Bounded and unbounded cohomology of homeomorphism and diffeomorphism groups (Q6040626)

Bounded group cohomology means the cohomology theory defined by admitting only bounded cochains in the usual definition of group cohomology via the bar resolution. In many cases bounded cohomology is either trivial (e.g., for amenable groups) or infinite-dimensional (e.g., for hyperbolic groups). So far, there seems to have been no example of groups with completely known nontrivial bounded cohomology. The paper under review proves that for the homeomorphism groups of the circle resp.\ the closed 2-disc, the bounded cohomology is generated (as a ring) by the Euler class, in particular the bounded cohomology groups are finite-dimensional and isomorphic to the ``unbounded'' cohomology. The proof adapts to groups of \(C^r\)-diffeomorphisms or PL-homeomorphisms to show that their bounded cohomology is still generated by the Euler class -- differently from the unbounded cohomology, where e.g.\ the Godbillon-Vey classes yield further nontrivial classes in the cohomology of diffeomorphism groups. The authors' approach to the bounded cohomology for homeomorphism groups of manifolds uses the following terminology. A fat point in an \(n\)-manifold \(M^n\) is, by definition, the germ at \(0\) of an orientation-preserving embedding \(f\colon B^n\to M^n\) of the open unit disc. The image of \(0\) is called the core of the fat point. For two fat points, \(x\perp y\) means that the cores are distinct. In general, a relation \(\perp\) on a set \(X\) is called generic if for any finite subset \(F\) there exists \(x\in X\) with \(x\perp y\) for all \(y\in F\). The tuples \((x_0,\ldots,x_n)\in X^{n+1}\) with \(x_i\perp x_j\) for all \(i<j\) form a semi-simplicial set \(X_*^\perp\). The authors prove that its bounded cohomology \(H_b^*(X_*^\perp)\) vanishes in all positive degrees if the relation \(\perp\) is generic. For the case of the fat points, the homeomorphism group \(G=Homeo_0(M)\) acts on \(X_*^\perp\). By a spectral sequence argument the authors prove an isomorphism \(H_b^p(G)\cong H_b^p(X_*^\perp/G)\) for \(0\le p<N\) if the following assumptions are satisfied for all \(p\ge 0,q>0\) with \(p+q<N\): the stabilizer of each element in \(X^\perp_p\) has vanishing \(H_b^q\), and the \(q\)-th vanishing moduli of all those stabilizers are uniformly bounded. Here the vanishing modulus of a group is the smallest norm of a cochain witnessing that a given cocycle of norm at most \(1\) (in the definition of bounded group cohomology) is a coboundary. The condition on uniform boundedness is trivially true if there are only finitely many isomorphism classes of stabilizers. For \(M=S^1\) and \(G=Homeo_0(S^1)\), the stabilizer of a fat point is isomorphic to \(Homeo_c(\mathbb{R})\), whose bounded cohomology is trivial by \textit{S. Matsumoto} and \textit{S. Morita} [Proc. Am. Math. Soc. 94, 539--544 (1985; Zbl 0536.57023)]. Similarly, the stabilizer of an element in \(X_p^\perp\) is isomorphic to a product of \(p+1\) copies of \(Homeo_c(\mathbb{R})\) and therefore has trivial bounded cohomology. Since there are only finitely many isomorphism classes of stabilizers for each \(p\), the above theorem can be applied to give an isomorphism \(H_b^*(G)\cong H_b^*(X_*^\perp/G)\). But \(X_*^\perp/G\) is -- for \(M=S^1\) -- the complex of cyclic orderings of finite sets, whose cohomology (bounded or unbounded) is generated by the Euler class by work of \textit{S. Jekel} [Adv. Math. 229, No. 3, 1949--1975 (2012; Zbl 1242.57021)]. The argument can be adapted to \(G=Diff^r(S^1)\) (and some other groups), where the authors use \(C^r\)-fat points defined as germs of \(C^r\)-embeddings and then need triviality of \(H_b^*(Diff_c^r(\mathbb{R}))\), which they prove by using a vast generalization of the Matsumoto-Morita argument due to the first author [Geom. Funct. Anal. 32, No. 3, 662--675 (2022; Zbl 1529.20077)]. Also the proof for the homeomorphism group of the 2-disc follows a similar strategy, but seems technically much harder. Here, the authors use notions of fat chords and strict transversality between them to construct several semi-simplicial sets \(X^\perp\) to which the before-mentioned theorem can be applied. Again the argument adapts to the \(C^r\)-case. The authors also have some results for homeomorphism groups of higher-dimensional spheres. Again they use the semi-simplicial set \(X_*^\perp\) of fat points with disjoint cores, which has trivial bounded cohomology. The problem is that the higher bounded cohomology groups of the stabilizers are in general not known and therefore the authors have to restrict to \(N=4\). So they prove \(H_b^2(Homeo_0(S^n))=0\) and \(H_b^3(Homeo_0(S^n))=0\) for all \(n\), and equally for the diffeomorphism groups. In the case of the \(3\)-sphere they can also handle the case \(N=5\) and prove vanishing of the relevant bounded cohomology groups of stabilizers to arrive at the results \(H_b^4(Homeo_0(S^3))=0\) and \(H_b^4(Diff_0^r(S^3))=0\) for \(r\not=4\). In particular, the Euler class and the first Pontryagin class are unbounded. This answers in the negative a question of E.\ Ghys in [\textit{R. Langevin}, Contemp. Math. 161, 59--80 (1994; Zbl 0844.57028)] about a possible generalization of the Milnor-Wood inequality from flat \(S^1\)-bundles to flat \(S^3\)-bundles.