3
$\begingroup$

Let $M$ be a Riemannian manifold with distance function $d$, $C \subset M$ a geodesically convex set, $a=(a_i)_{i=1}^n \in C^n$, $W \in \mathbb{R}_{\geq 0}^{n \times n}$ and $J\colon C^n \rightarrow \mathbb{R}$ be defined by $$J(u):=\sum_{i=1}^n d(a_i,u_i)^2 +\sum_{i,j}^n W_{i,j} d(u_i,u_j)^2 .$$ Does $J$ poses a unique critical point? This problem interests me because of some applications in total variation regularization. There, I want to take $C=S^2_{\geq 0} \subset S^2$, the positive octant of the two dimensional sphere, for denoising of color images, or $M=SPD(3)$ for denoising of diffusion tensor images. For Euclidean spaces, one can use convexity arguments or directly study the linear equations $DJ/du_i=0$ to prove uniqueness of a critical point. For manifolds of nonpositive curvature, such as $SPD(n)$ one can as well use convexity arguments (the squared distance function is convex on manifolds of non-positive curvature). However, for manifolds of positive curvature, such as the sphere, the squared distance function is not convex. WLOG we can assume $W$ symmetric. For the sphere I could prove uniqueness if $$\sum_{j\neq i} W_{i,j} <\frac{2(\pi-2)}{\pi^2+4}$$ for all $i$ by defining a retraction. However I wonder if there is some way to extend the result for arbitrary large weights $W$.

$\endgroup$
0
$\begingroup$

Found a proof for the half sphere by showing that the functional is convex at every critical point and using poincare-hopf theorem.

| cite | improve this answer | |
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.