Cohomologies on almost complex manifolds and the \(\partial \overline{\partial} \)-lemma (Q2286179)

The authors define and study three natural chain complexes associated to an almost complex manifold \((M,J)\), and their cohomology. The basic operator in the construction of the complexes is the \textit{algebraic derivation} \(\iota_K\) associated to a vector-valued \(k\) form \(K\) on \(M\), defined by sending a form \(\alpha\) to \(K^j \wedge (\iota_{e_j} \alpha)\), where \(K = K^j e_j\) in a local frame \(\{e_j\}\), and \(\iota_{e_j}\) is the usual interior product. One then forms the \textit{Nijenhuis-Lie derivation} \(\mathcal{L}_K\) by taking the graded commutator \([\iota_K, d]\) with the exterior derivative operator. On an almost complex manifold, we have two natural vector-valued forms: \(J\) itself, a vector-valued one-form, and its Nijenhuis tensor \(N\), a vector-valued two-form. One notes that \((\mathcal{L}_J)^2 = -\mathcal{L}_N\); by the Newlander-Nirenberg theorem, it follows that \(\mathcal{L}_J\) squares to zero if and only if \(J\) is induced by holomorphic charts. In general, neither \(\mathcal{L}_J\) nor \(\mathcal{L}_N\) square to zero. However, the authors notice that \([d, \mathcal{L}_J] = [d, \mathcal{L}_N] = [\mathcal{L}_J, \mathcal{L}_N]\), and hence any of the three operators \(d, \mathcal{L}_J, \mathcal{L}_N\) maps the kernel of any other of the three operators to itself. Hence we obtain three chain complexes: \begin{itemize} \item \((\ker \mathcal{L}_J, d)\), whose cohomology the authors call the \textit{\(J\)-cohomology} of \((M,J)\), \item \((\ker \mathcal{L}_N, d)\), whose cohomology is called the \textit{\(N\)-cohomology}, and \item \((\ker \mathcal{L}_N , \mathcal{L}_J)\), whose cohomology is called the \textit{\(J\)-twisted \(N\)-cohomology}. \end{itemize} If \(J\) is integrable, then clearly the \(N\)-cohomology reduces to the de Rham cohomology, and furthermore \(\mathcal{L}_J = -d^c = -J^{-1}dJ\), so the \(J\)-twisted \(N\)-cohomology is the \(d^c\)-cohomology, isomorphic to de Rham cohomology. These complexes are natural with respect to pseudoholomorphic maps of almost complex manifolds; hence these cohomologies give invariants of almost complex manifolds which are furthermore practically computable, as the authors go on to demonstrate. As an example of distinguishing diffeomorphic but not isomorphic almost complex manifolds, the authors employ the \(N\)-cohomology to distinguish several non-isomorphic almost complex structures on \(S^1 \times \mathbb{R}^3\). The remainder of the paper is spent studying the \(J\)-cohomology, which gives a new invariant even in the integrable case. Among other results, the authors prove the \(J\)-cohomology of a compact complex manifold is finite-dimensional, and that in the nonintegrable compact case, it is finite-dimensional in zeroth, first, and top degrees. The authors introduce a \(d\mathcal{L}_J\)-lemma for almost complex manifolds, which is equivalent to the much studied \(\partial \bar{\partial}\)-lemma (or equivalently the \(dd^c\)-lemma) in the integrable case. An almost complex manifold satisfies the \(d\mathcal{L}_J\)-lemma if and only if the natural map from \(J\)-cohomology to de Rham cohomology is an isomorphism. As an illustration of the applicability of this notion, the authors give a new proof that Hopf surfaces and the Iwasawa manifold do not satisfy the \(\partial \bar{\partial}\)-lemma (for Hopf surfaces, one would usually argue this by, say, observing that the \(\partial \bar{\partial}\)-lemma implies degeneration of the Hodge-de Rham spectral sequence, which further implies that the first Betti number must be even; for the Iwasawa manifold, one can show it admits a non-trivial triple Massey product, violating the rational homotopy theoretic formality that is guaranteed for \(\partial \bar{\partial}\)-manifolds by work of Deligne-Griffiths-Morgan-Sullivan). The authors compute that the rank of the first \(J\)-cohomology of Hopf surfaces and the Iwasawa manifold is strictly smaller than the corresponding first Betti number, thus showing the \(d\mathcal{L}_J\)-lemma is not satisfied. The authors then study in detail the \(J\)-cohomology of a family of almost complex structures on the four-dimensional torus, and end with some potential future directions for investigation.