---
title: Barycentric Optimal Transport
url: https://www.emergentmind.com/topics/barycentric-optimal-transport-problem
type: topic
---

# Barycentric Optimal Transport

The barycentric optimal transport problem is a class of weak optimal transport (WOT) problems characterized by cost functions that depend on the source and the barycenter (conditional mean) of the target variables under a chosen coupling. This formulation encompasses classical quadratic cost but generalizes to convex costs and finds applications in constrained transport map estimation, weak matching, robust barycenters, and aggregation models in economics and machine learning.

## 1. Mathematical Formulation and Static Characterization

Let $\mu \in \mathcal P(\mathcal X)$ and $\nu \in \mathcal P(\mathcal Y)$ be probability measures on compact, convex domains. For a coupling $\pi \in \Pi(\mu,\nu)$, with disintegration $\pi(dx,dy)=\mu(dx)\,\pi(dy|x)$, the barycentric map at $x$ is the conditional mean:
\[
b(\pi,x) = \int_{\mathcal Y} y\,\pi(dy|x) = \mathbb E_{y\sim\pi(\cdot|x)}[y].
\]
Given a cost $c(x,z)$, convex in $z$, the barycentric WOT problem is:
\[
\mathsf{WOT}_\mathrm{bary}(\mu,\nu) = \inf_{\pi\in\Pi(\mu,\nu)} 
\int_{\mathcal X} c\bigl(x,\,b(\pi,x)\bigr)\,\mu(dx).
\]
The most studied special case uses quadratic cost $c(x,z)=\|x-z\|^2$, yielding the quadratic barycentric optimal transport problem [2509.04935, 2511.21287, 2205.09825].

## 2. Dual Formulation and Convex Order Structure

Under the condition that $z\mapsto c(x,z)$ is convex and lower semicontinuous, the dual can be written as [2205.09825]:
\[
\mathsf{WOT}_\mathrm{bary}(\mu,\nu) = \inf_{\psi \in \mathrm{ConvLip}(\mathrm{conv}(\mathcal Y))} 
\left\{ 
\int_{\mathcal X} Q_c[\psi](x)\,\mu(dx) + \int_{\mathcal Y}\psi(y)\,\nu(dy)
\right\}
\]
where $Q_c[\psi](x) = \sup_{z} \{c(x,z) - \psi(z)\}$ is the Hamiltonian (sup-transform).

In the quadratic case, a convex order projection emerges: the barycentric OT cost is equal to the minimal $W_2^2$ distance among all projections of $\mu$ onto measures $\eta$ dominated by $\nu$ in convex order (i.e., $\eta \preceq_c \nu$) [2511.21287, 2509.04935]:
\[
\overline T_2(\mu,\nu) = \inf_{\eta\preceq_c\nu} W_2^2(\mu,\eta).
\]
The optimizer $\bar\eta$ is unique and given as the pushforward of $\mu$ under a 1-Lipschitz convex function (Brenier map).

## 3. Dynamic (Benamou–Brenier) and Martingale Analogues

The barycentric OT problem admits a dynamic formulation analogous to the Benamou–Brenier formula for classical OT, but only penalizing the drift component [2511.21287]:
\[
\overline T_2(\mu,\nu) = \inf_{(X_t)} \mathbb E\left[ \int_0^1 |v_t|^2dt \right],
\]
where processes $X_t$ satisfy $dX_t = v_tdt + \sigma_t dB_t$, $X_0 \sim \mu$, $X_1 \sim \nu$, and $\sigma_t$ is arbitrary (martingale part is cost-free). This focuses transport cost on movement of conditional expectations, with extra noise/martingales unconstrained.

An interpolation between barycentric and martingale cost yields the $\alpha$–$\beta$ functional [2511.21287]:
\[
\inf_{\pi\in\Pi(\mu,\nu)}
\int [\alpha|x-\mathbb E[Y|X=x]|^2 - \beta\,\mathsf{MCov}(\pi_x, \gamma^d_1)]\,\mu(dx)
\]
where $\mathsf{MCov}(\rho,\sigma)$ is the maximal covariance between measures $\rho$ and $\sigma$, and $\gamma^d_1$ is standard Gaussian.

In one dimension and for Gaussians, explicit formulae in terms of convex envelopes or linear SDEs are available [2509.04935, 2511.21287].

## 4. Numerical Algorithms: Mirror Descent and Barycentric Projections

For discrete or sampled measures, first-order primal and dual mirror descent algorithms with KL (entropic) geometry are efficient in high dimensions [2205.09825, 2107.01718]. The primal maximizes over the coupling matrix $P$ with objective:
\[
f(P) = \sum_{i=1}^n a_i\, c\left(x_i,\, \sum_{j=1}^m P_{ij}y_j\right)
\]
using gradient (in $P$) followed by KL-exponentiation and Sinkhorn projection to enforce marginals.

The dual works with convex potentials (parametrized on support) using subgradients derived from the Hamiltonian optimizer $z^*$.

Complexity per iteration is $O(nm)$ with convergence rate $O(1/\sqrt{T})$ for general convex cost and $O(1/T)$ if smooth/strongly convex. For quadratic cost, the barycentric projection of an optimal plan provides a plug-in estimator for maps, with nonparametric and smoothed estimators’ rates analyzed in [2107.01718].

In the context of constrained map estimation, the barycentric mapping for a plan $\pi$ provides the conditional mean $\bar\pi(x) = \mathbb{E}[Y|X=x]$, which is then projected onto a given function class $G$ (e.g., Lipschitz gradients, RKHS, neural nets) via $L^2$-minimization [2407.13445].

## 5. Theoretical Guarantees: Existence, Uniqueness, and Structure

- Existence of a barycentric OT optimizer is ensured by compactness and lower semicontinuity of the cost.
- In the quadratic barycentric setting, the optimizer splits as: Brenier map $\mu \to \bar\mu$ (convex order minimal measure) followed by any martingale coupling $\bar\mu \to \nu$, as characterized by Strassen-type theorems [2509.04935].
- Dual optimizers exist as convex potentials, providing a global duality [2205.09825, 2509.04935].
- In dimension one, explicit solutions exist in terms of quantile functions [2511.21287].

## 6. Applications Across Disciplines

- **Economics**: Models of labor market matching with aggregation effects use barycentric WOT for capturing production via team composition (firm as $x$, worker types $y$, cost is output as CES function on barycenters) [2205.09825].
- **Machine Learning**: Robust barycenters, low-regularity OT map estimation, and adversarial or continuous relaxations of statistical independence in generative modeling employ barycentric projection and related weak OT formulations [2407.13445, 2107.01718].
- **Statistics**: Barycentric projections underpin plug-in estimators for transport maps and test statistics for independence [2107.01718].
- **Numerical Linear Programming**: Fast combinatorial algorithms for the barycentric problem and its variants leverage the structure of barycentric costs and sparsity of supports [1704.05491, 1507.07218].

## 7. Connection to Classical OT and Beyond

- The barycentric OT is a generalization of classical OT; it recovers Monge-Kantorovich when the cost is linear in $y$.
- Its distinguishing feature is cost nonlinear in the conditional distribution, only convex in barycenters, which restricts original OT plans but provides a generalized matching framework.
- Its dynamic and martingale analogues connect optimal transport to stochastic processes, Markov representations, and more general convex geometries [2511.21287, 2509.04935].
- Variants include density-dependent minimal action (Lagrangian-weighted costs) [2511.02929], unbalanced extensions for robust barycenters [2410.08117], and constrained or regularized map estimation through projection [2407.13445].

---

**References:**  
- [2205.09825] Algorithms for Weak Optimal Transport with an Application to Economics  
- [2511.21287] Dynamic characterization of barycentric optimal transport problems and their martingale relaxation  
- [2509.04935] On the quadratic barycentric transport problem  
- [2407.13445] Constrained Approximate Optimal Transport Maps  
- [2107.01718] Rates of Estimation of Optimal Transport Maps using Plug-in Estimators via Barycentric Projections  
- [1704.05491] An LP-based, Strongly-Polynomial 2-Approximation Algorithm for Sparse Wasserstein Barycenters  
- [1507.07218] Discrete Wasserstein Barycenters: Optimal Transport for Discrete Data

Source: https://www.emergentmind.com/topics/barycentric-optimal-transport-problem