{"id":"7090d6b5-5671-4905-8ad0-3c4df67dfcf5","arxiv_id":"1908.05692","paper_version":3,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"For J_n(α)⊕βI_m, the k-th higher rank numerical range is always one of seven explicit forms: a disk, a disk plus a cone, a segment, a tilted segment, a singleton, or empty, depending on n, m, k, and |β−α|.","lead":"This paper computes the exact shape of the higher rank numerical range for every matrix made of one Jordan block plus a scalar block. The result condenses a convex set that is usually hard to visualize into a small table of disks, cones, segments, and empty sets.","discovery_kind":"new_application","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Theorem 3.7 is ill-posed when β=α because ψ=arg(β−α) is undefined, and the proof of the degenerate cases relies on δ_k=π/2 rather than δ_k=0; the fix is simple but the statement as written does not cover Remark 4.3's applications.","rationale":"The reader's weakest_assumption names the external Li–Sze theorem as the primary fragility. In my reading, that theorem is an established published result, so relying on it is not itself a correctness risk; the genuinely load-bearing issue is the internal β=α gap. The reader does identify this as a 'smaller fragility', so we agree on the substance. The gap is real: the theorem statement is undefined for β=α, and the proof of the degenerate cases uses δ_k=π/2 when the paper's own definition of δ_k gives δ_k=0 in the β=α case. However, direct application of Li–Sze with λ_k≡0 in the relevant cases confirms the table's predictions for β=α, so the flaw does not appear to produce a wrong classification; it should be repaired by stating a convention for ψ when β=α and rechecking cases 4, 6, and 7 separately. For this reason I would keep the reader's CONDITIONAL verdict rather than strengthening or overturning it.","tokens_in":17289,"tokens_out":23284,"duration_ms":219285,"concrete_test":"For T=J_3(0)⊕0_2, apply Theorem 1.1 directly: sample a fine grid of θ, compute λ_2(Re(e^{iθ}T)), and verify the intersection of halfplanes is exactly {0}. Repeat for T=J_5(0) with k=3. Then rerun the proof of Proposition 3.6 case 4 using the definition's δ_k=0 branch instead of δ_k=π/2 and check that the same conclusion follows. If the halfplane intersection is {0} in both cases, the β=α classification is confirmed and the concern is limited to the undefined ψ and an incorrect δ_k in the proof.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central classification is Theorem 3.7, which states 'Put ψ = arg(β−α)' with no exclusion of β=α. For β=α this argument is undefined, and the paper's own applications, including Remark 4.3 and Corollary 4.9, explicitly use the β=α case. The difficulty is not merely typographical: in Section 2, δ_k is defined so that when β=α the 'otherwise' branch gives δ_k=0, whereas the proof of Proposition 3.6 case 4 says 'now cos φ_k = 0, so δ_k = π/2'. That claim is false when r=|β−α|=0, and the same issue affects the proof of the nearby degenerate cases 6 and 7. The classification itself appears salvageable: for β=α, a direct application of Li–Sze with λ_k(Re(e^{iθ}T)) constant equal to 0 in the relevant cases gives Λ_k={α}, matching the table. But as written, Theorem 3.7 is not a well-formed statement for a case the paper explicitly relies on, and the supplied proof does not cover that case. This is a genuine rigor gap, though a localized and patchable one.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper gives a complete explicit description of the higher-rank numerical ranges Λ_k(J_n(α)⊕β I_m) for all k=1,...,n+m. The method applies the Li–Sze half-plane characterization: Λ_k(T) is the intersection of the half-planes Re(e^{iθ}μ)≤λ_k(Re(e^{iθ}T)). The authors reduce the eigenvalue computation to the known spectrum of Re(J_n(0)) and to the scalar block, and then assemble the resulting inequalities into a seven-case table (Theorem 3.7). The paper closes with applications to extremal examples of empty and singleton higher-rank numerical ranges, to corner behaviour, and to a unitary-equivalence rigidity statement (Corollary 4.9).","tokens_in":17529,"tokens_out":15713,"duration_ms":133957,"significance":"If the classification is correct, it is a genuine contribution: it is one of the very few complete higher-rank numerical range computations for a non-normal family, and the shapes that occur (disks, tangent cones, segments, singletons, empty sets) are substantially richer than in earlier examples such as powers of shifts or elliptic Toeplitz matrices. The applications are concrete and interesting, particularly the non-normal examples attaining emptiness for k as low as (n+m)/3+1 and the discussion of the failure of an Anderson-type theorem for k>1. The derivation is structurally sound: it relies on the external Li–Sze theorem and on exact eigenvalue formulas, with no fitted constants or hidden normalizations. The main weaknesses are localized degenerate-case gaps, described below, which do not appear to affect the final formulas but do affect the completeness of the proof as written.","major_comments":[{"comment":"The statement of Theorem 3.7 begins 'Put ψ = arg(β−α)' without excluding β=α; for β=α this is undefined. This is not merely cosmetic: the degenerate cases in Proposition 3.6 (cases 4, 6 and 7) are proved by asserting that cos φ_k=0 forces δ_k=π/2, but according to the definition in Section 2, δ_k=0 when β=α. The paper's own applications, including Remark 4.3 and Corollary 4.9, explicitly use β=α. A direct Li–Sze computation gives the same listed formulas in the β=α case, so the classification appears salvageable, but as stated Theorem 3.7 is not a well-formed statement for an important case the paper relies on, and the supplied proof does not establish that case.","section":"Theorem 3.7 and Proposition 3.6"},{"comment":"Lemma 2.3 is stated under the hypothesis 0<δ_k<π, and the cone R_{r,k} is defined using cot δ_k. However, δ_k attains the values 0 and π inside the parameter range of the theorem: δ_k=0 when β=α (or more generally when the 'otherwise' branch applies), and δ_k=π when k>(n+1)/2 and |β−α|=−cos φ_k>0, e.g., n=5, m=3, k=4, |β−α|=1/2. Proposition 3.2 and the proof of Proposition 3.6 cases 4–7 invoke Lemma 2.3 in precisely these regimes, so the proof does not cover them. The final claims are true (the cone degenerates to {r} in the relevant cases), but a separate argument or an extended statement of Lemma 2.3 is needed for these boundary values.","section":"Section 2, Lemma 2.3 and definition of R_{r,k}"}],"minor_comments":[{"comment":"The proposition statement reads 'Let T^0_{α,β}=e^{iψ}J_n(0)⊕|β−α|I_m', but throughout Section 3 and in the table inside Proposition 3.6 the correct expression is e^{-iψ}J_n(0)⊕|β−α|I_m; this is a sign typo.","section":"Proposition 3.6"},{"comment":"The description of the drawing script says 'the lines x cosθ−y cosθ = λ_k(T)'; this should be 'x cosθ−y sinθ'.","section":"Examples 3.8"},{"comment":"The text says that ~E_k consists of those μ with non-negative real part; since D_k=[π/2,3π/2], the complement consists of angles in (-π/2,π/2), which give positive real part. The point 0 also belongs to the set and needs to be accounted for separately, for instance by the definition of ~D_k∩B_0(0).","section":"Proposition 3.6, proof of case 4"},{"comment":"The reference is given as 'J.-L. de Lagrange'; the usual name is J.-L. Lagrange, and the entry should be checked for consistency with the citation in the text.","section":"Reference [19]"}],"recommendation":"major_revision","confidential_remarks":"The degenerate-case gaps are real but localized and patchable; my reading is that the main mathematical content is correct for β≠α and the β=α formulas are also correct. I would not reject on these grounds, but the manuscript needs a careful revision of the statement of Theorem 3.7 and of the proofs at the boundary values δ_k=0 and δ_k=π before it is suitable for publication."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The main thing to know: this paper gives the full classification of Λ_k(J_n(α)⊕βI_m), and Theorem 3.7 is genuinely new for this family. The method is standard—Li–Sze half-plane characterization plus explicit eigenvalues of Re(e^{iθ}J_n(0))—but it is applied cleanly, and the payoff is a set of concrete extreme examples that the higher-rank numerical range literature has been missing. If you work on numerical ranges, this is worth a careful read.\n\nWhat it does well: the case analysis in Section 3 is detailed and internally consistent for β≠α. The authors correctly note that previous explicit computations covered shift powers and 2×2 sums, but not this Jordan-plus-scalar family. The examples in Section 4 are useful: non-normal instances where Λ_k is empty at the threshold k = (n+m)/3+1, a natural counterexample to any Anderson-type theorem for k>1, and explicit projections witnessing Λ_{m+ℓ+1}(J_{2ℓ+1}(0)⊕0_m) = {0}. The citations look appropriate; no fitted constants or self-citation problems.\n\nThe soft spot is real, and the stress-test note has it right. Theorem 3.7 says \"put ψ = arg(β−α)\" with no exclusion, and Remark 4.3 uses the theorem with β=α (e.g., J_3(0)⊕0_m). For β=α, ψ is undefined, and the proof of case 4 in Proposition 3.6 claims δ_k = π/2 when cos φ_k = 0; under the paper's own definition δ_k = 0 when β=α. So the proof does not cover a case the statement and later examples rely on. This is not a fundamental flaw: for β=α the table still gives correct answers (a direct Li–Sze computation confirms case 4 yields {α}), and the β≠α classification is unaffected. But it needs to be stated carefully—either exclude β=α in the theorem and handle it separately, or define ψ arbitrarily and justify the formulas by a limiting argument.\n\nA few minor typos: the header of Proposition 3.6 says e^{iψ} where the body uses e^{-iψ}, and \"x cosθ−y cosθ\" in Examples 3.8 should presumably read \"x cosθ−y sinθ.\" These are trivial.\n\nWho it's for: people interested in higher-rank numerical ranges and concrete examples of their extreme behavior. It deserves a serious referee: the main classification is solid for β≠α, the examples are citable, and the gap is patchable. I'd suggest conditional acceptance with the β=α issue addressed.","headline":"A solid, specialized classification of higher-rank numerical ranges for a natural Jordan-plus-scalar family, with a real but localized and patchable gap in the β=α case.","tokens_in":18060,"tokens_out":5025,"would_cite":true,"duration_ms":43145,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["15A60","15B05"],"pacs":[],"model":"deepseek-v4-flash","headline":"The higher rank numerical range of a Jordan block plus a scalar matrix is fully described by an explicit seven-case table.","keywords":["higher rank numerical range","Jordan block","Jordan matrix","half-plane characterization","matrix compression","non-normal matrix","direct sum","numerical range"],"falsifier":"Take $T=J_4(0)\\oplus I_4$ and $k=2$. The table predicts $\\Lambda_2(T)=B_{\\cos(2\\pi/5)}(0)\\cup(\\widetilde E_2\\cap R_{1,2})$. Enumerating all rank-2 projections $P$ and computing the set of $\\lambda$ for which $PTP=\\lambda P$ would settle this case: any point outside the predicted union, or any predicted point that fails to appear, falsifies the classification.","tokens_in":17067,"feed_emoji":"📐","tokens_out":11851,"duration_ms":104196,"temperature":0.7,"pith_summary":"The paper gives a complete classification of the higher rank numerical ranges of matrices of the form $J_n(\\alpha)\\oplus \\beta I_m$, where $J_n(\\alpha)$ is a single Jordan block with eigenvalue $\\alpha$ and $\\beta I_m$ is a scalar multiple of the identity. Using the half-plane characterization of $\\Lambda_k$ together with an explicit eigenvalue computation for the real part of a rotated Jordan block, the authors show that every such range is one of seven shapes: a disk, a disk with a tangent wedge, the same wedge cut by a second disk, a segment, a scaled segment, a singleton, or the empty set. The deciding data are the rank $k$, the block sizes $n$ and $m$, and the distance $|\\beta-\\alpha|$; the thresholds that separate the cases are $\\cos(k\\pi/(n+1))$ and $\\cos((k-m)\\pi/(n+1))$. Because the list is complete and explicit, these matrices become a concrete laboratory for extreme behaviour of higher rank numerical ranges: emptiness can occur at ranks as low as $N/3+1$, nonemptiness can persist surprisingly close to the top rank, corners can appear that are not eigenvalues, and the full family of ranges determines the matrix.","feed_headline":"Seven shapes capture every numerical range in this matrix family","feed_subtitle":"Complete table: disks, wedges, segments, singletons, or empty, decided by k, n, m, and the eigenvalue gap.","key_machinery":"The machinery is the half-plane characterization of $\\Lambda_k$: for every $k$, $\\Lambda_k(T)$ is the set of $\\mu$ such that $\\operatorname{Re}(e^{i\\theta}\\mu)\\le \\lambda_k(\\operatorname{Re}(e^{i\\theta}T))$ for all $\\theta\\in[0,2\\pi]$. The paper combines this with a unitary-conjugation argument showing that the eigenvalues of $\\operatorname{Re}(e^{i\\theta}e^{-i\\psi}J_n(0))$ are exactly $\\cos(j\\pi/(n+1))$, $j=1,\\dots,n$, independent of $\\theta$ and $\\psi$. Thus $\\lambda_k(\\operatorname{Re}(e^{i\\theta}T^0))$ is an explicitly known piecewise function of $\\theta$ with pieces $\\cos\\phi_k$, $\\cos\\psi_{k,m}$, and $|\\beta-\\alpha|\\cos\\theta$. The angles $\\phi_k=k\\pi/(n+1)$, $\\psi_{k,m}=(k-m)\\pi/(n+1)$, the auxiliary angles $\\delta_k$ and $\\eta_{k,m}$, the sectors $D_k$ and $C_{k,m}$, and the cones $R_{r,k}$ serve as bookkeeping for the $\\theta$-intervals on which each piece governs; the final shapes are obtained by intersecting the corresponding half-planes.","core_discovery":"The central claim is Theorem 3.7. For $T=J_n(\\alpha)\\oplus\\beta I_m$ and $1\\le k\\le n+m$, $\\Lambda_k(T)$ is exactly one of the following: the disk $B_{\\cos\\phi_k}(\\alpha)$; that disk together with the wedge $\\widetilde E^\\psi_k\\cap R^\\psi_{|\\beta-\\alpha|,k}$; the same union with an additional cut $B_{\\cos\\psi_{k,m}}(\\alpha)$; the segment $[\\alpha,\\beta]$; the segment $\\{\\alpha+t(\\beta-\\alpha)\\cos\\eta_{k,m}: t\\in[0,1]\\}$; the singleton $\\{\\beta\\}$; or the empty set. Which case occurs is determined by comparing $k$ with $n/2$, $k$ with $m$, and $|\\beta-\\alpha|$ with $\\cos(k\\pi/(n+1))$ and $\\cos((k-m)\\pi/(n+1))$. The proof reduces $T$ by translation and rotation to $T^0=e^{-i\\psi}J_n(0)\\oplus|\\beta-\\alpha|I_m$; there the eigenvalues of $\\operatorname{Re}(e^{i\\theta}T^0)$ are $|\\beta-\\alpha|\\cos\\theta$ with multiplicity $m$ and $\\cos(j\\pi/(n+1))$, $j=1,\\dots,n$, so the $k$-th eigenvalue is a piecewise function with three possible values. Feeding this into the half-plane characterization turns $\\Lambda_k$ into an intersection of half-planes, one for each $\\theta$, and the paper evaluates that intersection explicitly in every regime.","pith_inferences":["The same half-plane strategy could plausibly handle direct sums of several Jordan blocks with a common eigenvalue plus a scalar block, since the spectrum of the real part would still be an ordered union of cosine sequences; the main difficulty would be combinatorial rather than conceptual.","The explicit projections exhibited for the singleton cases give a constructive membership certificate for those extreme ranges, which could be useful in compression problems where one must actually build the projection that witnesses $\\lambda\\in\\Lambda_k$.","The table can be read as an $O(1)$ decision procedure for emptiness and membership in this family; such a procedure could serve as a test oracle for conjectures about higher rank numerical ranges of more general matrices.","At the threshold $|\\beta-\\alpha|=\\cos(k\\pi/(n+1))$ the wedge term collapses into the disk, a phase transition in shape that may be a general phenomenon for sparse non-normal matrices."],"forward_implications":["For $k\\le n/2$ the range is a disk centered at $\\alpha$ whenever $|\\beta-\\alpha|\\le\\cos(k\\pi/(n+1))$; when the separation is larger it is a disk with a wedge attached along tangent lines.","For $k>n/2$ the range degenerates into a segment, a scaled segment, a singleton, or the empty set; in particular it is empty whenever $k>m$ and $|\\beta-\\alpha|>\\cos((k-m)\\pi/(n+1))$.","The class supplies explicit non-normal matrices, with explicit compressing projections, for which $\\Lambda_k$ is nonempty unusually close to the top rank for $n=2,3$ and empty for $n\\ge 4$ at rank $N-1$.","Corners of $\\Lambda_k$ need not be eigenvalues and can appear or disappear as $k$ crosses $m$, so the classical circle-boundary theorem for numerical ranges has no higher-rank analogue.","If two matrices of this form have identical $\\Lambda_k$ for every $k$, they are equal; the full hierarchy of higher rank numerical ranges is a complete invariant on this class."],"supporting_citations":[{"why":"Supplies the half-plane characterization $\\Lambda_k(T)=\\{\\mu:\\operatorname{Re}(e^{i\\theta}\\mu)\\le\\lambda_k(\\operatorname{Re}(e^{i\\theta}T))\\}$ that is the starting point of every computation.","marker":"[6]"},{"why":"Defines the higher-rank numerical range and provides the basic property that $\\Lambda_k$ is at most a singleton when $k>n/2$.","marker":"[12]"},{"why":"Gives the unitary-conjugation argument identifying the eigenvalues of $\\operatorname{Re}(e^{i\\theta}J_n(0))$ as $\\cos(j\\pi/(n+1))$.","marker":"[18]"},{"why":"Provides the corner characterization used to explain appearance and disappearance of corners in the computed ranges.","marker":"[20]"},{"why":"Supplies the $k<n/3+1$ nonemptiness threshold and normal examples whose non-normal counterparts are realized here.","marker":"[22]"}],"fun_headline_variants":["Seven shapes fully classify higher rank numerical ranges","Numerical ranges: disk, wedge, segment, singleton, or empty","Jordan-like matrices: exactly seven possible numerical range shapes","Higher rank ranges for J⊕βI: seven shapes, fully decided by k,n,m","Every higher rank numerical range here is one of seven explicit shapes"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The load-bearing premise is the external half-plane characterization of $\\Lambda_k(T)$ as the intersection over $\\theta$ of the half-planes $\\operatorname{Re}(e^{i\\theta}\\mu)\\le\\lambda_k(\\operatorname{Re}(e^{i\\theta}T))$; if that equality failed for arbitrary non-normal matrices the whole table collapses, and a smaller fragility is that the statement sets $\\psi=\\arg(\\beta-\\alpha)$ without excluding $\\beta=\\alpha$ even though several proof steps implicitly assume $\\beta\\ne\\alpha$.","fun_headline_variants_meta":{"raw":{"variants":["Seven shapes fully classify higher rank numerical ranges","Numerical ranges: disk, wedge, segment, singleton, or empty","Jordan-like matrices: exactly seven possible numerical range shapes","Higher rank ranges for J⊕βI: seven shapes, fully decided by k,n,m","Every higher rank numerical range here is one of seven explicit shapes"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000404,"raw_usage":{"total_tokens":2093,"prompt_tokens":926,"completion_tokens":1167,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":542,"completion_tokens_details":{"reasoning_tokens":1079}},"tokens_in":542,"tokens_out":1167,"duration_ms":9719,"temperature":1.0,"reasoning_tokens":1079,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:08:10.884199+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Take $T=J_4(0)\\oplus I_4$ and $k=2$. The table predicts $\\Lambda_2(T)=B_{\\cos(2\\pi/5)}(0)\\cup(\\widetilde E_2\\cap R_{1,2})$. Enumerating all rank-2 projections $P$ and computing the set of $\\lambda$ for which $PTP=\\lambda P$ would settle this case: any point outside the predicted union, or any predicted point that fails to appear, falsifies the classification.","supporting_citations":[{"cited_title":"Li and N.-S","cited_arxiv_id":null,"evidence_quote":"Supplies the half-plane characterization $\\Lambda_k(T)=\\{\\mu:\\operatorname{Re}(e^{i\\theta}\\mu)\\le\\lambda_k(\\operatorname{Re}(e^{i\\theta}T))\\}$ that is the starting point of every computation."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Defines the higher-rank numerical range and provides the basic property that $\\Lambda_k$ is at most a singleton when $k>n/2$."},{"cited_title":"Haagerup and P","cited_arxiv_id":null,"evidence_quote":"Gives the unitary-conjugation argument identifying the eigenvalues of $\\operatorname{Re}(e^{i\\theta}J_n(0))$ as $\\cos(j\\pi/(n+1))$."},{"cited_title":"Chang, H.-L","cited_arxiv_id":null,"evidence_quote":"Provides the corner characterization used to explain appearance and disappearance of corners in the computed ranges."},{"cited_title":"Li, Y.-T","cited_arxiv_id":null,"evidence_quote":"Supplies the $k<n/3+1$ nonemptiness threshold and normal examples whose non-normal counterparts are realized here."}],"review_version":1}