{"id":"998b23a2-1157-4c3b-8be8-2c764961998f","arxiv_id":"2505.12769","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"A unital graph C*-algebra is operator norm stable exactly when a certain reachability subgraph is finite, and is residually finite-dimensional exactly when the graph is finite and no cycle has an entry.","lead":"The paper proves that many unital graph C*-algebras can be broken apart into simpler building blocks called amalgamated free products, and uses this to say exactly when they have certain approximation properties. It gives complete criteria for when such algebras are residually finite-dimensional and when they are operator norm stable.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Lemma 5.4 contains a false equality (ν_{n-1}ν_{n-1}^* = p_{s(ν_n)}); the written proof of the central vanishing lemma is invalid, though a local inequality repair appears to work.","rationale":"The central claim Theorem 5.14 is likely correct: the definition of \\tilde G is coherent, the E-finiteness step in Theorem 5.6 is justified by condition (4), the quotient ∏M_n/⊕M_n is finite (every isometry is unitary), and the RFD facts used in Lemma 5.10 are true, though they should be stated or cited. The reader's weakest_assumption about Lemma 5.6's E-finiteness is not the real soft spot; that step is sound. The actual defect is the false equality in Lemma 5.4, which is load-bearing because Lemma 5.4 feeds directly into Lemma 5.5, Claim 1 of Theorem 5.6, and Corollary 5.7, and hence into the reductions in Theorem 5.11 and Lemma 5.12. The repair is local and elementary, so the verdict should remain conditional rather than being strengthened to rejection. I partially agree with the reader because both of us locate the risk in the vanishing lemmas of Section 5, but we identify different specific steps.","tokens_in":18247,"tokens_out":61477,"duration_ms":634195,"concrete_test":"Construct the minimal example where the equality fails: vertices v0, v1, v2; edges a:v0→v1, b:v2→v1, c:v1→v2, and a loop d:v2→v2. The path (c,a) leads to the cycle d. In Lemma 5.4's notation, ν_2=c and ν_1=a, so the displayed equality a a* = p_{v1} is false because v1 also receives b. Verify that replacing it by a a* ≤ p_{v1} = c^*c still yields π(a)=0 from π(c)=0, and check that every later step in Lemma 5.4 uses only subprojection inequalities, not the invalid equality. If the repaired induction goes through, Lemma 5.4 and the derived vanishing results stand; if not, identify the first induction step where the chain breaks.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Lemma 5.4, which underpins Lemma 5.5, Claim 1 of Theorem 5.6, and Corollary 5.7, contains an assertion that is false as written. For a path ν=(ν_n,...,ν_1) leading to a cycle, the proof states ν_{n-1}ν_{n-1}^* = r(ν_{n-1}) = s(ν_n) = ν_n^*ν_n. In a graph C*-algebra, e e^* is only a subprojection of p_{r(e)}, not equal to p_{r(e)} unless r(e) receives exactly one edge. The CK relation is p_v = sum over all incoming edges, so the displayed equality fails whenever the vertex s(ν_n) has more than one incoming edge, a case not excluded by the hypotheses. Since Lemma 5.4 is used to show that all edges in G′ vanish in any representation into a finite C*-algebra, the written proof of the vanishing theorem used in both directions of Theorem 5.14 is not formally complete. The repair is immediate: replace the equality by the CK inequality e e^* ≤ p_{r(e)}; then π(ν_n^*ν_n)=0 forces π(p_{s(ν_n)})=0, hence π(ν_{n-1}ν_{n-1}^*)=0. The same backward induction then works. This is a genuine but localized gap, not a disproof of the central claim.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies unital graph C*-algebras and gives two structural results. First, it proves decomposition theorems (Theorems 3.2 and 3.3) expressing C*(G) as an amalgamated free product over a finite-dimensional algebra when the graph is split into two subgraphs with no edge from the second entering the first. These decompositions are then used to characterize residually finite-dimensionality: Theorem 4.3 states that C*(G) is RFD if and only if G is finite and no cycle has an entry. The main new result is Theorem 5.14: C*(G) is matricially semiprojective (operator norm stable) if and only if a canonically defined subgraph \\tilde G is finite, where \\tilde G is built from four reachability conditions relative to the subgraph G' of paths leading to cycles. The proof introduces a vanishing lemma for homomorphisms into finite C*-algebras and uses it to reduce matricial semiprojectivity of C*(G) to that of C*(\\tilde G), together with an induction argument for the converse.","tokens_in":18533,"tokens_out":50989,"duration_ms":479242,"significance":"If the main results are correct, this is a substantial contribution. The characterization of RFD graph C*-algebras in Theorem 4.3 is clean and complete for the unital case, and the characterization of matricial semiprojectivity in Theorem 5.14 is a genuinely new and nontrivial graph condition that answers a natural question left open by earlier work on semiprojectivity. The decomposition theorems are of independent interest and are used in a coherent way through the Li--Shen criterion for RFD amalgamated free products. The paper also gives an explicit and checkable graph-theoretic condition, which is a strength. However, the proofs of several key lemmas in Section 5 contain localized errors, and one lemma used in the induction for the main theorem has a gap that needs a different argument. These issues are repairable, and the central claims appear defensible, but the manuscript requires careful revision.","major_comments":[{"comment":"The proof of Lemma 5.4 contains the false equality ν_{n-1}ν_{n-1}^* = p_{s(ν_n)}. In a graph C*-algebra the CK relation only gives ν_{n-1}ν_{n-1}^* ≤ p_{r(ν_{n-1})} = p_{s(ν_n)}, with equality only when s(ν_n) receives exactly one edge. Since this lemma is used in both directions of Theorem 5.14, the written proof is incomplete. The repair is immediate: from π(ν_n^*ν_n)=0 one obtains π(p_{s(ν_n)})=0, hence π(ν_{n-1}ν_{n-1}^*)=0, and the backward induction goes through. Please correct the argument.","section":"Section 5, Lemma 5.4"},{"comment":"In the proof of Claim 2 the displayed equality \"0 ≠ π(e)π(e)^* = π(s(e))\" is false; the CK relation is e^*e = p_{s(e)}, not ee^* = p_{s(e)}. The intended argument works with π(e)^*π(e) = π(s(e)) and then uses the CK relation s(e) = Σ_{h∈E} hh^* to find e_1 with π(e_1)≠0, but as written this is an invalid step in the central vanishing lemma. This should be corrected.","section":"Section 5, Theorem 5.6, Claim 2"},{"comment":"The proof of Lemma 5.12 tries to rule out the case where p is the range of an infinite-multiplicity edge belonging to G'. It asserts that then \"any edge with range p lies in G'\", which is false when p is on a cycle: the incoming cycle edge is in \\tilde G and is not in G'. The claimed contradiction therefore does not follow, and the subsequent conclusion that there are only finitely many edges of G\\tilde G with range p is not justified in that case. This is load-bearing because it is used to obtain the CK sum relation p = Σ_{r(e)=p} ee^* in the proof of Lemma 5.12. The lemma may still be true, but it requires a different argument for vertices satisfying condition 2), for instance using Corollary 5.7 to show that the infinite entry edges vanish in any representation into the finite C*-algebra ΠM_n/⊕M_n.","section":"Section 5, Lemma 5.12"},{"comment":"The statement of Theorem 5.13 as printed says \"If \\tilde G is finite, then C*(\\tilde G) is matricially semiprojective.\" This is already Lemma 5.10, while the proof actually proves that C*(G) is matricially semiprojective under the assumption that \\tilde G is finite. As printed, the theorem does not provide the converse direction needed in Theorem 5.14. The statement should be corrected to \"If \\tilde G is finite, then C*(G) is matricially semiprojective.\"","section":"Section 5, Theorem 5.13"}],"minor_comments":[{"comment":"In the displayed decomposition of C*(\\tilde G), the second factor should be C*(H_2)(⊕ possibly C), not a second copy of C*(H_1).","section":"Section 5, Lemma 5.10"},{"comment":"The fact that an RFD C*-algebra with finitely many ideals is finite-dimensional is used without proof or citation. This is true, but it should be justified or referenced, especially because it is essential for showing that G_2 is finite.","section":"Section 4, Theorem 4.3"},{"comment":"The implication \"Matricial semiprojectivity implies then that C*(\\tilde G) is RFD\" relies on the standard fact that a quasidiagonal C*-algebra embeds into ΠM_n/⊕M_n, and that an MSP lift of this embedding gives an embedding into ΠM_n. This should be stated explicitly or cited.","section":"Section 5, Lemma 5.10"},{"comment":"The cycle constructed in the final paragraph has an off-by-one indexing: if s(e_N)=s(e_i), the cycle should be (e_N,...,e_{i+1}) (or the indices chosen so that s(e_N)=s(e_{i-1})). The idea is clear, but the indexing should be fixed.","section":"Section 5, Theorem 5.6, final paragraph"},{"comment":"The notation u_i^*u_i = \\tilde\\psi(\\bar f_i^*\\bar f_i) is not meaningful because f_i is not an edge of H after removal; it should be u_i^*u_i = \\tilde\\psi(\\overline{s(f_i)}), the lift of the source projection.","section":"Section 5, Theorem 5.13, Claim 2"},{"comment":"The assertion \"Since there is no edge of G_2 entering G_1, {p_β}≠∅\" is false when G_2 has no edges, e.g., when G is a single cycle and G_2 is the same vertex set with no edges. The decomposition still holds in this case, but the argument should treat it explicitly or remove the claim.","section":"Section 3, Theorem 3.2"}],"recommendation":"major_revision","confidential_remarks":"The main results are likely correct and the paper is within scope for math.OA, but Section 5 needs careful revision: the false equality in Lemma 5.4, the false displayed equality in Theorem 5.6 Claim 2, and the flawed contradiction in Lemma 5.12 all affect load-bearing steps. I would check that the proof of Lemma 5.12 is rewritten to handle vertices on cycles with infinite entries, and that Theorem 5.13 is restated correctly. The paper's contribution is significant; the issues are localized and repairable."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The headline: this paper actually delivers. The amalgamated free product decomposition (Theorems 3.2 and 3.3) is new, the proof via universal properties is clean, and the paper uses it to give complete characterizations: C*(G) is RFD iff G is finite with no cycle entry (Theorem 4.3), and matricially semiprojective iff the subgraph G-tilde is finite (Theorem 5.14). These were open for the class, and the results are likely correct.\n\nThe RFD proof is a nice use of Li-Shen: decompose the graph into cycles plus a forest, embed the pieces into matrix algebras over a common finite-dimensional amalgam, use a dense family of circle representations for the cycles, then apply Li-Shen. The Section 5 machinery is dense but mostly careful, and the induction in Theorem 5.13 is a clever way to handle infinite multiplicities. The citation pattern is sensible: Raeburn-Tomforde for graph algebra basics, Li-Shen for RFD of amalgams, Eilers-Katsura for semiprojectivity.\n\nThere are three soft spots, none of which sink the central claims. First, two supporting facts are used without proof or citation: an RFD C*-algebra with finitely many ideals is finite-dimensional, and quasidiagonal plus matricially semiprojective implies RFD. Both are plausible and probably known, but the authors should provide references or short proofs.\n\nSecond, and more important, Lemma 5.4 contains a false equality. In the proof, for a path nu=(nu_n,...,nu_1) leading to a cycle, the paper writes nu_{n-1}nu_{n-1}^* = r(nu_{n-1}) = s(nu_n) = nu_n^*nu_n. In a graph C*-algebra, e e^* is only a subprojection of p_{r(e)}, equal to p_{r(e)} only when r(e) receives exactly one edge. The equality fails in general. The written proof of the vanishing lemma is therefore invalid. The repair is immediate: use e e^* <= p_{r(e)}. If pi(nu_n^*nu_n)=0 then pi(p_{s(nu_n)})=0, hence pi(nu_{n-1}nu_{n-1}^*)=0, and the same backward induction works. Since Lemma 5.4 underlies Lemma 5.5, Claim 1 of Theorem 5.6, and Corollary 5.7, this gap must be fixed explicitly, but it does not appear to falsify Theorem 5.14.\n\nThird, the minor typos and the terse proof of Lemma 5.1 are not serious.\n\nWho this is for: anyone working on graph C*-algebras, RFD, or semiprojectivity. The paper deserves a serious referee. I would send it to peer review and ask for a corrected proof of Lemma 5.4 plus the missing references.","headline":"A substantial paper that delivers a new decomposition tool and uses it to settle RFD and matricial semiprojectivity for unital graph C*-algebras, though one key lemma has a repairable proof gap.","tokens_in":19059,"tokens_out":3022,"would_cite":true,"duration_ms":27710,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["46L05"],"pacs":[],"model":"deepseek-v4-flash","headline":"Operator norm stability of a unital graph C*-algebra is equivalent to finiteness of a reachability subgraph \\tilde G, and residual finite-dimensionality is equivalent to the graph being finite with no cycle that has an entry.","keywords":["graph C*-algebras","amalgamated free products","residual finite-dimensionality","matricial semiprojectivity","operator norm stability","Cuntz-Krieger relations","reachability conditions","semiprojectivity"],"falsifier":"The theorem asserts two exact equivalences, so a single counterexample would settle it. Look for a graph with finitely many vertices whose $\\tilde G$ is infinite but whose $C^*(G)$ still lifts every approximate matrix representation; the paper predicts no such graph exists. Equally, a graph with finite $\\tilde G$ whose $C^*(G)$ admits an approximate matrix representation with no genuine lift would refute the other direction.","tokens_in":18046,"feed_emoji":"🕸️","tokens_out":11848,"duration_ms":108321,"temperature":0.7,"pith_summary":"The paper proves complete graph-theoretic characterizations of two approximation properties for unital graph C*-algebras. It shows that $C^*(G)$ is operator norm stable (equivalently, matricially semiprojective) exactly when a certain subgraph $\\tilde G$, built from paths leading to cycles and four reachability conditions, is finite. It also shows that $C^*(G)$ is residually finite-dimensional exactly when $G$ is finite and no cycle has an entry. The engine is a decomposition theorem: whenever a graph is split into two subgraphs with no edge of the second part entering the first, $C^*(G)$ is an amalgamated free product of the two subalgebras over a finite-dimensional C*-algebra. This reduces global approximation questions to questions about the pieces and a finite-dimensional amalgam.","feed_headline":"Finite subgraph decides operator-norm stability","feed_subtitle":"Both approximation properties are read off directly from the graph's shape","key_machinery":"The central mechanism is the decomposition theorem (Theorems 3.2 and 3.3). If $G = G_1 \\cup G_2$ is a graph with finitely many vertices, no edge of $G_2$ enters $G_1$, and the two pieces share $n$ vertices, then $C^*(G)$ is isomorphic to $(C^*(G_1) \\oplus \\mathbb{C}) *_{\\mathbb{C}^{n+1}} C^*(G_2)$ when $G_2$ contains all vertices, and to $(C^*(G_1) \\oplus \\mathbb{C}) *_{\\mathbb{C}^{n+2}} (C^*(G_2) \\oplus \\mathbb{C})$ otherwise. Amalgamating over a finite-dimensional C*-algebra lets the paper transfer the RFD criterion for amalgamated free products and the permanence of semiprojectivity under such products. For stability, the load-bearing object is the subgraph $\\tilde G$, whose edges are those reachable from a cycle not contained in $G'$, unreachable from $G'$, reachable from such an unreachable edge, or reachable from the common range of an infinite family of edges outside $G'$; the main theorem equates finiteness of $\\tilde G$ with matricial semiprojectivity of $C^*(G)$.","core_discovery":"The central claim is that for a graph with finitely many vertices, both the RFD property and operator norm stability are decided by the graph alone. Theorem 4.3 states that $C^*(G)$ is residually finite-dimensional if and only if $G$ is finite and no cycle has an entry; the proof decomposes $G$ into a union of disjoint cycles plus an acyclic forest, realizes $C^*(G)$ as an amalgamated free product over $\\mathbb{C}^{n+1}$ or $\\mathbb{C}^{n+2}$, and applies a known criterion for such products to be RFD. Theorem 5.14 states that $C^*(G)$ is matricially semiprojective if and only if the subgraph $\\tilde G$ is finite, where $\\tilde G$ is defined by four reachability conditions starting from the subgraph $G'$ of all paths that lead to cycles. The forward direction shows that any homomorphism from $C^*(G)$ into a finite C*-algebra kills every edge outside $\\tilde G$, so lifting problems descend to $C^*(\\tilde G)$; the reverse direction builds the required lifts directly by induction on vertices, successively removing edges that enter a distinguished vertex. The result is a complete, graph-readable description of operator norm stability for all unital graph C*-algebras.","pith_inferences":["The one-way no-entry split used in Theorems 3.2 and 3.3 is purely combinatorial, so the same decomposition should apply to any universal algebra generated by projections and partial isometries with local relations of Cuntz-Krieger type.","Condition (4) in the definition of $\\tilde G$ is a finiteness guard against edge multiplicity; one could implement the four conditions algorithmically and read operator norm stability directly off a finitely presented graph.","The contrast between the two theorems suggests a hierarchy: operator norm stability tolerates infinite acyclic parts of the graph, while residual finite-dimensionality requires the whole graph to be finite.","Remark 3.4 indicates where the unital assumption is essential, so the boundary of the theorem is already visible: extending to non-unital graph C*-algebras would need a genuinely non-unital amalgamated free product construction."],"forward_implications":["Every unital graph C*-algebra has a graph-only test for two properties: RFD means finite graph with no entered cycles, and operator norm stability means finite $\\tilde G$.","The decomposition theorem turns a one-way split of a graph into an amalgamated free product over a finite-dimensional algebra, so any property preserved by such products can be studied piece by piece.","Theorem 4.3 sharpens the known equivalence between \"no cycle has an entry\" and quasidiagonality: adding finiteness of the graph upgrades quasidiagonality to residual finite-dimensionality.","Since $C^*(\\tilde G)$ is matricially semiprojective exactly when $\\tilde G$ is finite, stability failures are visible inside $\\tilde G$; no larger portion of the graph can hide them.","The inductive proof of Theorem 5.13 gives a concrete construction that lifts approximate matrix representations whenever $\\tilde G$ is finite, so the characterization is effective rather than merely existential."],"supporting_citations":[{"why":"Supplies the structure theorem for finite acyclic graph C*-algebras as direct sums of matrix algebras, used to build the RFD embedding.","marker":"[10]"},{"why":"Supplies the identification of a single-cycle graph algebra with $M_n \\otimes C(\\mathbb{T})$ and the fact that acyclic graph algebras are AF, used in the RFD and $\\tilde G$-finiteness arguments.","marker":"[12]"},{"why":"Gives the criterion that an amalgamated free product over a finite-dimensional algebra is RFD exactly when the factors admit compatible matrix-algebra embeddings; this bridges the decomposition to RFD.","marker":"[8]"},{"why":"Establishes that \"no cycle has an entry\" is equivalent to quasidiagonality for graph C*-algebras, used to upgrade matricial semiprojectivity to RFD in Lemma 5.10.","marker":"[11]"},{"why":"Provides permanence of semiprojectivity under amalgamated free products over finite-dimensional subalgebras and the partial-isometry lifting result used in Lemmas 5.10 and 5.13.","marker":"[1]"},{"why":"Gives the fact about AF graph C*-algebras having finitely many ideals, used in Theorem 4.3 to force the forest part of an RFD graph to be finite.","marker":"[5]"}],"fun_headline_variants":["Graph shape alone decides RFD and operator stability","Finite subgraph condition determines both approximation properties","Unital graph C*-algebras: properties read from the graph","Two stability properties pinned down by graph's structure","Decomposition yields graph-only criteria for stability"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"Everything rests on Lemma 5.6, the claim that any homomorphism from $C^*(G)$ into a finite C*-algebra sends every edge outside $\\tilde G$ to zero; the finiteness argument in its proof depends on condition (4) of $\\tilde G$'s definition, which rules out an infinite family of edges entering a vertex from outside the path-to-cycle subgraph without forcing every edge reachable from their range into $\\tilde G$.","fun_headline_variants_meta":{"raw":{"variants":["Graph shape alone decides RFD and operator stability","Finite subgraph condition determines both approximation properties","Unital graph C*-algebras: properties read from the graph","Two stability properties pinned down by graph's structure","Decomposition yields graph-only criteria for stability"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000558,"raw_usage":{"total_tokens":2600,"prompt_tokens":835,"completion_tokens":1765,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":451,"completion_tokens_details":{"reasoning_tokens":1691}},"tokens_in":451,"tokens_out":1765,"duration_ms":11522,"temperature":1.0,"reasoning_tokens":1691,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-15T20:28:18.155798+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"The theorem asserts two exact equivalences, so a single counterexample would settle it. Look for a graph with finitely many vertices whose $\\tilde G$ is infinite but whose $C^*(G)$ still lifts every approximate matrix representation; the paper predicts no such graph exists. Equally, a graph with finite $\\tilde G$ whose $C^*(G)$ admits an approximate matrix representation with no genuine lift would refute the other direction.","supporting_citations":[{"cited_title":"Raeburn, Graph algebras, CBMS Regional Conference Series in Mathematics 103 Published for the Conference Board of the Mathematical Sciences, Washington D.C","cited_arxiv_id":null,"evidence_quote":"Supplies the structure theorem for finite acyclic graph C*-algebras as direct sums of matrix algebras, used to build the RFD embedding."},{"cited_title":"Tomforde, Graph C*-algebras, Theory, Technique, and Examples, 2011","cited_arxiv_id":null,"evidence_quote":"Supplies the identification of a single-cycle graph algebra with $M_n \\otimes C(\\mathbb{T})$ and the fact that acyclic graph algebras are AF, used in the RFD and $\\tilde G$-finiteness arguments."},{"cited_title":"Li and J","cited_arxiv_id":null,"evidence_quote":"Gives the criterion that an amalgamated free product over a finite-dimensional algebra is RFD exactly when the factors admit compatible matrix-algebra embeddings; this bridges the decomposition to RFD."},{"cited_title":"Schafhauser, AF-embeddings of graph C*-algebras, J","cited_arxiv_id":null,"evidence_quote":"Establishes that \"no cycle has an entry\" is equivalent to quasidiagonality for graph C*-algebras, used to upgrade matricial semiprojectivity to RFD in Lemma 5.10."},{"cited_title":"Blackadar, Shape theory for C*-algebras, Math","cited_arxiv_id":null,"evidence_quote":"Provides permanence of semiprojectivity under amalgamated free products over finite-dimensional subalgebras and the partial-isometry lifting result used in Lemmas 5.10 and 5.13."},{"cited_title":"Eilers, T","cited_arxiv_id":null,"evidence_quote":"Gives the fact about AF graph C*-algebras having finitely many ideals, used in Theorem 4.3 to force the forest part of an RFD graph to be finite."}],"review_version":1}