opt

opt is a Rust library for dense nonlinear optimization: BFGS, Newton trust-region, ARC, matrix-free trust-region, and fixed-point iteration, plus the shared policy primitives (trust-region radius control, Levenberg-Marquardt damping, ridge escalation, backtracking and bracketed root-finding) the solvers are built from.

On wolfe_bfgs

wolfe_bfgs was a second crate in this repository that re-exported opt's BFGS-focused API under a narrower name. It has been removed. It carried no code of its own — only a pub use list — and that list had to be kept in lockstep with opt's public surface by hand. It fell behind exactly as you would expect: the 0.6.0 release broke the workspace build on all three platforms because the re-export crate still required opt = "^0.5.15", and crates.io accepted the opt publish anyway because it builds each package independently. A crate whose only content is a list that can silently go stale is a liability, not a convenience.

Versions already on crates.io (up to wolfe_bfgs 0.7.0) are left published and keep working. New code should depend on opt directly; every name wolfe_bfgs exported is available from opt under the same path.

Repository layout

.
├── Cargo.toml
├── README.md
├── opt
│   ├── Cargo.toml
│   ├── README.md
│   ├── optimization_harness.py
│   └── src
│       └── lib.rs
└── .github
    └── workflows

Common commands

Run the full workspace tests:

cargo test --workspace

Run only the main crate tests:

cargo test -p opt

Publish:

cargo publish -p opt

The publish workflow in .github/workflows/publish.yml does the same on a crate-* tag, skipping the publish when the version is already on crates.io.