Computations of Nambu-Poisson cohomologies. (Q5945776)

Recall that a Poisson manifold \((M,\Pi)\) is an \(n\)-dimensional manifold \(M\) together with an antisymmetric 2-vector field \(\Pi\) such that its self Schouten bracket vanishes \([\Pi,\Pi]=0\). As a generalization, a Nambu-Poisson structure on an oriented manifold \((M,\omega)\) is an anti-symmetric \(k\)-vector field satisfying some condition. The author studies the special case of an \(n\)-vector \(\Lambda\) where this condition trivially holds. We will always suppose that \(\Lambda(x)\not=0\) on a dense open set. Poisson cohomology of \((M,\Pi)\) is just the cohomology of the graded Lie algebra \((\chi^{\bullet}(M),[,])\) of anti-symmetric multivector fields under the Schouten bracket with respect to the square zero element \(\Pi\), this formalism is due to Pierre Lecomte, see [\textit{C. Roger}, Algèbres de Lie graduées et quantification. Symplectic geometry and mathematical physics, Proc. Colloq., Aix-en- Provence/ Fr. 1990, Prog. Math. 99, 374-421 (1991; Zbl 0748.17028)]. Poisson cohomology has the reputation of being difficult to compute. In the \(n\)-vector case, there are at least two possibilities to generalize Poisson cohomology: the author defines for an \(n\)-vector \(\Lambda\) the cohomology groups \(H^{\bullet}_{\Lambda}(M)\), \({\bullet}=0,1,2\), as the cohomology of the complex \[ 0\to({\mathcal C}^{\infty}(M))^{n-1}\overset{\partial}{\to}\chi^1(M)\overset{\partial}{\to}\chi{^n(M)} 0 \] where \(\partial(g_1,\ldots,g_{n-1})=i_{dg_1\wedge\ldots\wedge dg_{n-1}}\Lambda\) and \(\partial(X)=[X,\Lambda]\) for \(g_i\in{\mathcal C}^{\infty}(M)\) and \(X\in\chi^1(M)\). This approach is an immediate generalization of the \(n=2\) dimensional case from the point of view of interpretation of the cohomology spaces, because \(H^2_{\Lambda}(M)\) gives the equivalence classes of infinitesimal deformations of \(\Lambda\) and \(H^1_{\Lambda}(M)\) is the quotient of the Lie algebra of \(\Lambda\)-preserving vector fields by the ideal of Hamiltonian vector fields. Another generalisation has been proposed by Ibáñez, de Léon, López, Marrero and Padrón. They define a cohomology for each Nambu-Poisson structure, defining for the special case of an \(n\)-vector a Lie algebra structure \([,]\) on \(\Omega^{n-1}(M)\) and an action of this Lie algebra on \({\mathcal C}^{\infty}(M)\), and setting then \(H_{\text{ NP}}^{\bullet}(M)\) to be the cohomology of the Lie algebra \((\Omega^{n-1}(M),[,])\) with values in \({\mathcal C}^{\infty}(M)\), using \({\mathcal C}^{\infty}(M)\)-linear cochains (thus close to the complex of differential forms). The key point of the article under review is that the author identifies \(H_{\text{ NP}}^{\bullet}(M)\) and \(H^{\bullet}_{\Lambda}(M)\) to the better accessible \(H_f^{\bullet}(M)\) and \(H_{f,p}^{\bullet}(M)\) which we will now describe: for \(f\in {\mathcal C}^{\infty}(M)\), define \[ d_f\,\,\,(\,\text{ resp.}\,\,\,d_f^{(p)}\,):\Omega^k(M)\to\Omega^{k+1}(M) \] by \[ d_f(\alpha)=fd\alpha-kdf\wedge\alpha\,\,\,(\,\text{ resp.}\,\,\,d_f^{(p)}(\alpha)=fd\alpha-(k-p)df\wedge\alpha\,\,\,\text{ for}\,\,\,p\in{\mathbb Z}\,) \] One has \(d_f^2=0\) (resp. \((d_f^{(p)})^2=0\)) and the corresponding cohomology spaces are denoted \(H_f^{\bullet}(M)\) (resp. \(H_{f,p}^{\bullet}(M)\)). Setting \(f=i_{\Lambda}\omega\), the author shows that \(H_{\text{ NP}}^{\bullet}(M)\cong H_f^{\bullet}(M)\), \(H^1_{\Lambda}(M)\cong H_{f,n-2}^{n-1}(M)\), and \(H^2_{\Lambda}(M)\cong H_{f,n-2}^{n}(M)\). Let us remark that the author studies in his thesis (which we recommend to the interested reader) \(H_f^{\bullet}(M)\) in much more detail, showing the existence of a long exact sequence in relative cohomology, excision and Mayer-Vietoris for \(H_f^{\bullet}(M)\) associated to an arbitrary \(f\in{\mathcal C}^{\infty}(M)\). The core of the article under review is devoted to the computation of \(H_f^{\bullet}(M)\) and \(H_{f,p}^{\bullet}(M)\) in a local setting, i.e. considering instead of spaces of objects globally defined on \(M\) only spaces of germs in \({\mathbb K}^n\) (\({\mathbb K=\mathbb R,\mathbb C}\)) of functions (\({\mathcal F}({\mathbb K}^n)\)), forms and multivectors. Under the assumption that \(f\) is a germ of a formal or analytic function in \({\mathbb K}^n\) with \(f(0)=0\) which is quasi-homogeneous of degree \(N\) (this is not too restrictive as all germs regular at 0 or with a singularity from Arnold's list are quasi-homogeneous) and of finite codimension (i.e. \(\text{ dim}\,\,(Q_f)<\infty\) where \(Q_f={\mathcal F}({\mathbb K}^n)\,/\,I_f\) and \(I_f\) is the ideal spanned by the partial derivatives of \(f\)), the author computes \(H_f^{\bullet}(M)\) entirely and some \(H_{f,p}^{\bullet}(M)\). It turns out that \(H_{\text{ NP}}^{k}(M)\) is possibly non zero only for \(k=0,1,n-1\) and \(n\). \(H_{\text{ NP}}^{0}(M)\) and \(H_{\text{ NP}}^{1}(M)\) are always 1-dimensional. On the other hand, the dimensions of \(H_{\text{ NP}}^{n}(M)\) and \(H_{\text{ NP}}^{n-1}(M)\) depend on the type of the singularity (via \(Q_f\)) and on the quasi-homogeneity type of \(f\). It turns out further that \(H^2_{\Lambda}(M)\) is finite dimensional with dependences on the properties of \(f\) like \(H_{\text{ NP}}^{n}(M)\). \(H^1_{\Lambda}(M)\) is not finite dimensional. Closing up, the author shows explicit examples for the cohomology spaces taking a germ of \(n\)-vector \(\Lambda\) which is either regular at 0 or has a singularity taken from Arnold's list.