Open ChrisRackauckas opened 6 years ago
Having banded jacobians would be great for CVODE. Right now explicit sparse jacobian + linear_solver=:KLU
is about twice as fast as no jacobian + linear_solver=:Band
. I think explicit jacobian + linear_solver=:Band
should be really good.
@dlfivefify Page 180 (163 in the document) of https://computation.llnl.gov/sites/default/files/public/cv_guide.pdf . Looking at https://github.com/JuliaMatrices/BandedMatrices.jl/blob/master/src/banded/BandedMatrix.jl#L17 . I want to do a conversion of the Sundials banded matrix to your type. data
is the same but I'd have to reshape it, l
and u
come from ml
and mu
, but what's raxis
?
We have user-defined dense Jacobians. Now we can use that to setup user-defined banded Jacobians. Ping me if you want this implemented.