⚛️ quantum physics

Coherent consensus for the estimation and classification of phases

Inspired by quantum metrology, this paper proposes and experimentally validates a "coherent consensus" primitive using GHZ states on trapped-ion hardware to amplify decision signals from independent quantum classifiers, achieving a $\sqrt{n}$ improvement in precision over traditional incoherent majority voting.

Original authors: Roberto Ruiz, Alessandro Luongo, Sergi Ramos-Calderer, José Ignacio Latorre

Published 2026-10-08

📖 5 min read🧠 Deep dive

CC BY 4.0

Original authors: Roberto Ruiz, Alessandro Luongo, Sergi Ramos-Calderer, José Ignacio Latorre

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

In the quiet, controlled world of quantum physics, scientists are constantly trying to measure things that are too small to see and too fragile to touch. Imagine trying to detect a magnetic field so weak it barely nudges a single atom. To do this, researchers often rely on a process called quantum metrology, which uses the strange rules of the quantum world to make measurements far more sensitive than anything possible with classical tools. At the heart of this field is a simple but powerful idea: if you have many independent sensors, you can combine their results to get a better answer. However, there is a limit to how much this helps. If you simply take the average of many separate measurements, your accuracy improves slowly, like adding more drops to a bucket to raise the water level. But if you can link those sensors together so they act as a single, unified entity, the improvement becomes dramatic. This is the difference between a crowd of people shouting their guesses and a choir singing in perfect harmony; the latter produces a sound that is far louder and clearer than the sum of individual voices.

A team of researchers at the Centre for Quantum Technologies in Singapore has now taken this concept of harmony and applied it to a new challenge: helping computers make better decisions when they are unsure. In machine learning, a computer often has to decide which category a piece of data belongs to, such as whether an image shows a cat or a dog. When the data is clear, the decision is easy. But when the data is right on the edge, near the boundary between categories, even a well-trained computer might hesitate or make a mistake. The researchers asked if they could use the power of quantum entanglement to help a group of these hesitant computers reach a better consensus. They developed a method they call "coherent consensus," which allows a group of quantum classifiers to work together not just by comparing notes, but by literally combining their signals into a single, amplified wave.

To test this idea, the team set up a scenario where a group of quantum classifiers, each slightly imperfect, tried to determine the sign of a tiny, hidden angle. Think of this angle as a secret direction that the classifiers must agree on. In a standard approach, known as majority voting, each classifier would make its own measurement, and the final answer would be decided by counting how many voted for one direction versus the other. This is the classical way of combining information. The researchers compared this to their new coherent consensus method, where the classifiers are first linked together in a special quantum state. In this state, the tiny signals from each classifier stack up on top of one another before being measured, effectively amplifying the total signal. The researchers ran these experiments on a real quantum computer built with trapped ions, which are individual atoms held in place by electric fields. They tested the system with groups of twenty, fifty, and eighty classifiers, using thousands of measurement attempts for each group.

The results showed that the new method worked exactly as the theory predicted, at least for smaller groups. When the team used twenty or fifty classifiers, the coherent consensus method was significantly more precise than the standard majority voting. The error in their measurements shrank much faster as they added more classifiers, confirming that the quantum linking successfully amplified the signal. However, the experiment also revealed a limit. When they increased the group size to eighty classifiers, the advantage began to fade. The system became so sensitive to the tiny imperfections and noise inherent in the physical hardware that the extra precision was lost to errors in the machine itself. The researchers found that while the quantum method could theoretically offer a massive boost in accuracy, the current technology used to build these quantum computers introduces enough noise to cancel out the benefits when the groups get too large.

Despite this hardware limitation, the study provides a clear path forward. The team proved that for a specific range of group sizes, the quantum method is not only more precise but also more accurate, meaning it is more likely to get the right answer. They also showed that this advantage holds true even when accounting for the extra cost of running the more complex quantum circuits. The work demonstrates that entanglement can be used as a practical tool to improve decision-making in quantum computing, effectively turning a collection of noisy, uncertain sensors into a single, highly sensitive instrument. While the current machines are not yet perfect enough to handle the largest possible groups, the experiment confirms that the principle works. It suggests that as quantum hardware improves and becomes less noisy, this technique of coherent consensus could become a standard way to boost the reliability of quantum algorithms, allowing them to solve problems that are currently too difficult for independent machines to handle alone.

Technical Summary: Coherent Consensus for Phase Estimation and Classification

Problem Statement
The paper addresses the challenge of efficiently learning information encoded in a quantum process, specifically the estimation and classification of an unknown phase $\phi$ . This problem arises in contexts such as quantum metrology, phase kickbacks, and the aggregation of outputs from an ensemble of imperfect binary classifiers. In a standard "black box" setting, learning the phase requires a minimum number of oracle calls to achieve a prescribed error. When using an ensemble of $n$ independent classifiers (or phase gates) that produce slightly imperfect estimates $\phi_j = \phi + \gamma_j$ , the goal is to determine the sign of $\phi$ (classification) or its value (estimation) with a finite shot budget. The core difficulty lies in distinguishing the target phase from statistical sampling noise and systematic deviations, particularly when the phase is small and near the decision boundary.

Methodology
The authors propose a primitive called coherent consensus, inspired by quantum metrology, to enhance the collaborative performance of independent quantum classifiers. The methodology contrasts two strategies:

Majority Voting (Incoherent): This classical strategy involves preparing $n$ qubits in parallel, applying individual phase gates $U_{\phi_j}$ , and measuring each qubit independently in the $X$ or $Y$ basis. The final estimate $\hat{\phi}_{MV}$ is the average of the individual estimates. This approach obeys the Standard Quantum Limit (SQL), where the statistical error scales as $O(1/\sqrt{nm})$ (with $m$ being the number of shots).
Coherent Consensus: This quantum strategy utilizes entanglement to amplify the signal. The circuit consists of three steps:
- Prepare a Greenberger-Horne-Zeilinger (GHZ) state on $n$ qubits.
- Apply the tensor product of the $n$ imperfect phase gates $U_{\phi_j}$ .
- Undo the GHZ preparation to collect the accumulated phases onto a single qubit.
- Measure the first qubit to estimate the sum of phases.
  The estimator $\hat{\phi}_{CC}$ is the normalized sum. This approach achieves the Heisenberg Limit, where the statistical error scales as $O(1/n\sqrt{m})$ .

The paper models the imperfect phase gates as unitaries $U_{\phi_j}$ where the phase $\phi_j$ deviates from the target $\phi$ by independent, identically distributed (i.i.d.) perturbations $\gamma_j$ bounded by $a$ . The total error is decomposed into statistical error (shot noise), systematic error (device noise), and theoretical error (training deviations/bias).

Key Contributions

Theoretical Guarantees: The authors derive rigorous bounds on the number of qubits ( $n$ $n$ ) and shots ( $m$ $m$ ) required to achieve a prescribed relative error $\epsilon$ $ϵ$ with failure probability $\delta$ $δ$ for both strategies (Theorems 1 and 2).
- For majority voting, the required resources scale linearly with $1/\epsilon^2$ .
- For coherent consensus, the required resources scale linearly with $1/(n\epsilon^2)$ , demonstrating a quadratic improvement in precision efficiency.
Decision Guarantee: The paper establishes that if the relative error is bounded within $(0, 1)$ , the estimated phase $\hat{\phi}$ will have the same sign as the true phase $\phi$ with high probability, ensuring correct binary classification (Corollary 1).
Resource Analysis: The authors introduce "Hardware Quantum Credits" (HQCs) to fairly compare the two methods. This metric accounts for the higher cost of two-qubit gates required for GHZ state preparation in coherent consensus versus the parallel single-qubit operations in majority voting.

Experimental Results
The authors implemented both strategies on a 98-qubit trapped-ion quantum computer and in noiseless simulations:

Precision: In noiseless simulations, coherent consensus demonstrated a narrower confidence interval and faster convergence to the target phase compared to majority voting, consistent with the predicted $\sqrt{n}$ gain in precision.
Hardware Performance: On the trapped-ion device, coherent consensus showed the expected $\sqrt{n}$ gain in precision for $n=20$ and $n=50$ . However, at $n=80$ , the precision gain was offset by increased dispersion due to device drift and systematic errors (specifically two-qubit gate errors and memory errors during the longer GHZ preparation).
Accuracy: Despite systematic errors, coherent consensus remained more accurate than majority voting for $n=20$ and $n=50$ across equal shot budgets and most equal credit budgets. For $n=80$ , the accumulation of systematic errors in the entangling circuit caused coherent consensus to lose its accuracy advantage over majority voting.

Significance and Claims
The paper claims that coherent consensus is a "ready-to-deploy primitive" for phase estimation and sign classification using imperfect phase gates. Its primary significance lies in:

Precision Enhancement: It provides a provable $\sqrt{n}$ reduction in sampling error compared to incoherent majority voting, effectively acting as a form of sensing inside a quantum computer.
Robustness: The method is well-suited for current quantum hardware, which offers many qubits but limited coherence time, as it minimizes circuit depth through parallelization (in the simulation) or logarithmic depth compilation (in hardware).
Practical Utility: The work demonstrates that entanglement can be leveraged to coordinate independent classifiers, reducing the sampling effort needed to reach a specified precision. The authors note that while systematic errors currently limit the advantage at very large $n$ , the primitive offers a clear advantage for moderate system sizes ( $n=20, 50$ ) typical of near-term devices.

The paper concludes that while universal approximation by single-qubit classifiers does not guarantee a computational advantage, the coordination of these classifiers via entanglement (coherent consensus) provides a non-classical gain in reducing sampling error.

More like this