Measure transport via pseudo-cones
Rolf Schneider · 2025
Read Measure transport via pseudo-cones by Rolf Schneider (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2503-10449.txt.
Introduction
Rolf Schneider gives a new proof for a measure transport problem in this short 2025 math paper. The problem is the Gauss image problem for pseudo-cones, a question about moving a measure on a sphere from one pattern to another. Schneider solves it with a special case of Kantorovich duality, a classic tool for transport problems. The note is brief and proof-focused, aimed at readers in geometry who know the Gauss image setting.
Catalog record
| CREATOR | Rolf Schneider |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 17 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-10449
Browse the full library · About the Tower of Babel
Full text
The opening of the work, from the library's hosted text:
Measure transport via pseudo-cones Rolf Schneider Abstract For the solution of the Gauss image problem for pseudo-cones, which can be considered as a measure transport problem for certain measures on the sphere, we give a new proof, using a special case of Kantorovich duality. Keywords: pseudo-cone, Gauss image problem, measure transport, reverse radial Gauss map, Kantorovich duality 2020 Mathematics Subject Classification: 52A20 49Q22 1 Introduction Let C\subset{\mathbb{R}}^{n} be a closed convex cone, pointed and with interior points. A C -pseudo-cone is a nonempty closed convex set K\subset C not containing the origin and satisfying K+C=K . This implies that C is the recession cone of K and that \lambda K\subseteq K for \lambda\geq 1 . The set of all C -pseudo-cones in {\mathbb{R}}^{n} is denoted by ps(C) . Denoting by {\mathbb{S}}^{n-1} the unit sphere of {\mathbb{R}}^{n} , we define the sets \Omega_{C}:={\mathbb{S}}^{n-1}\cap{\rm int}\,C and \Omega_{C^{\circ}}:={\mathbb{S}}^{n-1}\cap{\rm int}\,C^{\circ} , where C^{\circ}:=\{x\in{\mathbb{R}}^{n}:\langle x,y\rangle\leq 0\,\forall y\in C\} is the dual cone of C . (We denote by {\rm int} the interior, by {\rm cl} the closure, by {\rm bd} the boundary, by \langle\cdot\,,\cdot\rangle the scalar product, and by o the origin of {\mathbb{R}}^{n} .) Let K\in ps(C) . Then \rho_{K}(v):=\min\{r>0:rv\in K\} , v\in\Omega_{C} , defines the radial function \rho_{K}:\Omega_{C}\to(0,\infty) of K (the set \{r>0:rv\in K\} is not empty for v\in\Omega_{C} ; otherwise C would not be the recession cone of K ). Obviously, \rho_{K}(v)v\in{\rm bd}\,K . Let \omega_{K} be the set of all u\in{\rm cl}\,\Omega_{C^{\circ}} with the property that u is a normal vector (by which we always mean an outer unit normal vector) at more than one point of K . Then \omega_{K} has Hausdorff dimension at most n-2 (as follows from [ 6 , Thm. 2.2.5] and polarity). The reverse radial Gauss map \alpha_{K}^{*}:\Omega_{C^{\circ}}\setminus\omega_{K}\to\Omega_{C} is defined by letting \alpha_{K}^{*}(u) be the unique vector v\in\Omega_{C} for which u is attained as a normal vector of K at \rho_{K}(v)v . For a topological space X , we denote by P(X) the set of Borel probability measures on X . The main subject of this note is the following result. Theorem 1 . Let \mu\in P(\Omega_{C^{\circ}}) and \nu\in P(\Omega_{C}) and suppose that \mu is zero on sets of Hausdorff dimension at most n-2 . Then there exists a C -pseudo-cone K\in ps(C) such that (\alpha_{K}^{*})\sharp\mu=\nu . As usual, (\alpha_{K}^{*})\sharp\mu=\nu means that \mu((\alpha_{K}^{*})^{-1}(\eta))=\nu(\eta) for each Borel set \eta\subseteq\Omega_{C} , that is, \nu is the image measure of \mu under the reverse radial Gauss map of K . Theorem 1 was proved in [ 9 ] . It was motivated by the introduction and investigation of the Gauss image problem for convex bodies in [ 2 ] . The proof there used a variational argument, as was the case in [ 9 ] . A treatment of the Gauss image problem for convex bodies by mass transportation methods was presented by Bertrand [ 1 ] . The purpose of the following is to prove also Theorem 1 by such methods. The proof is much shorter than that in [ 9 ] , though less elementary. We do not try to carry over the argumentation of [ 1 ] to pseudo-cones, but apply the Kantorovich duality in a similar way as done by Rüschendorf and Rachev [ 5 ] . Theorem 1 is proved in Section 3 , while Section 2 contains some preparations. 2 Preparations involving pseudo-cones In this section, we collect and define some notions which will be essential for the proof of Theorem 1 . Let K\in ps(C) . The support function of K is defined by h_{K}(u)=\sup_{y\in K}\langle u,y\rangle\quad\mbox{for }u\in\Omega_{C^{\circ}}, from which it follows that h_{K}(u)=\sup_{v\in\Omega_{C}}\langle u,v\rangle\rho_{K}(v). Writing \overline{h}_{K}=-h_{K} , this is equivalent to \overline{h}_{K}(u)=\inf_{v\in\Omega_{C}}|\langle u,v\rangle|\rho_{K}(v). (1) The copolar set of K is defined by K^{*}=\{x\in{\mathbb{R}}^{n}:\langle x,y\rangle\leq-1\mbox{ for all }y\in K\}. It satisfies K^{*}\in ps(C^{\circ}) , and we have K^{**}=K . Moreover, \rho_{K}(v)=\frac{1}{\overline{h}_{K^{*}}(v)}=\frac{1}{\inf_{u\in\Omega_{C^{\circ}}}|\langle u,v\rangle|\rho_{K^{*}}(u)}, (2) where the first equality is shown, for example, in [ 12 ] and the second equality follows from ( 1 ). Let f:\Omega_{C}\to(0,\infty) be a function that is bounded away from 0 (that is, satisfies f\geq a for some a>0 ). We define the convexification of f by \langle f\rangle:=\bigcap\{K\in ps(C):f(v)v\in K\,\forall v\in\Omega_{C}\}. Since f is bounded away from 0 , there exists K\in ps(C) with f(v)v\in K for all v\in\Omega_{C} (for example, the intersection of C with a suitable closed halfspace). Clearly, \langle f\rangle\in ps(C) , \rho_{\langle f\rangle}\leq f (3) and \langle\rho_{K}\rangle=K\quad\mbox{for }K\in ps(C). (4) For the support function of \langle f\rangle we get, by ( 1 ) and ( 3 ), \overline{h}_{\langle f\rangle}(u)=\inf_{v\in\Omega_{C}}|\langle u,v\rangle|\rho_{\langle f\rangle}(v)\leq\inf_{v\in\Omega_{C}}|\langle u,v\rangle|f(v). Here strict inequality would contradict the definition of the convexification, hence we have \overline{h}_{\langle f\rangle}(u)=\inf_{v\in\Omega_{C}}|\langle u,v\rangle|f(v)\quad\mbox{for }u\in\Omega_{C^{\circ}}. (5) Similarly, of course, for g:\Omega_{C^{\circ}}\to(0,\infty) , bounded away from 0 , we define \langle g\rangle:=\bigcap\{K\in ps(C^{\circ}):g(u)u\in K\,\forall u\in\Omega_{C^{\circ}}\}, which is a C^{\circ} -pseudo-cone. For a function f:\Omega_{C}\to(0,\infty) , bounded away from zero, we define the pseudo-conjugate function f^{*}:\Omega_{C^{\circ}}\to(0,\infty) by f^{*}(u):=\sup_{v\in\Omega_{C}}\frac{1}{|\langle u,v\rangle|f(v)},\quad u\in\Omega_{C^{\circ}}. For fixed u\in\Omega_{C^{\circ}} , the product |\langle u,v\rangle|f(v) is bounded away from 0 , hence f^{*}(u)<\infty . It follows from ( 2 ) and ( 5 ) that \rho_{\langle f\rangle^{*}}(u)=\frac{1}{\overline{h}_{\langle f\rangle}(u)}=\frac{1}{\inf_{v\in\Omega_{C}}|\langle u,v\rangle|f(v)}=f^{*}(u), thus f^{*}=\rho_{\langle f\rangle^{*}}. (6) Since the pseudo-conjugate function of f is the radial function of the copolar set of the convexification of f , the function f^{*} is always the radial function of a C^{\circ} -pseudo-cone. It follows from ( 6 ) and ( 4 ) that (\rho_{K})^{*}=\rho_{\langle\rho_{K}\rangle^{*}}=\rho_{K^{*}} for K\in ps(C) . Of course, for functions g:\Omega_{C^{\circ}}\to(0,\infty) , we define g^{*} correspondingly. Again, let f:\Omega_{C}\to(0,\infty) be a function that is bounded away from 0 . Since f^{*}=\rho_{\langle f\rangle^{*}} by ( 6 ), we have \langle f^{*}\rangle=\langle\rho_{\langle f\rangle^{*}}\rangle=\langle f\rangle^{*} by ( 4 ), thus \langle f^{*}\rangle=\langle f\rangle^{*}. This gives f^{**}=(\rho_{\langle f\rangle^{*}})^{*}=\rho_{\langle f\rangle^{**}}=\rho_{\langle f\rangle}\leq f by ( 3 ), thus f^{**}\leq f. (7) Here equality holds if and only if f is the radial function of a C -pseudo-cone. From the definition of f^{*} it follows that f^{*}(u)\geq\frac{1}{|\langle u,v\rangle|f(v)}, hence f(v)f^{*}(u)\geq\frac{1}{|\langle u,v\rangle|}\quad\mbox{for }(v,u)\in\Omega_{C}\times\Omega_{C^{\circ}}. (8) Suppose that equality holds in ( 8 ). Then, by ( 6 ), ( 2 ) and ( 3 ), f(v)=\frac{1}{|\langle u,v\rangle|f^{*}(u)}=\frac{1}{|\langle u,v\rangle|\rho_{\langle f\rangle^{*}}(u)}\leq\frac{1}{\overline{h}_{\langle f\rangle^{*}}(v)}=\rho_{\langle f\rangle}(v)\leq f(v). Since equality holds here, we have \rho_{\langle f\rangle}(v)=f(v) and |\langle u,v\rangle|\rho_{\langle f\rangle^{*}}(u)=\overline{h}_{\langle f\rangle^{*}}(v) . The latter means that v is a normal vector of \langle f\rangle^{*} at \rho_{\langle f\rangle^{*}}(u)u . Equivalently (as follows, e.g., from [ 7 , Lem. 6] ), u is a normal vector of \langle f\rangle at \rho_{\langle f\rangle}(v)v . The converse also holds. We recall that in [ 10 ] we have defined the pseudo-subdifferential of K by \partial^{\bullet}K:=\{(v,u)\in\Omega_{C}\times\Omega_{C^{\circ}}:\mbox{$u$ is a normal vector of $K$ at $\rho_{K}(v)v$}\}. We can thus state the following lemma. Lemma 1 . Let f:\Omega_{C}\to(0,\infty) be a function bounded away from zero. The inequality f(v)f^{*}(u)\geq\frac{1}{|\langle u,v\rangle|} (9) holds for all (v,u)\in\Omega_{C}\times\Omega_{C^{\circ}} . Equality in (\ref{M9}) for some (v,u)\in\Omega_{C}\times\Omega_{C^{\circ}} holds if and only if \rho_{\langle f\rangle}(v)=f(v) and (v,u)\in\partial^{\bullet}\langle f\rangle . 3 Proof of Theorem 1 We define the function c:\Omega_{C^{\circ}}\times\Omega_{C}\to(0,\infty) by c(u,v):=-\log|\langle u,v\rangle|,\quad(u,v)\in\Omega_{C^{\circ}}\times\Omega_{C}. (10) This is the cost function found in [ 10 ] , but with a negative sign, to make it nonnegative. Given \mu\in P(\Omega_{C^{\circ}}) and \nu\in P(\Omega_{C}) , we denote by M(\mu,\nu) the set of all probability measures \pi on