On coordination strategies for the interaction prediction principle using gradient and multiplier methods