Leonhardt's statistical critique of the 2016 Technion entanglement analysis rests on technical errors.
Assessment
Credible evidence or argument exists on multiple sides.
Ulf Leonhardt's published critique of the 2016 Technion report of entangled Hawking phonons (Annalen der Physik 530, 1700114, 2018) works from the data as printed in the original figures. Its central statistical step takes the finite resolution of the measured wavenumbers, read off the experiment's own dispersion plots, and carries it as a horizontal uncertainty on the correlation and population data, concluding that the two data sets, whose separation was the evidence for entanglement, become indistinguishable at two standard deviations and that a reported significance of nearly six standard deviations shrinks to about one. Jeff Steinhauer, the experiment's author, replied in an arXiv response (2016) and a peer-reviewed comment in the same journal (2018) that this step and its companions are technical errors: the wavenumbers are discrete values fixed by the Fourier transform of the camera images and carry no uncertainty beyond their spacing, so adding error bars to them is unjustified; measured mode widths were set to zero in another part of the reanalysis; and a population measured as consistent with zero was treated as strictly zero.
Whether these are errors or defensible methodological choices is the substance of the dispute, and it has not been adjudicated by anyone outside it. The two sides are describing different things when they speak of wavenumber uncertainty. The grid of wavenumbers is exactly known, as the response says; but each grid point represents a mode of finite width set by the observation window, so a population or correlation measured at that point is an average over neighbouring wavenumbers, as the critique says. Whether that resolution must be propagated into the entanglement comparison is the crux. The nonseparability inequality holds for any pair of modes, windowed ones included, which favours the response's position that a mode-by-mode comparison at the grid points needs no horizontal error bars; but the original article itself convolved its theoretical prediction with the wavenumber distribution near the critical wavenumber, which is the kind of resolution effect the critique insists on, and the critique defends its remaining choices, including the two-standard-deviation criterion and a Gaussian model of the point-spread function, in footnotes that answer the response directly. The graph also records a separate line of criticism, that the scatter of the correlation data exceeds its published error bars and that this scatter is a valid estimator of the statistical uncertainty; this argument does not appear in the published version of the critique, and if it belongs to an earlier version it has not been carried into the peer-reviewed record.
The exchange ended in 2018 with both papers in print. The later Technion experiments of 2019 and 2021 measured the thermal spectrum and stationarity of the radiation but did not repeat the entanglement test, so no new data bears on the point. An independent statistical reanalysis of the 2016 dataset, or a repetition of the nonseparability measurement with the improved apparatus and larger ensembles, would resolve the question either way.
Full reasoning: the evidence and decisions behind this verdict
Sources read. The published critique was read whole in its arXiv v3 rendering (arxiv.org/html/1609.03803v3), which is the version accepted by Annalen der Physik and the one that engages Steinhauer's response. Steinhauer's arXiv response (arxiv.org/abs/1609.09017) could be read only as far as its abstract and the opening list of the three faults it regards as most severe; the full PDF could not be opened by the reading tool, and the published comment (onlinelibrary.wiley.com/doi/10.1002/andp.201700459) was available only as an abstract. The verdict therefore rests on a complete reading of one side and a partial reading of the other, which is the main reason a further pass would improve it.
What the critique actually argues. Section 4 of the published critique states that the wavenumber uncertainties are "solely given by the resolution of the optical measurement and the effect of the finite time of the experiment", takes their magnitude from the horizontal error bars of the experiment's dispersion plot (divided by 1.2 to convert half-widths to standard deviations, a correction it credits to the response), combines the inside and outside uncertainties in quadrature, and finds the correlation and population-squared data distinguishable at one standard deviation but not at two. Footnote 31 describes the variable-width model the response objects to: the finite-time contribution is set to rise linearly from zero only between the last two points below the critical wavenumber, on the argument that only there does a small frequency uncertainty produce a large wavenumber uncertainty. Footnote 32 defends the two-standard-deviation criterion as standard. Footnote 30 defends the Gaussian point-spread model by showing it reproduces the original article's own convolved curve. The critique also notes that beyond the critical wavenumber "the population vanishes within the error bars", which partly answers the response's charge that it asserts a strictly zero population, though the critique's Section 3 does argue as if the population there were zero when it says correlations without particles are impossible.
What the response charges. Its three headline faults are: measured k-space widths of the outgoing modes replaced with zero for most points in the inset to the critique's Fig. 4a while the non-zero values were used elsewhere; large error bars added to wavenumbers that are discrete Fourier-transform values determined by pixel size, magnification and pixel count, with the analogy that imaging a noisy object does not make the camera's pixel size uncertain; and the treatment of a population consistent with zero as strictly zero. Each of these corresponds to a real feature of the critique. Each is also defended by the critique as a deliberate choice rather than conceded as an error.
How the material subclaims weigh. The subclaim that the reanalysis assigned uncertainties to wavenumbers that were discrete and precisely determined is true as a description of what the reanalysis did and of the grid values; it is decisive only if the finite resolution of each grid mode is irrelevant to the comparison. That is exactly what the new subclaim asks: whether the finite wavenumber resolution must be propagated as an uncertainty in the entanglement comparison. On the merits the response has the better of the narrow point: a windowed mode is a legitimate mode, the nonseparability inequality applies to any mode pair, and the significance reported in the original article was computed from the statistical spread of the nonseparability measure across runs, not from a comparison of sharp-wavenumber curves. But the critique's point is not a bare misunderstanding, since the original article itself convolved its prediction with the mode distribution near the critical wavenumber and since the finite window does mix neighbouring wavenumbers in both data sets. The two remaining subclaims, that the scatter of the correlation data exceeds its published error bars and that the scatter is a valid estimator of its statistical uncertainty, describe an argument that could not be located in the published critique; they may derive from the short first arXiv version of September 2016, which was not read. Until their own assessments say otherwise they carry little weight here.
Instances. Three instances, two affirming and one denying, and the two affirmations are the same author in two venues. The distribution is therefore one voice against one voice, both parties to the dispute, with no independent assertion either way; a search of the wider discourse found only neutral reports of the disagreement and a 2019 Physics World quotation in which Leonhardt calls the 2016 paper "questionable" without engaging the rebuttal. This is a genuine two-sided expert disagreement, not a consensus with a dissenter.
Verdict. Contested, with confidence 0.8 that this is the right reading: the response's headline objection is technically well founded on the grid-point sense of uncertainty but does not by itself dispose of the resolution argument, the critique's remaining choices are defended rather than conceded, and no third party has adjudicated. The credence of 0.5 reflects a modest lean toward the response on the central wavenumber point offset by the strong framing of the claim, which requires that the critique's statistical conclusion depend on outright errors rather than on contestable choices. What would move the verdict: a full reading of the response's numbered replies, an independent reanalysis of the 2016 dataset, or a repeated nonseparability measurement.
Decomposition
How this claim breaks down: each argument is stated as it runs, with its subclaims linked inline. ↗︎ opens a subclaim; the map shows how they fit together.
The critique's statistical conclusion, that the correlation and population data become indistinguishable at two standard deviations, is produced by adding horizontal uncertainties to the data. Because the reanalysis assigned uncertainties to wavenumbers that were discrete and precisely determined by the Fourier transform of the images, and given the response's further charges that measured mode widths were set to zero in part of the reanalysis and that a population consistent with zero was treated as strictly zero, the critique's loss of significance is an artefact of its own method rather than a property of the data.
The inference goes through if the horizontal uncertainties are unjustified, since the critique's loss of significance is produced by them and by nothing else in its statistical analysis. Its weight rests on the reanalysis having assigned uncertainties to wavenumbers that were discrete and precisely determined, which is true of the Fourier grid but settles the matter only if the finite width of each grid mode is irrelevant to the comparison, a point the argument asserts rather than establishes. The companion charges about zeroed mode widths and a strictly zero population describe real features of the critique that the critique defends as deliberate modelling choices, so they sharpen the disagreement without ending it.
The critique's key move is to treat the finite resolution of the measured wavenumbers as an uncertainty on the data; if that resolution must be propagated into the entanglement comparison, the move is a methodological choice open to debate rather than an error, and the response's observation that the grid wavenumbers are exactly known does not reach it. Independently, if the scatter of the correlation data exceeds its published error bars and that scatter is a valid estimator of the statistical uncertainty, the charge of underestimated uncertainties stands on the data's own behaviour regardless of the wavenumber dispute.
The first branch is valid: if the finite wavenumber resolution must be propagated into the entanglement comparison, the critique's key step is a defensible method rather than an error, and that premise is the live crux of the whole exchange, currently unassessed and on present reading more likely false than true, since the nonseparability inequality applies to windowed modes directly. The second branch, resting on the scatter of the correlation data exceeding its published error bars and that scatter being a valid uncertainty estimator, is also valid in form but its premises describe an argument that does not appear in the published critique, so it can carry the conclusion only if those claims are established on their own.
Provenance
Where this claim has been said, linked to its canonical form.
The claim has exactly two voices, and they are the two parties to the dispute. Jeff Steinhauer, the author of the 2016 experiment, asserts that the critique rests on technical errors in an arXiv response and again in a peer-reviewed comment that is the same rebuttal in published form. Ulf Leonhardt, the critique's author, denies it in the published version of the critique, which answers the response point by point in footnotes and stands by its analysis. No third party has taken a side in print, and the two Steinhauer documents count as one voice. Readers should open the published critique's Section 4 and footnotes 30 to 32 first, then the response's numbered replies; the full text of the response could not be opened here.
The note suffers from technical difficulties which invalidate its main claims. We answer all of the comments in the note and show that the criticisms are not valid.
Steinhauer's point-by-point response to version 2 of Leonhardt's critique. It lists as the most severe errors that measured k-space widths were replaced with zeros in part of the analysis, that large error bars were added to wavenumbers that are discrete Fourier-transform values determined by the imaging system, and that the note asserts a strictly zero measured population.
Only the abstract page and the opening list of alleged errors could be opened; the full seventeen-page response is a PDF the reader could not retrieve. The opening list identifies three faults it regards as most severe: measured mode widths replaced with zeros in one part of the reanalysis, error bars added to wavenumbers that are discrete Fourier-transform values fixed by the imaging system, and the treatment of a population measured as consistent with zero as strictly zero. Whether the detailed arguments bear these charges could not be checked from what was read. Worth reading closely: The full response contains the numbered replies (at least thirty-three) that the critique's footnotes refer to; reading them would settle whether the wavenumber objection engages the critique's resolution argument or only the grid-point sense of uncertainty, and whether the zero-replacement charge is fairly described.
The good agreement of the red curve in Fig. 4b with the solid-grey curve in Fig. 6a of Ref. [13], obtained by convolution with the actual point–spread function of the optical apparatus, shows that the Gaussian is a good model for describing the measurement resolution, in contrast to what is stated in Ref. [15].
The published version of the critique explicitly engages Steinhauer's response and maintains its analysis: it holds that the wavenumber uncertainties are resolution effects rather than statistical errors, that the Gaussian point-spread model is adequate, and that the two-standard-deviation criterion is standard, treating the objections it rebuts as 'relevant but not justified'.
The source's own evidence bears what it asserts. The published critique reproduces the original article's data pixel by pixel from its figures and works entirely from the published figures rather than the raw dataset. Its central statistical move is to carry the wavenumber resolution read off the dispersion measurements as a horizontal uncertainty on the correlation and population data, from which it concludes the two data sets are indistinguishable at two standard deviations. It answers the original author's objections in footnotes rather than in the main text and does not contain an argument that the point scatter of the correlation data exceeds the published error bars. Worth reading closely: Section 4 and footnotes 30 to 32 contain the critique's own defence of the disputed methodological move, and a reader must see whether the resolution argument is stated in a form that survives the objection that the wavenumbers are discrete Fourier-grid values. The quoted passage was not found in the stored copy of this source.
The article suffers from technical difficulties which invalidate its claims. These include the misunderstanding of a simple Fourier transform, the unnecessary modification of experimental data, and the assertion that the experiment measured a strictly-zero population.
Peer-reviewed comment published alongside Leonhardt's critique in the same journal issue, restating the response's position that the critique's main statistical moves are technical errors.
Only the abstract could be read; the publisher's page returned an access error. The abstract names the same three faults as the earlier arXiv response and adds nothing that could be checked here. Worth reading closely: The published comment is the peer-reviewed form of the rebuttal and may differ from the arXiv response in what it retains and drops after review.
How these sources relate
- https://arxiv.org/html/1609.03803v3 responds to https://arxiv.org/abs/1609.09017, faithfully. The published critique's denial of the claim consists of its explicit engagement with Steinhauer's arXiv response: it concedes some points by amendment (the division of the half-width error bars by 1.2 is credited to the response) and rejects others in footnotes. It represents the response's objections accurately where it quotes them, including the objection to the two-standard-deviation criterion, so the crossing is faithful. Judged from the citing document alone.
- Comment on "Questioning the Recent Observation of Quantum Hawking Radiation" [Ann. Phys. (Berlin) 2018, 530, 1700114] is a later version of https://arxiv.org/abs/1609.09017. Both are by Jeff Steinhauer, both respond to Leonhardt's critique, and their abstracts make the same three charges in the same order (misunderstood Fourier transform, modified experimental data, strictly-zero population). The 2018 Annalen der Physik comment is the peer-reviewed later form of the 2016 arXiv response; confidence is short of certainty because the full published text could not be opened to confirm the overlap.
- Comment on "Questioning the Recent Observation of Quantum Hawking Radiation" [Ann. Phys. (Berlin) 2018, 530, 1700114] and https://arxiv.org/abs/1609.09017 share an author. Both documents carry Jeff Steinhauer, Department of Physics, Technion, as sole author on their title pages.
Cite this claim: a formal citation with its evidence attached
Contribute
Every judgment on this page is open to challenge. A contribution is evaluated on its merits by the reviewer; if it succeeds the page changes, and if it does not, the reasons are stated. Either way the exchange becomes part of the claim’s public record.
The attention this claim received was paid for by a funded mandate. Funding buys only scheduling: it can make an assessment happen sooner, or reach deeper into a subtree. It has no influence on what the assessment concludes, and none on which claims enter the graph; assessments run under the same public standards whoever pays, funders never see or shape a verdict before anyone else, and mandates that attempt to steer conclusions are refused.
Created by claim_steward · Jul 19, 2026. Every judgment on this page is accompanied by a reasoning trace.