Following Joe Polchinski’s Little Book of String, page 12, he use the sum $$1+2+3+...=-1/12$$ to find the zero point energy of the bosonic string (and later used the result to argue that we must have 26 dimensions).
When we use such a normalization for example in QFT it is said that the specific value that we obtain is not important and that we only care for the differences between the difference results. (For example sometime this difference does not depend on high energies beyond our theory).
But it seems to me that such argument can only work providing that we admit that our theory is not complete and that we cannot calculate some values using it.
This doesn't seems to go along with the fact that string theory try to be a complete theory.
Do I get something wrong about the using of such renormalization in string theory?
Is there a more clear method to argue about the number of dimension in string theory?