70

I stumbled across this question in a seminar-paper a long time ago:

Does there exist a positive integer $N$ such that if $G$ is a finite group with $\bigoplus_{i=1}^NH_i(G)=0$ then $G=\lbrace 1\rbrace$?

I believe this to still be an open problem. For $N=1$, any perfect group (ex: $A_5$) is a counterexample. For $N=2$, the binary icosahedral group $SL_2(F_5)$ suffices (perfect group with periodic Tate cohomology). And I found in one of Milgram's papers a result for $N=5$, the sporadic Mathieu group $M_{23}$. Note that this question is answered for infinite groups, because we can always construct a topological space (hence a $BG$ for some discrete group $G$) with prescribed homologies.

Is there another known group with a larger $N\ge 5$ before homology becomes nontrivial?
Are there any classifications of obstructions in higher homology groups?

[[Edit]]: Another view. A group is $\textit{acyclic}$ if it has trivial integral homology. There are no nontrivial finite acyclic groups. Indeed, a result of Richard Swan says that a group with $p$-torsion has nontrivial mod-$p$ cohomology in infinitely many dimensions, hence nontrivial integral homology.

Chris Gerig
  • 17,130
  • What do you mean about constructing a topological space with prescribed homologies, and then calling it 'a $BG$'. The homology of $BG$ can't be chosen or prescribed, it just is. And what, qualitatively, separates finite groups from infinite groups when constructing this so-called $BG$? Sorry if I'm being dense.. – David Roberts Jan 19 '11 at 22:38
  • 3
    @ David - For a connected CW complex $X$ the classifying space of the based loop space of $X$, i.e., $B\Omega X$ is homotopy equivalent to $X$. One can therefore think of $X$ as a classifying space of some H-space and I believe Milnor gave a model which realizes $\Omega X$ as a topological group. This is what the OP may be implying although I'm not so sure! – Somnath Basu Jan 19 '11 at 23:03
  • 18
    @David: Dan Kan and I once wrote a paper constructing, for any finite CW complex $X$, a space $Y = K(G,1)$, where $G$ is a finitely generated group, and a map $Y \rightarrow X$ which induces an isomorphism in homology over the fundamental group of $X$. So it can be done with homology of $G$ in the algebraic sense. But the construction actually depends on having a way to "cone off" the homology of any group within a bigger group. John Mather had proved that the group of compactly supported homeomorphisms of $\mathbb R^n$ is acyclic (in the algebraic sense)---same trick, I got it from that. – Bill Thurston Jan 19 '11 at 23:15
  • @Bill, ah, so in the question, the $G$ in $BG$ is not the $G$ in $H_i(G)$? That makes it clearer, even if not crystal. – David Roberts Jan 20 '11 at 00:23
  • @David: I think the point is that for every simply connected (to avoid dealing with the fundamental group) finite CW complex $X$ there exists a group $G$ such that $H_i(X) \cong H_i(BG)$ for every $i$. I have heard this result referred to as "morally, all homology is group homology." I think the point is that if I pick the homology groups I can construct a group $G$ s.t. the homology of the groups has the desired homology. (I hope I didn't screw this up) – Sean Tilson Jan 20 '11 at 01:19
  • (I should say that your choice of homology groups should not violate some sort of subtle homotopy theoretic condition that precludes it from being the homology of a finite CW complex.) – Sean Tilson Jan 20 '11 at 01:33
  • 7
    @David: for a discrete group, $BG = K(G,1)$ (as homotopy types), and the homology of the group is the same as the homology of this space. For a topological group, if $G_\delta$ denotes $G$ with discrete topology, then $K(G,1) = BG_\delta$, and often has quite different homology from $BG$. For instance $BSL_2(\mathbb R)$ is homotopy equivalent to $BSO_1$, which is $\mathbb{CP}^\infty$. But $BSL_2(\mathbb R)\delta$ has uncountably generated 2nd and 3rd homology. I guess even easier is $B\mathbb R$, which is trivial, vs $B\mathbb R\delta$, with homology rank $2^\omega$ in every dimension. – Bill Thurston Jan 20 '11 at 01:40
  • Arg, lost my comment. @Bill, Sean. Thanks. I was getting stuck on that $G$ (from which we get $BG$) is a priori a topological group, but now I read all the comments again.... – David Roberts Jan 20 '11 at 02:32
  • 2
    Theorem B in the paper On the integral homology of finitely-presented groups G. Baumslag, E. Dyer, and C. F. Miller Bull. Amer. Math. Soc. 4 (1981), 321-324. says roughly that for any recursively enumerable sequence of abelian groups $A_i$ with $A_1$, $A_2$ finitely generated, there exists a finitely presented group $G$ in which $H_n(G)=A_n$. More precisely, each $A_i$ must have a presentation of form $(X:R)_{ab}$, where $R$ is a free basis of the subgroup of FreeAb$(X)$ that it generates, the whole sequence of presentations must be r.e. – Derek Holt Jan 22 '11 at 13:15
  • 5
    If a finite group has vanishing homology in all degrees, it is trivial. There should be an earlier reference, but Quillen's ICM address on group cohomology announces that for element $g\in G$, $H^*(G)\to H^*(<g>)$ is nontrivial. I very much doubt that there is a uniform $n$ so that if a finite group has vanishing cohomology in degrees less than $n$, that it vanishes, but I can't give you any examples of groups with large vanishing ranges. – Ben Wieland Jan 26 '11 at 05:43
  • 8
    Just for the record: This question was posed ("an obvious question to ask") by A.Adem on page 810 of his 1997 paper here. – Misha Mar 16 '13 at 04:04

1 Answers1

26

Here is an approach to try to answer this question. To show the affirmative side [there is an N] look at the prime two and at swans argument that for a finite group with a two primary subgroup the cohomology mod two must be nonzero in infinitely many dimensions. I tried but didn't find it. if that results in a finite N with nonzero cohomology for all finte groups with a two primary part the problem is solved. because if the two primary component is absent, the group is solvable and N=1 results in non zero homology. On the other hand, to show the negative side, there is no such N. if swans argument does not yield such a concrete N for the prime two then one might well believe there is no such N, and that a string of examples might be constructed by thinking about the proof of swan's theorem.

Dennis Sullivan
  • 1,832
  • 21
  • 15