The complexity of antidifferentiation (Q1178488)

Denjoy showed how to reconstruct the primitive of a derivative using a process involving transfinite induction; and given any countable ordinal Denjoy gave an example for which the induction needed that ordinal. However it is possible that this is a property of the process rather than the problem; that is some reconstruction that does not need the totality of the countable ordinals may be possible. The main result of this paper shows that this is not the case. Let \(C^ \omega=C([0,1])^ \omega\) be the Polish space of sequences of real continuous functions in \([0,1]\); \(C([0,1])\) being given the uniform metric and \(C^ \omega\) the product topology. Let \(CN\) be the subset of \(C^ \omega\) consisting of everywhere convergent sequences; \(CN\) is not Borel, being co-analytic. As a result (i) there is no analytic set \(S\) such that if \(\tilde f\in CN\) then \(f\in S\) iff \(\lim\tilde f\) is a derivative; (ii) there is no Borel set \(B\subseteq C^ \omega\) such that for \(\tilde f\in CN\), \(f=\lim\tilde f\), \(f\) a derivative, \(f\in B\Leftrightarrow\int^ 1_ 0f>0\); i.e., the operation of antidifferentiation is not Borel.