Concatenated Quantum Tomography Protocols

• Concatenated quantum tomography protocols are advanced measurement strategies that blend deterministic and random primitives to reduce the exponential sample complexity of quantum characterization.
• They replace traditional QPD-based wire cuts with local tomography and rescaling-free estimation, achieving polynomial scaling in measurement cost.
• These protocols enable scalable benchmarking and shadow tomography in many-body quantum systems through efficient operator learning and tailored quantum circuit designs.

Concatenated quantum tomography protocols are advanced measurement strategies for quantum characterization and quantum computation, designed to systematically reduce sample complexity in large-scale quantum systems and hybrid quantum-classical algorithms. By replacing or augmenting fully random measurement protocols—long known to incur exponential resource blowup—with structured concatenations of deterministic and randomized tomography operations, these protocols achieve provable exponential or polynomial advantages in sample efficiency, particularly in contexts such as tree-structured circuit cutting and shadow tomography of many-body quantum states (Harada et al., 22 Dec 2025, Wu et al., 2024).

1. Overview and Motivation

Concatenated quantum tomography protocols (CQT protocols) are measurement procedures that combine local and global quantum tomography primitives in a compositional way, exploiting both deterministic and randomizing unitary operations between cuts or tomographic subroutines. The primary motivations are:

• Mitigating Exponential Overheads: Conventional tomography and circuit knitting techniques (e.g., those based on quasiprobability decompositions, QPDs) have sample complexity scaling exponentially with the number of cuts or subsystem size, typically $O(2^k)$ or worse.
• Enabling Scalable Quantum Benchmarking: Efficiently estimating expectation values and reconstructing local observables in quantum many-body systems is essential for both verification and algorithm execution beyond the reach of brute-force tomography.

The protocolic innovations summarized here demonstrate that by substituting key QPD steps with quantum tomography modules—constructed via local, biased, but rescaling-free estimators—these exponential bottlenecks can be alleviated or even replaced with polynomial scaling for important topologies (Harada et al., 22 Dec 2025).

2. Key Construction: Local Tomography-Based Wire Cuts

The essential step in CQT protocols is the replacement of QPD-based mid-circuit wire cuts by local tomography on the severed edge, followed by operator learning and rescaling-free effective channel simulation. In the tree-depth-1 case, given a state $\rho$ over registers $A$ and $B$ (dimension $d$), and a channel $\Phi$ on $C$, the protocol proceeds as follows (Harada et al., 22 Dec 2025):

• Learning the Effective Observable: Define $M = \Phi^\dagger(O_C)$, with $O_C$ the observable on $C$, satisfying $\|M\|_\infty \le 1$.
• Tomographic Estimation: Using $N_1 = O(d^3 \ln(d/\delta_1)/\eta^2)$ randomized experiments, construct an estimator $\tilde M$ with $\|\tilde M - M\|_\infty \le \eta$ with high probability.
• Rescaling-Free Simulation: Diagonalize $\tilde M = V \tilde D V^\dagger$ and simulate the effective map either quantumly (via a measure-and-prepare channel) or classically by post-processing, both incurring no multiplicative overhead and only additive bias $O(\eta)$.

This construction is iterated over $R$ distinct wire cuts in the circuit, giving a total measurement cost

$M_1 = O\left( d^3 R^3 \ln(dR)/\epsilon^2 \right)$

for additive error $\epsilon$ and confidence $\delta$.

3. Generalization to Tree-Structured Circuits

For general tree-structured circuits of depth $L$ (each node with $R$ children), the CQT protocol recursively learns and concatenates effective operators "bottom-up" using tomography at every cut:

• Propagated Error Budgeting: Let $M_{i_1 \cdots i_\ell}$ denote the effective observable at node $(i_1 \cdots i_\ell)$. Each is estimated to operator-norm error $\epsilon_\ell$, with error propagation per layer controlled so $x_\ell \le \epsilon_\ell + 2R x_{\ell+1}$.
• Sample Complexity: Per node, the required shots scale as $$N_{(i_1\cdots i_\ell)} = O\left( d^3 (2R)^{2\ell} L^2 / \epsilon^2 \ln(dR^L/\delta)\right)$$.
• Summed Measurement Cost: Over all $\sum_{\ell=1}^L R^\ell$ nodes, the total measurement requirement is

$$M_L = \tilde O \left( 4^L d^3 L^2 R^{3L} / \epsilon^2 \right)$$

or, for a complete $R$-ary tree with $K \approx R^L$,

$$M_L = \tilde O \left( d^3 K^5 / \epsilon^2 \right).$$

This polynomial in $K$ scaling is in sharp contrast to the exponential scaling in QPD-based circuit cutting methods, which scale at least as $\Omega((d+1)^R / \epsilon^2)$.

4. Contractive Unitary Constructions in Shadow Tomography

A complementary class of concatenated tomography protocols addresses the challenge of shadow tomography for high-weight Pauli observables. The protocol (Wu et al., 2024) introduces a deterministic global unitary, the contractive unitary $U_{\mathrm{ct}}$, sandwiched between layers of locally random single-qubit Clifford gates:

• Unitary Construction: $U_{\mathrm{ct}} = \prod_{i<j} U_{ij}$, where $U_{ij} = \exp[i (\pi/4) Z_i Z_j ]$ are all-to-all, mutually commuting two-qubit gates. In alternative form, $$U_{\mathrm{ct}} = (\prod_{i<j} {\mathrm{CZ}}_{ij}) \cdot \prod_i S_i^{k-1}$$, with $S = \mathrm{diag}(1,i)$.
• Protocol Steps: Each round involves independent single-qubit Clifford rotations, then $U_{\mathrm{ct}}$, then another Clifford layer, followed by computational-basis measurement.
• Reduced Shadow Norm: Under $U_{\mathrm{ct}}$, half of weight-$k$ Pauli strings contract to weight $\sim 2k/3$, while the remainder stay at weight $k$. The sample complexity scales as

$\|O\|^2_{E_{U_{\text{ct}}}} \approx 2 (1.8)^k$

as opposed to $2^k$ for fully random Clifford strategies.

The concatenation principle is general: further alternating such contractive layers with randomization can, in principle, drive Pauli-weight distributions towards lower average values, reducing the sample overhead further.

5. Experimental Implementation and Scalability

The practical realization of CQT protocols, particularly those employing contractive unitaries, is favored by hardware platforms with all-to-all connectivity, such as optical-tweezer-based atom arrays. Key features include (Wu et al., 2024):

• Parallel Gate Application: All-to-all CZ gates implementing $U_{\mathrm{ct}}$ can be grouped into $k-1$ rounds of disjoint gates, yielding $O(k)$ circuit depth for $k$-qubit subsystems.
• Gate Fidelity Requirements: Single-qubit Clifford rotations routinely exceed 99.97% fidelity, while 2-qubit gates achieve $\approx$99.5%.
• Feasible Subsystem Size: Error budgets remain modest for subsystem sizes $k$ up to at least 50–100, matching current atom-array hardware.

This suggests immediate applicability for large-scale snapshot tomography and efficient characterizations impossible with prior exponential-scaling protocols.

6. Theoretical Separation and Information-Theoretic Bounds

CQT protocols provably outperform QPD-based methods in settings where rescaling factors dominate sample-complexity costs:

• QPD Wire-Cutting Overhead: Standard Pauli mid-circuit cuts involve sampling overheads $\gamma = 4^n$, so $R$ wire cuts incur overhead $\gamma^{2R} = \Omega(d^{2R})$.
• CQT Polynomial Scaling: By replacing rescaling with controlled additive bias, overall measurement cost for CQT is polynomial in the number of cuts. Concretely, for tree-depth-1, CQT requires $O(d^3 R^3 \log(dR)/\epsilon^2)$ measurements, while QPD-based wire cutting requires at least $\Omega((d+1)^R/\epsilon^2)$ given high-rank observables (Harada et al., 22 Dec 2025).
• Information-Theoretic Lower Bounds: This exponential-versus-polynomial separation is rooted in fundamental distinguishability requirements, as established by information-theoretic analysis (Theorem 10 in (Harada et al., 22 Dec 2025)).

A plausible implication is that CQT methods are optimal, at least for polynomial-scaling applications where traditional QPD decompositions fail due to exponential rescaling.

7. Generalizations and Outlook

The conceptual framework of concatenated quantum tomography is extensible:

• Multi-Stage Contractive Circuits: In shadow tomography, further alternation of deterministic contraction and randomization layers can reduce Pauli-weight tails, though explicit sample-complexity formulas beyond single-stage contraction remain to be worked out (Wu et al., 2024).
• Higher-Connectivity Gadgets: Replacing the $U_{ij}$ building block with higher-rank Clifford gadgets may enable contraction of an even larger fraction of operator weight, contingent on available hardware connectivity.
• Hybrid Paradigms: More generally, optimal measurement ensembles for shadow- and circuit-cutting-based tomography need not be fully random but can be tailored ("random ⊗ deterministic") to contract operator size and maintain unbiasedness via interleaved randomization.

Concatenated quantum tomography protocols, by systematically assembling local, global, random, and deterministic tomography primitives, present a paradigm shift in both the theoretical and practical efficiency of quantum characterization and hybrid circuit simulation. The main distinguishing feature remains the transformation of exponential sample requirements into polynomial-complexity regimes for broad classes of quantum tasks.