---
title: Adapted Bures–Wasserstein Space
url: https://www.emergentmind.com/topics/adapted-bures-wasserstein-space
type: topic
---

# Adapted Bures–Wasserstein Space

The adapted Bures–Wasserstein (ABW) space is a recent extension of the classical Bures–Wasserstein geometry, incorporating causality or adaptedness constraints that are fundamental in multistage stochastic systems, especially discrete-time Gaussian processes. ABW space provides a natural setting for optimal transport with temporal structure, enabling explicit distances, geodesics, and tangent spaces, and exhibiting Alexandrov non-negative curvature. At its core, the ABW space organizes Gaussian process laws as orbits of block-lower-triangular (filtration-respecting) linear mappings, modulo adapted orthogonal transformations, with a metric determined by the minimal Frobenius norm in this orbit class. This construction simultaneously generalizes the classical Bures–Wasserstein geometry of Gaussian measures/covariances and allows for rigorous geometric analysis in multistage and adapted optimal transport [2602.00623, 2404.06625].

## 1. Formal Definition and Structure

The ABW space arises from endowing the space of multivariate, discrete-time Gaussian processes with the **adapted Wasserstein distance**. Given a fixed time horizon $T \ge 1$ and state dimension $d$, a centered Gaussian process $X = (X_1, \dots, X_T)$ is represented by
\[
X_t = \sum_{s=1}^t L_{t,s}\,G_s,\quad L = (L_{t,s})_{1 \le s \le t \le T}
\]
where $G_1, \dots, G_T$ are independent $N(0, I_d)$ random vectors, and $L$ is a block-lower-triangular matrix of size $dT \times dT$. Two processes, specified by $(a, L)$ and $(b, M)$, are equivalent if their laws coincide under adapted (block-diagonal) orthogonal transformations.

The adapted Bures–Wasserstein distance $d_{\mathrm{ABW}}$ between orbits $[L]$ and $[M]$ is given by
\[
d_{\mathrm{ABW}}([L], [M]) = \inf_{O \in \mathbf{O}} \|\,L - M O\,\|_F
\]
where $\mathbf{O}$ is the group of block-diagonal orthogonal matrices, and $\|\cdot\|_F$ denotes the Frobenius norm [2602.00623]. This metric is the minimizer of a Procrustes-type problem and is well-defined on the quotient space of all such orbits.

When $T = 1$ (single-stage), the ABW space reduces to the original Bures–Wasserstein geometry for positive definite matrices; that is, $d_{\mathrm{ABW}}$ coincides with the classical Bures–Wasserstein distance.

## 2. Metric Geometry and Alexandrov Curvature

The ABW space $(\mathbf{L} / \mathbf{O}, d_{\mathrm{ABW}})$ is an Alexandrov space with non-negative curvature. The fundamental semi-concavity (triangle comparison) inequality holds for any constant-speed geodesic $\gamma$ and point $Z$ in the space:
\[
d^2(\gamma(u), Z) \geq (1-u)d^2(\gamma(0), Z) + u\,d^2(\gamma(1), Z) - u(1-u)d^2(\gamma(0), \gamma(1))
\]
This property is inherited from the geometry of $L^2$ spaces under $AW_2$-interpolations and persists under the restriction to (regular) Gaussian processes and their orbits. The resulting space is geodesically convex: the set of all nondegenerate (regular) Gaussian processes forms a closed convex subset, with well-behaved tangent cones and exponential maps [2602.00623]. The non-negative curvature facilitates contraction properties and stability for optimization and sampling algorithms in the space.

## 3. Geodesics, Tangent Cones, and Exponential Maps

In ABW space, geodesics between points $[L]$ and $[M]$ are explicit and correspond to linearly interpolated block-lower-triangular matrices, up to adaptation via minimizing block-diagonal orthogonals. Specifically, for some $P \in \mathcal{O}^*(L, M)$ achieving the infimum,
\[
\gamma(u) = [(1-u)L + u\,M P],\quad u \in [0,1]
\]
Every geodesic arises from such an interpolation, and non-uniqueness emerges only when degeneracies in the conditional covariances are present [2602.00623].

The tangent cone at $[L]$ is given by the quotient $(\mathcal{V}(L) / \mathcal{O}^*(L,L), d_{\mathcal{O}^*(L,L)})$, where
\[
\mathcal{V}(L) = \{ V \in \mathbf{L} : (V^\top L)_{t,t} \text{ is symmetric for all } t \}
\]
with an explicit identification of the pseudo-metric. In the nondegenerate (regular) case, $\mathcal{O}^*(L,L) = \{\mathrm{Id}\}$, and the tangent cone becomes a Hilbert space with inner product $\langle V, W \rangle_F$.

The exponential map is linear: for $V \in \mathcal{V}(L)$ sufficiently small,
\[
\exp_{[L]}([V]) = [L + V]
\]
The logarithm is single-valued precisely when there is a unique geodesic, i.e., when the ranks of diagonal blocks match [2602.00623].

## 4. Explicit Formulas and the Link with Bicausal Optimal Transport

For two finite-dimensional Gaussian laws $\mu = \mathcal{N}(a, \Sigma)$ and $\nu = \mathcal{N}(b, \Lambda)$ on $\mathbb{R}^N$, the ABW distance specializes to [2404.06625]:
\[
d_{\mathrm{ABW}}^2(\Sigma, \Lambda) = \operatorname{Tr} \Sigma + \operatorname{Tr} \Lambda - 2 \|\operatorname{diag}(L^\top M)\|_1
\]
where $L$, $M$ are the Cholesky factors, and the infimum is over all bicausal couplings—i.e., couplings respecting the temporal filtration. The optimal bicausal coupling is characterized by block-diagonally signed permutations applied to $M$ such that each diagonal term $(L^\top M)_{tt}$ is replaced by its modulus.

This construction is fundamentally different from the classical Bures–Wasserstein structure, as it fails to coincide with any smooth Riemannian metric and is generally incomplete for processes of rank at most two [2602.00623, 2404.06625].

## 5. Comparison with Classical Bures–Wasserstein and Related Geometries

The ABW space recovers the classical Bures–Wasserstein metric when $T=1$, wherein
\[
d_{\mathrm{BW}}^2(\Sigma, \Xi) = \operatorname{tr}(\Sigma) + \operatorname{tr}(\Xi) - 2\,\operatorname{tr}\Big((\Sigma^{1/2} \Xi \Sigma^{1/2})^{1/2}\Big)
\]
and the geodesic, tangent, and metric structures coincide with the well-established Riemannian geometry of positive-definite matrices [2001.08056, 2204.09928].

In the adapted case, the strict filtration-respecting requirement modifies both the minimal transport cost and geodesic structure, making the adapted distance strictly greater than the classical Bures–Wasserstein distance whenever causality constraints are nontrivial. The lack of a global manifold structure and the presence of non-uniqueness in geodesics for degenerate multi-stage processes further underscores this distinction [2602.00623, 2404.06625].

## 6. Applications and Implications

The ABW space is central in problems of adapted optimal transport, sequential inference, and filtering of Gaussian processes:
- It provides a rigorous metric for distances between stochastic processes with adapted structures (e.g., filtering, smoothing, and control of time-series).
- Its non-negative Alexandrov curvature ensures stability properties useful for iterative algorithms, such as barycenter computation and optimal transport interpolation.
- The special role of the "regular" subspace ensures that, for nondegenerate processes, both geodesics and tangent cones are globally well-posed and tractable.
- Explicit metrization via the block-lower-triangular Procrustes problem enables algorithmic implementations exploiting matrix factorizations and singular value decompositions [2602.00623].

A plausible implication is that, in high-dimensional temporal models (including control and signal-processing applications), the ABW space may enable both computationally efficient and statistically robust transport-based algorithms, extending the reach of optimal transport and information geometry into stochastic process settings.

## 7. Connections to Quantum Information and Matrix-Valued Transport

At $T = 1$, the ABW construction is equivalent to the Bures metric relevant in quantum information, connecting it directly to quantum fidelity and the geometry of quantum density matrices [1901.00226, 2001.08056]. Extensions to matrix-valued (e.g., operator- or tensor-valued) transport with generalized Benamou–Brenier dynamics further underline the versatility of the underlying geometric architecture [2011.05845]. The regularized and dynamically weighted analogs in the ABW framework provide a path toward unified theories of unbalanced, adapted, and quantum transport geometries.

---

*Key references: [2602.00623], [2404.06625], [2204.09928], [2001.08056], [2011.05845], [1901.00226].*

Source: https://www.emergentmind.com/topics/adapted-bures-wasserstein-space