Phantom maps and the Gray index. (Q1426493)

Let \(X\) be a CW complex with skeleta \(X_n\) and let \(q_n: X\to X/X_n\) denote the quotient map. A phantom map \(f:X\to Y\) is a map with the property that, for each \(n\), there is \textit{at least one} map \(f_n:X/X_n\to Y\) such that \(f \simeq f_n \circ q_n\). It may or may not happen that \(f_n\) is itself a phantom map; the Gray index of \(f\) is the greatest integer \(n\) for which \(f_n\) can be chosen to be a phantom map. In the paper [Am. J. Math. 123, No. 4, 679--697 (2001; Zbl 0988.55009)], \textit{C. A. McGibbon} and the reviewer asked whether every essential phantom map between nilpotent finite type spaces has a finite Gray index. The paper under review has three main results. Theorem 1 is a new formula for the set of homotopy classes of phantom maps from one space to another; it is phrased in terms of homological algebra of inverse systems of groups. Theorem 2 characterizes phantom maps \(X\to Y\) with Gray index at most \(k\) in terms of the profinite completion \(\widehat Y\) of \(Y\). The third main result claims to answer the question referred to above: every essential phantom map between finite type spaces does have finite Gray index. Unfortunately, the proof of Theorem 3 appears to be incomplete in a fundamental way, as is described in some detail below. Let \(i:Y_\rho \to Y\) denote the inclusion of the homotopy fiber of \(Y\to \widehat Y\). Then \(f\) is a phantom map if and only if there is a map \(\bar f:X\to Y_\rho\) such that \(i\circ \bar f \simeq f\); the map \(\bar f\) is not generally unique. If \(\bar f\) is such a map, the let \(n(\bar f)\) be the least integer \(n\) for which \(\bar f| _{X_n}\not\simeq *\), or \(\infty\). According to Theorem 2, the Gray index of \(f\) is precisely \(\sup_{\bar f} \{ n(\bar f)\}\). The offered proof of Theorem 3 shows that \(n( \bar f) < \infty\) for each choice of \(\bar f\). The (apparent) gap in the proof is the following: there may be an infinite sequence of maps \(\bar f_k\) with \(n(\bar f_k) \to \infty\) as \(k\to \infty\). This possiblity is not addressed in the paper.