Answers are grounded in this paper and cite their sources.
This paper reveals properties of metric projections on Wasserstein spaces, suggesting implications for convex subsets.
In this paper, we investigate properties of metric projections onto specific closed and geodesically convex proper subsets of Wasserstein spaces <m:math xmlns:m="http://www.w3.org/1998/Math/MathML"> <m:mrow> <m:mo stretchy="false">(</m:mo> <m:mrow> <m:msub> <m:mi mathvariant="script">P</m:mi> <m:mi>p</m:mi> </m:msub> <m:mo></m:mo> <m:mrow> <m:mo stretchy="false">(</m:mo> <m:msup> <m:mi mathvariant="bold">R</m:mi> <m:mi>d</m:mi> </m:msup> <m:mo stretchy="false">)</m:mo> </m:mrow> </m:mrow> <m:mo>,</m:mo> <m:msub> <m:mi>W</m:mi> <m:mi>p</m:mi> </m:msub> <m:mo stretchy="false">)</m:mo> </m:mrow> </m:math> (Pₚ(Rᵈ),Wₚ) . When <m:math xmlns:m="http://www.w3.org/1998/Math/MathML"> <m:mrow> <m:mi>d</m:mi> <m:mo>=</m:mo> <m:mn>1</m:mn> </m:mrow> </m:math> d=1 , as <m:math xmlns:m="http://www.w3.org/1998/Math/MathML"> <m:mrow> <m:mo stretchy="false">(</m:mo> <m:mrow> <m:msub> <m:mi mathvariant="script">P</m:mi> <m:mn>2</m:mn> </m:msub> <m:mo></m:mo> <m:mrow> <m:mo stretchy="false">(</m:mo> <m:mi mathvariant="bold">R</m:mi> <m:mo stretchy="false">)</m:mo> </m:mrow> </m:mrow> <m:mo>,</m:mo> <m:msub> <m:mi>W</m:mi> <m:mn>2</m:mn> </m:msub> <m:mo stretchy="false">)</m:mo> </m:mrow> </m:math> (P₂(R),W₂) is isometrically isomorphic to a flat space with a Hilbertian structure, the corresponding projection operators are expected to be nonexpansive. We give a direct proof of this fact, relying on intrinsic analysis, which also implies nonexpansiveness in certain special cases in higher dimensions. When <m:math xmlns:m="http://www.w3.org/1998/Math/MathML"> <m:mrow> <m:mi>d</m:mi> <m:mo>></m:mo> <m:mn>1</m:mn> </m:mrow> </m:math> d>1 , we show the failure of this property in two regimes: when <m:math xmlns:m="http://www.w3.org/1998/Math/MathML"> <m:mrow> <m:mi>p</m:mi> <m:mo>></m:mo> <m:mn>1</m:mn> </m:mrow> </m:math> p>1 is either small enough or large enough. Finally, we prove some positive curvature properties of Wasserstein spaces <m:math xmlns:m="http://www.w3.org/1998/Math/MathML"> <m:mrow> <m:mo stretchy="false">(</m:mo> <m:mrow> <m:msub> <m:mi mathvariant="script">P</m:mi> <m:mi>p</m:mi> </m:msub> <m:mo></m:mo> <m:mrow> <m:mo stretchy="false">(</m:mo> <m:msup> <m:mi mathvariant="bold">R</m:mi> <m:mi>d</m:mi> </m:msup> <m:mo stretchy="false">)</m:mo> </m:mrow> </m:mrow> <m:mo>,</m:mo> <m:msub> <m:mi>W</m:mi> <m:mi>p</m:mi> </m:msub> <m:mo stretchy="false">)</m:mo> </m:mrow> </m:math> (Pₚ(Rᵈ),Wₚ) when <m:math xmlns:m="http://www.w3.org/1998/Math/MathML"> <m:mrow> <m:mi>d</m:mi> <m:mo>≥</m:mo> <m:mn>2</m:mn> </m:mrow> </m:math> d≥ 2 and <m:math xmlns:m="http://www.w3.org/1998/Math/MathML"> <m:mrow> <m:mi>p</m:mi> <m:mo>∈</m:mo> <m:mrow> <m:mo stretchy="false">(</m:mo> <m:mn>1</m:mn> <m:mo>,</m:mo> <m:mrow> <m:mo>+</m:mo> <m:mi mathvariant="normal">∞</m:mi> </m:mrow> <m:mo stretchy="false">)</m:mo> </m:mrow> </m:mrow> </m:math> p∈(1,+∞) are arbitrary: we show that Wasserstein spaces are nowhere locally Busemann NPC spaces, and they nowhere locally satisfy the so-called projection criterion . As a corollary of the former, they have nonnegative upper Alexandrov curvature , in a precise sense that we define here. In our analysis, a particular subset of probability measures having densities uniformly bounded above by a given constant plays a special role.
No takes yet. Share an insight, caveat, or question.
Adve et al. (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: