Submersions via projections (Q1282304)

Let \(TM = HM \oplus VM\) be an almost-product structure on the manifold \(M\), and let \({\mathcal H}, {\mathcal V}\) be the corresponding projections, i.e. \({\mathcal H} + {\mathcal V} =\text{ id}\), \({\mathcal H} \cdot {\mathcal V} = 0 = {\mathcal V} \cdot {\mathcal H}\). For any connection \(\nabla\) on \(M\), the author writes down the splitting of \(\nabla\) in terms of the derivative \(\nabla {\mathcal H} = - \nabla {\mathcal V}\) and the induced connections \(\nabla^H, \nabla^V\) for the subbundles. Namely, \(\nabla_X H = \nabla_X ^H H + \nabla_X {\mathcal H} \cdot H\), \(\nabla_X V = \nabla_X^V V - \nabla_X {\mathcal H} \cdot V\), for vector fields \(X\) (arbitrary), \(H\) (horizontal), \(V\) (vertical). Moreover, \(\nabla_Y {\mathcal H} \cdot {\mathcal V} = {\mathcal H} \cdot \nabla_Y {\mathcal H}\), \(\nabla_Y {\mathcal H} \cdot {\mathcal H} = {\mathcal V} \cdot \nabla_Y {\mathcal H}\), that is, \(\nabla_Y {\mathcal H}\) maps \(HM\) to \(VM\) and \(VM\) to \(HM\) for any vector field \(Y\). The identities above are simpler and more appealing than the corresponding properties of the O'Neill tensors [\textit{A. Gray}, J. Math. Mech. 16, 715-737 (1967; Zbl 0147.21201), and \textit{B. O'Neill}, Mich. Math. J. 13, 459-469 (1966; Zbl 0145.18602)]. The main simplification lies in the curvature computations, due to the following formulae: \(-{\mathcal H}\cdot \nabla_{X,Y}^2 {\mathcal H} \cdot H = (\nabla_X{\mathcal H} \cdot \nabla_Y{\mathcal H} + \nabla_Y{\mathcal H} \cdot \nabla_X {\mathcal H})\cdot H\), \(+{\mathcal V}\cdot \nabla_{X,Y}^2 {\mathcal H} \cdot V = (\nabla_X{\mathcal H} \cdot \nabla_Y{\mathcal H} + \nabla_Y{\mathcal H} \cdot \nabla_X {\mathcal H}) \cdot V\) and \({\mathcal V}\cdot R(X,Y)H = (\nabla_{X,Y}^2 {\mathcal H} - \nabla_{Y,X}^2 {\mathcal H})\cdot H\), \(-{\mathcal H}\cdot R(X,Y)V = (\nabla_{X,Y}^2 {\mathcal H} - \nabla_{Y,X}^2 {\mathcal H})\cdot V\). Finally, the integrability of \(VM\) is equivalent to the symmetry of \(\nabla {\mathcal H}\) on \(VM\), that is \(\nabla_U {\mathcal H}\cdot V = \nabla_V {\mathcal H} \cdot U\). As an application, the author obtains a nice direct proof of the Frobenius theorem.