Optimal Overlapping Tomography

• Optimal Overlapping Tomography is a framework that leverages overlapping measurement schemes to reconstruct subsystem marginals with significant savings in measurements and computation.
• In quantum systems, it establishes that 3^k global measurement settings are sufficient to recover all k-body marginals, with analogous extensions for qudit and classical imaging contexts.
• Methodologies such as combinatorial covering, optimization of measurement order, and SDP postprocessing are used to enhance accuracy and reduce experimental overhead.

Optimal Overlapping Tomography is a set of experimental and algorithmic frameworks designed to efficiently reconstruct information about subsystems of a large system by leveraging overlapping measurement schemes. While full tomography scales exponentially with system size—rendering it infeasible for even moderate $n$—overlapping tomography targets application-relevant marginals and achieves dramatic savings in both measurement and computational cost. This efficiency extends across quantum and classical imaging modalities. In quantum settings, optimal overlapping tomography achieves the information-theoretic minimum number of global measurement settings for reconstructing all $k$-body marginals, independent of the total number of constituents. In classical imaging, optimal overlapping coverage provides improved geometric flexibility and reduced scan time through nonlinear modeling of overlapping X-ray projections.

1. Mathematical Formulation and Core Principles

Overlapping tomography is fundamentally concerned with reconstructing all $k$-body marginals $\rho_S = \operatorname{Tr}_{\overline{S}} \rho$ (for all size-$k$ subsets $S$) of a global state $\rho$ via collective measurements. In quantum systems, global product measurements—where each qubit, qudit, or mode is measured independently in a certain basis—are reused so that outcomes simultaneously inform many marginals (Cotler et al., 2019, Wei et al., 2024, Hansenne et al., 2024).

The central combinatorial structure is as follows: for qubit systems, a measurement setting is a string $g \in \{X,Y,Z\}^{n}$, representing a global product observable. Measuring $g$ yields the expectation value of all local operators $g_{i_1} \otimes \dots \otimes g_{i_k}$ for each $k$-subset $\{i_1,\dots,i_k\}$, providing one local Pauli expectation for every possible subset. The incidence matrix between local observables and global settings is bipartite, and the covering problem becomes one of minimum clique cover in this graph (Wei et al., 2024).

For qudit systems of dimension $d$, generalized Gell-Mann (GGM) matrices are used, and measurement settings are indexed by tensor products $$M = \lambda_{i_1} \otimes \cdots \otimes \lambda_{i_n}$$, where $\lambda_{i_j}$ ranges over the $d^2-1$ traceless operators plus identity (Ma et al., 15 Jan 2026).

In classical imaging, notably X-ray tomography with overlapping Beams, the forward measurement model becomes nonlinear: the intensity $I_{Dj}$ at detector pixel $j$ is a sum of exponentials $$\psi_j(x) = \sum_{k=1}^{p} \exp(-\sum_{i} \xi_{ijk} x_i)$$, reflecting overlapping contributions from simultaneous emissions (Klodt et al., 2016).

2. Measurement-Efficient Protocols and Optimality

Quantum Tomography:

The optimal number of global product measurement settings to reconstruct all $k$-body marginals in an $n$-qubit system is $3^k$, independent of $n$: $M_{\rm opt}(N,k) = 3^k$ This bound is tight: any scheme necessarily requires at least $3^k$ settings by a simple counting argument—each global setting yields at most one unique $k$-body Pauli correlation per subset, and there are $3^k \binom{n}{k}$ distinct correlations (Wei et al., 2024, Hansenne et al., 2024). Explicit constructions match this bound by cycling all possible $k$-qubit Pauli patterns across measurement settings (Wei et al., 2024). For qudit systems, the optimal scaling generalizes to $(d^2-1)^k$ settings for all $k$-body marginals (Ma et al., 15 Jan 2026).

Previous overlapping tomography protocols—based on perfect hash families and covering arrays—achieved $O(3^k \log n)$ scaling (Cotler et al., 2019, Hansenne et al., 2024), and for selected local RDMs, this scaling can be further reduced via graph coloring arguments (Araújo et al., 2021).

Classical Overlapping Tomography:

In overlapping X-ray tomography, the goal is robust 3D image reconstruction from measurements in which rays overlap at the detector. The measurement model is fundamentally nonlinear, requiring a reformulation of reconstruction algorithms and regularization strategies (Klodt et al., 2016).

3. Algorithmic Constructions and Optimization Strategies

Quantum:

• Combinatorial Covering Models:
  • For full $k$-body marginal recovery, $3^k$ measurement settings suffice (Wei et al., 2024, Hansenne et al., 2024).
  • For systems with restricted locality (e.g., nearest-neighbor interactions in a lattice), graph coloring further reduces the required settings: e.g., for planar graphs $\chi(G)\le 4$, only $9$ Pauli settings are needed for all two-body marginals (Hansenne et al., 2024).
• Measurement-Order Optimization:

Excessive switching between measurement settings can incur substantial experimental overhead. An assignment-minimizing Hamiltonian path (Held–Karp algorithm) or cluster+2-opt heuristic yields up to $50\%$ reduction in switching cost (Ma et al., 15 Jan 2026).

• Semidefinite Programming (SDP) Postprocessing:

To suppress shot noise and maintain compatibility among overlapping marginals, overlapping tomography data can be postprocessed via polynomial-size SDPs, enforcing local and partial global physicality constraints (Wang et al., 30 Jan 2025). This approach achieves factor $2$–$5$ reduction in statistical error over naïve independent tomography.

Classical:

• Nonlinear Optimization:

Reconstruction from overlapping X-ray measurements is posed as a regularized nonlinear least-squares problem. Proximal gradient (forward–backward splitting) methods—incorporating $L^1$ or total variation priors, nonnegativity, and Armijo backtracking—yield provably convergent algorithms (Klodt et al., 2016).

• Sequential Bayesian Design:

For adaptive tomography (e.g., with projected overlap), acquisition parameters (angles, positions) are optimized sequentially using A- or D-optimality criteria, exploiting posterior variances to concentrate probing on poorly covered or high-variance regions (Burger et al., 2020).

4. Local and Structured Overlapping Tomography

In practical quantum and classical architectures, attention is often restricted to local marginals (e.g., nearest neighbors). Local overlapping tomography achieves resource requirements independent of global system size $N$:

• Scaling Theorem: For a $d$-dimensional lattice and fixed connected shape of size $k$, the total number of measurement settings required is $M(d,k) = O(k^d 3^k)$ (for qubits), saturating $M(d,k)=3^k$ for block shapes (Araújo et al., 2021).
• Tiling and Shift Technique:

The lattice is partitioned into cells, each cell is measured exhaustively, and shifts are applied to ensure every translate of the shape is covered. For hypercubic blocks, all local RDMs are recovered with $3^k$ settings (Araújo et al., 2021).

• Fermionic Systems:

For fermionic lattices with Majorana modes, local marginal tomography can be accomplished in $O(1)$ measurement settings, exploiting commutation structure and edge/plaquette coloring (Araújo et al., 2021).

5. Experimental Implementations and Performance

Experimental demonstrations span both quantum and classical platforms:

System/Modality	Optimal Settings (k=2)	Platform	Experimental Savings/Performance
Qubits (general)	$9$	NMR/Photonics	$>99\%$ two-RDM fidelity; $>30\%$ time reduction
Superconducting qubits	$9$	cQED	$26$–$58\%$ fewer samples for same reconstruction error (Wei et al., 2024)
Qutrits	$8 + 56\lceil\log_8 n\rceil$	General	$50\%$ reduction in switching overhead by optimized ordering (Ma et al., 15 Jan 2026)
X-ray imaging	—	Laboratory	Robust 3D recon; up to $2\times$ higher accuracy under overlap (Klodt et al., 2016)

Quantum overlapping tomography on four-qubit NMR systems reconstructed all two-qubit RDMs with $9$ settings and $99.10\%$ fidelity, reducing experimental time by $67\%$. On noisy six- and nine-qubit superconducting devices, optimal overlapping tomography required $26$–$58\%$ fewer samples compared to previous overlapping schemes for fixed error (Wei et al., 2024). In photonic Dicke- and GHZ-state experiments, the minimal settings were used to reconstruct all two-body marginals with equally high fidelity as full-state tomography but with a logarithmic factor fewer measurements (Hansenne et al., 2024, 2207.14488). In classical X-ray overlapping tomography, the optimal nonlinear reconstruction algorithm maintained accurate recovery up to moderate overlap ($\bar{p} \approx 2$) and degraded gracefully even for substantial photon overlap (Klodt et al., 2016).

6. Extensions, Limitations, and Theoretical Guarantees

Extensions

• Qudit and hybrid system generalization: The optimal scaling to higher-dimensional systems is established via covering array correspondence, with explicit constructions and bounds for qutrits and higher $d$ (Ma et al., 15 Jan 2026).
• Structured marginals and graphs: Efficient protocols extend to arbitrary locality graphs (lattices, chains, arbitrary connectivity) by computing clique covers or coloring the interaction graph (Hansenne et al., 2024, Wei et al., 2024).
• Bayesian/Adaptive Design: Sequential optimal design methods adapt to inhomogeneous samples in X-ray imaging, optimizing coverage and reducing posterior uncertainties with each step (Burger et al., 2020).
• Shot noise mitigation: SDP-based postprocessing robustly enforces physical marginals and yields significantly tighter estimation bounds in the presence of finite data (Wang et al., 30 Jan 2025).

Theoretical Guarantees

• Optimality: The $3^k$ (resp.\ $(d^2-1)^k$ for qudits) setting requirement is information-theoretically minimal for global product measurement protocols (Wei et al., 2024, Ma et al., 15 Jan 2026).
• Partial Convexity and Convergence: In nonlinear classical overlapping tomography, the optimization problem is partially convex and globally Lipschitz, guaranteeing convergence to a minimizer at rate $O(1/t)$ (Klodt et al., 2016).
• Locality-Independence: For local RDM recovery, the number of settings is strictly independent of system size, relying only on marginal size and geometric dimension (Araújo et al., 2021).

7. Practical Recommendations and Impact

Optimal overlapping tomography has transformative impact on the scalability of both classical and quantum characterization:

• Select measurement settings via explicit covering array constructions to minimize settings and switching.
• For local tomography in lattices, use $3^k$ settings per $k$-site block, regardless of total size; for arbitrary graphs, compute coloring and use pre-computed arrays.
• In X-ray or classical imaging, employ nonlinear forward models with tailored regularization; balance overlap, scan count, and noise via Bayesian or convex design (Klodt et al., 2016, Burger et al., 2020).
• Incorporate postprocessing noise mitigation (e.g., SDP constraints) to maximally exploit overlaps and suppress unphysical effects from finite sampling (Wang et al., 30 Jan 2025).
• Optimize measurement sequences to minimize laboratory reconfiguration cost, especially in high-dimensional ($d>2$) systems (Ma et al., 15 Jan 2026).
• Adapt prior information and design to the physical locality and application constraints (quantum computing, quantum chemistry, medical imaging).

Optimal overlapping tomography delivers exponential (technique-dependent) reduction in measurement and computational requirements for subsystem characterization and has become foundational in scalable quantum system verification, process tomography, and advanced imaging applications (Hansenne et al., 2024, Wei et al., 2024, Araújo et al., 2021, Klodt et al., 2016).