Tempus::AdjointSensitivityUtils Namespace Reference

Functions
AdjointResponseType	parseResponseType (const std::string &name)
	Convert the "Response Type" parameter value to the enum.
template<class Scalar>
std::vector< Scalar >	parseTimeList (const std::string &str)
template<class Scalar>
void	evalResponseGradientX (const Thyra::ModelEvaluator< Scalar > &model, const int g_index, const Thyra::ModelEvaluatorBase::InArgs< Scalar > &inArgs, Thyra::MultiVectorBase< Scalar > &dgdx, const bool dgdx_dot_check, const std::string &caller)
	Evaluate dg/dx^T (gradient form, x_space x num_g) of response g_index at the given inArgs.
template<class Scalar>
void	evalResponseGradientP (const Thyra::ModelEvaluator< Scalar > &model, const int g_index, const int p_index, const Thyra::ModelEvaluatorBase::InArgs< Scalar > &inArgs, Thyra::MultiVectorBase< Scalar > &dgdp)
	Evaluate dg/dp^T (gradient form, p_space x num_g) of response g_index with respect to parameter p_index at the given inArgs.
template<class Scalar>
void	applyDfDpTranspose (const Thyra::ModelEvaluator< Scalar > &model, const int p_index, const Thyra::ModelEvaluatorBase::InArgs< Scalar > &inArgs, const Thyra::MultiVectorBase< Scalar > &v, Thyra::MultiVectorBase< Scalar > &y, const Scalar alpha, const Scalar beta, Teuchos::RCP< Thyra::LinearOpBase< Scalar > > &dfdp_op, Teuchos::RCP< Thyra::MultiVectorBase< Scalar > > &dfdp_mv)
	Compute y = alpha * df/dp^T * v + beta * y.
template<class Scalar>
Teuchos::RCP< Thyra::LinearOpWithSolveBase< Scalar > >	createAdjointLinearOpWithSolve (const Thyra::ModelEvaluator< Scalar > &adjoint_model, const Thyra::ModelEvaluatorBase::InArgs< Scalar > &inArgs, const Scalar alpha, const Scalar beta)
	Create a solver for the adjoint model's W = (alpha*df/dx_dot + beta*df/dx)^T at inArgs.
template<class Scalar>
void	addResponseTimeEvents (TimeStepControl< Scalar > &tsc, const std::vector< Scalar > &times)
	Add the response times to the time step control as time events on which the time steps land exactly.

Function Documentation

parseResponseType()

AdjointResponseType Tempus::AdjointSensitivityUtils::parseResponseType ( const std::string & name )

inline

Convert the "Response Type" parameter value to the enum.

Definition at line 53 of file Tempus_AdjointSensitivityUtils.hpp.

parseTimeList()

template<class Scalar>

std::vector< Scalar > Tempus::AdjointSensitivityUtils::parseTimeList

(

const std::string &

str

)

Convert a comma-separated list of times (as in "Output Time List") to a sorted vector.

Definition at line 70 of file Tempus_AdjointSensitivityUtils.hpp.

evalResponseGradientX()

template<class Scalar>

void Tempus::AdjointSensitivityUtils::evalResponseGradientX	(	const Thyra::ModelEvaluator< Scalar > &	model,
		const int	g_index,
		const Thyra::ModelEvaluatorBase::InArgs< Scalar > &	inArgs,
		Thyra::MultiVectorBase< Scalar > &	dgdx,
		const bool	dgdx_dot_check,
		const std::string &	caller )

Evaluate dg/dx^T (gradient form, x_space x num_g) of response g_index at the given inArgs.

If dgdx_dot_check is true and the model supports dg/dx_dot in gradient form, dg/dx_dot is evaluated in the same call and an exception is thrown if it is nonzero, since the adjoint formulation requires responses that do not depend on x_dot.

Definition at line 110 of file Tempus_AdjointSensitivityUtils.hpp.

evalResponseGradientP()

template<class Scalar>

void Tempus::AdjointSensitivityUtils::evalResponseGradientP	(	const Thyra::ModelEvaluator< Scalar > &	model,
		const int	g_index,
		const int	p_index,
		const Thyra::ModelEvaluatorBase::InArgs< Scalar > &	inArgs,
		Thyra::MultiVectorBase< Scalar > &	dgdp )

Evaluate dg/dp^T (gradient form, p_space x num_g) of response g_index with respect to parameter p_index at the given inArgs.

Models that only provide the Jacobian form are supported by transposing, as done for the final-time response.

Definition at line 157 of file Tempus_AdjointSensitivityUtils.hpp.

applyDfDpTranspose()

template<class Scalar>

void Tempus::AdjointSensitivityUtils::applyDfDpTranspose	(	const Thyra::ModelEvaluator< Scalar > &	model,
		const int	p_index,
		const Thyra::ModelEvaluatorBase::InArgs< Scalar > &	inArgs,
		const Thyra::MultiVectorBase< Scalar > &	v,
		Thyra::MultiVectorBase< Scalar > &	y,
		const Scalar	alpha,
		const Scalar	beta,
		Teuchos::RCP< Thyra::LinearOpBase< Scalar > > &	dfdp_op,
		Teuchos::RCP< Thyra::MultiVectorBase< Scalar > > &	dfdp_mv )

Compute y = alpha * df/dp^T * v + beta * y.

v is a multivector in the residual (f) space and y a multivector in the space of parameter p_index, with the same number of columns. df/dp is evaluated at inArgs in the form supported by the model (linear operator, Jacobian or gradient multivector). dfdp_op and dfdp_mv cache the derivative objects between calls (pass null RCPs on the first call).

Definition at line 197 of file Tempus_AdjointSensitivityUtils.hpp.

createAdjointLinearOpWithSolve()

template<class Scalar>

Teuchos::RCP< Thyra::LinearOpWithSolveBase< Scalar > > Tempus::AdjointSensitivityUtils::createAdjointLinearOpWithSolve	(	const Thyra::ModelEvaluator< Scalar > &	adjoint_model,
		const Thyra::ModelEvaluatorBase::InArgs< Scalar > &	inArgs,
		const Scalar	alpha,
		const Scalar	beta )

Create a solver for the adjoint model's W = (alpha*df/dx_dot + beta*df/dx)^T at inArgs.

Uses the W out-arg of the adjoint model if supported, otherwise its W_op, its W_factory, and the preconditioner of the W_factory (or of the model).

Definition at line 240 of file Tempus_AdjointSensitivityUtils.hpp.

addResponseTimeEvents()

template<class Scalar>

void Tempus::AdjointSensitivityUtils::addResponseTimeEvents	(	TimeStepControl< Scalar > &	tsc,
		const std::vector< Scalar > &	times )

Add the response times to the time step control as time events on which the time steps land exactly.

This only adjusts the time steps for "Step Type" = "Variable"; with "Constant" time steps the response times must be on the time grid. The time steps are not flagged for output.

Definition at line 310 of file Tempus_AdjointSensitivityUtils.hpp.