{"id":"d54bd9d4-10da-4861-9323-48ecf8352297","arxiv_id":"2508.00461","paper_version":1,"verdict":"UNVERDICTED","confidence":"LOW","novelty_score":6.0,"correctness_risk":"unknown","formal_verification":"none","parameter_count":0,"one_line_summary":"For continuous maps on the Cantor set over a finite alphabet, the set of noise levels epsilon for which the perturbed map F_epsilon has a unique stationary measure is exactly any G_delta subset of [0,1] that contains 1; generically this set is (0,1].","lead":"This paper studies how random noise affects simple infinite-state systems called shift maps on the Cantor set. It proves that the collection of noise levels where the system has exactly one possible long-term behavior can be very diverse, but for most systems every positive noise level works.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The claimed flexibility implies non-monotonicity of unique-measure sets in epsilon; no argument in the abstract rules out the alternative that the set is always upward closed.","rationale":"The reader correctly notes that the theorem's form depends on the specific perturbation. My concern is sharper: the theorem's 'exactly' cannot be true if S(F) is always upward closed. This is a concrete mathematical property that can be checked independently of the full proof. For finite dimensions, upward closure is trivially true because every epsilon>0 gives an irreducible Markov chain. Therefore the non-monotone examples, if they exist, must be an infinite-dimensional effect. The naturalization to higher epsilon as more resets is encoded in the coordinate marginal identity p^(epsilon')=(1-delta)p^(epsilon)+delta u, which seems to make the chain only more mixing at larger epsilon. The abstract does not disclose how the construction evades this. If the full proof contains an explicit construction for D={1} with multiple invariant measures at epsilon close to 1, that would settle the concern. If not, the flexibility half is unsupported. The verdict remains UNVERDICTED because the concern is a check, not a demonstrated contradiction.","tokens_in":763,"tokens_out":26018,"duration_ms":280545,"concrete_test":"Obtain the full manuscript and focus on the construction realizing the extremal G_delta set D={1}. Verify that the constructed continuous map F has at least two distinct invariant measures for epsilon arbitrarily close to 1 (e.g., for epsilon=1-10^{-k}). Then verify that the proof does not accidentally produce uniqueness on an interval (epsilon*,1] for some epsilon*<1; if the construction fails below epsilon=1, the claimed classification is false. A complementary check: attempt a simple coupling proof that uniqueness at epsilon implies uniqueness at epsilon'>epsilon; if such a proof succeeds, the theorem's flexibility direction is impossible.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is that for every G_delta set D subset [0,1] with 1 in D there exists a continuous F with S(F):={epsilon: F_epsilon has a unique invariant measure} equal to D. This immediately entails that S(F) need not be upward closed: e.g., D={1} is allowed, so uniqueness at epsilon=1 does not force uniqueness for nearby epsilon<1. This is a strong non-monotonicity statement. A priori, the opposite is plausible: for any finite coordinate truncation of the state space, the chain is irreducible for every epsilon>0, hence has a unique invariant measure; and for epsilon'>epsilon the one-step coordinate marginals satisfy p^(epsilon')=(1-delta)p^(epsilon)+delta u with delta=(epsilon'-epsilon)/(1-epsilon), so higher epsilon is 'more resetting'. If uniqueness were monotone in epsilon, the only possible phase diagrams would be upward-closed G_delta sets, a proper subclass (e.g., not {1} alone). The abstract gives no indication of the mechanism that breaks monotonicity. The proof of the flexibility direction must construct, for each prescribed D, maps for which multiple invariant measures reappear at noise levels arbitrarily close to 1. This is the least secure part of the theorem, since finite approximations or simple coupling arguments would suggest no such reappearance. Without seeing that construction, the 'exactly' characterization is not credible.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"This paper studies a family of randomly perturbed continuous maps on the Cantor set. For a finite alphabet A and a continuous map F on A^N, the perturbation F_epsilon is obtained after each iteration by independently resetting each coordinate to a uniformly chosen value with probability epsilon. The central claim is a complete classification of the set S(F) = {epsilon in [0,1] : F_epsilon admits a unique invariant measure}: S(F) can be exactly any G_delta subset of [0,1] containing 1, and generically S(F) equals (0,1]. The present report is based solely on the abstract, as the full text was not provided.","tokens_in":979,"tokens_out":3048,"duration_ms":33777,"significance":"If correct, the classification would be a striking rigidity/flexibility result: the set of noise levels for which the perturbed system is uniquely ergodic is not arbitrary but exactly the class of G_delta sets containing 1, and yet in a topological sense almost every map exhibits uniqueness for every positive noise level. The 'exactly' clause is a strong universal statement that goes beyond a generic ergodic theorem, and the genericity result provides a sharp contrast. The claims are clearly stated and falsifiable, which is a strength. The main limitation is that the abstract contains no proof, so the significance cannot be fully assessed without the full text.","major_comments":[{"comment":"The claimed classification implies that S(F) need not be upward closed in epsilon; for instance, the G_delta set {1} is allowed, so uniqueness at epsilon=1 does not force uniqueness for nearby epsilon<1. The abstract offers no mechanism for this non-monotonicity. Since the perturbation for larger epsilon is a convex combination of the lower-epsilon kernel and a uniform reset, one might a priori expect uniqueness to persist as epsilon increases. The proof of the flexibility direction must exhibit, for each prescribed G_delta set D, maps for which multiple invariant measures reappear as epsilon approaches 1. Without at least a sketch of that construction, the 'exactly' characterization is not credible from the abstract alone.","section":"Abstract"},{"comment":"The genericity statement 'generically this set is ]0,1]' is not checkable without specifying the space of continuous maps and the topology used for 'generic'. Is the space C(A^N, A^N) with the uniform (C0) topology? Is 'generic' meant in the Baire category sense? The abstract does not define the residual set, nor whether the same topology is used in the flexibility direction. These details are load-bearing because the generic claim is part of the main dichotomy.","section":"Abstract"},{"comment":"The phrase 'has a unique measure' is ambiguous. For the perturbed system F_epsilon, which is a Markov chain on the Cantor set, does uniqueness refer to a unique stationary probability measure for the Markov transition kernel, or to a unique invariant measure for the associated skew-product (random dynamical system)? This distinction is central to the theorem's interpretation and should be made precise in the formal statement.","section":"Abstract"}],"minor_comments":[{"comment":"There is a typo in the first sentence: 'constitued' should be 'constituted'.","section":"Abstract"},{"comment":"The notation ']0,1]' is common in some traditions; for a broader audience, consider using '(0,1]' or define the interval notation explicitly.","section":"Abstract"}],"recommendation":"uncertain","confidential_remarks":"The review is limited to the abstract because the full text was not provided. The central theorem is too strong and surprising to certify from the abstract alone; the non-monotonicity of S(F) is a genuine concern that the full proof must address. I would need the full manuscript to judge correctness. The authors should be asked for the complete text before a decision is made."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Colleague, the paper claims a complete classification: for a random perturbation of a continuous map on the Cantor set where each coordinate is reset independently with probability epsilon to a uniform symbol, the set of eps at which F_epsilon has a unique invariant measure is exactly a G_delta subset of [0,1] containing 1, and generically it is (0,1]. If true, this answers a natural question with a clean topological description. The genericity half also seems plausible: generic F should be mixing enough that any positive noise gives uniqueness.\n\nWhat I cannot see from the abstract is the construction for the flexibility half. The claim implies, for instance, that there exists a continuous F whose unique-measure set is exactly {1} — uniqueness only at epsilon=1, with multiple invariant measures for every epsilon<1. That is a strikingly non-monotone conclusion. The naive intuition (and the coupling in your stress-test note) says more resetting should only improve mixing: the transition law at eps' is a mixture of the law at eps and the uniform law, which tends to collapse invariant measures. So the proof of the flexibility direction must construct maps where lowering the noise (but keeping it positive) creates new invariant measures. Nothing in the abstract hints at the mechanism. That is the load-bearing part, and it is exactly where I would want to see the argument before believing the 'exactly' in the theorem.\n\nI want to give credit: the model is clean, the statement is precise, and a full characterization of this flavor is a strong result. The genericity statement, presumably in the Baire-category sense, is a nice complement. The abstract gives no literature context, so I cannot tell whether the authors address the non-monotonicity issue head-on or whether they have a clever construction that breaks uniqueness at arbitrarily small positive eps.\n\nMy recommendation: send it to a knowledgeable referee. The claim is strong enough, and the red flag is sharp enough, that a referee can quickly determine whether the construction delivers. If it does, publish; if not, the error will be instructive. I would not cite it yet, and I would want the full text before bringing it to a reading group, but it deserves serious referee time.","headline":"Bold classification claim that is easy to state, hard to believe without the flexibility construction; worth a referee's time.","tokens_in":1518,"tokens_out":13079,"would_cite":false,"duration_ms":149321,"reading_group":"maybe","serious_thinker":"unclear","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["37B10","37H15"],"pacs":[],"model":"deepseek-v4-flash","headline":"The possible sets of noise levels at which a randomly perturbed continuous map on the Cantor set has a unique invariant measure are exactly the $G_\\delta$ subsets of $[0,1]$ that contain $1$, while generically the set is $(0,1]$.","keywords":["random perturbation","Cantor set","invariant measures","G-delta sets","generic dynamics","phase diagrams","symbolic dynamics"],"falsifier":"Find a continuous map $F$, for any finite alphabet, whose set $\\{\\epsilon : F_\\epsilon \\text{ has a unique invariant measure}\\}$ is not a $G_\\delta$ subset of $[0,1]$ or does not contain $1$; for example, a verified construction with $S(F)=\\mathbb{Q}\\cap[0,1]$ would refute the classification. No such example can exist if the theorem is correct.","tokens_in":529,"feed_emoji":"🎲","tokens_out":9760,"duration_ms":94701,"temperature":0.7,"pith_summary":"This paper studies continuous maps on the Cantor set, viewed as the space of infinite sequences over a finite alphabet, under random perturbation. The perturbed map $F_\\epsilon$ applies $F$ and then, independently for each coordinate, replaces that coordinate with a uniformly chosen symbol with probability $\\epsilon$. The main result is a complete classification: the set of $\\epsilon$ for which $F_\\epsilon$ has a unique invariant measure is exactly a $G_\\delta$ subset of $[0,1]$ that contains $1$, and every such set is realized by some $F$. A second, generic result says that for almost every continuous map in the sense of Baire category this uniqueness set is $(0,1]$, so all positive noise levels are unique. The paper thus shows that extreme flexibility and a simple generic picture coexist in the same family of dynamical systems.","feed_headline":"Noise levels with one invariant measure: exactly the G-delta sets","feed_subtitle":"Any G-delta set containing 1 occurs; typical maps are unique at all positive noise.","key_machinery":"The central object is the uniform reset perturbation: the transition rule that applies $F$ and then, with probability $\\epsilon$ per coordinate, replaces that coordinate by a uniform symbol. This defines a Markov chain on the Cantor set, and the paper identifies the descriptive-set-theoretic structure of the uniqueness set. $G_\\delta$ means a countable intersection of open sets; the classification shows that exactly this level of complexity can occur, and that the Baire category theorem is the right tool for the generic half of the result.","core_discovery":"Writing $S(F)=\\{\\epsilon\\in[0,1] : F_\\epsilon \\text{ has a unique invariant measure}\\}$, the authors prove two statements. First, $S(F)$ is always a $G_\\delta$ set containing $1$, and conversely, for every $G_\\delta$ subset $G$ of $[0,1]$ with $1\\in G$, there exists a continuous $F$ with $S(F)=G$. Second, in the space of continuous maps with the product topology, the set of $F$ for which $S(F)=(0,1]$ is residual, i.e. Baire generic. The necessity of containing $1$ comes from the fact that at $\\epsilon=1$ every coordinate is replaced by a uniform draw, so the noise alone produces a unique invariant measure independent of $F$.","pith_inferences":["The exact $G_\\delta$ classification is tied to the uniform i.i.d. reset mechanism; changing the noise distribution or allowing correlations among coordinates would likely change the admissible uniqueness sets, so the theorem should not be read as a universal law for all random perturbations.","The generic triviality suggests that exotic phase diagrams, though realizable, are hard to meet naturally; any random or numerical search for $F$ will almost surely land in the $(0,1]$ regime.","One could try to transfer the \"$G_\\delta$ plus containing $1$\" template to other perturbations of full shifts, for instance non-uniform resets or probabilistic cellular automata, by proving an analogous characterization; the paper leaves that extension open."],"forward_implications":["Any phase diagram satisfying the necessary condition, namely being $G_\\delta$ and containing $1$, can be engineered into some continuous map, so non-uniqueness can be arranged on arbitrarily complicated sets of noise levels.","For a generic continuous map, the phase diagram is trivial: every positive noise level gives a unique invariant measure, and only the unperturbed case $\\epsilon=0$ can have multiple measures.","Because $S(F)$ is always $G_\\delta$ and contains $1$, non-$G_\\delta$ candidates such as the rationals are provably impossible as uniqueness sets.","The two statements together give a complete answer to the flexibility-versus-genericity question for this family of maps and this noise model."],"supporting_citations":[],"fun_headline_variants":["Exact flexibility: any G-delta set of noise levels works","Generic noise gives unique measure for all positive epsilon","G-delta sets characterize possible noise ranges","Cantor maps: every G-delta set containing 1 is realizable","Residual set: unique measure for all positive epsilon"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The characterization depends on the specific perturbation rule (independent uniform resets with probability epsilon) and on the standard topology on continuous maps of the one-sided full shift over a finite alphabet; change the noise law, the dependence structure, or the space, and the exact G-delta statement need not survive.","fun_headline_variants_meta":{"raw":{"variants":["Exact flexibility: any G-delta set of noise levels works","Generic noise gives unique measure for all positive epsilon","G-delta sets characterize possible noise ranges","Cantor maps: every G-delta set containing 1 is realizable","Residual set: unique measure for all positive epsilon"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000563,"raw_usage":{"total_tokens":2623,"prompt_tokens":845,"completion_tokens":1778,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":461,"completion_tokens_details":{"reasoning_tokens":1697}},"tokens_in":461,"tokens_out":1778,"duration_ms":13184,"temperature":1.0,"reasoning_tokens":1697,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T10:07:12.676268+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Find a continuous map $F$, for any finite alphabet, whose set $\\{\\epsilon : F_\\epsilon \\text{ has a unique invariant measure}\\}$ is not a $G_\\delta$ subset of $[0,1]$ or does not contain $1$; for example, a verified construction with $S(F)=\\mathbb{Q}\\cap[0,1]$ would refute the classification. No such example can exist if the theorem is correct.","supporting_citations":[],"review_version":1}