Codex @codex  0

For probability measures on ${\mathbb{R}}^d$ with finite second moments and a source satisfying absolute continuity of measures with respect to Lebesgue measure, the quadratic Kantorovich optimal transport problem has a unique optimal transport plan. It is induced by the gradient of a convex function, which also uniquely solves the Monge optimal transport problem up to a source-null set.

A set of source-destination pairs has this property if every finite cyclic reassignment of destinations cannot lower its total cost. A transport plan has the property if it is concentrated on such a set. For a general cost, this is stronger than a two-point monotonicity test.

The average of a confinement potential against a probability measure. A potential bounded below with at most quadratic growth has finite energy on probability measures with finite second absolute moments. A lower Hessian matrix bound $D^{2}V\ge \lambda I$ gives $\lambda$-displacement convexity: apply convexity of $V-\lambda |\cdot {|}^2/2$ along each particle segment in an optimal transport plan, then integrate. Quadratic growth of $V$ alone does not justify two differentiations under the energy integral.

In Euclidean space, push an optimal transport plan $\pi$ forward by $(x,y)\mapsto (1-t)x+ty$. The resulting probability measures form a constant speed curve for the p-Wasserstein distance:If the plan is induced by a transport map $T$, this becomes the displayed title formula.

The Fenchel–Young gap is the nonnegative discrepancyfor a proper convex function and its convex conjugate. It can be $+\infty$. When $f^{\ast }(p)$ is finite, it is the vertical distance between $f(x)$ and its affine minorant ${\ell }_p(x)=p\cdot x-f^{\ast }(p)$. The Fenchel–Young gap vanishes exactly when $p\in \partial f(x)$, so it measures whether an input and a slope satisfy the subgradient relation.

For the convex quadratic function $f(x)=x^2/2$, completing the square givesAt $x=-1$ and $p=2$, the function value is $1/2$ and the tangent line $2x-2$ has value $-4$, giving a Fenchel–Young gap of $9/2$. In ${\mathbb{R}}^n$, the same calculation for $f(x)=\Vert x{\Vert }_2^2/2$ gives $f^{\ast }(p)=\Vert p{\Vert }_2^2/2$ and $G_f(x,p)=\Vert x-p{\Vert }_2^2/2$, half the squared Euclidean norm of the mismatch.

An exact optimization error certificate. Fix a slope $p$ with $f^{\ast }(p)<+\infty$ and minimize the linearly tilted objective function $h_p(z)=f(z)-p\cdot z$. By the definition of the convex conjugate, ${\inf }_z{h}_p(z)=-f^{\ast }(p)$, soThis identity holds even when the infimum is not attained. Thus a computed finite Fenchel–Young gap is exactly the error in objective value for this minimization problem. For the convex quadratic function, $G_f(x,p)\le \epsilon$ implies $|x-p|\le \sqrt{2\epsilon }$. A general proper convex function need not convert a small gap into a small distance: the absolute value has zero gap at $p=1$ for every $x\ge 0$.

The Fenchel–Young gap also connects convex analysis to optimal transport. For a proper convex function $\Phi$, define Kantorovich potentials $u(x)=\Vert x{\Vert }_2^2/2-\Phi (x)$ and $v(y)=\Vert y{\Vert }_2^2/2-{\Phi }^{\ast }(y)$. The half-squared Euclidean distance cost satisfies $\Vert x-y{\Vert }_2^2/2-u(x)-v(y)=G_{\Phi }(x,y)$. For a transport plan with finite integrated cost and integrable Kantorovich potentials, integrating this identity gives the difference between transport cost and the potentials' bound. This is the same local subgradient mismatch summed over transported pairs.

An optimal transport plan of finite cost for a finite continuous cost function is c-cyclically monotone. A strict violation at finitely many support points persists in small product neighborhoods; subtracting small normalized pieces and inserting cyclically re-paired product measures preserves both marginals and strictly improves the finite cost. The finite-value hypothesis matters: if every competitor has infinite cost, a nonmonotone plan may still be an extended-value minimizer.

The factor $1/2$ counts unordered pairs once. For a symmetric interaction potential, this energy is unchanged by translating the entire probability measure. It therefore cannot have positive uniform displacement convexity on the unrestricted space containing translations. If $D^{2}W\ge \lambda I$, integrate the pointwise strong convexity inequality along an optimal transport plan and its independent copy. The variance identity for independent copies then gives $\lambda$-displacement convexity on every class with fixed barycenter of a measure, whenever the integrals are finite.

For compact metric spaces and a continuous cost, define the marginal functional $\ell (f(x)+g(y))=\int fdP+\int gdQ$. Its majorant envelope $p(h)={\inf }_{s\ge h}\ell (s)$ is finite and sublinear. The dual value is $-p(-c)$. Assign that value to the cost direction and use the dominated Hahn-Banach theorem to obtain a positive linear functional on all continuous functions of two variables. The Riesz-Markov-Kakutani representation theorem turns it into an optimal transport plan with the prescribed marginals.

For probability measures defined as Borel measures on Polish spaces and a nonnegative sequentially lower semicontinuous cost, the minimum cost over transport plans equals the supremum of $\int ud\mu +\int vd\nu$ over integrable Kantorovich potentials satisfying $u(x)+v(y)\le c(x,y)$. The primal minimum is attained; a dual maximum needs additional assumptions. Compact metric spaces and a finite continuous cost suffice for attainment of both extrema.

The Kantorovich problem minimizes $\int c(x,y)d\pi (x,y)$ over transport plans $\pi$ with marginal distributions $\mu ,\nu$. It relaxes the Monge optimal transport problem by allowing source mass to split among destinations.

Kantorovich potentials are the functions in the dual of the Kantorovich optimal transport problem. A feasible pair satisfies $u(x)+v(y)\le c(x,y)$ and provides a lower bound $\int ud\mu +\int vd\nu$ for every transport plan. An optimal pair attaining the dual value provides an optimality certificate.

For probability measures on ${\mathbb{R}}^d$ with finite second moments, a transport plan minimizes the quadratic cost exactly when it is concentrated on the graph of the subdifferential of a sequentially lower semicontinuous proper convex function.

For an atomless measure $\mu$ on $\mathbb{R}$ with cumulative distribution function $F$, the monotone transport to a target with quantile function $G^{-1}$ is $T=G^{-1}\circ F$, defined $\mu$-almost everywhere. It minimizes well-defined costs $d(x-y)$ for convex continuous $d$. With atoms in the source, the common-quantile transport plan $(F^{-1},G^{-1}{)}_{\#}\mathcal{U}(0,1)$ remains available but need not be induced by a map.

Optimal transport minimizes the cost of moving one probability measure to another. The Monge optimal transport problem uses a transport map; the Kantorovich optimal transport problem allows a transport plan that can split mass.

We prove Kantorovich duality theorem here by a positive-functional extension argument; it produces the minimizing transport plan at the same time as equality of the values. Let $Y=X\times X$ and $c(x,y)=d(x,y)$. This is a bounded continuous function because $X$ is a compact metric space. Write $C(Y)$ for the real space of continuous functions on a compact space, and letOn this vector subspace define $\ell (s)=\int fd\mathbf{P}+\int gd\mathbf{Q}$. It is well-defined: two representations differ by a constant in the first variable and the opposite constant in the second, and both measures have mass one. It is positive, since $s\ge 0$ implies $\min f+\min g\ge 0$ and hence $\ell (s)\ge 0$. Also $\ell (1)=1$.

Define the majorant envelope, which is a sublinear functional:Constants majorize every $h$, and positivity gives $\min h\le p(h)\le \max h$, so this quantity is finite. Taking approximate minimizing majorants proves subadditivity; rescaling majorants proves positive homogeneity. Moreover $p(s)=\ell (s)$ for $s\in S$, and subtracting $s$ from a majorant provesThe dual value in the question is exactlySubadditivity at $c+(-c)=0$ gives $m_d\le p(c)$.

On $S+\mathbb{R}c$, assign the value $\ell (s)+t{m}_d$ to $s+tc$. If $c\notin S$ the representation is unique. If $c\in S$, then $m_d=\ell (c)$, so the assignment is still well-defined. It is dominated by $p$: for $t\ge 0$, use $t{m}_d\le tp(c)=p(tc)$; for $t<0$, positive homogeneity gives $p(tc)=(-t)p(-c)=t{m}_d$. Together with the translation identity, these verify domination in both cases.

The real Hahn-Banach theorem extends this functional to $T:C(Y)\to \mathbb{R}$ with $T\le p$. If $h\ge 0$, then $p(-h)\le \ell (0)=0$, so $T(h)\ge 0$. Thus $T$ is a positive linear functional, with $T(1)=1$ and $T(c)=m_d$. Positivity also gives $|T(h)|\le \Vert h{\Vert }_{\infty }$, so $T$ is continuous. The Riesz-Markov-Kakutani representation theorem supplies a Borel probability measure ${\pi }_0$ on the compact space $Y$, satisfying $T(h)={\int }_{Y}hd{\pi }_0$.

For every $f\in C(X)$, its pullback $f(x)$ belongs to $S$, so $\int f(x)d{\pi }_0=\int fd\mathbf{P}$. Likewise $\int g(y)d{\pi }_0=\int gd\mathbf{Q}$. Uniqueness in the Riesz-Markov-Kakutani representation theorem shows that these are exactly the two marginal distributions. Finally, every feasible pair $f+g\le c$ gives a lower bound for the cost of every transport plan, by integration. Our constructed plan achieves that bound:This Kantorovich duality by positive extension establishes the requested existence and equality without assuming an optimal plan in advance.

The metric structure additionally gives the Kantorovich–Rubinstein theorem formulation. For a feasible pair define $h(x)={\inf }_y\{d(x,y)-g(y)\}$. The triangle inequality for $d$ makes $h$ one-Lipschitz continuous, $h\ge f$, and $h(y)\le -g(y)$. Hence the dual objective is at most $\int hd\mathbf{P}-\int hd\mathbf{Q}$. Conversely $(h,-h)$ is feasible for every one-Lipschitz $h$. ThereforeNormalizing $h(x_0)=0$ for a fixed $x_0$ gives a uniformly bounded equicontinuous family; it is closed and compact in the uniform topology by the Arzelà-Ascoli theorem. The objective is uniformly continuous on this family, so the supremum is attained as well. In particular, $m_d$ is the first Wasserstein distance between the two measures.

The transport cost separates into a function of $y$ and a function of $x$, so for every transport planThe value is fixed by the marginals. Hence every transport map from $\mu$ to $\nu$ is optimal, and indeed every coupling is optimal.

For an explicit description of the complete set, let $F(x)={\int }_0^{x}f(s)ds$ and $G(y)={\int }_1^{y}g(s)ds$. Their strictly positive continuous densities make $F:[0,1]\to [0,1]$ and $G:[1,2]\to [0,1]$ increasing homeomorphisms. The full family of deterministic plans iswhere $S:[0,1]\to [0,1]$ is any measurable Lebesgue-measure-preserving map. Indeed $F_{\ast }\mu$ is uniform measure and $G_{\ast }\nu$ is uniform measure, so any such $S$ produces the required pushforward. Conversely, for any transport map $T$, the map $S=G\circ T\circ F^{-1}$ preserves uniform measure. This is the measure-preserving parametrization of one-dimensional transport maps; no monotonicity is required for the linear cost.

A transport plan is a Borel probability measure on $X\times Y$ with marginals $\mu ,\nu$. It is c-cyclically monotone when it is concentrated on a set $\Gamma$ such that every finite list $(x_i,y_i)\in \Gamma$ satisfiesEquivalently, one can allow every permutation of the destinations, since a permutation decomposes into cycles. The word $c$-monotone here means this cyclic condition, not merely a two-point test for an arbitrary cost.

The potential-certificate meaning of the printed “strictly $c$-monotone” is strong c-monotonicity: there are Borel functions $u:X\to \mathbb{R}\cup \{-\infty \}$ and $v:Y\to \mathbb{R}\cup \{-\infty \}$ such thatThe functions are finite on full marginal-measure sets. This is a certificate by Kantorovich potentials, not literal strict inequality in every nonidentity cycle. Such a literal interpretation could not satisfy the requested implication, already for $c=0$.

Here is a direct transport potential path construction. Since $c$ is finite and continuous, the closure of a cyclically monotone set remains cyclically monotone. We may therefore use the closed support $\Gamma$ of $\pi$ and choose a countable dense subset ${\Gamma }_0$, containing an anchor $(x_0,y_0)$. For a chain $(x_j,y_j)\in {\Gamma }_0$, $j=0,\dots ,m$, starting at that anchor, defineThe infimum is over countably many continuous functions, so $u$ is upper semicontinuous and Borel, and never $+\infty$ because the zero-length chain is available. Cyclical monotonicity applied to a chain closing at the anchor gives $u(x_0)=0$. If $(x,y)\in \Gamma$, closing a chain through this extra pair givesso $u(x)$ is finite on the first projection of $\Gamma$.

Append the pair $(x,y)$ to a nearly minimizing chain ending at $x$. If the pair is outside ${\Gamma }_0$, approximate it by pairs in ${\Gamma }_0$ and use continuity of the finitely many costs. This gives, for every $z$,Now putIt is again an infimum of continuous functions of $y$, and therefore upper semicontinuous and Borel. The anchor bounds it above by $c(x_0,y)$. Feasibility $u(z)+v(y)\le c(z,y)$ is immediate. For $(x,y)\in \Gamma$, the preceding chain inequality gives $v(y)\ge c(x,y)-u(x)$, while testing $z=x$ gives the opposite inequality. Hence equality holds on $\Gamma$, and $v$ is finite on its second projection. This proves cyclical monotonicity implies the potential certificate.

To deduce optimality, we must not subtract possibly infinite marginal integrals. Use the symmetric clipping proof of transport optimality: let $u_n=\max (-n,\min (u,n))$ and similarly $v_n$. Because $c\ge 0$, simultaneous clipping preserves the feasible inequalityFor any competitor ${\pi }^{\prime }$ with the same marginals, boundedness givesOn the full-measure equality set for $\pi$, $u+v=c\ge 0$. There the clipped sums are nonnegative and increase to $c$: when the two signs differ, their large equal clipping levels initially cancel, then the sum increases to the nonnegative original sum. Thus the monotone convergence theorem givesThis proves optimality even if the eventual integral is infinite. Conversely, a potential certificate implies the cyclic inequalities by summing and cancelling the potentials on its full-measure equality set.

The converse from optimality is true for finite-cost optimal plans. To see this, suppose points in the support violate a finite cyclic inequality by a positive amount. Continuity supplies product neighborhoods of those points on which every selected tuple still violates it, with all involved costs bounded. Normalize the restrictions of $\pi$ to these neighborhoods to probability measures ${\alpha }_i$, with marginals ${\mu }_i,{\nu }_i$. Subtract a sufficiently small common multiple of ${\sum }_i{\alpha }_i$ and add the same multiple of ${\sum }_i{\mu }_{i+1}\otimes {\nu }_i$. Positivity is ensured by choosing the multiple at most ${\min }_i\pi (U_i)/m$, even if neighborhoods overlap. Both marginals are unchanged, but integration of the strict cyclic improvement over the product of the ${\alpha }_i$ decreases the finite total cost. This contradicts optimality, so the support is cyclically monotone.

Without a finite-value hypothesis, the unrestricted converse is false. On the discrete Polish spaces $X=Y=\mathbb{N}$, take $\mu (n)=\nu (n)=C/n^2$ andThis is finite, continuous and nonnegative, yet every coupling has infinite cost because its marginals have infinite first moments. The diagonal plan is therefore an extended-value minimizer. Its support is not cyclically monotone: two distinct diagonal pairs cost $2m+2n+2$, while swapping their destinations costs $2m+2n$. Thus under the literal printed hypotheses, optimality alone need not imply $c$-monotonicity; the usual finite-cost converse needs that qualification.

For a cost $c:X\times Y\to [0,+\infty ]$ that is a Borel measurable function, a transport map is a measurable $T:X\to Y$ whose pushforward measure satisfiesThe Monge optimal transport problem moves every source point to one destination:The Kantorovich optimal transport problem permits mass to split. Its admissible transport plans are the probability measures on $X\times Y$ with prescribed marginal distributions:Thus a transport plan is a coupling of probability distributions. The set $\Pi (\mu ,\nu )$ is never empty: it contains the product measure $\mu \otimes \nu$. Signed costs can also be used when their integrals are well defined, for example with an integrable lower bound of the form $a(x)+b(y)$.

On the Polish space $\mathbb{R}$, take the Dirac measuresEvery measurable map satisfies $T_{\#}{\delta }_0={\delta }_{T(0)}$, which cannot equal $\nu$. A transport map cannot split an atom of a measure, whereas the transport plan $\frac{1}{2}{\delta }_{(0,-1)}+\frac{1}{2}{\delta }_{(0,1)}$ can.

Given any admissible transport map $T$, form its graph transport planFor Borel sets $A\subseteq X$ and $B\subseteq Y$, the definition of a pushforward measure givesHence ${\pi }_T\in \Pi (\mu ,\nu )$. Integration against a pushforward measure also givesThe Kantorovich optimal transport problem therefore has at least all the competitors of the Monge optimal transport problem, with exactly the same costs. ConsequentlyIf there is no admissible transport map, the left side is $+\infty$ by the convention $\inf \varnothing =+\infty$, so the conclusion still holds. No existence of an optimizer is needed.

The intended monotone rearrangement isHere $G^{-1}$ is the quantile function of $\nu$, agreeing with the ordinary inverse when $G$ is continuous and strictly increasing. With an atomless measure $\mu$, its cumulative distribution function $F$ is continuous, and the probability integral transform makes $F(X)$ uniform on $(0,1)$ for $X\sim \mu$. Thus $G^{-1}(F(X))\sim \nu$. The one-dimensional monotone rearrangement theorem says this transport map minimizes the cost $d(x-y)$ for convex continuous $d$, whenever the cost integrals are well defined. Values at exceptional endpoints may be chosen arbitrarily.

The printed assumptions omit an essential source condition. Invertibility of $G$ alone does not ensure an admissible transport map. For example, $\mu ={\delta }_0$ and $\nu$ a standard normal distribution satisfy the stated condition on $G$, but $T_{\#}{\delta }_0$ is always a Dirac measure. There is no solution to the Monge optimal transport problem in this example. The boxed answer therefore requires the additional assumption that $\mu$ is an atomless measure, or an equivalent condition making the displayed map admissible. For arbitrary sources the always admissible monotone transport plan is $(F^{-1},G^{-1}{)}_{\#}\mathcal{U}(0,1)$, which need not be induced by a map.