Convergence of Riemannian manifolds; Ricci and \(L^{n/2}\)-curvature pinching (Q1173855)

The well known compactness theorem of M. Gromov states, that the class of all closed \(C^{1,\alpha}\) Riemannian \(n\)-manifolds \(M\) \((0<\alpha<1)\) with (a) \(| \text{curv}|\leq K\), (b) \(\text{diam}\leq D\), (c) \(\text{Vol}\geq V\) is precompact, i.e. any sequence of such metrics has a subsequence, which is converging in the \(C^{1,\alpha}\)-sense. In the present paper, (a) is replaced by the weaker curvature conditions (\(a_ 1\)) \(|\text{Ric}|\leq H\), (\(a_ 2\)) \(\int| R|^{n/2}\leq K\), where \(R\) denotes the Riemannian curvature tensor and Ric the Ricci curvature. If further (c) is replaced by an injectivity radius estimate (c\('\)) \(\text{inj}\geq i_ 0\), it can be shown (Thm. 0.1), that such a class contains only finitely many diffeomorphism types, and on a fixed diffeomorphism class, any sequence of such metrics contains a subsequence, which is \(C^{1,\alpha}\)-convergent outside a finite number of points. In dimension 4, due to Gauß-Bonnet, (\(a_ 2\)) may be replaced by an upper bound for the Euler number. Moreover, consider the slightly stronger curvature assumption (\(a_ 2'\)) \(\int_{B(x,r)}| R|^{n/2}\leq\kappa\) for some \(r>0\) and any \(x\in M\), where \(\kappa > 0\) is a very small constant. Then (\(a_ 2'\)), (b) and (c) imply full \(C^{1,\alpha}\)-convergence on the whole manifold, cf. Thm. 0.8. (There are several misprints in the statement; compare pp. 351 and 360). From these theorems, various pinching results follow, e.g. Thm. 0.3: If a manifold satisfies (\(a_ 2\)), (\(b\)), (\(c'\)) and \(| \text{Ric}- 1|<\varepsilon\) for some \(\varepsilon > 0\) depending on \(n\), \(D\), \(K\), \(i_ 0\), then there exists an Einstein metric with \(\text{Ric}=1\) on \(M\). The main technical tool in the paper is an existence theorem for harmonic coordinates with good estimates on manifolds satisfying \((a_ 2')\) (Thm. 3.0). This is used to prove Thm. 0.1 as follows: One chooses a minimal covering of \(M\) by balls of fixed radius \(r<\text{inj}\). The number \(Q\) of these balls is bounded above by a constant depending only on \(r\), due to the lower Ricci curvature bound (Gromov-Bishop inequality). Moreover, the number \(h\) of balls with a common intersection has an upper bound, which is independent of \(r\) for small \(r\); apply Gromov-Bishop to balls of radius \(2r\) and \(r\). Due to (\(a_ 2\)) and since each ball overlaps less than \(h\) others, (\(a_ 2'\)) fails only on a number \(N\) of balls, where \(N\) is bounded independently of \(M\) and \(r\). Outside these ``bad'' balls, good harmonic coordinates exist (Thm. 3.0) and we get convergence. Since \(r\) can be made as small as we want, we get convergence outside \(N\) points. To prove Thm. 0.8, one deforms the given metric \(g\) by the evolution equation \[ \partial g(t)/\partial t=-2\text{ Ric}(g(t)),\qquad g(0)=g. \] Since \(g(t)\) is close to \(g\), one can use estimates for \(g(t)\) in order to get uniform harmonic coordinates at every point, which implies \(C^{1,\alpha}\)-convergence.