How to provide the VectorMemory<BlockVector<double> > for the class SchurMatrix?

[Jesse]

Deal all,

I'm using the mixed finite element to solve a Cahn-HIlliard type equation, after discretization I obtain a block matrix system. I plan to use the class SchurMatrix in deal.ii to obtain the Schur Complement.

The block matrix system is 

* / \ / \ / \ * | A Dt | | u | - | f | * | -B C | | p | - | g | * \ / \ / \ /

where A and C are sparse mass matrices, B Dt are sparse stiff matrices.

But when constructing the SchurMatrix object, I need to provide  'mem'. Could someone help me to provide 'mem'?

Thank you in advance.

Jesse 

template<class MA_inverse , class MB , class MDt , class MC >

SchurMatrix< MA_inverse, MB, MDt, MC >::SchurMatrix	(	const MA_inverse &	Ainv,
		const MB &	B,
		const MDt &	Dt,
		const MC &	C,
		VectorMemory< BlockVector< double > > &	mem,
		const std::vector< types::global_dof_index > &	signature = std::vector<types::global_dof_index>(0)
	)

[Jean-Paul Pelteret]

Dear Jesse,

You can probably just provide an instance of the GrowingVectorMemory class templated on BlockVector< double >. So this would look something like:

GrowingVectorMemory< BlockVector<double> > memory;
const SchurMatrix schur_matrix (A_inv, B, Dt, C, memory);

...

If its of any help to you, and if you have a C++11 compatible compiler, then you could alternatively use the (relatively) new LinearOperator class and the schur_complement operator. It provides a more flexible framework with which you could perform the same operation.

Kind regards,

Jean-Paul

[Jesse]

Dear Jean-Paul,

Thank you very much for your help. This is exactly what I needed. My compiler support c++11 features, I will also try the new schur_complement operator.

With best wishes,

Jesse

在 2017年4月26日星期三 UTC+2下午4:22:26，Jean-Paul Pelteret写道：