{"id":"88b72274-9dc7-4c80-bb9f-db00239f47d9","arxiv_id":"2412.10686","paper_version":3,"verdict":"REJECT","confidence":"MODERATE","novelty_score":4.0,"correctness_risk":"high","formal_verification":"none","parameter_count":4,"one_line_summary":"The paper reformulates Bellman's lost-in-a-forest problem as a traveling-salesman-style optimization over rotated and translated forest boundaries, but the promised general solution lacks a rigorous convergence proof and yields only known results.","lead":"This paper rewrites Bellman's lost-in-a-forest problem as a large optimization problem by rotating and shifting the forest boundary around the hiker. It claims a general solution, but the proof is a sketch and the examples repeat results from earlier work.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 3's Gamma-convergence is asserted, not proved, and the paper's own unit-strip run (2.24853 vs. known optimum 2.278292) shows the discretized problem can undercut the true optimum; the 'general solution' claim is therefore unsupported.","rationale":"The reader's weakest_assumption identifies the missing liminf inequality and recovery sequence in the proof of Theorem 3, which is precisely the load-bearing gap. I add a concrete numerical red flag: the paper's own unit-strip computation (2.24853 in Appendix III) is below the known continuous optimum (2.278292, cited in Section 4.2.10), demonstrating that the sampled constraints do not enforce escape for all continuous states and that the limit is not established by the paper's evidence. The manuscript does present a coherent discrete TSP-type reformulation and reproduces several known shapes, which gives the framework some interest, but those are solutions to restricted order-fixed problems, not to the general claim. The abstract promises a 'general solution' to Bellman's problem; the delivered content is a discretization scheme plus a theorem whose proof is a sketch, with no certified convergence and with numerical evidence that the discrete objects are not feasible escape paths. Rejecting the paper as a proof of the central claim is therefore appropriate; the interesting reformulation could be salvaged with a rigorous Gamma-convergence argument and correct numerical experiments, but that is not present.","tokens_in":58248,"tokens_out":11213,"duration_ms":111401,"concrete_test":"Solve the discrete Weak Form II for the unit strip to certified global optimality for increasing refinement, e.g., N=8,12,16,24 with M=2N, using exact TSP/order optimization (dynamic programming or branch-and-bound for these sizes), and compare the optimal discrete lengths against Zalgaller's known continuous optimum 2.278292. If the certified discrete optima do not increase monotonically toward 2.278292 from below---or if they converge to a strictly smaller value---then Theorem 3 is false and the central claim fails.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central claim is Theorem 3 (Section 6.1): as M and N grow, the solution to Weak Form II converges to the solution of the original Bellman problem. The proof is a sketch: it asserts the liminf inequality (60) from 'lower semicontinuity' and the recovery sequence (61) from continuity, but never constructs either. The discrete functionals live on spaces of different dimensions (MN continuous variables plus a permutation), so standard Gamma-convergence requires a common metric space, equi-coercivity, and control of the order variable; none is established. The failure of the finite approximations to enforce the continuum constraints is visible in the paper's own numerics: Appendix III reports length 2.24853 for the unit strip with N=12, M=26, while Section 4.2.10 cites Zalgaller's optimal 2.278292. A path shorter than the known optimum cannot be a valid escape strategy for all starting points and orientations, so the finite discrete solution is not a feasible continuous escape path; whether its limit is the true optimum is exactly what Theorem 3 must prove, and it is not proved. Moreover, nearly all examples fix the visiting order (e.g., Section 4.2.1: 'If assuming the order of points ranges from 0 to N-1'), so they solve a restricted problem, not the Weak Form I/II global optimum asserted in the theorem. Section 10 concedes the exact problems remain intractable, and Appendix III itself notes NMinimize may not return a global optimum, further weakening the numerical support.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a discretization framework for Bellman's lost-in-a-forest problem. The original minimax problem over starting points and orientations is converted into a constrained shortest-path problem through 'escape points' on rotated and translated copies of the forest boundary. Upon discretization this becomes a Hamiltonian-path/TSP-type optimization over the locations and visiting order of the escape points. The paper claims in Theorem 3 that as the numbers of orientation samples N and starting-point samples M tend to infinity, the discrete optima converge to the solution of the original continuous problem. The bulk of the paper consists of numerical examples for lines, circles, strips, triangles, sectors, opaque sets, and related problems, with Mathematica notebooks in the appendices. The examples reproduce several known optimal or near-optimal shapes and values.","tokens_in":58596,"tokens_out":2444,"duration_ms":26689,"significance":"If the claimed general solution were established, this would be a substantial contribution: it would reduce a long-standing minimax problem over paths with continuous uncertainty to a sequence of discrete TSP-type optimizations and would provide a computational route to Moser's worm problem and opaque-set variants. The paper also has a genuine methodological idea—the translation/rotation reduction to escape points—and the supplied notebooks make the discrete computations reproducible. However, the central convergence claim is not proved, and the paper's own numerics indicate that the discrete problem can produce paths shorter than the known continuum optimum, so the paper does not currently deliver a verified general solution. Its value at this stage is that of a heuristic framework with several consistency checks against known results.","major_comments":[{"comment":"The proof of Theorem 3 is only a sketch and is load-bearing for the paper's main claim. The liminf inequality (60) and the recovery sequence (61) are asserted from 'lower semicontinuity' and 'continuity' without being constructed, and the discrete functionals live on spaces of dimension 3MN plus a permutation variable, so standard Gamma-convergence requires a common topological space, equi-coercivity, and control of the order variable; none of these is established. As stated, the theorem does not follow from the cited Weierstrass and uniform-continuity arguments.","section":"Section 6.1, Theorem 3, Eqs. (60)-(61)"},{"comment":"The reported numerical result for the unit strip, length 2.24853 for N=12, M=26, is smaller than Zalgaller's known continuum optimum 2.278292 quoted in Section 4.2.10. Since the continuum problem is a minimax problem over all starting points and orientations, a value below the known optimum cannot correspond to a feasible continuous escape strategy; it is an artifact of the finite discretization. This directly contradicts the convergence asserted in Theorem 3 and shows that the discrete minimizer need not even be a valid approximate escape path for the original problem.","section":"Appendix III and Section 6.2.1"},{"comment":"Nearly all examples fix the visiting order of the escape points rather than optimizing it. For example, Section 4.2.1 states 'If assuming the order of points ranges from 0 to N-1', and Appendix III hard-codes a permutation BB in the unit-strip computation. The formulations in Definitions 3.15 and 3.22 explicitly include the permutation as an optimization variable, so these examples solve a restricted subproblem and cannot be used as evidence for the claimed global solution of Weak Form I or II.","section":"Sections 4.2.1, 4.2.10, and 6.2.1"},{"comment":"The continuous calculus-of-variations derivation for the half-plane example is not a derivation of the optimal path: the Euler-Lagrange equation (29) is left unsolved, and the boundary conditions are imposed by hand ('Approximate boundary value' y[1]=0.57735 and the condition y[nL]+1=0) to match the known line-search solution. The resulting length 6.39724 reproduces the known value only because those boundary values are fitted, not because the variational problem is solved from first principles.","section":"Appendix I"}],"minor_comments":[{"comment":"The heading reads 'Length of escape path for Weak Form I' but the definition is for Weak Form II; this is confusing and should be corrected.","section":"Definition 3.21"},{"comment":"The sentence 'Them are nontrivial and consistent with previous papers' contains a grammatical error and should be rewritten.","section":"Section 4.2, introductory paragraph"},{"comment":"Several references are Wikipedia pages ([2], [31], [46]) rather than archival sources; for a mathematical paper these should be replaced by standard bibliographic entries.","section":"References"},{"comment":"The constraint in the Mathematica code uses angle step 2π/(nn-1) while the main text Definition 3.10 uses 2π/N; the discrepancy between N and N-1 should be explained or reconciled.","section":"Appendix III"}],"recommendation":"reject","confidential_remarks":"The paper's central theorem is unsupported and the numerical evidence contains a direct counterexample to the convergence claim, so I cannot recommend publication in the current form. The discrete reformulation may still be a useful heuristic, and a revised paper that either proves a rigorous upper/lower-bound statement with a feasible recovery construction or explicitly reframes the contribution as a numerical method with verified error bounds could be worth considering."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Bottom line: this is a reformulation paper, not a solution paper. The idea of sampling orientations and starting points, rotating/translating the boundary, and turning escape-point selection into a TSP over boundary constraints is real and useful. The discrete problems are coherent, and the examples reproduce known results for the line, circle, strip, and point searches, with honest Mathematica notebooks attached. The connections to Moser's worm and opaque sets are reasonable to point out. So there is substance here.\n\nThe soft spot is the load-bearing claim. Theorem 3 says the discrete solutions Gamma-converge to the continuous minimax optimum. The proof is a sketch: it asserts liminf and recovery sequence from continuity and lower semicontinuity, but never builds either. The discrete functionals live on different dimensions depending on M,N, with an extra permutation variable, so standard Gamma-convergence needs a common metric space and control of the order variable; none of that is there.\n\nWorse, the paper's own Appendix III shows the danger: for the unit strip, the discrete run with N=12, M=26 gives length 2.24853, while Zalgaller's known optimum is 2.278292. A shorter path that is not a valid continuous escape strategy cannot be a feasible solution to the original problem, so the finite approximations are not doing what the theorem needs them to do. That doesn't disprove convergence in the limit, but it means the numerical evidence is not supporting the theorem, and the theorem itself is unproved.\n\nAlso, almost every example fixes the order of escape points (e.g., Section 4.2.1 assumes order 0,...,N-1). So they solve a restricted discrete problem, not the full Weak Form I/II. Appendix I uses boundary values chosen to match Isbell's known line solution. Section 10 concedes the exact problems remain intractable.\n\nI agree with the reader's verdict. The reformulation is a valid contribution to the toolkit, but the paper's title and abstract overstate it badly. For a reader interested in computational heuristics for these problems, the paper is worth a look. For a reader wanting a solution of Bellman's problem, it does not deliver.\n\nIf I were handling it, I would send it to a referee with expertise in Gamma-convergence and geometric optimization, because the core idea deserves serious evaluation and the author is honest about limitations. But I would expect the referee to reject the current version: either Theorem 3 needs a real proof, or the claims need to be scaled back to a numerical/heuristic framework. It should not be desk-rejected; it just needs to be judged by someone who can distinguish a genuinely interesting reformulation from a solved problem.","headline":"A genuinely interesting TSP reformulation of Bellman's problem, but the 'general solution' theorem is a sketch and the paper's own numerics show the discretization can undercut the true optimum.","tokens_in":59111,"tokens_out":3125,"would_cite":false,"duration_ms":29015,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["49K30","49Q10","52A40","90C27"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims that Bellman's lost-in-a-forest problem, for any known forest boundary, can be solved by discretizing starting points and orientations and solving a traveling-salesman-type optimization.","keywords":["Bellman's lost-in-a-forest problem","minimax escape path","traveling salesman problem","Hamiltonian path","Gamma-convergence","Moser's worm problem","shortest opaque set","discrete geometry"],"falsifier":"Take the unit circle, whose escape optimum is known to be the diameter of length 2. Solve the Weak Form II mixed-integer program to certifiable global optimality for growing M and N; if the optimal values do not converge to 2, or if the only optima require an escape path that meets a rotated boundary more than once, the general-solution claim collapses.","tokens_in":58012,"feed_emoji":"🧭","tokens_out":4599,"duration_ms":41691,"temperature":0.7,"pith_summary":"Bellman's lost-in-a-forest problem asks for the shortest path that guarantees escape from a forest of known shape when the hiker's starting point and facing direction are unknown. This paper aims to turn that continuous minimax problem into a finite optimization problem by fixing a fine grid of possible starting points and orientations and requiring the escape path to hit one point on each rotated and translated copy of the forest boundary. The discrete version becomes a variant of the traveling salesman problem, and the paper argues that as the grid and orientation steps shrink to zero, the discrete optima converge to the true optimal escape path. If the argument is right, any forest shape given by a boundary equation can in principle be solved by refining a grid and running discrete optimization, and the same machinery transfers to Moser's worm problem and shortest opaque set problems. The paper also reproduces known optimal paths for lines, circles, strips, and triangles, which supports the method.","feed_headline":"Lost-hiker puzzle recast as a traveling-salesman problem","feed_subtitle":"A grid of starting points and orientations turns Bellman's unsolved escape problem into discrete optimization.","key_machinery":"The central device is to freeze the escape path and rotate and translate the forest boundary instead: for each sampled starting point and initial heading, the boundary is moved so that the hiker's coordinate frame stays fixed, and the point where the path first hits the boundary becomes a constrained escape point lying on the transformed boundary Fki(x,y)=0. Collecting all such escape points turns the problem into finding the shortest polygonal path, starting from the origin, that visits one escape point per transformed boundary; this is an open traveling salesman or Hamiltonian path problem, which the paper writes with binary order variables and Miller–Tucker–Zemlin subtour-elimination constraints. The limiting step is the assertion that these discrete optima Γ-converge to the continuous minimax solution.","core_discovery":"The central claim is Theorem 3: when M starting points are evenly distributed in the region like grid points and N orientations are evenly distributed over [0, 2π], the solution to Weak Form II—the shortest path through the MN escape points on the corresponding rotated and translated boundary copies—yields the solution to the original Bellman problem as M and N tend to infinity. Weak Form I treats a known starting point with unknown orientation, while Weak Form II treats finitely many possible starting points. The escape path is sought as the shortest Hamiltonian path through escape points constrained to lie on the transformed boundaries, formulated with Miller–Tucker–Zemlin subtour elimination. The proof of convergence invokes Weierstrass existence, uniform continuity, Riemann-sum convergence, and Γ-convergence.","pith_inferences":["The paper's examples fix the visiting order of escape points, for instance assuming points are visited in increasing index; the full generality claim therefore depends on solving the complete permutation search, not the order-restricted runs shown, so global optima of the full Miller–Tucker–Zemlin problem are needed as the grid refines.","If the convergence theorem survives scrutiny, a natural next step is to quantify the discretization error—how large M and N must be to certify an epsilon-optimal escape path—which the paper leaves open.","The opaque-set connection suggests that the beam-detection constant for the unit circle could be attacked as a sequence of multiple-path mixed-integer programs, but only if the unconnected multi-curve variant is solved globally rather than by an imposed order.","The proposed extension to non-Euclidean geometry is speculative because the rotation and translation constraints rely on Euclidean distance; a testable extension would replace planar rigid motions with spherical or hyperbolic ones."],"forward_implications":["For any boundary expressible as F(x,y)=0, an escape path can be approximated by solving a mixed-integer program whose size grows with the number of sampled starting points and orientations.","Known optimal results—Isbell's half-plane search, Zalgaller's strip, and the circle and point-search cases—are recovered by the same formulation rather than by shape-specific geometry.","The framework gives a route to upper bounds for Moser's worm problem: any solved escape path of length L for a region of area A yields A/L^2 as an upper bound on the minimal covering area.","The same discrete formulation can be applied to opaque sets, replacing escape points with intersection points on all sampled lines, and to three-dimensional analogues.","Closed escape paths that return to the starting point fit the same framework by adding a return segment to the objective, connecting the method to closed-worm variants."],"supporting_citations":[{"why":"Isbell's optimal search pattern for a half-plane, which supplies the classical result the line-search example must reproduce.","marker":"[7]"},{"why":"Zalgaller's solution for the unit strip, used as the comparison target in Sections 4.2.10 and 6.2.1.","marker":"[10]"},{"why":"Finch and Wetzel's survey of known escape paths for the circle and other shapes, providing the benchmarks and the diameter optimality for the circle.","marker":"[11]"},{"why":"Besicovitch's zigzag path for the equilateral triangle, the known optimum the formulation is checked against.","marker":"[13]"},{"why":"Miller–Tucker–Zemlin integer programming formulation of TSP, which supplies the subtour-elimination constraints for the discrete problem.","marker":"[16]"},{"why":"Braides' Gamma-convergence monograph, the cited basis for the claim that discrete optima converge to the continuous minimum.","marker":"[23]"},{"why":"Karp's NP-completeness of combinatorial optimization, cited to explain why the resulting discrete problems are hard.","marker":"[17]"},{"why":"Melzak's point-search result with length 1+2π, reproduced in the one-point example.","marker":"[22]"}],"fun_headline_variants":["Bellman's forest escape recast as traveling salesman","Any forest, any start: a TSP formulation","Grid points turn escape route into a Hamiltonian path","Lost hiker puzzle discretized to optimal path finding"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The whole argument rests on the assumption that the shortest escape path is an open, simple, piecewise-smooth curve that crosses each rotated or translated forest boundary exactly once, and that the discrete optima genuinely converge to that continuous optimum; the convergence proof states the two Gamma-convergence inequalities but does not construct the recovery sequence.","fun_headline_variants_meta":{"raw":{"variants":["Bellman's forest escape recast as traveling salesman","Any forest, any start: a TSP formulation","Grid points turn escape route into a Hamiltonian path","Lost hiker puzzle discretized to optimal path finding"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000746,"raw_usage":{"total_tokens":3254,"prompt_tokens":802,"completion_tokens":2452,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":418,"completion_tokens_details":{"reasoning_tokens":2390}},"tokens_in":418,"tokens_out":2452,"duration_ms":16081,"temperature":1.0,"reasoning_tokens":2390,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-11T15:42:34.542086+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take the unit circle, whose escape optimum is known to be the diameter of length 2. Solve the Weak Form II mixed-integer program to certifiable global optimality for growing M and N; if the optimal values do not converge to 2, or if the only optima require an escape path that meets a rotated boundary more than once, the general-solution claim collapses.","supporting_citations":[{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Isbell's optimal search pattern for a half-plane, which supplies the classical result the line-search example must reproduce."},{"cited_title":"R., Wetzel, J","cited_arxiv_id":null,"evidence_quote":"Finch and Wetzel's survey of known escape paths for the circle and other shapes, providing the benchmarks and the diameter optimality for the circle."},{"cited_title":"E., Tucker, A","cited_arxiv_id":null,"evidence_quote":"Miller–Tucker–Zemlin integer programming formulation of TSP, which supplies the subtour-elimination constraints for the discrete problem."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Braides' Gamma-convergence monograph, the cited basis for the claim that discrete optima converge to the continuous minimum."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Karp's NP-completeness of combinatorial optimization, cited to explain why the resulting discrete problems are hard."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Melzak's point-search result with length 1+2π, reproduced in the one-point example."}],"review_version":1}