Generalised Pattern Search Based on Covariance Matrix Diagonalisation

Abstract Pattern Search is a family of gradient-free direct search methods for numerical optimisation problems. The characterising feature pattern the use multiple directions spanning problem domain to sample new candidate solutions. These compose matrix potential moves, that pattern. Although some fundamental studies theoretically indicate various can be used, selection remains an unaddressed problem. present article proposes procedure selecting guarantee high convergence/high performance search. proposed consists fitness landscape analysis characterise geometry by sampling points and those whose objective function values are below threshold. eigenvectors covariance this distribution then used as Numerical results show method systematically outperforms its standard counterpart competitive with modern complex metaheuristic methods.