With all numpy floating-point errors ignored, compute and return np.ones(1) / 0 (which is array([inf])).