---
title: Weak Optimal Transport (WOT)
url: https://www.emergentmind.com/topics/weak-optimal-transport-wot
type: topic
---

# Weak Optimal Transport (WOT)

Weak Optimal Transport (WOT) generalizes the classical optimal transport framework by allowing nonlinear cost functionals that depend on entire conditional distributions, rather than just pairwise matches. It unifies, extends, and regularizes numerous transport and stochastic order problems found in mathematics, statistics, economics, and data science. The problem admits a robust duality theory, supports general existence and structural theorems, and underpins both theoretical and computational advances in convex geometry, risk measurement, learning, and economic modeling.

## 1. Static Formulation and Primal Problem

In the general setting, let $X$ and $Y$ be Polish spaces, and $\mu\in\mathcal{P}(X)$, $\nu\in\mathcal{P}(Y)$ be Borel probability measures. The set of transport plans $\Pi(\mu,\nu)$ consists of couplings $\pi\in\mathcal{P}(X\times Y)$ with marginals $\mu$, $\nu$. Any $\pi$ admits a disintegration $\pi(dx,dy) = \mu(dx)\,\pi_x(dy)$, with each $\pi_x\in\mathcal{P}(Y)$ a regular conditional probability.

A *weak* cost is given by a map $C:X\times \mathcal{P}(Y)\to [0,+\infty]$, jointly lower semicontinuous and convex in the measure variable. The primal weak optimal transport problem is:

\[
\mathrm{WOT}_C(\mu, \nu) = \inf_{\pi\in\Pi(\mu, \nu)} \int_X C(x, \pi_x)\, \mu(dx).
\]

For $C(x,p) = \int_Y c(x, y)\,p(dy)$, the classical Kantorovich problem is recovered as a special case.

A core instance is the **barycentric weak transport**: for $X=Y=\mathbb{R}^d$ and convex $\theta$, set $C(x,p) = \theta(x - \int y\, p(dy))$. This permits mass-splitting and aggregation, and replaces strict point-to-point transport by barycentric projections, as in the Brenier–Strassen and convex order–constrained transport problems [2501.16316][2003.05338][2102.13380].

## 2. Duality and Structural Theorems

The dual problem involves maximizing over potential functions, often under monotonicity and convexity restrictions. For admissible test functions $g\in \mathcal{C}_{b,p}(Y)$, define the $C$–transform

\[
g^C(x) = \inf_{p\in\mathcal{P}_p(Y)} \left\{ \int g\,dp + C(x, p) \right\}.
\]

The *fundamental duality theorem* asserts [2501.16316]:

\[
\mathrm{WOT}_C(\mu,\nu) = \sup_{g\in \mathcal{C}_{b,p}(Y)} \left\{ \mu(g^C) - \nu(g) \right\}.
\]

Under additional $c$-decreasing or order-monotonicity constraints (i.e., $C$ decreases along convex order), the dual can be restricted to convex (or increasing–convex) test potentials, unifying the classical Brenier, Kantorovich–Rubinstein, Strassen, mechanism design, and martingale Benamou–Brenier dualities [2507.07200].

For "barycentric" WOT, the dual takes the form
\[
\max_{\psi\text{ convex}} \left\{ \mu(\psi^C) - \nu(\psi) \right\},
\]
where
\[
\psi^C(x) = \inf_{z} \{ \psi(z) + \theta(x - z) \},
\]
generalizing the $c$-transform [2507.07200][2501.16316].

Complementary slackness holds: the pair $(\pi^*,g^*)$ is optimal if and only if, $\mu$-a.e.,
\[
C(x,\pi^*_x) = g^C(x) + \int g^*(y)\, \pi^*_x(dy).
\]

## 3. Existence, Monotonicity, and Stability

The direct method of calculus of variations yields primal attainment provided $C$ has appropriate convexity, joint lower semicontinuity, and quadratic (or appropriate $p$-power) growth [2501.16316][2003.05338]. Existence and strong duality follow for any $C$ convex and l.s.c. in the measure variable, and lower bounded.

A key property is $C$-monotonicity: a plan $\pi$ is optimal iff it is supported on the $C$-monotone set—generalizing $c$-cyclical monotonicity:

A set $\Gamma\subset X\times\mathcal{P}(Y)$ is $C$-monotone if, for tuples $(x_i, p_i),\, (x_i, q_i)\in\Gamma$ with $\sum p_i = \sum q_i$,
\[
\sum C(x_i, p_i) \le \sum C(x_i,q_i).
\]
Stability holds: $C$-monotonicity is closed under weak convergence of couplings and cost perturbations [1904.04171].

## 4. Examples and Special Cases

Representative cost functions and order constraints recover diverse classical and modern transport and stochastic order problems:

- **Classical OT**: $C(x,p)=\int_Y c(x,y)p(dy)$.
- **Barycentric WOT** (Brenier–Strassen, convex-order projections): $C(x,p) = \theta(x-\mathbb{E}_p[y])$.
- **Martingale/Martingale Benamou–Brenier**: $C(x,p)$ infinity unless $\int y p(dy) = x$; otherwise a functional of $p$.
- **Mechanism Design**: $C(x,p)=\inf_{q\preceq_{ic} p}\theta(x-\mathrm{mean}(q))$.
- **Entropic/Schrödinger Problems**: $C(x,p)=H(p|\gamma_x)$, or includes regularization terms [2501.09362].

**Table: Core Weak Optimal Transport Costs**

| Cost $C(x,p)$                       | Induced Order      | Applications / Theorems         |
|--------------------------------------|--------------------|---------------------------------|
| $\int_Y c(x,y) p(dy)$               | None               | OT, Kantorovich–Rubinstein, Brenier |
| $\theta(x-\mathrm{mean}(p))$        | Convex order       | Brenier–Strassen, weak barycenters |
| $H(p|\gamma_x)$                     | None               | Schrödinger bridge, entropy OT  |
| $0$ if $\delta_x\preceq_c p$, $\infty$ else | Convex order   | Strassen's theorem             |
| $\inf_{q\preceq_{ic} p}\theta(x-\mathrm{mean}(q))$ | Increasing–convex order | Monopoly revenue, mechanism design |

## 5. Dynamic Formulations and PDE Links

Dynamic equivalents of WOT are established using generalizations of the Benamou–Brenier and Fokker–Planck equations. For convex, 1-homogeneous costs $f$ on measure-valued symmetric matrix flows $(\rho, \lambda)$ solving generalized Fokker–Planck evolutions, one writes [2311.13872]:

\[
\partial_t \rho = \mathrm{tr}\left(\tfrac12\nabla_x^2 \lambda\right), \quad \rho_{t=0} = \mu,\, \rho_{t=1} = \nu,
\]
and cost
\[
F(\mu,\nu) = \inf_{(\rho, \lambda)} \int_0^1 \int f\big(\frac{d\lambda_t}{d|\lambda_t|}\big) d|\lambda_t| dt.
\]
Static and dynamic problems are *equivalent*: $F(\mu, \nu) = H(\mu, \nu)$, with strong duality and uniqueness under regularity assumptions [2311.13872].

## 6. Computational Methods and Algorithms

Discrete and continuous algorithms have been developed including mirror descent, Sinkhorn-type methods, and neural parameterized saddle-point approaches [2205.09825][2201.12220].

- **Mirror Descent**: For entropic regularization, mirror descent in dual or primal variables with KL divergence yields $O(1/\sqrt{T})$ convergence in the duality gap [2205.09825].
- **Neural Adversarial Training**: Parameterizes primal maps (plans) and dual potentials with neural networks, enabling scalable computation in high dimensions and non-convex landscapes [2201.12220].
- **Sinkhorn-type Schemes**: Regularized dynamic WOT admits iterative improvement analogously to classical entropic transport [2604.02312].

For unnormalized-kernel variants (WOTUK), mass can be split non-proportionally, supporting unbalanced scenarios and flexible economic matching frameworks [2203.16227][2205.09825].

## 7. Applications and Extensions

WOT underpins a wide array of modern applications:

- **Barycenters and Aggregation**: WOT barycenters extract shared latent structures, are robust to outliers, and admit efficient fixed-point or streaming algorithms [2102.13380][1909.05517].
- **Risk Measures**: Penalty-based risk functionals via WOT robustly account for worst-case scenarios in insurance and finance [2312.05973].
- **Mechanism Design**: WOT duals deliver sharp pricing characterizations in revenue-maximizing mechanisms through convex or order-restricted potentials [2507.07200][2003.05338].
- **Stochastic Orderings**: Characterizations via kernels and convex (or positively 1-homogeneous) order functions generalize Strassen's theorem, extending to martingales and unbalanced case [2203.16227][2507.07200].
- **Numerics and Economics**: Labor market matching, production models, outlier-robust clustering, and Wasserstein barycenters in high dimensions all benefit from the WOT calculus [2205.09825][2201.12220][2102.13380].

---
In summary, Weak Optimal Transport forms a fundamental bridge between classical OT and a host of sophisticated, nonlinear and stochastic optimization problems. WOT theory unifies convex geometric, informational, and probabilistic paradigms; its duality, order, and regularity structures are central to modern transport, risk, and economic modeling [2501.16316][2311.13872][2205.09825][2507.07200][2003.05338].

Source: https://www.emergentmind.com/topics/weak-optimal-transport-wot