Jason Rader

I find either behavior reasonable. To me, multiple consistent values like this seems fine from the interpolation perspective, but maybe not from the diffeq perspective. This type of data came...

Okay!

So the issue in this example is `F` is a JAX array, but the output structure of `J` has the PyTree structure of `State`, as set by the `out_structure` in...

Just to be clear, are you handling a general minimisation problem, or a nonlinear least-squares problem? In your example problem, $F$ is a residual and $J$ the Jacobian, so the...

Gauss-Newton and backtracking Armijo with the mix-and-match API should work then. Gauss-Newton is mathematically equivalent to Newton for nonlinear systems. As Patrick said, they're just applied a little differently. The...

First off, excellent stuff!! Thank you very much for this PR. Regarding block matrices, now may be a good time to settle on an abstraction for them. General linear solves...

Hello! Thanks for the interest in implementing some new line searches. Just to be clear, for a minimisation problem of a function $f\colon \mathbb{R}^n \to \mathbb{R}$, an Optimistix descent $\delta\colon...

Newton CG is one of the algorithms which I think could be somewhat involved to implement. It is further from the existing solvers in Optimistix, so there's more custom work...

Yep, this looks like it may require a more significant rewrite. As you pointed out, the issue is how to get $\delta'(\alpha)$ without access to $\delta$. The most obvious solution...

I don't think this is what's happening. If you replace the matrix in `dense_operator` with the matrix containing only its diagonal, or with an `IdentityLinearOperator`, then no issue arises and...