This object serves for storing parameters and data used in numerical continuation of solution branches of models (for more details about continuation see the Getfem++ user documentation).
S = gf_cont_struct(model md, string dataname_parameter[,string dataname_init, string dataname_final, string dataname_current], scalar sc_fac[, ...])
General constructor for cont_struct objects.
This object serves for storing parameters and data used in numerical continuation of solution branches of models (for more details about continuation see the Getfem++ user documentation).
S = gf_cont_struct(model md, string dataname_parameter[,string dataname_init, string dataname_final, string dataname_current], scalar sc_fac[, ...])
The variable dataname_parameter
should parametrise the model given by
md
. If the parametrisation is done via a vector datum, dataname_init
and dataname_final
should store two given values of this datum
determining the parametrisation, and dataname_current
serves for actual
values of this datum. sc_fac
is a scale factor involved in the weighted
norm used in the continuation.
Additional options:
- 'lsolver', string SOLVER_NAME
name of the solver to be used for the incorporated linear systems
(the default value is 'auto', which lets getfem choose itself);
possible values are 'superlu', 'mumps' (if supported), 'cg/ildlt',
'gmres/ilu' and 'gmres/ilut';
- 'bifurcations'
activates tools for detection and treatment of bifurcation points;
- 'h_init', scalar HIN
initial step size (the default value is 1e-2);
- 'h_max', scalar HMAX
maximum step size (the default value is 1e-1);
- 'h_min', scalar HMIN
minimum step size (the default value is 1e-5);
- 'h_inc', scalar HINC
factor for enlarging the step size (the default value is 1.3);
- 'h_dec', scalar HDEC
factor for diminishing the step size (the default value is 0.5);
- 'max_iter', int MIT
maximum number of iterations allowed in the correction (the default
value is 10);
- 'thr_iter', int TIT
threshold number of iterations of the correction for enlarging the
step size (the default value is 4);
- 'max_res', scalar RES
target residual value of a new point on the solution curve (the
default value is 1e-6);
- 'max_diff', scalar DIFF
determines a convergence criterion for two consecutive points (the
default value is 1e-6);
- 'min_cos', scalar MCOS
minimal value of the cosine of the angle between tangents to the
solution curve at an old point and a new one (the default value is
0.9);
- 'max_res_solve', scalar RES_SOLVE
target residual value for the linear systems to be solved (the
default value is 1e-8);
- 'non-smooth'
determines that some special methods for non-smooth problems can be
used;
- 'delta_max', scalar DMAX
maximum size of division for evaluating the test function on the
convex combination of two augmented Jacobians that belong to different
smooth pieces (the default value is 0.005);
- 'delta_min', scalar DMIN
minimum size of division for evaluating the test function on the
convex combination (the default value is 0.00012);
- 'thr_var', scalar TVAR
threshold variation for refining the division (the default value is
0.02);
- 'nb_dir', int NDIR
number of linear combinations of vectors in one subspace when
searching for new tangent predictions during location of new one-sided
branches (the default value is 40);
- 'nb_comb', int NCOMB
maximum number of couples of reference vectors forming the linear
combinations (the default value is 1);
- 'noisy' or 'very_noisy'
determines how detailed information has to be displayed during the
continuation process (residual values etc.).
Y. Collette