On the Wasserstein distance between mutually singular measures (Q2174610)
From MaRDI portal
scientific article
Language | Label | Description | Also known as |
---|---|---|---|
English | On the Wasserstein distance between mutually singular measures |
scientific article |
Statements
On the Wasserstein distance between mutually singular measures (English)
0 references
21 April 2020
0 references
The authors investigate some minimization problems that arise from models for bilayer membranes. More precisely, if \(\mu\) is a given probability measure on \(\mathbb R^d\), the first problem under consideration is given by \[ W(\mu,A)=\min\{W(\mu,\nu) : \nu \in A\} \] where \(W\) is the \(p\)-Wasserstein distance, with \(p\in[1,+\infty)\), and \(A\) is family of probability measures singular w.r.t.\,\(\mu\). Some cases are treated: \[ A=\{\text{probability measures singular w.r.t. \(\mu\)}\} \Longrightarrow W(\mu,A)=0; \] \[ A=\{\text{probability measures singular w.r.t. \(\mu\) and w.r.t. \(\mathcal L^d\)}\}\cap L^1 \Longrightarrow W(\mu,A)=0; \] If \[ A=\{\text{probability measures singular w.r.t. \(\mu\)}\}\cap L^1 \] and \(\mu\) has a non-zero absolutely continuous part w.r.t.\,\(\mathcal L^d\) and \(W(\mu,A)\ne 0\), then \(W(\mu,A)=W(\mu,\nu)\) for a non-trivial measure \(\nu\). In this case, if \(\mu\in L^1\), then \[ \nu=(x-d(x)\nabla d(x))_\sharp \mu \] where \(d(x)=\mathrm{dist}(x,\partial S(\mu))\) and \(S(\mu)\) is a concentration set of \(\mu\). Next, the authors impose an upper bound on \(\mu\). Precisely, if \(\phi\) is a given non-negative integrable function with \(\int \phi(x)\,dx>1\) and \(\mathcal P_c\) is the set of all probability measures on \(\mathbb R^d\) with compact support, and \[ A_\phi=\{\rho \in \mathcal P_c \cap L^1 : \rho\perp \mu,\,\rho\le \phi\}, \] then \[ W(\mu,A_\phi)=W(\mu,\phi{\mathbf 1}_A) \] for a suitable set \(A\) (which has finite perimeter if \(\mu\) satisfies additional properties). Finally, the authors investigate a joint minimization problem: \[ \min_{A,B}\{P(B)+kW(A,B) : |A\cap B|=0,\,|A|=|B|=1\} \] where \(k>0\), \(W(A,B)=W({\mathbf 1}_A,{\mathbf 1}_B)\) and \(P\) denotes the perimeter. A solution \((A^*,B^*)\) of that problem exists and it is regular in the sense that \(A^*\) has finite perimeter and \(B^*\) is a quasi-minimizer of the perimeter. At the end of the paper, some numerical simulations are given when \(p=2\).
0 references
Wasserstein distance
0 references
singular measures
0 references
perimeter penalization
0 references
0 references
0 references