1 2 #include <petsc/private/snesimpl.h> /*I "petscsnes.h" I*/ 3 4 #undef __FUNCT__ 5 #define __FUNCT__ "SNESComputeJacobianDefault" 6 /*@C 7 SNESComputeJacobianDefault - Computes the Jacobian using finite differences. 8 9 Collective on SNES 10 11 Input Parameters: 12 + x1 - compute Jacobian at this point 13 - ctx - application's function context, as set with SNESSetFunction() 14 15 Output Parameters: 16 + J - Jacobian matrix (not altered in this routine) 17 - B - newly computed Jacobian matrix to use with preconditioner (generally the same as J) 18 19 Options Database Key: 20 + -snes_fd - Activates SNESComputeJacobianDefault() 21 . -snes_test_err - Square root of function error tolerance, default square root of machine 22 epsilon (1.e-8 in double, 3.e-4 in single) 23 - -mat_fd_type - Either wp or ds (see MATMFFD_WP or MATMFFD_DS) 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 SNESComputeJacobianDefault() 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 An alternative routine that uses coloring to exploit matrix sparsity is 33 SNESComputeJacobianDefaultColor(). 34 35 Level: intermediate 36 37 .keywords: SNES, finite differences, Jacobian 38 39 .seealso: SNESSetJacobian(), SNESComputeJacobianDefaultColor(), MatCreateSNESMF() 40 @*/ 41 PetscErrorCode SNESComputeJacobianDefault(SNES snes,Vec x1,Mat J,Mat B,void *ctx) 42 { 43 Vec j1a,j2a,x2; 44 PetscErrorCode ierr; 45 PetscInt i,N,start,end,j,value,root; 46 PetscScalar dx,*y,wscale; 47 const PetscScalar *xx; 48 PetscReal amax,epsilon = PETSC_SQRT_MACHINE_EPSILON; 49 PetscReal dx_min = 1.e-16,dx_par = 1.e-1,unorm; 50 MPI_Comm comm; 51 PetscErrorCode (*eval_fct)(SNES,Vec,Vec)=0; 52 PetscBool assembled,use_wp = PETSC_TRUE,flg; 53 const char *list[2] = {"ds","wp"}; 54 PetscMPIInt size; 55 const PetscInt *ranges; 56 57 PetscFunctionBegin; 58 ierr = PetscOptionsGetReal(((PetscObject)snes)->prefix,"-snes_test_err",&epsilon,0);CHKERRQ(ierr); 59 eval_fct = SNESComputeFunction; 60 61 ierr = PetscObjectGetComm((PetscObject)x1,&comm);CHKERRQ(ierr); 62 ierr = MPI_Comm_size(comm,&size);CHKERRQ(ierr); 63 ierr = MatAssembled(B,&assembled);CHKERRQ(ierr); 64 if (assembled) { 65 ierr = MatZeroEntries(B);CHKERRQ(ierr); 66 } 67 if (!snes->nvwork) { 68 snes->nvwork = 3; 69 70 ierr = VecDuplicateVecs(x1,snes->nvwork,&snes->vwork);CHKERRQ(ierr); 71 ierr = PetscLogObjectParents(snes,snes->nvwork,snes->vwork);CHKERRQ(ierr); 72 } 73 j1a = snes->vwork[0]; j2a = snes->vwork[1]; x2 = snes->vwork[2]; 74 75 ierr = VecGetSize(x1,&N);CHKERRQ(ierr); 76 ierr = VecGetOwnershipRange(x1,&start,&end);CHKERRQ(ierr); 77 ierr = (*eval_fct)(snes,x1,j1a);CHKERRQ(ierr); 78 79 ierr = PetscOptionsBegin(PetscObjectComm((PetscObject)snes),((PetscObject)snes)->prefix,"Differencing options","SNES");CHKERRQ(ierr); 80 ierr = PetscOptionsEList("-mat_fd_type","Algorithm to compute difference parameter","SNESComputeJacobianDefault",list,2,"wp",&value,&flg);CHKERRQ(ierr); 81 ierr = PetscOptionsEnd();CHKERRQ(ierr); 82 if (flg && !value) use_wp = PETSC_FALSE; 83 84 if (use_wp) { 85 ierr = VecNorm(x1,NORM_2,&unorm);CHKERRQ(ierr); 86 } 87 /* Compute Jacobian approximation, 1 column at a time. 88 x1 = current iterate, j1a = F(x1) 89 x2 = perturbed iterate, j2a = F(x2) 90 */ 91 for (i=0; i<N; i++) { 92 ierr = VecCopy(x1,x2);CHKERRQ(ierr); 93 if (i>= start && i<end) { 94 ierr = VecGetArrayRead(x1,&xx);CHKERRQ(ierr); 95 if (use_wp) dx = PetscSqrtReal(1.0 + unorm); 96 else dx = xx[i-start]; 97 ierr = VecRestoreArrayRead(x1,&xx);CHKERRQ(ierr); 98 if (PetscAbsScalar(dx) < dx_min) dx = (PetscRealPart(dx) < 0. ? -1. : 1.) * dx_par; 99 dx *= epsilon; 100 wscale = 1.0/dx; 101 ierr = VecSetValues(x2,1,&i,&dx,ADD_VALUES);CHKERRQ(ierr); 102 } else { 103 wscale = 0.0; 104 } 105 ierr = VecAssemblyBegin(x2);CHKERRQ(ierr); 106 ierr = VecAssemblyEnd(x2);CHKERRQ(ierr); 107 ierr = (*eval_fct)(snes,x2,j2a);CHKERRQ(ierr); 108 ierr = VecAXPY(j2a,-1.0,j1a);CHKERRQ(ierr); 109 /* Communicate scale=1/dx_i to all processors */ 110 ierr = VecGetOwnershipRanges(x1,&ranges);CHKERRQ(ierr); 111 root = size; 112 for (j=size-1; j>-1; j--) { 113 root--; 114 if (i>=ranges[j]) break; 115 } 116 ierr = MPI_Bcast(&wscale,1,MPIU_SCALAR,root,comm);CHKERRQ(ierr); 117 118 ierr = VecScale(j2a,wscale);CHKERRQ(ierr); 119 ierr = VecNorm(j2a,NORM_INFINITY,&amax);CHKERRQ(ierr); amax *= 1.e-14; 120 ierr = VecGetArray(j2a,&y);CHKERRQ(ierr); 121 for (j=start; j<end; j++) { 122 if (PetscAbsScalar(y[j-start]) > amax || j == i) { 123 ierr = MatSetValues(B,1,&j,1,&i,y+j-start,INSERT_VALUES);CHKERRQ(ierr); 124 } 125 } 126 ierr = VecRestoreArray(j2a,&y);CHKERRQ(ierr); 127 } 128 ierr = MatAssemblyBegin(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 129 ierr = MatAssemblyEnd(B,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 130 if (B != J) { 131 ierr = MatAssemblyBegin(J,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 132 ierr = MatAssemblyEnd(J,MAT_FINAL_ASSEMBLY);CHKERRQ(ierr); 133 } 134 PetscFunctionReturn(0); 135 } 136 137 138