Gradient-free distributed optimization with exact convergence