Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Log inSign up
Challenges
Datasets
Workspaces
Papers
Discussions
Leaderboard
Blog
Job Board
Q3AS

© 2026 Aqora Quantum S.A.S.

TermsPrivacyLegal Notice

Mathematics

Reference-Measure Geometry in Quantum Parameter Estimation: When Coordinate Surrogates Optimize the Wrong Objective

Christopher P. Fulton, Lawrence V. Fulton

Published Jul 5, 2026• 0 citations
Publisher

Abstract

Quantum gate estimation and tomography pipelines routinely combine intrinsically defined likelihoods with priors or regularization terms specified in local Euclidean coordinates. This practice implicitly replaces the Haar reference measure on SU(2) with Lebesgue measure, specifying a different statistical model rather than a reparametrization of the intended one. We show that omitting the associated chart-volume factor alters the optimization objective itself, modifying its gradient field and stationary-point structure. The gradient discrepancy ∇LG−∇LE=−∇logJexp is nonzero for all v≠0 so that flat-coordinate surrogate objectives can converge to points that are non-stationary for the corresponding Haar-consistent objective even in regimes where local Gaussian approximations are assumed valid. We prove a formal non-equivalence proposition and validate a leading-order Fisher-information correction analytically and numerically. Large-scale multi-start optimization experiments (N=11,900 runs) demonstrate that the discrepancy is regime dependent and most pronounced under moderate-to-strong regularization or limited data. The fix requires a single-line modification to any gradient-based optimizer. These results identify reference-measure selection as an explicit modeling decision with direct consequences for optimization and inference in gate-set tomography, randomized benchmarking, and Bayesian gate estimation on curved parameter manifolds; quantitative validation is restricted to single-qubit systems, though the mechanism extends to any regularized optimization on a curved parameter manifold.

DOI
10.3390/math14132405

Related papers

Information-Geometric Quantum Control: Convergence Limits, Brachistochrone Speed Limits, and Shot-Noise Concentration in Quantum Natural Gradient Optimization

Ahmed Hala

We formalize the exact optimization boundaries, physical time evolution minimums, and finite-resource measurement constraints that govern the performance of the Quantum Natural Gradient (QNG) update rule on parameterized variational manifolds. First, we state and prove Theorem 4.1 to characterize the exact convergence properties of QNG updates under bounded Riemannian metrics. Crucially, we isolate the causal mechanism linking ansatz expressivity to geometric degradation, proving that the pulled-back Fubini–Study metric elements inherit the exponential O(2−N ) variance collapse dictated by Haar measure integration over random unitary circuit layouts. This scaling is shown to drive an exponential inflation of the metric condition number, establishing a clear geometric origin for the numerical tracking instabilities that stall optimization vectors in large-scale parameter spaces. Second, we formalize the Information-Geometric Quantum Brachistochrone (Theorem 4.2), proving that under instantaneous velocity limitations derived from the system’s energy variance, the absolute minimum time T required to map a state vector between two configurations is bounded strictly below by the integrated Riemannian geodesic distance Lgeodesic. This result demonstrates that flat Euclidean metrics severely underestimate physical time requirements near highly entangled boundaries where local manifold curvature dominates. Finally, we derive a sharp Shot-Noise Metric Concentration Bound (Theorem 4.3), resolving an exact cyclic trace simplification to prove that finite-shot perturbation errors concentrate according to a structural noise alignment factor κ(θ) = Tr(Σg−1)/∥∆θ∥2g . Multi-trial multinomial simulations validate this relation across all shot budgets (102 to 105), verifying that the empirical breach probability tracks safely below the analytical trace constraint line. This bound exposes a fundamental engineering trade-off: as an optimization path approaches ill conditioned regions where the minimum eigenvalue collapses (λmin → 0), the tracking framework requires a quadratic expansion of measurement resources to suppress shot noise and stabilize control loops.

•Jun 30, 2026•0 citations

End-to-end Optimization of Single-Shot Quantum Machine Learning for Bayesian Inference

Theodoros Ilias, Fangjun Hu, Marti Vives +1 more

We introduce an end-to-end optimization strategy for quantum machine learning that directly targets performance under finite measurement resources, where learning objectives are defined directly at the level of task performance. The method is applied on a Bayesian quantum metrology task since it provides a natural testbed with known fundamental limits and scaling with system size. The sampling-aware hybrid algorithm achieves a single-shot risk within 1 dB of the -20 dB Bayesian limit using 32 qubits. We extend the Bayesian framework from parameter estimation to global function inference, where the task is to infer a target function of the sensor input drawn from an arbitrary prior, and we demonstrate a clear computational-sensing advantage for direct functional inference over indirect reconstruction. We relate the corresponding Bayesian risk to the Capacity metric and argue that the Resolvable Expressive Capacity provides a natural measure of the space of functions accessible in a single shot. The resulting eigentask analysis identifies noise-robust feature combinations that yield compact estimators with improved accuracy and reduced optimization cost in resource-limited or real-time on-device settings. Going beyond the irreducible statistical uncertainty associated with finite measurement records, we study implementation-induced distortions of the sensor response focusing on representative imperfections arising from readout errors and static coherent gate disorder. We show that when these imperfections are stationary across training and inference, end-to-end optimization can adapt the quantum circuit and classical estimator to the implemented hardware response, preserving high inference performance over a broad range of imperfection strengths.

quant-pharXiv:2512.20492•Dec 24, 2025•4 citations

Physical Gauge-Orbit Geometry in Gate Set Tomography: Nonunitary Representatives Beyond Unitary Gauge Fixing

Juan Harris

Gate Set Tomography determines experimental probabilities only up to similarity transformations of the fitted gate set. In practice, reported error metrics are therefore evaluated after choosing a gauge representative, often with gauge fixing restricted to unitary transformations.This work studies the geometry of the larger physical gauge orbit and asks whether nonunitary but still physically admissible representatives can improve reported gate-quality metrics beyond unitary gauge fixing.Analytically, the paper identifies a class of unital common-isotropic models for which nonunitary gauge transformations provide no advantage, and shows how a quadratic advantage appears when common nonunitality is introduced. Exact results are also derived for pure-translation gauges, together with a gauge-invariant lower bound.The analytical mechanism is tested numerically in two complementary settings. First, a historical GST model is audited conditionally, with the model-mismatch limitation stated explicitly. Second, a synthetic CPTP benchmark is generated, refitted by maximum likelihood, and shown not to be rejected by its goodness-of-fit test. In that well-specified setting the nonunitary gauge advantage reappears and agrees with the analytical prediction to numerical precision.A paired parametric bootstrap shows that the absolute value of the reported error metric can fluctuate substantially under sampling while the difference between unitary and nonunitary gauge fixing is much more stable because the two estimates are strongly correlated. An analytical bound also quantifies the maximum relative size of the nonunitary advantage within the solvable model family.The archive contains the manuscript, numerical certificates, reproducibility scripts, provenance information, package manifests, and automated verification tools. Version 1.8.1 differs from version 1.8 only by a clarification in the smoke-test output; the manuscript and scientific results are unchanged.

•Sep 21, 2026•0 citations

One Coordinate at a Time: Convergence Guarantees for Rotosolve in Variational Quantum Algorithms

Sayantan Pramanik, M Girish Chandra

In this paper, we resolve an open question in the field of optimization algorithms for training parametrized quantum circuits: Does the popular Rotosolve algorithm converge? Until now, interpolation-based coordinate descent methods such as Rotosolve have mostly been treated as heuristics, lacking any formal convergence guarantees. We rigorously analyze Rotosolve, and show that it converges to $\varepsilon$-stationary points if the optimization landscape is non-convex and smooth; and to $\varepsilon$-suboptimal points if the objective function additionally obeys the Polyak-Lojasiewicz (PL) condition. Further, we derive explicit worst-case rates of convergence in the finite quantum measurement regime. These rates are contrasted against those from a similar coordinate-based method: Randomized Coordinate Descent (RCD). Although in the worst case their rates are, prima facie, equivalent, we present arguments for a more nuanced comparison between the two. We highlight that Rotosolve is hyperparameter-free, and implicitly uses first and second derivatives in its updates. Finally, we supplement our theoretical findings with numerical experiments from Quantum Machine Learning; and compare the performance of Rotosolve against RCD, Stochastic Gradient Descent, Simultaneous Perturbation Stochastic Approximation, and Randomized Stochastic Gradient Free methods.

quant-pharXiv:2604.25613•Apr 29, 2026•0 citations

Optimization Methods for Quantum Machine Learning

Hamid D. Ismail

This chapter examines optimization as the central practical bottleneck in quantum machine learning (QML), emphasizing that the usefulness of parameterized quantum models depends not only on expressivity but also on whether they can be trained reliably under realistic conditions. It formulates QML training as a constrained, stochastic optimization problem on curved state manifolds, where parameters define unitary evolutions rather than unconstrained Euclidean weights. This geometric perspective is used to analyze the roles of parameterized quantum circuits, generator structure, circuit depth, and quantum Fisher information in shaping trainability and convergence behavior. This chapter develops the main classes of quantum loss functions and gradient-evaluation methods, including the parameter-shift rule, Hadamard-test-based approaches, and finite-shot estimation, and then surveys both gradient-based and gradient-free optimization methods relevant to hybrid quantum-classical training. Particular attention is given to major optimization pathologies, including barren plateaus, noise-induced flat landscapes, hyperparameter sensitivity, and parameter redundancy. The discussion also emphasizes hardware-aware optimization, accounting for finite shot budgets, noise, measurement overhead, and simulation strategies. Taken together, this chapter provides a systematic framework for understanding how geometry, stochastic measurement, ansatz design, optimizer choice, and hardware constraints jointly determine the practical limits of trainability in QML.

CRC Press eBooks•Nov 4, 2026•0 citations

Fisher geometry reshapes the effect of incompatibility in multiparameter quantum estimation

Jiayu He, Matteo G. A. Paris

Multiparameter quantum estimation faces two fundamental obstacles: sloppiness, i.e., anisotropy of the quantum Fisher information matrix (QFIM) that renders some parameter directions insensitive, and incompatibility, the non-commutativity of optimal measurements for different parameters. The trade-off bound $C_T$ captures their joint impact on precision, but it has remained unclear how the distribution of incompatibility across parameter planes affects its overall cost. Here we separate the total amount of incompatibility from its location. We introduce a dimensionless quantity $G_n^{(F)}$ that measures the alignment between the incompatibility distribution and the eigenvalues of the QFIM, and show how the Frobenius scale of the incompatibility contribution factorizes. We obtain a bound and prove the incompatibility cost lies between this bound and a rank-dependent multiple thereof. We also prove that at fixed sloppiness, or equivalently fixed Fisher volume, concentrating incompatibility into a single parameter plane reduces the optimized trade-off cost because the Fisher geometry can then be reshaped to allocate more Fisher area to that plane. A qutrit $SU(2)$ encoding numerically confirms that states with larger incompatibility strength can nevertheless incur a smaller cost if the matching factor $G$ is sufficiently small. Our results establish that the distribution of incompatibility relative to the Fisher eigenbasis is a central diagnostic for multiparameter estimation, beyond the total incompatibility strength.

quant-pharXiv:2606.11343•Jun 11, 2026•3 citations
Quantum gate estimation and tomography pipelines routinely combine intrinsically defined likelihoods with priors or regularization terms specified in local Euclidean coordinates. This practice implicitly replaces the Haar reference measure on SU(2) with Lebesgue measure, specifying a different statistical model rather than a reparametrization of the intended one. We show that omitting the associated chart-volume factor alters the optimization objective itself, modifying its gradient field and stationary-point structure. The gradient discrepancy ∇LG−∇LE=−∇logJexp is nonzero for all v≠0 so that flat-coordinate surrogate objectives can converge to points that are non-stationary for the corresponding Haar-consistent objective even in regimes where local Gaussian approximations are assumed valid. We prove a formal non-equivalence proposition and validate a leading-order Fisher-information correction analytically and numerically. Large-scale multi-start optimization experiments (N=11,900 runs) demonstrate that the discrepancy is regime dependent and most pronounced under moderate-to-strong regularization or limited data. The fix requires a single-line modification to any gradient-based optimizer. These results identify reference-measure selection as an explicit modeling decision with direct consequences for optimization and inference in gate-set tomography, randomized benchmarking, and Bayesian gate estimation on curved parameter manifolds; quantitative validation is restricted to single-qubit systems, though the mechanism extends to any regularized optimization on a curved parameter manifold.
We formalize the exact optimization boundaries, physical time evolution minimums, and finite-resource measurement constraints that govern the performance of the Quantum Natural Gradient (QNG) update rule on parameterized variational manifolds. First, we state and prove Theorem 4.1 to characterize the exact convergence properties of QNG updates under bounded Riemannian metrics. Crucially, we isolate the causal mechanism linking ansatz expressivity to geometric degradation, proving that the pulled-back Fubini–Study metric elements inherit the exponential O(2−N ) variance collapse dictated by Haar measure integration over random unitary circuit layouts. This scaling is shown to drive an exponential inflation of the metric condition number, establishing a clear geometric origin for the numerical tracking instabilities that stall optimization vectors in large-scale parameter spaces. Second, we formalize the Information-Geometric Quantum Brachistochrone (Theorem 4.2), proving that under instantaneous velocity limitations derived from the system’s energy variance, the absolute minimum time T required to map a state vector between two configurations is bounded strictly below by the integrated Riemannian geodesic distance Lgeodesic. This result demonstrates that flat Euclidean metrics severely underestimate physical time requirements near highly entangled boundaries where local manifold curvature dominates. Finally, we derive a sharp Shot-Noise Metric Concentration Bound (Theorem 4.3), resolving an exact cyclic trace simplification to prove that finite-shot perturbation errors concentrate according to a structural noise alignment factor κ(θ) = Tr(Σg−1)/∥∆θ∥2g . Multi-trial multinomial simulations validate this relation across all shot budgets (102 to 105), verifying that the empirical breach probability tracks safely below the analytical trace constraint line. This bound exposes a fundamental engineering trade-off: as an optimization path approaches ill conditioned regions where the minimum eigenvalue collapses (λmin → 0), the tracking framework requires a quadratic expansion of measurement resources to suppress shot noise and stabilize control loops.
We introduce an end-to-end optimization strategy for quantum machine learning that directly targets performance under finite measurement resources, where learning objectives are defined directly at the level of task performance. The method is applied on a Bayesian quantum metrology task since it provides a natural testbed with known fundamental limits and scaling with system size. The sampling-aware hybrid algorithm achieves a single-shot risk within 1 dB of the -20 dB Bayesian limit using 32 qubits. We extend the Bayesian framework from parameter estimation to global function inference, where the task is to infer a target function of the sensor input drawn from an arbitrary prior, and we demonstrate a clear computational-sensing advantage for direct functional inference over indirect reconstruction. We relate the corresponding Bayesian risk to the Capacity metric and argue that the Resolvable Expressive Capacity provides a natural measure of the space of functions accessible in a single shot. The resulting eigentask analysis identifies noise-robust feature combinations that yield compact estimators with improved accuracy and reduced optimization cost in resource-limited or real-time on-device settings. Going beyond the irreducible statistical uncertainty associated with finite measurement records, we study implementation-induced distortions of the sensor response focusing on representative imperfections arising from readout errors and static coherent gate disorder. We show that when these imperfections are stationary across training and inference, end-to-end optimization can adapt the quantum circuit and classical estimator to the implemented hardware response, preserving high inference performance over a broad range of imperfection strengths.
Gate Set Tomography determines experimental probabilities only up to similarity transformations of the fitted gate set. In practice, reported error metrics are therefore evaluated after choosing a gauge representative, often with gauge fixing restricted to unitary transformations.This work studies the geometry of the larger physical gauge orbit and asks whether nonunitary but still physically admissible representatives can improve reported gate-quality metrics beyond unitary gauge fixing.Analytically, the paper identifies a class of unital common-isotropic models for which nonunitary gauge transformations provide no advantage, and shows how a quadratic advantage appears when common nonunitality is introduced. Exact results are also derived for pure-translation gauges, together with a gauge-invariant lower bound.The analytical mechanism is tested numerically in two complementary settings. First, a historical GST model is audited conditionally, with the model-mismatch limitation stated explicitly. Second, a synthetic CPTP benchmark is generated, refitted by maximum likelihood, and shown not to be rejected by its goodness-of-fit test. In that well-specified setting the nonunitary gauge advantage reappears and agrees with the analytical prediction to numerical precision.A paired parametric bootstrap shows that the absolute value of the reported error metric can fluctuate substantially under sampling while the difference between unitary and nonunitary gauge fixing is much more stable because the two estimates are strongly correlated. An analytical bound also quantifies the maximum relative size of the nonunitary advantage within the solvable model family.The archive contains the manuscript, numerical certificates, reproducibility scripts, provenance information, package manifests, and automated verification tools. Version 1.8.1 differs from version 1.8 only by a clarification in the smoke-test output; the manuscript and scientific results are unchanged.
In this paper, we resolve an open question in the field of optimization algorithms for training parametrized quantum circuits: Does the popular Rotosolve algorithm converge? Until now, interpolation-based coordinate descent methods such as Rotosolve have mostly been treated as heuristics, lacking any formal convergence guarantees. We rigorously analyze Rotosolve, and show that it converges to ε\varepsilonε-stationary points if the optimization landscape is non-convex and smooth; and to ε\varepsilonε-suboptimal points if the objective function additionally obeys the Polyak-Lojasiewicz (PL) condition. Further, we derive explicit worst-case rates of convergence in the finite quantum measurement regime. These rates are contrasted against those from a similar coordinate-based method: Randomized Coordinate Descent (RCD). Although in the worst case their rates are, prima facie, equivalent, we present arguments for a more nuanced comparison between the two. We highlight that Rotosolve is hyperparameter-free, and implicitly uses first and second derivatives in its updates. Finally, we supplement our theoretical findings with numerical experiments from Quantum Machine Learning; and compare the performance of Rotosolve against RCD, Stochastic Gradient Descent, Simultaneous Perturbation Stochastic Approximation, and Randomized Stochastic Gradient Free methods.
This chapter examines optimization as the central practical bottleneck in quantum machine learning (QML), emphasizing that the usefulness of parameterized quantum models depends not only on expressivity but also on whether they can be trained reliably under realistic conditions. It formulates QML training as a constrained, stochastic optimization problem on curved state manifolds, where parameters define unitary evolutions rather than unconstrained Euclidean weights. This geometric perspective is used to analyze the roles of parameterized quantum circuits, generator structure, circuit depth, and quantum Fisher information in shaping trainability and convergence behavior. This chapter develops the main classes of quantum loss functions and gradient-evaluation methods, including the parameter-shift rule, Hadamard-test-based approaches, and finite-shot estimation, and then surveys both gradient-based and gradient-free optimization methods relevant to hybrid quantum-classical training. Particular attention is given to major optimization pathologies, including barren plateaus, noise-induced flat landscapes, hyperparameter sensitivity, and parameter redundancy. The discussion also emphasizes hardware-aware optimization, accounting for finite shot budgets, noise, measurement overhead, and simulation strategies. Taken together, this chapter provides a systematic framework for understanding how geometry, stochastic measurement, ansatz design, optimizer choice, and hardware constraints jointly determine the practical limits of trainability in QML.
Multiparameter quantum estimation faces two fundamental obstacles: sloppiness, i.e., anisotropy of the quantum Fisher information matrix (QFIM) that renders some parameter directions insensitive, and incompatibility, the non-commutativity of optimal measurements for different parameters. The trade-off bound CTC_TCT​ captures their joint impact on precision, but it has remained unclear how the distribution of incompatibility across parameter planes affects its overall cost. Here we separate the total amount of incompatibility from its location. We introduce a dimensionless quantity Gn(F)G_n^{(F)}Gn(F)​ that measures the alignment between the incompatibility distribution and the eigenvalues of the QFIM, and show how the Frobenius scale of the incompatibility contribution factorizes. We obtain a bound and prove the incompatibility cost lies between this bound and a rank-dependent multiple thereof. We also prove that at fixed sloppiness, or equivalently fixed Fisher volume, concentrating incompatibility into a single parameter plane reduces the optimized trade-off cost because the Fisher geometry can then be reshaped to allocate more Fisher area to that plane. A qutrit SU(2)SU(2)SU(2) encoding numerically confirms that states with larger incompatibility strength can nevertheless incur a smaller cost if the matching factor GGG is sufficiently small. Our results establish that the distribution of incompatibility relative to the Fisher eigenbasis is a central diagnostic for multiparameter estimation, beyond the total incompatibility strength.