Yi-Fan Gong and Dan-Bo Zhang of South China Normal University have developed a new benchmark to measure how well quantum circuits infer conserved charge. Their work extends the study of “charge learnability” to include probabilistic weak measurements, allowing independent tuning of both measurement probability and strength. The researchers found the boundary between successful and unsuccessful charge inference is determined by the informational power of local measurement. “Our results establish informational power of local measurement as a unifying principle for charge learnability under general monitoring protocols,” the paper reports.

U(1)-Symmetric Circuits and Charge Inference Setup

This work extends beyond previous studies limited to deterministic or purely probabilistic measurements, offering a more detailed understanding of how information is extracted from quantum systems. The researchers specifically focused on U(1)-symmetric circuits, a model system for exploring charge conservation, and how effectively a hidden charge label can be inferred from measurement records. The study introduces cross entropy as a diagnostic tool to differentiate between three distinct decoder types: unbiased, biased, and antibiased.

Each decoder variant interprets the measurement records differently, impacting the accuracy of charge inference. The unbiased decoder, for instance, disregards specific details of the quantum gates involved, relying instead on an averaged value, while the biased and antibiased decoders incorporate gate-specific information in a manner that either favors or disfavors certain charge configurations. This subtle approach allows for a detailed comparison of decoding strategies and their sensitivity to the underlying quantum dynamics.

The researchers found that the choice of decoder significantly influences performance, highlighting the importance of tailoring the decoding process to the specific characteristics of the monitored circuit. Beyond decoder comparisons, the team established record-label mutual information as a benchmark for quantifying the fundamental limits of charge inference. This metric assesses the maximum amount of information available in the measurement record, regardless of the decoding algorithm employed.

By comparing the performance of different decoders against this benchmark, scientists can determine how efficiently each decoder extracts information from the system. The experimental setup involved a one-dimensional U(1)-symmetric circuit, where the conserved total charge is calculated as the sum of individual qubit charges.

The circuit evolves through a series of random charge-conserving two-qubit gates, and the hidden charge label, representing one of two neighboring charge sectors, is inferred from the resulting measurement record. The researchers used a classical learner based on the symmetric-exclusion-process to model the coarse-grained charge dynamics, effectively simulating the information flow within the system. This approach allowed them to analyze the impact of different measurement protocols and decoding strategies on the accuracy of charge inference.

The researchers detail how they formulate charge learnability within this monitored circuit, introducing the concept of measurement informational power to compare different monitoring protocols. Their analysis reveals that the learnability phase boundary is consistently determined by this local informational power, regardless of whether the measurements are probabilistic, deterministic, or generally weak.

Learnability Phase Boundary Defined by Informational Power

The ability of quantum circuits to accurately infer conserved charge relies on the informational power of local measurements, a newly defined metric that unifies different monitoring approaches. This finding allows for direct comparison of monitoring protocols using a single, quantifiable value. The work extends the analysis beyond simple, direct measurements to encompass probabilistic weak measurements, where both the probability of a measurement occurring and its strength can be independently adjusted.

By tuning these parameters, scientists can explore a wider range of monitoring strategies and their impact on charge learnability. This flexibility allows for a more detailed control over the information gained from measurements, optimizing quantum circuits for information processing tasks. The team’s analysis revealed that the learnability phase boundary remains organized by the local informational power, even when transitioning between projective and weak measurement schemes.

Record-Label Mutual Information as a Benchmark

Beyond simply establishing whether a quantum circuit can learn to infer a conserved charge, scientists are now focusing on how much information is fundamentally available for that inference, independent of the decoding method used. This diagnostic relies on the variance of the assigned probability to the true label, offering a more robust transition estimate when decoder mismatch complicates the posterior entropy distribution.

Consequently, estimates derived from mismatched decoders can be rigorously checked against this label-sensitive diagnostic, ensuring reliability in the assessment of learnability. To quantify this fundamental limit, the research team introduced record-label mutual information as a metric assessing the maximum mutual information between the hidden charge label and the complete spacetime measurement record, providing a single value representing the inherent information accessible for charge inference.

The calculation of this mutual information involves determining the likelihood of observing a specific measurement record under each candidate charge sector, beginning with a uniform ensemble of computational-basis states. This approach allows for an exact, rather than approximate, evaluation of the information content, offering a more precise understanding of the circuit’s capabilities. The researchers further validated this benchmark by comparing projective and deterministic weak measurements using the exact record-label mutual information.

Their results reveal a strong correlation between decoding accuracy and the mutual information, suggesting that the latter is a reliable predictor of performance regardless of the specific measurement technique employed. The team used circuits with six and eight qubits, averaging over ten thousand independent trajectories to ensure statistical significance in their analysis.

This work builds on the established concept of informational power, initially introduced as a means of quantifying the information gleaned from a single readout in monitored. While previous studies focused on ensemble-averaged dynamics, this research emphasizes the importance of assessing the information contained within individual measurement records.

The team’s approach uses the mutual information between the input label and the measurement outcome, maximizing this value over all possible input ensembles to determine the single-readout informational power. This refined methodology provides a more accurate and insightful measure of the information available for charge inference, contributing to a deeper understanding of the underlying principles governing learnability in monitored quantum systems.

Unifying Measurement Protocols in Charge Learnability

This nuanced approach allows for a more detailed understanding of how different algorithms process the same data, revealing subtle variations in their ability to accurately infer the hidden charge. The team’s analysis demonstrates that the performance of these decoders is intrinsically linked to the informational power of the measurements themselves, highlighting the critical role of measurement design in achieving successful charge inference.

The team’s findings establish that the local informational power of a measurement, a metric quantifying the information gained about the charge from a single measurement, is an organizing principle for the learnability phase diagram. This means that both projective measurements, which provide a definitive outcome, and weak measurements, which yield probabilistic information, can achieve similar levels of learnability if they deliver comparable local information. This principle provides a valuable framework for comparing the effectiveness of different monitoring strategies, regardless of their specific implementation.

Further refining the methodology, the researchers identified the record-label mutual information as a fundamental limit on charge inference. The team found that decoders can exhibit tendencies to favor certain charge labels, leading to systematic errors in inference. By quantifying these biases, researchers can develop more robust algorithms that mitigate their effects and improve overall accuracy. This is particularly important in noisy quantum systems, where even small biases can accumulate and significantly degrade performance.

The team’s work highlights the importance of maximizing the information content of each measurement, regardless of whether it is a strong, definitive measurement or a weaker, probabilistic one. The team’s approach defines single-readout informational power as the difference between the entropy of the signal and the entropy of the signal given the value of the Pauli Z operator.

This value, when combined with the measurement probability, creates a local informational power metric that can be used to predict the learnability phase boundary. The results show that the finite-size learnability transition in the measurement probability and measurement strength plane approximately follows constant local informational power.

The researchers also found that probabilistic weak and projective monitoring protocols exhibit similar learnability when they supply comparable local information to the measurement record, further solidifying the role of informational power as a unifying principle. This work builds on previous investigations into monitored quantum circuits, which have revealed that measurement can act as a genuine dynamical ingredient rather than a passive probe.

The competition between unitary scrambling and local measurements can drive measurement-induced phase transitions, influencing the entanglement scaling within the circuit. This research extends these findings by specifically focusing on charge learnability and the role of informational power in governing the transition between different phases.

The team’s results establish that “measurement informational power is the organizing principle of the learnability phase diagram and provide a unified framework for comparing both monitoring protocols and decoder performance.” The paper details the specific methodology used to calculate these metrics, outlining the formulation of charge learnability under probabilistic weak measurements and defining the measurement informational power. Section II of the paper introduces these concepts, while Section III presents a numerical determination of the learnability phase boundary and a comparison of the three decoder variants.

👉 More information
🗞 Unifying charge-learnability transitions in monitored U(1)-symmetric quantum circuits through informational power of local measurement
✍️ Yi-Fan Gong and Dan-Bo Zhang
🧠 DOI: http://link.aps.org/doi/10.1103/63lg-r5fg

More like this