{"id":"14b0512d-4853-4848-a6d9-5e5e58f53dd4","arxiv_id":"2506.23112","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":5.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"For any signed graph, the positive inertia index is at least (n - p)/2 minus the cyclomatic number, and equality holds exactly for disjoint unions of certain signed cycles.","lead":"This math paper proves a lower bound on how many positive eigenvalues a signed graph must have, in terms of its number of vertices, pendant vertices, and independent cycles. It also lists exactly which signed graphs hit the bound, a complete answer to an extremal question in spectral graph theory.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The proof of Theorem 3.1, Case 3 contains an off-by-one error in applying Lemma 2.4(ii): Γ−y has s+1 components (the isolated pendant x is one), so θ((Γ−y)−x)=θ(Γ)−d(y)+s+1, not +s; the printed derivation of the stronger bound is therefore invalid.","rationale":"The reader's verdict is CONDITIONAL and that remains the right level: the central inequality and equality characterization appear correct, but the proof of the stronger bound in Theorem 3.1 has a genuine gap. The reader identified the unproved fact d(x)≥s+2 in Case 1; that fact is true, though it should be proved. My stress-test found a second, more definite defect in Case 3: Lemma 2.4(ii) is misapplied because the isolated pendant vertex x is not counted as a component of Γ−y. After correcting the off-by-one, the desired stronger bound still follows by combining the induction hypotheses with the stronger form of Lemma 2.4(i), so the theorem is not in doubt; the manuscript merely needs the Case 3 derivation fixed and, ideally, the d(x)≥s+2 fact stated explicitly. Since both issues are local proof repairs rather than counterexamples to the statements, no verdict change is warranted, but the paper should not be accepted without these corrections.","tokens_in":11409,"tokens_out":30380,"duration_ms":291188,"concrete_test":"Verify the disputed equality on a model graph: let Γ be two triangles joined by the path a1–c–d–b1, with a pendant leaf x attached to c. Then n=9, θ(Γ)=2, d(c)=3, and (Γ−c)−x has s=2 components. The printed formula gives θ((Γ−c)−x)=θ(Γ)−d(c)+s=2−3+2=1, while direct inspection gives θ((Γ−c)−x)=2. If the direct computation disagrees with the printed formula, the off-by-one is confirmed. Equivalently, re-derive the final display of Case 3 using θ((Γ−y)−x)=θ(Γ)−d(y)+s+1 and 2d(y)≥2s+m−r+2, and check that the desired bound i_+(Γ)≥n/2−θ(Γ) follows.","verdict_should_be":"UNCHANGED","load_bearing_attack":"In Theorem 3.1, Case 3 (unique pendant vertex, d(y)≥3), the proof defines H_1,...,H_s as the components of (Γ−y)−x. Since x is a pendant vertex, Γ−y consists of these s components plus the isolated vertex x, so Γ−y has s+1 components. Lemma 2.4(ii), applied to the connected graph Γ at the vertex y, therefore gives θ(Γ−y)=θ(Γ)−d(y)+s+1. Removing the isolated vertex x does not change the cyclomatic number, so θ((Γ−y)−x)=θ(Γ−y)=θ(Γ)−d(y)+s+1. The equality printed in the proof, θ((Γ−y)−x)=θ(Γ−y)=θ(Γ)−d(y)+s, is off by one. This false intermediate drives the displayed bound i_+(Γ)≥(n+1)/2−θ(Γ); with the correct +1, and using the correct consequence 2d(y)≥2s+m−r+2 of Lemma 2.4(i) applied to Γ−x, the same argument can be repaired to yield exactly the needed bound i_+(Γ)≥n/2−θ(Γ) when p(Γ)=1. Thus the 'moreover' part of Theorem 3.1 does not follow from the proof as written in the unique-pendant/d(y)≥3 subcase. This is load-bearing because the equality characterization in Theorem 3.2 and the corresponding results in Section 4 rely on the stronger bound. The defect is repairable and appears to be a proof gap rather than a false theorem. The reader's concern about the unproved assertion d(x)≥s+2 is real but the assertion itself is true; the off-by-one here is a more definite error.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the positive and negative inertia indices of signed graphs in terms of the order n, the cyclomatic number θ, and the number of pendant vertices p. The main result (Theorem 3.1) is the inequality i_+(Γ) ≥ (n−p(Γ))/2 − θ(Γ) for connected signed graphs of order n≥2, together with a stronger bound (n−p(Γ)+1)/2 − θ(Γ) when p(Γ)≥1 or when p(Γ)=0 and two distinct cycles share vertices. Theorem 3.2 characterizes the extremal graphs attaining equality as disjoint unions of signed cycles C_{n_i} with n_i≡0 mod 4 if balanced and n_i≡2 mod 4 if unbalanced. By applying the results to the negation of Γ, the paper obtains the analogous statements for i_−, and then derives the nullity bound η(Γ)≤p(Γ)+2θ(Γ), with the improved bound when the stronger hypotheses hold. The proof is by induction on the order, using interlacing, a pendant-vertex reduction lemma (Lemma 2.2), and elementary identities for the cyclomatic number (Lemma 2.4).","tokens_in":11791,"tokens_out":22661,"duration_ms":224258,"significance":"If the proof gaps identified below are repaired, the results are a solid and useful contribution to the spectral theory of signed graphs. The paper gives a clean, parameter-free inequality relating three natural graph invariants, completely characterizes the extremal graphs, and recovers known simple-graph nullity results from [9] as a by-product. The proof is largely elementary and self-contained, and it is not circular: the only self-citation ([4]) is background material and is not used in the proofs. The equality characterization in Theorem 3.2, if fully justified, is a genuine structural result rather than a merely numerical one. The main weaknesses are localized proof gaps: an off-by-one error in the application of Lemma 2.4(ii) in the unique-pendant case, and an unproved graph-theoretic assertion in the intersecting-cycles case.","major_comments":[{"comment":"There is an off-by-one error in the application of Lemma 2.4(ii). The proof defines s as the number of components of (Γ−y)−x. Since x is the pendant neighbor of y, Γ−y consists of these s components together with the isolated vertex x, so Γ−y has s+1 components. Lemma 2.4(ii) therefore gives θ(Γ−y)=θ(Γ)−d(y)+s+1, and deleting the isolated vertex x does not change θ, so θ((Γ−y)−x)=θ(Γ)−d(y)+s+1, not θ(Γ)−d(y)+s as printed. The displayed derivation of i_+(Γ)≥(n+1)/2−θ(Γ) in this subcase is consequently invalid. The gap is repairable: applying Lemma 2.4(i) to Γ at y, with Γ−y having s+1 components, gives d(y)≥m+(s+1)−r; combined with the trivial bound d(y)≥s+1 (every component of (Γ−y)−x contains at least one neighbor of y), one obtains 2d(y)≥2s+m−r+2, which together with the corrected θ yields exactly i_+(Γ)≥n/2−θ(Γ), i.e. the desired stronger bound since p(Γ)=1. The same off-by-one occurs in Theorem 4.1, Case 3, and must be fixed there as well.","section":"Theorem 3.1, Case 3 (d(y)≥3 subcase)"},{"comment":"The assertion that choosing x in the intersection of two distinct cycles with d_{C1∪C2}(x)≥3 implies d(x)≥s+2 is not proved in the text. The statement is true: writing a_k for the number of neighbors of x in the k-th component of Γ−x, the existence of two distinct cycles whose union has degree at least 3 at x forces Σ_k(a_k−1)≥2, hence d(x)=Σ_k a_k≥s+2. However, this fact is load-bearing: it is used to prove the 'moreover' bound in the case p(Γ)=0 with two intersecting cycles, and that bound is in turn used in the equality characterization (Theorem 3.2) and in Corollaries 4.1–4.2. The proof should supply the short argument instead of leaving it to the reader. The analogous step in Theorem 4.1, Case 1, needs the same clarification.","section":"Theorem 3.1, Case 1 (final paragraph)"}],"minor_comments":[{"comment":"There are several typos and grammatical slips: 'Denoted by' should be 'Denote by', 'spire us' should be 'inspire us', 'inequaliy' should be 'inequality', and 'well-know' should be 'well-known'.","section":"Throughout"},{"comment":"The wording 'deleting u together with the vertices adjacent to it' is ambiguous; the lemma is used for deleting a pendant vertex and its unique neighbor, so it should say 'deleting u and its unique neighbor'.","section":"Lemma 2.2"},{"comment":"The sentence 'y must belong to some cycle of Γ because d(y)≥3 and x is unique pendant vertex' is false: for example, take two cycles, connect each by a path to y, and attach the pendant vertex x to y. Then d(y)=3, x is the unique pendant vertex, and y lies on no cycle. The needed bound d(y)≥s+1 holds for the trivial reason that every component of (Γ−y)−x contains at least one neighbor of y, so the incorrect justification should be replaced.","section":"Theorem 3.1, Case 3"},{"comment":"The claim that contracting each induced signed cycle yields a tree is asserted with 'Clearly' but is part of the structural argument. A sentence explaining that a cycle in the contracted graph would lift to a cycle of Γ sharing vertices with at least two original cycles would make the proof self-contained.","section":"Theorem 3.2, converse"},{"comment":"The domain for signed cycles (n≥3) should be stated explicitly, since the formulas for C_2 are not defined in the simple-graph setting.","section":"Lemma 2.5"}],"recommendation":"major_revision","confidential_remarks":"The manuscript is within the scope of standard algebraic graph theory journals. The main theorem is very likely correct and the defects are localized, but the off-by-one error in the unique-pendant case directly affects the proof of the stronger bound that underpins the equality characterization. Since the fix is straightforward but must be written into Theorems 3.1 and 4.1, I recommend major revision rather than rejection. The authors should also add the missing proof of d(x)≥s+2 in the intersecting-cycles case."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Short version: this is a real extension of the unsigned result of Ma–Wong–Tian to signed graphs, with the right balanced/unbalanced distinction in the extremal characterization. The main inequality is very likely correct. But the proof as written has an off-by-one error in the unique-pendant subcase of Theorem 3.1, and that gap is load-bearing for the extremal theorem.\n\nWhat is new: the inequality i+(Γ) ≥ (n−p)/2−θ and the stronger version with +1/2, plus the equality characterization as disjoint unions of cycles with n≡0 mod 4 (balanced) or n≡2 mod 4 (unbalanced). These do not appear in the unsigned paper [9]; switching to signed graphs is not cosmetic because the inertia indices of cycles depend on balance. The corollaries for negative inertia and nullity are direct consequences once the main theorem is fixed.\n\nWhat is done well: Lemma 2.4 is correctly extended to signed graphs; the induction structure is mostly transparent; the extremal characterization is clean and checks out against Lemma 2.5. The self-citation [4] is background only, not used in the proof, so no circularity.\n\nThe soft spots, in proportion: First, the reader's complaint about Case 1 is real: the assertion that a vertex in the intersection of two cycles with degree at least 3 in their union satisfies d(x) ≥ s+2 is stated without proof. It is true, but should be justified. Second, and more serious, Case 3 of Theorem 3.1 contains an off-by-one. The proof defines s as the number of components of (Γ−y)−x and then writes θ((Γ−y)−x)=θ(Γ−y)=θ(Γ)−d(y)+s. Since x is a pendant vertex, Γ−y has s+1 components (the isolated x plus the H_i), so applying Lemma 2.4(ii) to Γ at y gives θ(Γ−y)=θ(Γ)−d(y)+s+1. The printed equality is off by one, and the displayed bound i+(Γ) ≥ (n+1)/2−θ uses exactly that wrong value. With the correct +1, the same algebra gives only the weaker n/2−θ bound. Thus the 'moreover' part is not proved in the unique-pendant/d(y)≥3 subcase. This matters because Theorem 3.2 and the Section 4 results rely on that stronger bound. The defect is likely repairable—I would be surprised if the theorem is false—but it has to be fixed rather than waved through.\n\nAudience: people working on inertia indices of signed graphs or nullity bounds; it is a modest but useful step. I would send it to a serious referee, because the result is worth having and the gap is specific and fixable. A revised version that fixes the off-by-one and makes the Case 1 assertion explicit would be acceptable.","headline":"A genuine signed-graph extension of the Ma–Wong–Tian bound, but the proof of the stronger bound contains an off-by-one error in the unique-pendant case that must be fixed.","tokens_in":12300,"tokens_out":11851,"would_cite":true,"duration_ms":119836,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["05C50","05C22"],"pacs":[],"model":"deepseek-v4-flash","headline":"The paper proves a sharp lower bound for the positive inertia index of signed graphs in terms of cyclomatic number and pendant vertices, and classifies equality as disjoint unions of signed cycles with lengths 0 mod 4 (balanced) or 2 mod…","keywords":["inertia indices","positive inertia index","negative inertia index","nullity","cyclomatic number","pendant vertices","signed graphs","extremal signed graphs"],"falsifier":"Enumerate all connected signed graphs of order at most 8 with $p(\\Gamma)=0$ and $\\theta(\\Gamma)=2$ and compute $i_+(\\Gamma)$ directly: any such graph with $i_+(\\Gamma)=n/2-\\theta(\\Gamma)$ whose cycles share a vertex would contradict the necessity direction of Theorem 3.2. Alternatively, check the unproved degree-counting step by looking for a vertex lying on two distinct cycles with $d(x)