefaultComputeJacobian() is not recommended for general use
29    in large-scale applications, It can be useful in checking the
30    correctness of a user-provided Jacobian.
31 
32 .keywords: SNES, finite differences, Jacobian
33 
34 .seealso: SNESSetJacobian()
35 @*/
36 int SNESDefaultComputeJacobian(SNES snes,Vec x1,Mat *J,Mat *B,MatStructure *flag,void *ctx)
37 {
38   Vec      j1,j2,x2;
39   int      i,ierr,N,start,end,j;
40   Scalar   dx, mone = -1.0,*y,scale,*xx,wscale;
41   double   amax, epsilon = 1.e-8; /* assumes double precision */
42   double   dx_min = 1.e-16, dx_par = 1.e-1;
43   MPI_Comm comm;
44   int      (*eval_fct)(SNES,Vec,Vec);
45 
46   if (snes->method_class == SNES_NONLINEAR_EQUATIONS)
47     eval_fct = SNESComputeFunction;
48   else if (snes->method_class == SNES_UNCONSTRAINED_MINIMIZATION)
49     eval_fct = SNESComputeGradient;
50   else SETERRQ(1,0,"Invalid method class");
51 
52   PetscObjectGetComm((PetscObject)x1,&comm);
53   MatZeroEntries(*J);
54   if (!snes->nvwork) {
55     ierr = VecDuplicateVecs(x1,3,&snes->vwork); CHKERRQ(ierr);
56     snes->nvwork = 3;
57     PLogObjectParents(snes,3,snes->vwork);
58   }
59   j1 = snes->vwork[0]; j2 = snes->vwork[1]; x2 = snes->vwork[2];
60 
61   ierr = VecGetSize(x1,&N); CHKERRQ(ierr);
62   ierr = VecGetOwnershipRange(x1,&start,&end); CHKERRQ(ierr);
63   VecGetArray(x1,&xx);
64   ierr = eval_fct(snes,x1,j1); CHKERRQ(ierr);
65 
66   /* Compute Jacobian approximation, 1 column at a time.
67       x1 = current iterate, j1 = F(x1)
68       x2 = perturbed iterate, j2 = F(x2)
69    */
70   for ( i=0; i<N; i++ ) {
71     ierr = VecCopy(x1,x2); CHKERRQ(ierr);
72     if ( i>= start && i<end) {
73       dx = xx[i-start];
74 #if !defined(PETSC_COMPLEX)
75       if (dx < dx_min && dx >= 0.0) dx = dx_par;
76       else if (dx < 0.0 && dx > -dx_min) dx = -dx_par;
77 #else
78       if (abs(dx) < dx_min && real(dx) >= 0.0) dx = dx_par;
79       else if (real(dx) < 0.0 && abs(dx) < dx_min) dx = -dx_par;
80 #endif
81       dx *= epsilon;
82       wscale = 1.0/dx;
83       VecSetValues(x2,1,&i,&dx,ADD_VALUES);
84     }
85     else {
86       wscale = 0.0;
87     }
88     ierr = eval_fct(snes,x2,j2); CHKERRQ(ierr);
89     ierr = VecAXPY(&mone,j1,j2); CHKERRQ(ierr);
90     /* Communicate scale to all processors */
91 #if !defined(PETSC_COMPLEX)
92     MPI_Allreduce(&wscale,&scale,1,MPI_DOUBLE,MPI_SUM,comm);
93 #else
94     MPI_Allreduce(&wscale,&scale,2,MPI_DOUBLE,MPI_SUM,comm);
95 #endif
96     VecScale(&scale,j2);
97     VecGetArray(j2,&y);
98     VecNorm(j2,NORM_INFINITY,&amax); amax *= 1.e-14;
99     for ( j=start; j<end; j++ ) {
100       if (PetscAbsScalar(y[j-start]) > amax) {
101         ierr = MatSetValues(*J,1,&j,1,&i,y+j-start,INSERT_VALUES); CHKERRQ(ierr);
102       }
103     }
104     VecRestoreArray(j2,&y);
105   }
106   ierr = MatAssemblyBegin(*J,MAT_FINAL_ASSEMBLY); CHKERRQ(ierr);
107   ierr = MatAssemblyEnd(*J,MAT_FINAL_ASSEMBLY); CHKERRQ(ierr);
108   *flag =  DIFFERENT_NONZERO_PATTERN;
109   return 0;
110 }
111 
112 #undef __FUNC__
113 #define __FUNC__ "SNESDefaultComputeHessian"
114 /*@C
115    SNESDefaultComputeHessian - Computes the Hessian using finite differences.
116 
117    Input Parameters:
118 .  x1 - compute Hessian at this point
119 .  ctx - application's gradient context, as set with SNESSetGradient()
120 
121    Output Parameters:
122 .  J - Hessian
123 .  B - preconditioner, same as Hessian
124 .  flag - matrix flag
125 
126    Options Database Key:
127 $  -snes_fd
128 
129    Notes:
130    This routine is slow and expensive, and is not currently optimized
131    to take advantage of sparsity in the problem.  Although
132    SNESDefaultComputeHessian() is not recommended for general use
133    in large-scale applications, It can be useful in checking the
134    correctness of a user-provided Hessian.
135 
136 .keywords: SNES, finite differences, Hessian
137 
138 .seealso: SNESSetHessian()
139 @*/
140 int SNESDefaultComputeHessian(SNES snes,Vec x1,Mat *J,Mat *B,MatStructure *flag,void *ctx)
141 {
142   return SNESDefaultComputeJacobian(snes,x1,J,B,flag,ctx);
143 }
144