⚛️ quantum physics

Certifying bipartite entanglement on a superconducting processor from a corrected QAOA cost layer

This study demonstrates the successful certification of bipartite entanglement in a corrected QAOA cost layer on a nine-qubit superconducting processor through rigorous tomographic negativity and causal manipulation of device physics, while confirming that the algorithm yields no optimization advantage on classically trivial instances at greater depths.

Original authors: Carlos-Miguel Lorenzo

Published 2026-09-10

📖 6 min read🧠 Deep dive

CC BY 4.0

Original authors: Carlos-Miguel Lorenzo

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 race to build useful quantum computers, scientists are currently working with machines that are powerful but fragile. These devices, often called noisy intermediate-scale quantum processors, are made of superconducting circuits that can hold information in a state called superposition, where a bit can be both zero and one at the same time. The real magic, however, happens when two or more of these bits become linked in a phenomenon known as entanglement. When particles are entangled, the state of one instantly influences the state of the other, no matter how far apart they are. This connection is the engine that drives quantum algorithms, allowing them to solve certain problems much faster than classical computers ever could. But because these machines are so sensitive to their environment, the links between particles often break or become weak before the computer can finish its work. A major question for the field has been whether the complex circuits researchers design actually create this special link on real hardware, or if the machine is simply failing in a way that looks like success.

A new study from a research team at CUNEF University in Madrid, working on a superconducting processor at the Barcelona Supercomputing Center, provides a definitive answer to that question for a specific setup. The researchers tested a standard quantum algorithm designed to solve optimization problems, which involves arranging a series of operations to find the best solution among many possibilities. They focused on a single layer of this algorithm, a step meant to create a link between two specific qubits. Instead of just checking if the computer gave a good answer, they performed a deep, forensic examination of the machine's output. They reconstructed the exact state of the two qubits after the operation and measured a specific property that proves they were truly entangled. The result was clear: on a well-functioning connection within the chip, the machine did create genuine entanglement. The strength of this link was measured at a value of 0.077, a number that, while small, was statistically significant and confirmed to be real across six different days of testing. Crucially, when they turned off the connection between the qubits, the entanglement vanished completely, proving that the link was not an artifact of the machine's background noise or a trick of the measurement.

The study goes further by showing exactly what controls this fragile link. The researchers treated the machine like a laboratory instrument they could tune, rather than a black box. They deliberately adjusted the strength of the electrical pulse used to create the connection between the qubits. By lowering the amplitude of this pulse, they observed a direct, causal drop in the strength of the entanglement. This confirmed that the physical drive of the machine, not just the abstract design of the code, was the deciding factor. However, they also discovered that this relationship is not a simple straight line that goes on forever. The connection strength peaks at a specific point and then begins to drop if the pulse is pushed too hard, much like a radio signal that gets clearer up to a point and then distorts if the volume is turned up too high. This nuance is vital because it shows that simply increasing the power of the machine does not guarantee better results; the control must be precise.

Perhaps the most important finding of the paper is what happens when the researchers try to make the algorithm deeper and more complex. In the world of quantum computing, adding more layers to an algorithm is often seen as a way to get closer to the perfect answer. In this study, however, the researchers found that adding just two more layers to their circuit caused the entanglement to disappear entirely. The link that was strong at the first step collapsed to zero by the third step, even though the theoretical design suggested it should still be there. This collapse happened so quickly that the machine could not sustain the complex circuit. The team also tested a larger group of problems at this deeper level and found no sign of the computer solving them better than random chance. They were careful to state that the problems they chose were actually very simple and could be solved instantly by a standard computer, so there was no claim of a speed advantage. The value of the work lies not in solving a hard problem, but in rigorously proving what the machine can and cannot do.

The researchers also addressed a common pitfall in the field: the risk that a computer might appear to work because of a mistake in how the instructions were written. They discovered a flaw in a previous attempt where the code they used accidentally canceled out the very connection they were trying to build. To prevent this from happening again, they built a safety check into their process that verifies the instructions are correct before the machine even starts running. They also compared their results to a simpler method of checking for entanglement that is faster but less precise. They found that while this faster method gave a hint that entanglement was present, it was not strong enough to prove it with certainty on this specific machine. This honest assessment of the limits of different measurement tools is a key part of their contribution.

Ultimately, this paper serves as a rigorous report card for a single quantum processor. It confirms that the machine can create the essential quantum link when the conditions are right and the circuit is simple. It proves that this link is controlled by the physical pulses sent to the chip and that it vanishes if those pulses are too weak or if the circuit becomes too deep for the machine to handle. By ruling out errors in the code, showing the direct cause-and-effect of the controls, and admitting where the machine fails, the study offers a clear, unvarnished picture of the current state of quantum hardware. It shows that while the technology is capable of the fundamental physics required for quantum computing, it is still very far from being able to run the complex, multi-step programs needed for practical advantage. The work stands as a model for how to measure and report on these machines, prioritizing proof and transparency over hype.

Technical Summary: Certifying Bipartite Entanglement on a Superconducting Processor from a Corrected QAOA Cost Layer

Problem and Motivation
The Quantum Approximate Optimisation Algorithm (QAOA) is a leading heuristic for near-term quantum devices, yet hardware demonstrations often fail to prove that the intended physical coupling is realized, that genuine entanglement is prepared, or which device parameters control that entanglement. Existing studies frequently report success probabilities against random baselines without verifying that the circuit does not degenerate into a separable product state due to compilation errors or hardware faults. Furthermore, claims of entanglement are often confounded by daily calibration drifts or lack causal links between physical control knobs and the resulting quantum state. This paper addresses these gaps by treating the QAOA cost layer not as an optimization benchmark, but as a state-preparation circuit subject to rigorous, device-independent certification.

Methodology
The study was conducted on the nine-qubit superconducting processor "Red" at the Barcelona Supercomputing Center, utilizing a connected five-qubit component. The experimental design relies on pre-registration, provenance control, and a "guardrail-and-twin" architecture:

Compilation Correction and Guardrail: The author identified and disclosed a prior compilation fault where the QAOA cost layer ( $CZ \cdot RZ \cdot CZ$ ) collapsed into a single-qubit phase ( $RZ$ ) due to commutation, effectively running a separable circuit. The corrected layer uses Hadamard conjugation ( $CNOT \cdot RZ \cdot CNOT$ ) to realize the intended $ZZ$ coupling. A per-circuit statevector guardrail verifies that the compiled noiseless state matches the intended unitary (worst deviation $8.4 \times 10^{-15}$ ) before any hardware shots are taken.
Entanglement Certification: The primary certificate is the negativity of the two-qubit reduced state on a specific edge, derived from full two-qubit tomography (36,864 shots per arm per day). The negativity is computed after projecting the reconstructed density matrix to the nearest physical (positive semidefinite) state.
Controls:
- Coupling-off Twin ( $J=0$ ): A matched circuit with zero coupling strength serves as a separable baseline.
- Degrated Coupler: A physically distinct, lower-fidelity coupler on the same chip serves as a negative control.
- Randomized-Measurement Witness: An independent cross-check using local randomized measurements (classical shadows) to estimate partial-transpose moments, reported honestly regarding its statistical power.
Causal Manipulation: To move beyond observational correlation, a separate pre-registered experiment deliberately varied the CZ pulse amplitude (the "drive") on a fixed Bell preparation across a ladder of settings. This isolates the causal effect of the two-qubit drive on entanglement, distinct from the non-monotone behavior of average gate fidelity ( $F_{avg}$ ) caused by phase wrapping.
Statistics: All claims are supported by bias-corrected-and-accelerated (BCa) bootstrap confidence intervals at the day-level (for tomography) or rung-level (for causal experiments), with multiplicity corrections (Holm/Benjamini–Hochberg) applied where necessary.

Key Results

Entanglement Certification: On the "good" coupler (edge 4-5) at depth one, the corrected QAOA cost layer prepares a state with a certified negativity of 0.077 (BCa 95% CI [0.065, 0.091]). This value excludes zero on all six independent calibration days.
Control Validation: The coupling-off twin and the degraded coupler (edge 6-8) both return negativity indistinguishable from zero, confirming that the signal arises from the realized coupling and not from single-qubit biases or artifacts.
Causal Dose-Response: In the manipulation experiment, reducing the CZ pulse amplitude causally reduces the certified negativity along a positive, well-fit slope ( $R^2 = 0.83\text{--}0.96$ ) across three couplers and two calibration days. The slope is approximately 0.085 units of negativity per 1% change in amplitude. The response is a local linearization of a peaked function (negativity peaks at conditional phase $|\zeta|=\pi$ ), not a global monotone law.
Depth Collapse: The certified entanglement present at depth one collapses to zero by depth three on the primary instance, even faster than the ideal noiseless target degrades.
Optimization Signal: A pre-registered ensemble of 16 instances at depth three shows no optimization signal (one-sided Wilcoxon $p=0.15$ ), consistent with the depth collapse.
Randomized-Measurement Witness: On the primary device, the witness is consistent with entanglement but is statistically underpowered at the measured magnitude ( $W \approx 0.016$ ) against a near-pure-separable null, requiring $\approx 12,672$ settings for full power. It is reported as a corroboration, not the primary certificate.

Significance and Claims
The paper explicitly claims no quantum advantage, noting that the problem instances are classically trivial. Its value is methodological and physical:

Rigorous Certification: It provides a reproducible, pre-registered account of how a real device realizes a corrected QAOA cost layer, proving that the coupling is physically present and that the state is genuinely entangled, protected against compilation faults and circular reasoning.
Causal Device Physics: It establishes a measured causal law: the two-qubit drive amplitude directly controls the certifiable entanglement on a superconducting processor. This is demonstrated through direct manipulation rather than observational correlation.
Honest Reporting: The study highlights the limitations of current entanglement witnesses on near-term hardware, showing that a witness can be clear of a "convenient" null (Werner boundary) yet underpowered against the "threatening" null (near-pure separable states).
Scope: The findings are bounded to a single device, a single certified instance, and a specific depth. The causal law is measured on a Bell preparation, not the QAOA cost layer itself, and is a local linearization near the nominal amplitude.

The paper concludes that the "value" of the device in this context lies in realizing the coupling correctly on a healthy edge at low depth, rather than in deeper circuits or post-hoc error mitigation.

More like this