sparse Jacobian and/or coloring

[Mauro]

For an implicit ode/dae solver, I want to provide automatic computation
of a sparse Jacobian, either through automatic differentiation or finite
differences. So, for given a function

f(y, ydot)

I need to compute the Jacobian in the following form:

J = df/dy + a* df/dydot

ReverseDiffSparse seems to do most of this (although Hessians are
symmetric) but I struggle to make sense of it. Given f and a sparsity
pattern of J, is there an incantation to get df/dy and df/dydot?

If I wanted to use the coloring of ReverseDiffSparse (or some other
package) for finite differencing, how would I go about it? What
function calls would give me a coloring of the colums?

Thanks! Mauro

[Miles Lubin]

Hi Mauro,

Could you open an issue on ReverseDiffSparse to continue this discussion?

Miles

[Mauro]

Opened issue
https://github.com/mlubin/ReverseDiffSparse.jl/issues/18

Tnx, M