Metaheuristic optimization algorithms have emerged as indispensable tools for solving complex mathematical problems that resist classical analytical and gradient-based methods. This paper presents a comprehensive systematic review of metaheuristic algorithms applied across the mathematical sciences, encompassing continuous function optimization, combinatorial optimization, constrained and multi-objective optimization, numerical methods enhancement, and integer programming. Following the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) framework, a total of 312 primary studies published between 2000 and 2025 were identified from five major databases, of which 187 met the inclusion criteria for detailed analysis. The review provides a rigorous examination of the theoretical foundations underlying metaheuristic convergence, including the No Free Lunch (NFL) Theorem and its mathematical implications. Twelve prominent algorithms Genetic Algorithm (GA), Particle Swarm Optimization (PSO), Differential Evolution (DE), Simulated Annealing (SA), Ant Colony Optimization (ACO), Artificial Bee Colony (ABC), Grey Wolf Optimizer (GWO), Whale Optimization Algorithm (WOA), Sine Cosine Algorithm (SCA), Moth-Flame Optimization (MFO), Harris Hawks Optimization (HHO), and Salp Swarm Algorithm (SSA) are evaluated on standard benchmark functions including Sphere, Rosenbrock, Rastrigin, Ackley, Schwefel, and Griewank in 30 dimensions. Performance comparisons on Traveling Salesman Problem (TSP) benchmark instances, Congress on Evolutionary Computation (CEC) constrained problems, and multi-objective test suites are presented with statistical significance testing via Friedman and Wilcoxon signed-rank tests. A bibliometric analysis of publication trends from 2000 to 2025 reveals accelerating growth in this field. The review identifies critical open challenges including theoretical convergence guarantees, scalability to ultra-high dimensions, and parameter sensitivity, while outlining future directions such as quantum-inspired metaheuristics, integration with deep reinforcement learning, and the pursuit of a theoretical unification framework.