Lagrange multipliers for constraints

[dbeurle]

Lagrange multipliers can be used to enforce constraints on the displacements of the system to machine precision. The introduction of these constraints change the structure and increase of the sparse matrix and restrict the type of solver to be used to only direct linear solvers.

It should be possible to implement this partitioned structure for displacement boundary conditions for non-linear problems and for using periodic boundary conditions for example. Tie constraints also naturally fall out of this.

A constraint matrix can be formulated for each of these describing the derivative of the gap function used for the particular constraint and provided under a different class of boundary conditions.

• Add interface class constraint
• Support constraint dofs in fem::mesh and fem::matrix
• Prohibit use of non-compatible linear solvers with zero-diagonal matrices