It's funny that you bring up mathematical notation because mathematical notation is completely optimized for write-only. Mathematics is so obsessed with one letter variable names that they had to gather letters from other alphabets to continue their opaque tradition.
Mathematical "code" isn't usually meant to be read without the surrounding variable and notation declarations, for example "Let x be ..." or "where • is the ... operation". If those are provided, short variable names and terse notation allow for the intent to be presented clearly, without it being obscured by line-length restrictions and a limited number of generic constructs (eg. for-loops instead of the summation operator).
That said, I'm finding this C implementation of J to be pretty cryptic. Of course, that's no reason to dismiss the J programming language itself.
Code like this needs to be read more slowly than usual, because unlike much C code, the information level isn't low enough that you can read it lines at a time. Math papers have all these weird sigma symbols and stuff, too; you can't read just them as fast as you would read "for (i=0; i<N; i++) {" for the millionth time.
Likewise, rather than writing "for (i=0; i<N; i++) {" yet again, K just calls that ' and moves on. In the source, there's a macro called DO, which is used extensively. DO(N, Blk) -> for(i=0, int _n=N; i<_n; i++) { Blk } . The APLs are about having a notation that doesn't constantly get in the way of thinking about your actual problem.