Semi-bounded probabilistic polynomial-time, is it equal to BPP?

Question

The complexity class $\mathsf{BPP}$ is typically defined as the class of all problems for which:

1. Running an algorithm once takes polynomial time at most.
2. The answer is correct with the probability at least $2/3$.

The complexity class $\mathsf{PP}$, on the other hand, has $1/2$ instead of $2/3$ in the second constraint. Now, we can create a class that lies between them:

1. Running an algorithm once takes polynomial time at most.
2. The answer "YES" is correct with the probability at least $2/3$.
3. The answer "NO" is correct with the probability at least $1/2$.

Is this complexity class equivalent to $\mathsf{BPP}$? If this is not known, is it in $\mathsf P/poly$ at least?

Answer 1

Your class is the same as BPP. If you run your algorithm $n$ times, then in the YES case, the expected number of YES answers is at least $(2/3)n$, while in the NO case, the expected number of YES answers is at most $(1/2)n$. Moreover, both of these values are concentrated by Chernoff bounds. Therefore if you answer YES if you see at least $0.6n$ many YES answers, then you get the correct answer with exponentially small error probability.