They use them because they're more comprehensive, and scientists want to spend their time doing science, not improving free software.
Also, in the case of Mathematica, the interface and feature set are far beyond what any other tool offers (and can be expected to offer, given the track-record of free software).
Wasnt the whole point of science to build their own chain of verifiable tools to rely on, instead of having to trust secret, unverified commercial black boxes?
... No it wasn't and isn't, what gave you that idea?
Arguably, the rough idea is to discover and prove interesting or useful facts.
That's a bit different from Stallman's focus. Stallman is mainly interested in the freedom of users to modify the software they use (and share those modifications). But for scientific software the real issue is whether scientists can look at how it works. It would be fine for it to be under a "look but don't redistribute" proprietary license.
But other things aren't equal, and the productivity difference is order(s) of magniture more important than the transparency of an apparatus that has been successfully used by many others and is believed to give correct results.