In Riemannian geometry, the cut locus of a point in a manifold is roughly the set of all other points for which there are multiple minimizing geodesics connecting them from , but it may contain additional points where the minimizing geodesic is unique, under certain circumstances. The distance function from p is a smooth function except at the point p itself and the cut locus.
Fix a point in a complete Riemannian manifold , and consider the tangent space . It is a standard result that for sufficiently small in , the curve defined by the Riemannian exponential map, for belonging to the interval is a minimizing geodesic, and is the unique minimizing geodesic connecting the two endpoints. Here denotes the exponential map from . The cut locus of in the tangent space is defined to be the set of all vectors in such that is a minimizing geodesic for but fails to be minimizing for for any . The cut locus of in is defined to be image of the cut locus of in the tangent space under the exponential map at . Thus, we may interpret the cut locus of in as the points in the manifold where the geodesics starting at stop being minimizing.
The least distance from p to the cut locus is the injectivity radius at p. On the open ball of this radius, the exponential map at p is a diffeomorphism from the tangent space to the manifold, and this is the largest such radius. The global injectivity radius is defined to be the infimum of the injectivity radius at p, over all points of the manifold.
Suppose is in the cut locus of in . A standard result is that either (1) there is more than one minimizing geodesic joining to , or (2) and are conjugate along some geodesic which joins them. It is possible for both (1) and (2) to hold.
On the standard round n-sphere, the cut locus of a point consists of the single point opposite of it (i.e., the antipodal point). On an infinitely long cylinder, the cut locus of a point consists of the line opposite the point.
The significance of the cut locus is that the distance function from a point is smooth, except on the cut locus of and itself. In particular, it makes sense to take the gradient and Hessian of the distance function away from the cut locus and . This idea is used in the local Laplacian comparison theorem and the local Hessian comparison theorem. These are used in the proof of the local version of the Toponogov theorem, and many other important theorems in Riemannian geometry.
One can similarly define the cut locus of a submanifold of the Riemannian manifold, in terms of its normal exponential map.