Not everything in math is suitably expressed (or, perhaps, discussed) in an algorithmic form. Much (most?) of mathematics is operating at a level of abstraction that is far removed from an underlying machine model (in contrast to algorithms, like, well, the cryptographic algorithms you mention).

Take matrix multiplication, imagine seeing this:

  C = zero(m,p)
  for k from 1 to n
    for i from 1 to m
      for j from 1 to p
        C_ij += A_ik * B_kj

Now suppose that B is invertible and we actually want to know A in terms of B and C. Oops, we can't figure it out because we've tied ourselves to this algorithmic expression. In contrast to a more typical algebraic expression which describes not a computation but a relation:

  C = AB
  CB^-1 = ABB^-1
  CB^-1 = A

Perform that manipulation with the algorithmic description that helpfully obfuscates the relationship between the parts.

[And this is a small example, algorithmic expressions of algebraic ideas, like plain English expressions of the same, does not scale very well.]

Sure, as a physicist, I have written papers with pseudocode for algorithms, and I think we would do well to express algorithms in this way. That's definitely an improvement that is slowly percolating various fields. But most mathematical theorems and their proofs are not algorithms, and not suitable for expressing this way.

The reason for single-letter variable names is that mathematics is best learned by manipulating ideas by hand on paper and pen. And that many mathematical expressions are long and if we started using longer names, we would end up writing only a couple of statements per page, which would be much annoying than just using symbols.