Take the pictured graph.

Now, look at how the two sides mirror each other from the "X" point. Think about it as if one side represents the past and one side represents the future. The limitation that I mentioned above speaks to the fact that there must be sufficient symmetry in the events occurring at a point in the past and its relative ("mirror") point in the future. As the scientists explain it in their paper:

"In fact, the entanglement can be completely destroyed by shifting one of the window functions sufficiently far away from its symmetrical location in time..."

"...Clearly, entanglement is maximized around the symmetrical point in time (corresponding to x = 0), while a sufficiently non-symmetrical choice for x can kill the extraction of timelike entanglement entirely."

Thus, while time travel would theoretically be possible under this scenario, there has to be a precise, symmetric interaction along that graph between the relevant "qubits" (points) existing in both the past (P) and relative its future (F).