Bures Speed in Quantum-State Geometry

• Bures speed is defined as the local geometric rate at which a parameterized quantum state changes, measured by the Bures metric derived from Uhlmann fidelity.
• It equals one quarter of the quantum Fisher information for full-rank states, but at rank-changing points, a continuous completion of the QFI is required.
• Bures speed is central to quantum metrology and quantum speed limits, offering experimentally accessible insights into state evolution and parameter estimation.

Bures speed is the infinitesimal geometric rate at which a parameterized quantum state moves in state space when distance is measured by the Bures metric. For a family of density operators $\rho_{\boldsymbol{\epsilon}}$, the local line element is induced from the Bures distance built from Uhlmann fidelity, and the corresponding speed is the norm of the tangent vector in this Riemannian geometry. In the formulation developed for quantum metrology, this speed is governed by the Bures metric tensor $g$ and, for full-rank states, coincides with one quarter of the quantum Fisher information matrix (QFIM); at rank-changing points, however, the equality fails, and the Bures metric becomes the continuous completion of the QFIM (Šafránek, 2016).

1. Definition in quantum-state geometry

The Bures construction starts from fidelity. For two density operators $\rho_1$ and $\rho_2$, one standard form used in the infinitesimal analysis is

$$\mathcal{F}(\rho_1,\rho_2):=\Big(\operatorname{tr}\sqrt{\sqrt{\rho_1}\,\rho_2\,\sqrt{\rho_1}}\Big)^2,$$

with associated Bures distance

$$d_B^2(\rho_1,\rho_2)=2\bigl(1-\sqrt{\mathcal{F}(\rho_1,\rho_2)}\bigr).$$

For an infinitesimal parameter displacement $d\boldsymbol{\epsilon}$, the metric tensor is defined by

$$\sum_{i,j} g^{ij}(\boldsymbol{\epsilon})\,d\epsilon_i\,d\epsilon_j :=d_B^2\!\bigl(\rho_{\boldsymbol{\epsilon}},\rho_{\boldsymbol{\epsilon}+d\boldsymbol{\epsilon}}\bigr),$$

so the Bures speed is the local norm determined by this quadratic form (Šafránek, 2016).

In the one-parameter case $\rho_\theta$, the line element reads

$$ds_B^2=g_{\theta\theta}\,d\theta^2=\frac{1}{4}H_c(\theta)\,d\theta^2,$$

where the paper introduces

$g$0

and interprets $g$1 as the “continuous quantum Fisher information matrix” (Šafránek, 2016). In this sense, Bures speed is not a separate distance function but the infinitesimal rate associated with the Bures metric.

A distinct but closely related convention appears in quantum-speed-limit work, where the geodesic distance is written as the Bures arccos distance

$g$2

For pure states this reduces to the Fubini–Study distance,

$g$3

so the Bures metric functions as the mixed-state generalization of Fubini–Study geometry (Carrasco et al., 4 Jun 2026). The local metric tensor and the finite geodesic angle therefore represent two scales of the same geometry: infinitesimal motion and endpoint separation.

2. Relation to quantum Fisher information

The standard QFIM is defined through symmetric logarithmic derivatives $g$4 satisfying

$g$5

with matrix elements

$g$6

In the eigenbasis $g$7, one has

$g$8

For non-singular $g$9, this is exactly $\rho_1$0, yielding the familiar identity

$\rho_1$1

in the single-parameter case and its matrix generalization in the multiparameter case (Šafránek, 2016).

A common misconception is that the QFIM and Bures metric are always the same object. The 2016 analysis shows that this identification is correct only away from rank changes. Where the density matrix remains full rank, there are no zero eigenvalues and no correction term, so

$\rho_1$2

At such points, Bures speed and the QFI-based infinitesimal distinguishability scale coincide exactly (Šafránek, 2016).

This equivalence underlies many metrological formulas. Because the quantum Cramér–Rao bound is written as

$\rho_1$3

the metric interpretation of the QFI is operational: it measures local distinguishability. Bures speed is the geometric expression of that distinguishability rate. The crucial qualification is that the identification requires care at singular points, where the naive QFI can jump while the Bures metric does not (Šafránek, 2016).

3. Rank-changing singularities and continuous completion

The central result of the discontinuity analysis is that the equality between QFIM and Bures metric fails exactly at points where the rank of $\rho_1$4 changes. The continuous completion is

$\rho_1$5

or equivalently

$\rho_1$6

The correction comes entirely from eigenvalues that vanish at the point of interest. Thus, the failure of the identity $\rho_1$7 is neither arbitrary nor perturbative; it is controlled precisely by the Hessians of the zero eigenvalues (Šafránek, 2016).

The same paper proves that the QFIM is discontinuous whenever

$\rho_1$8

so the jump occurs when eigenvalues are created or destroyed as parameters vary. By contrast, the continuous completion satisfies directional-limit relations such as

$\rho_1$9

which makes $\rho_2$0 the continuous version of the QFI matrix (Šafránek, 2016).

This point is the main correction to the earlier literature. Bures speed is the geometrically continuous object; the standard QFI formula is only a partial representation of it. The paper also gives a regularization procedure,

$\rho_2$1

with $\rho_2$2 a fixed full-rank state diagonal in the eigenbasis of $\rho_2$3, showing that

$\rho_2$4

This reinforces the interpretation of the Bures metric as the smooth extension of the QFI across singular strata (Šafránek, 2016).

The metrological significance is direct. The infinitesimal distinguishability relevant to estimation theory remains well defined in Bures geometry even when the naive QFI formula becomes pathological. The paper explicitly identifies this as relevant in decoherence estimation, thermal and mixed-state estimation, subsystem inference, and quantum criticality (Šafránek, 2016).

4. Geodesics, fastest evolution, and quantum speed limits

Bures speed has a global counterpart in geodesic motion. In the quantum-speed-limit formulation, the length of a mixed-state trajectory $\rho_2$5 is

$\rho_2$6

where $\rho_2$7 is the QFI along the trajectory. This yields the generalized Mandelstam–Tamm bound

$\rho_2$8

and the corresponding geometric statement is that shortest Bures geodesics are exactly the fastest evolutions when the energy-variance cost is fixed (Carrasco et al., 4 Jun 2026).

For faithful states, the shortest Bures geodesic arc between $\rho_2$9 and $$\mathcal{F}(\rho_1,\rho_2):=\Big(\operatorname{tr}\sqrt{\sqrt{\rho_1}\,\rho_2\,\sqrt{\rho_1}}\Big)^2,$$0 is known explicitly. The 2026 extension shows that non-faithful density matrices admit a limiting geodesic of the same general form, but with the polar-decomposition unitary split into a regularization-independent part and a regularization-dependent part,

$$\mathcal{F}(\rho_1,\rho_2):=\Big(\operatorname{tr}\sqrt{\sqrt{\rho_1}\,\rho_2\,\sqrt{\rho_1}}\Big)^2,$$1

The shortest-arc family is then parameterized by a contraction $$\mathcal{F}(\rho_1,\rho_2):=\Big(\operatorname{tr}\sqrt{\sqrt{\rho_1}\,\rho_2\,\sqrt{\rho_1}}\Big)^2,$$2 between kernel-related subspaces, so non-faithful states may support either a unique fastest path or infinitely many equally fast ones (Carrasco et al., 4 Jun 2026).

For pure states, the same construction reduces exactly to the Fubini–Study geodesic,

$$\mathcal{F}(\rho_1,\rho_2):=\Big(\operatorname{tr}\sqrt{\sqrt{\rho_1}\,\rho_2\,\sqrt{\rho_1}}\Big)^2,$$3

For orthogonal pure states, the family broadens to include boundary Fubini–Study geodesics, interior elliptic geodesics, and a straight diameter through mixed states (Carrasco et al., 4 Jun 2026).

The geometry of supports and kernels controls uniqueness. The paper states that a unique shortest geodesic exists when

$$\mathcal{F}(\rho_1,\rho_2):=\Big(\operatorname{tr}\sqrt{\sqrt{\rho_1}\,\rho_2\,\sqrt{\rho_1}}\Big)^2,$$4

equivalently when $$\mathcal{F}(\rho_1,\rho_2):=\Big(\operatorname{tr}\sqrt{\sqrt{\rho_1}\,\rho_2\,\sqrt{\rho_1}}\Big)^2,$$5 or $$\mathcal{F}(\rho_1,\rho_2):=\Big(\operatorname{tr}\sqrt{\sqrt{\rho_1}\,\rho_2\,\sqrt{\rho_1}}\Big)^2,$$6. In the detailed statement, this condition is proved as sufficient and conjectured to be necessary in full generality. When it fails, there are infinitely many shortest geodesic arcs, all with the same length

$$\mathcal{F}(\rho_1,\rho_2):=\Big(\operatorname{tr}\sqrt{\sqrt{\rho_1}\,\rho_2\,\sqrt{\rho_1}}\Big)^2,$$7

Thus, the global notion of “fastest Bures motion” can become non-unique on the boundary of state space, even for some non-orthogonal non-faithful states (Carrasco et al., 4 Jun 2026).

5. Operational, experimental, and stochastic manifestations

The geometric speed defined by the Bures metric can be accessed experimentally in systems with self-similar dynamics. In ultracold gases confined in a time-dependent harmonic trap with scale-invariant interactions, the scaling factor $$\mathcal{F}(\rho_1,\rho_2):=\Big(\operatorname{tr}\sqrt{\sqrt{\rho_1}\,\rho_2\,\sqrt{\rho_1}}\Big)^2,$$8 determines both fidelity and nonadiabatic energy fluctuations. For these states,

$$\mathcal{F}(\rho_1,\rho_2):=\Big(\operatorname{tr}\sqrt{\sqrt{\rho_1}\,\rho_2\,\sqrt{\rho_1}}\Big)^2,$$9

is the Bures angle, while the path length is

$$d_B^2(\rho_1,\rho_2)=2\bigl(1-\sqrt{\mathcal{F}(\rho_1,\rho_2)}\bigr).$$0

The excess Bures angle

$$d_B^2(\rho_1,\rho_2)=2\bigl(1-\sqrt{\mathcal{F}(\rho_1,\rho_2)}\bigr).$$1

quantifies departure from quantum-speed-limit saturation. The operational message is that cloud-size imaging gives $$d_B^2(\rho_1,\rho_2)=2\bigl(1-\sqrt{\mathcal{F}(\rho_1,\rho_2)}\bigr).$$2, and from $$d_B^2(\rho_1,\rho_2)=2\bigl(1-\sqrt{\mathcal{F}(\rho_1,\rho_2)}\bigr).$$3 and $$d_B^2(\rho_1,\rho_2)=2\bigl(1-\sqrt{\mathcal{F}(\rho_1,\rho_2)}\bigr).$$4 one can reconstruct the Bures angle, the energy fluctuations, the traversed path length, and the QSL itself (Campo, 2020).

Continuous quantum measurements reveal a different aspect of Bures speed. For conditioned stochastic trajectories $$d_B^2(\rho_1,\rho_2)=2\bigl(1-\sqrt{\mathcal{F}(\rho_1,\rho_2)}\bigr).$$5, the Bures angle along a single measurement record acquires a stochastic contribution, and the conditioned speed can exceed the ensemble-based QSL speed. The paper on continuous monitoring shows that standard QSLs remain valid for the ensemble-averaged state but can be violated by individual trajectories. Its differential-geometric result is

$$d_B^2(\rho_1,\rho_2)=2\bigl(1-\sqrt{\mathcal{F}(\rho_1,\rho_2)}\bigr).$$6

and, for continuous measurement of a Hermitian observable $$d_B^2(\rho_1,\rho_2)=2\bigl(1-\sqrt{\mathcal{F}(\rho_1,\rho_2)}\bigr).$$7,

$$d_B^2(\rho_1,\rho_2)=2\bigl(1-\sqrt{\mathcal{F}(\rho_1,\rho_2)}\bigr).$$8

Because the infinitesimal Bures distance scales as $$d_B^2(\rho_1,\rho_2)=2\bigl(1-\sqrt{\mathcal{F}(\rho_1,\rho_2)}\bigr).$$9 rather than $d\boldsymbol{\epsilon}$0, the induced motion is interpreted as Brownian dynamics in Hilbert space (García-Pintos et al., 2018).

A thermodynamic interpretation appears in equilibrium quantum statistical mechanics. There the Bures angle is connected to the Rényi $d\boldsymbol{\epsilon}$1-divergence, and the infinitesimal metric is expressed through amplitudes $d\boldsymbol{\epsilon}$2 obeying the parallel-transport condition

$d\boldsymbol{\epsilon}$3

with line element

$d\boldsymbol{\epsilon}$4

For commuting states,

$d\boldsymbol{\epsilon}$5

so geometric displacement is tied directly to a thermodynamic divergence measure (Hardal et al., 2016). This suggests a thermodynamic reading of Bures speed as the local rate of change of state-space distinguishability among equilibrium density operators.

6. Related extensions and generalized geometries

The local metric-speed viewpoint extends beyond finite-dimensional density operators. On the Bures–Wasserstein manifold of positive definite matrices $d\boldsymbol{\epsilon}$6, the metric tensor is

$d\boldsymbol{\epsilon}$7

where $d\boldsymbol{\epsilon}$8 solves the Sylvester equation

$d\boldsymbol{\epsilon}$9

For a smooth curve $$\sum_{i,j} g^{ij}(\boldsymbol{\epsilon})\,d\epsilon_i\,d\epsilon_j :=d_B^2\!\bigl(\rho_{\boldsymbol{\epsilon}},\rho_{\boldsymbol{\epsilon}+d\boldsymbol{\epsilon}}\bigr),$$0, the corresponding local speed is

$$\sum_{i,j} g^{ij}(\boldsymbol{\epsilon})\,d\epsilon_i\,d\epsilon_j :=d_B^2\!\bigl(\rho_{\boldsymbol{\epsilon}},\rho_{\boldsymbol{\epsilon}+d\boldsymbol{\epsilon}}\bigr),$$1

The same work expresses finite displacement through the logarithmic map,

$$\sum_{i,j} g^{ij}(\boldsymbol{\epsilon})\,d\epsilon_i\,d\epsilon_j :=d_B^2\!\bigl(\rho_{\boldsymbol{\epsilon}},\rho_{\boldsymbol{\epsilon}+d\boldsymbol{\epsilon}}\bigr),$$2

showing that the infinitesimal Bures-speed concept survives in the linearized Bures–Wasserstein geometry (Afham et al., 2024).

A transport-theoretic formulation appears in the study of Bures–Wasserstein barycenters. For $$\sum_{i,j} g^{ij}(\boldsymbol{\epsilon})\,d\epsilon_i\,d\epsilon_j :=d_B^2\!\bigl(\rho_{\boldsymbol{\epsilon}},\rho_{\boldsymbol{\epsilon}+d\boldsymbol{\epsilon}}\bigr),$$3,

$$\sum_{i,j} g^{ij}(\boldsymbol{\epsilon})\,d\epsilon_i\,d\epsilon_j :=d_B^2\!\bigl(\rho_{\boldsymbol{\epsilon}},\rho_{\boldsymbol{\epsilon}+d\boldsymbol{\epsilon}}\bigr),$$4

and the optimal transport map

$$\sum_{i,j} g^{ij}(\boldsymbol{\epsilon})\,d\epsilon_i\,d\epsilon_j :=d_B^2\!\bigl(\rho_{\boldsymbol{\epsilon}},\rho_{\boldsymbol{\epsilon}+d\boldsymbol{\epsilon}}\bigr),$$5

controls the local geometry. The differential identities

$$\sum_{i,j} g^{ij}(\boldsymbol{\epsilon})\,d\epsilon_i\,d\epsilon_j :=d_B^2\!\bigl(\rho_{\boldsymbol{\epsilon}},\rho_{\boldsymbol{\epsilon}+d\boldsymbol{\epsilon}}\bigr),$$6

make the linearization of the transport map the effective infinitesimal “velocity operator” in this geometry (Kroshnin et al., 2019).

For stochastic processes, the adapted Bures–Wasserstein space replaces density matrices by Gaussian processes represented through block-lower-triangular matrices. There the relevant notion of speed is explicitly metric-geometric: geodesics are constant-speed curves. If $$\sum_{i,j} g^{ij}(\boldsymbol{\epsilon})\,d\epsilon_i\,d\epsilon_j :=d_B^2\!\bigl(\rho_{\boldsymbol{\epsilon}},\rho_{\boldsymbol{\epsilon}+d\boldsymbol{\epsilon}}\bigr),$$7, then

$$\sum_{i,j} g^{ij}(\boldsymbol{\epsilon})\,d\epsilon_i\,d\epsilon_j :=d_B^2\!\bigl(\rho_{\boldsymbol{\epsilon}},\rho_{\boldsymbol{\epsilon}+d\boldsymbol{\epsilon}}\bigr),$$8

so the speed is the Frobenius norm of the tangent direction (Acciaio et al., 31 Jan 2026). This is not the same object as Bures speed for density operators, but it is a direct analogue in an adapted optimal-transport setting.

Finally, the Bures metric also has a global-topological role distinct from infinitesimal speed. On density spaces over a unital $$\sum_{i,j} g^{ij}(\boldsymbol{\epsilon})\,d\epsilon_i\,d\epsilon_j :=d_B^2\!\bigl(\rho_{\boldsymbol{\epsilon}},\rho_{\boldsymbol{\epsilon}+d\boldsymbol{\epsilon}}\bigr),$$9-algebra with faithful trace, the topology induced by the $\rho_\theta$0-norm is finer than the Bures topology, with equality of the induced topologies in finite dimensions, although metric equivalence can fail (Aguilar et al., 2024). A plausible implication is that local “speed” comparisons between Bures and norm-based geometries can be delicate even when the underlying topologies agree.

Bures speed is therefore best understood as a family of closely related notions centered on one principle: quantum-state motion is measured by the local norm induced by Bures geometry. In finite-dimensional quantum statistics this norm is $\rho_\theta$1, with $\rho_\theta$2 rather than always $\rho_\theta$3; in quantum speed limits it determines the geodesic paths that saturate Mandelstam–Tamm bounds; in experiments it can be reconstructed from accessible observables under special dynamics; and in broader Bures–Wasserstein settings it persists as the norm of a tangent vector, transport displacement, or geodesic direction (Šafránek, 2016).