Siberian Salamander Optimization Algorithm

To address premature convergence, search collapse, and structural degradation in complex high-dimensional nonconvex optimization problems, this paper proposes a population optimization method based on phase transition dynamics and a dual topology structure—the Siberian salamander optimization algorithm. The algorithm characterizes the individual activity state using a continuous phase transition function, introduces a learnable frozen boundary to achieve adaptive adjustment during the search phase, constructs a dual-layer topological search structure (reversible and irreversible) to enhance stability in complex curvature regions, and achieves direction-selective reinforcement through anisotropic diffusion tensors. Simultaneously, an energy-information coupling model is established to constrain ineffective oscillations during the search process, and a population structure tensor and covariance collapse detection mechanism are combined to prevent population degradation. This paper models the algorithm from a dynamical system perspective, providing convergence analysis and complexity derivation. Theoretical analysis shows that, under appropriate step-size decay conditions, the algorithm can converge to local extrema and improve saddle-point escape capability under irreversible mapping.