{"id":"e7206229-2b14-4433-9459-8ea548c27ca2","arxiv_id":"2505.11903","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A first-of-its-kind joint magnetic noise injection between LIGO Hanford and LIGO Livingston confirms that correlated magnetic noise couples into the strain channels and can be suppressed with Wiener filtering, though with residual side effects in a gravitational-wave background search.","lead":"Scientists injected a coordinated magnetic-noise pattern into the two LIGO gravitational-wave detectors separated by about 3,000 kilometers, and they observed correlated noise in both instruments. The test shows how such disturbances could contaminate future gravitational-wave background searches and demonstrates both the promise and the risks of cleaning the data afterward.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Wiener-filter claim rests on an in-sample, loud-injection demo; the paper's own subtracted-data GWB analysis shows residual artifacts, so 'effective and can be applied' is not yet established.","rationale":"The paper reports a genuinely novel dataset and an impressive experimental achievement: the first coherent broadband magnetic injection between two LIGO sites, with clear magnetometer and strain-channel coherence. The core observational claim is well supported. The reader's conditional verdict is appropriate, but the stated weakest assumption (phase-free coupling functions) is not the most load-bearing issue for the central claim, because the paper itself explicitly identifies and discusses the phase limitation in Sec. IV and the Conclusion. The more consequential issue is the Wiener-filter demonstration: the filter is trained and evaluated on the same dataset, and the paper's own GWB analysis of the subtracted data shows non-trivial residual artifacts. The abstract's final sentence claims that Wiener filtering is effective and can be applied to eventual GWB detection, but the evidence presented is a proof-of-concept on a loud, in-sample signal with acknowledged side effects. A simple out-of-sample test would settle whether the subtraction success is genuine or an artifact of in-sample fitting. My recommendation is therefore to keep the conditional verdict, with the condition explicitly tied to out-of-sample validation and to tempering the abstract's claims about applicability to realistic GWB searches.","tokens_in":14866,"tokens_out":4277,"duration_ms":47235,"concrete_test":"Split the 43-min injection into two equal halves; estimate the Wiener filter from the first half only and apply it to the second half. Then compare the subtracted ASD/CSD and the pygwb zero-lag and time-shifted results on the held-out half to the in-sample results. If the held-out residuals and the α=1.5 / α=−10.2 SNRs differ substantially from the in-sample values, the current demonstration is overfit and the abstract's claim should be conditioned on out-of-sample validation.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The most load-bearing concern is not the phase assumption (which the paper itself flags) but the in-sample validation of the Wiener filter and the residual artifacts in the GWB analysis. The central claim in the abstract—that Wiener filtering is effective and can be applied to GWB detection—requires that the demonstrated subtraction generalizes. The paper trains on the entire noise dataset (Sec. VI.A) and then evaluates on the same data, so the clean ASD/CSD recovery in Figs. 10–11 is not an out-of-sample test. Moreover, Table II and Fig. 12 show the subtracted data retain a 3.45σ residual at α=1.5 and a −7.32 SNR at α=−10.2; Sec. VII concedes the filter 'might alter the data in unknown and unforeseen ways.' Those residuals are not negligible for a GWB search, so the abstract's conclusion goes beyond what the experiment establishes. The phase-free coupling issue is a real but explicitly acknowledged limitation of the projection comparison; it does not threaten the injection claim, whereas the in-sample Wiener demonstration directly bears on the headline claim.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The manuscript reports the first coherent broadband magnetic-noise injection performed simultaneously at LIGO Hanford and LIGO Livingston, using synchronized injection coils to create correlated magnetic fields across ~3000 km. The authors characterize the injected magnetic field at multiple witness sensors, show that the injection produces significant strain-strain coherence between roughly 16 and 40 Hz, and compare the observed correlated strain with noise projections built from the independently measured O4a magnetic coupling functions. They then analyze the data with the LVK isotropic GWB search pipeline (pygwb), finding a loud power-law signal with a steep spectral index. Finally, they apply a time-domain Wiener filter using a single magnetometer as a witness, demonstrating that the injected noise can be strongly suppressed in the ASD and CSD, but they also report residual artifacts in the subsequent GWB analysis. The paper concludes that Wiener filtering is effective and can be applied in future GWB searches, while cautioning that the method may alter the data in unforeseen ways.","tokens_in":15063,"tokens_out":3445,"duration_ms":36164,"significance":"The experimental accomplishment is substantial: creating a controlled, coherent, broadband magnetic field across two widely separated gravitational-wave detectors and observing its coupling into the strain channel provides a unique testing ground for correlated-noise projection and subtraction methods that are central to future GWB searches. The noise projections use coupling functions measured in an independent O4a injection campaign, which avoids the circularity that often plagues witness-channel studies. The paper is also refreshingly honest about the limitations it uncovers, including the >1-sigma overestimation of the correlated magnetic budget and the non-negligible phase of the magnetic-to-strain coupling. The dataset and the analysis results will be valuable for future studies of Schumann-resonance noise and for developing spectral-separation techniques. However, the abstract's strong claim that Wiener filtering is 'effective and can be applied' to GWB detection goes beyond what the current in-sample, high-amplitude demonstration establishes, given the residual artifacts reported in the subtracted-data GWB analysis.","major_comments":[{"comment":"The Wiener-filter demonstration is an in-sample test: the filter coefficients κ_I,Wiener are computed from the full 43-minute noise-injection dataset using Eq. (12), and the suppression is then evaluated on the same dataset. The clean recovery of the reference ASD and CSD in Figs. 10 and 11 therefore does not demonstrate that the filter generalizes to independent data, where the SNR of the magnetic signal in the witness channel will be much lower. An out-of-sample evaluation—for example, training on one half of the injection segment and testing on the other, or training on the injection segment and testing on a separate time with similar coupling—is needed to support the claim that the method is applicable to realistic GWB searches.","section":"Sec. VI.A, Eq. (12)"},{"comment":"The noise-subtracted data still contain substantial artifacts in the GWB analysis: the zero-lag result shows a 3.45-sigma excess at α=1.5 and a signal-to-noise ratio of -7.32 at α=-10.2, and the text in Sec. VII concedes that the filter 'might alter the data in unknown and unforeseen ways.' These residuals are not negligible for a search aiming at a 5-sigma detection, and they show that the subtraction is not simply removing the correlated noise without side effects. The abstract's final sentence, which states that Wiener filtering is effective and can be applied in the eventual detection of the GWB, is therefore not supported by the evidence presented; at most, the paper demonstrates a proof-of-concept that requires further validation with lower-amplitude injections and out-of-sample tests.","section":"Sec. VI.B, Table II, and Sec. VII"},{"comment":"The comparison between the projected and observed correlated strain relies on coupling functions that are measured in amplitude only, while Eq. (6) uses the real part of the magnetic CSD. The paper itself notes that the phase of the magnetic-to-strain coupling appears to be site-dependent and non-zero, particularly above 30 Hz, and that this leads to a systematic overestimation of the projection. Because the phase is not characterized, the quantitative agreement between the projection and the observed strain-strain coherence cannot be fully assessed. This is an acknowledged limitation rather than a hidden flaw, but it should be stated more prominently in the abstract or introduction: the injection demonstrates the existence of coherent magnetic coupling, but the accuracy of the projection method remains uncertain at the level set by the >1-sigma discrepancy shown in Fig. 7.","section":"Sec. IV, Eq. (6), Fig. 7"}],"minor_comments":[{"comment":"There is a typo in the sentence introducing Eq. (7): 'detectror I' should be 'detector I'. Additionally, 'concretly' in the following paragraph should be 'concretely'.","section":"Sec. IV, Eq. (7)"},{"comment":"The corner plot in Fig. 9 shows a posterior for α that is quite broad; the text should state more explicitly whether this is consistent with the injected spectrum, since the injected signal has a known, non-power-law shape. This would help the reader connect the injected spectrum to the recovered spectral index.","section":"Sec. V, Fig. 9"},{"comment":"When describing the high-pass filter, the text notes that an 8th-order Butterworth filter with a cutoff of 16 Hz is applied, but it does not specify whether the same filter is applied to the witness channel before computing Eq. (12). If the witness is filtered differently, the transfer function should be described, as it affects the meaning of the Wiener filter coefficients.","section":"Sec. VI.A"},{"comment":"The text states that 'reweighting the frequentist results to α=1.5' gives a point estimate and standard deviation, but Table II does not list the α=1.5 row. Please add this row or clarify in the text, since the reader needs to see the numbers directly.","section":"Sec. VI.B"},{"comment":"The coil-field distance estimate using Eq. (2) neglects the coil radius and the square geometry; the text acknowledges this, but it would be helpful to state the expected uncertainty on the ratio (e.g., factor of 2) so the reader can judge how well the observed 8:1 ratio matches the model.","section":"Sec. II.B"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is a strong experimental contribution that will be of interest to the LIGO-Virgo-KAGRA community and to the broader gravitational-wave data-analysis community. The main weakness is the gap between the abstract's strong claim about Wiener filtering and the in-sample, high-amplitude demonstration, compounded by the residual artifacts in the subtracted-data GWB analysis. I would recommend major revision: the authors should either significantly temper the conclusion or add an out-of-sample test that demonstrates generalizability. The phase-uncertainty issue is honestly reported, but it too should be elevated in the presentation. I do not see grounds for rejection; the experimental result and the open questions are valuable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nThe thing to know: this is the first coherent broadband magnetic noise injection between two GW detectors separated by ~3000 km, and the strain-strain coherence they achieve between ~16 and 40 Hz is the real deal. That alone makes the paper worth a serious look. The authors also show that projections based on O4a coupling functions, measured independently, reproduce the observed magnetic contribution at the right order of magnitude, though with a consistent overestimate. The claim that fails to hold up, as stated in the abstract, is that Wiener filtering 'is effective and can be applied' to the GWB search. The body is more careful, and the body is right: the filter is trained on the same data it is evaluated on, the injection is far louder than realistic correlated noise, and the GWB analysis of the subtracted data leaves behind a 3.45 sigma excess at alpha=1.5 and a -7.32 SNR at alpha=-10.2. Those are not negligible. The paper itself acknowledges the filter 'might alter the data in unknown and unforeseen ways.' That is not a fatal flaw in a proof-of-concept, but it does mean the abstract should be tempered.\n\nWhat is genuinely good: the injection design, the use of multiple witness sensors, the independent coupling measurements, and the decision to run the actual pygwb pipeline on the data. The phase-free coupling assumption is a real limitation, and the authors flag it; the quadrature-sum projection overestimates, but they argue that this is expected from the method. The citation pattern is fine, and the logbook references are appropriately concrete.\n\nThe main soft spot, as the stress-test note says, is that the Wiener filter demonstration is in-sample. The paper's own residual artifacts confirm the concern. That said, the authors are unusually candid about these issues in Secs. VI-VII. The weakness is not that they hide problems; it is that the abstract promises more than the analysis supports.\n\nFor a reader: the dataset and the projection comparison are the lasting value. The Wiener filter part is a demonstration, not a validated method, and should be read as such.\n\nI would bring this to a reading group, and if I were an editor I would definitely send it to peer review. The referee's main job should be to push the authors to either provide out-of-sample validation of the filter or to scale down the abstract's claim. My recommendation: engage, but expect revision.","headline":"A genuinely new inter-site coherent magnetic noise injection with a solid projection comparison; the Wiener-filter conclusion overreaches and needs revision.","tokens_in":15636,"tokens_out":2434,"would_cite":true,"duration_ms":24207,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"The first coherent broadband magnetic noise injection between two gravitational-wave detectors created measurable strain-strain coherence between about 16 and 40 Hz, and a single-magnetometer Wiener filter removed it.","keywords":["gravitational-wave background","correlated magnetic noise","Schumann resonances","LIGO","Wiener filter","noise projection","cross-correlation statistic","noise injection"],"falsifier":"Measure the complex transfer function from injected coil current to strain at each site across 16-42 Hz. If the phase difference between Hanford and Livingston is nonzero and grows just where the real-part strain CSD drops, the paper's explanation holds; if the phase difference is zero, the over-projection would have to come from something else, such as the quadrature-sum method or unmodelled amplitude changes.","tokens_in":14678,"feed_emoji":"🧲","tokens_out":6589,"duration_ms":60404,"temperature":0.7,"pith_summary":"This paper reports the first controlled experiment in which broadband coherent magnetic noise was injected simultaneously into two gravitational-wave detectors roughly 3,000 km apart. The injection created correlated magnetic fields between LIGO Hanford and LIGO Livingston that were strong enough to leave a measurable imprint on the detectors' strain channels between about 16 and 40 Hz. That imprint matters because exactly this kind of correlated, non-astrophysical noise can masquerade as a gravitational-wave background signal in future searches. The paper uses the data to test the standard magnetic-noise projection procedure, run the data through an isotropic background search pipeline, and demonstrate that a single magnetometer witness channel feeding a Wiener filter removes the injected noise. A sympathetic reading is that these results validate the projection and subtraction tools the field plans to rely on, while exposing where those tools need refinement.","feed_headline":"Magnetic noise injected 3,000 km apart shows up in LIGO strain data","feed_subtitle":"A single magnetometer witness plus a Wiener filter removes the correlated magnetic noise from the strain channel.","key_machinery":"The central object is the coherent injection itself: the same prescribed broad-band magnetic spectrum, modelled on Schumann resonances observed at Sos Enattos and tapered to 16-42 Hz, played out synchronously through large injection coils at each central station. The argument is carried by the comparison between the observed magnetometer-to-magnetometer coherence and the strain-to-strain coherence, and by two transfer tools: the O4a magnetic coupling functions $\\kappa_I(f)$ used to project magnetic ASD into strain ASD, and the Wiener filter $\\kappa_{I,\\mathrm{Wiener}} = \\langle s_I w_I\\rangle/\\langle w_I w_I\\rangle$ estimated from one witness channel. The magnetic cross-correlation statistic of Eq. (6) connects the two, and the crucial limitation is that only the magnitude of $\\kappa_I$ is known, so the phase of the coupling is left out of the projection.","core_discovery":"The central claim is that the December 20, 2023 injection succeeded: it produced coherent magnetic fields at both sites, and the resulting strain-strain coherence between 16 and 40 Hz is attributable to the injected magnetic field. Projecting the observed tri-axial magnetic noise through the O4a coupling functions reproduces the magnitude of the strain effect within about a factor of two, though it overestimates the correlated strain above roughly 30 Hz because the coupling functions carry no phase information and the true coupling phase differs between sites. Running the data through the isotropic background search pipeline recovers a loud, steep power-law signal with spectral index around -10.2, as expected if the correlated magnetic noise were a foreground. Finally, a time-domain Wiener filter using one vertex magnetometer as a witness removes essentially all of the correlated strain noise, restoring the ASD and CSD to reference levels, with the caveat that the subtraction leaves residual effects visible in the background search at the 3-4 sigma level.","pith_inferences":["Editorial inference: The same experiment design, repeated with injection amplitude below the strain ASD floor but high enough to accumulate in cross-correlation, would directly test how Wiener filtering behaves in the realistic sub-threshold regime the authors flag as untested.","Editorial inference: The phase decorrelation above about 30 Hz implies that a network of three or more sites could, in principle, distinguish a common Schumann-type magnetic foreground from a genuine gravitational-wave background by comparing the phase structure of the cross-spectra across baselines.","Editorial inference: The authors' finding that subtraction leaves residual spectral-index structure suggests that any Wiener-filter-based cleaning pipeline should be validated end-to-end with the same background-search statistic it is meant to protect, not just by ASD and CSD agreement."],"forward_implications":["Correlated magnetic noise of the kind produced here would appear in an isotropic gravitational-wave background search as a steep, loud power-law signal; in this dataset the recovered spectral index is about -10.2, far steeper than any expected astrophysical background, so spectral separation should be possible.","The quadrature sum over all nine magnetometer orientation pairs overestimates the true correlated magnetic contribution by about a factor of two.","A single magnetometer witness at the same site, processed through a time-domain Wiener filter, can remove essentially all of the correlated broadband magnetic noise from the strain channel and restore the strain CSD to reference levels.","Noise subtraction is not risk-free: the background search on the subtracted data shows residual structure, such as a 3.45 sigma feature with spectral index about 1.5, so Wiener filtering should be applied with caution in real searches.","Because the coupling phase differs between sites and is not captured by current magnitude-only coupling functions, future noise projections should measure and include the phase of the magnetic-to-strain transfer function."],"supporting_citations":[{"why":"Supplies the Sos Enattos Schumann resonance spectrum used to construct the injected magnetic noise shape.","marker":"[12]"},{"why":"Supplies the O4a magnetic coupling functions for the three central-station locations used in the ASD projections.","marker":"[20]"},{"why":"Provides the measured magnetic coupling functions for LIGO Hanford and Livingston that the noise projections rely on.","marker":"[35]"},{"why":"Establishes the quadrature-sum noise projection method and the earlier simulated impact of Schumann resonances on isotropic background searches.","marker":"[13]"},{"why":"The isotropic background search pipeline used to estimate the GWB spectrum and run parameter estimation on the injection data.","marker":"[29]"},{"why":"Provides the expression for the bias of the cross-correlation estimator caused by correlated noise.","marker":"[11]"},{"why":"Introduces the magnetic cross-correlation statistic and the coupling of Schumann resonances to gravitational-wave detectors.","marker":"[5]"},{"why":"Previous Wiener-filter noise subtraction work based on simulated data or proxy channels, which this paper's real-data demonstration extends.","marker":"[6]"}],"fun_headline_variants":["Wiener filter erases injected magnetic noise in LIGO data","Wiener filter silences injected magnetic noise in LIGO","First coherent magnetic noise injection tests LIGO background search","One magnetometer suffices to filter LIGO's injected magnetic noise"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The projection rests on treating the O4a magnetic-to-strain coupling functions as still valid and as carrying no phase, when the data itself indicates the coupling phase is site-dependent and nonzero above about 30 Hz.","fun_headline_variants_meta":{"raw":{"variants":["Wiener filter erases injected magnetic noise in LIGO data","Wiener filter silences injected magnetic noise in LIGO","First coherent magnetic noise injection tests LIGO background search","One magnetometer suffices to filter LIGO's injected magnetic noise"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.001724,"raw_usage":{"total_tokens":6815,"prompt_tokens":940,"completion_tokens":5875,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":556,"completion_tokens_details":{"reasoning_tokens":5802}},"tokens_in":556,"tokens_out":5875,"duration_ms":35980,"temperature":1.0,"reasoning_tokens":5802,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T20:45:24.312409+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Measure the complex transfer function from injected coil current to strain at each site across 16-42 Hz. If the phase difference between Hanford and Livingston is nonzero and grows just where the real-part strain CSD drops, the paper's explanation holds; if the phase difference is zero, the over-projection would have to come from something else, such as the quadrature-sum method or unmodelled amplitude changes.","supporting_citations":[{"cited_title":"Detecting a stochastic gravitational-wave background in the presence of correlated magnetic noise","cited_arxiv_id":"2008.00789","evidence_quote":"Supplies the Sos Enattos Schumann resonance spectrum used to construct the injected magnetic noise shape."},{"cited_title":"Nguyenet al., Classical and Quantum Gravity38, 145001 (2021)","cited_arxiv_id":null,"evidence_quote":"Supplies the O4a magnetic coupling functions for the three central-station locations used in the ASD projections."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the measured magnetic coupling functions for LIGO Hanford and Livingston that the noise projections rely on."},{"cited_title":"Janssens, K","cited_arxiv_id":null,"evidence_quote":"Establishes the quadrature-sum noise projection method and the earlier simulated impact of Schumann resonances on isotropic background searches."},{"cited_title":"Abbott, T","cited_arxiv_id":null,"evidence_quote":"The isotropic background search pipeline used to estimate the GWB spectrum and run parameter estimation on the injection data."},{"cited_title":"Himemoto and A","cited_arxiv_id":null,"evidence_quote":"Provides the expression for the bias of the cross-correlation estimator caused by correlated noise."}],"review_version":1}