Diagnosing crosstalk in large-scale QPUs using zero-entropy classical shadows

TL;DR

This work addresses the challenge of diagnosing crosstalk in large-scale QPUs by introducing Zero-Entropy Classical Shadows (ZECS), a CS-based method that reconstructs small-subsystem density operators and then enforces a rank-one, pure-state representation. By computing fidelity, trace distance, and especially entanglement entropy S_{ab} from the ZECS-reconstructed states, ZECS reveals both local and non-local noise correlations that signal crosstalk. The authors validate ZECS on multiple platforms ( ibm_lagos, ibm_brisbane, ibm_fez, and ionq_forte ) with 1{,}000–6{,}000 samples, demonstrating substantial improvements in state reconstruction over standard CS and showing that ZECS-informed routing enhances a 20-qubit QAOA task by up to 33% in algorithmic lifetime and 10% in approximation ratio relative to Qiskit routing. This approach offers a scalable, measurement-efficient diagnostic tool for large QPUs, with practical impact on selecting low-crosstalk qubit subsets and guiding quantum optimization pipelines. ZECS also opens avenues for applying density-operator-level diagnostics to other platforms and for deeper investigation into the physical sources of residual noise beyond pure readout effects.

Abstract

As quantum processing units (QPUs) scale toward hundreds of qubits, diagnosing noise-induced correlations (crosstalk) becomes critical for reliable quantum computation. In this work, we introduce Zero-Entropy Classical Shadows (ZECS), a diagnostic tool that uses information of a rank-one quantum state tomography (QST) reconstruction from classical shadow (CS) information to make a crosstalk diagnosis. We use ZECS on trapped ion and superconductive QPUs, including ionq_forte (36 qubits), ibm_brisbane (127 qubits), and ibm_fez (156 qubits), using from 1,000 to 6,000 samples. With these samples, we use the ZECS to characterize crosstalk among disjoint qubit subsets across the full hardware. This information is then used to select low-crosstalk qubit subsets on ibm_fez for executing the Quantum Approximate Optimization Algorithm (QAOA) on a 20-qubit problem. Compared to the best qubit selection via Qiskit transpilation, our method improves solution quality by 10% and increases algorithmic coherence by 33%. ZECS offers a scalable and measurement-efficient approach to diagnosing crosstalk in large-scale QPUs.

Figures (13)

• Figure 1: (a) Shows the ZECS density state operator reconstruction workflow. (b) Shows the QPU layout with three sets of qubits chosen to implement some circuits. Vertices represent qubits and edges two-qubit native interactions of the QPU. (c) Shows the acquisition of the CS information to reconstruct the density state operator. After implementing the circuit, the measuring basis changes randomly from the Pauli set, and the process repeats $N$ times to get $N$ snapshots. (d) One ZECS application is the quantification of the entanglement entropy from disjoint circuits. First, the full-density state operator $\rho_T$ is reconstructed, and then the partial trace to the disjoint circuits is applied. If there is no leakage of information, the entanglement entropy $S_{AB} \approx 0$.
• Figure 2: Crosstalk simulation between subsystems $a$ and $b$. (a) A random circuit used to simulate crosstalk between subsystems $a$ and $b$. The noisy gate that represents the crosstalk connects both subsystems and has a strength parameter $\varepsilon$. (b) Entanglement entropy versus the number of shadows used for different crosstalk strengths for the simulation of the circuit in (a). The inset plot shows the fidelity for the same process.
• Figure 3: (a) The EfficientSU2 circuit used for the characterization of ibm_lagos. The circuit is repeated $n$ layers, and the parameters for each layer, the $\theta_i^n$, are randomly chosen from a uniform distribution between $0$ and $\pi/2$; (b) The layout of ibm_lagos where the vertices represent the qubits and the edges the physical connectivity between them.
• Figure 4: Infidelity versus the number of snapshots using the qasm_simulator and ibm_lagos for 7 repetitions of the protocol of Fig. \ref{['Fig:circuit_used']}(a) on qubits 0, 1, and 2. The dashed lines represent the CS and the solid lines the ZECS reconstructions of the density state operator.
• Figure 5: (a) ($1-F$) vs. $D$ using CS and ZECS for the 2-qubit experiment of the EfficientSU2 gates for $n=1$ and 10 repetitions on ibm_brisbane. The inset circuit is the EfficientSU2 gate repeated n times with random parameters $\theta_i$. Every marker represents a pair of qubit results. (b) Singular values of $\rho_{cs}$ for the bipartite subsystems in the $n=10$ EfficientSU2 experiment.