FEATool Multiphysics  v1.16.4
Finite Element Analysis Toolbox
assemblej.m File Reference


ASSEMBLEJ Finite element assembly of Jacobian forms.


Assembles a numerical Jacobian matrix representation of a finite element bilinear and linear forms.

IPERT indicates to perturb the solution vector (=0) with DPERT (default 1e-8) calculating the Jacobian J = dK/du*U - f, where each column J_i is given as

J_i = ( (K(U_i+DPERT)*(U_i+DPERT)-f(U_i+DPERT)) - (K(U)*U-f(U)) )/DPERT

Alternatively, if IPERT is equal to an integer (1, 2 or 3) the Jacobian with respect to space dimension/coordinate IPERT will be computed as

J_i = ( (K(U,x+DPERT)*(U,x+DPERT)-f(U,x+DPERT)) - (K(U,x)*U-f(U,x)) )/DPERT

The linearization point/solution vector U needs to be given and stored in the prob.sol.u field. If it is not present a zero solution will be used.

Each column in AFORM and FFORM specify bilinear and linear additive terms to compute. In AFORM the first row corresponds to the trial function space, and second test space. The form entries denote either function values or derivatives to evaluate.

C_SFUN indicates which shape functions to use for the trial and test function spaces. Optionally, a third entry is required for the trial function space when assembling off-diagonal (target) blocks for coupled problems. If only one or two C_SFUN entries is provided, the first entry will be used for target trial function space.

For off-diagonal blocks the dof OFFSET ( dof_offset_row, dof_offset_column ) is used to find and perturb the target block solution vector.

C_ACOEF and C_FCOEF specifies the coefficients to evaluate for the forms.

The computed matrix entries are returned in VJVALS where VROWINDS and VROWCOLS are corresponding row and column pointers. N_ROWS and N_COLS give the number of rows in the matrix. J sparse representation of the matrix can subsequently be constructed by calling SPARSE

J = sparse( vRowInds, vColInds, vJvals, n_rows, n_cols );

on the output of ASSEMLBLEJ.

Input       Value/[Size]           Description
aform       [2,n_aforms]           Bilinear forms to compute, first row indicates
                                   trial function space (second test function space)
                                    1     = function value
                                    2/3/4 = x/y/z -derivatives
fform       [n_fforms]             Linear forms to compute
c_sfun      strings {2/3}          Shape functions used for the assmebly trial,
                                   assembly test, and target trial function spaces
c_acoef     {n_aforms}             Coefficients for each bilinear form
c_fcoef     {n_fforms}             Coefficients for each linear form
i_cub       scalar                 Numerical integration rule
p           [n_sdim,n_p]           Array with grid point coordinates
c           [n_vert,n_cells]       Array with cell connectivities
a           [n_vert,n_cells]       Array with cell adjacency information
ind_s       [n_cells]              (optional) Subdomain indices
prob        struct                 (optional) FEATool problem struct, used
offset      [2]                    (optional) Offset for finding start in solution
                                   for matrix coef/vector product dof perturbation
ipert       scalar/{0/i_sdim}      0-Solution perturbation, >0 Coordinate dir. pert
dpert       scalar                 (optional) Perturbation parameter (default 1e-8)
n_cmax      scalar {50000}         Max number of cells per group to assemble
                                   for at once (to limit memory consumption)
Output      Value/[Size]           Description
vRowInds    [n_J]                  Row pointers to matrix entries
vColInds    [n_J]                  Column pointers to matrix entries
vJvals      [n_J]                  Values of the assembled matrix entries
n_rows      scalar                 Number of rows in matrix
n_cols      scalar                 Number of columns in matrix
See also
assemjac, mapdofbdr