A SELF-ADAPTIVE MULTIVERSAL METAHEURISTIC BASED ON TEMPORAL PARADOXES

A continuous metaheuristic based on a temporal multiverse scheme was proposed that captures, in a single evolutionary framework, mechanisms of aggressive exploitation, numerical refinement, and preservation of viability. The method operates with three simultaneous universes: (i) the Information Horizon, which acts as an explorer; Its logic mimics the evaporation of a black hole, exchanging subsets of variables according to population "temperature", favoring pronounced leaps to distant regions without sacrificing diversity; (ii) the Zeno-Quantum universe with Predestination, which plays the role of a local explorer, subdividing each displacement into geometric sections until it finds an acceptable improvement, while an adaptive attraction β pushes solutions to a focal point predicted in-line by light regression, providing fine-tuning even in non-differentiable functions; (iii) the Novikov universe with Bootstrap, which is responsible for global coherence, mitigating any backsliding that exceeds a δ threshold and storing significant advances as "seeds" that can be reinjected with φ probability, avoiding the loss of valuable discoveries and facilitating controlled escapes from local optimums.

These universes share only seven hyperparameters: total population size N, multiversal exchange interval L, initial noise amplitude σ₀, global exploration factor α, attraction strength β, initial entropic budget E₀, and restart probability φ. All secondary parameters (cut depth, memory exclusion rates, bifurcation cycles, etc.) are calculated internally through adaptive rules that depend on problem dimensionality and the population’s instantaneous dispersion, drastically reducing the calibration effort.

This structured combination of paradoxes is expected to produce search trajectories capable of dynamically balancing global and local exploration, reaching high-quality regions in fewer iterations than traditional mono-paradigm schemes. The approach was designed for high-dimensional continuous optimization problems. By combining mechanisms from chaos theory, time dilation, and self-consistency in an architecture with low tuning burden, the proposal aims to offer researchers and practitioners a versatile, reproducible, and easily integrable tool for scientific or industrial workflows that require high-quality numerical solutions under computational time constraints.