{"id":"a505d90d-3de6-4e9f-a097-adec6f20cab6","arxiv_id":"2607.28614","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"For tampered-memory elephant random walks, lim |D^c_n|/n = 1/2 is the sharp threshold separating persistence of diffusive/critical/superdiffusive phases from purely diffusive O(√n) behavior.","lead":"The paper shows that an elephant random walk keeps its anomalous diffusion phases only if more than half its memory is left untampered; below that threshold the walk is purely diffusive. This gives a sharp memory breakpoint for a long-open question in long-range dependent random walks.","discovery_kind":"extension","skeptic_critique":{"model":"grok-4.5","headline":"No significant objection identified","rationale":"The central proved claim (Theorem 1.10 + Corollary 1.11) is narrowly and correctly scoped to deterministic arithmetic partitions. The technical obstruction for general renewals is stated by the authors themselves (Remark 4), so it does not constitute an unacknowledged load-bearing flaw. The SA machinery, eigenvalue calculation, and variance formulae check out on inspection. The reader's CONDITIONAL verdict already reflects exactly this scope limitation; no adjustment is warranted.","tokens_in":32435,"tokens_out":437,"duration_ms":8212,"concrete_test":"Independently recompute the eigenvalues of -A in (4.37) and the critical value ρ*=1/2; confirm that ρ*=1/2 iff p=1/2+μ/(4k) and that this is equivalent to lim |D^c_n|/n=k/μ>1/2 precisely when a superdiffusive window exists. If the algebra matches Proposition 1.12 and Corollary 1.11, the proved claim is secure.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The reader's weakest-assumption note is accurate but already fully owned by the paper: Theorem 1.10 and Corollary 1.11 are proved only for arithmetic-progression partitions, and Remark 4 after Lemma 4.10 explicitly records that the error-rate conditions for the SA CLTs fail for general renewals. The abstract and Problems 1.14–1.15 label the general-renewal and random-D_n extensions as conjectures. The two-component SA argument (vector recursion (4.36), eigenvalues of -A in Lemma 4.4, Γ in Lemma 4.9, error rates in Lemma 4.10) is internally consistent for the deterministic arithmetic case that is claimed, and the LLN (Theorem 1.8) already holds under the full renewal assumptions (A1)–(A2). No hidden gap undermines the proved sharp-threshold statement.","agreement_with_reader":"agree"},"referee_report":{"model":"grok-4.5","summary":"The paper introduces the tampered-memory elephant random walk (TMERW), in which memory is partitioned into retained indices D_n^c (classical ERW dynamics) and tampered indices D_n (i.i.d. innovations). Using two-dimensional stochastic approximation on the restricted walks, the authors prove an SLLN under nested renewal structure with exponential moments (Theorem 1.8). For deterministic arithmetic-progression partitions they obtain a complete phase diagram (Theorem 1.10, Corollary 1.11, Proposition 1.12): the classical diffusive/critical/superdiffusive trichotomy persists precisely when lim |D_n^c|/n > 1/2, collapses to pure diffusion when the limit is < 1/2, and is fully characterised on the boundary = 1/2. When |D_n|/n → 0 they recover ERW-type fluctuations under mild growth restrictions (Theorems 1.6–1.7). General nested renewals and random partitions are left as conjectures (Problems 1.14–1.15).","tokens_in":32634,"tokens_out":1145,"duration_ms":38165,"significance":"The work directly addresses the open ‘memory breakpoint’ question posed by Gut–Stadtmüller. The emergence of the sharp threshold 1/2 is clean, non-circular, and arises naturally from the smallest eigenvalue of the mean-field matrix. The two-component SA analysis (vector recursion (4.36), eigenvalue Lemma 4.4, explicit Γ and variance formulae in Lemma 4.9 and Proposition 1.12, Appendices A–B) is carefully executed for the arithmetic case that is fully proved, and the authors correctly flag the remaining cases as conjectures. The model is novel and the LLN under general renewals already constitutes a solid contribution.","major_comments":[{"comment":"The abstract and the opening of §1.2 claim a sharp threshold for general non-random nested increasing partitions (‘if {D_n} is non-random \tau… with lim |D_n^c|/n > 1/2 then \tau…’). Theorem 1.10 and Corollary 1.11 are proved only for arithmetic-progression blocks (fixed \tau ≡ k, \tau ≡ \tau−k). Remark 4 after Lemma 4.10 explicitly records that the error-rate hypotheses needed for the SA CLTs fail for general renewals. The abstract must be aligned with the theorems; the general deterministic nested case should be stated as a conjecture (as already done for random D_n in Problem 1.14).","section":"Abstract; §1.2; Theorem 1.10; Remark 4"},{"comment":"In the non-trivial regime the fluctuation analysis relies on the deterministic block structure to obtain the precise rates in Lemma 4.10 (r_\tau n = O(1/n)). For the arithmetic case this is fine, but the paper should add a short remark clarifying that the same eigenvalue threshold \rho* = 1 - (2p-1)k/\tau already appears in the LLN mean-field matrix (Lemma 4.4) under general renewals, so the conjectured critical value p_c = 1/2 + \tau/(4E[\tau]) is at least consistent with the existing SLLN.","section":"Lemma 4.4; Lemma 4.10; Problem 1.14"}],"minor_comments":[{"comment":"Several typos and notational slips: ‘indenpendent’ (A1), ‘superdifuisive’ (Corollary 1.11), ‘critial’ (Problem 1.13), duplicated ‘{D_n}n≥1, and’ in the abstract, and inconsistent bold/roman for vectors.","section":"Throughout"},{"comment":"Figures 1–4 are mentioned in §1.3 but the captions and axis labels are not self-contained; a one-sentence description of the simulation parameters (p, \tau, number of paths) would help.","section":"§1.3; Appendix C"},{"comment":"The constant \tau in the arithmetic case is used both for the period and for the generic inter-arrival random variable; a distinct symbol for the period would improve readability.","section":"Theorem 1.10; §4.3"},{"comment":"Reference [29] is cited for the unresolved linear setting; a one-sentence comparison of the present D_n with the linear memory window M_n = {1,\tau…,m_n} would clarify the precise relationship.","section":"§1"}],"recommendation":"minor_revision","confidential_remarks":"The mathematical core (LLN + arithmetic phase diagram) is sound and the 1/2-threshold is attractive. The only substantive issue is the mismatch between the abstract’s claim and the theorems; once the abstract and introduction are tightened the paper is ready for a strong journal in probability. I see no reason to demand the general-renewal CLT before acceptance."},"author_rebuttal":null,"desk_editor":{"model":"grok-4.5","letter":"The one thing worth knowing: they give a sharp, explicit answer to the Gut–Stadtmüller memory-breakpoint question inside a natural competitive model. For deterministic nested partitions that grow by fixed blocks (arithmetic progression), if the retained-memory density stays strictly above 1/2 the classical three regimes survive (with a shifted critical p), while below 1/2 you get only ordinary √n diffusion for every p. The boundary case is also fully characterized. That is new and cleanly proved.\n\nWhat they do well. The model itself (tamper a subset of the memory with independent innovations) is a sensible way to probe how much long-range dependence is actually needed; it sits next to the still-open linear setting of Roy–Takei–Tanemura. The two-component stochastic-approximation setup (restricted walks on D and D^c, vector recursion, eigenvalue analysis of the mean-field matrix) is the right tool and is carried through carefully. LLN under renewal + exponential moments is solid (Thm 1.8). For the arithmetic case the CLTs, critical log scaling, and superdiffusive a.s. limit are complete, with explicit variances (Prop 1.12 and the appendices). No free parameters, no circularity; the 1/2 threshold emerges from the sign of the smallest eigenvalue, not by construction. Citations are appropriate and the open problems are honestly flagged.\n\nSoft spots, in proportion. The fluctuation theorems stop at arithmetic partitions. Remark 4 after Lemma 4.10 states plainly that the error-rate conditions needed for the SA CLTs fail for general renewals, so the general-renewal and random-D_n statements remain conjectures (Problems 1.14–1.15). That is a technical limitation, not a conceptual hole, and the paper owns it. The growth restrictions in the trivial-size CLT (Thm 1.7) look like artifacts of the method; the authors say so and the simulations support the conjecture that they can be removed. Nothing load-bearing is broken.\n\nThis is for people who already work on elephant walks, reinforced processes, or stochastic approximation. A serious referee in that circle will find it worth the time. I would accept it for peer review, expect the conjectures to stay conjectures, and cite the arithmetic threshold result.","headline":"Solid, carefully proved 1/2-memory threshold for ERW phase transition in a clean competitive model; fluctuations fully settled only for arithmetic partitions, with the rest cleanly labeled conjecture.","tokens_in":33275,"tokens_out":569,"would_cite":true,"duration_ms":11156,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60G50","60F05","60F15","60K35"],"pacs":[],"model":"grok-4.5","headline":"One-half of retained memory is the sharp breakpoint that keeps elephant random walks anomalously diffusive.","keywords":["elephant random walk","tampered memory","phase transition","anomalous diffusion","stochastic approximation","memory breakpoint","renewal structure","law of large numbers"],"falsifier":"Construct a nested renewal partition whose retained-memory density converges to a value strictly above (respectively below) one-half but whose inter-arrival times are not constant, and check whether the superdiffusive regime still appears exactly when the density exceeds one-half; any counter-example at a density other than one-half would refute the claimed breakpoint.","tokens_in":33299,"feed_emoji":"🐘","tokens_out":984,"duration_ms":23199,"temperature":0.7,"pith_summary":"Elephant random walks remember the past and can switch from ordinary diffusion to superdiffusion once a memory parameter crosses a critical value. This paper asks how much of that memory can be erased or replaced by fresh noise before the three-regime picture collapses. It introduces a tampered-memory walk that still samples the whole past, but on a growing set D_n replaces the remembered step by an independent coin flip, leaving the complementary set as classical elephant memory. For nested deterministic partitions the authors prove a clean threshold: if the retained-memory fraction stays strictly above one-half, the diffusive, critical and superdiffusive regimes all survive (with a shifted critical point); if it stays strictly below one-half, only ordinary square-root diffusion remains for every memory strength. The boundary case equal to one-half is fully classified. The same threshold is conjectured to govern random partitions. The result gives a concrete answer to how much memory is truly necessary for anomalous diffusion to persist.","feed_headline":"Half the memory keeps elephant walks superdiffusive","feed_subtitle":"Below one-half retained past, only ordinary square-root diffusion survives for every memory strength","key_machinery":"A two-dimensional stochastic-approximation recursion for the pair of restricted walks (the walk summed only over retained-memory indices and the walk summed only over innovation indices). The joint mean-field matrix and its eigenvalues determine the law of large numbers and the three fluctuation regimes.","core_discovery":"For non-random nested memory partitions with both the tampered set and its complement increasing, the phase transition of the tampered-memory elephant random walk into diffusive, critical and superdiffusive regimes persists precisely when the asymptotic density of retained memory exceeds one-half; when that density is less than one-half the walk is purely diffusive of order square-root n for every p in [0,1]; the equality case is completely characterised. Thus one-half is the sharp breakpoint for persistence of anomalous diffusion.","pith_inferences":["The competition between elephant memory and independent noise is essentially a density contest: the side that occupies more than half the indices asymptotically dictates the macroscopic scaling.","The unresolved linear partial-memory model of earlier work is likely governed by the same one-half rule once the remembered block length grows linearly with n.","In higher dimensions the analogous breakpoint should be the classical multi-dimensional critical value scaled by the retained-memory density.","Practical memory-limited implementations of long-range dependent walks can safely discard almost half their history without losing superdiffusive behaviour."],"forward_implications":["If more than half the past is retained as classical elephant memory, anomalous diffusion cannot be destroyed by independent noise on the complementary set.","If less than half is retained, the walk behaves like a simple random walk at every memory strength p, so superdiffusion is impossible.","The critical memory parameter itself shifts upward from the classical 3/4 to 1/2 + μ/(4k), widening the diffusive window.","The same one-half threshold is predicted to control random nested partitions and non-nested partitions whose density limit exists.","Recurrence versus transience when innovations are unbiased remains open and is expected to depend jointly on the density ratio and p."],"fun_headline_variants":["One-half retained memory is the sharp breakpoint for elephant-walk phases","Below half memory, tampered elephant walks stay purely diffusive","Retained density above 1/2 keeps diffusive-critical-superdiffusive transition","Memory fraction 1/2 decides if anomalous diffusion survives tampering","Tampered elephant walks: phase transition persists only above half memory"],"cache_read_input_tokens":16512,"weakest_assumption_plain":"When the tampered set is a positive fraction of the past, the sharp fluctuation theorems are proved only for perfectly regular arithmetic-progression blocks; the same threshold for general renewal or random partitions is left as a conjecture.","fun_headline_variants_meta":{"raw":{"variants":["One-half retained memory is the sharp breakpoint for elephant-walk phases","Below half memory, tampered elephant walks stay purely diffusive","Retained density above 1/2 keeps diffusive-critical-superdiffusive transition","Memory fraction 1/2 decides if anomalous diffusion survives tampering","Tampered elephant walks: phase transition persists only above half memory"]},"model":"grok-4.5","effort":"low","cost_usd":0.00421,"raw_usage":{"total_tokens":1406,"prompt_tokens":1002,"num_sources_used":0,"completion_tokens":79,"cost_in_usd_ticks":42104000,"prompt_tokens_details":{"text_tokens":1002,"audio_tokens":0,"image_tokens":0,"cached_tokens":128},"completion_tokens_details":{"audio_tokens":0,"reasoning_tokens":325,"accepted_prediction_tokens":0,"rejected_prediction_tokens":0}},"tokens_in":1002,"tokens_out":79,"duration_ms":6289,"temperature":1.0,"reasoning_tokens":325,"cache_read_input_tokens":128,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-07-31T02:05:57.049544+00:00","model_set":{"reader":"grok-4.5"},"falsifier":"Construct a nested renewal partition whose retained-memory density converges to a value strictly above (respectively below) one-half but whose inter-arrival times are not constant, and check whether the superdiffusive regime still appears exactly when the density exceeds one-half; any counter-example at a density other than one-half would refute the claimed breakpoint.","supporting_citations":[],"review_version":1}