Crate lbfgsb_rs_pure 

A compact, self-contained reimplementation of an L-BFGS-B style solver in Rust.

This crate focuses on a safe, idiomatic Rust port of the algorithmic core of the classical L-BFGS-B package. It aims for algorithmic parity and reproducibility rather than maximal performance — for production use link a tuned BLAS/LAPACK provider.

The main public API is the LBFGSB type (in solver), together with the Status and Solution types returned by the solver.

Example

use lbfgsb_rs::{LBFGSB, Status};

let mut x = vec![0.0];
let lower = vec![-1.0];
let upper = vec![1.0];

let mut solver = LBFGSB::new(5);
let sol = solver
    .minimize(&mut x, &lower, &upper, |x| {
        let dx = x[0] - 0.5;
        (dx * dx, vec![2.0 * dx])
    })
    .unwrap();

assert_eq!(sol.status, Status::Converged);
assert!((sol.x[0] - 0.5).abs() < 1e-6);

Modules

• blas : lightweight level-1 BLAS helpers used throughout the crate.
• linesearch : line-search helpers (dcstep / lnsrlb / convenience methods).
• linpack : small linpack routines (Cholesky / triangular solve).
• subalgorithms: helper routines (active-set initialization, bmv partial port).
• solver : high-level LBFGS-B driver (uses the above modules).

The crate re-exports LBFGSB from solver and provides the Status and Solution types here so solver can reference them as crate::Status and crate::Solution (matching the original design).

Re-exports

pub use crate::solver::IterationControl;

pub use crate::solver::IterationInfo;

pub use crate::solver::LBFGSB;

Modules

linpack

solver

High-level LBFGS-B solver implementation using the helper modules.

subalgorithms

Structs

Solution

Solution returned by the solver.

Enums

Status

Enum describing solver termination.

Constants

CURVATURE_EPS

Minimum curvature s’y to accept BFGS update.

MAX_BACKTRACK

Maximum number of backtracking line search iterations used by convenience routines.

Functions

daxpy

Compute y := y + a * x in-place.

dcopy

Copy vector x into y (y := x).

dcsrch

Minimal port of dcsrch (More’-Thuente controller).

dcstep

Compute a safeguarded step update for the More’-Thuente algorithm.

ddot

Compute the dot product of x and y.

dscal

Scale vector x by scalar a in-place (x := a * x).

line_search_and_update

Convenience line search for use by the simplified solver.

lnsrlb

Safeguarded More’-Thuente style line search driver.

lnsrlb_search

Convenience wrapper around lnsrlb that mirrors the previous line_search_and_update signature used by the solver. It sets standard More’-Thuente parameters and calls lnsrlb.