{"id":"7e7b62e7-cde0-4d0a-928d-556c1f467a95","arxiv_id":"2506.21481","paper_version":1,"verdict":"ACCEPT","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A single, simple search algorithm decides the escape problem for all robust instances, and is provably complete for the weakest representation of continuous functions.","lead":"This paper gives an algorithm that decides when a point escapes a region under repeated application of a continuous function, in a setting where the function is known only approximately. It proves the algorithm is optimal: it halts exactly on the cases whose answer is stable under small perturbations, which is the best possible.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified; the central proof is sound, with only a repairable presentation gap in Lemma 29's written argument.","rationale":"The reader's verdict of ACCEPT with moderate confidence is appropriate. My stress-test focused on the central completeness theorem, Theorem 15, and its key ingredient, Lemma 13. I checked the chain construction, the use of Lemma 29, the perturbation bounds, and the extension from compact sets to closed sets in Lemma 14. The mathematics is sound; the only genuine issue is a logical misstatement in the proof of Lemma 29, but the lemma is easily repaired and does not threaten the conclusions. The reader's weakest assumption, the representation's compact-open topology, is an explicit and well-motivated modeling choice rather than a hidden inconsistency; it limits the scope of the completeness claim but does not undermine its internal correctness. I therefore see no need to change the verdict, though a careful revision should fix the Lemma 29 proof and the minor typos.","tokens_in":30385,"tokens_out":43050,"duration_ms":439622,"concrete_test":"Independently prove Lemma 29 using the standard pigeonhole argument: let {x_1,...,x_N} be a delta/2-net of K. For each E_i choose a net point x_{j(i)} within delta/2, noting that x_{j(i)} cannot lie in any earlier delta/2-neighborhood. If two distinct E_i, E_j shared the same net point, points of these sets would be within delta, contradicting disjointness. Since infinitely many E_i and finitely many net points exist, this is impossible. Confirm that this repair validates Lemma 13 and hence Theorem 15.","verdict_should_be":"UNCHANGED","load_bearing_attack":"I find no load-bearing flaw in the central completeness proof. The key perturbation lemma (Lemma 13, Appendix B) is sound: the chain construction of E_i is finite via a standard pigeonhole argument, and the inductive perturbation f_1,...,f_N maps x0 out of K with sup-norm error < epsilon. Lemma 14's passage to closed sets, including the fixed-A case, is also correct: the compact approximations K_n and the modifications outside [-2^n,2^n]^d yield compact-open convergence. The completeness theorem then follows as stated. The main caveat, already acknowledged by the reader, is the explicit modeling choice of the weakest function representation (compact-open topology); this is not an internal inconsistency and the paper flags it in Section 4. The only issue I noticed is a garbled 'Hence' step in the written proof of Lemma 29 in Appendix B: from 'for each i there exists j' the text concludes 'for each j there exists i', which is not logically valid. However, the lemma itself is true and the intended pigeonhole proof is straightforward, so this is a presentation gap rather than a substantive flaw. Minor typos (Theorem 15 'I ⊆ A^o', 'f_{i,n}→f_n', Lemma 5's 'F_n' vs 'F_m') do not affect the argument.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the Point Escape Problem: given a continuous map f:R^d->R^d, a closed set A, and x0 in A, decide whether some iterate f^n(x0) leaves A. Working in the bit-model of computable analysis, the author proposes a partial decision method (Algorithm 1) that searches either for a certificate of escape or for a robust invariant set witnessing that the orbit is trapped. The central result, Theorem 15, states that this algorithm is complete in the maximal-partial-decidability sense: it halts exactly on the robust instances, where robustness is understood with respect to the compact-open topology on continuous functions (the topology induced by the weakest representation making evaluation computable). The paper also proves that the halting set is dense, analyzes the linear escape problem via a compactification reduction, and relates the quadratic family to the Mandelbrot set, giving a conditional alternative proof of Hertling's result subject to the density of hyperbolicity conjecture.","tokens_in":30582,"tokens_out":28502,"duration_ms":361667,"significance":"If the results hold, the paper gives a clean characterization of what is decidable about the escape problem for arbitrary continuous maps under a natural perturbation topology: the maximal partial decision method is exactly the search for robust invariants, and every non-robust instance is a genuine decision-boundary instance. This is a strong and surprising result, and the extension to affine linear systems and to the quadratic family shows that the general algorithm can compete with bespoke methods. The paper is also careful about the modeling assumptions: the compact-open topology is explicitly identified as the source of the notion of robustness, and the conditional nature of the Mandelbrot application is stated clearly. The main theorems are proved in detail in the appendices, and the central perturbation argument (Lemma 13) is substantial and, in substance, correct. The density result (Lemma 16 and Corollary 17) is an additional nice contribution.","major_comments":[{"comment":"The proof of Lemma 14 invokes Lemma 13 on compact sets K_n and C_n with only the assumption x0 in K_n (or x0 in C_n), but Lemma 13 requires x0 to lie in the interior of the compact set. When x0 lies on the boundary of A, no compact subset of A can contain x0 in its interior, so the invocation is unjustified as written. The statement is likely true and repairable—for the first part one may enlarge K_n with a small ball around x0 and take A_n to be that enlarged compact set, and for the fixed-A part one must argue via modifications supported outside [-2^n,2^n]^d—but the current proof does not supply this argument. Since Theorem 15 depends directly on Lemma 14, this gap is load-bearing and should be fixed.","section":"Lemma 14 (main text, Section 3.1)"},{"comment":"The written proof of Lemma 29 contains a logical inversion: from the claim that for each i there exists j with x_j in B(E_i, delta/2) and no earlier B(E_k, delta/2) containing x_j, the text concludes 'Hence, for all j there exists i' with x_j in B(E_i, delta/2). That inference is not valid. The lemma itself is true, and the intended argument is a pigeonhole principle: the sets B(E_i, delta/2) are pairwise disjoint, so an infinite sequence would require infinitely many distinct points x_j among finitely many candidates. The proof should be rewritten accordingly.","section":"Appendix B, Lemma 29"}],"minor_comments":[{"comment":"The statement says 'with I ⊆ A^o'; the symbol I should be V, the robust invariant introduced in the preceding clause.","section":"Theorem 15, item 3"},{"comment":"The phrase 'fi,n → fn for i = 0,1' should read 'fi,n → f for i = 0,1', and the sequences A0,n and A1,n are not defined; the text should specify that A0,n = A1,n = A_n in the general case and A0,n = A1,n = A in the fixed-A case.","section":"Theorem 15, item 4"},{"comment":"The justification of convergence fn → f is too terse: the displayed inequality only controls the sup-norm on C_n, which is not enough for the compact-open topology. The intended argument is that the modification is supported outside [-2^n,2^n]^d, so on every fixed B(0,2^m) the functions fn agree with gn for all n >= m and ||gn - f|| on that ball tends to zero. This should be stated explicitly.","section":"Lemma 14, second half"},{"comment":"The proof uses F_n in the statement but discusses F_m and F_n interchangeably; the indexing should be fixed to avoid confusion.","section":"Lemma 5"},{"comment":"In the non-hyperbolic case, the proof asserts that an unbounded orbit disjoint from K_c with 0 in K_c admits a uniform lower bound delta with |f_c^n(z_epsilon)| > delta for all n. This is true because an escaping orbit eventually leaves every compact set and has only finitely many early iterates, but the argument should be spelled out, since the current one-line justification is too compressed.","section":"Section 5, Theorem 22"},{"comment":"The proof refers to 'routine exercise' derivations of effective bounds in cases 1--3. Since these bounds are used to justify robustness, it would be helpful to state at least the form of the bounds or cite the corresponding lemmas in [25] more precisely.","section":"Appendix D, Lemma 19"}],"recommendation":"major_revision","confidential_remarks":"The central theorem is sound in substance and the paper is a good fit for a journal in computable analysis or theoretical computer science. The two main proof gaps—Lemma 14's application of Lemma 13 at boundary points and the garbled proof of Lemma 29—are repairable and do not undermine the main claims, but they do need to be fixed before publication. The Mandelbrot part is clearly labeled as conditional, and the reduction theorem is a nice contribution."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The paper proves something real: Algorithm 1 is a complete decision method for the Point Escape Problem under the weakest representation of continuous functions that makes evaluation computable. The key perturbation lemma (Lemma 13) is the load-bearing piece, and as far as I can tell it is correct. The appendix proof is detailed, and the stress-test pass agrees: the chain construction is finite by a pigeonhole argument, and the passage from compact to closed sets in Lemma 14 holds. This is not a case of the main theorem being oversold.\n\nWhat is actually new: the general completeness result for arbitrary continuous maps, and the characterization that non-halting exactly means the instance lies on the decision boundary, with A and x0 even fixed when A is not all of R^d. The linear-system classification overlaps with the author's prior work, and the Mandelbrot application is an alternative proof of Hertling's conditional result, but these are clearly presented as such. The denseness result for the halting set is a nice bonus.\n\nSoft spots, in proportion: the written proof of Lemma 29 contains a garbled step—it concludes \"for each j there exists i\" from \"for each i there exists j\"—which is not logically valid as written. The lemma itself is true and the intended pigeonhole argument is straightforward, so this is a presentation gap, not a flaw. A referee should ask for that paragraph to be rewritten. The effective bounds in Lemma 19 are left as \"routine exercise\"; that is acceptable for this audience but could be tightened. Theorem 22 is terse, though the argument goes through once you note the escaping orbit avoids the filled Julia set. Typos (e.g., 'I ⊆ A^o' in Theorem 15, 'F_n' vs 'F_m' in Lemma 5) are minor.\n\nThe one substantive modeling choice is the representation of functions: the compact-open topology makes perturbations near infinity arbitrarily small, so unbounded-set instances like f(x)=2x, A=[0,∞) are not robust. The paper flags this explicitly in Section 4, and the linear-system section shows how compactification can recover such cases. This is not an internal inconsistency; it is a clearly stated convention that defines what \"robust\" means.\n\nBottom line: the central argument holds up, the new result is real, and the paper is worth a serious referee. I would bring it to reading group and cite it. Send it to peer review with a request to fix the Lemma 29 presentation and the small typos.","headline":"A sound and genuinely new completeness theorem for the Point Escape Problem, with a careful but not flawless write-up; deserves peer review.","tokens_in":31123,"tokens_out":1142,"would_cite":true,"duration_ms":16053,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["03D78"],"pacs":[],"model":"deepseek-v4-flash","headline":"For continuous dynamical systems, the point-escape problem is maximally decidable: a cube-overapproximation search halts on exactly the instances whose answer is stable under small perturbations.","keywords":["dynamical systems","computability in analysis","escape problem","robust decidability","partial decision methods","invariant sets","Mandelbrot set","hyperbolicity conjecture"],"falsifier":"Find a trapped triple $(f, A, x_0)$ for which the algorithm never halts but, for some fixed $n$, every $g$ with $d(f, g) \\le 2^{-n}$ in the compact-open metric (1) also traps $x_0$; that would directly contradict the lemma that absence of a robust invariant yields an arbitrarily small perturbed map $g$ under which $x_0$ escapes.","tokens_in":30140,"feed_emoji":"🌀","tokens_out":11419,"duration_ms":121859,"temperature":0.7,"pith_summary":"The paper studies a general escape question for dynamical systems: given a closed set $A \\subseteq \\mathbb{R}^d$, a continuous map $f \\colon \\mathbb{R}^d \\to \\mathbb{R}^d$, and a point $x_0 \\in A$, decide whether some iterate of $f$ leaves $A$. Exact decision is impossible for continuous data, so the paper adopts the bit-model of real computation and redefines completeness: a decision method is complete when it halts on every instance whose answer is stable under all sufficiently small perturbations of the input. The paper gives one such method—a stage-by-stage search over dyadic cubes that either witnesses escape or finds a finite invariant set inside $A$—and proves it complete: no sound partial decision method can halt on more instances. The key dichotomy is that a trapped orbit is certifiable exactly when the point $f(x_0)$ lies in the interior of a robust invariant $V$ with $f(V) \\subseteq V^\\circ$; without such a set, arbitrarily small perturbations make the point escape. The same algorithm then yields complete methods for affine-linear systems (by compactifying the unbounded state space) and, conditional on the hyperbolicity conjecture, for the quadratic family that defines the Mandelbrot set.","feed_headline":"A cube search decides all stable escape questions","feed_subtitle":"It certifies trapped or escaping exactly when small perturbations can't change the answer, and no sound method can do more.","key_machinery":"The load-bearing object is the robust invariant: a compact set $V$ with $f(V) \\subseteq V^\\circ$, that is, $f$ maps $V$ into its own interior. The algorithm never searches for $V$ abstractly; instead, at each stage $n$ it works with dyadic cube partitions of $[-2^n,2^n]^d$, maintains cube overapproximations $Q_{n,i}$ of the orbit points and $O_{n,i}$ of the orbit segment, and checks two conditions: that some $Q_{n,i}$ is separated from $A$ (escape), or that $F_n(O_{n,i}) \\subseteq O_{n,i}$ and $O_{n,i} \\Subset A$ (trapped). The completeness proof shows that the cube-level inclusion $F_n(V_n) \\subseteq V_n$ forms a sound and complete certificate for the existence of a robust invariant: such a $V$ exists exactly when the finite-cube search eventually stabilises.","core_discovery":"The paper's central discovery is that the point-escape problem is maximally decidable at the level of robust instances: Algorithm 1 terminates on every escaping instance and on every trapped instance for which $f(x_0)$ is contained in the interior of a robust invariant $V \\subseteq A^\\circ$, and if it fails to terminate the instance sits on the decision boundary, with arbitrarily close perturbations answering both ways. Equivalently, its halting set contains the halting set of every sound partial decision method for the problem. The proof builds on a perturbation lemma: whenever $x_0$ is trapped but no robust invariant contains $f(x_0)$, the map $f$ can be modified by an arbitrarily small amount (in the topology of uniform convergence on compact sets) so that $x_0$ escapes; when $A \\neq \\mathbb{R}^d$, this can be done while keeping $A$ and $x_0$ fixed. From this, the halting behaviour of the algorithm is exactly the boundary structure of the problem.","pith_inferences":["If the completeness dichotomy is correct, then every sound trap-certification procedure that works from black-box evaluations of a continuous map is implicitly searching for a robust invariant; one could try to compile other certificate formats such as Lyapunov functions, barrier certificates, or reachability proofs into the cube-inclusion certificate the algorithm constructs.","The representation-dependence of the result is itself a predictive tool: adding one piece of information—a modulus of continuity, a derivative bound, or algebraic data—should enlarge the halting set exactly by the instances that become robust under the corresponding stronger topology, and this can be tested instance-by-instance.","The quadratic-family result suggests a template for higher-degree families: any family of maps for which hyperbolic parameters are dense would inherit a complete decision method from the same invariant-search algorithm, so proving such density would immediately yield new maximal decision procedures for Julia-type escape sets.","The compactification used for affine linear maps suggests that 'well-behaved at infinity' is the right hypothesis under which the general algorithm can be extended; polynomial maps or rational maps on projective space are natural candidates where the same conjugation idea might apply."],"forward_implications":["Every trapped orbit that is robust—stable under small perturbations of the function—is automatically certifiable by a finite cube computation; no additional mathematical insight about the map is needed once the invariant exists.","The halting set of the algorithm is dense in the space of all instances, and every boundary trapped instance has arbitrarily small perturbations that make it robustly trapped, so the undecidable cases are confined to decision boundaries rather than occupying open regions.","For any fixed closed set $A \\neq \\mathbb{R}^d$ and initial point $x_0$, the algorithm is already complete when only the map $f$ varies; no specialised method for that fixed $A$ and $x_0$ can decide more functions than the general algorithm.","For affine linear maps the robust trapped instances admit an explicit spectral classification, and a compactification reduction makes the unbounded linear escape problem decidable by the same algorithm.","For the quadratic family $z \\mapsto z^2 + c$, the origin-escape problem reduces to the point-escape problem in a way that preserves robustness exactly if the hyperbolicity conjecture holds; under that conjecture the algorithm decides the Mandelbrot set, giving a new proof of the conditional computability result.",""],"supporting_citations":[{"why":"Provides the standard representations of real numbers and continuous functions used to define the escape problem and its compact-open topology.","marker":"[43]"},{"why":"Defines complete and maximally partial decision methods and proves that completeness is equivalent to halting on all robust instances; the linear-recurrence robustness results whose classification the linear section mirrors.","marker":"[25]"},{"why":"Gives the conditional computability result for the Mandelbrot set whose proof the quadratic-family section recovers via the escape algorithm.","marker":"[17]"},{"why":"Shows the Mandelbrot set is undecidable in an alternative real-arithmetic model, the contrasting setting that motivates the computable-analysis formulation of decidability.","marker":"[8]"},{"why":"Raises the question whether the Mandelbrot set might be uncomputable, which the paper's escape-problem reduction addresses in conditional form.","marker":"[33]"},{"why":"Supplies the theorem that the origin is attracted to an attracting cycle, used to build a robust invariant at hyperbolic parameters.","marker":"[9]"},{"why":"Establishes that non-hyperbolic interior parameters lie in the boundary of their filled Julia set, used to show the algorithm cannot halt at such parameters.","marker":"[15]"},{"why":"Shows the escape problem is undecidable for exactly specified piecewise-linear systems, motivating robust decidability; also supplies the Jordan-normal-form escape fact used in the linear classification.","marker":"[41]"}],"fun_headline_variants":["Escape questions: robust cases fully decided","Maximal decidability for escape queries","Every sound escape method is topped","All stable escape decisions, and no more"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The completeness result holds only for the weakest representation of continuous functions that makes evaluation computable; if functions are given with extra information—say a modulus of continuity or derivative bounds—then more instances are robust and the algorithm is no longer claimed to be complete.","fun_headline_variants_meta":{"raw":{"variants":["Escape questions: robust cases fully decided","Maximal decidability for escape queries","Every sound escape method is topped","All stable escape decisions, and no more"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000848,"raw_usage":{"total_tokens":3695,"prompt_tokens":953,"completion_tokens":2742,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":569,"completion_tokens_details":{"reasoning_tokens":2690}},"tokens_in":569,"tokens_out":2742,"duration_ms":25160,"temperature":1.0,"reasoning_tokens":2690,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-06T22:26:25.542620+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Find a trapped triple $(f, A, x_0)$ for which the algorithm never halts but, for some fixed $n$, every $g$ with $d(f, g) \\le 2^{-n}$ in the compact-open metric (1) also traps $x_0$; that would directly contradict the lemma that absence of a robust invariant yields an arbitrarily small perturbed map $g$ under which $x_0$ escapes.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Gives the conditional computability result for the Mandelbrot set whose proof the quadratic-family section recovers via the escape algorithm."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Shows the Mandelbrot set is undecidable in an alternative real-arithmetic model, the contrasting setting that motivates the computable-analysis formulation of decidability."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Establishes that non-hyperbolic interior parameters lie in the boundary of their filled Julia set, used to show the algorithm cannot halt at such parameters."}],"review_version":1}