PageRank implementations produce unexpected results

[BamboWu]

I tried to print the top 3 vertex IDs with their rank value (p_ans in the most numa-PageRanks and p_global in numa-PageRankDelta).
Having run it on different graphs (road network TX, orkut social network), I found the results are different from Ligra's.
For example, on orkut, Polymer finds vertex 3036689, 2994689, 3065023 to be the top 3, with their ranks are about 5e-05.
While Ligra tells that vertex 3072587, 3072573, 3063686 are the top 3, with ranks around 1e-09.
The difference is not caused by hashing, because even I change the hash function to return the original vertex ID, I would end up with the same result.
I am confused because Polymer's PageRank implementation seems to follow the same update formula as Ligra uses, the only difference may because the edgeMapDenseForwardOTHER() function?
I will appreciate it if you can point out the reason why I would get different results.