The Lorentzian splitting theorem without the completeness assumption (Q1117481)

The purpose of the paper is the proof of the following Lorentzian analogue of the Cheeger-Gromoll splitting theorem: Let (M,g) be a connected globally hyperbolic Lorentzian space-time manifold with Ric(X,X)\(\geq 0\) for all timelike vectors X. If M contains a maximal timelike geodesic \(\gamma\) : \({\mathbb{R}}\to M\) (``line''), then M is isometric to (\({\mathbb{R}}\times S\), \(dt^ 2\oplus h)\), where (S,h) is a complete Riemannian manifold. As in the Riemannian case, the fibres \(\{\) \(t\}\) \(\times S\) of the splitting arise as the horospheres, i.e. the levels of the Busemann functions \(b^+\) and \(b^-\) corresponding to the two (future and past oriented) rays of \(\gamma\). A main step in the proof is to show that near \(\gamma\), the future and past horospheres are smooth and agree, i.e. \(b^-=-b^+\). To see this, one uses a small maximal spacelike hypersurface \(\Sigma\) with \(\partial \Sigma \subset \{b^-=0\}\) which exists by a theorem of \textit{R. Bartnik} [cf. Acta Math. 161, No.3/4, 145-181 (1988; see below)]. Due to the curvature assumption, the Busemann functions are subharmonic in a weak sense, so by some maximum principle, they are constant along \(\Sigma\). Previously, the splitting theorem was proved only under the crucial assumption of timelike geodesic completeness [cf. the reviewer, J. Differ. Geom. 27, No.3, 477-491 (1988; Zbl 0647.53043)]; this assumption now is avoided by introducing \(\Sigma\). Recently, \textit{R. P. Newman} has proved the splitting result also by assuming timelike geodesic completeness, but avoiding the assumption of global hyperbolicity [A proof of the splitting conjecture of S. T. Yau (preprint)].