{"id":"5309fd86-f639-4a0b-bad8-776114ed41e3","arxiv_id":"1908.00578","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Visibility from a viewpoint is characterized as the subzero level set of the viscosity solution of the local obstacle problem min{u-g, (x-x*)·∇u}=0.","lead":"The authors show that the visibility set from a point, with obstacles represented as level sets, can be obtained as the subzero level set of a solution to a simple local nonlinear obstacle PDE. This offers a simpler alternative to prior nonlocal PDE formulations and may ease optimization and inverse problems in visibility.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Proposition 3.9's comparison proof is circular: it assumes supersolutions lie above SS_+(g), the very statement being proved, so the convergence theorem 5.8 lacks a proven foundation.","rationale":"My assessment agrees with the reader: the central PDE characterization of visibility is plausible and supported by the explicit envelope formula and numerical experiments, but the manuscript's proof of uniqueness/convergence contains a circular step. The concern is load-bearing because Theorem 5.8 explicitly invokes Proposition 3.9 to replace the strong comparison principle that the Barles–Souganidis framework requires, and without a valid comparison principle the numerical solutions could converge to a different viscosity solution or fail to converge. I do not see a deeper inconsistency: Proposition 3.8 can be verified directly by testing the explicit envelope formula against smooth test functions, and the ray-wise comparison suggested in the test should repair Proposition 3.9. The proof of Proposition 3.6 is also sketchy, but Proposition 3.9 is the critical dependency. Because the gap is localized and likely fixable, the appropriate verdict remains CONDITIONAL, with the condition being a non-circular proof of the comparison principle.","tokens_in":14496,"tokens_out":15895,"duration_ms":116609,"concrete_test":"Restrict (7) to a single ray γ(s)=x*+s v, s∈[0,L], reducing to the one-dimensional obstacle ODE min{U(s)-G(s), s U'(s)}=0, U(0)=g(x*), where G(s)=g(γ(s)). Prove by elementary ODE comparison (without invoking envelopes) that any viscosity subsolution with U(0)≤g(x*) satisfies U(s)≤max_{τ∈[0,s]} G(τ) and any supersolution satisfies U(s)≥max_{τ∈[0,s]} G(τ). If this one-dimensional comparison holds for every ray, Proposition 3.9 follows directly and Theorem 5.8's weakened comparison argument is validated; if a counterexample is found, the convergence claim fails.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"In the proof of Proposition 3.9, the authors assert that for a viscosity supersolution v of (7), 'v(x)≥SS_+(g)(x)' follows from 'Perron's characterization of SS_+(g) and Proposition 2.8.' Proposition 2.8 only establishes that SS_+(g) is the infimum of all star-shaped functions above g; it does not say that every viscosity supersolution of (7) is star-shaped or lies above this envelope. That inequality is precisely the comparison statement Proposition 3.9 is meant to prove, so the argument is circular. Theorem 5.8 then uses Proposition 3.9 in place of the strong comparison principle required by Barles–Souganidis, so the claimed convergence of the finite-difference scheme is not established as written. The result is likely repairable: a supersolution of (7) is a viscosity supersolution of (x−x*)·∇u=0, which should force monotonicity along rays from x*; combined with v≥g this would give v≥SS_+(g) without circularity. The same ray-monotonicity argument, together with u(x*)≤g(x*), should give u≤SS_+(g) for subsolutions. But these steps are absent from the manuscript.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proposes a local PDE obstacle problem for computing visibility sets in a level-set framework. For a signed distance function g (negative outside obstacles, positive inside), the visibility set from a point x* is claimed to be the subzero level set of the solution u of min{u(x)-g(x), (x-x*)·∇u(x)}=0 with u(x*)=g(x*) (Eq. (7)). The solution is identified with the upper star-shaped envelope SS_+(g) whose explicit ray-tracing formula is (6). The paper establishes regularity properties of star-shaped envelopes, derives PDEs for lower and upper envelopes, states a comparison principle, introduces a monotone consistent finite-difference scheme with a one-pass fast sweeping solver, proves first-order convergence in a smooth example, and presents 2D and 3D numerical examples, including a multi-viewpoint extension.","tokens_in":14723,"tokens_out":13135,"duration_ms":133230,"significance":"The proposed PDE is substantially simpler than the earlier nonlocal formulation (1), and the numerical scheme is attractive: the update u(x)=max{g(x), I_h u(\\tilde{x})} is a one-pass sweep. The explicit formula (6) and the convergence table (Table 1) provide good evidence that the characterization is correct. If the comparison argument is supplied, this would be a useful contribution for forward and inverse visibility problems. The multi-viewpoint extension in Section 3.5 is a nice addition.","major_comments":[{"comment":"The proof of the comparison principle is circular. It states that a supersolution v of (7) satisfies v(x)≥SS_+(g)(x), citing 'Perron's characterization of SS_+(g) and Proposition 2.8'; but the statement that every supersolution dominates SS_+(g) is precisely the comparison result to be proved. Proposition 2.8 only gives the explicit formula for SS_+(g); it does not show that viscosity supersolutions belong to the admissible class of star-shaped functions. Since Theorem 5.8 uses Proposition 3.9 in place of the strong comparison principle required by Barles and Souganidis, the convergence claim in Section 5 is not established as written. A repair would be to prove that every supersolution of (7) is nondecreasing along rays from x* and lies above g, which would imply v≥SS_+(g) without circularity; this argument is absent.","section":"Section 3.4, Proposition 3.9"},{"comment":"The proof contains a false assertion: if v is star-shaped with respect to x* in the sense of Definition 2.3, then -v is not star-shaped; it satisfies the opposite inequality along rays. The statement that '-u is the supremum of functions v with -v star-shaped' therefore does not reduce the problem to Proposition 3.7. The result can be recovered by verifying directly that the explicit function (6) is a viscosity sub- and supersolution of (7), and I recommend rewriting the proof in that way.","section":"Section 3.3, Proposition 3.8"},{"comment":"Even with a corrected comparison principle, the Barles-Souganidis argument is not fully written out. One needs to show that the relaxed semilimits of the numerical solutions are sub- and supersolutions of (7) on the whole domain and that the boundary condition is preserved; the manuscript only checks that one semilimit satisfies \\bar{u}(x*)≤g(x*). This is likely routine, but it should be done explicitly.","section":"Section 5.2, Theorem 5.8"},{"comment":"The converse direction of the characterization of star-shaped functions as viscosity subsolutions is not justified in detail. The proof assumes 'without loss of generality' that y is a global maximizer of u and asserts the existence of a linear test function with the required properties; for merely upper semicontinuous functions this needs an argument (localization or sup/inf convolution). Since Proposition 3.6 is used in the Perron arguments of Propositions 3.7 and 3.8, the proof should be expanded.","section":"Section 3.2, Proposition 3.6"}],"minor_comments":[{"comment":"The text contains several typos: 'We star by recalling' should be 'We start by recalling', and 'arrays' should be 'rays' in the proof of Proposition 3.6.","section":"Throughout"},{"comment":"The case x=x* is not handled in the definition of u_h_v; if x* is a grid point, the denominator |x-\\tilde{x}| vanishes. The paper should state that the equation is imposed for x≠x* and that u(x*)=g(x*).","section":"Section 4.1"},{"comment":"The sweep order is described by four loops, but the notation 'i=i*,...,N, j=j*,...,1' is easy to misread; a concrete ordering or pseudocode would improve clarity.","section":"Section 4.2"},{"comment":"The correspondence between the inner/outer star-shaped envelopes and the visibility set should be stated explicitly, since the paper ultimately uses the upper envelope for the visibility set while the figure labels the inner envelope as the visibility set.","section":"Section 2, Definition 2.2 and Figure 1"}],"recommendation":"major_revision","confidential_remarks":"The paper is in scope and the core idea is likely correct; the numerical results and explicit formula support this. I would not reject over the circularity, because a direct proof of Proposition 3.9 is plausible. However, the convergence theorem as stated is not proven, so the revision should be substantive."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"First thing you should know: the genuinely new item is the local obstacle problem (7), min{u-g, (x-x*)·∇u}=0 with u(x*)=g(x*), whose subzero set is the visibility set. The authors show the solution is the upper star-shaped envelope, and the explicit formula (6) is just ray tracing—so the PDE is a reformulation of a known object, not a new visibility model. That is not a flaw: a local PDE with a one-pass fast sweeping solver is a genuinely useful way to compute visibility in level-set and optimization settings, and the multi-viewpoint formula u = min_i u_i is clean.\n\nWhat they do well: the derivation from the envelope definition is clear; the numerical scheme is simple, degenerate elliptic, consistent, and stable in the Barles–Souganidis framework; the convergence table shows clean first order; and the 2D/3D examples, including the Stanford bunny, are convincing. The explicit envelope formulas are correct.\n\nSoft spots. The big one is Proposition 3.9. The proof asserts that any viscosity supersolution v of (7) satisfies v≥SS_+(g), citing Perron and Proposition 2.8. But Proposition 2.8 only identifies SS_+(g) as the infimum of star-shaped functions above g; it does not say supersolutions are star-shaped or lie above the envelope. That inequality is exactly what the comparison principle is supposed to prove. As written, the argument is circular. Theorem 5.8 then uses Proposition 3.9 in place of the strong comparison principle, so the convergence claim lacks a proven foundation. This is likely repairable: a supersolution should be nondecreasing along rays from x*, and together with v≥g that gives v≥SS_+(g); a similar ray-monotonicity argument should place subsolutions below. But those steps are not in the paper. Proposition 3.6 is also sketchy, and no code or data are posted, though that is common for this type of paper.\n\nWho this is for: anyone working on PDE-based visibility, level-set methods, or fast sweeping. The central characterization is probably correct and the proof gap is fixable. I would send this to a serious referee with a request for a repaired comparison proof, not desk reject.","headline":"A clean local obstacle-PDE reformulation of level-set visibility; the characterization is likely correct, but the comparison-principle proof assumes the conclusion it needs to prove, so the convergence theorem is not fully established as written.","tokens_in":15271,"tokens_out":3465,"would_cite":true,"duration_ms":34864,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["35D40","35F21","49L25","65M06"],"pacs":[],"model":"deepseek-v4-flash","headline":"Visibility sets solve a local obstacle PDE","keywords":["visibility","level set method","viscosity solutions","obstacle problem","star-shaped envelope","fast sweeping method","signed distance function"],"falsifier":"Take a bounded continuous obstacle function $g$ for which $x_*$ is not a global minimum, compute the explicit envelope $SS_+(g)$ from (6), and check directly whether any viscosity supersolution of $\\min\\{u-g,(x-x_*)\\cdot\\nabla u\\}=0$ can drop below that envelope at some point; exhibiting one would disprove the comparison principle and thereby the convergence theorem.","tokens_in":14275,"feed_emoji":"👁️","tokens_out":6574,"duration_ms":60303,"temperature":0.7,"pith_summary":"This paper claims that the set of points visible from a viewpoint $x_*$, with obstacles encoded by a signed distance function $g$, is exactly the subzero level set $\\{u \\le 0\\}$ of the viscosity solution $u$ of the local nonlinear obstacle problem $\\min\\{u(x)-g(x), (x-x_*)\\cdot \\nabla u(x)\\}=0$, with $u(x_*)=g(x_*)$. The older formulation in the literature is a nonlocal PDE; this one is local, falls into standard viscosity-solution theory, and leads to a one-pass fast sweeping finite difference scheme. The paper proves the characterization through an explicit star-shaped envelope formula, establishes consistency, stability, and ellipticity of the scheme, and confirms first-order convergence numerically in two and three dimensions. If correct, the result turns ray tracing into a PDE that computes visibility for every level set of the obstacle function at once.","feed_headline":"Visibility sets solve a local obstacle PDE","feed_subtitle":"A subzero level set of a viscosity solution equals the view from a point; one-pass sweeping computes it.","key_machinery":"The load-bearing object is the upper star-shaped envelope $SS_+(g)$ with respect to the viewpoint $x_*$, defined pointwise as the maximum of $g$ along the ray segment from $x_*$ to $x$. The paper proves this envelope is exactly the viscosity solution of the obstacle problem $\\min\\{u-g, (x-x_*)\\cdot \\nabla u\\}=0$: the first term enforces $u\\ge g$, and the second enforces monotonicity along rays from the viewpoint, which is the analytic signature of star-shapedness. This identity is what converts ray tracing into a local PDE and at the same time provides the explicit solution formula (6) that underlies the one-pass fast sweeping scheme.","core_discovery":"The central discovery is that the upper star-shaped envelope with respect to the viewpoint, $SS_+(g)(x) = \\max\\{g(y) : y = x_* + t(x-x_*) \\in \\Omega, t\\in[0,1]\\}$, is the viscosity solution of (7). Consequently the subzero level set of the solution recovers exactly the visibility set from $x_*$, and each sublevel set $\\{u\\le \\alpha\\}$ is the visibility set of the corresponding superlevel set $\\{g\\le \\alpha\\}$. The paper further shows the multi-viewpoint case where a point is visible if seen by at least one viewpoint is solved by taking the minimum of the single-viewpoint solutions, so the same local PDE structure persists across multiple sources.","pith_inferences":["If a fully independent comparison proof is supplied, the same envelope identity likely yields explicit Lipschitz estimates or quantitative error bounds for the fast sweeping scheme, going beyond the qualitative convergence guarantee.","The same obstacle-PDE mechanism could apply to other ray-based queries on continuous functions, such as horizon computation or line-of-sight in terrain, wherever the query set is star-shaped from the source.","The paper's observation that 'visible by all viewpoints' is not star-shaped suggests that boolean combinations of visibility beyond 'at least one' cannot be captured by a single local PDE of this type; a different operator would be needed for such constraints.","For obstacle functions that change in time, the PDE needs to be solved only once; extracting the visibility set at any time is then a lookup of the sublevel set, which could speed up time-dependent visibility problems."],"forward_implications":["All sublevel sets of the solution $u$ are visibility sets of the corresponding superlevel sets of $g$, so one PDE solve gives visibility for every threshold of the obstacle function at once.","Multi-viewpoint visibility under 'seen by at least one' is obtained as $\\min_i u_i$ where each $u_i$ solves the single-viewpoint PDE; the cost is one solve per viewpoint.","For obstacles given as the graph of a height function, the solution of the PDE provides the horizontal visibility set from a given height directly, reducing the dimension of the computation.","The explicit envelope formula (6) gives an exact solution operator that can serve as a reference solution for numerical error measurement, as the paper does in its convergence test."],"supporting_citations":[{"why":"proposes the nonlocal PDE formulation for visibility that this paper simplifies","marker":"[TCO+04]"},{"why":"establishes viscosity-solution properties for the earlier formulation, providing the comparison context","marker":"[KT08]"},{"why":"supplies the convergence framework for monotone, consistent, stable schemes used in the convergence proof","marker":"[BS91]"},{"why":"provides the viscosity solution definitions and Perron method background underpinning the PDE arguments","marker":"[CIL92]"},{"why":"gives the equivalence between degenerate elliptic finite difference schemes and monotone schemes used to verify ellipticity","marker":"[Obe06]"}],"fun_headline_variants":["Visibility set from a PDE obstacle problem","Local PDE solves visibility, no ray tracing needed","Obstacle PDE yields exact visibility sets","One PDE, many views: sublevel sets are visibility"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof assumes, in Proposition 3.9, that every viscosity supersolution of the obstacle problem is pointwise at least as large as the explicit ray-max envelope $SS_+(g)$; that comparison statement is itself what the proof is meant to establish, so the argument needs an independent proof of this comparison to hold without circularity.","fun_headline_variants_meta":{"raw":{"variants":["Visibility set from a PDE obstacle problem","Local PDE solves visibility, no ray tracing needed","Obstacle PDE yields exact visibility sets","One PDE, many views: sublevel sets are visibility"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000126,"raw_usage":{"total_tokens":1046,"prompt_tokens":819,"completion_tokens":227,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":435,"completion_tokens_details":{"reasoning_tokens":169}},"tokens_in":435,"tokens_out":227,"duration_ms":2881,"temperature":1.0,"reasoning_tokens":169,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T15:48:05.441320+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a bounded continuous obstacle function $g$ for which $x_*$ is not a global minimum, compute the explicit envelope $SS_+(g)$ from (6), and check directly whether any viscosity supersolution of $\\min\\{u-g,(x-x_*)\\cdot\\nabla u\\}=0$ can drop below that envelope at some point; exhibiting one would disprove the comparison principle and thereby the convergence theorem.","supporting_citations":[],"review_version":1}