Mathematische Zeitschrift

, Volume 268, Issue 1, pp 371–407

A proof of Sudakov theorem with strictly convex norms


DOI: 10.1007/s00209-010-0677-6

Cite this article as:
Caravenna, L. Math. Z. (2011) 268: 371. doi:10.1007/s00209-010-0677-6


The paper establishes a solution to the Monge problem in \({\mathbb {R}^n}\) for a possibly asymmetric norm cost function and absolutely continuous initial measures, under the assumption that the unit ball is strictly convex—but not necessarily differentiable nor uniformly convex. The proof follows the strategy initially proposed by Sudakov in 1976, found to be incomplete in 2000; the missing step is fixed in the above case adapting a disintegration technique introduced for a variational problem. By strict convexity, mass moves along rays, and we also investigate the divergence of the vector field of rays.

List of symbols


The identity function, Id(x) = x

The function vanishing out of S, equal to one on S (where \({S\subset \mathbb {R}^n}\))

⦇ a, b ⦈

The segment in \({\mathbb {R}^n}\) from a to b, without the endpoints

\({[\kern-0.15em[ a, b ]\kern-0.15em]}\)

The segment in \({\mathbb {R}^n}\) from a to b, including the endpoints


The symmetric difference between two sets


{e1, . . . , en} fixed orthonormal basis of \({\mathbb {R}^n}\)

| · |

The Euclidean norm of a vector

| · |

The maximum of the component of a vector

\({\mathcal {B}(X)}\)

The Borel subsets of \({X\subset\mathbb {R}^n}\)

\({\mathcal {H}^{\alpha}}\)

The α-dimensional Hausdorff measure in \({\mathbb {R}^n}\)

\({\fancyscript {L}^n}\)

The Lebesgue measure on \({\mathbb {R}^n}\)


Denotes that a measure is absolutely continuous w.r.t. another one


Denotes the linear span


\({\langle\theta,\varphi\rangle=\int \varphi d\theta}\), where θ is a measure and \({\varphi}\) is θ-integrable


The push forward with a measurable map τ, see Appendix A

Π(μ, ν)

The set of transport plans between two probability measures μ and ν

\({f \upharpoonright {S}}\)

The restriction of the function f to a set S

\({\theta \llcorner {S}}\)

For Aθ-measurable, \({\theta \llcorner {S}(A)=\theta(A \cap S)}\),where S is θ-measurable

\({\int\theta_{s}\, m}\)

\({\theta=\int\theta_{s}\, dm(s)}\) denotes the disintegration of a measure θ, see Appendix A


Probability measures on \({\mathbb {R}^n, \mu\ll\fancyscript {L}^n}\)


A possibly asymmetric norm on \({\mathbb {R}^n}\) whose unit ball is strictly convex


The cost function \({c(x,y)=\|y-x\|}\)


The unit ball \({\{x\in\mathbb {R}^n:\|x\|\leq 1\}}\)


The dual convex set of \({D^{*}\,:\,D=\{\ell\,:\,\ell\cdot v\le 1\quad \forall v {\in} ^{*}\}}\)


The boundary of D


\({\delta D^*(\ell)=\{v: (\|\ell\|v)\in\partial D}\) and v· ℓ = 1}


See Definition 2.1 and (3.1)

\({\mathcal{T}, \mathcal{T}_{\rule[.2pt]{.2pt}{3pt}\!\rm e}}\)

See Definition 2.2


The \({c}\)-subdifferential of \({\phi\,:\, \partial_{c}\phi =\Big\{(x,y):\phi x)-\phi(y)= c(x,y)\Big\}}\)


The sub-differential of a function \({\phi:\mathbb {R}^n\mapsto\mathbb {R}: \quad \partial^{-}\phi(x)=\Big\{ v^{*}:\phi(y)\geq \phi(x) +v^{*}\cdot(y-x) \forall y \Big\} }\)


See Definition 2.11, and also P. 15 for \({\sigma^{\cdot}_{d_{I}}(\cdot)}\)

\({\mathcal {Z}, Z}\)

Sheaf set and its basis, see Definition 2.9

\({\mathcal {K}}\)

d-cylinder, see Definition 2.11

\({\mathcal {P}, \mathcal {R}}\)

See Definition 2.3

\({\mathcal{D}, d}\)

Directions of the rays, see (2.4) and (2.7)

\({\mathcal {S}}\)

See Theorems 2.25, 3.2


See Lemma 2.21


See Corollary 2.23, Lemmata 2.21, 2.24

c(t, z)

See Theorem 2.25, Lemma 2.30, (2.23)

Copyright information

© Springer-Verlag 2010

Authors and Affiliations

  1. 1.SISSATriesteItaly