{"id":"fcd5ee52-7975-4b78-b22a-98a80b274410","arxiv_id":"2411.08943","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":4,"one_line_summary":"A new z_reion-based merger-tree algorithm applied to the THESAN simulations finds three growth phases for ionized bubbles and a single dominant bubble that emerges by z≈9-10, with a deficit near 10 cMpc in the final segmented size distribution.","lead":"Using the THESAN simulation suite, this paper builds merger histories for ionized hydrogen bubbles by growing regions from maps of reionization time, and finds that one dominant bubble emerges by redshift 9-10, well before reionization completes. The new tree-based framework gives 21cm observers and theorists a way to connect bubble sizes, merger activity, and source models during the epoch of reionization.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The R_eff≈10 cMpc deficit and the 'main bubble dominates growth' narrative both follow from the tree's freeze-on-merge rule; without a symmetric watershed or instantaneous bubble-size comparison, the central claim may be an artifact of the segmentation.","rationale":"The reader's weakest-assumption analysis and my own reading converge on the same load-bearing point: Sec. 2.3's asymmetric merge rule, which freezes smaller groups upon merging, is what produces both the early 'dominance' of the main bubble in the active-growth statistics and the sharp deficit at ~10 cMpc in the final segmented size distribution. The paper provides robustness checks across smoothing scales, grid resolutions, neighbor definitions, and physics variations, but all of these vary inputs while holding the tree algorithm fixed. They confirm that the qualitative three-phase picture and the timing of the main merging event are stable to those choices; they do not test whether the headline size-distribution deficit is a property of reionization or a property of the segmentation. The comparison to watershed in Sec. 2.3 is explicitly qualitative and even emphasizes that the watershed would behave differently, so the missing quantitative comparison is not a minor omission. I agree with the reader's conditional verdict: the paper is a competent and novel analysis of a merger-tree construction, and the three-stage description is likely robust, but the abstract's 'characteristic scale' claim and the claim that the main bubble is one of the few actively growing bubbles are not yet shown to be method-independent. The proposed test, a watershed segmentation on the same field plus instantaneous FOF size distributions at fixed redshifts, would settle whether the deficit is physical. This is not an ad hominem concern; it is a request for a specific control calculation that the paper itself identifies as the key difference between its method and existing approaches.","tokens_in":23781,"tokens_out":2722,"duration_ms":27858,"concrete_test":"Run the standard watershed transform (basins continue growing after contact, boundaries at first contact) on the same 512^3 z_reion grid at 125 ckpc smoothing, and also compute instantaneous FOF bubble-size distributions from x_HII at z = 9, 8, and 7. If the watershed final distribution and the instantaneous FOF distributions both show a deficit or absence of R_eff≈10 cMpc bubbles after the main percolation event, the finding is physical. If intermediate-size bubbles appear in either test, the deficit and the 'few actively growing bubbles' claim are artifacts of the freeze-on-merge rule and should be reframed as properties of the tree-based segmentation.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The paper's central claims (abstract; Sec. 5 items ii-iii) rest on Sec. 2.3's asymmetric merge rule: when groups meet, the largest by volume becomes the parent and all smaller groups are removed from further consideration, 'no longer expand independently.' This freeze rule guarantees that any bubble that merges into the eventual main bubble stops growing at the merger moment. Consequently, the 'sharp decline' at R_eff≈10 cMpc in the final segmented distribution (Sec. 3.2.2, Fig. 14) and the statement that the main bubble is 'one of the few bubbles actively growing' (Sec. 5 ii) are built into the tree construction rather than being independent properties of reionization. The paper explicitly contrasts this with watershed segmentation in Sec. 2.3 ('regions continue to grow even after merging') but never computes a watershed or any instantaneous, non-freezing bubble finder on the same z_reion field. Physical bubble sizes at fixed redshift, e.g., FOF or spherical-average on x_HII fields, could show intermediate ~10 cMpc regions existing at z~8-9 while they are still ionizing; the final frozen segmentation hides them. Thus the abstract's 'characteristic scale' deficit may be an artifact of the algorithm rather than a physical feature of reionization.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper introduces a 'bubble tree' algorithm that operates on the spatially resolved reionization-redshift field (z_reion) of the THESAN radiation-hydrodynamic simulations. The algorithm identifies local maxima in z_reion as bubble seeds, expands groups into neighboring cells in descending z_reion order, and records merger events in which the largest group by volume absorbs smaller groups, which then stop expanding independently. Applying this tree to the z_reion field, the authors partition the simulation volume into a 'natural ionization history' and identify three growth stages: isolated expansion (z ≳ 11), a merging/percolation phase (z ≈ 9-11), and a final phase dominated by a single large bubble (z ≲ 9). The central claims are that the largest bubble emerges by z ≈ 9-10 and comes to dominate the box well before the midpoint of reionization, and that the final segmented bubble size distribution shows a sharp deficit or characteristic scale near R_eff ≈ 10 cMpc because the main bubble absorbs all intermediate-sized groups. The paper also studies the dependence of these results on smoothing scale, grid resolution, neighbor definition, and the different physics variants in the THESAN suite.","tokens_in":24059,"tokens_out":5799,"duration_ms":56721,"significance":"If the central claims hold, the paper offers a new chronological framework for EoR bubble growth that is analogous to dark-matter merger trees and could be useful for interpreting 21-cm observations. The manuscript has clear strengths: the algorithm is described precisely enough to be reimplemented; the analysis code is publicly released; the resolution, neighbor-choice, and smoothing tests (Appendices A and B) are appropriate; and no physical parameters are fitted, with THESAN serving as an independent input. The three-stage picture and the 10 cMpc deficit are, however, generated by a central algorithmic choice -- the asymmetric freeze-on-merge rule in Sec. 2.3 -- and the paper does not demonstrate that these features survive in an alternative, non-freezing segmentation. Because the abstract and conclusions present the dominant bubble and the ~10 cMpc deficit as physical properties of reionization, the significance of the work stands or falls on whether those features are robust to the segmentation choice.","major_comments":[{"comment":"The freeze-on-merge rule is load-bearing for all three headline results, not a neutral bookkeeping choice. Because a group that merges into a larger one 'no longer expand[s] independently,' the merging-rate peak in Fig. 6, the convergence of the effective number of bubbles to one in Fig. 8, the early dominance of the main bubble in Fig. 9, and the disappearance of ~10 cMpc bubbles in Figs. 10 and 14 all directly reflect the tree rule. The manuscript explicitly contrasts this rule with watershed segmentation (where 'regions continue to grow even after merging') but never computes a watershed or any instantaneous, non-freezing bubble finder on the same z_reion field. I request a concrete comparison: run a watershed segmentation on the same z_reion field, or compute instantaneous bubble-size statistics (e.g., FOF or spherical-average on the x_HII fields at fixed redshifts z ~ 8-9), and show whether a dominant largest region and the ~10 cMpc deficit survive in that representation. Without such a check, the central claims are statements about the tree construction rather than established properties of reionization.","section":"Sec. 2.3 ('Expansion and Merging'); Sec. 5 items (i)-(iii)"},{"comment":"The sharp cutoff of the R_eff(z_merge) curve near 10 cMpc and the spike at R_box are partly built into the data products. The main bubble never merges, so 'there is essentially only one bubble past that range' is true by construction of the tree. In addition, the spike at the high-radius end is inflated by the rule in Sec. 2.3 that all cells still neutral at z = 5.5 are assigned to the final group 'as if they were ionized at the final redshift.' The abstract's statement that the deficit 'indicates a characteristic scale' is therefore not yet established as a physical feature of the simulated reionization. A physical interpretation requires an algorithm-independent size measure at fixed redshift, together with a statement of how the final-redshift neutral-cell assignment affects the high-radius peak.","section":"Sec. 3.2.2, Fig. 14"},{"comment":"The 'emergence' of the main bubble at z ≈ 9-10 is defined inside the tree by the parent-selection rule, under which the largest group by volume becomes the parent and all secondaries freeze. The paper itself notes that at z ≳ 11 'there are several bubbles of relatively equal size that compete to emerge as the one that happens to be the largest,' which indicates that the early-time identity of the main bubble is not driven by a physically distinct object. Since the volumes used in the parent selection are produced by the algorithm's expansion ordering, the claim that the largest bubble 'emerges well before the midpoint of reionization' needs to be demonstrated in an instantaneous or watershed segmentation before it can be presented as a finding about reionization rather than about the tree.","section":"Sec. 3.1.3, Fig. 9"}],"minor_comments":[{"comment":"The redshift at which the largest bubble establishes dominance is quoted inconsistently: the abstract and Sec. 5 say z ≈ 9-10, while Sec. 3.1.3 says f_main increases rapidly 'by z ≈ 10-11' and the Fig. 9 caption says the main bubble 'becomes the main bubble at z ~ 10.5.' Please harmonize these numbers.","section":"Abstract vs. Sec. 3.1.3 and Fig. 9"},{"comment":"The legend label 'GlobalGlobal' in the top panel appears to be a typo for 'Global'; the duplicated word should be removed.","section":"Fig. 6"},{"comment":"The sentence 'smaller bubbles do have the opportunity to get large enough to get a centre of volume displacement as large as shown in the figure' appears to be missing a negative; as written it contradicts the preceding sentence that only larger bubbles achieve large displacements.","section":"Sec. 3.2, Fig. 15 discussion"},{"comment":"The triaxiality formula is typeset in a confusing way; it should read T = (λ_3^2 - λ_2^2)/(λ_3^2 - λ_1^2), and the current rendering makes the squared eigenvalues unclear.","section":"Eq. (2)"},{"comment":"The reference 'Einasto, J. Suhhonenko, I. Liivamägi, L. J. Einasto, M. 2018' does not follow the journal's author citation style and should be formatted as 'Einasto J., Suhhonenko I., Liivamägi L. J., Einasto M., 2018.'","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The stress-test concern raised by the reader lands squarely: the two headline results -- early dominance of the largest bubble and the ~10 cMpc deficit -- are direct consequences of the asymmetric freeze-on-merge rule in Sec. 2.3, and the paper provides no alternative-segmentation check. This is not an irreparable flaw: the algorithm is well specified, the code is public, and the required test (a watershed segmentation of the same z_reion field, or instantaneous FOF bubble sizes at z ~ 8-9) is well within the scope of the manuscript. I therefore recommend major revision rather than rejection. The paper would also be strengthened by an explicit statement that the 'characteristic scale' is a property of the merger-tree segmentation unless confirmed by an instantaneous measurement."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: the z_reion-based bubble merger tree is genuinely new, and the three-stage growth picture is credible. But the two headline claims—early dominance of one percolating bubble and a sharp deficit around 10 cMpc—both follow from the tree's freeze-on-merge rule, and the paper never compares to a watershed or any instantaneous bubble finder. That is the main thing to know.\n\nWhat is actually new: the chronological merger tree, with expansion and merging separated, is not in the cited bubble-finder literature (MFP, SPA, FOF, granulometry, watershed). The algorithm is precisely defined, the code is on GitHub, and the THESAN simulation is an independent, state-of-the-art input. They run smoothing, resolution, and neighbor tests, and the three-stage expansion/merger pattern is robust across those. Separating expansion volume from merger volume and building a persistence-based reionization history is a useful diagnostic for EoR simulations and for interpreting future 21cm tomographic data.\n\nWhere it is soft: the stress-test concern lands. Section 2.3 freezes smaller groups once they merge into the largest group, which by construction stops intermediate bubbles from growing further. The early emergence of the main bubble around z≈9–10 and the sharp suppression near 10 cMpc in Fig. 14 are therefore properties of the segmentation, not obviously of the ionized field itself. The paper even contains a sentence contrasting this with watershed segmentation, where regions continue to grow after merging, but never runs that comparison. A watershed or FOF on the same z_reion or x_HII fields, or simply instantaneous bubble sizes at fixed redshifts, would settle it. If the deficit survives, the claim is physical; if it disappears, the abstract overreaches. Either way, the authors should do the test and frame the deficit as tree-dependent until then.\n\nThe natural ionization history is also partly defined by the tree, though the comparison to the global reionization history mitigates that. The triaxiality comparison to Thélie+22 is admittedly not apples-to-apples, which is minor.\n\nBottom line: this is a competent, inventive analysis and a valuable framework. It deserves serious peer review; the code is reproducible and the required comparison is cheap. I would send it out and ask for that watershed/instantaneous-size test, and for the claims to be reframed or confirmed accordingly. The rest of the paper holds up.","headline":"A genuinely new bubble merger tree with a real soft spot: the headline ~10 cMpc deficit and early main-bubble dominance are built into the freeze-on-merge rule, so the paper needs a watershed comparison.","tokens_in":24664,"tokens_out":2702,"would_cite":true,"duration_ms":24176,"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":"A single ionized bubble emerges by z≈9–10, well before reionization's midpoint, and grows to dominate the entire simulated volume.","keywords":["Epoch of Reionization","ionized hydrogen bubbles","reionization redshift","bubble merger tree","percolation","radiation-hydrodynamics simulation","bubble size distribution"],"falsifier":"Re-run the segmentation with a symmetric watershed rule in which both bubbles keep growing after first contact, and check whether the largest bubble still emerges by z ≈ 9–10 and whether the ~10 cMpc deficit still appears in the final size distribution; if either feature vanishes, it is a product of the tree's merge rule rather than of reionization. A complementary check is to measure directly, from time-sliced ionization snapshots, the fraction of ionized volume contained in the largest connected region at each redshift.","tokens_in":23543,"feed_emoji":"🫧","tokens_out":8810,"duration_ms":83114,"temperature":0.7,"pith_summary":"This paper introduces a merger-tree view of ionized hydrogen bubbles during the Epoch of Reionization, built from the redshift at which each simulation cell becomes ionized. Applied to the THESAN radiation-hydrodynamics simulations, the tree shows bubble growth in three stages: slow isolated expansion, accelerated merging and percolation, and a final rapid expansion by a single dominant region. The central finding is that the largest bubble emerges by z ≈ 9–10, before reionization is 10 per cent complete, and that it then absorbs essentially all intermediate-sized bubbles. That produces a sharp deficit of bubbles with effective radii around 10 cMpc in the final size distribution. If correct, reionization is hierarchical and its late, observable phase is shaped by one percolating region rather than by many comparable bubbles.","feed_headline":"One ionized bubble takes over reionization by z≈10","feed_subtitle":"One region dominates by z≈10, so the late 21-cm signal is essentially the story of a single bubble.","key_machinery":"The central object is the bubble tree algorithm, a chronological watershed-like segmentation of the reionization-redshift field. Local maxima in z_reion seed bubble groups; each group grows by taking the neighbouring cell with the highest z_reion value; and when two groups claim the same cell, the largest group by volume becomes the parent and all smaller groups stop growing independently and are recorded as merged. This asymmetric merge rule is what produces a tree rather than a static partition, and it is the mechanism from which the three growth stages, the early emergence of the main bubble, and the ~10 cMpc deficit all follow.","core_discovery":"The paper claims that the spatially resolved reionization redshift, defined as the last time a cell's ionized hydrogen fraction crosses 0.5, carries enough chronological information to reconstruct how individual ionized bubbles grow and merge. The bubble tree built on this field yields a 'natural ionization history': local maxima of z_reion seed bubbles that expand into the neighbouring cell with the highest z_reion, with mergers resolved by absorbing smaller groups into the largest group. Using this tree on THESAN, the authors find three growth stages and show that a single largest bubble emerges by z ≈ 9–10, well before the global midpoint of reionization; it dominates the merging phase, becomes the only significant growing region during the final expansion, and eventually fills the whole box. They additionally find that the final segmented bubble size distribution is underrepresented at R_eff ≈ 10 cMpc, because the main bubble's main merging event absorbs all other large groups before they can grow through that size, while the main bubble itself passes through that range too quickly for many cells to be assigned to it.","pith_inferences":["Editorial extension: Because the bubble tree records the volume, centre-of-mass, and moment-of-inertia tensor at each merger, it can be used to test whether the main bubble is seeded in an unusually overdense region or simply wins a random race among comparable early bubbles; the paper does not address this classification.","Editorial extension: The predicted ~10 cMpc gap should be visible to forthcoming 21-cm experiments as a scale-dependent feature in the power spectrum or in reconstructed bubble-size statistics near x_HI ≈ 0.9; computing those observational forecasts is a natural next step not taken in the paper.","Editorial extension: The same z_reion-based tree could be applied to fast semi-numeric reionization models to map how the main-bubble emergence redshift depends on source efficiency, escape fraction, and minimum halo mass, turning the z ≈ 9–10 result into a model-discriminating statistic."],"forward_implications":["If a single largest bubble is established by z ≈ 9–10, then the observable later stages of reionization, including 21-cm maps and the ionized-volume fraction around x_HI ~ 0.9, are dominated by the expansion of one percolating region rather than by many competing bubbles.","The sharp deficit at R_eff ≈ 10 cMpc is a characteristic scale in the bubble size distribution that can serve as a diagnostic of the main merging event in reionization simulations.","Smoothing at 125 ckpc changes the timing of the growth stages but not the qualitative picture, so the three-stage sequence is robust at that resolution; only very large smoothing near 1 cMpc materially alters the early bubble-growth timing.","Because the main bubble is already established by the time reionization is 10 per cent complete across the physics-variation runs once histories are matched, the emergence of a dominant bubble is a common feature rather than a peculiarity of the fiducial model."],"supporting_citations":[{"why":"Supplies the THESAN simulation suite used throughout the analysis.","marker":"Garaldi et al. 2024"},{"why":"Describes the THESAN setup and the physics-variation runs compared in Section 4.","marker":"Kannan et al. 2022a"},{"why":"Presents the THESAN radiation-hydrodynamics implementation that produces the ionization fields.","marker":"Smith et al. 2022"},{"why":"Defines the watershed method against which the bubble tree is explicitly compared.","marker":"Lin et al. 2016"},{"why":"Provides the reionization-redshift definition and the triaxiality comparison used for bubble shapes.","marker":"Thélie et al. 2022"},{"why":"Establishes z_reion as an effective tracer of bubble sizes at sub-cMpc scales, motivating the method.","marker":"Neyer et al. 2024"},{"why":"Supplies the percolation framework for accelerated bubble growth through merging.","marker":"Furlanetto & Oh 2016"}],"fun_headline_variants":["A single ionized bubble dominates reionization by z≈10","THESAN: one bubble takes over the final reionization stage","Reionization's largest bubble emerges at z≈10 and grows alone","Bubble tree reveals reionization is a single-bubble takeover","Tracking bubble mergers shows one giant wins by z≈10"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is the asymmetric merge rule: when two growing ionized regions touch, the larger one absorbs the smaller one and the smaller one freezes and never expands again, and the paper's headline results would change or disappear if real bubbles instead continue to expand after contact.","fun_headline_variants_meta":{"raw":{"variants":["A single ionized bubble dominates reionization by z≈10","THESAN: one bubble takes over the final reionization stage","Reionization's largest bubble emerges at z≈10 and grows alone","Bubble tree reveals reionization is a single-bubble takeover","Tracking bubble mergers shows one giant wins by z≈10"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000275,"raw_usage":{"total_tokens":1715,"prompt_tokens":1090,"completion_tokens":625,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":706,"completion_tokens_details":{"reasoning_tokens":548}},"tokens_in":706,"tokens_out":625,"duration_ms":134979,"temperature":1.0,"reasoning_tokens":548,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-12T21:14:13.470343+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Re-run the segmentation with a symmetric watershed rule in which both bubbles keep growing after first contact, and check whether the largest bubble still emerges by z ≈ 9–10 and whether the ~10 cMpc deficit still appears in the final size distribution; if either feature vanishes, it is a product of the tree's merge rule rather than of reionization. A complementary check is to measure directly, from time-sliced ionization snapshots, the fraction of ionized volume contained in the largest connected region at each redshift.","supporting_citations":[],"review_version":1}