The geometry of optimal transportation

DOI10.1007/BF02392620MaRDI QIDQ1373007zbMATH OpenOpenAlexWikidataFDO

Authors Wilfrid Gangbo, Robert J. McCann

Publication date 5 November 1997

Published in Acta Mathematica (Search for Journal in Brave)

Full work available at URL https://doi.org/10.1007/bf02392620

cost functions Kantorovich's problem measure-preserving transformation minimum-cost transformations Monge's problem

Existence theories in calculus of variations and optimal control (49J99) Transportation, logistics and supply chain management (90B06) Trade models (91B60)

Review text

This paper treats a classical (1781) problem of Monge whose essence is: Let \((X,\mu)\) and \((Y,\nu)\) be (probability) measure spaces, where \(\mu\) represents the distribution of the production of some commodity at various ``localities \(x\in X\) and \(\nu\) of the consumption of that commodity at localities \(y\in Y\). There is a cost function \(c(x,y)\) on \(X\times Y\) which represents the cost of transporting the commodity from \(x\) to \(y\). A measure-preserving transformation \(s:X\to Y\) represents a ``distribution scheme for the commodity, resulting in a total transport cost of \(C_1(s)= \int c(x,s(x))d\mu(x)\). \textit{Monge's problem} is to find, where possible, the (or an) \(s\) which gives the minimum \(M_1\) of \(C_1(s)\) over all such choices of \(s\). \textit{Kantorovich's problem}, simpler than Monge's one, is to find, where possible, a measure \(\gamma\) on \(X\times Y\) which has \(\mu\) and \(\nu\) as marginals, and which gives the minimum \(M_2\) of the transport cost \(C_2(\gamma)= \int c(x,y)d\gamma(x, y)\) over all such choices of \(\gamma\). It is easy to see that \(M_2\leq M_1\), but there are cases where \(M_1= M_2\) and where \(s\) can be accessed through \(\gamma\). Sometimes \(s\) is even uniquely determined a.e. For their main (positive) results, the authors restrict themselves to \(X, Y\subseteq\mathbb{R}^d\), \(d= 1,2,3,\dots\), and to two classes of cost functions: (i) \(c(x,y)= h(x-y)\), where \(h\) is strictly convex (relatively easy); (ii) \(c(x,y)= l(|x-y|)\), where \(|\cdot|\) denotes Euclidean distance on \(\mathbb{R}^d\), and where \(l\geq 0\) is strictly concave (harder but more realistic in the economic context). They also give numerous examples where the (unique) map \(s\) can be explicitly constructed, and their treatment is largely self-contained.

Recommendations

Cites work

Cited in

(only showing first 100 items - show all)

This page was built for publication: The geometry of optimal transportation

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q1373007)