Statistical and Algorithmic Foundations of Probing Quantum Systems with Compressive Measurements: A Review

Abstract: Quantum state tomography (QST) is a fundamental task in quantum information science that aims to reconstruct unknown quantum states from measurement data. However, the exponential growth of Hilbert-space dimension with system size makes full tomography of general quantum states statistically and computationally prohibitive. This challenge has motivated extensive research on structured quantum state tomography, where prior structure, such as low-rankness, tensor-network representations, shallow quantum circuits, and neural quantum states, can substantially reduce the effective degrees of freedom and enable scalable recovery. In this review, we provide a unified perspective on QST for structured quantum states through three closely related themes: compact state representations, measurement design, and computational algorithms. After reviewing common models for structured quantum states, we survey existing work on geometric preservation properties of measurement frameworks, ranging from informationally complete POVMs to randomized measurements, and their implications for sample complexity. On the algorithmic side, we review optimization methods for reconstructing structured quantum states from empirical measurements. By connecting QST with broader principles from compressive sensing, matrix sensing, and structured inverse problems, this survey highlights common theoretical foundations underlying sample complexity, measurement efficiency, and scalable recovery.

• The paper demonstrates that leveraging structured state representations like low-rank matrices, tensor networks, and shallow circuits significantly reduces the complexity of quantum state reconstruction.
• It details various measurement designs—including IC-POVMs and random Pauli measurements—that provide stable embeddings and RIP guarantees for efficient compressive recovery.
• It analyzes unified sample complexity bounds and optimization methods, while highlighting open challenges in neural quantum states and local measurement strategies.

Review of "Statistical and Algorithmic Foundations of Probing Quantum Systems with Compressive Measurements: A Review" (2605.27191)

Introduction and Motivation

Quantum state tomography (QST) is a fundamental protocol for reconstructing quantum states from experimental measurement data. The exponential scaling of Hilbert space with system size renders traditional tomography infeasible for large-scale systems. The paper provides a comprehensive review of structured QST, addressing how prior structure in physically relevant quantum states—such as low-rankness, tensor networks, shallow circuits, and neural quantum states—enables scalable recovery. The review organizes the landscape according to three tightly coupled pillars: compact state representations, measurement design, and computational algorithms. By connecting QST to broader frameworks like compressive sensing and matrix sensing, the paper elucidates common statistical and geometric principles underpinning measurement efficiency and algorithmic scalability.

Compact Representations and Structured Quantum States

The review delineates several families of structured quantum states that admit parameterizations far below the ambient Hilbert dimension. These include:

• Low-rank states: Pure or nearly pure states are described by low-rank density matrices, leading to parameterization via Burer–Monteiro factorization. The degrees of freedom scale as $O(d^n r)$ for rank-$r$ states, offering substantial complexity reduction relative to $O(d^{2n})$.
• Tensor networks: One-dimensional systems often adopt matrix product operators (MPO) and matrix product states (MPS), while projected entangled-pair operators (PEPO) generalize to higher-dimensional lattices. MPO/PEPO parameterization scales polynomially in system size ($O(nd^2 r^2)$ for MPO; $O(nd^2 r^4)$ for PEPO), contingent on bond dimension.
• Neural quantum states: Quantum states represented by neural networks (RBM, CNN, RNN, transformers, variational autoencoders) offer flexible expressivity, but their complexity and representational properties remain less systematically understood. The paper notes the open question of sample complexity and optimization in this regime.
• Shallow quantum circuits: States generated by shallow and local circuits can be parameterized efficiently, as the locality of reduced density matrices scales with circuit depth, thus avoiding exponential complexity.

The review formalizes geometric complexity via covering numbers, providing a unified scaling law: $\log N(X) = O(d^n r)$ for low-rank, $O(nd^2 r^2 \log n)$ for MPO, $O(nd^2 r^4 \log n)$ for PEPO, and $O(d^{2n})$ for generic physical states.

Measurement Design and Geometric Preservation

Measurement design fundamentally influences sample complexity and statistical efficiency. The paper characterizes several POVMs and measurement frameworks:

• Informationally complete (IC) POVMs: Spherical $t$-designs induce measurement operators that provide exact Hilbert–Schmidt isometry for all physical states, enabling recovery guarantees governed solely by state complexity.
• Random measurements: Haar-random projective measurements and unitary $r$0-designs offer compressive alternatives, with stable embeddings for structured quantum states. Higher-order designs (e.g., unitary 3- or 4-designs) yield stronger concentration and improved sample efficiency.
• Pauli measurements: Random Pauli observable measurements satisfy restricted isometry property (RIP) for low-rank states, enabling scalable tomography.

Local measurements (tensor products of local POVMs, Pauli basis measurements): These are experimentally accessible but analytical guarantees are weaker. The sample complexities typically scale exponentially with system size.

The paper formalizes distinguishability in measurement via information-geometric stable embeddings, quantifying the KL divergence between outcome distributions and connecting it to classical RIP-type bounds via local Gaussian approximation. The review highlights that, for structured models, sample complexity is governed by intrinsic parameter dimension and covering number, contingent on the measurement ensemble admitting a stable embedding.

Sample Complexity Analysis

The review provides unified sample complexity bounds for broad classes of structured states:

• IC-POVMs & spherical $r$1-designs: For state class $r$2, empirical measurements of size $r$3 suffice for $r$4-accurate reconstruction via constrained least squares. This holds for low-rank, MPO, PEPO, and other structured state families.
• Random Pauli measurements: With $r$5, Pauli observables yield RIP and enable efficient recovery. The optimal allocation of budget between variety ($r$6) and repetition ($r$7) shows that maximizing measurement diversity improves bounds.
• Unitary $r$8-designs: Higher $r$9 reduces both measurement settings and state copies required. Sample complexity scales as $O(d^{2n})$0 for unitary 3-designs and improves with higher $O(d^{2n})$1.
• Haar-random measurements for MPO/PEPO: Sample complexity scales polynomially in $O(d^{2n})$2 and bond dimensions, due to the tensor network structure, although current bounds are conservative owing to independence approximations.
• Local measurements: Sample complexity for local unitary 4-designs is $O(d^{2n})$3 and for Pauli basis measurements is $O(d^{2n})$4, reflecting lower measurement efficiency compared to global measurements.

Algorithmic Foundations and Optimization

Optimization-based approaches are surveyed, with emphasis on:

• Constrained optimization: Projected gradient descent (IHT) is canonical for low-rank QST, relying on efficient projections (via eigendecomposition) and gradient updates.
• Parameterization and nonconvex optimization: Direct gradient-based optimization over parameter space (neural networks, MPO/PEPO factors) circumvents explicit projection but yields nonconvex landscapes. Recent analyses demonstrate benign geometry and efficient convergence for low-rank/MPO models. Neural quantum states present more challenging landscapes.
• Projected classical shadows: The classical shadow method constructs unbiased estimators from random measurements and projects onto structured state space. Sample complexity for predicting $O(d^{2n})$5 observables scales as $O(d^{2n})$6 in the number of properties, but linearly in the shadow norm of the observable.

Implications, Limitations, and Open Directions

The review asserts several strong claims supported by unified sample complexity bounds:

• Structured states admit polynomial sample complexity (for MPO, PEPO, low-rank) in contrast with exponential scaling for general states.
• IC-POVMs preserve geometry exactly, yielding optimal guarantees; random measurement ensembles suffice for compressive recovery of structured states if they admit stable embedding.
• Measurement allocation and diversity significantly impact recovery bounds, with single-shot measurements maximizing theoretical guarantees.

Several limitations and open problems are highlighted:

• Neural quantum states: Sample complexity, covering numbers, and optimization landscape for neural network parameterizations remain poorly understood and present fertile ground for further research.
• Local measurements: Unified geometry and concentration theory for local measurement ensembles are lacking. Whether local measurements provide uniform guarantees across structured families is an open question, with evidence of state-dependent recovery efficiency.
• Optimization theory: Benign landscape properties are established for low-rank/tensor models; generalization to PEPO, neural states, and convergence under practical nonconvex formulations is unresolved.
• Adaptive measurement strategies: Recent work indicates exponential advantages for adaptive tomography in structured states, suggesting systematic exploration of adaptivity, especially for neural and tensor network models.
• Noise robustness: Structured priors can enhance robustness to adversarial corruption and physical measurement noise; further theoretical and algorithmic development for error mitigation is warranted.

Conclusion

This review synthesizes statistical and algorithmic theory for QST in compressive and structured measurement regimes. By elucidating the interplay of state structure, measurement geometry, and optimization, it unifies the sample complexity landscape for a broad spectrum of quantum state models. Strong recovery guarantees are established for low-rank and tensor network representations under global/random measurement designs. Outstanding challenges remain in the characterization and algorithmic exploitation of neural quantum states, the theory of local measurements, robustness, and adaptive methodologies. Progress in these areas will further enhance practical scalability and physical interpretability in quantum information science and structured inverse problems.