A characterization of Dobrushin's coefficient of ergodicity (Q915266): Difference between revisions

Let \(\| \cdot \|\) be an arbitrary norm on \({\mathbb{R}}^ n\) and \(S_ n(n\geq 2)\) the set of all (n\(\times n)\) stochastic matrices. The ergodicity coefficient of \(P\in S_ n\), with respect to the vector norm \(\| \cdot \|\), is defined by \[ \tau (P)=\sup \{\underset{\tilde{}} x'P;\quad \| \underset{\tilde{}} x\| =1,\quad \underset{\tilde{}} x'\underset{\tilde{}} 1=0\}. \] When the norm is \(\ell_ 1\), the resulting coefficient (the Dobrushin coefficient) has the property \(\tau\) (P)\(\leq 1\), \(P\in S_ n\). The authors show that this property holds only for the Dobrushin coefficient. The key result is that the following assertions (i) and (ii) are equivalent, where \(\lambda >0:\) (i) \(\| \underset{\tilde{}} x\|_ 1=\lambda \| \underset{\tilde{}} x\|,\) (ii) \(\lambda =\sup \{\| \underset{\tilde{}} x\|_ 1\); \(\underset{\tilde{}} x\in B\}\) and \(\| \underset{\tilde{}} e\| =1/\lambda\) (\(\underset{\tilde{}} e\in E XB_ 1).\) Here B and \(B_ 1\) are unit balls according to \(\| \cdot \|\) and \(\| \cdot \|_ 1\), and E XB\({}_ 1\) is the set of extremal points of \(B_ 1\).