{"id":"b8f3601d-eb29-4f8c-9872-39c30f6178ab","arxiv_id":"1908.07582","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"A unified likelihood reconstruction combining IceTop and in-ice data reconstructs cosmic rays with better pointing and expands the usable event sample to showers that miss the surface array.","lead":"This paper describes a new computer reconstruction that combines surface and deep-ice measurements at the IceCube detector to better locate cosmic ray air showers and their directions. If it performs on real data as it does on simulations, it could add many high-energy cosmic ray events to physics analyses and sharpen measurements of cosmic ray properties.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"IT-Uncontained pointing comparison is uncontrolled: 3-D QC requires IceTop stations, so its advantage over InIce-alone may be selection, not likelihood.","rationale":"The reader's weakest assumption focuses on Monte Carlo fidelity, which is important because all performance claims are evaluated against simulation. My concern is complementary and more localized: even if the Monte Carlo is perfect, the paper's key comparison for IT-Uncontained events does not control for event selection. The 3-D reconstruction and InIce-alone reconstruction use different quality cuts, and the 3-D cuts explicitly require IceTop stations. Since the claimed advantage is largest for IT-Uncontained events, where this selection difference is most pronounced, the evidence as presented does not isolate the effect of the unified likelihood from the effect of the cuts. This is a load-bearing issue because the strongest claim is that the 3-D reconstruction provides better pointing resolution than InIce-alone in exactly this regime. It is also testable: recomputing the right panel of Figure 3 with a common event sample would settle whether the advantage survives. I do not see this as grounds for rejection; the method may well be sound, and the paper itself is transparent about many limitations. But the controlled comparison is necessary before the headline performance claim is accepted, which is consistent with the reader's CONDITIONAL verdict. Hence I recommend no change to the verdict.","tokens_in":6498,"tokens_out":7028,"duration_ms":71264,"concrete_test":"Reproduce Figure 3 (right) using one common event set for IT-Uncontained proton events: select events passing both the InIce-alone QC and the 3-D QC, or better, take the QC of the 3-D reconstruction and apply all three reconstructions to exactly those events. Compute the 68% space-angle resolution in the same energy bins, and also report the number of events per bin. In addition, restate the 'successful reconstruction rates' with an explicit success criterion and binomial uncertainties. If the 3-D advantage over InIce-alone disappears or drops below about 1 sigma in the high-energy bins, the claimed pointing advantage is a selection effect rather than evidence for the unified likelihood.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The largest risk to the central claim is not the likelihood machinery itself but the uncontrolled event-selection comparison used to demonstrate it for IT-Uncontained events. In Section 3, Figure 3 (right) shows the 68% space-angle resolution for IT-Uncontained proton events as solid lines 'after their own QC separately'; the identical-QC dotted-line comparison shown for IT-Contained events (left) is absent for the IT-Uncontained case. The 3-D QC explicitly requires five triggered IceTop stations plus standard InIce QC, whereas the InIce-alone reconstruction is not required to have IceTop information. Low-energy IT-Uncontained events, which the text says trigger fewer IceTop stations, are exactly the region where 3-D is worse than InIce-alone (below about 10^6.6 GeV). At higher energies, the apparent 3-D advantage may therefore reflect that its QC preferentially selects events with more IceTop information, rather than the unified likelihood itself. The claimed success rates (62.6%, 58.1%, 43.5%, 25.3% versus 63.5%, 38.1%, 22.5%, 13.7% for InIce-alone) are also quoted without a definition of 'successful' or statistical uncertainties, so the recovery improvement is not yet quantitatively established. A controlled comparison on an identical event sample is needed before attributing the improvement to the 3-D method.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper presents a new three-dimensional (3-D) maximum-likelihood reconstruction for cosmic-ray events in IceCube that jointly uses IceTop surface signals and in-ice muon signals. The algorithm combines a charge and time likelihood for IceTop tanks with an in-ice photon-arrival likelihood, and is tested on CORSIKA/Sibyll2.1 Monte Carlo with proton and iron primaries for the 2012 detector configuration. Two event classes are considered: IT-Contained (core inside IceTop, muon contained in the in-ice detector) and IT-Uncontained (core outside/at edge, muon contained). The paper reports that the 3-D reconstruction achieves better 68% pointing resolution than IceTop-alone or InIce-alone reconstructions for both event classes (except below ~10^6.6 GeV for IT-Uncontained), recovers more IT-Uncontained events, and provides additional EAS parameters, at the cost of a slightly worse core position resolution (2–3 m) and a wider S125–energy spread.","tokens_in":6908,"tokens_out":7253,"duration_ms":515392,"significance":"If the reported performance holds, the 3-D reconstruction would increase the number of usable high-energy cosmic-ray events in IceCube by recovering events with cores outside IceTop, and would improve angular resolution through the IceTop–in-ice lever arm. The paper clearly describes the software architecture and makes concrete, falsifiable MC-based predictions. However, the central performance comparisons for IT-Uncontained events are not controlled for event selection, and the success-rate claims lack a definition and uncertainties. With the missing identical-QC analysis added, the paper would be a useful methods contribution to the field.","major_comments":[{"comment":"The IT-Uncontained pointing resolution comparison is uncontrolled: the solid lines are results after each reconstruction's own QC, and the identical-QC dotted-line comparison shown for IT-Contained events (left panel) is absent for the IT-Uncontained case. Because the 3-D QC requires at least five triggered IceTop stations while the InIce-alone reconstruction has no such requirement, the apparent 3-D advantage above about 10^6.6 GeV may be a selection effect rather than a property of the unified likelihood. Please provide the identical-QC comparison for IT-Uncontained events, or explicitly limit the claim to the separate-QC scenario.","section":"Section 3, Figure 3 (right)"},{"comment":"The quoted success rates for IT-Uncontained events (e.g., 62.6% vs 63.5% in the first bin) are not defined, are quoted without statistical uncertainties, and are evaluated after different QC for each reconstruction. This prevents a quantitative assessment of the claimed event-recovery improvement. Please define what constitutes a 'successful' reconstruction, provide uncertainties (the lower-energy bins appear to have limited MC statistics), and present a comparison on a common event set.","section":"Section 3, 'successful reconstruction rates' paragraph"},{"comment":"The statement that 'only the 3-D reconstruction ... is able to correctly find the core' is based on events passing each method's own QC; for example, IceTop-alone retains only 2.4% of IT-Uncontained events in the lowest energy bin. The figure thus conflates reconstruction accuracy with selection effects. A comparison on an identical event sample is required to support the claim that the 3-D method removes the systematic core bias.","section":"Section 3, Figure 2"},{"comment":"The combined likelihood function is not explicitly defined: the paper does not state whether the IceTop charge, IceTop time, and in-ice likelihoods are summed in log-space with unit weights, or how the relative normalizations are determined. Since the combination is the core of the new method, a precise definition is needed for reproducibility. Also, the derivation of C1 and C2 in Eq. (2.3) from 'a MC study' is described only briefly; please specify the procedure and whether these constants are fixed before evaluating the performance metrics.","section":"Section 2, 'Likelihood Combiner'"}],"minor_comments":[{"comment":"The notation '2· (118.1))2' is ambiguous; please clarify the intended expression, likely 2(118.1)^2.","section":"Equation (2.2)"},{"comment":"Please state explicitly that the right panel does not include the identical-QC comparison, or add it.","section":"Figure 3 caption"},{"comment":"The sentence 'The values of C1 and C2 derived from a MC study' should state the MC sample and fitting procedure.","section":"Section 2, after Eq. (2.3)"},{"comment":"The phrase 'Events are weighted to a E^-2.7 spectrum' should read 'an E^-2.7 spectrum'.","section":"Section 3"},{"comment":"The phrase 'extraordinary improvement' is subjective; replace it with a quantitative statement of the systematic-core-bias reduction.","section":"Section 3, Figure 2 text"},{"comment":"The abstract and Section 4 state the advantages in present tense without noting that all performance numbers are from Monte Carlo; please add a qualifier such as 'in simulation'.","section":"Abstract and Section 4"},{"comment":"Reference [11] (arXiv:1906.04317) should be cited with its title and venue if available.","section":"Reference [11]"},{"comment":"The colorbar label 'relative intensity after the spectrum is weighted to E^-2.7' is unclear; please define the normalization.","section":"Figure 5"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is an ICRC2019 proceedings paper being considered for journal publication. The central technical weakness is the uncontrolled comparison for IT-Uncontained events; the missing identical-QC analysis is straightforward to produce from the collaboration's existing MC samples. If the authors provide it, the result will likely support the stated conclusions, so I recommend major revision rather than rejection."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague,\n\nWhat you should know: this ICRC proceedings paper presents a genuinely new integration—a single likelihood that fits IceTop charge and time together with in-ice photon arrival probabilities, with a free curvature parameter and a per-event time-fluctuation term. It is a sensible step: instead of running IceTop-alone and InIce-alone separately and then combining, you get one joint fit that uses the vertical lever arm between the surface and deep ice. The paper shows on Monte Carlo that this joint fit gives better pointing than either alone for IT-Contained events, and that it can reconstruct events whose cores land outside the IceTop array, which the standard IceTop-alone analysis loses. That would be a real gain in high-energy statistics for IceCube cosmic-ray physics, and the claim is plausible.\n\nCredit where due: the method section is clear, the MC setup (CORSIKA/Sibyll2.1, 2012 detector configuration, proton and iron, out to large core distances) is standard and adequate for a first look, and the paper is honest about the main weakness it found: core resolution is 2–3 m worse than IceTop-alone, and they say they do not yet understand the mechanism. They also flag the wider S125-energy spread and call for more study. That is the right tone for a methods note.\n\nSoft spots, in order of importance. The IT-Uncontained pointing comparison in Figure 3 (right) is not controlled. The 3-D reconstruction's quality cuts require five triggered IceTop stations; the InIce-alone reconstruction does not. The dotted \"identical QC\" comparison that appears for IT-Contained events is missing from the right-hand plot. So part of the apparent advantage of 3-D at high energies could simply be selection: events that pass the 3-D QC are preferentially those with more IceTop information. The paper itself notes that at low energies, where IceTop stations are scarce, 3-D is worse—that is consistent with the selection explanation. The success rates quoted (e.g. 62.6% vs 63.5%, then 58.1% vs 38.1%) are given without a definition of \"successful\" or any uncertainties. These numbers are the quantitative core of the \"more events\" claim, and right now they are not fully pinned down. Finally, everything rests on MC; there is no data validation. For a conference proceedings that is acceptable as a status report, but the performance claims should not be taken as established until the controlled comparison and data check are done.\n\nWho this is for: IceCube folks and anyone comparing cosmic-ray reconstruction strategies in large hybrid detectors. It is worth a serious referee if it is expanded into a full paper, provided the IT-Uncontained comparison is rerun on identical events and the success-rate metric is defined. I would encourage engaging with it now, but treat the headline gain as conditional.\n\nRecommendation: peer review yes with requests for the controlled comparison.","headline":"Useful methods paper from IceCube on a joint 3-D reconstruction, but the headline IT-Uncontained gain is asserted on an uncontrolled QC comparison and needs a matched-sample check.","tokens_in":7289,"tokens_out":2296,"would_cite":true,"duration_ms":114541,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":[],"pacs":[],"model":"deepseek-v4-flash","headline":"IceCube's new three-dimensional reconstruction, which fits surface and in-ice signals with one unified likelihood, recovers cosmic-ray showers whose cores land outside IceTop and points them as well as or better than either array alone.","keywords":["cosmic ray air showers","IceCube","IceTop","in-ice array","event reconstruction","likelihood maximization","hybrid detector analysis","muon bundles"],"falsifier":"Apply the 3-D reconstruction to real IceCube data for IT-Uncontained events and compare reconstructed core positions against an independent in-ice-only muon track: if reconstructed cores still cluster toward the IceTop center, or if the expected gain in recovered events over in-ice-alone reconstruction does not appear, the central claim is refuted. A sharper test would check the improved pointing resolution using the Moon shadow or a known source as an independent arrival-direction reference.","tokens_in":6339,"feed_emoji":"🧊","tokens_out":6904,"duration_ms":61774,"temperature":0.7,"pith_summary":"The paper introduces a three-dimensional reconstruction for cosmic-ray air showers detected by IceCube's surface array IceTop and the deep in-ice array, jointly fitting signals from both detectors by minimizing one unified likelihood function. The authors claim that this combined fit reconstructs the shower core and arrival direction of events whose cores land inside, at the edge of, or outside IceTop, recovering a substantial fraction of IT-Uncontained events that standard single-array reconstructions lose. A sympathetic reader would care because cosmic-ray rates fall steeply with energy, so every recovered event matters for spectrum, anisotropy, and composition studies at high energy. On Monte Carlo events with the 2012 detector configuration, the 3-D reconstruction achieves pointing resolution better than or comparable to IceTop-alone or in-ice-alone reconstructions, extends usable zenith angles beyond those of contained events, and adds new shower parameters that may sharpen energy and mass estimators.","feed_headline":"Hybrid 3-D fit recovers cosmic-ray showers landing outside IceTop","feed_subtitle":"Combining surface tanks and deep muon tracks nearly doubles the usable highest-energy event sample.","key_machinery":"The load-bearing object is the unified likelihood function $L = L^{\\rm IceTop}_Q \\otimes L^{\\rm IceTop}_t \\otimes L^{\\rm InIce}_{\\rm SPE}$, whose three terms describe the signal charge and arrival time in IceTop tanks and the photon arrival times in the in-ice DOMs. The shower signal is modeled by the lateral distribution function $S(r) = S_{\\rm ref}(r/r_{\\rm ref})^{-\\beta - 0.30264\\log_{10}(r/r_{\\rm ref})}$ and a curvature term $\\delta t(r) = c_t r^2 + 19.41(1-e^{-r^2/(2(118.1)^2)})$, with $c_t$ left free per event; the per-station time fluctuation is redefined as $\\sigma_{t_i}=C_1\\sqrt{\\sum_{j=1}^2(t_{ij}-\\frac{t_{i1}+t_{i2}}2)^2}/(\\sum_j Q_{ij})+C_2$ with MC-derived constants. A likelihood combiner feeds these terms to Minuit or SIMPLEX, seeded by parameter and boundary services, and the reconstructed track is then passed to a Millipede-based energy estimator for muon energy losses. The mechanism works because the in-ice term provides a long lever arm that anchors the direction, while the IceTop terms constrain the core and lateral profile, so events outside the surface array are no longer stranded.","core_discovery":"The paper's central claim is that a single likelihood function combining IceTop charge and time information with in-ice photon-arrival information reconstructs the full three-dimensional air-shower footprint more reliably than either detector alone. On Monte Carlo showers from CORSIKA with Sibyll2.1, the 3-D reconstruction determines the shower core, arrival direction, lateral distribution parameter $S_{125}$, slope $\\beta$, and curvature parameter $c_t$ by maximizing this combined likelihood. For cores contained in IceTop it gives the best 68% pointing resolution of the three methods across the studied energy range, while its core resolution is about 2-3 m worse than IceTop-alone, which widens the $S_{125}$-energy relation. For cores on or outside IceTop, where IceTop-alone systematically pulls reconstructed cores inward and in-ice-alone is biased outward, the combined fit finds the true core and raises the successful reconstruction rate after quality cuts from 13.7% (in-ice alone) to 25.3% in the highest energy bin, with similar large gains at lower energies. This event recovery is the paper's central discovery: the two arrays together see the three-dimensional footprint that neither alone reconstructs reliably.","pith_inferences":["If the Monte Carlo-demonstrated pointing improvement holds in data, the 3-D reconstruction should sharpen IceCube's cosmic-ray anisotropy maps, especially for events whose cores are outside IceTop, since those events currently enter anisotropy analyses with much larger directional errors.","A testable extension would be a two-stage fit that first obtains the direction from the unified likelihood and then re-fits the core using only IceTop charge information; this could recover the IceTop-alone core resolution while keeping the improved pointing.","The paper describes the likelihood-combiner architecture as open to additional observables, so the same approach could be applied to future surface/deep hybrid arrays where the surface density and in-ice lever arm differ.","Because the mean $S(r_{\\rm ref})$ versus energy for IT-Uncontained events is shown to separate by primary mass and zenith, a mass estimator built from these relations may extend composition measurements to energies and angles previously inaccessible."],"forward_implications":["For IT-Contained proton Monte Carlo events, the 3-D reconstruction gives the best 68% pointing resolution of the three methods over the whole studied energy range, at the cost of a 2-3 m worse core resolution than IceTop-alone.","For IT-Uncontained events, the successful reconstruction rate after quality cuts rises from 13.7% (in-ice alone) to 25.3% in the $10^9$-$10^{9.5}$ GeV bin, directly reducing statistical uncertainty at high energy.","Recovered IT-Uncontained events extend to zenith angles near 60 degrees, versus about 30 degrees for contained events, letting the detector sample higher-energy showers near their $X_{\\rm max}$.","The reconstruction yields per-event curvature $c_t$, an electromagnetic/muonic LDF option, and muon-bundle energy-loss information, opening new handles on primary mass and composition.","The widened $S_{125}$-energy relation caused by the slightly worse core resolution marks the main price of the method and a target for future optimization."],"supporting_citations":[{"why":"Describes the IceCube detector geometry and triggering that underlies both surface and in-ice measurements.","marker":"[1]"},{"why":"Supplies the standard IceTop charge LDF and the IceTop-alone reconstruction that the 3-D method extends.","marker":"[4]"},{"why":"Provides the in-ice likelihood formalism for photon arrival probabilities used in the combined fit.","marker":"[5]"},{"why":"Introduces Millipede, the muon energy-loss reconstruction reused as the in-ice energy estimator.","marker":"[7]"},{"why":"Documents the Millipede implementation and in-ice likelihood options adopted by the 3-D fit.","marker":"[8]"},{"why":"Generates the CORSIKA air showers used to build the Monte Carlo events for performance tests.","marker":"[9]"},{"why":"Provides the Sibyll2.1 high-energy hadronic interaction model used in the Monte Carlo production.","marker":"[10]"},{"why":"Defines the IceTop-InIce coincidence quality cuts used to select comparable event samples.","marker":"[11]"}],"fun_headline_variants":["3-D fit nearly doubles cosmic-ray events beyond IceTop","Combined detector fit recovers more cosmic-ray showers","Unified likelihood boosts cosmic-ray event recovery 2x","3-D reconstruction expands IceCube's cosmic-ray reach","Two arrays, one fit: more cosmic-ray events outside IceTop"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that the CORSIKA/Sibyll2.1 Monte Carlo and the 2012 IceCube detector simulation reproduce the real spatial and temporal signals of showers with cores outside IceTop, so that the gains measured on Monte Carlo events transfer to real data.","fun_headline_variants_meta":{"raw":{"variants":["3-D fit nearly doubles cosmic-ray events beyond IceTop","Combined detector fit recovers more cosmic-ray showers","Unified likelihood boosts cosmic-ray event recovery 2x","3-D reconstruction expands IceCube's cosmic-ray reach","Two arrays, one fit: more cosmic-ray events outside IceTop"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000414,"raw_usage":{"total_tokens":2132,"prompt_tokens":929,"completion_tokens":1203,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":545,"completion_tokens_details":{"reasoning_tokens":1122}},"tokens_in":545,"tokens_out":1203,"duration_ms":11487,"temperature":1.0,"reasoning_tokens":1122,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T12:02:14.739958+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Apply the 3-D reconstruction to real IceCube data for IT-Uncontained events and compare reconstructed core positions against an independent in-ice-only muon track: if reconstructed cores still cluster toward the IceTop center, or if the expected gain in recovered events over in-ice-alone reconstruction does not appear, the central claim is refuted. A sharper test would check the improved pointing resolution using the Moon shadow or a known source as an independent arrival-direction reference.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Describes the IceCube detector geometry and triggering that underlies both surface and in-ice measurements."},{"cited_title":"Muon Track Reconstruction and Data Selection Techniques in AMANDA","cited_arxiv_id":"astro-ph/0407044","evidence_quote":"Provides the in-ice likelihood formalism for photon arrival probabilities used in the combined fit."},{"cited_title":"The IceCube Collaboration: contributions to the 30th International Cosmic Ray Conference (ICRC 2007)","cited_arxiv_id":"0711.0353","evidence_quote":"Introduces Millipede, the muon energy-loss reconstruction reused as the in-ice energy estimator."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Generates the CORSIKA air showers used to build the Monte Carlo events for performance tests."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the Sibyll2.1 high-energy hadronic interaction model used in the Monte Carlo production."}],"review_version":1}