From the article, she says they tried "1.0 / (1 << mj)", which turned out to be slower than the lookup table. May or may not be equivalent to java.lang.Math.scalb, I dunno.
I'm confused by the discussion of multi-lingual corpora. Is it common in topic modeling to consider documents drawn from disjoint vocabularies, or is it just a kind of thought experiment?
I love project euler, but I've come to the realization that its purpose is to beat programmers soundly about the head and neck with a big math stick. At work last week, we were working on project euler at lunch, and had the one CS PhD in our midst not jumped up and explained the chinese remainder theorem to us, we wouldn't have had a chance.