{"id":"27a323a6-fba1-4887-809b-3ba3b251dea0","arxiv_id":"1908.04499","paper_version":2,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":6.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"The paper proves a strictly stronger lower bound on the numerical radius of any nonzero Hilbert space operator, using the Crawford number of its square, and derives new bounds for operator matrices.","lead":"This mathematics paper finds tighter limits on the numerical radius of a linear operator, a quantity that measures how much a transformation can stretch vectors. It improves a classic 1963 lower bound by adding a term that is positive for many operators, and it derives new limits for block-shaped operator matrices.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: the central inequality (2.3) is sound, and the unproved equality used in its proof is true.","rationale":"The stress-test pass focused on the central claim, Theorem 2.8 inequality (2.3). The reader identified the unproved equality w([[0,T],[T,0]]) = w(T) as the weakest assumption. Investigation shows this equality is true and easily proven via unitary conjugation, and that (2.3) can be derived even more directly from Lemma 2.1 by a supremum argument. No counterexample to the inequality was found; it is consistent with known numerical-radius bounds. The secondary issues (Corollary 2.5's substitution wording and Theorem 2.10's reliance on a citation) are real but peripheral to the abstract's headline claim. Since the central argument holds up, the reader's conditional verdict is preserved; no adjustment is needed.","tokens_in":12680,"tokens_out":19594,"duration_ms":169824,"concrete_test":"Verify analytically that U* [[0,T],[T,0]] U = diag(T, -T) with U = (1/sqrt(2))[[I, I], [I, -I]]; since w(diag(T,-T)) = max(w(T), w(-T)) = w(T), the equality used in Theorem 2.8 is established. Optionally, also confirm (2.3) numerically for a sample of non-normal matrices with m(T^2) > 0, e.g., T = [[1,1],[0,2]].","verdict_should_be":"UNCHANGED","load_bearing_attack":"The proof of Theorem 2.8 (2.3) relies on the equality w([[0,T],[T,0]]) = w(T), which is stated without proof. This equality is in fact correct: for U = (1/sqrt(2))[[I, I], [I, -I]], one gets U* [[0,T],[T,0]] U = [[T, 0], [0, -T]], and since w(X ⊕ Y) = max(w(X), w(Y)) and w(-T) = w(T), the equality follows. Moreover, the bound itself can be derived directly from Lemma 2.1: for unit x, ||Tx||^2 + m(T^2) ≤ 2w(T)||Tx||, and taking the supremum over x immediately gives ||T||^2 + m(T^2) ≤ 2w(T)||T||, i.e., (2.3). Thus the central claim does not depend on any questionable bridge. The minor issues noted by the reader (Corollary 2.5 substitution wording, Theorem 2.10 reliance on a citation) are secondary and do not affect the main inequality.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper presents new upper and lower bounds for the numerical radius of bounded Hilbert space operators and of operator matrices. Section 2 generalizes the Bernau–Smithies parallelogram-law inequality to three operators (Lemma 2.2), then derives product bounds (Theorem 2.4, Corollary 2.5), lower bounds for 2×2 operator matrices (Theorem 2.7), and the main result Theorem 2.8: for every nonzero T, w(T) ≥ ‖T‖/2 + m(T²)/(2‖T‖), together with a companion inequality involving w(T²) and c(T). Section 3 gives upper bounds for n×n and 2×2 operator matrices, and Section 4 gives lower bounds, with numerical examples comparing the bounds to earlier ones.","tokens_in":12899,"tokens_out":27837,"duration_ms":258159,"significance":"The central inequality of Theorem 2.8 is a genuine, strict improvement of the classical lower bound w(T) ≥ ‖T‖/2 whenever m(T²) > 0, and it is derived from first principles without free parameters or circular use of the target result. The operator-matrix bounds are concrete and are illustrated by numerical examples. The proof of the main inequality is checkable and self-contained except for one standard unitary-invariance fact; the remaining concerns are local clarity issues rather than correctness problems.","major_comments":[],"minor_comments":[{"comment":"The proof uses the equality w([[0,T],[T,0]]) = w(T) without proof or reference. Since this equality is the bridge between Theorem 2.7 and the main bound (2.3), please add a one-line proof (e.g., via the unitary U = 2^{-1/2}[[I,I],[I,-I]]) or a precise citation.","section":"Theorem 2.8"},{"comment":"The sentence “Taking B = I, T = A and A = B” explains the first displayed inequality but not the second. The second inequality follows from the first inequality of Theorem 2.4 with (A,T,B) = (A*, B*, I), using w((AB)*)=w(AB) and m(A*B*)=m(BA*); please correct the derivation sentence or state the substitution explicitly.","section":"Corollary 2.5"},{"comment":"The proof of Theorem 2.10 is reduced to a citation for the necessity direction and the word “obvious” for sufficiency. Because this is a full characterization, please either quote the exact statement from [8, Th. 1.3-5] or provide a self-contained proof of the necessity direction.","section":"Theorem 2.10"},{"comment":"In the displayed computation of Re²(e^{iθ}T)+Im²(e^{iθ}T), the top-left block of the first RHS matrix is written as Re²(e^{iθ}T)+Im²(e^{iθ}T); it should be Re²(e^{iθ}A)+Im²(e^{iθ}A).","section":"Theorem 3.7"},{"comment":"The proof uses the power inequality w(S²) ≤ w(S)² for the numerical radius, for S = T and S = U*TU, without proof or citation. Please add a reference to this standard fact or include a brief proof.","section":"Theorem 4.1"}],"recommendation":"minor_revision","confidential_remarks":null},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The headline result is the lower bound w(T) ≥ ‖T‖/2 + m(T²)/(2‖T‖) for nonzero T on a Hilbert space. It is correct. I checked the proof and the stress-test note is right: the equality w([[0,T],[T,0]]) = w(T) is standard and true, and the bound can even be derived directly from Lemma 2.1. This is a genuine improvement over the classical 1963 inequality whenever m(T²) > 0, and it gives a clean necessary condition for equality in the old bound. The paper does this honestly, with a self-contained proof using the parallelogram law.\n\nWhat else is new: the authors prove several upper and lower bounds for operator matrices, compare them with existing bounds, and give explicit numerical examples. They are also transparent that Theorem 2.14 was already proved by Hirzallah et al. The method is not a new technology; it is a careful refinement of known techniques, but the main bound is new and useful.\n\nWhere the soft spots are, and they are minor. Corollary 2.5's second displayed inequality does not obviously follow from the stated substitutions into Theorem 2.4. It is likely a typo or a missing intermediate step, but a referee should ask the authors to spell it out. Theorem 2.10's proof is deferred entirely to a citation; acceptable for a known characterization, but thin. Theorem 2.16's proof contains a step where |a² − b²| is bounded using quantities involving m(ReT) and m(ImT); that step is not immediate and should be checked. None of these affect the central inequality. There are also some small typos and awkward phrases, but nothing that undermines the mathematics.\n\nThe citation pattern looks normal for the subfield. The authors cite their own related work, but the key new inequality is not secretly recycled. No code or data, which is standard for pure functional analysis.\n\nWho this is for: people working on numerical radius inequalities and operator matrices. It will be a useful reference for the improved lower bound and for comparison inequalities. It is not a blockbuster, but it is a solid, honest piece of work.\n\nMy recommendation: yes, send it to peer review. A careful referee can fix the secondary issues and the paper will make a decent contribution to the literature.","headline":"The main numerical radius bound (2.3) is correct and genuinely new; the paper deserves a serious referee, though a few secondary inequalities need cleanup.","tokens_in":721,"tokens_out":1620,"would_cite":true,"duration_ms":40828,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["47A12","47A63","47A30"],"pacs":[],"model":"deepseek-v4-flash","headline":"For every nonzero Hilbert-space operator, the numerical radius is at least half the operator norm plus a Crawford-number correction.","keywords":["numerical radius","Crawford number","operator norm","operator matrix","Hilbert space","numerical range","inequalities","bounded linear operator"],"falsifier":"Search for a nonzero finite matrix $T$ with $w(T)<\\lVert T\\rVert/2 + m(T^2)/(2\\lVert T\\rVert)$; the theorem asserts that none exists, so a single such matrix would refute it. A natural check is to take $2\\times2$ or $3\\times3$ matrices with $m(T^2)>0$ and compute both sides numerically.","tokens_in":12495,"feed_emoji":"📐","tokens_out":9624,"duration_ms":91763,"temperature":0.7,"pith_summary":"The paper proves sharper upper and lower estimates for the numerical radius $w(T)$ of a bounded linear operator $T$ on a complex Hilbert space, the largest modulus of an expectation value $\\langle Tx,x\\rangle$ over unit vectors. Its main lower bound is $w(T)\\ge \\lVert T\\rVert/2 + m(T^2)/(2\\lVert T\\rVert)$ for every nonzero $T$, where $m(T^2)$ is the Crawford number of $T^2$, the smallest modulus attained by $\\langle T^2x,x\\rangle$. Because $m(T^2)\\ge 0$, this strictly improves the classical inequality $w(T)\\ge \\lVert T\\rVert/2$ whenever $T^2$ has numerical range away from zero. The paper also obtains upper and lower bounds for numerical radii of operator matrices and gives examples where the new bounds beat earlier estimates.","feed_headline":"Numerical radius bound sharpened by Crawford number","feed_subtitle":"For every nonzero operator, the classical half-norm lower bound gains a correction term; equality forces a rigid block form.","key_machinery":"The central object is the numerical range $W(T)=\\{\\langle Tx,x\\rangle: \\lVert x\\rVert=1\\}$; its extremal radii are the numerical radius $w(T)=\\sup\\{|\\lambda|:\\lambda\\in W(T)\\}$ and the Crawford number $m(T)=\\inf\\{|\\lambda|:\\lambda\\in W(T)\\}$. The argument runs through a three-operator inequality $|\\langle A^*TBx,x\\rangle|+|\\langle B^*TAx,x\\rangle|\\le 2w(T)\\lVert Ax\\rVert\\lVert Bx\\rVert$, proved by expressing the sum of two expectation values as the difference of two numerical-range values and choosing a parameter optimally. Applying this with $A=B=T$ and using the standard identity $w\\left(\\begin{smallmatrix}0&T\\\\T&0\\end{smallmatrix}\\right)=w(T)$ turns the inequality into the lower bound for $T$ alone; the same lemma, specialized to operator matrices, yields the paper's upper and lower bounds for matrix entries.","core_discovery":"The paper's central discovery is that the classical lower bound is not tight: the gap between numerical radius and half the operator norm is controlled by the Crawford number of the square. More precisely, for every nonzero $T$, $w(T)\\ge \\lVert T\\rVert/2 + m(T^2)/(2\\lVert T\\rVert)$, with a second bound $w(T)\\ge c^2(T)/(2\\lVert T\\rVert)+w(T^2)/(2\\lVert T\\rVert)$ using the minimum norm $c(T)$. The two bounds are combined into a single maximum in Corollary 2.12. For $n\\times n$ matrices, the paper further proves that equality $w(T)=\\lVert T\\rVert/2$ holds exactly when $T$ is unitarily similar to the direct sum of a nilpotent block of norm $\\lVert T\\rVert$ and a remainder of numerical radius at most $1/2$, so the minimal ratio is attained only by operators whose square has Crawford number zero.","pith_inferences":["Not in the paper: a natural analogue with higher powers is $w(T)\\ge \\lVert T\\rVert/2 + m(T^n)/(2\\lVert T\\rVert^{n-1})$ for $n>2$, worth testing since the square case arises because the off-diagonal $2\\times2$ matrix squares to $T^2$.","Not in the paper: the equality condition suggests that minimal-numerical-radius operators are exactly those whose square has zero in the closure of its numerical range, a characterization that could be linked to existing results on numerical-radius attainability.","Not in the paper: for finite matrices the bound supplies a cheap numerical lower test to screen matrices for near-minimal numerical radius in optimization or randomized searches."],"forward_implications":["If correct, the classical estimate $w(T)\\ge \\lVert T\\rVert/2$ is strict for every operator whose square has positive Crawford number.","Any operator attaining the minimal value $w(T)=\\lVert T\\rVert/2$ must satisfy $m(T^2)=0$, and for matrices this forces the explicit block structure of Theorem 2.10.","The two lower bounds combine to give the stronger estimate $w(T)\\ge \\frac{1}{2\\lVert T\\rVert}\\max\\{\\lVert T\\rVert^2+m(T^2),\\,c^2(T)+w(T^2)\\}$.","The new upper bounds for $2\\times2$ operator matrices can be strictly better than existing bounds, as the paper's worked examples show.","In applications that estimate norms through the numerical radius, these bounds narrow the interval in which the true operator norm can lie."],"supporting_citations":[{"why":"Supplies the one-operator inequality that Lemma 2.2 generalizes and from which the main lower bound is derived.","marker":"[2]"},{"why":"Provides the standard numerical-range facts used throughout, including the off-diagonal matrix identity and the equality case quoted in Theorem 2.10.","marker":"[8]"},{"why":"Gives earlier numerical-radius bounds for 2x2 operator matrices that the paper's new bounds strengthen or complement.","marker":"[9]"},{"why":"Contains a competing upper bound for 2x2 operator matrices that Theorem 3.7 is shown to beat on an example.","marker":"[14]"},{"why":"Yields the equality $w(T)=\\sup_\\theta \\lVert\\operatorname{Re}(e^{i\\theta}T)\\rVert$ used to derive several upper bounds.","marker":"[15]"}],"fun_headline_variants":["Crawford number sharpens numerical radius bound","Numerical radius lower bound gains Crawford correction","Equality in half-norm bound forces rigid block form","Improved numerical radius via Crawford term"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The proof of the main lower bound assumes without proof the standard identity $w\\left(\\begin{smallmatrix}0&T\\\\T&0\\end{smallmatrix}\\right)=w(T)$; if that identity failed, the improvement term $m(T^2)/(2\\lVert T\\rVert)$ would not follow.","fun_headline_variants_meta":{"raw":{"variants":["Crawford number sharpens numerical radius bound","Numerical radius lower bound gains Crawford correction","Equality in half-norm bound forces rigid block form","Improved numerical radius via Crawford term"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000234,"raw_usage":{"total_tokens":1468,"prompt_tokens":890,"completion_tokens":578,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":506,"completion_tokens_details":{"reasoning_tokens":521}},"tokens_in":506,"tokens_out":578,"duration_ms":6072,"temperature":1.0,"reasoning_tokens":521,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T13:42:22.069117+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Search for a nonzero finite matrix $T$ with $w(T)<\\lVert T\\rVert/2 + m(T^2)/(2\\lVert T\\rVert)$; the theorem asserts that none exists, so a single such matrix would refute it. A natural check is to take $2\\times2$ or $3\\times3$ matrices with $m(T^2)>0$ and compute both sides numerically.","supporting_citations":[{"cited_title":"Gustafson and D.K.M","cited_arxiv_id":null,"evidence_quote":"Provides the standard numerical-range facts used throughout, including the off-diagonal matrix identity and the equality case quoted in Theorem 2.10."},{"cited_title":"Hirzallah, F","cited_arxiv_id":null,"evidence_quote":"Gives earlier numerical-radius bounds for 2x2 operator matrices that the paper's new bounds strengthen or complement."},{"cited_title":"Yamazaki, On upper and lower bounds of the numerical radius and an equality condition, Studia Math","cited_arxiv_id":null,"evidence_quote":"Yields the equality $w(T)=\\sup_\\theta \\lVert\\operatorname{Re}(e^{i\\theta}T)\\rVert$ used to derive several upper bounds."}],"review_version":1}