Sharper Convergence Rates for Nonconvex Optimisation via Reduction Mappings