{"id":"b05ea329-c261-4e7a-a27f-e9983017d080","arxiv_id":"2608.22139","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Every simple graph of order at least five satisfies E(G) ≥ r(G) + d̄(G) - 1, with equality only for complete graphs and perfect matchings.","lead":"This paper proves that the energy of any graph with at least five vertices is at least the rank of its adjacency matrix plus its average degree minus one, and it identifies exactly which graphs hit that minimum. The result settles five previously open conjectured lower bounds for graph energy in one go, giving spectral graph theory a single organizing inequality.","discovery_kind":"unification","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Main p≥3 argument rests on unverified preprints [10], [11], [2]; additionally, Proposition 6.1 contains a repairable but real gap after the bipartite conclusion.","rationale":"The reader's conditional verdict is appropriate: the proof of the main p≥3 case is conditional on the unverified preprints [10], [11], and [2], and those dependencies are genuinely load-bearing. At the same time, the reader's statement that there is no obvious internal derivation error is not quite right. Proposition 6.1 contains a non-sequitur: from the fact that a bipartite component of ar G with a cycle produces the desired strict inequality, the proof concludes that every component of ar G is acyclic. This conclusion does not follow, since non-bipartite cyclic components of ar G were not ruled out. The gap is easily repaired because the p=2 case is actually impossible for n≥6: the inertia bound α(G)≤2, together with bipartiteness, forces each bipartition class to have size at most two, hence n≤4. Therefore the repair does not change the truth of the theorem, and the external preprints remain the more serious obstacle to certifying the central claim. The verdict should stay CONDITIONAL until those citations are verified and the Section 6 proof is corrected.","tokens_in":19276,"tokens_out":34743,"duration_ms":317736,"concrete_test":"To settle the external dependency, obtain [10], [11], and [2] and independently re-derive the exact statements used in (4), (5), and the two-range nonsingular bounds, or verify them exhaustively for all graphs of order n≤9. For the internal gap, replace the forest paragraph in Proposition 6.1 with the observation that α(G)≤2 and bipartiteness imply n≤4, and confirm that no later section uses the forest construction.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The paper's central claim is only as secure as three cited external results that are not checked here: (4) from [10], (5) from [11], and the two-range nonsingular bounds from [2]. These are used in Section 4 to restrict any putative counterexample (ρ>7.11, a<log n+1) and to derive the determinant-variance contradiction (Proposition 5.6). If any one of these preprints has a gap, the proof of the p≥3 case does not go through, and Theorem 1.1 is not established. A separate internal issue appears in Proposition 6.1: after proving that a non-bipartite G with p=2 is impossible, the proof turns to components of ar G and concludes 'Therefore every component of ar G is acyclic, and ar G is a forest.' This does not follow: only bipartite components with cycles were treated, and a bipartite graph's complement need not be a forest. The case is actually vacuous for n≥6, since α(G)≤2 and bipartiteness force both bipartition classes to have size at most two, so n≤4; adding this one-line argument repairs the proof. As written, however, the p=2 case is not rigorously completed.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper proves a rank–average-degree energy bound: for every finite simple graph G of order n≥5, E(G) ≥ r(G)+d̄(G)−1, with equality exactly for K_n and, when n is even, (n/2)K_2. The proof proceeds by establishing a vertex-deletion monotonicity for E−d̄, treating the nonsingular case with a determinant–variance reduction and an analytic scalar inequality, handling the p=2 case through the complement, and then reducing singular graphs to induced nonsingular subgraphs. As consequences, the paper derives five previously conjectured lower bounds for nonsingular graph energy, including the Δ+δ and Zagreb-index bounds.","tokens_in":19455,"tokens_out":23258,"duration_ms":214433,"significance":"If sound, this is a substantial result: it unifies several open conjectures under one sharp inequality and gives a complete extremal characterization. The paper is genuinely self-contained in its analytic parts: the scalar inequalities are proved with explicit elementary estimates and no fitted parameters, and the base cases are handled carefully. The main caveat is that the nonsingular proof leans on three recent preprints ([2], [10], [11]) and the equality characterization on the author's own preprint [12]; those dependencies are outside the verification offered here.","major_comments":[{"comment":"The p(G)=2 case is not rigorously completed. The interlacing argument in the first paragraph is valid only if the induced odd cycle is in G; if, as the surrounding notation suggests, it is in the complement of G, then p(complement)≥3 does not contradict p(G)=2, because interlacing for induced subgraphs of the complement controls the inertia of the complement, not of G. The same problem occurs in the 2P3 paragraph: the displayed characteristic polynomial is that of two disjoint copies of P3 in the complement, whereas interlacing in G would require the positive inertia of the complementary induced subgraph, which is never computed. Consequently, the conclusion that the complement is a forest with at most one component of order at least three is not established, and the subsequent double-star/star analysis rests on an unproved structural assumption. The case may be repairable, but as written it is a load-bearing gap.","section":"Section 6 (Proposition 6.1)"},{"comment":"Theorem 1.1 depends essentially on three external results that are used as black boxes: inequality (4) from [10], inequality (5) from [11], and the two-range nonsingular bounds from [2], which are used to restrict counterexamples to ρ>7.11 and a<logn+1 and to control the determinant–variance contradiction in Proposition 5.6. Since [2], [10], and [11] are recent preprints and a gap in any one of them would invalidate the central theorem, the paper should either prove these results or explicitly state that the main theorem is conditional on them.","section":"Sections 2, 4, and 8"},{"comment":"The equality characterization of Theorem 1.2 relies on the author's own preprint [12, Theorem 1.3] for the classification of graphs with E(G)=2(n−α(G)). Because Theorem 1.2 is a central claim and [12] is not independently verified in this manuscript, the equality result should not be presented as unconditional; at minimum the relevant classification theorem should be stated in full or proved.","section":"Section 8 (Lemma 8.2)"}],"minor_comments":[{"comment":"The overline notation for the complement is missing in several displays (for example, 'B=A( G)' and 'm=|E( G)|'), making it hard to follow which graph is meant; please correct the notation throughout.","section":"Section 6"},{"comment":"After showing that a cyclic component yields strict inequality, the sentence 'Therefore every component of the complement is acyclic' should be phrased as 'It remains to consider the case where every component is acyclic'; without an explicit counterexample assumption, the strict inequality already finishes that case.","section":"Section 6"},{"comment":"The phrase 'in their stated ranges' for the five consequences is vague; please specify the exact order restriction n≥5 for each of the five inequalities.","section":"Section 1"},{"comment":"The proof is dense but self-contained; consider adding a short table of the endpoint values for a=1,2,3 to improve readability.","section":"Section 8 (Lemma 8.4)"}],"recommendation":"major_revision","confidential_remarks":"The main risk is the dependence on several 2026 preprints, and in particular on the author's own [12] in the equality proof. Independent verification of [10] and [11] would substantially increase confidence. The Proposition 6.1 gap seems repairable, so I do not recommend rejection on that ground alone."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The takeaway: this paper deserves a serious referee, but with a clear caveat. It proves a new inequality E(G) ≥ r(G) + d̄(G) − 1 for all simple graphs of order n≥5, and that inequality is strong enough to settle five open conjectures on nonsingular graph energy. That is a real result, not a repackaging. The genuinely new pieces are a deletion-monotonicity lemma for E − d̄, a determinant-variance reduction that turns integrality of det into a scalar inequality, and an all-analytic proof of that scalar bound. The equality characterization (complete graphs and perfect matchings only) is clean, and the Zagreb-index and structural corollaries are natural.\n\nWhere it is vulnerable: the p≥3 part rests on three external results — [10]'s E ≥ 2(n−α), [11]'s square-energy bound min{s+,s−} ≥ n−1, and [2]'s two-range nonsingular bounds. None are verified in the paper. If any of those has a gap, Theorem 1.1 is not established. That is the main reason to be conditional rather than accepting outright. The equality characterization also depends on the author's own [12], though only in the equality case.\n\nThere is also a real, though repairable, gap in Proposition 6.1. After showing G is bipartite, the proof considers a bipartite cyclic component of the complement and derives a strict energy bound, then concludes every component of the complement is acyclic. That conclusion does not follow: non-bipartite cyclic components were not ruled out. The fix is trivial — α(G)≤2 plus bipartiteness forces both bipartition classes to have size at most 2, so n≤4, contradicting n≥6 — but as written the p=2 case is not fully proved.\n\nWould I cite this? Yes, if the external pillars hold up; it is the right inequality to cite for these conjectures. For a reading group, it would be a useful case study in how far determinant-integrality arguments can go, though I would want to check the cited preprints first. Send it to peer review — the main theorem is important enough that a referee should spend time on it, and the gaps I found are fixable.","headline":"Proves a strong new rank-average-degree inequality that settles five conjectures; sound modulo unverified external preprints and one repairable gap in the p=2 case.","tokens_in":19990,"tokens_out":3569,"would_cite":true,"duration_ms":33412,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"pith_extraction":{"msc":["05C50"],"pacs":[],"model":"deepseek-v4-flash","headline":"Every finite simple graph of order at least five has energy at least its adjacency rank plus its average degree minus one, with equality only for complete graphs and perfect matchings.","keywords":["graph energy","adjacency rank","average degree","nonsingular graph","first Zagreb index","extremal graphs","spectral bounds","determinant-variance reduction"],"falsifier":"A concrete calculation: the path $P_4$ fails the inequality because $\\mathcal{E}(P_4)=2\\sqrt5\\approx 4.472$ is just below $4.5 = r(\\bar d)(P_4)-1$, which is why the theorem requires $n\\ge 5$; a computer search over all graphs on five and six vertices would verify or refute the claimed universal bound in the smallest cases.","tokens_in":19036,"feed_emoji":"⚡","tokens_out":5768,"duration_ms":47357,"temperature":0.7,"pith_summary":"This paper establishes a single lower bound for the energy of any finite simple graph $G$ of order $n\\ge 5$: the energy $\\mathcal{E}(G)$ is at least $r(G)+\\bar d(G)-1$, where $r(G)$ is the rank of the adjacency matrix and $\\bar d(G)$ is the average degree. It also characterizes the extremal graphs: equality holds exactly for complete graphs $K_n$ and, when $n$ is even, for perfect matchings $\\frac{n}{2}K_2$. If the theorem is right, five previously conjectured lower bounds for the energy of nonsingular graphs follow immediately in their stated ranges, including bounds involving maximum and minimum degree, the geometric mean of average degree and $n-1$, and the first Zagreb index. The proof reduces the nonsingular case to a determinant-variance scalar inequality and then extends to singular graphs by a deletion monotonicity lemma.","feed_headline":"Rank and average degree bound settles five graph-energy conjectures","feed_subtitle":"A single inequality, E ≥ rank + average degree − 1, unifies five lower bounds and pins down equality cases.","key_machinery":"The load-bearing objects are the vertex-energy deletion lemma, which shows that $\\mathcal{E}(G)-\\bar d(G)$ does not increase when a vertex is removed while the adjacency rank is held fixed; the determinant-integrality principle, which says that the pseudodeterminant of an integral adjacency matrix of rank $r$ is a nonzero integer, hence the product of the nonzero eigenvalues has absolute value at least $1$; and a determinant-variance reduction that turns the assumed failure of the bound into a purely analytic inequality in parameters $k,a,w$. The resulting scalar inequality $F(k,a,w)<0$ is then proved by elementary calculus, with the two-positive-eigenvalue case handled separately through a sparse-complement argument.","core_discovery":"The central claim is that for every finite simple graph with $n\\ge 5$ vertices, the energy $\\mathcal{E}(G)=\\sum_i |\\lambda_i|$ is bounded below by the adjacency rank plus the average degree minus one. This is a unification: for nonsingular graphs, whose adjacency matrix is invertible, the rank equals $n$, giving $\\mathcal{E}(G)\\ge n-1+\\bar d(G)$, and since $n-1+\\bar d(G)$ dominates both $\\Delta(G)+\\delta(G)$ and $2\\sqrt{\\bar d(G)(n-1)}$, the older conjectures follow. The author further claims that equality occurs only for complete graphs and, in even orders, for perfect matchings, and uses this to prove that certain Zagreb-index inequalities hold with equality only for $K_n$.","pith_inferences":["Beyond the paper: if the rank bound is accepted as a master inequality, then any future improvement in lower bounds for the adjacency rank, for example via independence or domination parameters, automatically becomes an energy bound through Theorem 1.1.","Beyond the paper: the determinant-variance reduction suggests a template for converting a nonsingular energy lower bound whose proof uses $|\\det A|\\ge 1$ into a rank bound by deletion monotonicity; applying the same template to other spectral invariants is a testable scheme.","Beyond the paper: the sharp order restriction $n\\ge 5$ raises the question of a corrected statement for $n=4$; the paper notes that $P_4$ fails, so enumerating all four-vertex graphs would show whether a simple exception list exists."],"forward_implications":["Every nonsingular graph of order at least five satisfies $\\mathcal{E}(G)\\ge n-1+\\bar d(G)$, settling the Akbari-Dabirian-Ghasemi conjecture.","Every nonsingular graph satisfies $\\mathcal{E}(G)\\ge \\Delta(G)+\\delta(G)$ and $\\mathcal{E}(G)\\ge 2\\sqrt{\\bar d(G)(n-1)}$, with equality only for $K_n$.","For nonsingular graphs, $\\mathcal{E}(G)\\ge M_1(G)/m$ and $\\mathcal{E}(G)\\ge M_1(G)/(2m)+2m/n$, with equality only for $K_n$.","Combining Theorem 1.1 with rank bounds yields energy lower bounds in terms of clique number, induced matching number, total domination number, and diameter.","The equality characterization is sharp: the only graphs attaining equality in the rank-average-degree bound are complete graphs and perfect matchings."],"supporting_citations":[{"why":"Supplies the two-range nonsingular energy estimates used to restrict counterexamples and the conjectured bound $n-1+\\bar d(G)$.","marker":"[2]"},{"why":"Supplies the lower bound $\\mathcal{E}(G)\\ge 2(n-\\alpha(G))$ used for inertia estimates and base cases.","marker":"[10]"},{"why":"Supplies the lower bound $\\min\\{s_+(G),s_-(G)\\}\\ge n-1$ for connected graphs, used in the determinant-variance reduction.","marker":"[11]"},{"why":"Supplies the author's prior classification of equality in $\\mathcal{E}(G)=2(n-\\alpha(G))$, used in the equality characterization.","marker":"[12]"},{"why":"Supplies de Caen's degree-square inequality used to derive the Zagreb-index corollaries.","marker":"[7]"},{"why":"Introduces graph energy as the sum of absolute adjacency eigenvalues, the quantity being bounded.","marker":"[8]"},{"why":"Supplies the rank-based upper bound on total domination number used in the structural corollaries.","marker":"[1]"}],"fun_headline_variants":["Rank-average degree inequality settles five graph-energy conjectures","Graph energy bound: rank plus average degree minus one unifies five results","One inequality proves five conjectures for graph energy","New graph energy inequality confirms five lower-bound conjectures","Rank and average degree inequality resolves five graph energy open cases"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof leans on three previously established results about graph spectra that this paper does not re-derive; if any of those results is wrong or incomplete, the main inequality is not established.","fun_headline_variants_meta":{"raw":{"variants":["Rank-average degree inequality settles five graph-energy conjectures","Graph energy bound: rank plus average degree minus one unifies five results","One inequality proves five conjectures for graph energy","New graph energy inequality confirms five lower-bound conjectures","Rank and average degree inequality resolves five graph energy open cases"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000663,"raw_usage":{"total_tokens":3030,"prompt_tokens":948,"completion_tokens":2082,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":564,"completion_tokens_details":{"reasoning_tokens":2002}},"tokens_in":564,"tokens_out":2082,"duration_ms":12345,"temperature":1.0,"reasoning_tokens":2002,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-27T18:38:07.749903+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"A concrete calculation: the path $P_4$ fails the inequality because $\\mathcal{E}(P_4)=2\\sqrt5\\approx 4.472$ is just below $4.5 = r(\\bar d)(P_4)-1$, which is why the theorem requires $n\\ge 5$; a computer search over all graphs on five and six vertices would verify or refute the claimed universal bound in the smallest cases.","supporting_citations":[{"cited_title":"A lower bound of the energy of non-singular graphs in terms of average degree","cited_arxiv_id":"2207.04599","evidence_quote":"Supplies the two-range nonsingular energy estimates used to restrict counterexamples and the conjectured bound $n-1+\\bar d(G)$."},{"cited_title":"The positive and negative square-energy conjecture","cited_arxiv_id":"2607.18031","evidence_quote":"Supplies the lower bound $\\min\\{s_+(G),s_-(G)\\}\\ge n-1$ for connected graphs, used in the determinant-variance reduction."},{"cited_title":"Extremal Graphs for the Energy-Independence Number Inequality","cited_arxiv_id":"2608.04367","evidence_quote":"Supplies the author's prior classification of equality in $\\mathcal{E}(G)=2(n-\\alpha(G))$, used in the equality characterization."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies de Caen's degree-square inequality used to derive the Zagreb-index corollaries."},{"cited_title":"Math.-Statist","cited_arxiv_id":null,"evidence_quote":"Introduces graph energy as the sum of absolute adjacency eigenvalues, the quantity being bounded."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Supplies the rank-based upper bound on total domination number used in the structural corollaries."}],"review_version":1}