Abstract: Based on the analysis of Markov chain on mind evolution based machine learning（MEBML）,it is proved that the population generated by the similartaxis operation converges to the global optimum with probability 1 in discrete space.But because of the local property of the similartaxis operation, the transition probability from a local optimum to the global optimum is very small.To increase this transition probability,the dissimilation operation is introduced.Moreover,with the concepts of P optimal state and convergent region,theoretical and practical values of similartaxis and dissimilation operations are analyzed.