Computes the Jacobian using finite differences.


#include "petscsnes.h"  
PetscErrorCode SNESComputeJacobianDefault(SNES snes, Vec x1, Mat J, Mat B, void *ctx)


Input Parameters#

  • snes - the SNES context

  • x1 - compute Jacobian at this point

  • ctx - application’s function context, as set with SNESSetFunction()

Output Parameters#

  • J - Jacobian matrix (not altered in this routine)

  • B - newly computed Jacobian matrix to use with preconditioner (generally the same as J)

Options Database Keys#

  • -snes_fd - Activates SNESComputeJacobianDefault()

  • -snes_fd_coloring - Activates a faster computation that uses a graph coloring of the matrix

  • -snes_test_err - Square root of function error tolerance, default square root of machine epsilon (1.e-8 in double, 3.e-4 in single)

  • -mat_fd_type - Either wp or ds (see MATMFFD_WP or MATMFFD_DS)


This routine is slow and expensive, and is not currently optimized to take advantage of sparsity in the problem. Although SNESComputeJacobianDefault() is not recommended for general use in large-scale applications, It can be useful in checking the correctness of a user-provided Jacobian.

An alternative routine that uses coloring to exploit matrix sparsity is SNESComputeJacobianDefaultColor().

This routine ignores the maximum number of function evaluations set with SNESSetTolerances() and the function evaluations it performs are not counted in what is returned by of SNESGetNumberFunctionEvals().

This function can be provided to SNESSetJacobian() along with a dense matrix to hold the Jacobian

Developer Note#

The function has a poorly chosen name since it does not mention the use of finite differences

See Also#

SNES: Nonlinear Solvers, SNES, SNESSetJacobian(), SNESComputeJacobianDefaultColor(), MatCreateSNESMF()





Index of all SNES routines
Table of Contents for all manual pages
Index of all manual pages