Seeking a robust solution for large nonlinear optimization problems? This paper introduces L-BFGS-B, a limited-memory algorithm specifically designed for solving large-scale nonlinear optimization problems with simple bounds on the variables. Intended for scenarios where Hessian matrix information is difficult to obtain or where problems are large and dense, L-BFGS-B proves particularly valuable. The algorithm also serves effectively for unconstrained problems, exhibiting performance comparable to its predecessor, algorithm L-BFGS. Implemented in Fortran 77, L-BFGS-B offers a practical and efficient tool for tackling complex optimization challenges. This algorithm enhances computational capabilities in various fields, from engineering design to machine learning, where optimization is crucial for solving real-world problems.
Published in ACM Transactions on Mathematical Software, this paper aligns perfectly with the journal's focus on numerical algorithms and mathematical software. By presenting a new algorithm and its implementation details, the paper contributes to the journal's ongoing discourse on efficient and reliable computational tools.