Arpack.jl

Arpack.jl copied to clipboard

I have two symmetric PD sparse matrices, and I call

d, v, nconv = eigs(Symmetric(K+OmegaShift*M), Symmetric(M); nev=neigvs, which=:SM)

Clearly it is reasonable to expect the results to be normalized relative to the matrix B. This however does not happen since the designation of the matrices as symmetric is lost due to explicittransform being set internally to :shiftinvert.
This is a violation of backward compatibility.

Also, allowing explicit transform to happen appears to carry with it a significant performance penalty. In the above case the execution time more than doubled.

See #120