A space lower bound of two-dimensional probabilistic turing machines

This paper shows a sublogarithmic space lower bound for two-dimensional probabilistic Turing machines (2-ptm's) over square tapes with bounded error, and shows, using this space lower bound theorem, that a specific set is not recognized by any o(log n) space-bounded 2-ptm. Furthermore, the paper investigates a relationship between 2-ptm's and two-dimensional Turing machines with both nondeterministic and probabilistic states, which we call "two-dimensional stochastic Turing machines (2-stm's)", and shows that for any loglog n less than or equal to L(n) = o(log n), L(n) space-bounded 2-ptm's with bounded error are less powerful than L(n) space-bounded 2-stm's with bounded error which start in nondeterministic mode, and make only one alternation between nondeterministic and probabilistic modes.