Multilevel fill strategy

  In the elimination of a level we may allow for a certain amount of fill-in. Without any fill-in or with fill-in only on the diagonal, the original sparsity structure of the matrix is preserved; with other fill strategies the matrix on subsequent levels may become more and more dense.

These are the fill strategies defined:

AMLFillNone

no fill-in, this results in an SSOR type factorisation. Basically, the factorisation is only concerned with identifying the level sets.

AMLFillDiag

allow fill-in on the diagonal. This gives an ILU(0)-type factorisation.

AMLFillStrong

allow large fill-in elements. This is in effect a drop tolerance method. The tolerance is out of user control for the time being.

AMLFillFull

we accept the full fill matrix.