Curvature of differentiable Hilbert modules and Kasparov modules (Q2136540)

In this paper, the notion of curvature of densely defined universal connections on Hilbert \(C^*\)-modules relative to a spectral triple (or unbounded Kasparov module) is defined, obtaining a well-defined curvature operator. For a fixed spectral triple, modulo ``junk forms'', the curvature only depends on the represented form of the universal connection. This notion of curvature is further releated to factorizations of unbounded Kasparov modules, recovering curvature data explored in previous works on Riemannian submersions of compact manifolds [\textit{J. Kaad} and \textit{W. D. van Suijlekom}, J. Noncommut. Geom. 12, No. 3, 1133--1159 (2018; Zbl 1405.19002)]. Numerous other approaches to curvature have appeared independently in many works (for example, [\textit{J. Arnlind} and \textit{M. Wilson}, J. Noncommut. Geom. 11, No. 2, 507--536 (2017; Zbl 1378.46053); \textit{J. Bhowmick} et al., Int. J. Math. 32, No. 13, Article ID 2150088, 45 p. (2021; Zbl 1491.58004); \textit{M. Dubois-Violette} et al., J. Math. Phys. 37, No. 8, 4089--4102 (1996; Zbl 0873.58006)]). The paper focuses on assuming minimal conditions for defining curvature, which involves several technical results controlling the domain of unbounded operators and various analytical bounds. Below we summatize the main definitions, with a few simplifications for brevity (one can work with less stringent conditions). Recall that an \textit{unbounded Kasparov module} is a triple \((\mathcal{B},Y,T)\), where \(\mathcal{B}\) is a complex \(\ast\)-algebra faithfully represented on a \(\mathbb{Z}/2\mathbb{Z}\)-graded Hilbert module \(Y\), together with an odd self-adjoint regular operator \(T\) with locally compact resolvent, satisfying a certain ``\(C^1\)-condition'' (roughly speaking, operators in \([T,\mathcal{B}]\) need to extend to bounded operators). Inspired by this, we can require a \(C^2\)-condition involving the commutator of \(T^2\) and elements \(b\in\mathcal{B}\). These conditions allow us to define completions \(\mathcal{B}_i\) (\(i=1,2\)) of \(\mathcal{B}\), and spaces of \emph{universal one-forms} as kernel of multiplications maps \(\mathcal{B}_1\otimes \mathcal{B}_2\to \mathcal{B}_1\) (it is essential here that the Haagerup tensor product is used). Universal \(2\)-forms are defined analogously, for instance \(\Omega^2_u(B,\mathcal{B}_2)=\Omega^1_u(B,\mathcal{B}_1)\otimes\Omega^1_u(\mathcal{B}_1,\mathcal{B}_2)\). There are ``represented'' versions of these spaces (denoted \(\Omega^i_T\)), obtained by considering the maps \(a\otimes b\mapsto a[T,b]\), \(a\otimes b\mapsto a[T^2,b]\). However, it is important to note that typically there is no compatible differential going between \emph{represented} forms, due to the existence of \(1\)-forms \(\omega\) such that the candidate \(d\omega\) is nonzero with \(\omega\) itself being zero. The definition of \emph{junk forms} captures this phenomenon. If \(X\) is a (graded) \(C^*\)-module over \(B\), we say that a dense \(\mathcal{B}_i\)-submodule \(\mathcal{X}\) is horizontally \(C^i\)-differentiable if it has \(\mathcal{B}_i\)-valued inner products. A universal connection is then a linear map \(\nabla\colon \mathcal{X}\to X\otimes_B \Omega^1(B,\mathcal{B}_1)\) satisfying the Leibniz rule and compatible with the inner product. Associated to \(\nabla\) is its represented version \(\nabla_T\), and a natural completion of \(\mathcal{X}\) denoted \(\mathcal{X}_{\nabla_T}\). Then \[ 1\otimes_\nabla T(x\otimes y) = \gamma(x)\otimes Ty+\nabla_T(x)y \] is a densely-defined odd symmetric operator on \(X\otimes_B Y\) (above, \(\gamma\) is the grading operator). A \(C^2\)-module for the \(C^2\)-Kasparov module \((\mathcal{B},Y,T)\) is a pair \((\mathcal{X},S)\) where \(\mathcal{X}\) is horizontally \(C^2\)-differentiable, and \(S\) is an essentially self-adjoint and regular operator, with \(\mathcal{X}\subseteq \mathrm{Dom}\, S^2\). We can think of \(S\) as defining a \emph{vertical} differentiable structure on \(X\). A \(C^2\)-connection for such module is a pair of connections \((\nabla,\nabla^1)\), with \(\nabla\) as above and \(\nabla^1\colon \mathcal{X}\to \mathcal{X}_{\nabla_T}\otimes_{\mathcal{B}_1} \Omega^1_u(\mathcal{B}_1,\mathcal{B}_2)\), satisfying: \begin{itemize} \item \(\nabla\) and \(\nabla^1\) coincide up to the natural inclusion \( \mathcal{X}_{\nabla_T}\hookrightarrow \mathcal{X}\); \item \((S\otimes 1, 1\otimes_\nabla T)\) form a ''vertically anticommuting pair''. This is a compatibility condition between the vertical structure defined by \(S\) and the horizontal structure associated to \(T\). It has the effect that the sum of the operators is self-adjoint and regular on the natural intersection domain. It is also a sufficient condition for the sum to represent the Kasparov product in \(KK\)-theory. Variants of this notion appeared previously in [\textit{J. Kaad} and \textit{M. Lesch}, J. Funct. Anal. 262, No. 10, 4540--4569 (2012; Zbl 1251.46030)] and [\textit{M. Lesch} and \textit{B. Mesland}, J. Math. Anal. Appl. 472, No. 1, 947--980 (2019; Zbl 07056463)]. \end{itemize} Finally we can define the \textbf{curvature} of a \(C^2\)-connection as the symmetric operator \(R_{\nabla_T}:= (1\otimes_\nabla T)^2-1\otimes_\nabla T^2\). The paper goes on to describe how simpler situations, like the case of Fréchet modules or finitely generated projective modules, can be incorporated in the \(C^2\)-setting outlined above. Finally, it is shown that the case of Riemannian submersions nicely fits this framework, with the curvature operator encoding information of the second fundamental form of the submersion, the mean curvature associated to it, and the curvature \(3\)-form of the fibration \(M\to B\). It is interesting to note that the latter appears as an obstruction to the realization of the Dirac operator \(D_M\) as an unbounded (internal) Kasparov product between a ``vertical Dirac family'' and the Dirac operator on the base.