⚛️ quantum physics

Direct fidelity estimation through joint fiducial grouping

This paper introduces "joint fiducial grouping," a technique that partitions Pauli pairs into commuting sets to significantly reduce the measurement overhead of direct fidelity estimation for continuously parameterized quantum gates while preserving circuit context.

Original authors: Júlia Barberà-Rodríguez, Arthur Strauss

Published 2026-08-20

📖 4 min read🧠 Deep dive

CC BY 4.0

Original authors: Júlia Barberà-Rodríguez, Arthur Strauss

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Quantum computers promise to solve problems that are impossible for today's machines, but they are incredibly fragile. The bits of information they use, called qubits, are easily disturbed by their surroundings, causing them to make mistakes. To build a machine that can run complex calculations without collapsing, scientists must ensure that every single operation performed by the computer is nearly perfect. This requirement is so strict that even tiny errors, which might seem negligible in a normal circuit, can accumulate and ruin the entire process. The challenge is not just making the gates that move these bits work correctly in isolation, but ensuring they work correctly when surrounded by other operations, idle periods, and the specific wiring of the chip itself. This surrounding environment, known as the context, changes how errors behave, meaning a gate that works well in one setting might fail in another.

To fix these errors, researchers need a way to measure how accurately a quantum gate is performing within its actual working environment. Traditional methods often involve averaging out the noise or running the circuit in reverse to check for mistakes, but these approaches can hide the very context-dependent errors that need fixing. A more direct method exists, which involves preparing specific simple states before the gate and measuring them immediately after. However, for the complex, continuously adjustable gates used in modern quantum processors, this direct method becomes prohibitively expensive. It requires a unique setup for every single combination of input and output states being tested, leading to a massive number of experimental configurations that take too long to run.

A team of researchers has developed a new strategy to make this direct measurement practical again. They introduced a technique called joint fiducial grouping, which allows them to test multiple combinations of states at the same time. Instead of treating every possible input and output pair as a separate experiment, they group together those pairs that can be measured using the same basic settings. By organizing the tests this way, they can estimate the performance of the gate with far fewer distinct experimental setups. This reduction is crucial because it lowers the overhead of reconfiguring the machine, which is often the bottleneck in calibration workflows.

The researchers tested their idea using a specific family of two-qubit gates known as fSim gates, which are commonly used in superconducting quantum processors. These gates are continuously adjustable, meaning their behavior changes smoothly as their parameters are tuned, making them difficult to benchmark with older methods. Through detailed computer simulations, the team showed that their grouping strategy consistently reduces the number of unique settings required to characterize the gate. In many cases, this also reduced the total number of times the gate needed to be run to achieve a precise measurement. The improvement was most significant when the gate's behavior was concentrated in a few specific patterns rather than spread out evenly, a common trait in the structured operations used in real quantum algorithms.

Beyond just saving time, the new method offers a more honest view of the gate's performance. Because it does not require randomizing the circuit or running it in reverse, it preserves the exact context in which the gate operates. This makes it an ideal tool for calibration systems that use machine learning to automatically tune quantum hardware. In these systems, the computer needs to evaluate the gate's quality thousands of times to find the best settings. The researchers demonstrated that their grouped estimator could serve as a reliable reward signal for such a learning loop, guiding the system toward better performance without the heavy burden of reconfiguring the hardware for every single test.

The study confirms that while the fundamental difficulty of measuring quantum gates remains, the way we organize those measurements can be optimized. The grouping technique does not change the physics of the gate or the nature of the errors, but it changes the efficiency of the measurement process. It provides a practical path forward for characterizing the continuously parameterized gates that are essential for the next generation of quantum applications. By reducing the experimental overhead, this approach makes it feasible to monitor and correct errors in real-world quantum circuits, keeping the context intact while ensuring the machine operates at the high levels of precision required for fault-tolerant computing.

Technical Summary: Direct Channel Fidelity Estimation through Joint Fiducial Grouping

Problem Statement
Fault-tolerant quantum computation requires physical error rates below a specific threshold, necessitating the characterization of circuit-dependent noise inherent to the execution context of quantum gates. While Randomized Benchmarking (RB) offers scalability, it averages out coherent and context-dependent errors, obscuring the specific noise shaping of a circuit. Direct Fidelity Estimation (DFE) provides a context-preserving alternative by estimating fidelity using local Pauli preparations and measurements without altering the target circuit. However, standard DFE suffers from a significant experimental overhead: each sampled input-output Pauli pair typically requires a distinct state-preparation and measurement (SPAM) configuration. This overhead becomes prohibitive for non-Clifford gates (e.g., continuously parameterized gates like fSim), where the number of required settings grows rapidly, hindering their use in closed-loop calibration workflows.

Methodology
The authors introduce Joint Fiducial Grouping, a protocol that partitions the support of the target Pauli transfer matrix into sets of Pauli pairs with commuting input and output operators.

Grouping Mechanism: Instead of treating every Pauli pair $(\alpha, \beta)$ as a unique setting, the method groups pairs where the input Paulis $\{P_\beta\}$ and output Paulis $\{P_\alpha\}$ are mutually commuting (specifically, qubit-wise commuting). This allows a single preparation-measurement basis pair to estimate multiple Pauli-transfer coefficients simultaneously.
Estimator Construction: The authors derive an unbiased grouped estimator. They define a probability distribution over groups based on the squared norm of the target coefficients within each group. For a sampled group, they perform a finite number of shots to estimate the contribution of all pairs in that group simultaneously.
Theoretical Bounds: The paper derives finite-sample guarantees for the grouped estimator. The resource complexity is shown to depend on the Rényi-1/2 effective support of the target weights within each compatible group, rather than the nominal group size. This links the cost of estimation to entropic measures of the target's non-stabilizerness conditioned on the grouping.

Key Contributions

Joint Fiducial Grouping Protocol: A method to reduce the number of distinct input-output settings required for DFE by exploiting commutativity among Pauli pairs.
Unbiased Estimator and Guarantees: Derivation of an unbiased estimator for channel fidelity with finite-sample bounds. The authors prove that grouping never increases the required number of channel uses compared to ungrouped DFE and can reduce it when target weights are concentrated within compatible groups.
Resource Separation: The work distinguishes between two resources:
- Configuration Compression: A guaranteed reduction in the number of distinct input-output settings (control configurations), independent of the target's coefficient distribution.
- Shot Reduction: A potential reduction in the total number of channel uses (shots), which occurs only when the target Pauli-transfer weight is non-uniform within the groups (i.e., concentrated on a small effective support).
Application to Calibration: The grouped estimator is utilized as a context-sensitive reward function in a reinforcement-learning-based gate calibration workflow.

Results
The authors validate their approach through numerical simulations on the continuously parameterized two-qubit fSim( $\theta, \phi$ ) gate family:

Analytic Landscape: Across the fSim parameter space, the configuration compression factor (reduction in settings) ranges from 1.67 to 2.70. The shot improvement factor (reduction in channel uses) varies from 1 to 1.43, with significant gains appearing where target weights are concentrated within commuting groups.
Fixed-Point Estimation: At a representative point ( $\theta=\pi/8, \phi=\pi/8$ ), the grouped estimator demonstrates lower Root Mean Squared Error (RMSE) than standard DFE under both ideal and readout-noisy conditions (with linear-inversion Readout Error Mitigation). The improvement is primarily driven by variance reduction.
Context-Aware Comparison: When compared to Context-Aware Fidelity Estimation (CAFE) for repeated-cycle fidelity, grouped DFE avoids the "bootstrapping" error associated with CAFE's requirement for a reference inverse and two-design state preparation. While CAFE showed lower trial-to-trial variability in fitted parameters, grouped DFE provided a more accurate fitted one-cycle fidelity at matched budgets, particularly because it relies on local single-qubit primitives rather than entangling reference circuits.
Calibration Workflow: In a Proximal Policy Optimization (PPO) calibration loop, the grouped DFE reduced the cumulative number of distinct execution configurations (a proxy for compilation and controller-loading overhead) compared to standard DFE, while maintaining comparable convergence in the final gate fidelity.

Significance and Claims
The paper claims that joint fiducial grouping provides a practical route to lower-overhead, context-preserving fidelity estimation for continuously parameterized quantum gates.

Context Preservation: Unlike RB, the method retains information about specific circuit contexts and error mechanisms, making it suitable for adaptive calibration strategies.
Efficiency: It offers a dual advantage: it always reduces the number of distinct experimental configurations (alleviating compilation and control stack pressure) and, for structured non-Clifford targets, reduces the total channel budget required for a given accuracy.
Calibration Utility: By serving as a reward function for machine-learning-driven calibration, grouped DFE enables iterative suppression of coherent gate errors without requiring trusted noise models or randomized surrogates.
Limitations: The authors note that the shot advantage is conditional on the target's weight distribution; for targets with uniform weight across groups, only configuration compression is realized. Furthermore, the current results are simulation-based, and experimental deployment requires quantifying these resource savings on specific control stacks and hardware.

The work positions grouped DFE as a necessary tool for the characterization of near-term, utility-scale quantum processors where gate operations are non-Clifford and context-dependent errors are significant.

Technical Summary: Direct Channel Fidelity Estimation through Joint Fiducial Grouping

More like this