Paper 2019/049

The Relationship between the Construction and Solution of the MILP Models and Applications

Lingchen Li, Wenling Wu, Yafei Zheng, and Lei Zhang

Abstract

The automatic search method based on Mix-integer Linear Programming (MILP) is one of the most common tools to search the distinguishers of block ciphers. For differential analysis, the byte-oriented MILP model is usually used to count the number of differential active s-boxes and the bit-oriented MILP model is used to search the optimal differential characteristic. In this paper, we present the influences between the construction and solution of MILP models solved by Gurobi : 1). the number of variables; 2). the number of constraints; 3). the order of the constraints; 4). the order of variables in constraints. We carefully construct the MILP models according to these influences in order to find the desired results in a reasonable time. As applications, we search the differential characteristic of PRESENT,GIFT-64 and GIFT-128 in the single-key setting. We do a dual processing for the constraints of the s-box. It only takes 298 seconds to finish the search of the 8-round optimal differential characteristic based on the new MILP model. We also obtain the optimal differential characteristic of the 9/10/11-round PRESENT. With a special initial constraint, it only takes 4 seconds to obtain a 9-round differential characteristic with probability $2^{-42}$. We also get a 12/13-round differential characteristic with probability $2^{-58}/2^{-62}$. For GIFT-128, we improve the probability of differential characteristic of $9 \sim 21$ rounds and give the first attack on 26-round GIFT-128 based on a 20-round differential characteristic with probability $2^{-121.415}$.