sfepy.discrete.problem module¶
-
class
sfepy.discrete.problem.
Problem
(name, conf=None, functions=None, domain=None, fields=None, equations=None, auto_conf=True, active_only=True)[source]¶ Problem definition, the top-level class holding all data necessary to solve a problem.
It can be constructed from a
ProblemConf
instance using Problem.from_conf() or directly from a problem description file using Problem.from_conf_file()For interactive use, the constructor requires only the equations, nls and ls keyword arguments, see below.
Parameters: - name : str
The problem name.
- conf : ProblemConf instance, optional
The
ProblemConf
describing the problem.- functions : Functions instance, optional
The user functions for boundary conditions, materials, etc.
- domain : Domain instance, optional
The solution
Domain
.- fields : dict, optional
The dictionary of
Field
instances.- equations : Equations instance, optional
The
Equations
to solve. This argument is required when auto_conf is True.- auto_conf : bool
If True, fields and domain are determined by equations.
- active_only : bool
If True, the (tangent) matrices and residual vectors (right-hand sides) contain only active DOFs, see below.
Notes
The Problem is by default created with active_only set to True. Then the (tangent) matrices and residual vectors (right-hand sides) have reduced sizes and contain only the active DOFs, i.e., DOFs not constrained by EBCs or EPBCs.
Setting active_only to False results in full-size vectors and matrices. Then the matrix size non-zeros structure does not depend on the actual E(P)BCs applied. It must be False when using parallel PETSc solvers.
The active DOF connectivities contain all DOFs, with the E(P)BC-constrained ones stored as -1 - <DOF number>, so that the full connectivities can be reconstructed for the matrix graph creation. However, the negative entries mean that the assembled matrices/residuals have zero values at positions corresponding to constrained DOFs.
The resulting linear system then provides a solution increment, that has to be added to the initial guess used to compute the residual, just like in the Newton iterations. The increment of the constrained DOFs is automatically zero.
When solving with a direct solver, the diagonal entries of a matrix at positions corresponding to constrained DOFs has to be set to ones, so that the matrix is not singular, see
sfepy.discrete.evaluate.apply_ebc_to_matrix()
, which is called automatically insfepy.discrete.evaluate.Evaluator.eval_tangent_matrix()
. It is not called automatically inProblem.evaluate()
. Note that setting the diagonal entries to one might not be necessary with iterative solvers, as the zero matrix rows match the zero residual rows, i.e. if the reduced matrix would be regular, then the right-hand side (the residual) is orthogonal to the kernel of the matrix.-
block_solve
(state0=None, status=None, save_results=True, step_hook=None, post_process_hook=None, verbose=True)[source]¶ Call
Problem.solve()
sequentially for the individual matrix blocks of a block-triangular matrix. It is called byProblem.solve()
if the ‘block_solve’ option is set to True.
-
create_evaluable
(expression, try_equations=True, auto_init=False, preserve_caches=False, copy_materials=True, integrals=None, ebcs=None, epbcs=None, lcbcs=None, ts=None, functions=None, mode=’eval’, var_dict=None, strip_variables=True, extra_args=None, active_only=True, verbose=True, **kwargs)[source]¶ Create evaluable object (equations and corresponding variables) from the expression string. Convenience function calling
create_evaluable()
with defaults provided by the Problem instance self.The evaluable can be repeatedly evaluated by calling
eval_equations()
, e.g. for different values of variables.Parameters: - expression : str
The expression to evaluate.
- try_equations : bool
Try to get variables from self.equations. If this fails, variables can either be provided in var_dict, as keyword arguments, or are created automatically according to the expression.
- auto_init : bool
Set values of all variables to all zeros.
- preserve_caches : bool
If True, do not invalidate evaluate caches of variables.
- copy_materials : bool
Work with a copy of self.equations.materials instead of reusing them. Safe but can be slow.
- integrals : Integrals instance, optional
The integrals to be used. Automatically created as needed if not given.
- ebcs : Conditions instance, optional
The essential (Dirichlet) boundary conditions for ‘weak’ mode. If not given, self.ebcs are used.
- epbcs : Conditions instance, optional
The periodic boundary conditions for ‘weak’ mode. If not given, self.epbcs are used.
- lcbcs : Conditions instance, optional
The linear combination boundary conditions for ‘weak’ mode. If not given, self.lcbcs are used.
- ts : TimeStepper instance, optional
The time stepper. If not given, self.ts is used.
- functions : Functions instance, optional
The user functions for boundary conditions, materials etc. If not given, self.functions are used.
- mode : one of ‘eval’, ‘el_avg’, ‘qp’, ‘weak’
The evaluation mode - ‘weak’ means the finite element assembling, ‘qp’ requests the values in quadrature points, ‘el_avg’ element averages and ‘eval’ means integration over each term region.
- var_dict : dict, optional
The variables (dictionary of (variable name) : (Variable instance)) to be used in the expression. Use this if the name of a variable conflicts with one of the parameters of this method.
- strip_variables : bool
If False, the variables in var_dict or kwargs not present in the expression are added to the actual variables as a context.
- extra_args : dict, optional
Extra arguments to be passed to terms in the expression.
- active_only : bool
If True, in ‘weak’ mode, the (tangent) matrices and residual vectors (right-hand sides) contain only active DOFs.
- verbose : bool
If False, reduce verbosity.
- **kwargs : keyword arguments
Additional variables can be passed as keyword arguments, see var_dict.
Returns: - equations : Equations instance
The equations that can be evaluated.
- variables : Variables instance
The corresponding variables. Set their values and use
eval_equations()
.
Examples
problem is Problem instance.
>>> out = problem.create_evaluable('ev_volume_integrate.i1.Omega(u)') >>> equations, variables = out
vec is a vector of coefficients compatible with the field of ‘u’ - let’s use all ones.
>>> vec = nm.ones((variables['u'].n_dof,), dtype=nm.float64) >>> variables['u'].set_data(vec) >>> vec_qp = eval_equations(equations, variables, mode='qp')
Try another vector:
>>> vec = 3 * nm.ones((variables['u'].n_dof,), dtype=nm.float64) >>> variables['u'].set_data(vec) >>> vec_qp = eval_equations(equations, variables, mode='qp')
-
create_materials
(mat_names=None)[source]¶ Create materials with names in mat_names. Their definitions have to be present in self.conf.materials.
Notes
This method does not change self.equations, so it should not have any side effects.
-
create_subproblem
(var_names, known_var_names)[source]¶ Create a sub-problem with equations containing only terms with the given virtual variables.
Parameters: - var_names : list
The list of names of virtual variables.
- known_var_names : list
The list of names of (already) known state variables.
Returns: - subpb : Problem instance
The sub-problem.
-
create_variables
(var_names=None)[source]¶ Create variables with names in var_names. Their definitions have to be present in self.conf.variables.
Notes
This method does not change self.equations, so it should not have any side effects.
-
eval_equations
(names=None, preserve_caches=False, mode=’eval’, dw_mode=’vector’, term_mode=None, active_only=True, verbose=True)[source]¶ Evaluate (some of) the problem’s equations, convenience wrapper of
eval_equations()
.Parameters: - names : str or sequence of str, optional
Evaluate only equations of the given name(s).
- preserve_caches : bool
If True, do not invalidate evaluate caches of variables.
- mode : one of ‘eval’, ‘el_avg’, ‘qp’, ‘weak’
The evaluation mode - ‘weak’ means the finite element assembling, ‘qp’ requests the values in quadrature points, ‘el_avg’ element averages and ‘eval’ means integration over each term region.
- dw_mode : ‘vector’ or ‘matrix’
The assembling mode for ‘weak’ evaluation mode.
- term_mode : str
The term call mode - some terms support different call modes and depending on the call mode different values are returned.
- verbose : bool
If False, reduce verbosity.
Returns: - out : dict or result
The evaluation result. In ‘weak’ mode it is the vector or sparse matrix, depending on dw_mode. Otherwise, it is a dict of results with equation names as keys or a single result for a single equation.
-
evaluate
(expression, try_equations=True, auto_init=False, preserve_caches=False, copy_materials=True, integrals=None, ebcs=None, epbcs=None, lcbcs=None, ts=None, functions=None, mode=’eval’, dw_mode=’vector’, term_mode=None, var_dict=None, strip_variables=True, ret_variables=False, active_only=True, verbose=True, extra_args=None, **kwargs)[source]¶ Evaluate an expression, convenience wrapper of
Problem.create_evaluable()
andeval_equations()
.Parameters: - dw_mode : ‘vector’ or ‘matrix’
The assembling mode for ‘weak’ evaluation mode.
- term_mode : str
The term call mode - some terms support different call modes and depending on the call mode different values are returned.
- ret_variables : bool
If True, return the variables that were created to evaluate the expression.
- other : arguments
See docstrings of
Problem.create_evaluable()
.
Returns: - out : array
The result of the evaluation.
- variables : Variables instance
The variables that were created to evaluate the expression. Only provided if ret_variables is True.
-
static
from_conf_file
(conf_filename, required=None, other=None, init_fields=True, init_equations=True, init_solvers=True)[source]¶
-
get_dim
(get_sym=False)[source]¶ Returns mesh dimension, symmetric tensor dimension (if get_sym is True).
-
get_evaluator
(reuse=False)[source]¶ Either create a new Evaluator instance (reuse == False), or return an existing instance, created in a preceding call to Problem.init_solvers().
-
get_integrals
(names=None)[source]¶ Get integrals, initialized from problem configuration if available.
Parameters: - names : list, optional
If given, only the named integrals are returned.
Returns: - integrals : Integrals instance
The requested integrals.
-
get_nls_functions
()[source]¶ Returns functions to be used by a nonlinear solver to evaluate the nonlinear function value (the residual) and its gradient (the tangent matrix) corresponding to the problem equations.
Returns: - fun : function
The function
fun(x)
for computing the residual.- fun_grad : function
The function
fun_grad(x)
for computing the tangent matrix.- iter_hook : function
The optional (user-defined) function to be called before each nonlinear solver iteration iteration.
-
get_output_name
(suffix=None, extra=None, mode=None)[source]¶ Return default output file name, based on the output directory, output format, step suffix and mode. If present, the extra string is put just before the output format suffix.
-
get_restart_filename
(ts=None)[source]¶ If restarts are allowed in problem definition options, return the restart file name, based on the output directory and time step.
-
get_tss_functions
(state0, update_bcs=True, update_materials=True, save_results=True, step_hook=None, post_process_hook=None)[source]¶ Get the problem-dependent functions required by the time-stepping solver during the solution process.
Parameters: - state0 : State
The state holding the problem variables.
- update_bcs : bool, optional
If True, update the boundary conditions in each prestep_fun call.
- update_materials : bool, optional
If True, update the values of material parameters in each prestep_fun call.
- save_results : bool, optional
If True, save the results in each poststep_fun call.
- step_hook : callable, optional
The optional user-defined function that is called in each poststep_fun call before saving the results.
- post_process_hook : callable, optional
The optional user-defined function that is passed in each poststep_fun to
Problem.save_state()
.
Returns: - init_fun : callable
The initialization function called before the actual time-stepping.
- prestep_fun : callable
The function called in each time (sub-)step prior to the nonlinear solver call.
- poststep_fun : callable
The function called at the end of each time step.
-
init_solvers
(status=None, ls_conf=None, nls_conf=None, ts_conf=None, force=False)[source]¶ Create and initialize solver instances.
Parameters: - status : dict-like, IndexedStruct, optional
The user-supplied object to hold the time-stepping/nonlinear solver convergence statistics.
- ls_conf : Struct, optional
The linear solver options.
- nls_conf : Struct, optional
The nonlinear solver options.
- force : bool
If True, re-create the solver instances even if they already exist in self.nls attribute.
-
load_restart
(filename, state=None, ts=None)[source]¶ Load the current state and time step from a restart file.
Alternatively, a regular output file in the HDF5 format can be used in place of the restart file. In that case the restart is only approximate, because higher order field DOFs (if any) were stripped out. Files with the adaptive linearization are not supported. Use with caution!
Parameters: - filename : str
The restart file name.
- state : State instance, optional
The state instance. If not given, a new state is created using the variables in problem equations. Otherwise, its variables are modified in place.
- ts : TimeStepper instance, optional
The time stepper. If not given, a default one is created. Otherwise, it is modified in place.
Returns: - new_state : State instance
The loaded state.
-
refine_uniformly
(level)[source]¶ Refine the mesh uniformly level-times.
Notes
This operation resets almost everything (fields, equations, …) - it is roughly equivalent to creating a new Problem instance with the refined mesh.
-
save_ebc
(filename, ebcs=None, epbcs=None, force=True, default=0.0)[source]¶ Save essential boundary conditions as state variables.
Parameters: - filename : str
The output file name.
- ebcs : Conditions instance, optional
The essential (Dirichlet) boundary conditions. If not given, self.conf.ebcs are used.
- epbcs : Conditions instance, optional
The periodic boundary conditions. If not given, self.conf.epbcs are used.
- force : bool
If True, sequential nonzero values are forced to individual ebcs so that the conditions are visible even when zero.
- default : float
The default constant value of state vector.
-
save_regions
(filename_trunk, region_names=None)[source]¶ Save regions as meshes.
Parameters: - filename_trunk : str
The output filename without suffix.
- region_names : list, optional
If given, only the listed regions are saved.
-
save_regions_as_groups
(filename_trunk, region_names=None)[source]¶ Save regions in a single mesh but mark them by using different element/node group numbers.
See
Domain.save_regions_as_groups()
for more details.Parameters: - filename_trunk : str
The output filename without suffix.
- region_names : list, optional
If given, only the listed regions are saved.
-
save_restart
(filename, state=None, ts=None)[source]¶ Save the current state and time step to a restart file.
Parameters: - filename : str
The restart file name.
- state : State instance, optional
The state instance. If not given, a new state is created using the variables in problem equations.
- ts : TimeStepper instance, optional
The time stepper. If not given, a default one is created.
Notes
Does not support terms with internal state.
-
save_state
(filename, state=None, out=None, fill_value=None, post_process_hook=None, linearization=None, file_per_var=False, **kwargs)[source]¶ Parameters: - file_per_var : bool or None
If True, data of each variable are stored in a separate file. If None, it is set to the application option value.
- linearization : Struct or None
The linearization configuration for higher order approximations. If its kind is ‘adaptive’, file_per_var is assumed True.
-
set_conf_solvers
(conf_solvers=None, options=None)[source]¶ Choose which solvers should be used. If solvers are not set in options, use first suitable in conf_solvers.
-
set_equations
(conf_equations=None, user=None, keep_solvers=False, make_virtual=False)[source]¶ Set equations of the problem using the equations problem description entry.
Fields and Regions have to be already set.
-
set_equations_instance
(equations, keep_solvers=False)[source]¶ Set equations of the problem to equations.
-
set_mesh_coors
(coors, update_fields=False, actual=False, clear_all=True, extra_dofs=False)[source]¶ Set mesh coordinates.
Parameters: - coors : array
The new coordinates.
- update_fields : bool
If True, update also coordinates of fields.
- actual : bool
If True, update the actual configuration coordinates, otherwise the undeformed configuration ones.
-
set_output_dir
(output_dir=None)[source]¶ Set the directory for output files.
The directory is created if it does not exist.
-
set_solver
(solver, status=None)[source]¶ Set a time-stepping or nonlinear solver to be used in
Problem.solve()
call.Parameters: - solver : NonlinearSolver or TimeSteppingSolver instance
The nonlinear or time-stepping solver.
Notes
A copy of the solver is used, and the nonlinear solver functions are set to those returned by
Problem.get_nls_functions()
, if not set already. If a nonlinear solver is set, a default StationarySolver instance is created automatically as the time-stepping solver. Also sets self.ts attribute.
-
setup_default_output
(conf=None, options=None)[source]¶ Provide default values to Problem.setup_output() from conf.options and options.
-
setup_hooks
(options=None)[source]¶ Setup various hooks (user-defined functions), as given in options.
Supported hooks:
- matrix_hook
- check/modify tangent matrix in each nonlinear solver iteration
- nls_iter_hook
- called prior to every iteration of nonlinear solver, if the solver supports that
- takes the Problem instance (self) as the first argument
- matrix_hook
-
setup_output
(output_filename_trunk=None, output_dir=None, output_format=None, float_format=None, file_per_var=None, linearization=None)[source]¶ Sets output options to given values, or uses the defaults for each argument that is None.
-
solve
(state0=None, status=None, force_values=None, var_data=None, update_bcs=True, update_materials=True, save_results=True, step_hook=None, post_process_hook=None, post_process_hook_final=None, verbose=True)[source]¶ Solve the problem equations by calling the top-level solver.
Before calling this function the top-level solver has to be set, see
Problem.set_solver()
. Also, the boundary conditions and the initial conditions (for time-dependent problems) has to be set, seeProblem.set_bcs()
,Problem.set_ics()
.Parameters: - state0 : State or array, optional
If given, the initial state satisfying the initial conditions. By default, it is created and the initial conditions are applied automatically.
- status : dict-like, optional
The user-supplied object to hold the solver convergence statistics.
- force_values : dict of floats or float, optional
If given, the supplied values override the values of the essential boundary conditions.
- var_data : dict, optional
A dictionary of {variable_name : data vector} used to initialize parameter variables.
- update_bcs : bool, optional
If True, update the boundary conditions in each prestep_fun call. See
Problem.get_tss_functions()
.- update_materials : bool, optional
If True, update the values of material parameters in each prestep_fun call. See
Problem.get_tss_functions()
.- save_results : bool, optional
If True, save the results in each poststep_fun call. See
Problem.get_tss_functions()
.- step_hook : callable, optional
The optional user-defined function that is called in each poststep_fun call before saving the results. See
Problem.get_tss_functions()
.- post_process_hook : callable, optional
The optional user-defined function that is passed in each poststep_fun to
Problem.save_state()
. SeeProblem.get_tss_functions()
.- post_process_hook_final : callable, optional
The optional user-defined function that is called after the top-level solver returns.
Returns: - state : State
The final state.
-
time_update
(ts=None, ebcs=None, epbcs=None, lcbcs=None, functions=None, create_matrix=False, is_matrix=True)[source]¶
-
update_equations
(ts=None, ebcs=None, epbcs=None, lcbcs=None, functions=None, create_matrix=False, is_matrix=True)[source]¶ Update equations for current time step.
The tangent matrix graph is automatically recomputed if the set of active essential or periodic boundary conditions changed w.r.t. the previous time step.
Parameters: - ts : TimeStepper instance, optional
The time stepper. If not given, self.ts is used.
- ebcs : Conditions instance, optional
The essential (Dirichlet) boundary conditions. If not given, self.ebcs are used.
- epbcs : Conditions instance, optional
The periodic boundary conditions. If not given, self.epbcs are used.
- lcbcs : Conditions instance, optional
The linear combination boundary conditions. If not given, self.lcbcs are used.
- functions : Functions instance, optional
The user functions for boundary conditions, materials, etc. If not given, self.functions are used.
- create_matrix : bool
If True, force the matrix graph computation.
- is_matrix : bool
If False, the matrix is not created. Has precedence over create_matrix.
-
update_materials
(ts=None, mode=’normal’, verbose=True)[source]¶ Update materials used in equations.
Parameters: - ts : TimeStepper instance
The time stepper.
- mode : ‘normal’, ‘update’ or ‘force’
The update mode, see
Material.time_update()
.- verbose : bool
If False, reduce verbosity.