libMesh::TwostepTimeSolver Class Reference

This class wraps another UnsteadySolver derived class, and compares the results of timestepping with deltat and timestepping with 2*deltat to adjust future timestep lengths. More...

#include <twostep_time_solver.h>

Inheritance diagram for libMesh::TwostepTimeSolver:

Public Types
typedef AdaptiveTimeSolver	Parent
	The parent class. More...
typedef DifferentiableSystem	sys_type
	The type of system. More...

Public Member Functions
	TwostepTimeSolver (sys_type &s)
	Constructor. More...
	~TwostepTimeSolver ()
	Destructor. More...
virtual void	solve () override
	This method solves for the solution at the next timestep. More...
virtual std::pair< unsigned int, Real >	adjoint_solve (const QoISet &qoi_indices) override
	This method solves for the adjoint solution at the next adjoint timestep (or a steady state adjoint solve) More...
virtual void	integrate_qoi_timestep () override
	A method to integrate the system::QoI functionals. More...
virtual void	integrate_adjoint_sensitivity (const QoISet &qois, const ParameterVector &parameter_vector, SensitivityData &sensitivities) override
	A method to integrate the adjoint sensitivity w.r.t a given parameter vector. More...
virtual void	integrate_adjoint_refinement_error_estimate (AdjointRefinementEstimator &adjoint_refinement_error_estimator, ErrorVector &QoI_elementwise_error) override
	A method to compute the adjoint refinement error estimate at the current timestep. More...
virtual void	init () override
	The initialization function. More...
virtual void	reinit () override
	The reinitialization function. More...
virtual void	advance_timestep () override
	This method advances the solution to the next timestep, after a solve() has been performed. More...
virtual void	adjoint_advance_timestep () override
	This method advances the adjoint solution to the previous timestep, after an adjoint_solve() has been performed. More...
virtual void	retrieve_timestep () override
	This method retrieves all the stored solutions at the current system.time. More...
virtual Real	last_completed_timestep_size () override
	Returns system.deltat if fixed timestep solver is used, the complete timestep size (sum of all substeps) if the adaptive time solver is used. More...
virtual Real	error_order () const override
	This method is passed on to the core_time_solver. More...
virtual bool	element_residual (bool get_jacobian, DiffContext &) override
	This method is passed on to the core_time_solver. More...
virtual bool	side_residual (bool get_jacobian, DiffContext &) override
	This method is passed on to the core_time_solver. More...
virtual bool	nonlocal_residual (bool get_jacobian, DiffContext &) override
	This method is passed on to the core_time_solver. More...
virtual std::unique_ptr< DiffSolver > &	diff_solver () override
	An implicit linear or nonlinear solver to use at each timestep. More...
virtual std::unique_ptr< LinearSolver< Number > > &	linear_solver () override
	An implicit linear solver to use for adjoint and sensitivity problems. More...
virtual unsigned int	time_order () const override
virtual void	init_adjoints () override
	Add adjoint vectors and old_adjoint_vectors as per the indices of QoISet. More...
virtual void	init_data () override
	The data initialization function. More...
void	update ()
Number	old_nonlinear_solution (const dof_id_type global_dof_number) const
virtual Real	du (const SystemNorm &norm) const override
	Computes the size of ||u^{n+1} - u^{n}|| in some norm. More...
virtual bool	is_steady () const override
	This is not a steady-state solver. More...
void	set_first_adjoint_step (bool first_adjoint_step_setting)
	A setter for the first_adjoint_step boolean. More...
void	set_first_solve (bool first_solve_setting)
virtual void	before_timestep ()
	This method is for subclasses or users to override to do arbitrary processing between timesteps. More...
const sys_type &	system () const
sys_type &	system ()
void	set_solution_history (const SolutionHistory &_solution_history)
	A setter function users will employ if they need to do something other than save no solution history. More...
SolutionHistory &	get_solution_history ()
	A getter function that returns a reference to the solution history object owned by TimeSolver. More...
bool	is_adjoint () const
	Accessor for querying whether we need to do a primal or adjoint solve. More...
void	set_is_adjoint (bool _is_adjoint_value)
	Accessor for setting whether we need to do a primal or adjoint solve. More...

Static Public Member Functions
static std::string	get_info ()
	Gets a string containing the reference information. More...
static void	print_info (std::ostream &out_stream=libMesh::out)
	Prints the reference information, by default to libMesh::out. More...
static unsigned int	n_objects ()
	Prints the number of outstanding (created, but not yet destroyed) objects. More...
static void	enable_print_counter_info ()
	Methods to enable/disable the reference counter output from print_info() More...
static void	disable_print_counter_info ()

Public Attributes
std::unique_ptr< UnsteadySolver >	core_time_solver
	This object is used to take timesteps. More...
SystemNorm	component_norm
	Error calculations are done in this norm, DISCRETE_L2 by default. More...
std::vector< float >	component_scale
	If component_norms is non-empty, each variable's contribution to the error of a system will also be scaled by component_scale[var], unless component_scale is empty in which case all variables will be weighted equally. More...
Real	target_tolerance
	This tolerance is the target relative error between an exact time integration and a single time step output, scaled by deltat. More...
Real	upper_tolerance
	This tolerance is the maximum relative error between an exact time integration and a single time step output, scaled by deltat. More...
Real	max_deltat
	Do not allow the adaptive time solver to select deltat > max_deltat. More...
Real	min_deltat
	Do not allow the adaptive time solver to select deltat < min_deltat. More...
Real	max_growth
	Do not allow the adaptive time solver to select a new deltat greater than max_growth times the old deltat. More...
Real	completed_timestep_size
	The adaptive time solver's have two notions of deltat. More...
bool	global_tolerance
	This flag, which is true by default, grows (shrinks) the timestep based on the expected global accuracy of the timestepping scheme. More...
std::shared_ptr< NumericVector< Number > >	old_local_nonlinear_solution
	Serial vector of _system.get_vector("_old_nonlinear_solution") This is a shared_ptr so that it can be shared between different derived class instances, as in e.g. More...
bool	quiet
	Print extra debugging information if quiet == false. More...
unsigned int	reduce_deltat_on_diffsolver_failure
	This value (which defaults to zero) is the number of times the TimeSolver is allowed to halve deltat and let the DiffSolver repeat the latest failed solve with a reduced timestep. More...

Protected Types
typedef bool(DifferentiablePhysics::*	ResFuncType) (bool, DiffContext &)
	Definitions of argument types for use in refactoring subclasses. More...
typedef void(DiffContext::*	ReinitFuncType) (Real)
typedef std::map< std::string, std::pair< unsigned int, unsigned int > >	Counts
	Data structure to log the information. More...

Protected Member Functions
virtual Real	calculate_norm (System &, NumericVector< Number > &)
	A helper function to calculate error norms. More...
void	prepare_accel (DiffContext &context)
	If there are second order variables in the system, then we also prepare the accel for those variables so the user can treat them as such. More...
bool	compute_second_order_eqns (bool compute_jacobian, DiffContext &c)
	If there are second order variables, then we need to compute their residual equations and corresponding Jacobian. More...
void	increment_constructor_count (const std::string &name) noexcept
	Increments the construction counter. More...
void	increment_destructor_count (const std::string &name) noexcept
	Increments the destruction counter. More...

Protected Attributes
bool	first_solve
	A bool that will be true the first time solve() is called, and false thereafter. More...
bool	first_adjoint_step
	A bool that will be true the first time adjoint_advance_timestep() is called, (when the primal solution is to be used to set adjoint boundary conditions) and false thereafter. More...
std::vector< std::unique_ptr< NumericVector< Number > > >	old_adjoints
	A vector of pointers to vectors holding the adjoint solution at the last time step. More...
Real	last_step_deltat
	We will need to move the system.time around to ensure that residuals are built with the right deltat and the right time. More...
Real	next_step_deltat
std::unique_ptr< DiffSolver >	_diff_solver
	An implicit linear or nonlinear solver to use at each timestep. More...
std::unique_ptr< LinearSolver< Number > >	_linear_solver
	An implicit linear solver to use for adjoint problems. More...
sys_type &	_system
	A reference to the system we are solving. More...
std::unique_ptr< SolutionHistory >	solution_history
	A std::unique_ptr to a SolutionHistory object. More...
Real	last_deltat
	The deltat for the last completed timestep before the current one. More...

Static Protected Attributes
static Counts	_counts
	Actually holds the data. More...
static Threads::atomic< unsigned int >	_n_objects
	The number of objects. More...
static Threads::spin_mutex	_mutex
	Mutual exclusion object to enable thread-safe reference counting. More...
static bool	_enable_print_counter = true
	Flag to control whether reference count information is printed when print_info is called. More...

Detailed Description

This class wraps another UnsteadySolver derived class, and compares the results of timestepping with deltat and timestepping with 2*deltat to adjust future timestep lengths.

Currently this class only works on fully coupled Systems

This class is part of the new DifferentiableSystem framework, which is still experimental. Users of this framework should beware of bugs and future API changes.

Author

Roy H. Stogner

Date

2007

Definition at line 50 of file twostep_time_solver.h.

Member Typedef Documentation

Counts

typedef std::map<std::string, std::pair<unsigned int, unsigned int> > libMesh::ReferenceCounter::Counts

protectedinherited

Data structure to log the information.

The log is identified by the class name.

Definition at line 119 of file reference_counter.h.

Parent

typedef AdaptiveTimeSolver libMesh::TwostepTimeSolver::Parent

The parent class.

Definition at line 56 of file twostep_time_solver.h.

ReinitFuncType

typedef void(DiffContext::* libMesh::TimeSolver::ReinitFuncType) (Real)

protectedinherited

Definition at line 327 of file time_solver.h.

ResFuncType

typedef bool(DifferentiablePhysics::* libMesh::TimeSolver::ResFuncType) (bool, DiffContext &)

protectedinherited

Definitions of argument types for use in refactoring subclasses.

Definition at line 325 of file time_solver.h.

sys_type

typedef DifferentiableSystem libMesh::TimeSolver::sys_type

inherited

The type of system.

Definition at line 69 of file time_solver.h.

Constructor & Destructor Documentation

TwostepTimeSolver()

libMesh::TwostepTimeSolver::TwostepTimeSolver ( sys_type &  s )

explicit

Constructor.

Requires a reference to the system to be solved.

Definition at line 40 of file twostep_time_solver.C.

References libMesh::AdaptiveTimeSolver::core_time_solver.

  : AdaptiveTimeSolver(s)

{
  // We start with a reasonable time solver: implicit Euler
  core_time_solver = std::make_unique<EulerSolver>(s);
}

~TwostepTimeSolver()

libMesh::TwostepTimeSolver::~TwostepTimeSolver ( )

default

Destructor.

Member Function Documentation

adjoint_advance_timestep()

void libMesh::AdaptiveTimeSolver::adjoint_advance_timestep ( )

overridevirtualinherited

This method advances the adjoint solution to the previous timestep, after an adjoint_solve() has been performed.

This will be done before every UnsteadySolver::adjoint_solve().

Reimplemented from libMesh::UnsteadySolver.

Definition at line 126 of file adaptive_time_solver.C.

References libMesh::TimeSolver::_system, libMesh::AdaptiveTimeSolver::core_time_solver, libMesh::DifferentiableSystem::deltat, libMesh::UnsteadySolver::first_adjoint_step, libMesh::System::get_adjoint_solution(), libMesh::System::get_dof_map(), libMesh::DofMap::get_send_list(), libMesh::System::get_vector(), libMesh::NumericVector< T >::localize(), libMesh::make_range(), libMesh::System::n_qois(), libMesh::UnsteadySolver::old_local_nonlinear_solution, and libMesh::System::time.

{
  // Store the computed full step adjoint solution for future use (sub steps are handled internally by the core time solver)
  core_time_solver->get_solution_history().store(true, _system.time);

  // For the first adjoint step ensure that we use the last primal timestep.
  if (first_adjoint_step)
  {
    _system.deltat = dynamic_cast<DifferentiableSystem &>(_system).time_solver->TimeSolver::last_completed_timestep_size();
    first_adjoint_step = false;
  }

  // Before moving to the next time instant, copy over the current adjoint solutions into _old_adjoint_solutions
  for(auto i : make_range(_system.n_qois()))
  {
   std::string old_adjoint_solution_name = "_old_adjoint_solution";
   old_adjoint_solution_name+= std::to_string(i);
   NumericVector<Number> & old_adjoint_solution_i = _system.get_vector(old_adjoint_solution_name);
   NumericVector<Number> & adjoint_solution_i = _system.get_adjoint_solution(i);
   old_adjoint_solution_i = adjoint_solution_i;
  }

  _system.time -= _system.deltat;

  // For the adaptive time solver, all SH operations
  // are handled by the core_time_solver's SH object
  // Retrieve the primal solution for the next adjoint calculation,
  // by using the core time solver's solution history object.
  core_time_solver->get_solution_history().retrieve(true, _system.time);

  // We also need to tell the core time solver that the adjoint initial conditions have been set
  core_time_solver->set_first_adjoint_step(false);

  // Dont forget to localize the old_nonlinear_solution !
  _system.get_vector("_old_nonlinear_solution").localize
    (*old_local_nonlinear_solution,
    _system.get_dof_map().get_send_list());
}

adjoint_solve()

std::pair< unsigned int, Real > libMesh::TwostepTimeSolver::adjoint_solve ( const QoISet &  qoi_indices )

overridevirtual

This method solves for the adjoint solution at the next adjoint timestep (or a steady state adjoint solve)

Implements libMesh::AdaptiveTimeSolver.

Definition at line 298 of file twostep_time_solver.C.

References libMesh::TimeSolver::_system, libMesh::System::calculate_norm(), libMesh::AdaptiveTimeSolver::completed_timestep_size, libMesh::AdaptiveTimeSolver::core_time_solver, libMesh::DifferentiableSystem::deltat, libMesh::System::get_adjoint_solution(), libMesh::H1, libMesh::TimeSolver::last_deltat, libMesh::make_range(), libMesh::System::n_qois(), libMesh::System::n_vars(), libMesh::out, libMesh::Real, and libMesh::System::time.

{
  // Take the first adjoint 'half timestep'
  core_time_solver->adjoint_solve(qoi_indices);

  // We print the forward 'half solution' norms and we will do so for the adjoints if running in dbg.
  #ifdef DEBUG
    for(auto i : make_range(_system.n_qois()))
    {
      for(auto j : make_range(_system.n_vars()))
        libMesh::out<<"||Z_"<<"("<<_system.time<<";"<<i<<","<<j<<")||_H1: "<<_system.calculate_norm(_system.get_adjoint_solution(i), j,H1)<<std::endl;
    }
  #endif

  // Record the sub step deltat we used for the last adjoint solve.
  last_deltat = _system.deltat;

  // Adjoint advance the timestep
  core_time_solver->adjoint_advance_timestep();

 // We have to contend with the fact that the delta_t set by SolutionHistory will not be the
 // delta_t for the adjoint solve. At time t_i, the adjoint solve uses the same delta_t
 // as the primal solve, pulling the adjoint solution from t_i+1 to t_i.
 // FSH however sets delta_t to the value which takes us from t_i to t_i-1.
 // Therefore use the last_deltat for the solve and reset system delta_t after the solve.
  Real temp_deltat = _system.deltat;
  _system.deltat = last_deltat;

  // The second half timestep
  std::pair<unsigned int, Real> full_adjoint_output = core_time_solver->adjoint_solve(qoi_indices);

  // Record the sub step deltat we used for the last adjoint solve and reset the system deltat to the
  // value set by SolutionHistory.
  last_deltat = _system.deltat;
  _system.deltat = temp_deltat;

  // Record the total size of the last timestep, for a 2StepTS, this is
  // simply twice the deltat for each sub(half) step.
  this->completed_timestep_size = 2.0*_system.deltat;

  return full_adjoint_output;
}

advance_timestep()

void libMesh::AdaptiveTimeSolver::advance_timestep ( )

overridevirtualinherited

This method advances the solution to the next timestep, after a solve() has been performed.

Often this will be done after every UnsteadySolver::solve(), but adaptive mesh refinement and/or adaptive time step selection may require some solve() steps to be repeated.

Reimplemented from libMesh::UnsteadySolver.

Definition at line 89 of file adaptive_time_solver.C.

References libMesh::TimeSolver::_system, libMesh::AdaptiveTimeSolver::completed_timestep_size, libMesh::AdaptiveTimeSolver::core_time_solver, libMesh::UnsteadySolver::first_solve, libMesh::System::get_dof_map(), libMesh::DofMap::get_send_list(), libMesh::System::get_vector(), libMesh::NumericVector< T >::localize(), libMesh::UnsteadySolver::old_local_nonlinear_solution, libMesh::System::solution, and libMesh::System::time.

Referenced by solve().

{
  // The first access of advance_timestep happens via solve, not user code
  // It is used here to store any initial conditions data
  if (!first_solve)
    {
      _system.time += this->completed_timestep_size;
    }
  else
    {
    // We are here because of a call to advance_timestep that happens
    // via solve, the very first solve. All we are doing here is storing
    // the initial condition. The actual solution computed via this solve
    // will be stored when we call advance_timestep in the user's timestep loop
    first_solve = false;
    core_time_solver->set_first_solve(false);
    }

  // For the adaptive time solver, all SH operations
  // are handled by the core_time_solver's SH object
  // Sub solution storage is handled internally by the core time solver,
  // but the 'full step' solution is stored here to maintain consistency
  // with the fixed timestep scheme.
  core_time_solver->get_solution_history().store(false, _system.time);

  NumericVector<Number> & old_nonlinear_soln =
    _system.get_vector("_old_nonlinear_solution");
  NumericVector<Number> & nonlinear_solution =
    *(_system.solution);

  old_nonlinear_soln = nonlinear_solution;

  old_nonlinear_soln.localize
    (*old_local_nonlinear_solution,
     _system.get_dof_map().get_send_list());
}

before_timestep()

virtual void libMesh::TimeSolver::before_timestep ( )

inlinevirtualinherited

This method is for subclasses or users to override to do arbitrary processing between timesteps.

Definition at line 205 of file time_solver.h.

{}

calculate_norm()

Real libMesh::AdaptiveTimeSolver::calculate_norm ( System &  s, NumericVector< Number > &  v  )

protectedvirtualinherited

A helper function to calculate error norms.

Definition at line 225 of file adaptive_time_solver.C.

References libMesh::System::calculate_norm(), and libMesh::AdaptiveTimeSolver::component_norm.

Referenced by solve().

{
  return s.calculate_norm(v, component_norm);
}

compute_second_order_eqns()

bool libMesh::FirstOrderUnsteadySolver::compute_second_order_eqns ( bool  compute_jacobian, DiffContext &  c  )

protectedinherited

If there are second order variables, then we need to compute their residual equations and corresponding Jacobian.

The residual equation will simply be \( \dot{u} - v = 0 \), where \( u \) is the second order variable add by the user and \( v \) is the variable added by the time-solver as the "velocity" variable.

Definition at line 32 of file first_order_unsteady_solver.C.

References libMesh::TimeSolver::_system, compute_jacobian(), libMesh::DiffContext::get_dof_indices(), libMesh::DiffContext::get_elem_jacobian(), libMesh::DiffContext::get_elem_residual(), libMesh::DiffContext::get_elem_solution_derivative(), libMesh::DiffContext::get_elem_solution_rate_derivative(), libMesh::FEMContext::get_element_fe(), libMesh::FEMContext::get_element_qrule(), libMesh::DifferentiableSystem::get_second_order_dot_var(), libMesh::DiffContext::get_system(), libMesh::FEMContext::interior_rate(), libMesh::FEMContext::interior_value(), libMesh::DifferentiablePhysics::is_second_order_var(), libMesh::libmesh_assert(), libMesh::make_range(), libMesh::QBase::n_points(), libMesh::DiffContext::n_vars(), libMesh::Variable::type(), and libMesh::System::variable().

Referenced by libMesh::EulerSolver::_general_residual(), libMesh::Euler2Solver::_general_residual(), libMesh::EulerSolver::element_residual(), libMesh::Euler2Solver::element_residual(), libMesh::EulerSolver::nonlocal_residual(), and libMesh::Euler2Solver::nonlocal_residual().

{
  FEMContext & context = cast_ref<FEMContext &>(c);

  unsigned int n_qpoints = context.get_element_qrule().n_points();

  for (auto var : make_range(context.n_vars()))
    {
      if (!this->_system.is_second_order_var(var))
        continue;

      unsigned int dot_var = this->_system.get_second_order_dot_var(var);

      // We're assuming that the FE space for var and dot_var are the same
      libmesh_assert( context.get_system().variable(var).type() ==
                      context.get_system().variable(dot_var).type() );

      FEBase * elem_fe = nullptr;
      context.get_element_fe( var, elem_fe );

      const std::vector<Real> & JxW = elem_fe->get_JxW();

      const std::vector<std::vector<Real>> & phi = elem_fe->get_phi();

      const unsigned int n_dofs = cast_int<unsigned int>
        (context.get_dof_indices(dot_var).size());

      DenseSubVector<Number> & Fu = context.get_elem_residual(var);
      DenseSubMatrix<Number> & Kuu = context.get_elem_jacobian( var, var );
      DenseSubMatrix<Number> & Kuv = context.get_elem_jacobian( var, dot_var );

      for (unsigned int qp = 0; qp != n_qpoints; ++qp)
        {
          Number udot, v;
          context.interior_rate(var, qp, udot);
          context.interior_value(dot_var, qp, v);

          for (unsigned int i = 0; i < n_dofs; i++)
            {
              Fu(i) += JxW[qp]*(udot-v)*phi[i][qp];

              if (compute_jacobian)
                {
                  Number rate_factor = JxW[qp]*context.get_elem_solution_rate_derivative()*phi[i][qp];
                  Number soln_factor = JxW[qp]*context.get_elem_solution_derivative()*phi[i][qp];

                  Kuu(i,i) += rate_factor*phi[i][qp];
                  Kuv(i,i) -= soln_factor*phi[i][qp];

                  for (unsigned int j = i+1; j < n_dofs; j++)
                    {
                      Kuu(i,j) += rate_factor*phi[j][qp];
                      Kuu(j,i) += rate_factor*phi[j][qp];

                      Kuv(i,j) -= soln_factor*phi[j][qp];
                      Kuv(j,i) -= soln_factor*phi[j][qp];
                    }
                }
            }
        }
    }

  return compute_jacobian;
}

diff_solver()

std::unique_ptr< DiffSolver > & libMesh::AdaptiveTimeSolver::diff_solver ( )

overridevirtualinherited

An implicit linear or nonlinear solver to use at each timestep.

Reimplemented from libMesh::TimeSolver.

Definition at line 211 of file adaptive_time_solver.C.

References libMesh::AdaptiveTimeSolver::core_time_solver.

{
  return core_time_solver->diff_solver();
}

disable_print_counter_info()

void libMesh::ReferenceCounter::disable_print_counter_info ( )

staticinherited

Definition at line 100 of file reference_counter.C.

References libMesh::ReferenceCounter::_enable_print_counter.

{
  _enable_print_counter = false;
  return;
}

du()

Real libMesh::UnsteadySolver::du ( const SystemNorm &  norm ) const

overridevirtualinherited

Computes the size of ||u^{n+1} - u^{n}|| in some norm.

Note

While you can always call this function, its result may or may not be very meaningful. For example, if you call this function right after calling advance_timestep() then you'll get a result of zero since old_nonlinear_solution is set equal to nonlinear_solution in this function.

Implements libMesh::TimeSolver.

Definition at line 348 of file unsteady_solver.C.

References libMesh::TimeSolver::_system, libMesh::System::calculate_norm(), libMesh::System::get_vector(), libMesh::TensorTools::norm(), and libMesh::System::solution.

{

  std::unique_ptr<NumericVector<Number>> solution_copy =
    _system.solution->clone();

  solution_copy->add(-1., _system.get_vector("_old_nonlinear_solution"));

  solution_copy->close();

  return _system.calculate_norm(*solution_copy, norm);
}

element_residual()

bool libMesh::AdaptiveTimeSolver::element_residual ( bool  get_jacobian, DiffContext &  context  )

overridevirtualinherited

This method is passed on to the core_time_solver.

Implements libMesh::TimeSolver.

Definition at line 181 of file adaptive_time_solver.C.

References libMesh::AdaptiveTimeSolver::core_time_solver, and libMesh::libmesh_assert().

{
  libmesh_assert(core_time_solver.get());

  return core_time_solver->element_residual(request_jacobian, context);
}

enable_print_counter_info()

void libMesh::ReferenceCounter::enable_print_counter_info ( )

staticinherited

Methods to enable/disable the reference counter output from print_info()

Definition at line 94 of file reference_counter.C.

References libMesh::ReferenceCounter::_enable_print_counter.

{
  _enable_print_counter = true;
  return;
}

error_order()

Real libMesh::AdaptiveTimeSolver::error_order ( ) const

overridevirtualinherited

This method is passed on to the core_time_solver.

Implements libMesh::UnsteadySolver.

Definition at line 172 of file adaptive_time_solver.C.

References libMesh::AdaptiveTimeSolver::core_time_solver, and libMesh::libmesh_assert().

{
  libmesh_assert(core_time_solver.get());

  return core_time_solver->error_order();
}

get_info()

std::string libMesh::ReferenceCounter::get_info ( )

staticinherited

Gets a string containing the reference information.

Definition at line 47 of file reference_counter.C.

References libMesh::ReferenceCounter::_counts, and libMesh::Quality::name().

Referenced by libMesh::ReferenceCounter::print_info().

{
#if defined(LIBMESH_ENABLE_REFERENCE_COUNTING) && defined(DEBUG)

  std::ostringstream oss;

  oss << '\n'
      << " ---------------------------------------------------------------------------- \n"
      << "| Reference count information                                                |\n"
      << " ---------------------------------------------------------------------------- \n";

  for (const auto & [name, cd] : _counts)
    oss << "| " << name << " reference count information:\n"
        << "|  Creations:    " << cd.first    << '\n'
        << "|  Destructions: " << cd.second << '\n';

  oss << " ---------------------------------------------------------------------------- \n";

  return oss.str();

#else

  return "";

#endif
}

get_solution_history()

SolutionHistory & libMesh::TimeSolver::get_solution_history ( )

inherited

A getter function that returns a reference to the solution history object owned by TimeSolver.

Definition at line 124 of file time_solver.C.

References libMesh::TimeSolver::solution_history.

Referenced by libMesh::AdaptiveTimeSolver::init().

{
  return *solution_history;
}

increment_constructor_count()

void libMesh::ReferenceCounter::increment_constructor_count ( const std::string &  name )

inlineprotectednoexceptinherited

Increments the construction counter.

Should be called in the constructor of any derived class that will be reference counted.

Definition at line 183 of file reference_counter.h.

References libMesh::err, libMesh::BasicOStreamProxy< charT, traits >::get(), libMesh::Quality::name(), and libMesh::Threads::spin_mtx.

Referenced by libMesh::ReferenceCountedObject< RBParametrized >::ReferenceCountedObject().

{
  libmesh_try
  {
    Threads::spin_mutex::scoped_lock lock(Threads::spin_mtx);
    std::pair<unsigned int, unsigned int> & p = _counts[name];
    p.first++;
  }
  libmesh_catch (...)
  {
    auto stream = libMesh::err.get();
    stream->exceptions(stream->goodbit); // stream must not throw
    libMesh::err << "Encountered unrecoverable error while calling "
                 << "ReferenceCounter::increment_constructor_count() "
                 << "for a(n) " << name << " object." << std::endl;
    std::terminate();
  }
}

increment_destructor_count()

void libMesh::ReferenceCounter::increment_destructor_count ( const std::string &  name )

inlineprotectednoexceptinherited

Increments the destruction counter.

Should be called in the destructor of any derived class that will be reference counted.

Definition at line 207 of file reference_counter.h.

References libMesh::err, libMesh::BasicOStreamProxy< charT, traits >::get(), libMesh::Quality::name(), and libMesh::Threads::spin_mtx.

Referenced by libMesh::ReferenceCountedObject< RBParametrized >::~ReferenceCountedObject().

{
  libmesh_try
  {
    Threads::spin_mutex::scoped_lock lock(Threads::spin_mtx);
    std::pair<unsigned int, unsigned int> & p = _counts[name];
    p.second++;
  }
  libmesh_catch (...)
  {
    auto stream = libMesh::err.get();
    stream->exceptions(stream->goodbit); // stream must not throw
    libMesh::err << "Encountered unrecoverable error while calling "
                 << "ReferenceCounter::increment_destructor_count() "
                 << "for a(n) " << name << " object." << std::endl;
    std::terminate();
  }
}

init()

void libMesh::AdaptiveTimeSolver::init ( )

overridevirtualinherited

The initialization function.

This method is used to initialize internal data structures before a simulation begins.

Reimplemented from libMesh::UnsteadySolver.

Definition at line 54 of file adaptive_time_solver.C.

References libMesh::AdaptiveTimeSolver::core_time_solver, libMesh::TimeSolver::get_solution_history(), libMesh::libmesh_assert(), libMesh::UnsteadySolver::old_local_nonlinear_solution, and libMesh::TimeSolver::set_solution_history().

{
  libmesh_assert(core_time_solver.get());

  // We override this because our core_time_solver is the one that
  // needs to handle new vectors, diff_solver->init(), etc
  core_time_solver->init();

  // Set the core_time_solver's solution history object to be the same one as
  // that for the outer adaptive time solver
  core_time_solver->set_solution_history((this->get_solution_history()));

  // Now that we have set the SolutionHistory object for the coretimesolver,
  // we set the SolutionHistory type for the timesolver to be NoSolutionHistory
  // All storage and retrieval will be handled by the coretimesolver directly.
  NoSolutionHistory outersolver_solution_history;
  this->set_solution_history(outersolver_solution_history);

  // As an UnsteadySolver, we have an old_local_nonlinear_solution, but it
  // isn't pointing to the right place - fix it
  old_local_nonlinear_solution = core_time_solver->old_local_nonlinear_solution;
}

init_adjoints()

void libMesh::UnsteadySolver::init_adjoints ( )

overridevirtualinherited

Add adjoint vectors and old_adjoint_vectors as per the indices of QoISet.

Reimplemented from libMesh::TimeSolver.

Definition at line 68 of file unsteady_solver.C.

References libMesh::TimeSolver::_system, libMesh::System::add_vector(), libMesh::GHOSTED, libMesh::TimeSolver::init_adjoints(), libMesh::make_range(), and libMesh::System::n_qois().

{
  TimeSolver::init_adjoints();

  // Add old adjoint solutions
  // To keep the number of vectors consistent between the primal and adjoint
  // time loops, we will also add the adjoint rhs vector during initialization
  for(auto i : make_range(_system.n_qois()))
  {
    std::string old_adjoint_solution_name = "_old_adjoint_solution";
    old_adjoint_solution_name+= std::to_string(i);
    _system.add_vector(old_adjoint_solution_name, false, GHOSTED);

    std::string adjoint_rhs_name = "adjoint_rhs";
    adjoint_rhs_name+= std::to_string(i);
    _system.add_vector(adjoint_rhs_name, false, GHOSTED);
  }

}

init_data()

void libMesh::UnsteadySolver::init_data ( )

overridevirtualinherited

The data initialization function.

This method is used to initialize internal data structures after the underlying System has been initialized

Reimplemented from libMesh::TimeSolver.

Reimplemented in libMesh::SecondOrderUnsteadySolver.

Definition at line 89 of file unsteady_solver.C.

References libMesh::TimeSolver::_system, libMesh::System::get_dof_map(), libMesh::DofMap::get_send_list(), libMesh::GHOSTED, libMesh::TimeSolver::init_data(), libMesh::System::n_dofs(), libMesh::System::n_local_dofs(), libMesh::UnsteadySolver::old_local_nonlinear_solution, and libMesh::SERIAL.

Referenced by libMesh::SecondOrderUnsteadySolver::init_data().

{
  TimeSolver::init_data();

#ifdef LIBMESH_ENABLE_GHOSTED
  old_local_nonlinear_solution->init (_system.n_dofs(), _system.n_local_dofs(),
                                      _system.get_dof_map().get_send_list(), false,
                                      GHOSTED);
#else
  old_local_nonlinear_solution->init (_system.n_dofs(), false, SERIAL);
#endif
}

integrate_adjoint_refinement_error_estimate()

void libMesh::TwostepTimeSolver::integrate_adjoint_refinement_error_estimate ( AdjointRefinementEstimator &  adjoint_refinement_error_estimator, ErrorVector &  QoI_elementwise_error  )

overridevirtual

A method to compute the adjoint refinement error estimate at the current timestep.

int_{tstep_start}^{tstep_end} R(u^h,z) dt The user provides an initialized ARefEE object. Fills in an ErrorVector that contains the weighted sum of errors from all the QoIs and can be used to guide AMR. CURRENTLY ONLY SUPPORTED for Backward Euler.

Implements libMesh::AdaptiveTimeSolver.

Definition at line 400 of file twostep_time_solver.C.

References libMesh::TimeSolver::_system, libMesh::AdaptiveTimeSolver::core_time_solver, libMesh::System::get_qoi_error_estimate_value(), libMesh::QoISet::has_index(), libMesh::index_range(), libMesh::make_range(), libMesh::System::n_qois(), libMesh::AdjointRefinementEstimator::qoi_set(), and libMesh::System::set_qoi_error_estimate().

{
  // We use a numerical integration scheme consistent with the theta used for the timesolver.

  // Create first and second half error estimate vectors of the right size
  std::vector<Number> qoi_error_estimates_first_half(_system.n_qois());
  std::vector<Number> qoi_error_estimates_second_half(_system.n_qois());

  // First half timestep
  ErrorVector QoI_elementwise_error_first_half;

  core_time_solver->integrate_adjoint_refinement_error_estimate(adjoint_refinement_error_estimator, QoI_elementwise_error_first_half);

  // Also get the first 'half step' spatially integrated errors for all the QoIs in the QoI set
  for (auto j : make_range(_system.n_qois()))
  {
    // Skip this QoI if not in the QoI Set
    if (adjoint_refinement_error_estimator.qoi_set().has_index(j))
    {
      qoi_error_estimates_first_half[j] = _system.get_qoi_error_estimate_value(j);
    }
  }

  // Second half timestep
  ErrorVector QoI_elementwise_error_second_half;

  core_time_solver->integrate_adjoint_refinement_error_estimate(adjoint_refinement_error_estimator, QoI_elementwise_error_second_half);

  // Also get the second 'half step' spatially integrated errors for all the QoIs in the QoI set
  for (auto j : make_range(_system.n_qois()))
  {
    // Skip this QoI if not in the QoI Set
    if (adjoint_refinement_error_estimator.qoi_set().has_index(j))
    {
      qoi_error_estimates_second_half[j] = _system.get_qoi_error_estimate_value(j);
    }
  }

  // Error contribution from this timestep
  for (auto i : index_range(QoI_elementwise_error))
    QoI_elementwise_error[i] = QoI_elementwise_error_first_half[i] + QoI_elementwise_error_second_half[i];

  for (auto j : make_range(_system.n_qois()))
  {
    // Skip this QoI if not in the QoI Set
    if (adjoint_refinement_error_estimator.qoi_set().has_index(j))
    {
      _system.set_qoi_error_estimate(j, qoi_error_estimates_first_half[j] + qoi_error_estimates_second_half[j]);
    }
  }
}

integrate_adjoint_sensitivity()

void libMesh::TwostepTimeSolver::integrate_adjoint_sensitivity ( const QoISet &  qois, const ParameterVector &  parameter_vector, SensitivityData &  sensitivities  )

overridevirtual

A method to integrate the adjoint sensitivity w.r.t a given parameter vector.

int_{tstep_start}^{tstep_end} dQ/dp dt = int_{tstep_start}^{tstep_end} ( / p) - ( R (u,z) / p ) dt The midpoint rule is used to integrate each substep

Implements libMesh::AdaptiveTimeSolver.

Definition at line 377 of file twostep_time_solver.C.

References libMesh::TimeSolver::_system, libMesh::AdaptiveTimeSolver::core_time_solver, libMesh::make_range(), libMesh::ParameterVector::size(), and libMesh::QoISet::size().

{
  // We are using the trapezoidal rule to integrate each timestep, and then pooling the contributions here.
  // (f(t_j) + f(t_j+1/2))/2 (t_j+1/2 - t_j) + (f(t_j+1/2) + f(t_j+1))/2 (t_j+1 - t_j+1/2)

  // First half timestep
  SensitivityData sensitivities_first_half(qois, _system, parameter_vector);

  core_time_solver->integrate_adjoint_sensitivity(qois, parameter_vector, sensitivities_first_half);

  // Second half timestep
  SensitivityData sensitivities_second_half(qois, _system, parameter_vector);

  core_time_solver->integrate_adjoint_sensitivity(qois, parameter_vector, sensitivities_second_half);

  // Get the contributions for each sensitivity from this timestep
  const auto pv_size = parameter_vector.size();
  for (auto i : make_range(qois.size(_system)))
    for (auto j : make_range(pv_size))
      sensitivities[i][j] = sensitivities_first_half[i][j] + sensitivities_second_half[i][j];
}

integrate_qoi_timestep()

void libMesh::TwostepTimeSolver::integrate_qoi_timestep ( )

overridevirtual

A method to integrate the system::QoI functionals.

Implements libMesh::AdaptiveTimeSolver.

Definition at line 341 of file twostep_time_solver.C.

References libMesh::TimeSolver::_system, libMesh::AdaptiveTimeSolver::core_time_solver, libMesh::System::get_qoi_value(), libMesh::make_range(), libMesh::System::n_qois(), and libMesh::System::set_qoi().

{
  // Vectors to hold qoi contributions from the first and second half timesteps
  std::vector<Number> qois_first_half(_system.n_qois(), 0.0);
  std::vector<Number> qois_second_half(_system.n_qois(), 0.0);

  // First half contribution
  core_time_solver->integrate_qoi_timestep();

  for (auto j : make_range(_system.n_qois()))
  {
    qois_first_half[j] = _system.get_qoi_value(j);
  }

  // Second half contribution
  core_time_solver->integrate_qoi_timestep();

  for (auto j : make_range(_system.n_qois()))
  {
    qois_second_half[j] = _system.get_qoi_value(j);
  }

  // Zero out the system.qoi vector
  for (auto j : make_range(_system.n_qois()))
  {
    _system.set_qoi(j, 0.0);
  }

  // Add the contributions from the two halftimesteps to get the full QoI
  // contribution from this timestep
  for (auto j : make_range(_system.n_qois()))
  {
    _system.set_qoi(j, qois_first_half[j] + qois_second_half[j]);
  }
}

is_adjoint()

bool libMesh::TimeSolver::is_adjoint ( ) const

inlineinherited

Accessor for querying whether we need to do a primal or adjoint solve.

Definition at line 277 of file time_solver.h.

References libMesh::TimeSolver::_is_adjoint.

Referenced by libMesh::FEMSystem::build_context().

  { return _is_adjoint; }

is_steady()

virtual bool libMesh::UnsteadySolver::is_steady ( ) const

inlineoverridevirtualinherited

This is not a steady-state solver.

Implements libMesh::TimeSolver.

Definition at line 194 of file unsteady_solver.h.

{ return false; }

last_completed_timestep_size()

virtual Real libMesh::AdaptiveTimeSolver::last_completed_timestep_size ( )

inlineoverridevirtualinherited

Returns system.deltat if fixed timestep solver is used, the complete timestep size (sum of all substeps) if the adaptive time solver is used.

Returns the change in system.time, deltat, for the last timestep which was successfully completed. This only returns the outermost step size in the case of nested time solvers. If no time step has yet been successfully completed, then returns system.deltat.

Reimplemented from libMesh::TimeSolver.

Definition at line 91 of file adaptive_time_solver.h.

References libMesh::AdaptiveTimeSolver::completed_timestep_size.

{ return completed_timestep_size; };

linear_solver()

std::unique_ptr< LinearSolver< Number > > & libMesh::AdaptiveTimeSolver::linear_solver ( )

overridevirtualinherited

An implicit linear solver to use for adjoint and sensitivity problems.

Reimplemented from libMesh::TimeSolver.

Definition at line 218 of file adaptive_time_solver.C.

References libMesh::AdaptiveTimeSolver::core_time_solver.

{
  return core_time_solver->linear_solver();
}

n_objects()

static unsigned int libMesh::ReferenceCounter::n_objects ( )

inlinestaticinherited

Prints the number of outstanding (created, but not yet destroyed) objects.

Definition at line 85 of file reference_counter.h.

References libMesh::ReferenceCounter::_n_objects.

Referenced by libMesh::LibMeshInit::~LibMeshInit().

  { return _n_objects; }

nonlocal_residual()

bool libMesh::AdaptiveTimeSolver::nonlocal_residual ( bool  get_jacobian, DiffContext &  context  )

overridevirtualinherited

This method is passed on to the core_time_solver.

Implements libMesh::TimeSolver.

Definition at line 201 of file adaptive_time_solver.C.

References libMesh::AdaptiveTimeSolver::core_time_solver, and libMesh::libmesh_assert().

{
  libmesh_assert(core_time_solver.get());

  return core_time_solver->nonlocal_residual(request_jacobian, context);
}

old_nonlinear_solution()

Number libMesh::UnsteadySolver::old_nonlinear_solution ( const dof_id_type  global_dof_number ) const

inherited

Returns

The old nonlinear solution for the specified global DOF.

Definition at line 337 of file unsteady_solver.C.

References libMesh::TimeSolver::_system, libMesh::System::get_dof_map(), libMesh::DofMap::n_dofs(), and libMesh::UnsteadySolver::old_local_nonlinear_solution.

Referenced by libMesh::EulerSolver::_general_residual(), libMesh::Euler2Solver::_general_residual(), libMesh::NewmarkSolver::_general_residual(), and libMesh::FEMPhysics::eulerian_residual().

{
  libmesh_assert_less (global_dof_number, _system.get_dof_map().n_dofs());
  libmesh_assert_less (global_dof_number, old_local_nonlinear_solution->size());

  return (*old_local_nonlinear_solution)(global_dof_number);
}

prepare_accel()

void libMesh::FirstOrderUnsteadySolver::prepare_accel ( DiffContext &  context )

protectedinherited

If there are second order variables in the system, then we also prepare the accel for those variables so the user can treat them as such.

Definition at line 25 of file first_order_unsteady_solver.C.

References libMesh::DiffContext::elem_solution_accel_derivative, libMesh::DiffContext::get_elem_solution_accel(), libMesh::DiffContext::get_elem_solution_rate(), and libMesh::DiffContext::get_elem_solution_rate_derivative().

Referenced by libMesh::EulerSolver::_general_residual(), and libMesh::Euler2Solver::_general_residual().

{
  context.get_elem_solution_accel() = context.get_elem_solution_rate();

  context.elem_solution_accel_derivative = context.get_elem_solution_rate_derivative();
}

print_info()

void libMesh::ReferenceCounter::print_info ( std::ostream &  out_stream = libMesh::out )

staticinherited

Prints the reference information, by default to libMesh::out.

Definition at line 81 of file reference_counter.C.

References libMesh::ReferenceCounter::_enable_print_counter, and libMesh::ReferenceCounter::get_info().

Referenced by libMesh::LibMeshInit::~LibMeshInit().

{
  if (_enable_print_counter)
    out_stream << ReferenceCounter::get_info();
}

reinit()

void libMesh::AdaptiveTimeSolver::reinit ( )

overridevirtualinherited

The reinitialization function.

This method is used to resize internal data vectors after a mesh change.

Reimplemented from libMesh::UnsteadySolver.

Definition at line 79 of file adaptive_time_solver.C.

References libMesh::AdaptiveTimeSolver::core_time_solver, and libMesh::libmesh_assert().

{
  libmesh_assert(core_time_solver.get());

  // We override this because our core_time_solver is the one that
  // needs to handle new vectors, diff_solver->reinit(), etc
  core_time_solver->reinit();
}

retrieve_timestep()

void libMesh::AdaptiveTimeSolver::retrieve_timestep ( )

overridevirtualinherited

This method retrieves all the stored solutions at the current system.time.

Reimplemented from libMesh::UnsteadySolver.

Definition at line 165 of file adaptive_time_solver.C.

References libMesh::AdaptiveTimeSolver::core_time_solver.

{
  // Ask the core time solver to retrieve all the stored vectors
  // at the current time
  core_time_solver->retrieve_timestep();
}

set_first_adjoint_step()

void libMesh::UnsteadySolver::set_first_adjoint_step ( bool  first_adjoint_step_setting )

inlineinherited

A setter for the first_adjoint_step boolean.

Needed for nested time solvers.

Definition at line 199 of file unsteady_solver.h.

References libMesh::UnsteadySolver::first_adjoint_step.

  {
    first_adjoint_step = first_adjoint_step_setting;
  }

set_first_solve()

void libMesh::UnsteadySolver::set_first_solve ( bool  first_solve_setting )

inlineinherited

Definition at line 209 of file unsteady_solver.h.

References libMesh::UnsteadySolver::first_solve.

  {
    first_solve = first_solve_setting;
  }

set_is_adjoint()

void libMesh::TimeSolver::set_is_adjoint ( bool  _is_adjoint_value )

inlineinherited

Accessor for setting whether we need to do a primal or adjoint solve.

Definition at line 284 of file time_solver.h.

References libMesh::TimeSolver::_is_adjoint.

Referenced by libMesh::DifferentiableSystem::adjoint_solve(), libMesh::FEMSystem::postprocess(), and libMesh::DifferentiableSystem::solve().

  { _is_adjoint = _is_adjoint_value; }

set_solution_history()

void libMesh::TimeSolver::set_solution_history ( const SolutionHistory &  _solution_history )

inherited

A setter function users will employ if they need to do something other than save no solution history.

Definition at line 119 of file time_solver.C.

References libMesh::SolutionHistory::clone(), and libMesh::TimeSolver::solution_history.

Referenced by libMesh::AdaptiveTimeSolver::init().

{
  solution_history = _solution_history.clone();
}

side_residual()

bool libMesh::AdaptiveTimeSolver::side_residual ( bool  get_jacobian, DiffContext &  context  )

overridevirtualinherited

This method is passed on to the core_time_solver.

Implements libMesh::TimeSolver.

Definition at line 191 of file adaptive_time_solver.C.

References libMesh::AdaptiveTimeSolver::core_time_solver, and libMesh::libmesh_assert().

{
  libmesh_assert(core_time_solver.get());

  return core_time_solver->side_residual(request_jacobian, context);
}

solve()

void libMesh::TwostepTimeSolver::solve ( )

overridevirtual

This method solves for the solution at the next timestep.

Usually we will only need to solve one (non)linear system per timestep, but more complex subclasses may override this.

Implements libMesh::AdaptiveTimeSolver.

Definition at line 54 of file twostep_time_solver.C.

References libMesh::TimeSolver::_system, libMesh::AdaptiveTimeSolver::advance_timestep(), libMesh::AdaptiveTimeSolver::calculate_norm(), libMesh::NumericVector< T >::clone(), libMesh::AdaptiveTimeSolver::completed_timestep_size, libMesh::AdaptiveTimeSolver::core_time_solver, libMesh::DifferentiableSystem::deltat, libMesh::UnsteadySolver::first_solve, libMesh::System::get_vector(), libMesh::AdaptiveTimeSolver::global_tolerance, libMesh::TimeSolver::last_deltat, libMesh::libmesh_assert(), libMesh::AdaptiveTimeSolver::max_deltat, libMesh::AdaptiveTimeSolver::max_growth, libMesh::AdaptiveTimeSolver::min_deltat, libMesh::out, libMesh::Utility::pow(), libMesh::TimeSolver::quiet, libMesh::Real, libMesh::TimeSolver::reduce_deltat_on_diffsolver_failure, libMesh::System::solution, libMesh::AdaptiveTimeSolver::target_tolerance, libMesh::System::time, and libMesh::AdaptiveTimeSolver::upper_tolerance.

{
  libmesh_assert(core_time_solver.get());

  // All actual solution history operations are handled by the outer
  // solver, so the outer solver has to call advance_timestep and
  // call solution_history->store to store the initial conditions
  if (first_solve)
    {
      advance_timestep();
      first_solve = false;
    }

  // We may have to repeat timesteps entirely if our error is bad
  // enough
  bool max_tolerance_met = false;

  // Calculating error values each time
  Real single_norm(0.), double_norm(0.), error_norm(0.),
    relative_error(0.);

  // The loop below will change system time and deltat based on calculations.
  // We will need to save these for calculating the deltat for the next timestep
  // after the while loop has converged.
  Real old_time;
  Real old_deltat;

  while (!max_tolerance_met)
    {
      // If we've been asked to reduce deltat if necessary, make sure
      // the core timesolver does so
      core_time_solver->reduce_deltat_on_diffsolver_failure =
        this->reduce_deltat_on_diffsolver_failure;

      if (!quiet)
        {
          libMesh::out << "\n === Computing adaptive timestep === "
                       << std::endl;
        }

      // Use the double-length timestep first (so the
      // old_nonlinear_solution won't have to change)
      core_time_solver->solve();

      // Save a copy of the double-length nonlinear solution
      // and the old nonlinear solution
      std::unique_ptr<NumericVector<Number>> double_solution =
        _system.solution->clone();
      std::unique_ptr<NumericVector<Number>> old_solution =
        _system.get_vector("_old_nonlinear_solution").clone();

      double_norm = calculate_norm(_system, *double_solution);
      if (!quiet)
        {
          libMesh::out << "Double norm = " << double_norm << std::endl;
        }

      // Then reset the initial guess for our single-length calcs
      *(_system.solution) = _system.get_vector("_old_nonlinear_solution");

      // Call two single-length timesteps
      // Be sure that the core_time_solver does not change the
      // timestep here.  (This is unlikely because it just succeeded
      // with a timestep twice as large!)
      // FIXME: even if diffsolver failure is unlikely, we ought to
      // do *something* if it happens
      core_time_solver->reduce_deltat_on_diffsolver_failure = 0;

      old_time = _system.time;
      old_deltat = _system.deltat;
      _system.deltat *= 0.5;

      // Attempt the 'half timestep solve'
      core_time_solver->solve();

      // If we successfully completed the solve, let the time solver know the deltat used
      this->last_deltat = _system.deltat;

      // Increment system.time, and save the half solution to solution history
      core_time_solver->advance_timestep();

      core_time_solver->solve();

      single_norm = calculate_norm(_system, *_system.solution);
      if (!quiet)
        {
          libMesh::out << "Single norm = " << single_norm << std::endl;
        }

      // Reset the core_time_solver's reduce_deltat... value.
      core_time_solver->reduce_deltat_on_diffsolver_failure =
        this->reduce_deltat_on_diffsolver_failure;

      // Find the relative error
      *double_solution -= *(_system.solution);
      error_norm  = calculate_norm(_system, *double_solution);
      relative_error = error_norm / old_deltat /
        std::max(double_norm, single_norm);

      // If the relative error makes no sense, we're done
      if (!double_norm && !single_norm)
        return;

      if (!quiet)
        {
          libMesh::out << "Error norm = " << error_norm << std::endl;
          libMesh::out << "Local relative error = "
                       << (error_norm /
                           std::max(double_norm, single_norm))
                       << std::endl;
          libMesh::out << "Global relative error = "
                       << (error_norm / old_deltat /
                           std::max(double_norm, single_norm))
                       << std::endl;
          libMesh::out << "old delta t = " << old_deltat << std::endl;
        }

      // If our upper tolerance is negative, that means we want to set
      // it based on the first successful time step
      if (this->upper_tolerance < 0)
        this->upper_tolerance = -this->upper_tolerance * relative_error;

      // If we haven't met our upper error tolerance, we'll have to
      // repeat this timestep entirely
      if (this->upper_tolerance && relative_error > this->upper_tolerance)
        {
          // If we are saving solution histories, the core time solver
          // will save half solutions, and these solves can be attempted
          // repeatedly. If we failed to meet the tolerance, erase the
          // half solution from solution history.
          core_time_solver->get_solution_history().erase(_system.time);

          // We will be retrying this timestep with deltat/2, so restore
          // all the necessary state.
          // FIXME: this probably doesn't work with multistep methods
          _system.get_vector("_old_nonlinear_solution") = *old_solution;
          _system.time = old_time;
          _system.deltat = old_deltat;

          // Update to localize the old nonlinear solution
          core_time_solver->update();

          // Reset the initial guess for our next try
          *(_system.solution) =
            _system.get_vector("_old_nonlinear_solution");

          // Chop delta t in half
          _system.deltat /= 2.;

          if (!quiet)
            {
              libMesh::out << "Failed to meet upper error tolerance"
                           << std::endl;
              libMesh::out << "Retrying with delta t = "
                           << _system.deltat << std::endl;
            }
        }
      else
        max_tolerance_met = true;

    }

  // We ended up taking two half steps of size system.deltat to
  // march our last time step.
  this->last_deltat = _system.deltat;
  this->completed_timestep_size = 2.0*_system.deltat;

  // TimeSolver::solve methods should leave system.time unchanged
  _system.time = old_time;

  // Compare the relative error to the tolerance and adjust deltat
  _system.deltat = old_deltat;

  // If our target tolerance is negative, that means we want to set
  // it based on the first successful time step
  if (this->target_tolerance < 0)
    this->target_tolerance = -this->target_tolerance * relative_error;

  const Real global_shrink_or_growth_factor =
    std::pow(this->target_tolerance / relative_error,
             static_cast<Real>(1. / core_time_solver->error_order()));

  const Real local_shrink_or_growth_factor =
    std::pow(this->target_tolerance /
             (error_norm/std::max(double_norm, single_norm)),
             static_cast<Real>(1. / (core_time_solver->error_order()+1.)));

  if (!quiet)
    {
      libMesh::out << "The global growth/shrink factor is: "
                   << global_shrink_or_growth_factor << std::endl;
      libMesh::out << "The local growth/shrink factor is: "
                   << local_shrink_or_growth_factor << std::endl;
    }

  // The local s.o.g. factor is based on the expected **local**
  // truncation error for the timestepping method, the global
  // s.o.g. factor is based on the method's **global** truncation
  // error.  You can shrink/grow the timestep to attempt to satisfy
  // either a global or local time-discretization error tolerance.

  Real shrink_or_growth_factor =
    this->global_tolerance ? global_shrink_or_growth_factor :
    local_shrink_or_growth_factor;

  if (this->max_growth && this->max_growth < shrink_or_growth_factor)
    {
      if (!quiet && this->global_tolerance)
        {
          libMesh::out << "delta t is constrained by max_growth" << std::endl;
        }
      shrink_or_growth_factor = this->max_growth;
    }

  _system.deltat *= shrink_or_growth_factor;

  // Restrict deltat to max-allowable value if necessary
  if ((this->max_deltat != 0.0) && (_system.deltat > this->max_deltat))
    {
      if (!quiet)
        {
          libMesh::out << "delta t is constrained by maximum-allowable delta t."
                       << std::endl;
        }
      _system.deltat = this->max_deltat;
    }

  // Restrict deltat to min-allowable value if necessary
  if ((this->min_deltat != 0.0) && (_system.deltat < this->min_deltat))
    {
      if (!quiet)
        {
          libMesh::out << "delta t is constrained by minimum-allowable delta t."
                       << std::endl;
        }
      _system.deltat = this->min_deltat;
    }

  if (!quiet)
    {
      libMesh::out << "new delta t = " << _system.deltat << std::endl;
    }
}

system() [1/2]

const sys_type& libMesh::TimeSolver::system ( ) const

inlineinherited

Returns

A constant reference to the system we are solving.

Definition at line 210 of file time_solver.h.

References libMesh::TimeSolver::_system.

Referenced by libMesh::TimeSolver::adjoint_solve(), libMesh::TimeSolver::reinit(), and libMesh::TimeSolver::solve().

{ return _system; }

system() [2/2]

sys_type& libMesh::TimeSolver::system ( )

inlineinherited

Returns

A writable reference to the system we are solving.

Definition at line 215 of file time_solver.h.

References libMesh::TimeSolver::_system.

{ return _system; }

time_order()

virtual unsigned int libMesh::FirstOrderUnsteadySolver::time_order ( ) const

inlineoverridevirtualinherited

Returns

The maximum order of time derivatives for which the UnsteadySolver subclass is capable of handling.

For example, EulerSolver will have time_order() = 1 and NewmarkSolver will have time_order() = 2.

Implements libMesh::UnsteadySolver.

Definition at line 90 of file first_order_unsteady_solver.h.

  { return 1; }

update()

void libMesh::UnsteadySolver::update ( )

inherited

Definition at line 361 of file unsteady_solver.C.

References libMesh::TimeSolver::_system, libMesh::System::get_dof_map(), libMesh::DofMap::get_send_list(), libMesh::System::get_vector(), libMesh::NumericVector< T >::localize(), and libMesh::UnsteadySolver::old_local_nonlinear_solution.

{
  // Dont forget to localize the old_nonlinear_solution !
  _system.get_vector("_old_nonlinear_solution").localize
    (*old_local_nonlinear_solution,
     _system.get_dof_map().get_send_list());
}

Member Data Documentation

_counts

ReferenceCounter::Counts libMesh::ReferenceCounter::_counts

staticprotectedinherited

Actually holds the data.

Definition at line 124 of file reference_counter.h.

Referenced by libMesh::ReferenceCounter::get_info().

_diff_solver

std::unique_ptr<DiffSolver> libMesh::TimeSolver::_diff_solver

protectedinherited

An implicit linear or nonlinear solver to use at each timestep.

Definition at line 302 of file time_solver.h.

Referenced by libMesh::NewmarkSolver::compute_initial_accel(), libMesh::TimeSolver::diff_solver(), and libMesh::UnsteadySolver::solve().

_enable_print_counter

bool libMesh::ReferenceCounter::_enable_print_counter = true

staticprotectedinherited

Flag to control whether reference count information is printed when print_info is called.

Definition at line 143 of file reference_counter.h.

Referenced by libMesh::ReferenceCounter::disable_print_counter_info(), libMesh::ReferenceCounter::enable_print_counter_info(), and libMesh::ReferenceCounter::print_info().

_linear_solver

std::unique_ptr<LinearSolver<Number> > libMesh::TimeSolver::_linear_solver

protectedinherited

An implicit linear solver to use for adjoint problems.

Definition at line 307 of file time_solver.h.

Referenced by libMesh::TimeSolver::linear_solver(), and libMesh::TimeSolver::reinit().

_mutex

Threads::spin_mutex libMesh::ReferenceCounter::_mutex

staticprotectedinherited

Mutual exclusion object to enable thread-safe reference counting.

Definition at line 137 of file reference_counter.h.

_n_objects

Threads::atomic< unsigned int > libMesh::ReferenceCounter::_n_objects

staticprotectedinherited

The number of objects.

Print the reference count information when the number returns to 0.

Definition at line 132 of file reference_counter.h.

Referenced by libMesh::ReferenceCounter::n_objects(), libMesh::ReferenceCounter::ReferenceCounter(), and libMesh::ReferenceCounter::~ReferenceCounter().

_system

sys_type& libMesh::TimeSolver::_system

protectedinherited

A reference to the system we are solving.

Definition at line 312 of file time_solver.h.

Referenced by libMesh::EulerSolver::_general_residual(), libMesh::Euler2Solver::_general_residual(), libMesh::SteadySolver::_general_residual(), libMesh::NewmarkSolver::_general_residual(), libMesh::AdaptiveTimeSolver::adjoint_advance_timestep(), libMesh::UnsteadySolver::adjoint_advance_timestep(), adjoint_solve(), libMesh::UnsteadySolver::adjoint_solve(), libMesh::TimeSolver::adjoint_solve(), libMesh::NewmarkSolver::advance_timestep(), libMesh::AdaptiveTimeSolver::advance_timestep(), libMesh::UnsteadySolver::advance_timestep(), libMesh::NewmarkSolver::compute_initial_accel(), libMesh::FirstOrderUnsteadySolver::compute_second_order_eqns(), libMesh::UnsteadySolver::du(), libMesh::EulerSolver::element_residual(), libMesh::Euler2Solver::element_residual(), libMesh::EigenTimeSolver::element_residual(), libMesh::SecondOrderUnsteadySolver::init(), libMesh::UnsteadySolver::init(), libMesh::TimeSolver::init(), libMesh::EigenTimeSolver::init(), libMesh::UnsteadySolver::init_adjoints(), libMesh::TimeSolver::init_adjoints(), libMesh::SecondOrderUnsteadySolver::init_data(), libMesh::UnsteadySolver::init_data(), libMesh::TimeSolver::init_data(), libMesh::Euler2Solver::integrate_adjoint_refinement_error_estimate(), integrate_adjoint_refinement_error_estimate(), libMesh::EulerSolver::integrate_adjoint_refinement_error_estimate(), integrate_adjoint_sensitivity(), libMesh::SteadySolver::integrate_adjoint_sensitivity(), libMesh::UnsteadySolver::integrate_adjoint_sensitivity(), libMesh::Euler2Solver::integrate_qoi_timestep(), integrate_qoi_timestep(), libMesh::EulerSolver::integrate_qoi_timestep(), libMesh::SteadySolver::integrate_qoi_timestep(), libMesh::EulerSolver::nonlocal_residual(), libMesh::Euler2Solver::nonlocal_residual(), libMesh::EigenTimeSolver::nonlocal_residual(), libMesh::UnsteadySolver::old_nonlinear_solution(), libMesh::SecondOrderUnsteadySolver::old_solution_accel(), libMesh::SecondOrderUnsteadySolver::old_solution_rate(), libMesh::NewmarkSolver::project_initial_accel(), libMesh::SecondOrderUnsteadySolver::project_initial_rate(), libMesh::SecondOrderUnsteadySolver::reinit(), libMesh::UnsteadySolver::reinit(), libMesh::TimeSolver::reinit(), libMesh::UnsteadySolver::retrieve_timestep(), libMesh::EigenTimeSolver::side_residual(), solve(), libMesh::UnsteadySolver::solve(), libMesh::EigenTimeSolver::solve(), libMesh::TimeSolver::system(), and libMesh::UnsteadySolver::update().

completed_timestep_size

Real libMesh::AdaptiveTimeSolver::completed_timestep_size

inherited

The adaptive time solver's have two notions of deltat.

The deltat the solver ended up using for the completed timestep. And the deltat the solver determined would be workable for the coming timestep. The latter gets set as system.deltat. We need a variable to save the deltat used for the completed timestep.

Definition at line 205 of file adaptive_time_solver.h.

Referenced by adjoint_solve(), libMesh::AdaptiveTimeSolver::advance_timestep(), libMesh::AdaptiveTimeSolver::last_completed_timestep_size(), and solve().

component_norm

SystemNorm libMesh::AdaptiveTimeSolver::component_norm

inherited

Error calculations are done in this norm, DISCRETE_L2 by default.

Definition at line 135 of file adaptive_time_solver.h.

Referenced by libMesh::AdaptiveTimeSolver::calculate_norm().

component_scale

std::vector<float> libMesh::AdaptiveTimeSolver::component_scale

inherited

If component_norms is non-empty, each variable's contribution to the error of a system will also be scaled by component_scale[var], unless component_scale is empty in which case all variables will be weighted equally.

Definition at line 143 of file adaptive_time_solver.h.

core_time_solver

std::unique_ptr<UnsteadySolver> libMesh::AdaptiveTimeSolver::core_time_solver

inherited

This object is used to take timesteps.

Definition at line 130 of file adaptive_time_solver.h.

Referenced by libMesh::AdaptiveTimeSolver::adjoint_advance_timestep(), adjoint_solve(), libMesh::AdaptiveTimeSolver::advance_timestep(), libMesh::AdaptiveTimeSolver::diff_solver(), libMesh::AdaptiveTimeSolver::element_residual(), libMesh::AdaptiveTimeSolver::error_order(), libMesh::AdaptiveTimeSolver::init(), integrate_adjoint_refinement_error_estimate(), integrate_adjoint_sensitivity(), integrate_qoi_timestep(), libMesh::AdaptiveTimeSolver::linear_solver(), libMesh::AdaptiveTimeSolver::nonlocal_residual(), libMesh::AdaptiveTimeSolver::reinit(), libMesh::AdaptiveTimeSolver::retrieve_timestep(), libMesh::AdaptiveTimeSolver::side_residual(), solve(), and TwostepTimeSolver().

first_adjoint_step

bool libMesh::UnsteadySolver::first_adjoint_step

protectedinherited

A bool that will be true the first time adjoint_advance_timestep() is called, (when the primal solution is to be used to set adjoint boundary conditions) and false thereafter.

Definition at line 226 of file unsteady_solver.h.

Referenced by libMesh::AdaptiveTimeSolver::adjoint_advance_timestep(), and libMesh::UnsteadySolver::set_first_adjoint_step().

first_solve

bool libMesh::UnsteadySolver::first_solve

protectedinherited

A bool that will be true the first time solve() is called, and false thereafter.

Definition at line 220 of file unsteady_solver.h.

Referenced by libMesh::NewmarkSolver::advance_timestep(), libMesh::AdaptiveTimeSolver::advance_timestep(), libMesh::UnsteadySolver::advance_timestep(), libMesh::UnsteadySolver::set_first_solve(), solve(), and libMesh::UnsteadySolver::solve().

global_tolerance

bool libMesh::AdaptiveTimeSolver::global_tolerance

inherited

This flag, which is true by default, grows (shrinks) the timestep based on the expected global accuracy of the timestepping scheme.

Global in this sense means the cumulative final-time accuracy of the scheme. For example, the backward Euler scheme's truncation error is locally of order 2, so that after N timesteps of size deltat, the result is first-order accurate. If you set this to false, you can grow (shrink) your timestep based on the local accuracy rather than the global accuracy of the core TimeSolver.

Note

By setting this value to false you may fail to achieve the predicted convergence in time of the underlying method, however it may be possible to get more fine-grained control over step sizes as well.

Definition at line 222 of file adaptive_time_solver.h.

Referenced by solve().

last_deltat

Real libMesh::TimeSolver::last_deltat

protectedinherited

The deltat for the last completed timestep before the current one.

Definition at line 332 of file time_solver.h.

Referenced by adjoint_solve(), libMesh::UnsteadySolver::adjoint_solve(), libMesh::TimeSolver::last_completed_timestep_size(), solve(), and libMesh::UnsteadySolver::solve().

last_step_deltat

Real libMesh::UnsteadySolver::last_step_deltat

protectedinherited

We will need to move the system.time around to ensure that residuals are built with the right deltat and the right time.

Definition at line 237 of file unsteady_solver.h.

Referenced by libMesh::Euler2Solver::integrate_adjoint_refinement_error_estimate(), and libMesh::EulerSolver::integrate_adjoint_refinement_error_estimate().

max_deltat

Real libMesh::AdaptiveTimeSolver::max_deltat

inherited

Do not allow the adaptive time solver to select deltat > max_deltat.

If you use the default max_deltat=0.0, then deltat is unlimited.

Definition at line 183 of file adaptive_time_solver.h.

Referenced by solve().

max_growth

Real libMesh::AdaptiveTimeSolver::max_growth

inherited

Do not allow the adaptive time solver to select a new deltat greater than max_growth times the old deltat.

If you use the default max_growth=0.0, then the deltat growth is unlimited.

Definition at line 197 of file adaptive_time_solver.h.

Referenced by solve().

min_deltat

Real libMesh::AdaptiveTimeSolver::min_deltat

inherited

Do not allow the adaptive time solver to select deltat < min_deltat.

The default value is 0.0.

Definition at line 189 of file adaptive_time_solver.h.

Referenced by solve().

next_step_deltat

Real libMesh::UnsteadySolver::next_step_deltat

protectedinherited

Definition at line 238 of file unsteady_solver.h.

Referenced by libMesh::Euler2Solver::integrate_adjoint_refinement_error_estimate(), and libMesh::EulerSolver::integrate_adjoint_refinement_error_estimate().

old_adjoints

std::vector< std::unique_ptr<NumericVector<Number> > > libMesh::UnsteadySolver::old_adjoints

protectedinherited

A vector of pointers to vectors holding the adjoint solution at the last time step.

Definition at line 231 of file unsteady_solver.h.

Referenced by libMesh::Euler2Solver::integrate_adjoint_refinement_error_estimate(), libMesh::EulerSolver::integrate_adjoint_refinement_error_estimate(), and libMesh::UnsteadySolver::UnsteadySolver().

old_local_nonlinear_solution

std::shared_ptr<NumericVector<Number> > libMesh::UnsteadySolver::old_local_nonlinear_solution

inherited

Serial vector of _system.get_vector("_old_nonlinear_solution") This is a shared_ptr so that it can be shared between different derived class instances, as in e.g.

AdaptiveTimeSolver.

Definition at line 178 of file unsteady_solver.h.

Referenced by libMesh::AdaptiveTimeSolver::adjoint_advance_timestep(), libMesh::UnsteadySolver::adjoint_advance_timestep(), libMesh::AdaptiveTimeSolver::advance_timestep(), libMesh::UnsteadySolver::advance_timestep(), libMesh::AdaptiveTimeSolver::init(), libMesh::UnsteadySolver::init_data(), libMesh::UnsteadySolver::old_nonlinear_solution(), libMesh::UnsteadySolver::reinit(), libMesh::UnsteadySolver::retrieve_timestep(), and libMesh::UnsteadySolver::update().

quiet

bool libMesh::TimeSolver::quiet

inherited

Print extra debugging information if quiet == false.

Definition at line 230 of file time_solver.h.

Referenced by solve(), libMesh::UnsteadySolver::solve(), and libMesh::EigenTimeSolver::solve().

reduce_deltat_on_diffsolver_failure

unsigned int libMesh::TimeSolver::reduce_deltat_on_diffsolver_failure

inherited

This value (which defaults to zero) is the number of times the TimeSolver is allowed to halve deltat and let the DiffSolver repeat the latest failed solve with a reduced timestep.

Note

This has no effect for SteadySolvers.

You must set at least one of the DiffSolver flags "continue_after_max_iterations" or "continue_after_backtrack_failure" to allow the TimeSolver to retry the solve.

Definition at line 259 of file time_solver.h.

Referenced by solve(), and libMesh::UnsteadySolver::solve().

solution_history

std::unique_ptr<SolutionHistory> libMesh::TimeSolver::solution_history

protectedinherited

A std::unique_ptr to a SolutionHistory object.

Default is NoSolutionHistory, which the user can override by declaring a different kind of SolutionHistory in the application

Definition at line 319 of file time_solver.h.

Referenced by libMesh::UnsteadySolver::adjoint_advance_timestep(), libMesh::UnsteadySolver::advance_timestep(), libMesh::TimeSolver::get_solution_history(), libMesh::UnsteadySolver::retrieve_timestep(), and libMesh::TimeSolver::set_solution_history().

target_tolerance

Real libMesh::AdaptiveTimeSolver::target_tolerance

inherited

This tolerance is the target relative error between an exact time integration and a single time step output, scaled by deltat.

integrator, scaled by deltat. If the estimated error exceeds or undershoots the target error tolerance, future timesteps will be run with deltat shrunk or grown to compensate.

The default value is 1.0e-2; obviously users should select their own tolerance.

If a negative target_tolerance is specified, then its absolute value is used to scale the estimated error from the first simulation time step, and this becomes the target tolerance of all future time steps.

Definition at line 160 of file adaptive_time_solver.h.

Referenced by solve().

upper_tolerance

Real libMesh::AdaptiveTimeSolver::upper_tolerance

inherited

This tolerance is the maximum relative error between an exact time integration and a single time step output, scaled by deltat.

If this error tolerance is exceeded by the estimated error of the current time step, that time step will be repeated with a smaller deltat.

If you use the default upper_tolerance=0.0, then the current time step will not be repeated regardless of estimated error.

If a negative upper_tolerance is specified, then its absolute value is used to scale the estimated error from the first simulation time step, and this becomes the upper tolerance of all future time steps.

Definition at line 177 of file adaptive_time_solver.h.

Referenced by solve().

---

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

• include/solvers/twostep_time_solver.h
• src/solvers/twostep_time_solver.C