Student or educator? Sign up with your school email and get 6 months of Pro free.
CoChat

The DOI matches its Crossref record; 61 of the 61 references it cites were found in a registry or an index; 12 of the 32 sources its claims cite were checked for support.

Towards violations of Local Friendliness with quantum computers

Open access
Zeng, William J., Labib, Farrokh, Russo, Vincent · 2025 · Quantum

Abstract

Local Friendliness (LF) inequalities follow from seemingly reasonable assumptions about reality: (i) ``absoluteness of observed events'' (e.g., every observed event happens for all observers) and (ii) ``local agency'' (e.g., free choices can be made uncorrelated with other events outside their future light cone). Extended Wigner's Friend Scenario (EWFS) thought experiments show that textbook quantum mechanics violates these inequalities. Thus, experimental evidence of these violations would make these two assumptions incompatible. In [Nature Physics 16, 1199 (2020)], the authors experimentally implemented an EWFS, using a photonic qubit to play the role of each of the ``friends'' and measured violations of LF. One may question whether a photonic qubit is a physical system that counts as an ``observer'' and thereby question whether the experiment's outcome is significant. Intending to measure increasingly meaningful violations, we propose using a statistical measure called the ``branch factor'' to quantify the ``observerness'' of the system. We then encode the EWFS as a quantum circuit such that the components of the circuit that define the friend are quantum systems of increasing branch factor. We run this circuit on quantum simulators and hardware devices, observing LF violations as the system sizes scale. As errors in quantum computers reduce the significance of the violations, better quantum computers can produce better violations. Our results extend the state of the art in proof-of-concept experimental violations from branch factor 0.0 to branch factor 16.0. This is an initial result in an experimental program for measuring LF violations at increasingly meaningful branch factors using increasingly more powerful quantum processors and networks. We introduce this program as a fundamental science application for near-term and developing quantum technology.

Overview

CoChat Summary

## TL;DR

The paper proposes and demonstrates an experimental program for testing Local Friendliness (LF) violations with quantum computers, using the branch factor to quantify how “observer-like” the friend system is. It reports LF violations on simulators and hardware, extending proof-of-concept violations from branch factor 0.0 to branch factor 16.0.

## Key findings

  • The paper uses the semi-Brukner LF inequality, stated as

\(-\langle A_1B_2\rangle + \langle A_1B_3\rangle - \langle A_3B_2\rangle - \langle A_3B_3\rangle - 2 \le 0\),

and reports a theoretical maximum left-hand side of 0.82843 in Table 1.

  • Under a depolarizing-noise model, the paper derives that to preserve a violation the prepared friend state must have fidelity greater than 93.66%; equivalently, it states \(X \ge 8 - 6q\) and that the minimum fidelity supporting violation is 93.66%.
  • On real hardware, the paper reports that ibm_torino obtained a violation at branch factor 4, while Quantinuum H1-1 obtained a violation at branch factor 16; the other IBM devices shown (ibm_osaka, ibm_sherbrooke) did not show violations beyond branch factor 0.
  • The authors state that their results extend proof-of-concept experimental violations from branch factor 0.0 to branch factor 16.0.

## Methods

  • The paper studies Extended Wigner’s Friend Scenarios (EWFS) and focuses on the three-setting, two-party semi-Brukner inequality; it also analyzes a simplified single-friend variant where Bob has no friend register beyond a single qubit.
  • It implements EWFS as quantum circuits using GHZ friend states, and also discusses controlled-random-unitary and Dicke-state friends as alternative branch-factor constructions; experiments are run on noiseless simulators, noisy simulators/emulators, and real quantum hardware.
  • To infer friend outcomes under noise, the paper uses a majority-vote measurement strategy and validates branch-factor preparation via fidelity estimates, depolarizing-noise modeling, and circuit compilation/transpilation to IBM and Quantinuum hardware.

## Limitations

  • The paper explicitly states that the LF violations shown are not loophole-free; it notes, for example, that the experiments do not use speed-of-light separation to isolate the relevant parties.
  • It acknowledges that the results are limited by noise and hardware errors, which reduce the significance of the violations and restrict the maximum branch factor on current devices.
  • Evident from the text: the strongest hardware result is still a proof-of-concept demonstration, and the paper says that better quantum computers will be needed to reach larger, more meaningful branch factors.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.22331/q-2025-09-05-1851

Cites

61 of 61 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Figures

Figure 1 Conceptual rendering of the Wigner's Friend scenario. A system is sent into Charlie's sealed lab. Alice has different measurement settings labeled by x to observe the sealed lab that contains her friend Charlie and his measurement outcome c . Alice's measurement outcome has the value labeled a . Figure replicated and modified with permission from the authors from [3].
Figure 1 Conceptual rendering of the Wigner's Friend scenario. A system is sent into Charlie's sealed lab. Alice has different measurement settings labeled by x to observe the sealed lab that contains her friend Charlie and his measurement outcome c . Alice's measurement outcome has the value labeled a . Figure replicated and modified with permission from the authors from [3].CC-BY · From the PDF
Figure 2 Conceptual rendering of an Extended Wigner's Friend Scenario (EWFS). A system is split and sent into two sealed labs. Alice has different measurement settings labeled by x to observe the sealed lab that contains her friend Charlie and his measurement outcome c . Similarly, Bob has measurement settings labeled by y for the sealed lab containing Debbie and her measurement outcome d . Alice's measurement outcome has the value labeled a , and Bob's has the value labeled b . Figure replicated with permission from the authors from [3].
Figure 2 Conceptual rendering of an Extended Wigner's Friend Scenario (EWFS). A system is split and sent into two sealed labs. Alice has different measurement settings labeled by x to observe the sealed lab that contains her friend Charlie and his measurement outcome c . Similarly, Bob has measurement settings labeled by y for the sealed lab containing Debbie and her measurement outcome d . Alice's measurement outcome has the value labeled a , and Bob's has the value labeled b . Figure replicated with permission from the authors from [3].CC-BY · From the PDF
Figure 3 Single-friend Extended Wigner's Friend Scenario. Alice's friend Charlie, locked in an isolated lab, measures half of an entangled pair; Alice may either 'PEEK' at Charlie's recorded outcome or undo his measurement and re-measure in a rotated basis, while Bob directly measures the second half of the pair.
Figure 3 Single-friend Extended Wigner's Friend Scenario. Alice's friend Charlie, locked in an isolated lab, measures half of an entangled pair; Alice may either 'PEEK' at Charlie's recorded outcome or undo his measurement and re-measure in a rotated basis, while Bob directly measures the second half of the pair.CC-BY · From the PDF
Figure 4 A program designed to test Local Friendliness violations on a progressively larger scale increases the significance of the observers used as friends. Our work gives the first experimental violations of Local Friendliness with friends larger than a photonic qubit. In [30], the authors estimate the size of a QPU needed to perform experiments using a human-level artificial intelligence simulated on a QPU as the friend observer at approximately 10 19 logical qubits and logical depth of 10 14 operations. We mark this as a white dot as it is a theoretical target while our work gives experimental results (black dot).
Figure 4 A program designed to test Local Friendliness violations on a progressively larger scale increases the significance of the observers used as friends. Our work gives the first experimental violations of Local Friendliness with friends larger than a photonic qubit. In [30], the authors estimate the size of a QPU needed to perform experiments using a human-level artificial intelligence simulated on a QPU as the friend observer at approximately 10 19 logical qubits and logical depth of 10 14 operations. We mark this as a white dot as it is a theoretical target while our work gives experimental results (black dot).CC-BY · From the PDF
Image 5
CC-BY · From the PDF
Figure 5 Circuit depiction of the PEEK-PEEK setting for the extended Wigner's Friend scenario. Alice and Bob begin by preparing a bipartite state as defined in Equation (2) . Alice then performs her measurement setting on Charlie's qubit(s); likewise, Bob performs his measurement on Debbie's (single) qubit. The settings performed by Alice and Bob is either PEEK, REVERSE-1, or REVERSE-2. In this particular depiction of the circuit, we assume that Alice and Bob are performing the PEEK setting. Finally, the system qubits of Charlie and Debbie are measured.
Figure 5 Circuit depiction of the PEEK-PEEK setting for the extended Wigner's Friend scenario. Alice and Bob begin by preparing a bipartite state as defined in Equation (2) . Alice then performs her measurement setting on Charlie's qubit(s); likewise, Bob performs his measurement on Debbie's (single) qubit. The settings performed by Alice and Bob is either PEEK, REVERSE-1, or REVERSE-2. In this particular depiction of the circuit, we assume that Alice and Bob are performing the PEEK setting. Finally, the system qubits of Charlie and Debbie are measured.CC-BY · From the PDF
Figure 7 Example EWFS circuit with PEEK and REVERSE-1 settings. An example of the EWFS circuit with GHZ friends using the random strategy, where the friend Charlie consists of two qubits. Alice uses a PEEK setting in this example circuit, while Bob uses the REVERSE-1 setting.
Figure 7 Example EWFS circuit with PEEK and REVERSE-1 settings. An example of the EWFS circuit with GHZ friends using the random strategy, where the friend Charlie consists of two qubits. Alice uses a PEEK setting in this example circuit, while Bob uses the REVERSE-1 setting.CC-BY · From the PDF
Figure 8 EWFS circuit where Alice measures all qubits of Charlie. An example of the EWFS circuit with GHZ friends using the majority vote strategy, where the friend Charlie consists of three qubits. Alice uses a PEEK setting in this example circuit, while Bob uses the REVERSE-1 setting. Note that Alice is measuring all the qubits of her friend in this case.
Figure 8 EWFS circuit where Alice measures all qubits of Charlie. An example of the EWFS circuit with GHZ friends using the majority vote strategy, where the friend Charlie consists of three qubits. Alice uses a PEEK setting in this example circuit, while Bob uses the REVERSE-1 setting. Note that Alice is measuring all the qubits of her friend in this case.CC-BY · From the PDF
Figure 9 A comparison between the 'random' and 'majority vote' strategies for EWFS. Each plot considers how each strategy performs for progressively increasing depolarizing noise levels with fixed 1% readout error on all qubits. Each data point is run using 10000 shots and averaged over 10 trials. The top x-axis ranges over the number of qubits in the quantum system size of Charlie, while the bottom x-axis shows the corresponding branch factor. One can see there is minimal difference between the two approaches, except that the majority vote approach has less variance.
Figure 9 A comparison between the 'random' and 'majority vote' strategies for EWFS. Each plot considers how each strategy performs for progressively increasing depolarizing noise levels with fixed 1% readout error on all qubits. Each data point is run using 10000 shots and averaged over 10 trials. The top x-axis ranges over the number of qubits in the quantum system size of Charlie, while the bottom x-axis shows the corresponding branch factor. One can see there is minimal difference between the two approaches, except that the majority vote approach has less variance.CC-BY · From the PDF
Figure 10 A comparison between simulator, noisy emulator, and hardware for majority vote EWFS. The gray-scaled lines show how increasing the depolarizing noise reduces the maximum friend size for which a violation occurs. The lines in color show the ibm_osaka , ibm_sherbrooke , ibm_torino IBM hardware devices and the H1-1 Quantinuum hardware device. Note that the only IBM hardware device to obtain violations beyond branch factor 0 is ibm_torino , showing a violation at branch factor 4, while on the H1-1 device, we obtain a violation at branch factor 16. We transpile the circuit before running for the specific coupling map topology for the emulated and real hardware. Further details on the transpilation are provided in Appendix B. For each line in the plot, the top x-axis ranges over the number of qubits in the quantum system size of Charlie, while the bottom x-axis shows the corresponding branch factor. All IBM data points are run with 10000 shots over 10 trials. For the H1-1 device, we ran each experiment with 200 shots and 4 trials, except for the experiment of size 17, which we ran for 7 trials. Error bars are 3 standard deviations.
Figure 10 A comparison between simulator, noisy emulator, and hardware for majority vote EWFS. The gray-scaled lines show how increasing the depolarizing noise reduces the maximum friend size for which a violation occurs. The lines in color show the ibm_osaka , ibm_sherbrooke , ibm_torino IBM hardware devices and the H1-1 Quantinuum hardware device. Note that the only IBM hardware device to obtain violations beyond branch factor 0 is ibm_torino , showing a violation at branch factor 4, while on the H1-1 device, we obtain a violation at branch factor 16. We transpile the circuit before running for the specific coupling map topology for the emulated and real hardware. Further details on the transpilation are provided in Appendix B. For each line in the plot, the top x-axis ranges over the number of qubits in the quantum system size of Charlie, while the bottom x-axis shows the corresponding branch factor. All IBM data points are run with 10000 shots over 10 trials. For the H1-1 device, we ran each experiment with 200 shots and 4 trials, except for the experiment of size 17, which we ran for 7 trials. Error bars are 3 standard deviations.CC-BY · From the PDF
Figure 11 (Left) A plot of the estimated fidelities of the friend state when prepared on different backends using the depolarizing model to estimate fidelity. The IBM Torino device supports valid LF violations up to branch factor 7, while the H1 device could theoretically support much larger violations. (Right) The single and two-qubit gate counts for compiled EWFS circuits on different backends.
Figure 11 (Left) A plot of the estimated fidelities of the friend state when prepared on different backends using the depolarizing model to estimate fidelity. The IBM Torino device supports valid LF violations up to branch factor 7, while the H1 device could theoretically support much larger violations. (Right) The single and two-qubit gate counts for compiled EWFS circuits on different backends.CC-BY · From the PDF
Figure 12 Number of single and two-qubit gates and number of qubits required to prepare random unitary and GHZ states vs. branch factor. For the basis gate set, we use the single qubit X , Y , and Z rotations and CNOT gate, and we use Qiskit to transpile the given circuits. We see a tradeoff available for increasing branch factors depending on whether one is minimizing qubit number or gate count. Focusing on lower gate counts makes GHZ friends preferable while focusing on smaller numbers of qubits makes random unitary friends more preferable.
Figure 12 Number of single and two-qubit gates and number of qubits required to prepare random unitary and GHZ states vs. branch factor. For the basis gate set, we use the single qubit X , Y , and Z rotations and CNOT gate, and we use Qiskit to transpile the given circuits. We see a tradeoff available for increasing branch factors depending on whether one is minimizing qubit number or gate count. Focusing on lower gate counts makes GHZ friends preferable while focusing on smaller numbers of qubits makes random unitary friends more preferable.CC-BY · From the PDF
Figure 13 Inequality violation values for the semi-Brukner inequality over an increasing range of depolarizing noise. On the left, we consider a controlled-random unitary acting as the friends. On the right, we consider a controlled Dicke state as a friend. For the controlled-random unitary friends, we sample a Haar random unitary 100 times and take the average and error bars resulting from those runs. In the random unitary case, one sees that we don't obtain the optimal violation even in the ideal simulations. The reason is that inferring branches is probabilistic and fails with higher probability in the low friend sizes.
Figure 13 Inequality violation values for the semi-Brukner inequality over an increasing range of depolarizing noise. On the left, we consider a controlled-random unitary acting as the friends. On the right, we consider a controlled Dicke state as a friend. For the controlled-random unitary friends, we sample a Haar random unitary 100 times and take the average and error bars resulting from those runs. In the random unitary case, one sees that we don't obtain the optimal violation even in the ideal simulations. The reason is that inferring branches is probabilistic and fails with higher probability in the low friend sizes.CC-BY · From the PDF
Figure 14 The calibrations coupling map for the CNOT readout error of CNOT gates for the ibm_torino device. Lighter colors indicate a higher CNOT error, while darker shades indicate a lower CNOT error. When targeting this device, we aim to maintain a chain-like orientation for our initial layout and also pick a layout that optimizes for the lowest readout error. Specifically, we assign nodes labeled 43 to Debbie (highlighted in green), 45 to Alice (highlighted in red), and 44 to Bob (highlighted in blue). The remaining nodes [46, 47, 48, 49, 50, 51, 52, 37, 33, 32, 31] (highlighted in yellow)represent the size of Charlie's system for progressively higher qubit systems. The nodes with the outer layer highlight indicate the qubit layout used for our experiments. This image was produced with device readout assignment error data obtained via the IBM Quantum Platform on 8-3-2024.
Figure 14 The calibrations coupling map for the CNOT readout error of CNOT gates for the ibm_torino device. Lighter colors indicate a higher CNOT error, while darker shades indicate a lower CNOT error. When targeting this device, we aim to maintain a chain-like orientation for our initial layout and also pick a layout that optimizes for the lowest readout error. Specifically, we assign nodes labeled 43 to Debbie (highlighted in green), 45 to Alice (highlighted in red), and 44 to Bob (highlighted in blue). The remaining nodes [46, 47, 48, 49, 50, 51, 52, 37, 33, 32, 31] (highlighted in yellow)represent the size of Charlie's system for progressively higher qubit systems. The nodes with the outer layer highlight indicate the qubit layout used for our experiments. This image was produced with device readout assignment error data obtained via the IBM Quantum Platform on 8-3-2024.CC-BY · From the PDF

Verification

Claims the paper makes

10 Valid · 4 Not valid · 19 Not checked · 91 Own results and interpretations

54 more claims

Create a free account to see all 124 claims and to verify the 9 references still open.

See all claims — free account
  • It may state two separate claims as one. 23
  • Probably not a substantial claim. 47

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

2 of 2

Citations

Cites

Show all 53 references

Cited by

See all 5

Metadata

Authors
Zeng, W. J., Labib, F., & Russo, V.
Affiliations
Unitary Foundation
Venue
Quantum
Published
2025-09-05
Volume
9
Pages
1851
ISSN
2521-327X (ISSN-L)
OpenAlex
W4414019682
Cited by
0
References
53
FWCI
0.00
Type
journal article
Language
en
Open access
Locations
  • Quantum (publishedVersion, cc-by)
  • arXiv (Cornell University) (submittedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
  • arXiv (Cornell University) (cc-by)
Funders
Australian Research Council (DP250102162)
Topics
  • Physical Sciences › Physics and Astronomy › Atomic and Molecular Physics, and Optics › Quantum Mechanics and Applications
  • Physical Sciences › Physics and Astronomy › Atomic and Molecular Physics, and Optics › Quantum Information and Cryptography
  • Physical Sciences › Computer Science › Artificial Intelligence › Quantum Computing Algorithms and Architecture
Keywords
Wigner's friend, quantum computing hardware, quantum simulators, quantum circuits, quantum foundations, local agency, photonic qubits

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Towards violations of Local Friendliness with quantum computers. https://app.cochat.ai/papers/10.22331/q-2025-09-05-1851
The paper
Zeng, W. J., Labib, F., & Russo, V. (2025). Towards violations of Local Friendliness with quantum computers. Quantum. https://doi.org/10.22331/q-2025-09-05-1851

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.