FEATool Multiphysics
v1.17.2
Finite Element Analysis Toolbox
|
ASSEMJAC Assemble monolithic Jacobian matrix.
[ J, T_J, T_SP ] = ASSEMJAC( PROB, U, METHOD, SOLCOMP, DPERT, I_CUB, F_SPARSE )
Called from assemblenjac to assemble a monolithic Jacobian matrix J.
The METHOD flag indicates computing the Jacobian J = d(K(U)*U-f(U))/dU with respect to the solution and linearization point U (METHOD=0), where each column J_i is computed as
J_i = ( (K(U_i+DPERT)*(U_i+DPERT)-f(U_i+DPERT)) - (K(U)*U-f(U)) )/DPERT
or alternatively a coordinate direction (METHOD=j), Jj = d(K(U,xj)*U-f(U,xj))/dxj where
J_i = ( (K(U,xj+DPERT)*(U,xj+DPERT)-f(U,xj+DPERT)) - (K(U,x)*U-f(U,xj)) )/DPERT
The optional input argument SOLCOMP can for (METHOD=0) be specified as a 2 x n (cell) array indicating which row and column (dependent variable) combinations should be computed and returned. The first column corresponds to the row and second column for the blocks. SOLCOMP can also equivalently be prescribed as a cell array of pairwise dependent variable name strings. For (METHOD=1/2/3) SOLCOMP can be a single vector to specify the blocks/dependent variable to compute and return. If SOLCOMP is not prescribed or empty all blocks and dependent variable will be computed.
Input Value/{Default} Description ----------------------------------------------------------------------------------- prob struct Problem definition struct u vector [n_u,1] Solution vector/linearization point, size(n_u,1) method scalar {0/i_sdim} Jacobian computation type 0 - Jacobian with respect to solution 1/2/3 - Jacobian with respect to direction solcomp {all dvars/subd} Dependent variables/subdomains to assemble for dpert scalar/{1e-8} Perturbation parameter icub scalar/{2} Numerical integration rule f_sparse logical/{true} Return sparse/struct matrix format . Output Value Description ----------------------------------------------------------------------------------- J sparse/triplet Assembled Jacobian matrix (size(n_u,n_u)) t_j scalar Time spent assembling Jacobian matrix t_sp scalar Time for sparse matrix conversion