PIAT This is the Proof and Tightness of Theorem 1 for paper "Perturbation-Invariant Adversarial Training for Neural Ranking Models: Improving the Effectiveness-Robustness Trade-Off".