Todo List

Member checkInside (Dune::GeometryType const &gt, LocalCoordinate const &x)

The current implementation is just for cubes and simplices. Pyramids and prisms are not covered.

Class Kaskade::AdditiveMultiGrid< Smoother, Prolongation, CoarsePreconditioner >

: consider inheriting from Kaskade::SymmetricPreconditioner

Class Kaskade::BDDC::Subdomain< m, Scalar_, CoarseScalar, Transfer >

UMFPACK does not provide float factorization. Extract L and U factors and convert to float?

Class Kaskade::BoundaryTangents< dim, Index, ctype >

consider to turn this into a function returning the edge set

Class Kaskade::CacheBase< Functional, Cache, ScalarT >

docme better

Class Kaskade::DirichletNitscheBoundary< GridView, components, ScalarType >

Consider extending to M. Juntunen, R. Stenberg: NITSCHE’S METHOD FOR GENERAL BOUNDARY CONDITIONS, Math. Comp. 78(267):1353-1374, 2009.

Member Kaskade::DynamicMatrix< K >::operator() (RRange const &ridx, CRange const &cidx) const

change the implementation to work with forward ranges

Member Kaskade::EntityNumbering< GridView >::SortOrder

shall this be extendable by deriving from the class, or by policies, or ...?

Member Kaskade::FEFunctionSpace< LocalToGlobalMapper >::gridManager () const

why is this a const function returning a non-const grid manager reference?

Class Kaskade::FunctionViewAdaptor< Space_, Functor >

implement this as derived class of WeakFunctionViewAdaptor

Member Kaskade::GridManagerBase< Grd >::entityNumbering (typename Grid::LeafGridView const &) const

allow to specify the sorting

Member Kaskade::HdivSimplexShapeFunction< ctype, dimension, T >::HdivSimplexShapeFunction ()=default

makes only sense if it is assignable - any real need for default constructibility?

Class Kaskade::LagrangeOperator< NormalStepAssembler, TangentialStepAssembler, stateId, controlId, adjointId >

docme

Class Kaskade::Limex< Eq >

Currently the matrix \( B \) has to be constant.

Member Kaskade::Limex< Eq >::step (State const &x, double dt, int order, std::vector< std::pair< double, double > > const &tolX, std::vector< std::vector< bool > > &refinements)

(i) check for B constant, do not reassemble matrix in this case (ii) implement fixed point iteration instead of new factorization in case B is not constant

Member Kaskade::Limex< Eq >::step (State const &x, double dt, int order, std::vector< std::pair< double, double > > const &tolX)

(i) check for B constant, do not reassemble matrix in this case (ii) implement fixed point iteration instead of new factorization in case B is not constant

Member Kaskade::Limex< Eq >::step (State const &x, double dt, int order)

(i) check for B constant, do not reassemble matrix in this case (ii) implement fixed point iteration instead of new factorization in case B is not constant

Class Kaskade::LocalGeometryInCoarseGridAncestor< Grid >

doc me

Member Kaskade::makeAlgebraicMultigridStack (NumaBCRSMatrix< Entry, Index > &&A, Index n=0, bool onlyLowerTriangle=false)

Currently, the interpolation is just with factor 0.5 from both coarse parents - not really clever for unstructured grids or jumping coefficients. Implement a more clever weighting.

Member Kaskade::makeAuxiliarySpaceMultigridStack (Space const &space, NumaBCRSMatrix< Entry, Index > const &A, double volumeRatio=100)

: implement for general spaces

Class Kaskade::MatrixRepresentedOperator< Matrix_, Domain_, Range_ >

docme: How does this relate to Dune::MatrixAdapter and Dune::AssembledLinearOperator?

Class Kaskade::MultiLevelLocalGeometry< Grid >

: implement pull-back and pull-forward operation for gradients

Member Kaskade::MultiLevelLocalGeometry< Grid >::MultiLevelLocalGeometry (Grid const &grid, Cell const &child_, Cell const &ancestor_, MultiLevelLocalGeometry< Grid >::Direction direction_, bool geometricallyNested_=(dim==dimw), ctype tol_=-100)

: (i) Simplify code by using entity pointer's comparison operator instead of ids (caveat: can ids lead to shorter hierarchies? Remember that entities copied to higher grid levels have the same id. If the comparison operators between entity pointers evaluate to "not equal" in this case, comparing ids may be more efficient. (ii) Check whether spanwidth can be computed as difference of the entity levels instead of using the loop. The same consideration as for (i) applies.

Class Kaskade::Mutex

: design appropriate semantics for move construction/assignment

Member Kaskade::NumaBCRSMatrix< Entry, Index >::frobenius_norm () const

parallelize

Member Kaskade::NumaBCRSMatrix< Entry, Index >::frobenius_norm2 () const

parallelize

Member Kaskade::NumaCRSPatternCreator< Index >::addAllElements ()

This should be statically encoded in a different matrix type.

Member Kaskade::QPALSolver< d, sparsity, Real >::QPALSolver (MatrixA const &A, MatrixB const &B)

write a move constructor

Member Kaskade::QPBoundSolver< R >::QPBoundSolver (Matrix const &A, QPConvexificationStrategy convex=QPConvexificationStrategy::DONOTHING)

write a move constructor

Member Kaskade::QPDirectSparse< R >::QPDirectSparse (SparseMatrix const &A, QPConvexificationStrategy convex=QPConvexificationStrategy::DONOTHING)

write a move constructor

Member Kaskade::QPPenalizedSolver< d, sparsity, Real >::multiplierEstimate (VectorX const &x, VectorX c) const

Check if \( BA^{-1}B^T \lambda = BA + A^{-1}c \) is better or more appropriate, as it minimizes the residual in the problem's energy norm.

Member Kaskade::QPPenalizedSolver< d, sparsity, Real >::QPPenalizedSolver (MatrixA const &A, MatrixB const &B, Real gamma)

write a move constructor

Member Kaskade::QPPenalizedSolver< d, sparsity, Real >::QPPenalizedSolver (MatrixA const &A, MatrixB const &B)

write a move constructor

Member Kaskade::QPSlackSolver< sparsity, R >::QPSlackSolver (Matrix const &A, Matrix const &B, Real gamma, QPConvexificationStrategy convex=QPConvexificationStrategy::DONOTHING)

write a move constructor

Member Kaskade::QPSolver< d, Real >::QPSolver (MatrixA const &A, MatrixB const &B, QPConvexificationStrategy convex=QPConvexificationStrategy::DONOTHING)

write a move constructor

Member Kaskade::QPSolver< d, Real >::solve (VectorX const &c, VectorB const &b, double tol) const

add option to provide an initial guess for \( \lambda \).

Class Kaskade::RestrictedShapeFunctionSet< ShapeFunctionSet_ >

docme: what is is good for? Which ordering? Why "restricted"?

Class Kaskade::ScalarConverter< Cell, Scalar >

: The Hessian depends on the curvature of the grid - this is not taken into account yet.

Class Kaskade::ShapeFunctionCache< G, T, ComponentsEnd >

make shape function cache independent of grid type - only coordinate type and dimension are required

Class Kaskade::ThreadedMatrixDetail::NumaBCRSMatrixConstRowIterator< Entry, Index >

Implement a proper iterator, e.g., using boost::iterator_adaptor

Class Kaskade::Timings

Implement merging of timer statistics

Class Kaskade::TransposedOperator< Operator >

docme

Member Kaskade::twoGridProlongation (FromSpace const &from, ToSpace const &to)

generalize to arbitrary spaces, not only P1?

Member Kaskade::VariationalFunctionalAssembler< F, SparseIndex, BoundaryDetector, QuadRule >::getMatrix () const

check const correctness - why is this method const and returns a mutable matrix?

Class PartialIndexSet< Grid, IndexSet, Part >

recompute internal representation whenever grid changes.

Class RestartedFGMResSolver< X >

F durch rebind erzeugen und nur den field_type für F übergeben

Member two_norm

Currently, the norm is computed inefficiently via the signular value decomposition, and takes \( O(n^3) \) time. Consider implementing a power iteration.