{"id":"71a9bcb5-e869-4681-87a6-381ee39b8c35","arxiv_id":"2502.04869","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For a two-parameter skew product family, the basins of the two chaotic attractors are shown to be intermingled in an open parameter region, with a separating invariant graph and a multifractal spectrum for the stability index.","lead":"This paper studies a two-parameter family of maps on the unit square with two chaotic attractors living on two invariant lines. It identifies exact parameter regions where the basins of the two attractors are intermingled, locates blowout bifurcation curves, and computes the fractal dimensions of the sets where the local stability index takes given values.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The multifractal spectrum is not proved: (5.3) identifies the stability-index level sets with Birkhoff-average sets without pointwise estimates on the null sets A_i(σ), and §5.1 invokes [Kel17, Prop. 1.6] for νq without checking both normal Lyapunov exponents.","rationale":"The reader's weakest assumption was condition (H1) and the imported equality of the separating graphs. Those are real gaps, but the most load-bearing defect is in the multifractal proof itself: Section 5 asserts the bridge (5.3) between the stability-index level sets and Birkhoff-average level sets. The paper's own estimates in Section 4 are a.e. statements that discard νψ-null sets, while the level sets in Theorems 5 and 6 are precisely νψ-null for all non-ergodic values of the stability index. No pointwise version of Lemma 4.11 or of the estimates leading to (4.15)-(4.16) is supplied for such points. The text calls the argument a sketch and cites standard results from [PW97], but those results apply to equilibrium measures and do not automatically transfer to the sets defined by 'for every y below φ*'. A second concrete gap occurs in §5.1: Proposition 1.6 of [Kel17] is invoked for νq after checking only ∫log dg(φ1)dνq<0, whereas the paper's own use of that proposition in Lemma 3.5 requires negative exponents for both bounding graphs; the required negativity of λνq(φ0) is not shown. This does not invalidate Theorem 1, whose computations for the SRB exponents are explicit and convincing, nor the conditional Theorem 4. However, the claimed multifractal spectra are the central novelty of the title and abstract, so the proof gap is substantial. The verdict should remain conditional: the concern is addressable by adding a genuine proof of (5.3), or by restricting the multifractal results to the SRB case and proving the νq conditions there, but the current manuscript does not yet contain that argument.","tokens_in":29169,"tokens_out":16249,"duration_ms":167773,"concrete_test":"Verify (5.3) on a periodic base point where the Birkhoff average differs from the νψ-ergodic value, e.g. x=0 for a=0.45, b=0.33, with λ(φ0)=log(1+a)>0. Compute σνac(0,y) directly from Definition 2.15 for all y with φ0<y<φ*(0) and compare with the right-hand side of (5.3); if the values differ, or if the vertical interval is empty, the identification fails. Additionally, for the SRB case compute the pressure-based dimension T(q)+qσ(q) at q=0.2,0.5,0.8 via a transfer-operator calculation and compare with an independent estimation of the dimension of the original level set (2.16); a mismatch would refute (5.3).","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Theorems 5 and 6 rest on the unproved equality (5.3), which asserts that the stability-index level set A_{i,νψ}(σ) defined in (2.16)-(2.17) equals the Birkhoff-average set {x : lim t*_i S_n log dg(φ_i) / (-S_n log|df|) = σ}. Theorem 4 and Section 4 evaluate σν only for νψ-a.e. x, using Birkhoff's theorem and sequences ε_k, n_k satisfying (4.14). For any σ different from the single ergodic value of νψ, the relevant set is νψ-null, so those a.e. estimates do not apply; no pointwise control is given for points of that null set. Thus (5.3) is an additional assertion, not a consequence of the preceding estimates. Moreover, in §5.1 the measure νq is used to apply [Kel17, Proposition 1.6], but only ∫ log dg(φ1) dνq < 0 is verified. In the earlier use of Proposition 1.6 in Lemma 3.5 and Corollary 2.12, both bounding graphs φ0 and φ1 are required to have negative Lyapunov exponents; the paper does not prove λνq(φ0) < 0. Consequently, even in the SRB case ψ = -log|df|, the dimension formula is not established by the text. This is separate from, and more direct than, the unresolved question of whether condition (H1) holds for non-SRB potentials.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies a two-parameter family of skew-product maps F_{a,b}(x,y)=(2x mod 1, g_{a,b,x}(y)) with two invariant lines. It identifies an open region Γ in the (a,b)-plane in which the two invariant chaotic attractors have intermingled basins separated by an invariant graph φ*, proves the complementary chaotic-saddle regime, and locates blowout bifurcations on the two boundary curves. The second half uses thermodynamic formalism to compute Loynes exponents and the stability index for Gibbs measures satisfying condition (H1), and claims multifractal spectra of the stability-index level sets as Legendre transforms of pressure functions.","tokens_in":29558,"tokens_out":14443,"duration_ms":146015,"significance":"Theorems 1 and 2 rest on explicit derivative computations and give a clean, checkable example of intermingled basins and blowout bifurcations. The paper is also honest that the existence of the separating graph is imported from Keller's work. No parameters are fitted: the stability index and the pressure functions are defined directly from the dynamics. If the multifractal dimension formulas were established, they would be a valuable concrete illustration of the Legendre-transform formalism for stability-index level sets. As it stands, however, the multifractal part contains a load-bearing gap, so the full significance claimed in the abstract is not yet realized.","major_comments":[{"comment":"The displayed identification of A_{i,νψ}(σ) with a Birkhoff-average set is asserted rather than proved, and as written it is also incorrect in sign. For νψ-a.e. x and y < φ*(x), Theorem 4(2) gives σνψ(x,y) > 0, whereas the quotient Q0(x) = lim t*0 S_n log dg(φ0) / (-S_n log|df|) converges to t*0 λνψ(φ0) / ∫ log|df| dνψ < 0. Thus A0,νψ(σ) would need to be identified with {Q0 = -σ}, not {Q0 = σ}. The same sign issue occurs for A1 because (2.17) defines A1,νψ(σ) through σνψ = -σ. Moreover, Theorem 4 only computes the stability index on a νψ-full-measure set via the sequence (4.14); for σ different from the ergodic value the level set is νψ-null, and no pointwise estimates on that null set are supplied. Therefore (5.3) is an additional load-bearing assertion, and Theorems 5(2) and 6(2) do not follow from the preceding estimates.","section":"Section 5, Eq. (5.3)"},{"comment":"In Case (1) the paper concludes from ∫ log dg(φ1) dνq < 0 that φ* is defined νq-a.e., citing [Kel17, Proposition 1.6]. In the earlier use of Proposition 1.6, namely Corollary 2.12, the condition is applied after verifying negative Lyapunov exponents for both bounding graphs φ0 and φ1 with respect to the relevant measure. For q ≠ 0 the measure νq differs from νψ, and the text does not verify λνq(φ0) < 0. Hence the use of Proposition 1.6 is missing a hypothesis, and the Moran-cover argument that relies on the νq-defined graph is unsupported. The same omission affects Theorem 5, whose proof is only said to be similar.","section":"Section 5.1, proof of Theorem 6"},{"comment":"The theorems are stated as dimension formulas for the stability-index level sets A0,νψ(σ) and A1,νψ(σ), but the proof only attempts to analyze Birkhoff-average level sets after replacing them by (5.3). Even after correcting the sign, the paper would need to prove that the stability-index condition 'for every y < φ*(x)' is equivalent to a pointwise Birkhoff-average condition. No such equivalence is shown: Theorem 4 controls σνψ only for νψ-a.e. x and all y in the respective half-neighborhood, and the exceptional null sets are not controlled. The final dimension claims therefore need a genuinely new argument or a restricted statement, e.g. for the Birkhoff-average sets defined in (5.3).","section":"Section 5, proof of Theorems 5 and 6"}],"minor_comments":[{"comment":"The formula 'σ0ν(x,y) = t*1 · -log λ_{a,b,ν}(φ1) / ∫ log|df| dν' should read '-λ_{a,b,ν}(φ1)' instead of '-log λ_{a,b,ν}(φ1)'.","section":"Section 4.2, Case (2)"},{"comment":"Hypothesis (i) is mis-stated: the integral should be written as ∫ e^{t log dg^n_{a,b,x}(φ0)} dν, and the claimed bound cannot follow from boundedness of dg alone; the text should specify the intended inequality and its domain in t.","section":"Section 4, Proposition 4.7"},{"comment":"There are minor typos: 'Lemm 12.1.4' appears twice, and 'P(psi)' should be 'P(ψ)'.","section":"Section 4, Lemma 4.5 and Corollary 4.6"},{"comment":"The chain of equalities in item (1) is confusing because σ0νψ is used with different signs in the statement and in the proof; the notation should be aligned so that σνψ = σ1νψ - σ0νψ is transparent in both places.","section":"Theorem 4 statement"},{"comment":"Condition (H1) is verified only for the SRB potential ψ = -log|df|. The paper should state explicitly that Theorems 3-6 for general ψ are conditional on this unverified hypothesis, or provide an example of a non-SRB potential satisfying it.","section":"Section 2, Condition (H1)"},{"comment":"The word 'parametetric' in the title is a typo, and in the abstract 'these attractors either exhibits' should be 'these attractors either exhibit'.","section":"Title and abstract"}],"recommendation":"major_revision","confidential_remarks":"The first half of the paper, Theorems 1-4, is credible and appropriate for the journal. The multifractal section, however, needs substantial revision: the sign error in (5.3), the missing two-sided Lyapunov verification for the auxiliary measures νq, and the lack of pointwise control on νψ-null sets together mean that Theorems 5 and 6 are not supported by the present text. I would not require new numerics, but the authors must either prove the corrected identification or explicitly restrict the dimension results to the Birkhoff-average sets for which the standard thermodynamic argument applies."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: the paper is worth taking seriously as an explicit worked example, and the intermingled-basin/chaotic-saddle part (Theorems 1 and 2) largely holds up. The multifractal part (Theorems 5 and 6) does not: equation (5.3) is an unproved identification, and the dimension formulas are not established as written. If you cite this paper, cite it for the example and the parameter regions, not for the multifractal spectra.\n\nWhat's genuinely new: the two-parameter family with exact regions Γ0, Γ1, Γ and the blowout curves b = a/(1 ± a). The proofs of Theorems 1 and 2 are direct: explicit derivative computations give the SRB normal Lyapunov exponents, and the dense-periodic-point argument supplies the local riddling. The paper is also honest about what it imports from Keller, and Remark 2.16 correctly notes that Theorems 3 and 4 reduce to [Kel17] in the SRB case. That is the right way to use prior work.\n\nThe soft spots are concentrated in Section 5. The sets A_{i,ν}(σ) are defined pointwise: σν(x,y) = σ for every y below or above φ*. Theorem 4 only determines σν for ν-a.e. x. For any σ that is not the single ergodic value, the set is a ν-null set, and no pointwise control is given. So (5.3), which identifies A_{i,ν}(σ) with a Birkhoff-average set, is an additional assertion, not a consequence of Birkhoff's theorem. Relatedly, in the proof of Theorem 6 the paper invokes [Kel17, Prop. 1.6] for νq after checking only λνq(φ1) < 0; that proposition requires both bounding graphs to have negative normal Lyapunov exponents, and λνq(φ0) < 0 is not shown. So even in the SRB case the proof does not close. The abstract's \"fractal boundary curve\" is also stronger than what is proved: φ* is a measurable invariant graph, with no fractal dimension or regularity established. And the subcritical/supercritical labels on the blowout bifurcations are asserted from the sign of Λ_SRB; the bifurcation type is not actually analyzed. These are fixable, but they are real gaps, not cosmetic ones.\n\nWho gets value: anyone working on riddled or intermingled basins who wants a clean explicit family with computable parameter boundaries. The example deserves to be known. But the multifractal claims should not be relied on until the Section 5 gaps are addressed.\n\nRecommendation: send it to peer review, with a referee specifically asked to check Section 5. A major revision is needed – either prove (5.3) with genuine pointwise control, or restrict the claims and drop the full Legendre spectrum.","headline":"A solid explicit example of intermingled basins, but the multifractal spectrum is not proved as written; send to review with Section 5 flagged.","tokens_in":30078,"tokens_out":6856,"would_cite":true,"duration_ms":66959,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["37C70","37C40","37H15","37C45"],"pacs":[],"model":"deepseek-v4-flash","headline":"For an open region of parameters, a two-parameter skew product family has two chaotic attractors whose basins intermingle, separated by a single invariant graph whose stability-index level sets have dimensions given by a Legendre transform.","keywords":["invariant graph","chaotic Milnor attractor","riddled basin","intermingled basin","blowout bifurcation","stability index","multifractal analysis","thermodynamic formalism"],"falsifier":"Compute $\\ell_{a,b}(x)$ and $r_{a,b}(x)$ numerically for a dense grid of $x$ and for fine parameter grids inside $\\Gamma$; if for some set of positive $\\nu_{\\mathrm{ac}}$-measure the two functions differ, the separating graph $\\phi^*$ of Theorem 1 fails and the dimension formulas collapse. A cheaper check is to measure the stability index at points just above and below the numerically located boundary for a fixed parameter in $\\Gamma$ and compare the values with the constants predicted by Theorem 4.","tokens_in":28975,"feed_emoji":"🌀","tokens_out":8236,"duration_ms":78139,"temperature":0.7,"pith_summary":"The paper studies the family $F_{a,b}(x,y)=(2x\\bmod 1,\\,g_{a,b,x}(y))$ on the unit square, where the fiber map pushes up or down depending on which half of the square $x$ lies in. It tries to establish that for an open region $\\Gamma$ in the $(a,b)$-plane the two invariant lines $y=0$ and $y=1$ each support a chaotic Milnor attractor, and that the two basins are intermingled: every neighbourhood of almost every point in one basin contains positive measure of the other basin. It further claims that a single invariant graph $\\phi^*$ separates the basins, that crossing the boundary curves of $\\Gamma$ produces blowout bifurcations in which one attractor becomes a chaotic saddle, and that the local stability index and the Hausdorff dimension of its level sets are given by explicit thermodynamic-pressure formulas. A sympathetic reader would care because this gives a concrete, two-parameter model in which riddled basins, intermingled basins, and their multifractal spectra are described quantitatively rather than only observed.","feed_headline":"Intermingled basins fill an open region of a two-parameter map family","feed_subtitle":"Near almost every point in either basin, arbitrarily close initial conditions head to the other attractor.","key_machinery":"The central object is the skew product map $F_{a,b}(x,y)=(f(x),g_{a,b,x}(y))$ with base map $f(x)=2x\\bmod 1$ and fiber maps $g_a(y)=y+ay(1-y)$ for $x\\in[0,1/2)$ and $g_b(y)=y-by(1-y)$ for $x\\in[1/2,1]$. The two constant graphs $\\phi_0(x)=0$ and $\\phi_1(x)=1$ are invariant, and the proof of Theorem 1 rests on the normal Lyapunov exponents of the SRB measures on them: for $\\Gamma_0$, $\\Lambda_{\\mathrm{SRB}}(A_0)=\\tfrac12\\log(1+a)+\\tfrac12\\log(1-b)<0$, and for $\\Gamma_1$, $\\Lambda_{\\mathrm{SRB}}(A_1)=\\tfrac12\\log(1-a)+\\tfrac12\\log(1+b)<0$. Riddling is obtained by applying a cited criterion that combines negative SRB normal exponent with a dense set of periodic points having positive normal exponent; the separating graph $\\phi^*$ is identified with the common value of the lower boundary function $\\ell_{a,b}(x)=\\sup\\{y:\\lim_n g^n_{a,b,x}(y)=0\\}$ and the upper boundary function $r_{a,b}(x)=\\inf\\{y:\\lim_n g^n_{a,b,x}(y)=1\\}$, whose equality is imported from a cited proposition. The stability-index and multifractal results are carried by the pressure functions $p_{i,\\psi}(t)$ and by the implicitly defined curves $T(q)$, $S(q)$, whose Legendre transforms give the Hausdorff dimensions of the level sets.","core_discovery":"For $F_{a,b}$ defined by (2.2)--(2.3), if $(a,b)$ lies in $\\Gamma=\\{0<a<1/2,\\,0<b<1/2,\\,b>a/(1+a),\\,b<a/(1-a)\\}$, then $F_{a,b}$ has two chaotic essential Milnor attractors $A_0\\subset I\\times\\{0\\}$ and $A_1\\subset I\\times\\{1\\}$ whose basins are locally riddled and intermingled (Theorem 1). The basins are separated by a $\\nu_{\\mathrm{ac}}$-almost everywhere invariant graph $\\phi^*$ with $0<\\phi^*<1$. Crossing $b=a/(1+a)$ or $b=a/(1-a)$ turns one of the attractors into a chaotic saddle and produces a blowout bifurcation, one supercritical and one subcritical (Theorem 2 and Corollary 2.13). Under condition (H1), the stability index $\\sigma_\\nu(x,y)$ is constant $\\nu_\\psi$-almost everywhere on each side of $\\phi^*$, with values written in terms of positive zeros $t^*_0,t^*_1$ of pressure functions $p_{i,\\psi}(t)=P(\\psi+t\\log dg_{a,b}(\\phi_i))$ (Theorems 3 and 4). The level sets $A_{i,\\nu}(\\sigma)$ have Hausdorff dimension $T(q)+q\\sigma(q)$ or $S(q)+q\\sigma(q)$, where $T$ and $S$ are defined implicitly by pressure equations, so the multifractal spectrum of the stability index is the Legendre transform of a thermodynamic function (Theorems 5 and 6).","pith_inferences":["The same two-parameter mechanism should work for any piecewise expanding Markov base map with two branches whose fiber maps have negative Schwarzian derivative, since the paper's Remark 2.1 states the arguments extend and the only quantitative input is the normal Lyapunov exponents.","If the stability index is measured numerically at points just above and below the numerically located boundary curve for a parameter inside $\\Gamma$, the predicted constant values from Theorem 4 give a direct, quantitative test of the thermodynamic formalism.","The same pressure-based Legendre-transform construction could be applied to other skew products with finitely many invariant graphs to produce dimension spectra for riddled basins, although the existence of a single separating graph would need to be established separately in each case.","If condition (H1) fails for a potential other than $-\\log|df|$, the dimension formulas may still hold for the SRB measure, but the claimed generality over Gibbs measures would require an additional argument."],"forward_implications":["For every $(a,b)\\in\\Gamma$, almost every point of the square is eventually attracted to one of the two invariant lines $y=0$ or $y=1$, yet each basin has positive measure and is riddled with the other, so the asymptotic future of a point cannot be inferred from the futures of arbitrarily close neighbours.","The parameter plane splits into three dynamical regimes, with $b=a/(1+a)$ and $b=a/(1-a)$ as the two blowout bifurcation curves; crossing one turns the corresponding attractor into a chaotic saddle while the other attractor retains a locally riddled basin.","For $\\nu_\\psi$-almost every $x$, the stability index $\\sigma_\\nu(x,y)$ is constant on each side of the separating graph $\\phi^*$, with explicit values $t^*_0\\,(-\\lambda_{\\nu_\\psi}(\\phi_0))/\\int\\log|df|\\,d\\nu_\\psi$ and $t^*_1\\,(-\\lambda_{\\nu_\\psi}(\\phi_1))/\\int\\log|df|\\,d\\nu_\\psi$.","The Hausdorff dimension of the stability-index level sets equals $T(q)+q\\sigma(q)$ for the $A_0$ side and $S(q)+q\\sigma(q)$ for the $A_1$ side, so the full multifractal spectrum is described by the Legendre transform of a pressure function.","In the SRB case $\\psi=-\\log|df|$, the stability-index theorems reduce to the previously known intermingled-basin results, confirming that the new pressure-based formulas contain the earlier theory as a special case."],"supporting_citations":[{"why":"Supplies Proposition 3.1, which converts negative SRB normal exponent plus dense positive-exponent periodic points into a locally riddled basin, and negative-minimum plus positive-SRB normal exponent into a chaotic saddle.","marker":"[ABS96]"},{"why":"Supplies Proposition 1.6, used to identify the separating graph $\\phi^*$ and to transfer intermingled-basin conclusions to general Gibbs measures, and the stability-index formalism that Theorems 3 and 4 generalise.","marker":"[Kel17]"},{"why":"Supplies the multifractal analysis of equilibrium measures for expanding maps, including the Moran-cover argument used to compute the Hausdorff dimensions of level sets.","marker":"[PW97]"},{"why":"Introduces riddled basins and provides the original definition and criteria cited for locally riddled basins.","marker":"[Ale+92]"},{"why":"Provides the first example of maps with intermingled basins, the phenomenon this paper studies in a explicitly parametrised family.","marker":"[Kan94]"},{"why":"Supplies the argument used in Lemma 3.5 to show $\\ell_{a,b}(x)>0$ and $r_{a,b}(x)<1$ for $\\nu_{\\mathrm{ac}}$-almost every $x$.","marker":"[BM08]"},{"why":"Provides the definition of Milnor attractor used to characterise $A_0$ and $A_1$ as chaotic essential Milnor attractors.","marker":"[Mil85]"},{"why":"Supplies the large-deviations theorem used in the proof of Theorem 3 to establish the Loynes-exponent limits.","marker":"[PS75]"},{"why":"Provides the absolutely continuous invariant measure $\\nu_{\\mathrm{ac}}$ for the doubling map and the Perron-Frobenius facts used to show the Gibbs measure is fully supported.","marker":"[VO16]"}],"fun_headline_variants":["Two chaotic attractors with intermingled basins","Multifractal spectra of stability index in skew maps","Open region has intermingled riddled basins","Blowout bifurcations in a two-parameter map family"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is that for parameters in $\\Gamma$ the lower and upper boundary functions $\\ell_{a,b}(x)$ and $r_{a,b}(x)$ coincide almost everywhere, so that one separating graph $\\phi^*$ exists, and that condition (H1) holds: a Gibbs measure $\\nu_\\psi$ with negative normal Lyapunov exponents on both invariant lines; the paper verifies the equality only by importing a cited proposition, and verifies (H1) explicitly only for the SRB potential.","fun_headline_variants_meta":{"raw":{"variants":["Two chaotic attractors with intermingled basins","Multifractal spectra of stability index in skew maps","Open region has intermingled riddled basins","Blowout bifurcations in a two-parameter map family"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000274,"raw_usage":{"total_tokens":1688,"prompt_tokens":1042,"completion_tokens":646,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":658,"completion_tokens_details":{"reasoning_tokens":581}},"tokens_in":658,"tokens_out":646,"duration_ms":6565,"temperature":1.0,"reasoning_tokens":581,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-08T21:11:02.115442+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Compute $\\ell_{a,b}(x)$ and $r_{a,b}(x)$ numerically for a dense grid of $x$ and for fine parameter grids inside $\\Gamma$; if for some set of positive $\\nu_{\\mathrm{ac}}$-measure the two functions differ, the separating graph $\\phi^*$ of Theorem 1 fails and the dimension formulas collapse. A cheaper check is to measure the stability index at points just above and below the numerically located boundary for a fixed parameter in $\\Gamma$ and compare the values with the constants predicted by Theorem 4.","supporting_citations":[],"review_version":1}