1: \begin{abstract}

2:     Multi-objective optimization (MOO) aims to optimize multiple, possibly

3:     conflicting objectives with widespread applications. We introduce a novel

4:     interacting particle method for MOO inspired by molecular dynamics

5:     simulations. Our approach combines overdamped Langevin and birth-death

6:     dynamics, incorporating a ``dominance potential'' to steer particles toward

7:     global Pareto optimality. In contrast to previous methods, our method is

8:     able to relocate dominated particles, making it particularly adept at

9:     managing Pareto fronts of complicated geometries. Our method is also

10:     theoretically grounded as a Wasserstein-Fisher-Rao gradient flow with

11:     convergence guarantees. Extensive experiments confirm that our approach

12:     outperforms state-of-the-art methods on challenging synthetic and real-world

13:     datasets.

14: \end{abstract}

15: