Combine fixed BlockTensor vector algebra with an owned matrix-free apply operation. More...

#include <block_tensor_vector.hpp>

Inheritance diagram for uni20::krylov::BlockTensorMatrixFreeOps< Tensor, Operator >:

Collaboration diagram for uni20::krylov::BlockTensorMatrixFreeOps< Tensor, Operator >:

Public Types
using	base_type = BlockTensorVectorOps< Tensor >

using	operator_type = std::remove_cvref_t< Operator >

using	tensor_type = typename base_type::tensor_type

Public Types inherited from uni20::krylov::BlockTensorVectorOps< Tensor >
using	codomain_type = typename tensor_type::codomain_type

using	domain_type = typename tensor_type::domain_type

using	key_type = typename tensor_type::key_type

using	real_type = make_real_t< scalar_type >

using	scalar_type = typename tensor_type::value_type

using	tensor_type = std::remove_cvref_t< Tensor >

Public Member Functions
	BlockTensorMatrixFreeOps (tensor_type const &prototype, operator_type operation)
	Freeze a vector structure and take ownership of its apply operation.

void	matvec (tensor_type &output, tensor_type const &input)
	Apply the stored matrix-free operation as `output <- OP*input`.

Public Member Functions inherited from uni20::krylov::BlockTensorVectorOps< Tensor >
auto	allocate_like (tensor_type const &vector) const -> tensor_type
	Allocate a vector with the frozen structure and unspecified values.

template<BorrowedBlockTensorView View> requires detail::CompatibleBlockVector<View, tensor_type>
auto	allocate_like (View const &view) const -> tensor_type
	Allocate owning workspace from retained view metadata and its leaf context.

void	axpy (tensor_type &output, scalar_type factor, tensor_type const &input) const
	Add a scalar multiple of one input to an output.

	BlockTensorVectorOps (tensor_type const &prototype)
	Freeze the Krylov vector space represented by a prototype tensor.

void	copy (tensor_type &output, tensor_type const &input) const
	Copy one vector into another without changing either structure.

auto	inner_product (tensor_type const &lhs, tensor_type const &rhs) const -> scalar_type
	Return the conjugate-linear inner product as a host scalar.

auto	norm (tensor_type const &vector) const -> real_type
	Return the Euclidean norm as a host scalar.

auto	problem_dimension () const noexcept -> std::size_t
	Return the number of stored numerical degrees of freedom.

template<class Vector > requires detail::CompatibleBlockVector<Vector, tensor_type> && (std::is_lvalue_reference_v<Vector&&> \|\| BorrowedBlockTensorView<Vector>)
auto	prototype_view (Vector &&vector) const
	Own prototype metadata while borrowing its numerical payload and allocation context.

void	scal (tensor_type &vector, scalar_type factor) const
	Scale one vector in place.

void	set_zero (tensor_type &vector) const
	Set every stored numerical block to zero.

template<class Vector > requires detail::CompatibleBlockVector<Vector, tensor_type>
auto	vector_dimension (Vector const &vector) const -> std::size_t
	Validate a vector and return the frozen problem dimension.

Detailed Description

template<class Tensor, class Operator>
class uni20::krylov::BlockTensorMatrixFreeOps< Tensor, Operator >

Combine fixed BlockTensor vector algebra with an owned matrix-free apply operation.

The operation is invoked as operation(output, input) only after both vectors have been validated against the exact structure frozen by the prototype. The operation object may retain immutable Hamiltonian or environment state needed by repeated Krylov applies.

Template Parameters

Tensor	Owning sparse BlockTensor value type.
Operator	Output-first matrix-free operation object.