To address the problems of premature convergence, dimensional coupling imbalance, and topological degradation in the search space in complex high-dimensional nonconvex optimization problems, this paper proposes a novel swarm intelligence optimization method—the Bluefin Tuna Optimization Algorithm. Inspired by the high-speed migration, hunting cooperation, and energy regulation behavior of bluefin tuna, the algorithm constructs a nonlinear manifold dynamics model, a phase transition search mechanism, an information flow coupled tensor field, and a hybrid invertible-irreversible evolutionary structure. The algorithm achieves adaptive search step size adjustment in the metric space by constructing a direction-dependent anisotropic advancement tensor; it achieves automatic phase transition by defining a macroscopic order parameter; it constructs a continuous information flow network through inter-individual coupling kernel functions; it establishes an irreversible topological pruning mechanism through a risk accumulation function; and it combines fractal migration memory to achieve long-range time dependence. Theoretical analysis shows that the algorithm constitutes a nonlinear coupled discrete dynamical system, satisfying boundedness and weak convergence under certain conditions. This method provides a new approach for constructing swarm intelligence optimization frameworks with physical structural interpretations.