Stochastic Quantum Information Geometry

• Stochastic quantum information geometry is a differential-geometric framework that integrates information theory, stochastic processes, and quantum mechanics to model fluctuations and state distinguishability.
• It employs metrics, curvature, and connection structures to capture both ensemble and single-trajectory dynamics in complex quantum systems.
• Applications include quantum thermodynamics, control, and optimization, enabling advanced algorithms and protocols for efficient quantum state management.

Stochastic quantum information geometry is the synthesis of information geometry, stochastic process theory, and quantum mechanics. It provides a differential-geometric framework for modeling and analyzing fluctuations, entropy production, and dynamical evolution in quantum systems—both at the ensemble and single-trajectory levels—by endowing quantum state spaces, manifolds of statistical models, and paths of dynamical processes with appropriate metrics, curvature, and connection structures. The core insight is that quantum, classical, and stochastic phenomena can be described in terms of geometric invariants of state spaces and their evolution, with stochasticity and irreversibility emerging naturally from intrinsic curvature and metric properties rather than from external randomization.

1. Geometric Structures of Probability and Quantum State Spaces

The foundation of information geometry is the Fisher–Rao metric on the probability simplex, which quantifies the distinguishability of nearby distributions. For the discrete simplex $\Delta^{N-1}$ parameterized as $p \equiv p(\theta)$,

$$G_{ij}(p) = \sum_{k=1}^N \frac{1}{p_k} \frac{\partial p_k}{\partial \theta^i} \frac{\partial p_k}{\partial \theta^j}$$

with infinitesimal distance

$$ds^2 = \tfrac{1}{4} \sum_{k=1}^N \frac{dp_k^2}{p_k}$$

which is equivalent, under the transformation $q_k = \sqrt{p_k}$, to the Euclidean metric on the positive orthant of the unit sphere (0805.2770). This structure generalizes from the statistical manifold of classical probabilities (classical Fisher information) to quantum state spaces via monotone quantum Fisher information metrics, e.g., the Bogoliubov–Kubo–Mori (BKM) metric, symmetric logarithmic derivative (SLD), and Wigner–Yanase metrics.

For quantum systems, the quantum Fisher information matrix (QFIM) appears in several forms. In the context of parameterized quantum circuits or mixed states, it is typically defined via either the Bures/Fubini–Study metric (pure states) or, for mixed states $\rho(\theta)$, as the Hessian of quantum relative entropy or via operator monotone functions $f$, with the BKM metric corresponding to the logarithmic mean choice. These quantum metrics extend the notion of statistical distance to the geometry of the manifold of quantum states and drive the construction of natural-gradient, mirror descent, and geodesic-based optimization and dynamical protocols (Sbahi et al., 2022, Sohail et al., 2024).

2. From Classical to Quantum: Reconstruction via Information Geometry

The quantum formalism can be derived from classical information geometry by supplementing the Fisher–Rao metric with additional postulates: complementarity (doubling of degrees of freedom), metric invariance (reversible transformations as orthogonal/unitary maps), global gauge invariance (unobservability of overall phase), and measurement simulability (equivalence under composition with unitaries) (0805.2770).

The key parametrization proceeds from probabilities $p_i$ to "square-root coordinates" $Q_{2i-1} = \sqrt{p_i} \cos \theta_i$, $Q_{2i} = \sqrt{p_i} \sin \theta_i$, leading to a normalized complex vector $p \equiv p(\theta)$0. Transformations that preserve the metric in $p \equiv p(\theta)$1-space correspond to unitary (or antiunitary) maps in $p \equiv p(\theta)$2-space, establishing the emergence of complex Hilbert space, the Born rule, and Hermitian generators of evolution, uniquely determined by classical distinguishability (the Fisher–Rao metric) augmented by quantum postulates (0805.2770).

3. Stochastic Information Geometry: Ensemble and Trajectory Levels

Stochastic quantum information geometry extends beyond ensemble-averaged quantities to describe the fluctuating metrics and action associated with individual quantum trajectories. At the ensemble level, the quantum Fisher information $p \equiv p(\theta)$3 for a time-dependent mixed state $p \equiv p(\theta)$4 can be universally decomposed as

$p \equiv p(\theta)$5

where the incoherent component

$p \equiv p(\theta)$6

captures classical population dynamics, and

$p \equiv p(\theta)$7

is the metric-dependent coherent term (Bettmann et al., 2024). For individual pure-state trajectories, the Conditional Quantum Fisher Information (CQFI) is defined as

$p \equiv p(\theta)$8

and admits a decomposition into incoherent, coherent, and interference cross-terms (Melo et al., 18 Jan 2026). The trajectory-level line element becomes random: $p \equiv p(\theta)$9, and the thermodynamic length/action of individual histories obey strict Cauchy–Schwarz bounds, refining ensemble quantum speed limits.

4. Geometric Origins of Stochasticity and Quantum Dynamics

A geometric unification of stochastic and quantum dynamics is achieved by associating the stochasticity, fluctuation theorems, and entropy production to curvature invariants of evolving (possibly time-dependent) manifolds embedded in an ambient Minkowski space. The geometric Fokker–Planck equation,

$$G_{ij}(p) = \sum_{k=1}^N \frac{1}{p_k} \frac{\partial p_k}{\partial \theta^i} \frac{\partial p_k}{\partial \theta^j}$$0

features the curvature–noise correspondence: $$G_{ij}(p) = \sum_{k=1}^N \frac{1}{p_k} \frac{\partial p_k}{\partial \theta^i} \frac{\partial p_k}{\partial \theta^j}$$1 so that flat regions admit large fluctuations, while strongly curved regions suppress noise (Svintradze, 30 Mar 2026). The path probability functional (Onsager–Machlup) for stochastic dynamics and the Feynman path integral for quantum evolution both arise from the same curvature–kinetic quadratic form, which reduces to a Boltzmann weight in Euclidean signature and an oscillatory phase in Lorentzian signature. Entropy production and the Second Law are geometrically encoded via monotonicity of curvature-driven entropy functionals, with quantum irreversibility and superposition likewise stemming from geometric invariants.

5. Quantum Statistical Manifolds and Thermodynamic Geometry

Quantum generalizations of classical statistical manifolds, such as the grand-canonical ensembles for Fermi–Dirac and Bose–Einstein gases, are equipped with Fisher–Rao metrics that quantify statistical distinguishability and encode exchange effects via scalar curvature. For the two-parameter statistical manifolds $$G_{ij}(p) = \sum_{k=1}^N \frac{1}{p_k} \frac{\partial p_k}{\partial \theta^i} \frac{\partial p_k}{\partial \theta^j}$$2, the metric components are functions of polylogarithms of the fugacity, and the scalar curvature $$G_{ij}(p) = \sum_{k=1}^N \frac{1}{p_k} \frac{\partial p_k}{\partial \theta^i} \frac{\partial p_k}{\partial \theta^j}$$3—negative for Fermi–Dirac (repulsive), positive for Bose–Einstein (attractive), zero for classical gases—may or may not diverge at phase transitions, depending on the completeness of the manifold description (e.g., proper inclusion of the ground state in BEC avoids curvature singularities) (Pessoa et al., 2021). Thus, stochastic quantum information geometry offers a unified framework for analyzing phase behavior, criticality, and statistical correlations in quantum gases.

6. Applications to Quantum Thermodynamics and Quantum Control

The geometric framework enables a precise quantification of entropy production, thermodynamic length, and speed limits for transitions between quantum steady states and during quantum stochastic thermodynamic processes. For slow transitions between nonequilibrium steady states, the leading-order nonadiabatic entropy production is the quadratic path action with respect to the Kubo–Mori–Bogoliubov quantum Fisher information metric: $$G_{ij}(p) = \sum_{k=1}^N \frac{1}{p_k} \frac{\partial p_k}{\partial \theta^i} \frac{\partial p_k}{\partial \theta^j}$$4 where $$G_{ij}(p) = \sum_{k=1}^N \frac{1}{p_k} \frac{\partial p_k}{\partial \theta^i} \frac{\partial p_k}{\partial \theta^j}$$5 is identified with the KMB QFI. Minimally dissipative protocols are obtained by solving the associated geodesic equation. Upper bounds for the excess entropy flux during arbitrarily fast quantum processes are supplied by quantum geometric speed limits involving the instantaneous QFIM (Lacerda et al., 15 Jan 2025, Bettmann et al., 2024). Furthermore, geometric diagnostics cleanly distinguish and quantify quantum thermodynamic phenomena, as in the quantum Mpemba effect, by separating incoherent and coherent contributions to QFI, path lengths, and completion ratios.

7. Algorithms, Optimization, and Practical Realizations

Quantum natural gradient methods leveraging the geometry of quantum state spaces via QFIMs have demonstrated superior convergence, saddle avoidance, and sample complexity. Recent advances have enabled stochastic, single-shot compatible, and unbiased estimation of ensemble-based QFIMs (e.g., E-QFIM), facilitating efficient variational quantum optimization without costly tomography, even in the context of variational quantum algorithms and generative modeling. For instance, the Quantum Natural Stochastic Pairwise Coordinate Descent (2-QNSCD) algorithm uses batch-based estimators of the ensemble QFIM to robustly identify descent directions that are adaptive to the local geometry, yielding exponential convergence guarantees under weak regularity conditions and empirical improvements over Euclidean and block-diagonal metrics (Sohail et al., 2024). In sequential quantum stochastic processes, chained initialization strategies exploit geometric locality for efficient tracking of state sequences (Sbahi et al., 2022).

---

The theory and applications of stochastic quantum information geometry encompass the full spectrum from foundational reconstruction of quantum theory to geometric analysis of quantum statistical models, stochastic thermodynamics, manifold-based algorithms, and quantum control. The unifying perspective is that both classical stochasticity and quantum indeterminacy are not imposed by external randomness, but are a direct consequence of, and encoded by, the intrinsic geometry—curvature, metrics, and topology—of quantum state spaces and their evolution.