Varieties for modules and a problem of Steenrod (Q1822088)

Let G be a finite group and R a commutative ring. A 1-connected G-space X is said to realize an RG-module M (in dimension \(n>0)\) if \(\tilde H_ n(X;R)=M\) and \(\tilde H_ i(X;R)=0\), \(i\neq n\). Steenrod's problem asks whether every M is so realizable. Early work on this problem, by R. Swan and J. Arnold, concentrated on an additional finiteness question but also gave an affirmative answer for G cyclic primary. Since 1980, with work of G. Carlsson, P. Kahn, and J. Smith, the emphasis has been negative: various examples of nonrealizable M have been produced. An exception has been P. Vogel's theorem which extends the Swan-Arnold result to all G of squarefree order. The main contribution of the present paper is to provide a criterion whereby large numbers of nonrealizable M may be constructed. Furthermore, the criterion allows us for the first time to obtain a glimpse of the general, global situation (as opposed to a more narrow focus on this or that example or computation). For example, it leads to the observation (not at all obvious from definitions or from earlier work) that nonrealizable modules are in general more plentiful than realizable ones. To describe (a weak version of) the criterion, choose a field k of characteristic \(p>0\) and let \(X_ G\) be the affine variety \(Spec(H^.(G,k))\), where \(\cdot\) ranges over all indices \(\geq 0\) when \(p=2\) and only the even ones when \(p>2\). If \(H\leq G\), the restriction \(H^.(G,k)\to H^.(H,k)\) induces a map \(t_{H,G}: X_ H\to X_ G\). The authors associate with every kG-module V a certain ideal \(I(V)\subseteq H^.(G,k)\) and then define the variety \(X_ G(V)\subseteq X_ G\) to be \(Spec(H^.(G,k)/I(V)).\) Now let \(I\subseteq R\) be a maximal ideal, and let k be a field of characteristic \(p>0\) containing R/I. Suppose that V is a realizable RG- module. The authors then show that \(X_ G(V \otimes_ R k)\) is of the form \(\cup_{E}t_{E,G}(X_ E)\), where E ranges over some collection of elementary abelian subgroups of G. The promised (weak) nonrealizability criterion for an RG-module V, then, is simply that \(X_ G(V \otimes_ R k)\) not has the very special form above. To make this criterion useful for generating nonrealizable modules, the authors restrict to the case \(R={\mathbb{Z}}\) and show that any homogeneous subvariety \(X\subseteq X_ G\) which is closed under the Frobenius map is of the form \(X_ G(V \otimes_{{\mathbb{Z}}} k)\), for some \({\mathbb{Z}}\)-free \({\mathbb{Z}}G\)-module V. Thus, nonrealizable modules are (at least) as numerous as Frobenius-closed homogeneous subvarieties of \(X_ G\) not of the above form. An additional useful contribution of the paper is a sequence of results which show that the realizability of a module depends only on its cohomological equivalence class.