Is that a natural barrier for the current method or we could also have n(log n)^epsilon ?