---
title: Bures–Wasserstein Distance
url: https://www.emergentmind.com/topics/bures-wasserstein-distance
type: topic
---

# Bures–Wasserstein Distance

The Bures–Wasserstein distance is a canonical metric on the space of symmetric (or Hermitian) positive-definite or positive semi-definite matrices, unifying notions from quantum information theory, optimal transport, and Riemannian geometry. Formally, it arises as the geodesic distance on the manifold of positive-definite matrices equipped with a specific Riemannian structure, and coincides with the 2-Wasserstein distance between centered Gaussian laws parameterized by those matrices. It also connects directly to quantum fidelity and appears as a central object in statistical learning, optimization, and quantum statistical inference.

## 1. Formal Definition and Equivalent Characterizations

Let $A,B$ be $n \times n$ symmetric (or Hermitian) positive-definite, or more generally positive semi-definite, matrices. The Bures–Wasserstein distance is given by
\[
d_{BW}(A,B)^2 = \operatorname{Tr}(A) + \operatorname{Tr}(B) - 2\,\operatorname{Tr}\Bigl((A^{1/2} B\,A^{1/2})^{1/2}\Bigr).
\]
Equivalent characterizations include:
- **Procrustes form:** $d_{BW}(A,B) = \min_{U\in O(n)} \|A^{1/2} - B^{1/2} U\|_F$, where $\|\cdot\|_F$ is the Frobenius norm and $O(n)$ is the orthogonal (or unitary) group [2001.08056, 1712.01504].
- **Optimal transport:** $d_{BW}^2(A,B)$ coincides with the $2$-Wasserstein distance between zero-mean Gaussians with covariances $A$ and $B$ [2210.14671, 2509.23162].
- **Quantum information:** For density matrices (trace-one Hermitian PSD matrices) $\rho, \sigma$, the Bures distance specializes to $d_{B}^2(\rho,\sigma) = 2\bigl(1 - \sqrt{\mathcal{F}(\rho,\sigma)}\bigr)$, where $\mathcal{F}$ is the quantum fidelity [1901.00226].
- **SDP formulation:** An exact convex semidefinite program representation is available for $d_{BW}$ and its barycenter [2303.03883].

## 2. Riemannian Geometry and Geodesics

The manifold of positive-definite matrices carries a canonical Riemannian metric such that $d_{BW}$ is its geodesic distance. Tangent vectors $H,K$ at $X$ have pairing
\[
g_X(H,K) = \frac{1}{2} \operatorname{Tr}(H\, (\mathcal{L}_X(K))),
\]
where $\mathcal{L}_X$ is the Lyapunov operator: $X \mathcal{L}_X(H) + \mathcal{L}_X(H) X = H$ [2001.08056, 2110.10464]. The geodesic from $A$ to $B$ is given by
\[
\Sigma(t) = \bigl( (1-t)I + t\,M \bigr) A \bigl( (1-t)I + t\,M \bigr), \qquad M = (A^{-1/2} B A^{-1/2})^{1/2},
\]
and its length realizes $d_{BW}(A,B)$ [2001.08056]. For covariance matrices of different ranks, the set of BW-minimizing geodesics can be parametrized explicitly; uniqueness fails precisely when the intersection rank is not minimal [2204.09928].

## 3. Fundamental Metric and Analytic Properties

The Bures–Wasserstein distance is a genuine metric:
- **Nonnegativity and symmetry:** $d_{BW}(A,B) \ge 0$, $d_{BW}(A,B) = d_{BW}(B,A)$;
- **Triangle inequality:** $d_{BW}(A,C) \le d_{BW}(A,B) + d_{BW}(B,C)$;
- **Strict convexity:** $A \mapsto d_{BW}^2(A,B)$ is strictly convex on the cone of positive semidefinite matrices;
- **Extension to PSD matrices:** The mapping remains well-defined on the boundary (rank-deficient case) by continuity of the principal square root [2302.14618, 1901.00226].

On the trace-one (quantum state) subset, the BW metric induces the so-called Bures–Wasserstein angle,
\[
d_{BW}^{(1)}(\rho_1,\rho_2) = \arccos\left( \operatorname{Tr} \left( (\rho_2^{1/2} \rho_1 \rho_2^{1/2})^{1/2} \right) \right),
\]
paralleling the Fisher–Rao angle for classical measures [2001.08056].

## 4. Bures–Wasserstein Barycenter (Fréchet Mean) Theory

Given matrices $A_1, \dots, A_m$ (or, in the probabilistic setting, a law $\mathbb{P}$ on positive semidefinite matrices), the BW barycenter is the unique minimizer
\[
\overline{A} = \arg\min_{X \succeq 0} \sum_{i=1}^m w_i d_{BW}^2(X,A_i),
\]
which can be characterized by the fixed-point equation
\[
\overline{A} = \sum_{i=1}^m w_i (\overline{A}^{1/2} A_i \overline{A}^{1/2})^{1/2}.
\]
Practical computation is achieved by fixed-point or gradient-based iteration with guaranteed global convergence due to strict convexity [1712.01504, 2210.14671]. Statistical theory (CLT, concentration, empirical barycenters) is fully developed when the barycenter is estimated from random samples, extending the classical $W_2$ framework to matrices [1901.00226].

## 5. Optimal Transport and Connections to Gaussian Geometry

The Bures–Wasserstein metric is intimately tied to optimal mass transport: for $\mu_1 = \mathcal{N}(m_1,\Sigma_1)$, $\mu_2 = \mathcal{N}(m_2,\Sigma_2)$,
\[
W_2^2(\mu_1, \mu_2) = \|m_1 - m_2\|^2 + d_{BW}^2(\Sigma_1, \Sigma_2),
\]
where optimal coupling is realized by a linear map determined by $\Sigma_1^{1/2} \Sigma_2 \Sigma_1^{1/2}$ [2509.23162, 2210.14671].

Geodesics and the Riemannian structure associated with $d_{BW}$ equip the Gaussian family with explicit geodesic curves, closed-form distances, and tractable mean and barycenter computations, affording concrete algorithmic advantages in applications [2304.05398, 2602.04272].

## 6. Generalizations and Weighted Variants

Multiple generalizations of the Bures–Wasserstein geometry have been developed:
- **Generalized Bures–Wasserstein geometry (GBW):** A one-parameter deformation parameterized by $M \in \mathrm{SPD}_n$, with metric and distance
\[
d_{GBW}^M(A,B)^2 = \operatorname{Tr}(M^{-1}A) + \operatorname{Tr}(M^{-1}B) - 2\,\operatorname{Tr}\left( (M^{-1}A M^{-1}B)^{1/2} \right),
\]
adapting the cost from Euclidean to Mahalanobis norm; explicit formulas for gradients, exponential/logarithm maps, and curvature are available [2110.10464].
- **Weighted and matrix-valued optimal transport:** The weighted Wasserstein–Bures distance incorporates reaction and transport dynamics for matrix-valued measures, reducing to $d_{BW}$ in the reaction-only regime and providing a metric on matrix-valued measure spaces [2011.05845].

## 7. Applications and Computational Practices

The Bures–Wasserstein distance underpins numerous applications:
- **Machine Learning:** Classification, clustering, and averaging of covariance descriptors (SPD matrices) utilize the Bures mean and barycenter in BCI, vision, and domain adaptation tasks—offering robustness and speed advantages over affine-invariant metrics [2302.14618].
- **GANs:** Penalizing generators by $d_{BW}$ in feature space improves diversity and mitigates mode collapse [2006.09096].
- **Variational Inference:** Optimizing evidence lower bounds in Bures–Wasserstein space yields more stable and effective algorithms, particularly with importance sampling, as gradients remain stable for large sample sizes [2602.04272].
- **Low-rank and convex optimization:** Fixed-point and SDP-based formulations allow efficient computation in matrix recovery and learning tasks [2303.03883].
- **Quantum Information:** Bures–Wasserstein distance characterizes distances and fidelities between quantum states and is central in quantum statistical inference [1901.00226, 2410.04937].

Algorithmic schemes for barycenter and geodesic computation exploit explicit geodesics, gradient flows, and proximal/JKO steps, often accelerated compared to alternative metrics. Empirical studies support the theoretical robustness of BW-based averaging and inference in high-dimensional, low-rank, or near-singular regimes [2302.14618].

## 8. Relationships with Classical and Quantum Geometries

The Bures–Wasserstein paradigm offers a non-commutative analog of Fisher–Hellinger geometry:
- **Classical limit:** For diagonal matrices (probability vectors), $d_{BW}$ recovers the Hellinger distance, and the angle on the trace-one subset yields the Fisher–Rao metric [2001.08056].
- **Quantum generalization:** Lifts from diagonal to general Hermitian matrices replace the orthogonal by unitary group, leading to a Riemannian submersion framework where quantum fidelity, Bures angle, and generalized quantum Rényi divergences are all sources of invariant distances along the BW geometry [2410.04937].
- **Comparative geometry:** The BW structure provides a flat cone metric over the normalized sphere, a result mirrored in the cone construction for matrix-valued measure spaces [2011.05845].

## 9. Computational and Algorithmic Considerations

Efficient computation of $d_{BW}$ leverages eigendecomposition, matrix square roots, and convex optimization:
- $O(n^3)$ complexity for dense matrices via eigen/SVD methods; reduced complexity for low-rank or diagonal cases.
- Newton–Schulz and polynomial iterations improve practical performance in large dimensions [2006.09096].
- Robutness: BW barycenters do not require full rank, avoid inverses, and are stable under near-singular perturbations [2302.14618, 2210.14671].
- SDP formulations make $d_{BW}$ readily incorporable as a constraint or regularizer in convex programs [2303.03883].

## 10. Extensions: Block Structures, Generalized Fidelities, and Multivariate Generalizations

Recently, the geometric underpinnings of $d_{BW}$ have been extended to support:
- **Generalized fidelities:** Unified block-matrix and Riemannian-geometric formalisms yield new families of quantum fidelities, with invariance and overlap properties, extending the Uhlmann, Holevo, and Matsumoto fidelities [2410.04937].
- **Multivariate and $\alpha$-Rényi divergences:** The generalized BW metric by linearization at a base point underpins a class of distances applicable to multivariate and quantum Rényi divergence settings, broadening the analytic scope [2410.04937].

---

**References:**  
- "Bures–Wasserstein geometry for positive-definite Hermitian matrices and their trace-one subset" [2001.08056]  
- "Learning with symmetric positive definite matrices via generalized Bures–Wasserstein geometry" [2110.10464]  
- "Statistical inference for Bures–Wasserstein barycenters" [1901.00226]  
- "Barycenter Estimation of Positive Semi-Definite Matrices with Bures–Wasserstein Distance" [2302.14618]  
- "Bures–Wasserstein Barycenters and Low-Rank Matrix Recovery" [2210.14671]  
- "Critical Points and Convergence Analysis of Generative Deep Linear Networks Trained with Bures–Wasserstein Loss" [2303.03027]  
- "The Bures Metric for Generative Adversarial Networks" [2006.09096]  
- "A note on the Bures–Wasserstein metric" [2303.03883]  
- "Dense associative memory on the Bures–Wasserstein space" [2509.23162]  
- "On Wasserstein distances for affine transformations of random vectors" [2310.03945]  
- "Riemannian-geometric generalizations of quantum fidelities and Bures–Wasserstein distance" [2410.04937]  
- "Forward-backward Gaussian variational inference via JKO in the Bures–Wasserstein Space" [2304.05398]  
- "Bures–Wasserstein minimizing geodesics between covariance matrices of different ranks" [2204.09928]  
- "On a general matrix-valued unbalanced optimal transport problem" [2011.05845]  
- "Bures–Wasserstein Importance-Weighted Evidence Lower Bound: Exposition and Applications" [2602.04272]  
- "On the Bures–Wasserstein distance between positive definite matrices" [1712.01504]

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