{"id":"0eda18dd-75ee-4cec-977a-fca773391e16","arxiv_id":"1908.08897","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":7.0,"correctness_risk":"low","formal_verification":"none","parameter_count":0,"one_line_summary":"The complement of the union of the spectra of A + tB, with B bounded and non-negative, is always a discrete set, and every discrete subset of the real line can be obtained this way.","lead":"This paper classifies, for any self-adjoint operator A and bounded non-negative perturbation B, which real numbers can be kept out of the spectrum of A + tB for every strength t. The answer is that the protected points are exactly the discrete subsets of the real line, which means the union of all such spectra is always dense.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"No significant objection identified: the central classification is internally sound; only a harmless sign typo in an example.","rationale":"The reader's verdict of CONDITIONAL rests on a sign error in an illustrative example, not on the central argument. My stress-test of the main proof found no gap. The B ≥ 0 assumption is essential, as the paper itself demonstrates with an indefinite example, but it is an explicit hypothesis rather than a hidden weakness. The proof of Lemma 2.10, which constructs operators realizing any discrete protected set, was checked in detail: the kernel computation correctly uses cyclicity of v to force all βλ ≠ 0, and the converse direction correctly uses Proposition 2.1 to conclude that any protected point must be an eigenvalue of K. The corollary about spectrum independence also follows cleanly from the discreteness of the complement and closedness of the spectrum. The sign typo in Example 2.4 is real but inconsequential for the theorem; conditional acceptance after a minor correction remains an appropriate editorial decision.","tokens_in":6124,"tokens_out":19616,"duration_ms":194006,"concrete_test":"Recompute the eigenvalues in Example 2.4 with the corrected formula λ±(t) = t/2 ± sqrt(t^2/4 + 1) and verify that ∪_{t∈R} {λ+(t), λ-(t)} = R \\ {0}; this settles that the sign typo is harmless.","verdict_should_be":"UNCHANGED","load_bearing_attack":"No load-bearing concern identified. The proof of Theorem 1.1 is internally consistent: Proposition 2.1 correctly uses B ≥ 0 only through the existence of B^{1/2}; Lemma 2.7's monotonicity argument gives countability; Lemma 2.8's analytic-zero argument correctly excludes accumulation points; Lemma 2.10's cyclic-vector construction realizes every discrete set P. The assumption B ≥ 0 is genuinely load-bearing, but it is explicit in the theorem, and the paper itself acknowledges that indefinite B permits protected intervals. The only defect found is a sign typo in Example 2.4's eigenvalue formula: the eigenvalues should be t/2 ± sqrt(t^2/4 + 1), not -t/2 ± sqrt(t^2/4 + 1). This typo does not affect the claimed union R \\ {0} or any theorem, since t ranges over all real numbers.","agreement_with_reader":"partial"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper classifies the possible sets of protected points for real linear operator pencils A + tB, where A and B are self-adjoint, B is bounded, non-negative, and non-zero. Theorem 1.1 states that the complement of the union over t of spec(A+tB) is always a discrete subset of R (at most countable and without accumulation points), and conversely every discrete set P can be realized by suitable A and B. The proof is elementary and self-contained: Proposition 2.1 characterizes when 0 is protected via the condition 0 in rho(A) and BA^{-1}B=0; Lemma 2.7 proves countability by strict monotonicity of a suitable function; Lemma 2.8 excludes accumulation points via analyticity; Lemma 2.10 gives a constructive realization by a cyclic-vector construction. Corollary 1.2 then shows that if spec(A+tB) is independent of t, then B=0 or spec(A)=R.","tokens_in":6305,"tokens_out":6761,"duration_ms":69385,"significance":"If the main result is correct, it gives a complete and surprisingly clean characterization of all possible protected sets, and it strengthens the intuition that sign-definite perturbations cannot create protected intervals. The proof is a model of concision: the key condition BA^{-1}B=0 is derived from first principles, and the converse construction is explicit. The paper also carefully notes that the sign-definiteness of B is essential, with a 2x2 indefinite example showing that protected intervals can occur. The main theorem and corollary are supported by complete proofs; I found no circularity, hidden assumptions, or fitted parameters. The paper is a valuable short contribution to the spectral theory of operator pencils.","major_comments":[],"minor_comments":[{"comment":"The displayed eigenvalue formula has a sign error: the eigenvalues of A+tB are t/2 ± sqrt(t^2/4 + 1), not -t/2 ± sqrt(t^2/4 + 1). Since t ranges over all real numbers, the equality of the union with R\\{0} is unaffected, but the formula as printed is incorrect and should be fixed.","section":"2, Example 2.4"},{"comment":"In the line after equation (2.6), the identity should read <v,x> = overline{beta_lambda} <psi_lambda, x> (or equivalently note that beta_lambda != 0 and <psi_lambda,x> = 0 force x=0). The current notation is slightly imprecise, though the conclusion is correct.","section":"2, Lemma 2.10"},{"comment":"The remark is terse: it states that BA^{-1}B=0 holds if and only if a certain map defines a pseudo-resolvent. Since this notion is not defined in the paper, a one-sentence explanation or a more precise pointer to [5] would improve readability.","section":"2, Remark 2.2"},{"comment":"Reference [11] is cited as 'in preparation'. Since it is only mentioned in a 'See also' remark and is not used in any proof, the authors may wish to update its status or omit the citation if it remains unpublished.","section":"References, [11]"}],"recommendation":"minor_revision","confidential_remarks":"This is a sound and appropriately scoped note. The only mathematical blemish I found is the sign typo in Example 2.4, which does not affect any theorem. I recommend acceptance after a minor revision. The self-citation [11] is not used in the proofs, so there is no concern about circularity."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Dear colleague,\n\nThe short version: this is a correct, complete classification of protected points for A+tB with B bounded, nonnegative, and nonzero. Theorem 1.1 says the complement of the union of spectra is exactly a discrete subset of R, and conversely every discrete set occurs. That is new. I checked the cited papers [4] and [7]; they treat parameter sets and eigenvalue simplicity, not the union itself.\n\nWhat I like: the proof is elementary and self-contained. Proposition 2.1 reduces the protected point condition to BA^{-1}B = 0, Lemma 2.7 gives countability via strict monotonicity of f, Lemma 2.8 excludes accumulation points using analyticity, and Lemma 2.10 is a nice cyclic-vector construction that realizes any discrete set as the protected set. The converse direction is the most interesting part. The paper is also honest about B ≥ 0 being load-bearing: they explicitly show that for indefinite B you can get a whole interval protected. No fitted parameters or circular arguments anywhere.\n\nWhere are the weak spots? Only a harmless sign typo in Example 2.4. The displayed eigenvalues should be t/2 ± sqrt(t^2/4 + 1), not -t/2 ± sqrt(t^2/4 + 1). It does not affect any theorem because t ranges over all reals, but it will confuse someone reading through. The reference to Veselić's in-preparation paper [11] is only a \"see also\" and is not used in any proof, which is fine. Lemma 2.10's proof is a bit terse, but it works.\n\nWho is this for? Spectral theorists working on operator pencils, homogeneous operators, or relativistic Hamiltonians. It answers a natural question cleanly and goes in one sitting. I would send this to a referee without hesitation. The only required change is fixing the Example 2.4 eigenvalue formula before publication.\n\nBest,\n[You]","headline":"A clean, correct classification of protected points for A+tB with nonnegative bounded B: the union's complement is exactly a discrete set, and every discrete set is realized.","tokens_in":6788,"tokens_out":1531,"would_cite":true,"duration_ms":15924,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["47A56","47A10","47B15"],"pacs":[],"model":"deepseek-v4-flash","headline":"The protected set of any pencil A+tB with B non-negative and non-zero is a discrete subset of the real line, and every discrete set arises this way.","keywords":["union of spectra","operator pencil","protected points","self-adjoint operators","non-negative perturbation","discrete set","homogeneous operator"],"falsifier":"Try to realize a non-discrete protected set, for instance $P=\\{1/n:n\\in\\mathbb{N}\\}\\cup\\{0\\}$, with self-adjoint $A$ and non-negative, bounded $B$; the theorem says this is impossible. A more direct check: for any candidate protected point $\\lambda$, verify numerically whether $B(A-\\lambda)^{-1}B=0$ holds, since a violation would contradict Proposition 2.1 and hence the classification.","tokens_in":5963,"feed_emoji":"🎯","tokens_out":15476,"duration_ms":124136,"temperature":0.7,"pith_summary":"This paper classifies the sets of real numbers that can be protected from a whole line of operators: given a self-adjoint operator $A$ and a bounded, non-negative, non-zero operator $B$, consider the union of the spectra of $A+tB$ as $t$ runs over all reals. The complement of this union, the points that belong to no spectrum of the pencil, is shown to be discrete: at most countable and with no accumulation point. Conversely, every discrete subset of the real line can be realized as exactly this complement by a suitable choice of $A$ and $B$. The result also implies that the union of the spectra is always dense, and that if the spectrum of $A+tB$ is independent of $t$, then either $B=0$ or $\\operatorname{spec}(A)=\\mathbb{R}$.","feed_headline":"Only discrete sets can be protected from an operator pencil","feed_subtitle":"The union of spectra is always dense; only discrete gaps survive.","key_machinery":"The load-bearing mechanism is the equivalence in Proposition 2.1: a real point $\\lambda$ lies outside every spectrum of the pencil exactly when $\\lambda\\in\\rho(A)$ and $B(A-\\lambda)^{-1}B=0$. This reduces the protected set to the real zeros of the analytic function $f(z)=\\langle y,(A-z)^{-1}y\\rangle$ for any $y=Bx$ with $Bx\\neq 0$. Because $f$ is strictly monotone on each connected component of $\\rho(A)\\cap\\mathbb{R}$, it has at most one zero per spectral gap of $A$, which yields that the protected set is at most countable; a separate argument (Lemma 2.8) shows it cannot have an accumulation point. For the converse, the proof constructs, for any discrete $P$, a diagonal operator $K$ with simple spectrum $P$ and a cyclic vector $v$, then sets $A=\\begin{pmatrix} K & v \\\\ \\langle v,\\cdot\\rangle & 0 \\end{pmatrix}$ and $B=\\begin{pmatrix} 0 & 0 \\\\ 0 & 1 \\end{pmatrix}$ on $H\\oplus\\mathbb{C}$, so that the resulting pencil has spectrum exactly $R\\setminus P$.","core_discovery":"The central discovery is that the protected set $R \\setminus \\bigcup_{t\\in\\mathbb{R}} \\operatorname{spec}(A+tB)$ has no structure beyond being discrete. Theorem 1.1 states that for every self-adjoint $A$ and bounded, non-negative, non-zero $B$, this complement is at most countable and contains none of its accumulation points; and, in the converse direction, that for every discrete $P \\subset \\mathbb{R}$ there exist self-adjoint $A$ and such a $B$ with $P$ exactly equal to the complement. In particular, no interval can be protected when $B$ is non-negative, and the union of the pencil's spectra is dense. The paper also derives Corollary 1.2: if $\\operatorname{spec}(A+tB)$ is independent of $t$, then $B=0$ or $\\operatorname{spec}(A)=\\mathbb{R}$.","pith_inferences":["The resolvent test $B(A-\\lambda)^{-1}B=0$ suggests a computational way to certify protected points without forming the full pencil spectrum.","The converse construction might generalize to prescribed multiplicities or to more general pencils by trading the single extra dimension for a finite-rank perturbation.","For indefinite $B$, protected intervals appear; classifying exactly which subsets of $\\mathbb{R}$ arise in the indefinite case would be a natural next problem.","The same resolvent condition appears in the theory of pseudo-resolvents; one could ask whether the discreteness conclusion persists under weaker sign conditions on $B$."],"forward_implications":["The union of spectra $\\bigcup_{t\\in\\mathbb{R}}\\operatorname{spec}(A+tB)$ is dense in $\\mathbb{R}$ whenever $B\\neq 0$.","Each connected component of $\\rho(A)\\cap\\mathbb{R}$ can contain at most one protected point, so a spectral gap of $A$ shields at most one real number from the whole pencil.","If $\\operatorname{spec}(A+tB)$ is the same for all $t$, then either $B=0$ or $\\operatorname{spec}(A)=\\mathbb{R}$ (Corollary 1.2).","Every finite or countably infinite discrete set of real numbers can be realized as the protected set, so the classification is exhaustive.","The sign-definiteness of $B$ is essential: for indefinite $B$, intervals can be protected, as shown by the example $A=\\operatorname{diag}(1,-1)$, $B=\\begin{pmatrix}0&1\\\\1&0\\end{pmatrix}$."],"supporting_citations":[{"why":"Supplies the existence of a cyclic vector for operators with simple spectrum, a key ingredient in the converse construction.","marker":"[8]"},{"why":"Provides the general result that a spectral gap is preserved under off-diagonal perturbations, backing the counterexample showing sign-definiteness of B is essential.","marker":"[1]"},{"why":"Also cited for the preservation of spectral gaps under off-diagonal perturbations, reinforcing the same sign-definiteness counterexample.","marker":"[2]"}],"fun_headline_variants":["Operator pencils leave only discrete gaps","Dense spectrum union, discrete protected set","No interval survives an operator pencil","Spectra cover all but discrete points"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The classification depends crucially on $B$ being non-negative; if $B$ is allowed to be indefinite, intervals can be protected and the discrete-set result fails.","fun_headline_variants_meta":{"raw":{"variants":["Operator pencils leave only discrete gaps","Dense spectrum union, discrete protected set","No interval survives an operator pencil","Spectra cover all but discrete points"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000161,"raw_usage":{"total_tokens":1151,"prompt_tokens":777,"completion_tokens":374,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":393,"completion_tokens_details":{"reasoning_tokens":324}},"tokens_in":393,"tokens_out":374,"duration_ms":4250,"temperature":1.0,"reasoning_tokens":324,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T11:26:59.428614+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Try to realize a non-discrete protected set, for instance $P=\\{1/n:n\\in\\mathbb{N}\\}\\cup\\{0\\}$, with self-adjoint $A$ and non-negative, bounded $B$; the theorem says this is impossible. A more direct check: for any candidate protected point $\\lambda$, verify numerically whether $B(A-\\lambda)^{-1}B=0$ holds, since a violation would contradict Proposition 2.1 and hence the classification.","supporting_citations":[{"cited_title":"Schm¨ udgen,Unbounded Self-Adjoint Operators on Hilbert Space , Grad","cited_arxiv_id":null,"evidence_quote":"Supplies the existence of a cyclic vector for operators with simple spectrum, a key ingredient in the converse construction."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Provides the general result that a spectral gap is preserved under off-diagonal perturbations, backing the counterexample showing sign-definiteness of B is essential."},{"cited_title":"Davis, W","cited_arxiv_id":null,"evidence_quote":"Also cited for the preservation of spectral gaps under off-diagonal perturbations, reinforcing the same sign-definiteness counterexample."}],"review_version":1}