1: \begin{abstract}

2: Least squares form one of the most prominent classes of optimization problems, 

3: with numerous applications in scientific computing and data fitting. When 

4: such formulations aim at modeling complex systems, the optimization process 

5: must account for nonlinear dynamics by incorporating constraints. In addition, 

6: these systems often incorporate a large number of variables, 

7: which increases the difficulty of the problem, and motivates the need for 

8: efficient algorithms amenable to large-scale implementations.

9: 

10: In this paper, we propose and analyze a Levenberg-Marquardt algorithm for 

11: nonlinear least squares subject to nonlinear equality constraints. Our 

12: algorithm is based on inexact solves of linear least-squares problems, that

13: only require Jacobian-vector products. Global convergence is 

14: guaranteed by the combination of a composite step approach and a 

15: nonmonotone step acceptance rule. We illustrate the performance of our 

16: method on several test cases from data assimilation and 

17: inverse problems: our algorithm is able to reach the vicinity of a 

18: solution from an arbitrary starting point, and can outperform the most 

19: natural alternatives for these classes of problems.

20: \end{abstract}

21: