{"id":"ed0b9dca-1ac4-417a-ab45-65c937f402d9","arxiv_id":"2606.13987","paper_version":3,"verdict":"REJECT","confidence":"HIGH","novelty_score":5.0,"correctness_risk":"high","formal_verification":"none","parameter_count":1,"one_line_summary":"The paper asserts a certified computational scheme for Bellman's lost-in-a-forest problem via TSPN discretization and convergence, plus new numerical tables for two-line and closed-path variants, but the proof of the certificate is internally inconsistent.","lead":"This paper claims a general proof that Bellman's 'lost in a forest' search problem can be discretized as a traveling-salesman-with-neighborhoods problem whose solutions converge to the true answer. It also tabulates new numerical search paths for two-line forests and closed-loop variants; the proof's central equivalence step is incomplete.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 8's certificate collapses: by Theorem 5, OPT_TSPN(m)=L*_m, and by Theorem 7, L*_m is an increasing lower bound to L*, so the asserted upper bound L̄_m equals the lower bound and cannot be ≥ L* for finite m. The claimed ε-certificate is therefore invalid.","rationale":"The paper's central claim is a certified approximation scheme for Bellman's lost-in-a-forest problem. Theorems 5 and 7 establish that the discretized optimum L*_m equals the TSPN optimum and increases to the continuous optimum L*. The fatal flaw is in Theorem 8: it uses the same quantity OPT_TSPN(m) for both the lower bound and the upper bound. Since L*_m is a lower bound (Lemma 3/Theorem 7), OPT_TSPN(m) cannot also be an upper bound unless L* is attained at finite m, which is not generally true and is nowhere proved. The asserted decreasing upper bound L̄_m directly contradicts the monotone increasing lower bound from Theorem 7; no sequence can do both. Without a valid upper-bound construction (e.g., an explicit path feasible for all g∈G, or a fattening argument), the ε-certificate in Theorem 8 is unsupported and internally inconsistent. The reader's strongest claim is correct; the rejection is warranted. The weaker equivalence concern (Theorem 1 pointwise equivalence vs minimax) is also open, but the internal contradiction is the most decisive and independent failure. No additional evidence in the paper — numerical tables or the 3D reproductions — resolves this, since those numbers are upper-bound constructions for specific families, not for the finite-sample TSPN certificate.","tokens_in":14679,"tokens_out":4942,"duration_ms":51602,"concrete_test":"Take the paper's own two-line example: for β=0 the reported continuous optimum is ≈1.62782. For a small finite orientation set G_m with m orientations (e.g., N=4,8), formulate the MIP in §3 exactly as the paper prescribes and solve to zero gap, obtaining OPT_TSPN(m). By Theorem 5, OPT_TSPN(m)=L*_m. Check whether OPT_TSPN(m) ≥ 1.62782 (as Theorem 8's upper bound would require) or ≤ 1.62782 (as Theorem 7's lower bound would require). The latter will hold for finite m, disproving the claimed two-sided certificate. Alternatively, an even simpler check: with m=1, compare L*_1 to a known L* for a unit line; L*_1 is just the distance from origin to the single transformed line, generally less than the known optimal escape length, so Theorem 8's interval does not contain L*.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Section 3 defines L*_m as the optimum of (P_m), the problem with constraints only for g∈G_m. Lemma 3 shows L*_m is nondecreasing and L*_m ≤ L*. Theorem 5 proves L*_m equals OPT_TSPN(m). Theorem 7 proves L*_m ↑ L*. In Theorem 8, the paper defines Lm := L*_m and L̄_m := OPT_TSPN(m). By Theorem 5 these are the same number. Asserting Lm ≤ L* ≤ L̄_m therefore asserts L* = L*_m. For generic finite G_m, L*_m < L* (the optimal continuous path is not required to pass through the finite set of sampled transformed boundaries), so the claimed inclusion fails. Moreover, the claim that L̄_m ↓ L* contradicts Theorem 7's monotone increase from below; a sequence cannot both increase to L* and decrease to L*. The proof gives no construction of a feasible path for all g∈G — only for the finite sample — so there is no upper-bound certificate. This is not a gap that can be patched by 'm large enough'; it is a logical contradiction.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a formalization of Bellman's lost-in-a-forest problem as a transformed-boundary intersection problem (P), discretizes the parameter space, proves that the discretized problem is equivalent to TSPN (Theorem 5), proves monotone convergence of discrete optima to the continuous optimum (Theorem 7), and claims a certified approximation scheme (Theorem 8). Additional sections present numerical results for searching two lines, closed-path variants, and 3D extensions, and discuss connections to Wetzel's problem and Moser's worm problem.","tokens_in":15012,"tokens_out":5396,"duration_ms":57328,"significance":"If the central theorems were correct, the paper would provide a rigorous computational solution to a classic unsolved minimax problem, a notable result. It also reproduces known benchmarks and collects numerical data. However, the certification theorem is internally inconsistent, and the equivalence between the original minimax objective and the intersection formulation is not proven. The paper therefore does not establish its headline claim.","major_comments":[{"comment":"In Theorem 8 the upper bound is defined as L̄_m := OPT_TSPN(m). Theorem 5 states L*_m = OPT_TSPN(m), so L̄_m = L*_m. Theorem 7 shows L*_m ↑ L*, so L*_m ≤ L* and increases. A number cannot simultaneously be a lower bound and an upper bound unless L*_m = L* for every m, which is false for a generic finite sample. The asserted 'L̄_m ↓ L*' is equally incompatible with Theorem 7. No construction of a path feasible for all g∈G is given, so no upper-bound certificate exists. This invalidates the claimed ε-certificate.","section":"Theorem 8 (§3)"},{"comment":"The original minimax objective is a supremum of first-hitting lengths over starting configurations. The reformulation (P) minimizes total length of a curve intersecting every transformed boundary. Theorem 1 proves only pointwise equivalence of intersection; it does not show that the worst-case first hitting time equals the total length of a feasible sweep path. The proof of equivalence therefore stops short of the load-bearing step. Without a truncation/restart argument, the TSPN discretization need not solve Bellman's problem.","section":"Eq. (1) → (P), §3"},{"comment":"The statement that 'intersecting all forest boundaries' is identical to the curve's convex hull containing the shape is asserted without proof. This premise is not used in the formal development of Section 3, where the constraint is d(f,B_g)=0 for all g. As stated, it is doubtful for bounded targets under pure translations, since a bounded curve cannot intersect every translate of a bounded set. This unproven assertion is load-bearing for the claimed connection to Moser's worm problem and universal covers.","section":"§2, convex-hull equivalence"},{"comment":"Theorem 6 is not a mathematical proof but a conditional statement: if an MIP solver returns zero gap, then the finite TSPN is solved exactly. As stated, it adds no content beyond assuming the solver is correct; in a proof of a general theorem it should be phrased as an assumption or a computational oracle.","section":"Theorem 6"}],"minor_comments":[{"comment":"The proof cites 'Lemma 2' when it means Lemma 4 (polyline optimality).","section":"Theorem 5 proof"},{"comment":"The analytical formula for the arc+tangent path lacks derivation; the expression 1/2 + √3/2 + π/12 + β/2 should be justified.","section":"Eq. (6), §4"},{"comment":"The variables γ_i and ϕ_i are used without a precise definition of their feasible ranges, and the constraint 'mod 2π = [0,2π]' is not explained at the numerical level.","section":"Eqs. (9), (11), (14)"},{"comment":"The Zalgaller 3D length is reported as both 11.605 and 10.605427999164 in the same passage without clear reconciliation.","section":"Section 7, Table 3"},{"comment":"Weak Form IB is described informally; it lacks a formal definition and a theorem comparable to those in Section 3.","section":"Section 5"}],"recommendation":"reject","confidential_remarks":"The central theorems are not merely underproved; Theorem 8 is self-contradictory with Theorem 5 and Theorem 7, and the equivalence step is missing. The computational experiments depend on the author's prior arXiv paper [8] and are not reproducible from this manuscript. The paper would require a complete rewrite of the proof strategy to be salvageable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the headline result, a certified approximation scheme for Bellman's problem, is not correct as written. Theorem 8 is internally inconsistent: Theorem 5 says L*_m = OPT_TSPN(m), and Theorem 8 sets L_m = L*_m and Lbar_m = OPT_TSPN(m), so the two bounds are the same number. A sequence cannot both increase to L* from below (Theorem 7) and decrease to L* from above. The claimed interval [L_m, Lbar_m] has zero length for every m, and containing L* would force L* = L*_m, which fails for generic finite discretizations. No feasible path for all g in G is ever constructed, so there is no upper-bound certificate.\n\nWhat deserves credit: Theorem 7's monotone lower-bound convergence is a standard compactness/density argument and, as far as I can tell, correct. The two-line and closed-path tables, plus the reproduction of Zalgaller's 11.605, add some data not in the cited literature. The arc+tangent formula (6) is a nice closed-form check. But the tables are not reproducible: no code, no discretization sizes, no solver or tolerances.\n\nThere is also a gap at the front end. Theorem 1 shows pointwise intersection equivalence, but the original minimax first-hitting-time problem is not shown to be equivalent to minimizing the total length of a curve intersecting all transformed boundaries. That conversion is load-bearing; without it, the TSPN reduction solves a different problem. The paper's claim that 'intersecting all forest boundaries' is identical to the convex hull containing the shape is also asserted, not proved.\n\nIf the author reframed this as a one-sided convergence result with exploratory numerics, there might be a legitimate modest paper. As it stands, the main theorem is a contradiction, not a gap. The right editorial call is to reject, not send to referees; the flaw is immediate and does not require deep expertise. I would not cite this in its current form, and I would not spend a reading group on it. If a revised version drops the certificate claim and honestly separates the weaker results, I'd take another look.","headline":"Theorem 8's certified bounds are identical by Theorem 5, so the main claim collapses; the one-sided convergence and numerical tables are the only reusable pieces.","tokens_in":15456,"tokens_out":5469,"would_cite":false,"duration_ms":60934,"reading_group":"no","serious_thinker":"no","would_accept_peer_review":false},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["49K30","49Q10","52A40"],"pacs":[],"model":"deepseek-v4-flash","headline":"This paper claims a certified computational solution to Bellman's lost-in-a-forest problem by reducing it to a discrete traveling-salesman-with-neighborhoods problem and proving convergence.","keywords":["Bellman's lost-in-a-forest problem","traveling salesman problem with neighborhoods","discrete geometry","curve covering","Moser's worm problem","convex hull","computational optimization","escape path"],"falsifier":"Construct two candidate paths for a simple forest, such as a unit circle: one that has shorter total length but larger worst-case first-hitting time, and another that is longer in total length but hits every possible boundary earlier. If such a pair exists, the minimax escape problem and the total-length boundary-intersection problem cannot be equivalent, and the TSPN discretization would solve a different problem.","tokens_in":1339,"feed_emoji":"🌲","tokens_out":2094,"duration_ms":72791,"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 but unknown starting position and orientation. This paper attempts to prove that the problem can be recast as a finite traveling-salesman-with-neighborhoods problem on transformed copies of the forest boundary, solved with certified global optimization, and that the discrete optima converge to the true continuous optimum. If correct, this would turn a long-open geometry problem with only a handful of solved shapes into a computationally tractable approximation problem with rigorous error bounds. The paper also extends the framework to searching for two lines, closed paths, translation-only versions linked to Wetzel's unit-arc cover problem, and three-dimensional sphere inspection.","feed_headline":"Bellman's forest problem reduced to solvable discrete searches","feed_subtitle":"A discretized TSPN reformulation with proven convergence would give certified bounds for a classic open geometry problem.","key_machinery":"The load-bearing construction is the rigid-motion transformed boundary B_g = Rot(-alpha)(B - s), which converts the adversarial minimax search over starting positions and orientations into a fixed constraint: the canonical path must intersect every B_g. Discretizing the parameter space into nested finite sets G_m turns this into a finite TSPN with neighborhoods S_h = B_{g_h}, and the polyline optimality lemma reduces the continuous curve problem to a permutation-based shortest-path problem. Monotone convergence and existence of minimizers are carried by compactness arguments via the Arzelà–Ascoli theorem and lower semicontinuity of path length.","core_discovery":"The paper's central assertion is that every possible starting position and orientation of the hiker can be encoded as a rigidly transformed forest boundary, so that a canonical path escapes if and only if it intersects all transformed boundaries. It then claims that the discretized version of this boundary-intersection problem is exactly a TSPN instance, that the discrete optimal values form a nondecreasing sequence converging to the continuous optimum, and that solving the finite TSPN to zero MIP gap yields a two-sided certificate bracketing the true escape length within any prescribed tolerance. The same discretization is applied to several variants, producing numerical candidate paths for","pith_inferences":["If the equivalence between minimax first-hitting time and total-length boundary intersection is made fully rigorous, the framework would make Bellman's problem computationally tractable for arbitrary polygons, not just lines and strips.","One could test the convergence rate numerically on shapes with known exact answers, using the monotone lower bounds to measure how quickly the gap closes as the grid refines.","The rigid-motion discretization generalizes to other transformation groups, so analogous 'search with unknown pose' problems—rotation-only, translation-only, or in higher dimensions—could be cast as finite TSPN instances.","Editorial caution: as printed, Theorem 8's upper-bound half is not derived and appears inconsistent with the monotone lower-bound conclusion of Theorems 5 and 7; the certified two-sided approximation therefore depends on an argument not present in the manuscript."],"forward_implications":["For any compact forest, a finite set of transformed boundary constraints yields a rigorous lower bound on the true escape length, and refining the grid gives a nondecreasing sequence of lower bounds converging to the continuous optimum.","Solving the discretized TSPN with a certified global optimizer would, in principle, produce an epsilon-certificate for the original Bellman problem, making it computationally accessible rather than purely theoretical.","The same machinery gives numerical candidate paths for two-line searches, closed escape paths, and three-dimensional sphere inspection, connecting these variations to known constants such as the Zalgaller sphere-inspection value and the 4π closed-curve result.","Adding a closing segment to the objective turns the problem into a Hamiltonian-path variant, so the framework covers closed search paths as well as open ones.","The translation-only variant matches Wetzel's unit-arc covering problem, giving a TSPN formulation for a problem previously treated separately from Bellman's."],"fun_headline_variants":["Bellman's forest problem: proof and TSPN bridge","Forest escape solved via TSPN convergence","Bellman's lost hiker: discrete proof yields bounds","New proof for the lost-in-a-forest problem","TSPN discretization closes Bellman's forest proof"],"cache_read_input_tokens":16768,"weakest_assumption_plain":"The central claim rests on passing from the minimax first-hitting-time objective (Eq. 1) to the total-length boundary-intersection problem (P); Theorem 1 establishes pointwise intersection equivalence but not that minimizing total length equals minimizing the worst-case distance traveled before first hitting the boundary.","fun_headline_variants_meta":{"raw":{"variants":["Bellman's forest problem: proof and TSPN bridge","Forest escape solved via TSPN convergence","Bellman's lost hiker: discrete proof yields bounds","New proof for the lost-in-a-forest problem","TSPN discretization closes Bellman's forest proof"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000131,"raw_usage":{"total_tokens":897,"prompt_tokens":609,"completion_tokens":288,"prompt_tokens_details":{"cached_tokens":256},"prompt_cache_hit_tokens":256,"prompt_cache_miss_tokens":353,"completion_tokens_details":{"reasoning_tokens":224}},"tokens_in":353,"tokens_out":288,"duration_ms":4389,"temperature":1.0,"reasoning_tokens":224,"cache_read_input_tokens":256,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-02T11:34:01.672085+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Construct two candidate paths for a simple forest, such as a unit circle: one that has shorter total length but larger worst-case first-hitting time, and another that is longer in total length but hits every possible boundary earlier. If such a pair exists, the minimax escape problem and the total-length boundary-intersection problem cannot be equivalent, and the TSPN discretization would solve a different problem.","supporting_citations":[],"review_version":2}