CQuasiNewtonInvLeastSquares< Scalar_t, WithMPI > Class Template Reference

A quasi-Newton fixed-point (FP) accelerator based on IQN-ILS. More...

#include <CQuasiNewtonInvLeastSquares.hpp>

Public Types
using	Scalar = Scalar_t
using	Index = typename su2matrix< Scalar >::Index

Public Member Functions
	CQuasiNewtonInvLeastSquares ()=default
	Default construction without allocation.
	CQuasiNewtonInvLeastSquares (Index nsample, Index npt, Index nvar, Index nptdomain=0)
	Construction with allocation, see "resize".
void	resize (Index nsample, Index npt, Index nvar, Index nptdomain=0)
	Resize the object.
Index	size () const
	Size of the object, the number of samples.
void	reset ()
	Discard all history, keeping the current solution.
su2matrix< Scalar > &	FPresult ()
	Access the current fixed-point result.
const su2matrix< Scalar > &	FPresult () const
Scalar &	FPresult (Index iPt, Index iVar)
const Scalar &	FPresult (Index iPt, Index iVar) const
su2matrix< Scalar > &	solution ()
	Access the current solution approximation.
const su2matrix< Scalar > &	solution () const
Scalar &	operator() (Index iPt, Index iVar)
const Scalar &	operator() (Index iPt, Index iVar) const
const su2matrix< Scalar > &	compute ()
	Compute and return a new approximation.

Detailed Description

template<class Scalar_t, bool WithMPI = true>
class CQuasiNewtonInvLeastSquares< Scalar_t, WithMPI >

A quasi-Newton fixed-point (FP) accelerator based on IQN-ILS.

Note

The implementation prioritizes storage, the LS problem is solved via the normal equations as that is easy to make parallel over MPI, it may however be unstable with large sample sizes (compared to QR decomp). Usage: Allocate, store the initial solution (operator (i,j), default is 0), run the FP, store its result ("FPresult"), compute new solution, use it as the new input of the FP, run the FP, etc.

Member Typedef Documentation

Index

template<class Scalar_t , bool WithMPI = true>

using CQuasiNewtonInvLeastSquares< Scalar_t, WithMPI >::Index = typename su2matrix<Scalar>::Index

Scalar

template<class Scalar_t , bool WithMPI = true>

using CQuasiNewtonInvLeastSquares< Scalar_t, WithMPI >::Scalar = Scalar_t

Constructor & Destructor Documentation

CQuasiNewtonInvLeastSquares() [1/2]

template<class Scalar_t , bool WithMPI = true>

CQuasiNewtonInvLeastSquares< Scalar_t, WithMPI >::CQuasiNewtonInvLeastSquares ( )

default

Default construction without allocation.

CQuasiNewtonInvLeastSquares() [2/2]

template<class Scalar_t , bool WithMPI = true>

CQuasiNewtonInvLeastSquares< Scalar_t, WithMPI >::CQuasiNewtonInvLeastSquares ( Index  nsample, Index  npt, Index  nvar, Index  nptdomain = 0  )

inline

Construction with allocation, see "resize".

Member Function Documentation

compute()

template<class Scalar_t , bool WithMPI = true>

const su2matrix< Scalar > & CQuasiNewtonInvLeastSquares< Scalar_t, WithMPI >::compute ( )

inline

Compute and return a new approximation.

Note

To be used after storing the FP result.

FPresult() [1/4]

template<class Scalar_t , bool WithMPI = true>

su2matrix< Scalar > & CQuasiNewtonInvLeastSquares< Scalar_t, WithMPI >::FPresult ( )

inline

Access the current fixed-point result.

Note

Use these to STORE the result of running the FP.

FPresult() [2/4]

template<class Scalar_t , bool WithMPI = true>

const su2matrix< Scalar > & CQuasiNewtonInvLeastSquares< Scalar_t, WithMPI >::FPresult ( ) const

inline

FPresult() [3/4]

template<class Scalar_t , bool WithMPI = true>

Scalar & CQuasiNewtonInvLeastSquares< Scalar_t, WithMPI >::FPresult ( Index  iPt, Index  iVar  )

inline

FPresult() [4/4]

template<class Scalar_t , bool WithMPI = true>

const Scalar & CQuasiNewtonInvLeastSquares< Scalar_t, WithMPI >::FPresult ( Index  iPt, Index  iVar  ) const

inline

operator()() [1/2]

template<class Scalar_t , bool WithMPI = true>

Scalar & CQuasiNewtonInvLeastSquares< Scalar_t, WithMPI >::operator() ( Index  iPt, Index  iVar  )

inline

operator()() [2/2]

template<class Scalar_t , bool WithMPI = true>

const Scalar & CQuasiNewtonInvLeastSquares< Scalar_t, WithMPI >::operator() ( Index  iPt, Index  iVar  ) const

inline

reset()

template<class Scalar_t , bool WithMPI = true>

void CQuasiNewtonInvLeastSquares< Scalar_t, WithMPI >::reset ( )

inline

Discard all history, keeping the current solution.

resize()

template<class Scalar_t , bool WithMPI = true>

void CQuasiNewtonInvLeastSquares< Scalar_t, WithMPI >::resize ( Index  nsample, Index  npt, Index  nvar, Index  nptdomain = 0  )

inline

Resize the object.

Parameters

[in]	nsample	- Number of samples used to build the FP history.
[in]	npt	- Size of the solution including any halos.
[in]	nvar	- Number of solution variables.
[in]	nptdomain	- Local size (<= npt), if 0 it defaults to npt.

size()

template<class Scalar_t , bool WithMPI = true>

Index CQuasiNewtonInvLeastSquares< Scalar_t, WithMPI >::size ( ) const

inline

Size of the object, the number of samples.

solution() [1/2]

template<class Scalar_t , bool WithMPI = true>

su2matrix< Scalar > & CQuasiNewtonInvLeastSquares< Scalar_t, WithMPI >::solution ( )

inline

Access the current solution approximation.

Note

Use these, after calling compute, to GET the new FP solution estimate.

solution() [2/2]

template<class Scalar_t , bool WithMPI = true>

const su2matrix< Scalar > & CQuasiNewtonInvLeastSquares< Scalar_t, WithMPI >::solution ( ) const

inline

---

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

• Common/include/toolboxes/CQuasiNewtonInvLeastSquares.hpp