Means in metric spaces and the center of mass (Q541263): Difference between revisions

Let \(\left( X,d\right) \) be a metric space. The length \(L(\gamma )\) of a curve \(\gamma :I\rightarrow X\) is defined as the supremum of \( \sum_{i=1}^{n}d\left( \gamma \left( t_{i-1}\right) ,\gamma \left( t_{i}\right) \right) \) where \(t_{0}\leq t_{1}\leq \cdots \leq t_{n}\in I.\) A curve \(\gamma :I\rightarrow X\) is called a geodesic if \(d\left( \gamma \left( s\right) ,\gamma \left( t\right) \right) =d\left( \gamma \left( s\right) ,\gamma \left( r\right) \right) +d\left( \gamma \left( r\right) ,\gamma \left( t\right) \right) \) for all \(s<r<t\in I.\) The metric space \(X\) is called a geodesic metric space if any two points \(x,y\) can be connected with a geodesic \(\gamma \) such that \(L(\gamma )=d(x,y).\) A geodesic metric space is called uniquely geodesic if any two points \(x,y\) can be connected by a unique geodesic \(\gamma _{x,y}\). It follows that we have unique metric midpoints, the midpoints of these geodesics, denoted as \(x\sharp y=\gamma _{x,y}(1/2)\). A geodesic metric space is \(k\)-convex if for any three points \( x,y,z,\) any geodesic \(\gamma :[0,1]\rightarrow X\) between \(x\) and \(y\) and for all \( t\in [0,1]\) we have \[ d(z,\gamma (t))^{2}\leq (1-t)d(z,x)^{2}+td(z,y)^{2}-\frac{k}{2} t(1-t)d(x,y)^{2}. \] A \(k\)-convex space is uniquely geodesic. The author gives an extension of midpoint maps as means between two points to several variables. Let \((X,d)\) be a complete \(k\)-convex geodesic metric space. Let \(Q_{1}^{0},\dots,Q_{n}^{0}\) be points in \(X\) and \(\pi =(\pi _{0},\pi _{1},\dots)\) be an infinite sequence of permutations of the letters \(\{1,\dots,n\}\). Let \[ Q_{i}^{l+1}=\begin{cases} Q_{\pi _{i}(i)}^{l}\sharp Q_{\pi _{i}(i+1)}^{l}\;\text{if }1\leq i<n \\ Q_{\pi _{i}(n)}^{l}\sharp Q_{\pi _{i}(1)}^{l}\qquad \text{else.} \end{cases} \] Then the sequences \(Q_{i}^{l}\) converge to a common limit point which is the centre of mass of the points \(Q_{1}^{0},\dots,Q_{n}^{0}\) defined by \(\underset{ x\in X}{\arg \min }\sum_{i=1}^{n}d(x,Q_{i}^{0})^{2},\) where \(\underset{x\in X }{\arg \min }C(x)\) denotes the unique point that minimizes the function \( C(x).\) The case of the symmetric space on the convex cone of positive definite matrices related to the geometric mean and the special orthogonal group are also studied as examples of \(k\)-convex metric spaces.