{"id":"b7e37b10-6461-42a2-92d8-960021be27d6","arxiv_id":"2508.20739","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"A new formalism, vertex and edge replacement systems, gives a common construction of limit spaces for self-similar groups, iterated function systems, and edge replacement systems.","lead":"This paper introduces VERS, a general rule-based system for recursively expanding graphs, and shows when the expanding process produces a hyperbolic history graph whose boundary is a limit space. It unifies three known families: self-similar group Schreier graphs, iterated function system attractors, and Belk-Forrest edge replacement limit spaces.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The IFS homeomorphism proof in Theorem 5.17 defines χ on vertical rays only; the Gromov boundary of an augmented tree also contains rays with horizontal edges, and no argument shows every boundary class has a vertical representative or that χ is well-defined on those rays.","rationale":"The central mechanism of Theorem 3.12 is sound in outline: the no-squares argument works once the grid indices in Lemma 3.11 are corrected (B'_n instead of B'_{n-1}), and this appears to be a typo-level issue rather than a structural flaw. The self-similar-group application is direct because Theorem 4.3 identifies the history graph with the known self-similarity graph. The ERS application is also plausible; Theorem 6.12's four-step expansion argument is compressed but consistent with the no-isolated-vertices and no-direct-ιτ-edge conditions. The IFS application is where the proof has a genuine gap. The map to the attractor is only explicitly defined and shown continuous for rays parameterized by left-infinite words. Since the Gromov boundary is a quotient of all geodesic rays, one needs a lemma that every ray in HΦ, including those with horizontal edges, is asymptotic to such a vertical ray, or else χ must be extended and proved well-defined on the full set of rays. The appeal to [LW09, Theorem 4.3] may cover this, but the transfer is not demonstrated; 'works with minimal modifications' is exactly the kind of delegated step that a conditional acceptance should require. The reader's weakest assumption pointed in the same direction, namely the delegated continuity argument and the nondegenerate ratio in Proposition 5.14, so there is partial agreement. I would keep the manuscript conditional: the main construction is credible, but Theorem 5.17 needs a complete proof for non-vertical geodesic rays before the IFS application is fully established.","tokens_in":25374,"tokens_out":41709,"duration_ms":405825,"concrete_test":"Write a complete proof that every geodesic ray γ of HΦ is asymptotic to a vertical ray: for each depth n let w_n be the first vertex of γ at depth n, and use Theorem 5.7 plus the no-big-squares bound to show there is a word v_n with d_H(w_n, v_n) bounded and v_{n+1} extending v_n. Then re-prove Lemma 5.16 and the continuity step of Theorem 5.17 for arbitrary geodesic rays rather than only ...a2a1 rays. If a pcf IFS admits a geodesic ray with no such bounded vertical projection, exhibit that ray explicitly and show it has no limit point in K, which would falsify Theorem 5.17.","verdict_should_be":"CONDITIONAL","load_bearing_attack":"Theorem 5.17 claims ∂HΦ is homeomorphic to K. Its proof, and Lemma 5.16 before it, starts with 'geodesic rays with vertices ...a2a1', i.e., purely vertical rays in the augmented tree. But Definition 2.3 forms ∂HΦ from all geodesic rays, and HΦ has geodesic rays that use horizontal edges. For the interval IFS φ1(x)=x/2, φ2(x)=(x+1)/2, the vertical rays 1,12,122,... and 2,21,211,... are adjacent at every level and hence asymptotic, so a boundary point can have several vertical representatives and rays with horizontal edges; one must prove that at least one vertical representative exists. The text neither proves that every asymptotic class has a vertical representative nor defines χ for a ray that moves horizontally; the statement that [LW09, Theorem 4.3] 'works with minimal modifications' comes only after χ is assumed to be a map on all geodesic rays. If some boundary class has no vertical representative, χ is undefined there and Theorem 5.17 fails. This is the load-bearing step for the IFS application. The off-by-one in Lemma 3.11's grid proof is repairable and is not the main issue.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper introduces vertex and edge replacement systems (VERSs), a general formalism for recursively expanding colored, typed graphs. For a VERS it defines a history graph, an augmented tree whose n-th level records the n-th expansion, and, when this graph is hyperbolic, defines the limit space of the VERS as its Gromov boundary. The main structural result is Theorem 3.12, stating that an expanding VERS has a hyperbolic history graph. The framework is then applied to three settings: finitely generated contracting self-similar groups, where the history graph is shown to coincide with Nekrashevych's self-similarity graph and the limit space with the group's limit space (Section 4); injective post-critically finite iterated function systems, where the history graph is shown to be hyperbolic and its boundary homeomorphic to the attractor (Section 5); and Belk-Forrest edge replacement systems, where the barycentric subdivision construction yields a VERS whose limit space is homeomorphic to the ERS limit space (Section 6).","tokens_in":25674,"tokens_out":22547,"duration_ms":189029,"significance":"The VERS framework is a useful unifying language for several families of recursively constructed graphs and their limit spaces. The hyperbolicity criterion in Theorem 3.12 is simple, explicit, and semi-decidable for a fixed expansion bound, and the three applications demonstrate the breadth of the formalism. The self-similar-group and ERS applications are largely reformulations or adaptations of known results, but the framework itself is new and the ERS homeomorphism in Theorem 6.17 gives a fresh perspective on Belk-Forrest limit spaces. The IFS section recovers and extends existing self-similar-set boundary constructions, though its homeomorphism proof is not fully self-contained. Overall, if the proof gaps identified below are closed, the paper would be a solid contribution to the literature on graph expansions and fractal limit spaces.","major_comments":[{"comment":"The proof of the homeomorphism between the Gromov boundary of HΦ and the attractor K is incomplete. The map χ is defined and analyzed only on geodesic rays of the form ...a2a1, and the paper does not explicitly justify that every boundary point is represented by such a ray. In an augmented tree, any geodesic ray starting at the root must be purely vertical, because a horizontal edge connects vertices of the same level and would make the corresponding prefix longer than the distance from the root; however, this observation is not stated. More importantly, the continuity of χ is delegated to [LW09, Theorem 4.3] with the remark that its proof 'works with minimal modifications', but the specific modifications and the verification that the listed ingredients (Equation 2.1, Propositions 2.2, Theorem 2.3, Lemma 4.1 of [LW09]) apply verbatim to the different augmented tree HΦ are not provided. Since this is the load-bearing step for the IFS application, the authors should either supply a complete proof of continuity in their setting or give a precise statement of which parts of [LW09] transfer and why the differences between the two augmented trees are immaterial.","section":"§5.5, Theorem 5.17"},{"comment":"The hyperbolicity proof for the history graph of an injective pcf IFS uses the ratio R := min{d(p,q) : p,q ∈ PCrit} / max{d(p,q) : p,q ∈ PCrit} as a threshold for the contracting ratio. This ratio is undefined when PCrit is empty or a singleton, and the manuscript does not discuss these degenerate cases. For an IFS with no critical intersections, the history graph is a purely vertical tree and the claim is trivially true; for a singleton PCrit a separate argument is needed. Theorem 5.15 asserts hyperbolicity for every injective pcf IFS, so the proof should either rule out these cases or treat them separately.","section":"§5.4, Proposition 5.14 and Theorem 5.15"}],"minor_comments":[{"comment":"In the description of edge expansion, the phrase 'the edge xa→ya of R_{c(e)} (a∈{i,t})' is confusing and likely a typo: since edges of R_c connect vertices of the form x_i or x_t, the edge should be written x_a → y_b with a,b ∈ {i,t}, and the replacement of i by u and t by v should be applied to each endpoint independently.","section":"§2.2, Definition 2.4 and expansion description"},{"comment":"The notation T_{i≥1}K_{a_i} should be T_{i≥1}K_{a_i...a_1} (or an explicit parenthetical clarification) to be consistent with the definition of χ in Subsection 5.1, where K_w is defined for finite words w.","section":"§5.5, Lemma 5.16 and Theorem 5.17"},{"comment":"In the proof of Lemma 3.11, the endpoints of the top side T of a geodesic (n+1)-square are set to be P^n(u) and P^n(v), but since the vertical sides have length n+1, these endpoints should be P^{n+1}(u) and P^{n+1}(v). The apparent off-by-one propagates through the construction of the subsquare; the lemma itself is true and the argument is repairable, but the written proof should be corrected.","section":"§3.2, Lemma 3.11"},{"comment":"The statement of Proposition 5.14 should explicitly assume that PCrit has at least two distinct points so that the ratio min/max is defined and positive; otherwise the condition is vacuous or undefined for degenerate IFSs.","section":"§5.4, Proposition 5.14"},{"comment":"The paper would benefit from a brief explicit statement that in an augmented tree, every geodesic ray starting from the root is purely vertical (the argument is one sentence: a horizontal edge preserves the level, so its use would make a prefix longer than the distance from the root). Such a remark would clarify the discussion of geodesic rays in Sections 5 and 6.","section":"§3.2, Theorem 3.12 and Remark 3.13"}],"recommendation":"major_revision","confidential_remarks":"The paper's main novelty is the VERS framework itself; the three applications are largely reformulations or adaptations of known results (Nekrashevych's limit spaces, Lau-Wang boundaries for self-similar sets, and Belk-Forrest ERS limit spaces). The framework is attractive and the central hyperbolicity theorem is clean, but the IFS section's homeomorphism proof is not fully self-contained and the degenerate post-critical case is not handled. I believe these are fixable, but they are load-bearing for the paper's claims. The off-by-one in Lemma 3.11 is a typo-level issue, not a fundamental obstacle."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: the VERS machinery is a real contribution and the main hyperbolicity theorem reads clean, but the IFS application has a hole that the authors will need to patch. The proof of Theorem 5.17 defines the boundary map χ only on geodesic rays that go straight up the tree (the \"vertical\" ones), then claims it descends to the Gromov boundary. There is no argument that every boundary point has a vertical representative. The history graph contains geodesic rays with horizontal edges, and the text doesn't define χ on those. This is not a tiny omission: it's the step that identifies the boundary with the attractor for the main new example class. For the interval IFS, two different vertical rays are asymptotic to the same boundary point, and there are also non-vertical rays in the same class; the paper gives no reason that the class has a vertical representative. I think the gap is repairable—one could define χ on any ray by intersecting the K-sets of its vertices, then prove it's well-defined—but as written, Theorem 5.17 is not proven.\n\nWhat's good: Definition 2.4 is a reasonable formalization of graph replacement systems; Theorem 3.12 is a nice sufficient condition, proved from Kaimanovich's criterion without hand-waving. The encoding of self-similar groups (Section 4) is a clean translation of Nekrashevych's self-similarity graph, with a careful proof. The ERS part (Section 6) is the most original: the barycentric subdivision trick to turn edge replacements into vertex replacements is clever and gives the ERS limit space as a Gromov boundary. The examples are well chosen.\n\nSoft spots besides the IFS gap: Proposition 5.14 assumes the post-critical distance ratio is well-defined and nonzero; degenerate cases are brushed aside. The continuity argument in Theorem 5.17 is delegated to [LW09] with \"minimal modifications\", and that delegation comes after χ is already assumed to be defined on all boundary points. The ERS expansivity proof (Theorem 6.12) is compressed—some \"cannot feature\" claims need more justification. Minor off-by-one issue in Lemma 3.11's grid proof, but that is easily fixed.\n\nOverall: this deserves a serious referee. The framework is worth discussing, and the applications are largely correct in spirit. If the IFS gap is patched, this is a solid paper. I'd send it to review.","headline":"Solid VERS framework, but the IFS homeomorphism proof has an unpatched gap: the boundary map is only defined on vertical rays and the paper never shows every boundary point has one.","tokens_in":26190,"tokens_out":5326,"would_cite":true,"duration_ms":47869,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["20F65","05C25","28A80","37B10","05C12"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper claims that an expanding vertex and edge replacement system always yields a hyperbolic history graph, and that the Gromov boundary of that graph unifies the limit spaces of self-similar groups, iterated function systems, and…","keywords":["vertex and edge replacement system","history graph","Gromov boundary","hyperbolic graph","self-similar group","iterated function system","edge replacement system","augmented tree"],"falsifier":"Take a candidate expanding replacement system and expand, n times, every length-n chain of edges that the rules allow; if any such expanded level still contains a shortest path of length n running between the two endpoint descendants, the system is not expanding, and if such a square appears in a system that is expanding, the main hyperbolicity theorem is false.","tokens_in":25194,"feed_emoji":"🌀","tokens_out":9669,"duration_ms":88196,"temperature":0.7,"pith_summary":"The paper introduces VERSs, a general recipe for building graphs level by level: a starting bouquet of loops is expanded by finitely many rules that replace vertices and edges according to their types and colors. Each VERS has a history graph, an augmented tree whose vertical edges record ancestry and whose horizontal edges record adjacency at each level. The paper proves that if the VERS is expanding—informally, if iterated replacement eventually lengthens every possible shortest path—then this history graph is hyperbolic. The Gromov boundary of that graph is then defined to be the limit space of the VERS. The payoff is that three established limit-space constructions, from contracting self-similar groups, injective post-critically finite iterated function systems, and expanding edge replacement systems, all appear as special cases of this one boundary construction.","feed_headline":"Expanding replacements build hyperbolic history graphs","feed_subtitle":"One framework turns self-similar group limits, IFS attractors, and edge-replacement fractals into the same kind of boundary.","key_machinery":"The load-bearing object is the history graph of the VERS, an augmented tree: vertical edges form a rooted tree recording each expansion step, and horizontal edges join vertices on the same level that are adjacent in that level's graph. The paper uses two tools on this structure. The first is the spanning lift of a horizontal edge, the unique edge one level down whose expansion produced it; lifting cannot increase distances, and this links lengths of shortest paths across levels. The second is the no-big-squares criterion for augmented trees: an augmented tree is hyperbolic exactly when the side lengths of its geodesic squares are uniformly bounded. The expanding condition for VERSs is engineered so that any geodesic square of side n would force a length-n path whose n-th expansion still contains a length-n shortest path, contradicting Definition 3.8. A separate bridge, barycentric subdivision—inserting a new vertex inside every edge—converts edge replacement expansions into VERS expansions and carries the ERS limit space over.","core_discovery":"The central claim is Theorem 3.12: an expanding VERS has a hyperbolic history graph. Since a hyperbolic graph has a well-defined Gromov boundary, this lets the paper define the limit space of a VERS as that boundary. For a finitely generated contracting self-similar group, the associated VERS reproduces the known self-similarity graph, so its boundary is the group's limit space. For an injective post-critically finite iterated function system, the horizontal edges of the history graph correspond exactly to non-empty intersections of cylinder sets, and the boundary is homeomorphic to the attractor. For an expanding edge replacement system, the VERS expansions are barycentric subdivisions of the ERS expansions, and the boundary is homeomorphic to the ERS limit space. In all three cases the same abstract object—the boundary of one hyperbolic augmented tree—recovers the previously studied space.","pith_inferences":["Inference: since the IFS part of the construction never uses similarity ratios, the identification of boundary with attractor is purely topological; two attractors that are metrically different could still have the same VERS boundary, and the framework offers no metric control.","Inference: replacing the full shift by an arbitrary edge shift in the IFS VERS would likely produce a graph-directed version of the correspondence, connecting the construction to graph-directed fractal families.","Inference: the barycentric-subdivision bridge looks flexible enough to carry VERS technology to hyperedge replacement systems, which would give limit spaces with isolated points and almost-expanding behaviors.","Inference: if the suggested VERS structure on atoms of hyperbolic groups exists, then Gromov boundaries of hyperbolic groups would come with a recursive expansion description, potentially making boundary homeomorphism questions computational."],"forward_implications":["Every expanding VERS has a well-defined, compact, metrizable limit space: the visual Gromov boundary of its history graph.","Schreier graphs of any finitely generated contracting self-similar group can be generated by one uniform recursive procedure instead of ad hoc constructions.","For injective pcf IFSs, the attractor is homeomorphic to a single hyperbolic graph boundary, with horizontal edges encoding exactly which cylinder intersections are non-empty.","Expanding edge replacement systems, including the standard examples of rearrangement-group fractals, have limit spaces that are homeomorphic to VERS boundaries, so they inherit compactness and metrizability.","Hyperbolicity of a history graph is strictly more general than the expanding condition: the authors give an explicit VERS whose history graph is hyperbolic but which is not expanding."],"supporting_citations":[{"why":"Supplies the theorem that an augmented tree is hyperbolic exactly when its geodesic squares have bounded size; Theorem 3.12 applies this criterion.","marker":"[Kai03]"},{"why":"Introduces the self-similarity graph whose Gromov boundary is the limit space of a contracting self-similar group, reproduced here by the VERS R(G,S).","marker":"[Nek03]"},{"why":"Provides the theory of contracting self-similar groups, restrictions, and limit spaces used to build the group VERS.","marker":"[Nek05]"},{"why":"Defines expanding edge replacement systems and their limit spaces, which Section 6 realizes as VERS boundaries.","marker":"[BF19]"},{"why":"Supplies the argument that an attractor of a self-similar IFS is the hyperbolic boundary of an augmented tree, including the continuity step imported in Theorem 5.17.","marker":"[LW09]"},{"why":"Gives the standard IFS and attractor framework, including existence and uniqueness of the attractor and the address map used in Section 5.","marker":"[Fal90]"},{"why":"Contributes the address map and the critical and post-critical symbol terminology used to define the IFS VERS R_Phi.","marker":"[Kig01]"}],"fun_headline_variants":["Expanding replacements unify group, IFS, and fractal limits","One hyperbolic tree boundary recovers three fractal families","Expanding VERS: hyperbolic history graphs define limit spaces","A single augmented tree yields Gromov boundaries for many fractals","Hyperbolic history graphs turn replacements into limit spaces"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The construction for iterated function systems assumes the contracting maps never overlap except at finitely many special points, those special points stay at positive distances from one another, and a continuity argument borrowed from an earlier construction still works on this particular tree; if any of those assumptions fails, the claimed homeomorphism may fail.","fun_headline_variants_meta":{"raw":{"variants":["Expanding replacements unify group, IFS, and fractal limits","One hyperbolic tree boundary recovers three fractal families","Expanding VERS: hyperbolic history graphs define limit spaces","A single augmented tree yields Gromov boundaries for many fractals","Hyperbolic history graphs turn replacements into limit spaces"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000356,"raw_usage":{"total_tokens":1861,"prompt_tokens":802,"completion_tokens":1059,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":418,"completion_tokens_details":{"reasoning_tokens":978}},"tokens_in":418,"tokens_out":1059,"duration_ms":8010,"temperature":1.0,"reasoning_tokens":978,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T16:44:54.889065+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take a candidate expanding replacement system and expand, n times, every length-n chain of edges that the rules allow; if any such expanded level still contains a shortest path of length n running between the two endpoint descendants, the system is not expanding, and if such a square appears in a system that is expanding, the main hyperbolicity theorem is false.","supporting_citations":[],"review_version":2}