\((\mathrm{BV},L^p)\)-decomposition, \(p = 1,2\), of functions in metric random walk spaces (Q2156037)

This paper studies the \((BV,L^p)\) decomposition for \(p=1,2\) of functions in a metric random walk space \([X,d,m]\), where \((X,d)\) is a metric space and \(m=(m_x)_{x\in X}\) is a set of probability measures encoding the jumps of a Markov chain. The \(m\)-ROF model in the space \([X,d,m]\) with invariant and reversible measure \(\nu\) is given by \(\min\big\{\frac12\int_X\int_X|u(y)-u(x)|dm_x(y)d\nu(x)+\frac{\lambda}{2} \int_X|u(x)-f(x)|^2d\nu(x):u\in L^2(X,\nu)\big\}\). The existence and uniqueness, and two characterizations of the minimizers of it are derived. The problem can be approximated by a gradient descent method and a rate of convergence relying on \(\lambda\) is given. Another problem studied is the associated \((BV,L^1)\) minimization in the metric random walk space given as \(\min\big\{\frac12\int_X\int_X|u(y)-u(x)|dm_x(y)d\nu(x)+\lambda \int_X|u(x)-f(x)|d\nu(x):u\in L^1(X,\nu)\big\}\). The geometric energy functional associated with this problem is studied. The existence of solutions to this problem is established. The minimizers of this problem for a general input image \(f\) in \(L^1(X,\nu)\) can be characterized by the minimizers of the associated geometric problem.