Generalized Orlicz spaces and Wasserstein distances for convex-concave scale functions (Q645950)

Let \(\vartheta\) be a function which has a convex-concave factorization \((\varphi,\psi)\), i.e., \(\varphi, \psi:\mathbb{R}_+ \rightarrow \mathbb{R}_+\) are strictly increasing continuous functions such that \(\varphi\) is convex, \(\psi\) is concave, \(\varphi(0)=\psi(0)=0\) and \(\vartheta=\varphi\circ\psi\). It is shown that the factorization can be chosen minimal, i.e., for any other factorization \((\widetilde{\varphi},\widetilde{\psi})\) the function \(\varphi^{-1}\circ\widetilde{\varphi}\) is convex. Let now \((\varphi,\psi)\) be a minimal convex-concave factorization of \(\vartheta\). Let \((X,\Sigma,\mu)\) be a \(\sigma\)-finite measure space and \(L^\vartheta\) the space of all measurable functions \(f:X\rightarrow \mathbb{R}\) such that \(\int\varphi(\frac{1}{t}\psi(|f|))d\mu<\infty\) for some \(t>0\). It is shown that \(d_\vartheta(f,g):=\inf\{t>0:\int\varphi(\frac{1}{t}\psi(|f-g|))d\mu\leq 1\}\) defines a metric on \(L^\vartheta\) and that \((L^\vartheta,d_\vartheta)\) is complete. Let \((X,d)\) be a complete separable metric space and \(\mathcal{P}_\vartheta\) the \(L^\vartheta\)-Wasserstein space of all probability Borel measures \(\mu\) on \(X\) such that \(\int\varphi(\frac{1}{t}\psi(d(x,y)))d\mu(x)<\infty\) for some \(y\in X\) and for some \(t>0\). For \(\mu,\nu\in\mathcal{P}_\vartheta\) let \(W_\vartheta(\mu,\nu):=\inf\{t>0: \inf_{q\in\Pi(\mu,\nu)}\int\varphi(\frac{1}{t}\psi(d(x,y)))dq(x,y)\leq 1\}\) where \(\Pi(\mu,\nu)\) is the set of all probability Borel measres \(q\) on \(X\times X\) such that \(q(A\times X)=\mu(A)\) and \(q(X\times A)=\nu(A)\). It is proved that \(W_\vartheta\) is a metric on \(\mathcal{P}_\vartheta\) and that \((\mathcal{P}_\vartheta,W_\vartheta)\) is complete.