Probabilistic automata complexity of languages depends on language structure and error probability

New lower bound of complexity for probabilistic automata with error bounded probability was proved. It depends on the language structure and on error probability of recognition. It is shown that for languages which are "rich" with formulated property a new lower bound of probabilistic complexity is more precise than that of Rabin's lower bound. In addition with the help of a new lower bound we improve the hierarchy of complexity of probabilistic automata depending on error probability previously shown by the author.