SolverControl Class Reference
[Linear solver classes]

Inheritance diagram for SolverControl:

Inheritance graph
[legend]

List of all members.

Classes

class  ExcHistoryDataRequired
class  NoConvergence

Public Types

enum  State { iterate = 0, success, failure }

Public Member Functions

 SolverControl (const unsigned int n=100, const double tol=1.e-10, const bool log_history=false, const bool log_result=true)
virtual ~SolverControl ()
void parse_parameters (ParameterHandler &param)
virtual State check (const unsigned int step, const double check_value)
State last_check () const
double initial_value () const
double last_value () const
unsigned int last_step () const
unsigned int max_steps () const
unsigned int set_max_steps (const unsigned int)
void set_failure_criterion (const double rel_failure_residual)
void clear_failure_criterion ()
double tolerance () const
double set_tolerance (const double)
void enable_history_data ()
double average_reduction () const
double final_reduction () const
double step_reduction (unsigned int step) const
void log_history (const bool)
bool log_history () const
unsigned int log_frequency (unsigned int)
void log_result (const bool)
bool log_result () const

Static Public Member Functions

static void declare_parameters (ParameterHandler &param)

Protected Attributes

unsigned int maxsteps
double tol
State lcheck
double initial_val
double lvalue
unsigned int lstep
bool check_failure
double relative_failure_residual
double failure_residual
bool m_log_history
unsigned int m_log_frequency
bool m_log_result
bool history_data_enabled
std::vector< doublehistory_data


Detailed Description

Control class for iterative solvers.

Used by iterative methods to determine whether the iteration should be continued. To this respect, the virtual function check() is called in each iteration with the current iteration step and the value indicating convergence (usually the residual).

After the iteration has terminated, the functions last_value and last_step can be used to obtain information about the final state of the iteration.

check() can be replaced in derived classes to allow for more sophisticated tests.

State

The return states of the check function are of type State, which is an enum local to this class. It indicates the state the solver is in.

The possible values of State are


Member Enumeration Documentation

Enum denoting the different states a solver can be in. See the general documentation of this class for more information.

Enumerator:
iterate  Continue iteration.
success  Stop iteration, goal reached.
failure  Stop iteration, goal not reached.


Constructor & Destructor Documentation

SolverControl::SolverControl ( const unsigned int  n = 100,
const double  tol = 1.e-10,
const bool  log_history = false,
const bool  log_result = true 
)

Constructor. The parameters n and tol are the maximum number of iteration steps before failure and the tolerance to determine success of the iteration.

log_history specifies whether the history (i.e. the value to be checked and the number of the iteration step) shall be printed to deallog stream. Default is: do not print. Similarly, log_result specifies the whether the final result is logged to deallog. Default is yes.

virtual SolverControl::~SolverControl (  )  [virtual]

Virtual destructor is needed as there are virtual functions in this class.


Member Function Documentation

static void SolverControl::declare_parameters ( ParameterHandler param  )  [static]

Interface to parameter file.

Reimplemented in ReductionControl.

void SolverControl::parse_parameters ( ParameterHandler param  ) 

Read parameters from file.

Reimplemented in ReductionControl.

virtual State SolverControl::check ( const unsigned int  step,
const double  check_value 
) [virtual]

Decide about success or failure of an iteration. This function gets the current iteration step to determine, whether the allowed number of steps has been exceeded and returns failure in this case. If check_value is below the prescribed tolerance, it returns success. In all other cases iterate is returned to suggest continuation of the iterative procedure.

The iteration is also aborted if the residual becomes a denormalized value (NaN). Note, however, that this check is only performed if the isnan function is provided by the operating system, which is not always true. The configure scripts checks for this and sets the flag HAVE_ISNAN in the file Make.global_options if this function was found.

check() additionally preserves step and check_value. These values are accessible by last_value() and last_step().

Derived classes may overload this function, e.g. to log the convergence indicators (check_value) or to do other computations.

Reimplemented in ReductionControl.

Referenced by EigenInverse< VECTOR >::solve(), and EigenPower< VECTOR >::solve().

State SolverControl::last_check (  )  const

Return the result of the last check operation.

double SolverControl::initial_value (  )  const

Return the initial convergence criterion.

double SolverControl::last_value (  )  const

Return the convergence value of last iteration step for which check was called by the solver.

unsigned int SolverControl::last_step (  )  const

Number of last iteration step.

unsigned int SolverControl::max_steps (  )  const

Maximum number of steps.

unsigned int SolverControl::set_max_steps ( const unsigned  int  ) 

Change maximum number of steps.

void SolverControl::set_failure_criterion ( const double  rel_failure_residual  ) 

Enables the failure check. Solving is stopped with ReturnState failure if residual>failure_residual with failure_residual:=rel_failure_residual*first_residual.

void SolverControl::clear_failure_criterion (  ) 

Disables failure check and resets relative_failure_residual and failure_residual to zero.

double SolverControl::tolerance (  )  const

Tolerance.

double SolverControl::set_tolerance ( const   double  ) 

Change tolerance.

void SolverControl::enable_history_data (  ) 

Enables writing residuals of each step into a vector for later analysis.

double SolverControl::average_reduction (  )  const

Average error reduction over all steps.

Requires enable_history_data()

double SolverControl::final_reduction (  )  const

Error reduction of the last step; for stationary iterations, this approximates the norm of the iteration matrix.

Requires enable_history_data()

double SolverControl::step_reduction ( unsigned int  step  )  const

Error reduction of any iteration step.

Requires enable_history_data()

void SolverControl::log_history ( const   bool  ) 

Log each iteration step. Use log_frequency for skipping steps.

bool SolverControl::log_history (  )  const

Returns the log_history flag.

unsigned int SolverControl::log_frequency ( unsigned  int  ) 

Set logging frequency.

void SolverControl::log_result ( const   bool  ) 

Log start and end step.

bool SolverControl::log_result (  )  const

Returns the log_result flag.


Member Data Documentation

Maximum number of steps.

Prescribed tolerance to be achieved.

Result of last check operation.

Initial value.

Last value of the convergence criterion.

Last step.

Is set to true by set_failure_criterion and enables failure checking.

failure_residual equals the first residual multiplied by relative_crit set by set_failure_criterion (see there).

Until the first residual is known it is 0.

Log convergence history to deallog.

Log only every nth step.

Log iteration result to deallog. If true, after finishing the iteration, a statement about failure or success together with lstep and lvalue are logged.

Control over the storage of history data. Set by enable_history_data().

std::vector<double> SolverControl::history_data [protected]

Vector storing the result after each iteration step for later statistical analysis.

Use of this vector is enabled by enable_history_data().


The documentation for this class was generated from the following file:

deal.II documentation generated on Sat Aug 15 16:52:18 2009 by doxygen 1.5.9