Minimal eigenvalue of large sparse matrices by an efficient reverse power-conjugate gradient scheme