Sometimes what variables represent is too complicated to be expressed in a variable name. If you insist that all variables be semantically intelligible without reference to the part they play in the entire algorithm, you may be committing yourself to a less efficient algorithm that isn't any more flexible.

Is there something wrong with code that is too dense to be self-documenting, especially if you're willing to write and distribute a paper to explain it? Would you understand "pivot" in quicksort if you hadn't read a description of how quicksort works?