{"id":"0148b1d6-93f4-42f1-beba-a91c616f1620","arxiv_id":"2412.06008","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":2,"one_line_summary":"If the symbolic local dimension is >1 and the perturbation has Fourier decay, the random self-similar measure has an almost-surely Hölder continuous density, implying the random fractal contains an interior point almost surely.","lead":"Randomly perturbed self-similar measures on the line are shown to be absolutely continuous with Hölder continuous density almost surely, whenever the measure's symbolic local dimension exceeds 1 and the perturbation has fast Fourier decay. In particular, the randomly perturbed fractal contains an interior point almost surely when its similarity dimension is greater than 1.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Change-of-variables estimate in §3 is not justified: for a realized splitting configuration U with p=4, the integral along w1=w2=w3=0, w4=t behaves as ∫ t^{2γ}(1+λt)^{-M} dt, requiring M>1+2γ, not merely M>1+γ.","rationale":"The paper proves an interesting and likely correct result: random perturbations with sufficiently fast Fourier decay turn measures of symbolic dimension s′>1 into measures with Hölder continuous density, giving an interior point when s>1. The proof uses reasonable tools—Carleson's theorem, Kolmogorov's continuity theorem, and the structure of random self-similar sums—and the overall strategy is plausible. The main weakness is the core quantitative estimate in Section 3. The reader already flagged the loose statement of the M-dependent integrability condition and the sketchy change-of-variables step. My stress-test sharpens this into a concrete failure of the displayed estimate: for a specific attainable triangular configuration, the p-fold integral needs M>1+2γ rather than M>1+γ. This is not a fatal objection to the theorem, because a smaller choice of γ can restore integrability, but it is a genuine gap in the proof as written. The counting argument in (3.7) also relies on the product form of the bound and would need to be redone with the corrected exponent; the final convergence for s′>1+γ may become s′>1+2γ, which is still satisfiable since s′>1 by choosing γ small, though the required p becomes larger. I therefore keep the verdict at CONDITIONAL: the central claim is not disproven, but the manuscript needs a repaired or expanded derivation of the key estimate before it can be accepted as fully verified.","tokens_in":11048,"tokens_out":18699,"duration_ms":185544,"concrete_test":"Fix p=4 and the splitting configuration corresponding to U = [[1,1,1,1],[0,1,1,1],[0,0,1,1],[0,0,0,1]]. Choose M=1.1 and γ=0.05, so M>1+γ but M<1+2γ. Compute the integral over ξ of ∏|ξ_k|^γ times the product of |Θhat| factors along the path w1=w2=w3=0, w4=t. Show that after integrating the transverse variables the integral is finite only if M>1+2γ, so it diverges logarithmically for these parameters. This verifies that the estimate (3.5)–(3.6) is false as stated and that the proof must impose γ < (M−1)/2 (or M>1+2γ) and redo the summation in (3.7).","verdict_should_be":"CONDITIONAL","load_bearing_attack":"The load-bearing step is the estimate leading to (3.5)–(3.6). The paper claims that after the change of variables to the coordinate system w_k = Σ_{ℓ∈N(ω(k),N_k)} ξ_ℓ, the integral ∫∏|ξ_k|^γ |W| dξ is bounded by C′ ∏ λ_{ω(k),N_k}^{-(1+γ)} using only M=s′>1+γ. This is not what the displayed integrand gives. The inverse of the triangular matrix U has off-diagonal entries, so |ξ_k| is not controlled by |w_k| alone. For a possible splitting configuration with p=4 and U = [[1,1,1,1],[0,1,1,1],[0,0,1,1],[0,0,0,1]], take w1=w2=w3=0 and w4=t. Then ξ4=t, ξ3=-t, ξ2=0, ξ1=0, and |W| ≈ (1+λ4|t|)^{-M}. After integrating the transverse variables w1,w2,w3 (a finite integral for M>1), the remaining t-integral is C∫ |t|^{2γ}(1+λ4|t|)^{-M} dt, which converges only if M>1+2γ, not if M>1+γ. Since the paper later chooses only γ<s′−1, it can choose γ with 1+γ<M<1+2γ, making the claimed estimate false. The argument may be repairable by choosing γ < (s′−1)/2 and p large enough that pγ>1, but that is not what the text proves, and the counting sum in (3.7) would need to be rechecked under that stronger condition.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies randomly perturbed self-similar iterated function systems on the line. The main result (Theorem 1.1) asserts that if the similarity dimension satisfies s > 1 and a symbolic measure satisfies the cylinder bound μ([ω]) ≤ K λ_ω^{s'} with s' > 1, then for a random perturbation whose Fourier transform decays like (1+|x|)^{-M} with M ≥ s', the projected measure is almost surely absolutely continuous with Hölder continuous density. Corollary 1.2 concludes that the randomly perturbed attractor contains an interior point almost surely when M ≥ s. The proof combines Carleson's theorem with Kolmogorov's continuity theorem: it estimates p-th moments of the density process, using a change of variables and a counting argument over the levels at which the p symbolic sequences first separate. The paper also contains a measurability lemma and relies on earlier results of Jordan–Pollicott–Simon and Gu–Miao for the almost sure absolute continuity with L² density.","tokens_in":11374,"tokens_out":47083,"duration_ms":443876,"significance":"If the main theorem is correct, it would extend recent work of Dekking–Simon–Székely–Szekeres and Gu–Miao by producing Hölder regularity of the density rather than only absolute continuity or dimension information, and it would give a new proof of the interior-point corollary. The method is interesting: the use of Carleson's theorem restricts the argument to dimension one, and the authors state this limitation explicitly. The proof is not circular and relies on established external results. The authors also explicitly acknowledge that the most natural uniform-on-an-interval perturbation does not satisfy the Fourier decay condition (1.3), which is an honest limitation. However, the central change-of-variables estimate in Section 3 is not justified as written and needs correction; with a local repair the theorem may stand, but the present proof is incomplete.","major_comments":[{"comment":"The displayed estimate that leads to (3.6) is false under the stated assumption M = s' > 1 + γ. After the change of variables w_k = λ_{ω(k),N_k} Σ_{ℓ∈N(ω(k),N_k)} ξ_ℓ, one has ξ = U^{-1} D^{-1} w, so the factor ∏ |ξ_k|^γ introduces powers λ_{ω(k),N_k}^{-γ} that are not accounted for in the text. For a configuration with p = 2, N_1 = 0, λ_{ω(1),N_1} = 1, and λ_{ω(2),N_2} = λ, the line w_1 = 0, w_2 = t gives ξ_1 = -t/λ and ξ_2 = t/λ. The transverse integral over w_1 contributes λ^{-1} from the Jacobian and a bounded constant, leaving an integral of order λ^{-1-2γ} ∫ |t|^{2γ} (1+|t|)^{-M} dt. This converges only if M > 1 + 2γ, not merely M > 1 + γ. Since the text later chooses γ < s' - 1, there are admissible parameters (e.g., s' = 1.4, γ = 0.3, M = 1.4) for which the asserted bound fails. The estimate must be reworked: using the correct exponent 1 + 2γ (or else the full product W rather than the crude bound by the selected N_k factors) and then choosing γ < (s' - 1)/2 makes the counting estimate (3.7) converge, but this is not what the manuscript proves.","section":"§3, estimate preceding (3.6)"},{"comment":"The final choice of constants ('choose γ < s' - 1, then choose some even p large enough that pγ > 1') is tied to the erroneous exponent. With the corrected estimate, one must require M > 1 + 2γ, and since the theorem only assumes M ≥ s', the admissible choice is γ < (s' - 1)/2. This is compatible with the later requirement pγ > 1 by taking p large, but the current text does not state this condition. In addition, the sentence 'we also assume M = s' > 1 + γ' should be replaced by a precise statement such as 'M ≥ s' > 1 + 2γ' after fixing γ, and the summation in (3.7) should be reread with the exponent s' - 1 - 2γ > 0.","section":"§3, final choice of constants"}],"minor_comments":[{"comment":"Equation (3.7) has a notational mismatch: the left-hand side is written for ∏_{k=1}^p λ_{ω(k),N_k}, while the right-hand side uses ℓ and K^{ℓ-1}; please define ℓ = p and use the same symbol throughout.","section":"§3, equation (3.7)"},{"comment":"The sets N_{ω(1),...,ω(p)}(τ^n) and N(ω(k),N_k) are used before being defined explicitly; a short formal definition would prevent confusion, especially because the change-of-variables argument depends on their exact meaning.","section":"§3, notation"},{"comment":"There are numerous typographical issues (e.g., 't he' in the title, 'Micha/suppress l' in the author line, 'e very' and 'con tains' in the abstract, 'te function' in the proof of Theorem 1.1, and a missing closing parenthesis in the proof of Lemma 2.4). These should be corrected in a revised version.","section":"Throughout"},{"comment":"The theorem assumes M ≥ s', but the proof says 'M = s''. Since larger M gives stronger decay, one can take the worst case, but the text should say so explicitly to avoid the impression that the theorem's hypothesis is changed.","section":"Theorem 1.1 and §3"}],"recommendation":"major_revision","confidential_remarks":"The main theorem is plausible and the gap appears repairable by a local modification of the constant choice in Section 3. I would not recommend rejection. The authors should be asked to redo the change-of-variables estimate with explicit control of the powers of λ and to state the integrability condition on M precisely, then to recheck the counting sum (3.7) under the corrected exponent."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the paper has a real new theorem—Hölder continuous density for projected random self-similar measures under the Fourier decay condition (1.3)—but the main estimate in Section 3 is not justified as written. The stress-test note is right, and it lands on a load-bearing step.\n\nWhat is genuinely good: Theorem 1.1 is new and not in the cited papers. Carleson plus Kolmogorov is a sensible route, the measurability lemma is handled carefully, and the authors are honest that (1.3) excludes the uniform perturbation studied in [5]. No circularity, no data fitting. Corollary 1.2 is a modest repackaging of the interior-point result from [5], but as a corollary it is fine.\n\nWhere the proof breaks: the change-of-variables step around (3.5)–(3.6) assumes |ξ_k| can be treated like |w_k| after passing to w_k = Σ_{ℓ∈N_k} ξ_ℓ. The triangular matrix has an off-diagonal inverse. For a realized nested splitting with p=4, N1={1,2,3,4}, N2={2,3,4}, N3={3,4}, N4={4}, the slice w1=w2=w3=0, w4=t gives ξ4=t, ξ3=-t, and the integrand decays like |t|^{2γ}(1+λ4|t|)^{-M}. That integral needs M>1+2γ, not M>1+γ. The text explicitly chooses γ<s′−1 and M=s′, so it can pick γ in ((s′−1)/2, s′−1), where the stated estimate fails.\n\nThis is repairable: choose γ < (s′−1)/2 and then p large enough that pγ>1. The counting estimate (3.7) would need to be rechecked under that condition, but it looks like it can absorb it. As written, the proof overclaims its constant regime.\n\nBottom line: this deserves a serious referee, but not acceptance yet. The main idea is likely right, the gap is localized, and the authors should be asked to fix Section 3 and state the M,γ condition precisely.","headline":"A real new Hölder-density theorem for random self-similar measures, but the central estimate in Section 3 has a gap that needs fixing before the proof is accepted.","tokens_in":11999,"tokens_out":6755,"would_cite":false,"duration_ms":65151,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["28A80","60G30","60G57"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper proves that, under a Fourier-decay condition on a random perturbation, any symbolic measure with local dimension greater than 1 pushes forward to an absolutely continuous measure with Hölder-continuous density almost surely…","keywords":["random self-similar sets","iterated function systems","absolute continuity","Hölder continuous density","interior point","Fourier transform","local dimension","random perturbation"],"falsifier":"A concrete check is to take the IFS with maps $f_1(x)=x/2$, $f_2(x)=x/2+1/3$, $f_3(x)=x/2+2/3$ (similarity dimension $\\log 3/\\log 2>1$), the natural Bernoulli measure, and a compactly supported smooth perturbation with fast Fourier decay, then compute the random density by Fourier inversion at many nearby pairs of points over many realizations. The theorem predicts a Hölder-continuous density almost surely; any realization whose computed density has unbounded oscillations or fails to satisfy a power-law modulus at fine scales would contradict the central claim.","tokens_in":10762,"feed_emoji":"📐","tokens_out":16789,"duration_ms":160437,"temperature":0.7,"pith_summary":"Self-similar iterated function systems on the line have a similarity dimension $s$; when $s>1$, the paper asks whether a random perturbation of the system smooths the projected measure. The answer is yes under two conditions: the symbolic measure must satisfy the cylinder bound $\\mu([\\omega])\\le K\\lambda_\\omega^{s'}$ with $s'>1$, and the perturbation's distribution must have Fourier decay $|\\hat\\Theta(x)|\\le C(1+|x|)^{-M}$ with $M\\ge s'$. Then $\\nu^\\Theta=(\\Pi^\\Theta)_*\\mu$ is almost surely absolutely continuous with respect to Lebesgue measure and has a Hölder-continuous density. Because a continuous density that is positive somewhere forces an open interval in the support, the same theorem gives an interior point of the random self-similar set whenever $s>1$ and $M\\ge s$. This extends earlier interior-point results that were confined to uniform perturbations and yields regularity of the density, not only its existence.","feed_headline":"Fractal measures get Hölder-continuous densities almost surely","feed_subtitle":"When similarity dimension exceeds one, suitable random noise forces a continuous density and an interior point.","key_machinery":"The argument runs through the random perturbation applied independently at every finite word of the symbolic tree, $\\Pi^\\Theta(\\omega)=\\sum_{j\\ge0}\\lambda_{\\omega_j}(t_{\\omega_j}+\\Theta_{\\omega_j})$, and through the Fourier-domain quantity $W(\\xi_1,\\dots,\\xi_p)=|\\int\\prod_{k=1}^p\\widehat{\\nu^\\Theta}(\\xi_k)\\,d\\mathbb{P}(\\Theta)|$. Because the perturbations at different nodes are independent, this quantity factorizes over the tree, and the Fourier decay assumption bounds each factor by $(1+|\\lambda\\xi|)^{-M}$. A combinatorial estimate over the first levels at which the $p$ sampled symbolic sequences separate shows that $\\int\\prod_{k=1}^p|\\xi_k|^\\gamma W(\\xi_1,\\dots,\\xi_p)\\,d\\xi_1\\cdots d\\xi_p$ is finite whenever $\\gamma<s'-1$ and $p$ is chosen large and even. That integrability is what converts the Fourier-decay condition into the polynomial moment estimate $\\mathbb{E}|\\vartheta(a)-\\vartheta(b)|^p\\le C|a-b|^{p\\gamma}$, the input for the Hölder-continuity criterion that delivers the continuous density.","core_discovery":"The central claim is Theorem 1.1: for any self-similar IFS with similarity dimension $s>1$, any measure $\\mu$ on the symbolic space with $\\mu([\\omega])\\le K\\lambda_\\omega^{s'}$ for some $s'>1$, and any perturbation $\\Theta$ satisfying $|\\hat\\Theta(x)|\\le C(1+|x|)^{-M}$ with $M\\ge s'$, the projected measure $\\nu^\\Theta=(\\Pi^\\Theta)_*\\mu$ is absolutely continuous and its density is Hölder continuous almost surely. The proof first obtains almost-sure absolute continuity with an $L^2$ density, then uses one-dimensional Fourier inversion to represent the density pointwise as a limit of truncated inverse Fourier integrals, and finally views the density as a stochastic process in the spatial variable $x$. A moment bound of the form $\\mathbb{E}|\\vartheta(a)-\\vartheta(b)|^p\\le C|a-b|^{p\\gamma}$ with $p\\gamma>1$ feeds a Hölder-continuity criterion from stochastic process theory, which produces a continuous version of the density. Corollary 1.2 applies the theorem to the natural Bernoulli measure with weights $\\lambda_i^s$; since this measure satisfies the assumption with $s'=s$, the random attractor $\\Lambda^\\Theta$ contains an interior point almost surely.","pith_inferences":["One natural extension is to test whether the same Hölder conclusion holds for the uniform perturbation that the theorem excludes; a numerical experiment with a smooth approximation to the uniform distribution would show whether the Fourier decay condition is truly necessary or merely a technical convenience.","The one-dimensional Fourier-inversion step appears to be essential; in higher dimensions the analogous almost-everywhere inversion is false, so extending the result to random self-affine sets on $\\mathbb{R}^d$ would likely require a different mechanism and may fail.","Since the theorem works for any measure satisfying (1.4), the interior-point corollary should also hold for any symbolic measure with full support and local dimension above 1, not just the natural Bernoulli measure.","The proof's dependence on $\\gamma<s'-1$ suggests the Hölder exponent one can extract is controlled by the excess of the symbolic dimension over 1, so measures with larger $s'$ should yield smoother densities; this quantitative dependence is not stated as a theorem and would be worth checking."],"forward_implications":["Corollary 1.2: whenever the similarity dimension $s>1$ and the Fourier decay exponent satisfies $M\\ge s$, the random self-similar set $\\Lambda^\\Theta$ contains an open interval almost surely.","The density of $\\nu^\\Theta$ is not merely absolutely continuous but Hölder continuous, so it has a continuous representative defined at every point of the line.","The result applies to every symbolic measure satisfying the cylinder bound (1.4), not only to the natural Bernoulli measure, as long as its symbolic local dimension is strictly above 1.","If the symbolic local dimension drops to 1 or below, no bounded density can exist; the paper shows the condition $s'>1$ is close to optimal for a bounded density, not merely sufficient.","The proof also confirms, through the Fourier representation, that the $L^2$-based absolute continuity of the projected measure is upgraded to a pointwise continuous object almost surely."],"supporting_citations":[{"why":"Supplies Proposition 2.3, the almost-sure absolute continuity of $\\nu^\\Theta$ with an $L^2$ density for measures satisfying (1.4).","marker":"[12]"},{"why":"Provides the one-dimensional almost-everywhere Fourier inversion result used to represent the density pointwise.","marker":"[8]"},{"why":"Gives the Hölder-continuity criterion for stochastic processes used to produce the continuous version of the density.","marker":"[15]"},{"why":"Supplies the differentiation theorem identifying the density as the limit of averaged ball measures, used in the definition of $\\vartheta^\\Theta$.","marker":"[17]"},{"why":"Establishes the interior-point result for uniform perturbations, the result that the present corollary extends to a broader class of perturbations.","marker":"[5]"},{"why":"Relates the cylinder condition (1.4) to lower $L^q$ dimensions and gives the necessary dimension context used in the introduction.","marker":"[9]"},{"why":"Provides the stochastic-process viewpoint of local density as a process in the spatial variable, which the proof adapts.","marker":"[6]"}],"fun_headline_variants":["Random self-similar sets with dimension >1 get interior points a.s.","Hölder-smooth densities almost surely from random self-similar measures","When similarity dimension exceeds 1, random fractals host interior points a.s.","Hölder densities and interior points: almost sure for random self-similar measures","Random self-similar sets: interior points a.s. when dimension >1"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof rests on the Fourier decay assumption $|\\hat\\Theta(x)|\\le C(1+|x|)^{-M}$ with $M\\ge s'$; the most natural perturbation, uniform on an interval, does not satisfy it, so the theorem leaves that case open.","fun_headline_variants_meta":{"raw":{"variants":["Random self-similar sets with dimension >1 get interior points a.s.","Hölder-smooth densities almost surely from random self-similar measures","When similarity dimension exceeds 1, random fractals host interior points a.s.","Hölder densities and interior points: almost sure for random self-similar measures","Random self-similar sets: interior points a.s. when dimension >1"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000966,"raw_usage":{"total_tokens":4081,"prompt_tokens":887,"completion_tokens":3194,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":503,"completion_tokens_details":{"reasoning_tokens":3104}},"tokens_in":503,"tokens_out":3194,"duration_ms":22851,"temperature":1.0,"reasoning_tokens":3104,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T20:08:26.971766+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A concrete check is to take the IFS with maps $f_1(x)=x/2$, $f_2(x)=x/2+1/3$, $f_3(x)=x/2+2/3$ (similarity dimension $\\log 3/\\log 2>1$), the natural Bernoulli measure, and a compactly supported smooth perturbation with fast Fourier decay, then compute the random density by Fourier inversion at many nearby pairs of points over many realizations. The theorem predicts a Hölder-continuous density almost surely; any realization whose computed density has unbounded oscillations or fails to satisfy a power-law modulus at fine scales would contradict the central claim.","supporting_citations":[{"cited_title":"Jordan, M","cited_arxiv_id":null,"evidence_quote":"Supplies Proposition 2.3, the almost-sure absolute continuity of $\\nu^\\Theta$ with an $L^2$ density for measures satisfying (1.4)."},{"cited_title":"Grafakos","cited_arxiv_id":null,"evidence_quote":"Provides the one-dimensional almost-everywhere Fourier inversion result used to represent the density pointwise."},{"cited_title":"Khoshnevisan","cited_arxiv_id":null,"evidence_quote":"Gives the Hölder-continuity criterion for stochastic processes used to produce the continuous version of the density."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the differentiation theorem identifying the density as the limit of averaged ball measures, used in the definition of $\\vartheta^\\Theta$."},{"cited_title":"Dekking, K","cited_arxiv_id":null,"evidence_quote":"Establishes the interior-point result for uniform perturbations, the result that the present corollary extends to a broader class of perturbations."},{"cited_title":"Generalized $q$-dimensions of measures on nonautonomous fractals","cited_arxiv_id":"2411.17298","evidence_quote":"Relates the cylinder condition (1.4) to lower $L^q$ dimensions and gives the necessary dimension context used in the introduction."},{"cited_title":"Erraoui and Y","cited_arxiv_id":null,"evidence_quote":"Provides the stochastic-process viewpoint of local density as a process in the spatial variable, which the proof adapts."}],"review_version":1}