{"id":"c9d89e1e-40da-420b-8e55-9916f5819a32","arxiv_id":"2607.18179","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":3,"one_line_summary":"In a chiral Cayley-tree hopping model, eigenstates at zero energy are semi-fractal (extended support, multifractal higher moments) for a<0.25 and localized for a>0.25.","lead":"A quantum particle hopping on a tree-shaped graph with random link strengths and a special 'checkerboard' symmetry shows a new type of wave-function statistics called semi-fractal. The result ties together recent experiment and theory on non-ergodic states and predicts a sharp transition to a 'semi-localized' state.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"LDoS-to-eigenfunction conversion (Eqs. 11–13) is the weakest link: for β<1 the singular mean LDoS forces η_c∼N^{−1/β} and an ad hoc β adjustment in ED (Fig. 10); if the conversion fails, the localized phase and a_c≈0.25 are not established.","rationale":"The central claim is a phase diagram derived from the LDoS distribution. All quantitative statements, Eqs. (19)–(23), hinge on Eq. (13). I considered other potential objections: the chiral-mechanism claim is interpretive and supported by the E≠0 control; the two-power-law form is directly visible over many decades; the continuously varying γ is consistent with Ref. [38]. The weakest step is the exchange of limits η→0 and N→∞. For β<1, the mean LDoS diverges as η^{β−1}, so η_c is far smaller than the naive 1/N; at that scale the ED data show the exponent β itself shifts (Fig. 10). The authors' solution, using a shifted β in Eq. (23), introduces an adjustable parameter exactly where the theory needs a prediction. If Eq. (13) fails, the localized phase and a_c are not demonstrated. The semi-fractal side (β>1) is supported by the ED f(α) extrapolations, so the concern does not warrant rejection; it makes the paper a strong conjecture rather than a proof, matching the CONDITIONAL verdict. The proposed direct check of Eq. (13) in the same finite system would settle whether the conversion is valid and whether the phase boundary survives.","tokens_in":16335,"tokens_out":20029,"duration_ms":181342,"concrete_test":"Perform a direct finite-N test of Eq. (13) in a chiral RRG for a=0.6 and N=2^L (L=8,...,14): compute ⟨ρ^q⟩ at the self-consistent η_c=N^{−1/β} (β taken from population dynamics) and I_q=N⟨|ψ|^{2q}⟩ for the level closest to E=0; verify whether I_q=(Nη_c)η_c^{q−1}⟨ρ^q⟩ holds for q in (0,2.5). If the equality fails for q>β, the LDoS-to-eigenfunction conversion is invalid, and the localized-phase prediction is not grounded without a theory of the η_c shift.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central D_q and f(α) results rest on converting finite-η LDoS moments on the infinite tree into finite-N eigenfunction moments via Eq. (13). This uses (i) the two-power-law form (14) with η-independent β,γ, and (ii) the self-consistency condition η_c N⟨ρ⟩∼1, Eq. (11), which for β<1 gives η_c∼N^{−1/β}. Assumption (ii) is not independently checked. In fact, Fig. 10 shows that for a=0.7 the ED tail exponent β changes by ~0.1 when η crosses N^{−1/β}, and the authors then insert this 'adjusted' β into Eq. (23). This is a circular repair: the prediction is matched only after redefining its input. If Eq. (13) is invalid in the β<1 regime, the localized phase and the quantitative phase boundary a_c≈0.25 (read off from β(a)=1) are not supported. The semi-fractal side (β>1) has independent support from Fig. 11, so the concern is specifically the β<1 side and the precise location of a_c. Note that the mean-LDoS=mean-DoS relation invoked in footnote [37] is actually exact for the site-averaged LDoS; the real issue is the single-level dominance in Eq. (5) and the η-independence of the exponents used in the fits.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies a single-particle hopping model on a Cayley tree / random regular graph with chiral symmetry (no on-site disorder, bipartite hopping amplitudes drawn from p(t) ∝ e^{-t^2}/|t|^a). Using cavity equations solved by population dynamics, the authors find that the local density of states at E=0 has a piecewise power-law distribution P(ρ) with ρ^{1+β−γ} at small ρ and ρ^{−(1+β)} at large ρ, with β(a), γ(a) fitted numerically. They convert finite-η LDoS moments to finite-N eigenfunction moments using the self-consistency condition η_c N⟨ρ⟩∼1, obtaining D_q = 1 for q<β and D_q=(β−1)/(q−1) for q>β when β>1 (semi-fractal), D_q=0 for q>β and a non-ergodic form when β<1 (localized), and a jump from 1 to 0 at q=1 when β=1 (semi-localized). The fitted β(a) gives the transition at a_c≈0.25. Exact diagonalization on random regular graphs is used to support the f(α) spectra, and rank-ordered eigenfunction weights are argued to exhibit a power-law hierarchy.","tokens_in":16715,"tokens_out":6996,"duration_ms":69131,"significance":"If correct, the paper would provide a concrete chiral mechanism for the recently reported 'semi-fractal' statistics in quantum simulators, Erdős–Rényi graphs, and β-ensembles, and would sharpen the classification of non-ergodic extended phases. The population-dynamics implementation and the comparison with exact diagonalization on finite random regular graphs are valuable, and the release of code is a plus. The β>1 (semi-fractal) side appears to have reasonably direct numerical support from the f(α) extrapolations in Fig. 11. The main significance, however, depends on the reliability of the LDoS-to-eigenfunction conversion and on the fitted exponents; those are the points that need attention before the broad claims can be accepted.","major_comments":[{"comment":"The localized phase for β<1 rests on Eq. (13) combined with η_c∼N^{−1/β}. This conversion is not independently verified in the β<1 regime. The only direct check, Fig. 10, shows that when η crosses N^{−1/β} the ED tail exponent β changes by ~0.1, and the authors then use this 'adjusted' β in Eq. (23) to match the f(α) data. That is circular: the prediction is brought into agreement by changing one of its inputs. A non-circular test would be to compute I_q from ED eigenfunctions at finite N and compare with Eq. (22) using the population-dynamics β, or to derive the expected η-shift of β from the model. Without this, the quantitative phase boundary a_c≈0.25 and the β<1 side are not supported.","section":"Sec. III B and Sec. V A (Eqs. 21–23, Fig. 10)"},{"comment":"The transition point a_c≈0.25 is read off from the condition β=1 on a fitted curve β(a), with no error bars, no finite-η extrapolation, and no account of the fitting range used for the power-law tails. Since the entire phase diagram depends on these fitted exponents, the absence of uncertainty estimates is load-bearing. The authors should provide bootstrap or range-scan errors for β(a) and γ(a), and show explicitly that the exponent determination is stable as η→0 (or how the finite-η drift affects a_c).","section":"Sec. IV, Fig. 6"},{"comment":"There is an internal inconsistency in the rank-ordered weight analysis. Fig. 13 shows a slope of −0.70, i.e. ⟨w(r)⟩∼N^{μ−1}r^{−μ} with μ≈0.7, but Eq. (29) prints r^{μ}. More seriously, substituting the correct r^{−μ} form gives τ(q)=q−1 for q<1/μ and τ(q)=q(1−μ) for q>1/μ, whereas the LDoS-derived semi-fractal spectrum from Sec. III A gives τ(q)=q−1 for q<β and τ(q)=β−1 (a plateau) for q>β. The statement 'identifies β=1/μ' only matches the crossover point, not the shape of τ(q). For a=−0.5, the paper's own numbers (β≈2.07 from Fig. 11 and μ≈0.7) give 1/μ≈1.43, not 2.07, and D_2 differs (1 from the LDoS theory versus ≈0.6 from the rank profile). This discrepancy must be resolved or the rank-profile section should be explicitly presented as a separate, qualitative observation rather than a derivation of the same D_q.","section":"Sec. V C, Eqs. (29)–(33) and Fig. 13"},{"comment":"The piecewise power-law form (14) is an empirical fit to population-dynamics data, not a derived result, and the subsequent D_q and f(α) calculations assume that both exponents β and γ are independent of η in the entire range η≪ρ≪η^{-1}. The paper does not provide a check that β and γ extracted at finite η (e.g. η=10^{-8}) remain valid as η→0 for all a, nor a discussion of why the two-power-law form should be exact or asymptotically exact. Since Eq. (14) is the input to the Mellin-type moment calculation, the authors should state clearly the numerical/analytical evidence for η-independence, or qualify the results as conditional on this assumption.","section":"Sec. II–III (Eqs. 11–14)"}],"minor_comments":[{"comment":"The displayed formula ⟨w(r)⟩∼N^{−(1−μ)}r^{μ} has the wrong sign in the rank exponent; the figure and the subsequent summation use r^{−μq}. Please correct to r^{−μ}.","section":"Sec. V C, Eq. (29)"},{"comment":"The heading 'CHIRAL OR THOGONAL SYMMETR Y CLASS' contains typos ('OR' should be 'ORTHOGONAL' and 'SYMMETR Y' should be 'SYMMETRY').","section":"Sec. II heading"},{"comment":"The notation ρ^q is overloaded: it is used both for the q-th power of the LDoS and for the disorder-averaged moment ⟨ρ^q⟩. Please use a distinct symbol for the averaged moment, e.g. M_q.","section":"Sec. II, Eq. (5)"},{"comment":"The value γ=2.437 for a=0 is quoted in the caption but the main text/Fig. 6 uses fitted values without explicit table. A small table of β(a), γ(a) with fitting ranges and errors would be useful.","section":"Fig. 2 caption"},{"comment":"Refs. [29] and [42] are marked 'To be posted'; if they are essential to the Discussion, please provide arXiv IDs or checkable details. Otherwise, referencing unpublished works weakens the reproducibility of the claims.","section":"References"}],"recommendation":"major_revision","confidential_remarks":"The paper is potentially important, but the β<1 side and the precise a_c are not yet established because the comparison with exact diagonalization uses an adjusted β. The rank-ordered weight section also contains an inconsistency that needs to be fixed or clearly separated from the main D_q claim. I think the authors can address these with additional analysis and by softening the over-strong statements, so rejection is not warranted; however, as it stands the central claim is not fully supported."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: the semi-fractal side of this paper is in good shape, the localized side less so. The model—nearest-neighbor hopping on a Cayley tree with p(t)∝|t|^{-a}e^{-t^2}—is new and clean, and the population-dynamics data showing two power-law tails in P(ρ) with a-dependent exponents β(a), γ(a) is credible. The continuous evolution of the Mirlin-Fyodorov exponent γ with a is a genuinely new observation, and the E≠0 control restoring γ=3 is a useful check. The step from the two-power-law P(ρ) to D_q is elementary and correct given the ansatz. For β>1, the comparison with exact diagonalization works: the f(α) slope extrapolates to about 2.1, close to β≈2.07 from population dynamics. That is the core evidence for semi-fractality, and it holds up. Credit also for the code on GitHub.\n\nThe soft spots are concentrated on the β<1 side. The conversion from infinite-tree LDoS moments at finite η to finite-N eigenfunction moments, Eq. (13), is the load-bearing step. For β>1 it is standard and harmless. For β<1 the mean LDoS diverges, η_c scales as N^{-1/β}, and the single-level-dominance assumption behind Eq. (5) becomes genuinely delicate. The paper's own Fig. 10 shows that the ED tail slope changes by about 0.1 once η drops below N^{-1/β}; the authors then insert that adjusted β into Eq. (23) to get agreement. That is close to circular: the prediction is matched by redefining the input on the side it is supposed to predict. The stress-test note is right that footnote [37] is not the real problem—the site-averaged mean LDoS does equal the mean DoS exactly; the real question is whether the η→0 limit of the LDoS exponents and the η_c self-consistency give the same statistics as a finite-N eigenfunction. I don't think that is settled for β<1. So the qualitative transition at a_c≈0.25 is plausible, but the location and the localized-side D_q should be treated as provisional.\n\nAlso: the exponents β(a), γ(a) are fitted without error bars, and a_c is read off where β(a)=1. That is a minor issue but worth noting. The chirality-unification claim is a conjecture, supported by the E≠0 control and by two unpublished references; I would ask the authors to mark that clearly. The rank-ordered profile section is suggestive, but the claimed β=1/μ identification does not reproduce the D_q shape from the LDoS derivation—for constant μ the high-q D_q tends to 1-μ, not zero. Perhaps it is a typo (Eq. (29) writes r^μ where it should be r^{-μ}) or a separate typical-moment story, but as written the two sections don't line up quantitatively.\n\nWho should read this: anyone working on non-ergodic extended phases, chiral symmetry, or the recent quantum simulator results. It deserves a serious referee. I would send it out, but tell the referee to focus on Eq. (5) and the β<1 conversion, and ask for error bars on β(a) and a stability check on a_c.","headline":"Semi-fractal side is solid; localized side rests on a conversion step that the paper itself adjusts by hand.","tokens_in":17287,"tokens_out":10433,"would_cite":true,"duration_ms":92254,"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 paper claims that chiral symmetry — the bipartite sublattice structure of a hopping model with no on-site disorder — produces a new class of eigenfunction statistics, called semi-fractal, at zero energy, with a sharp crossover to locali","keywords":["chiral symmetry","Cayley tree","local density of states","semi-fractality","multifractality","localization","cavity method","random regular graph"],"falsifier":"Compute the inverse participation ratios I_q = N⟨|ψ|^{2q}⟩ for the chiral random regular graph at the level closest to E = 0 for sizes up to N = 2^{16}, setting η to the level spacing; if the crossover from N^{1−q} to N^{1−β} occurs at q differing from β(a), or if the rank-ordered weight exponent µ fails to satisfy β = 1/µ, the semi-fractal scenario is falsified.","tokens_in":16128,"feed_emoji":"🌳","tokens_out":4786,"duration_ms":44686,"temperature":0.7,"pith_summary":"The paper argues that chiral symmetry, realized by the bipartite structure of an infinite Cayley tree with random nearest-neighbor hoppings, generates a previously missing type of wave-function statistics at the band center. Solving the cavity equations by population dynamics, it finds that the distribution of the local density of states has two power-law tails whose exponents β(a) and γ(a) vary continuously with the parameter a controlling the abundance of weak links. From these tails it derives a 'semi-fractal' phase, where eigenstates spread over an extensive fraction of the system yet their higher moments scale like a multifractal. As a crosses a critical value a_c ≈ 0.25, the system becomes localized, and exactly at the transition the eigenstates are 'semi-localized': fractal dimensions jump from 1 to 0 at q = 1. The paper connects this mechanism to recent observations in quantum simulators, Erdős–Rényi graphs, and related random-matrix ensembles.","feed_headline":"Chirality spawns a half-ergodic, half-fractal eigenstate phase","feed_subtitle":"A tunable hopping distribution puts random-tree eigenstates in a new 'semi-fractal' class between ergodic and localized.","key_machinery":"The organizing mechanism is the chiral (sublattice) symmetry of the bipartite graph, which makes the retarded Green's function pure imaginary at E = 0 and leads to the cavity recursion G_{i→j} = (−E−iη − Σ t² G_{a→i})^{−1}. The paper solves this recursion by population dynamics and finds that the LDoS distribution has two power-law tails connected by the Mirlin–Fyodorov symmetry P(1/ρ) = ρ^γ P(ρ). The bridge from P(ρ) to eigenfunction multifractality is Eq. (13), which converts finite-broadening LDoS moments into finite-size participation moments through the self-consistency condition η_c N⟨ρ⟩ ∼ 1. This yields the piecewise-linear spectrum τ(q) and hence the fractal dimensions D_q and the sp","core_discovery":"The central claim is that at zero energy the distribution P(ρ) of the local density of states on the chiral Cayley tree obeys a piecewise power law, P(ρ) ∼ ρ^{1+β−γ} for small ρ and P(ρ) ∼ ρ^{−(1+β)} for large ρ, with β and γ depending continuously on the exponent a of the hopping distribution p(t) ∝ e^{−t²}/|t|^{a}. These tails satisfy the Mirlin–Fyodorov reflection symmetry P(1/ρ) = ρ^γ P(ρ), but γ is not fixed by the symmetry class; here it varies with a. The authors show that such a two-tail distribution implies a piecewise-linear spectrum of fractal dimensions f(α) with a linear segment of slope β, leading to D_q = 1 for q < β and D_q = (β−1)/(q−1) for q > β when β > 1 — the semi-fracta","pith_inferences":["One could test the chirality mechanism directly: in the quantum-simulator experiment, measuring participation entropies for eigenstates near the center of the band should reveal the plateau D_q = 1 for q < β followed by a multifractal branch, with β controlled by the effective disorder strength.","The power-law hierarchy of eigenstate weights suggests an analogy to log-correlated random potentials and 'frozen' multifractality; if that analogy holds, the semi-fractal phase might be characterized by a freezing transition in the statistics of wave-function amplitudes.","The semi-localized point β = 1 may be a generic critical point for disordered chiral systems; a finite-size scaling collapse of the LDoS distribution around the most probable value, which drops by about ten orders of magnitude across the transition, could sharpen the estimate a_c ≈ 0.25.","Because the derivation relies on the equality of mean LDoS and mean DoS, a numerical check of level-spacing fluctuations at the band center for β < 1 would clarify whether the localized-phase boundary is exactly at β = 1 or shifted by rare events."],"forward_implications":["If chirality is the organizing principle, semi-fractal statistics should appear at band center in any disordered system with exact or approximate sublattice symmetry, not only on trees.","At the critical value a_c ≈ 0.25 the wave functions realize the 'semi-localized' phase, which is simultaneously extended in support and localized in higher moments; this phase was previously anticipated but had no known realization.","The exponent γ varies continuously with the disorder parameter, in contrast to the standard Dyson symmetry classes where γ is fixed (γ = 3) — chiral classes thus carry a non-universal Mirlin–Fyodorov exponent.","Away from E = 0 the two power-law tails disappear and the standard Dyson value γ = 3 is restored, so the phenomenon is a zero-energy, chirality-protected effect.","The rank-ordered eigenstate probabilities exhibit a power-law hierarchy, giving a microscopic picture of semi-fractality as a hierarchy of wave-function weights rather than a fractal support set."],"fun_headline_variants":["Chiral tree eigenstates split between ergodic and fractal","Tunable hopping yields semi-fractal phase on Cayley tree","New eigenstate class: semi-fractal on a Cayley tree","Hopping exponent tunes chiral tree into semi-fractal","Chiral Cayley tree hosts semi-fractal eigenstates"],"cache_read_input_tokens":2304,"weakest_assumption_plain":"The derivation of the multifractal spectrum assumes that the two-power-law form of the LDoS distribution holds all the way to vanishing broadening η and that the mean LDoS equals the mean density of states, so that the self-consistency condition η_c N⟨ρ⟩ ∼ 1 faithfully converts infinite-size LDoS moments into finite-size eigenfunction moments; if level-spacing fluctuations or the divergent mean LDoS for β < 1 break this conversion, the predicted D_q and the phase boundary a_c","fun_headline_variants_meta":{"raw":{"variants":["Chiral tree eigenstates split between ergodic and fractal","Tunable hopping yields semi-fractal phase on Cayley tree","New eigenstate class: semi-fractal on a Cayley tree","Hopping exponent tunes chiral tree into semi-fractal","Chiral Cayley tree hosts semi-fractal eigenstates"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000752,"raw_usage":{"total_tokens":3215,"prompt_tokens":810,"completion_tokens":2405,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":554,"completion_tokens_details":{"reasoning_tokens":2331}},"tokens_in":554,"tokens_out":2405,"duration_ms":16590,"temperature":1.0,"reasoning_tokens":2331,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-01T15:44:51.668146+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute the inverse participation ratios I_q = N⟨|ψ|^{2q}⟩ for the chiral random regular graph at the level closest to E = 0 for sizes up to N = 2^{16}, setting η to the level spacing; if the crossover from N^{1−q} to N^{1−β} occurs at q differing from β(a), or if the rank-ordered weight exponent µ fails to satisfy β = 1/µ, the semi-fractal scenario is falsified.","supporting_citations":[],"review_version":1}