Soul theorem

Soul theorem

Cheeger and Gromoll's soul theorem states:

If (M, g) is a complete connected Riemannian manifold with nonnegative sectional curvature, then there exists a closed totally convex, totally geodesic embedded submanifold whose normal bundle is diffeomorphic to M.

Such a submanifold is called a soul of (M, g). By the Gauss equation and total geodesicity, the induced Riemannian metric on the soul automatically has nonnegative sectional curvature. Gromoll and Meyer had earlier studied the case of positive sectional curvature, where they showed that a soul is given by a single point, and hence that M is diffeomorphic to Euclidean space.

Very simple examples, as below, show that the soul is not uniquely determined by (M, g) in general. However, Vladimir Sharafutdinov constructed a 1-Lipschitz retraction from M to any of its souls, thereby showing that any two souls are isometric. This mapping is known as the Sharafutdinov's retraction.

Cheeger and Gromoll also posed the converse question of whether there is a complete Riemannian metric of nonnegative sectional curvature on the total space of any vector bundle over a closed manifold of positive sectional curvature. The answer is now known to be negative, although the existence theory is not fully understood.

Examples.

• As can be directly seen from the definition, every compact manifold is its own soul. For this reason, the theorem is often stated only for non-compact manifolds.
• As a very simple example, take M to be Euclidean space Rⁿ. The sectional curvature is 0 everywhere, and any point of M can serve as a soul of M.
• Now take the paraboloid M = {(x, y, z) : z = x² + y²}, with the metric g being the ordinary Euclidean distance coming from the embedding of the paraboloid in Euclidean space R³. Here the sectional curvature is positive everywhere, though not constant. The origin (0, 0, 0) is a soul of M. Not every point x of M is a soul of M, since there may be geodesic loops based at x, in which case ${\displaystyle \{x\}}$ wouldn't be totally convex.
• One can also consider an infinite cylinder M = {(x, y, z) : x² + y² = 1}, again with the induced Euclidean metric. The sectional curvature is 0 everywhere. Any "horizontal" circle {(x, y, z) : x² + y² = 1} with fixed z is a soul of M. Non-horizontal cross sections of the cylinder are not souls since they are neither totally convex nor totally geodesic.