Initialize solved/zero_resid flags to false

[farhadrclass]

Set solved, zero_resid, and zero_resid_lim to false during initialization instead of computing them from rNorm ≤ ε. This avoids marking the problem as solved at startup (and potential use of rNorm before it's set), ensuring stopping flags are only updated after the residual is computed.

[Copilot]

Pull request overview

This PR updates MINRES-QLP initialization so solved, zero_resid, and zero_resid_lim are set to false initially rather than being derived from the initial residual check (rNorm ≤ ε), with the intent to only update stopping flags after the residual has been computed.

Changes:

• Change stopping-criterion initialization in minres_qlp! to set solved/zero_resid/zero_resid_lim to false at startup.

---

💡 Add Copilot custom instructions for smarter, more guided reviews. Learn how to get started.

[Copilot]

Initializing solved to false removes the pre-loop convergence check (rNorm ≤ ε). If the initial (possibly warm-start) residual already satisfies the user tolerances, the solver will now run at least one iteration and may exit with a different status (e.g., breakdown / nonpositive curvature in linesearch mode) instead of cleanly returning at iter = 0. Since rNorm and ε are already computed before this block, consider restoring the initial solved = (rNorm ≤ ε) check (or an early return when it holds) while keeping zero_resid/zero_resid_lim initialized to false if the concern is about those flags’ semantics.

Suggested change

	solved = zero_resid = zero_resid_lim = false
	solved = (rNorm ≤ ε)
	zero_resid = zero_resid_lim = false

[dpo]

Set zero_resid = zero_resid_lim = rNorm ≤ ε, but solved = false.

[amontoison]

I don't understand your modification @farhadrclass.
Why do you want to do an iteration in all cases if the residual is small enough?
It includes cases like warm-start and we want to stop if ||r_0|| = ||b - A x_0|| is already very small.

[farhadrclass]

@amontoison it should be as @dpo mentioned, I made a mistake and asssign all of them

[amontoison]

You now go in the while loop with your modification:
https://github.com/JuliaSmoothOptimizers/Krylov.jl/blob/main/src/minres_qlp.jl#L238

We don't want that.

[amontoison]

Set solved, zero_resid, and zero_resid_lim to false during initialization instead of computing them from rNorm ≤ ε. This avoids marking the problem as solved at startup (and potential use of rNorm before it's set), ensuring stopping flags are only updated after the residual is computed.

I don't understand the goal of the PR.
You want to check rNorm == 0 or rNorm ≤ ε before the main loop in many methods.
rNorm is well-defined here before the main loop and it seems to me that we don't need any modification.