Breaking substitution ciphers using a relaxation algorithm

Abstract

Substitution ciphers are codes in which each letter of the alphabet has one fixed substitute, and the word divisions do not change. In this paper the problem of breaking substitution ciphers is represented as a probabilistic labeling problem. Every code letter is assigned probabilities of representing plaintext letters. These probabilities are updated in parallel for all code letters, using joint letter probabilities. Iterating the updating scheme results in improved estimates that finally lead to breaking the cipher. The method is applied successfully to two examples.