Speeding up MILP Aided Differential Characteristic Search with Matsui's Strategy

Being the first generic algorithm for finding the best differential and linear characteristics, Matsui’s branch and bound search algorithm (EUROCRYPT 1994) and its variants have played an important role in the security analysis of symmetric-key primitives. However, Matsui’s algorithm is difficult to implement, optimize, and be applied to different ciphers with reusable code. Another approach getting popular in recent years is to encode the search problem as a Mixed Integer Linear Programming (MILP) model which can be solved by open-source or commercially available optimizers. In this work, we show how to tweak the objective functions of the MILP models for finding differential characteristics such that a set of constraints derived from the bounding condition of Matsui’s algorithm can be incorporated into the models. We apply the new modeling strategy to PRESENT (S-box based SPN design), SIMON (Feistel structure), and SPECK (ARX construction). For PRESENT, the resolution time is significantly reduced. For example, the time to prove that the exact lower bound of the probabilities of the differential characteristics for 7-round PRESENT is reduced from 48638 s to 656 s. For SIMON, obvious acceleration is also observed, and for the ARX cipher SPECK, the new model is unable to speed up the resolution. In the future, it is interesting to investigate how to integrate other search heuristics proposed in the literature of symmetric-key cryptanalysis into the MILP models, and how to accelerate the resolution of MILP models for finding characteristics of ARX ciphers.