{"id":"70587ba0-0154-405a-ade2-ae6c0e6a141a","arxiv_id":"2505.15330","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":7.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":0,"one_line_summary":"Finite sums of consecutive Hermite polynomials have real-rootedness governed by a coefficient polynomial: real roots of P force real roots of the sum, and non-real roots of P appear one-for-one in high-degree sums.","lead":"Hermite polynomials are a classic family of functions used across physics and probability, and this paper determines exactly when finite sums of neighboring Hermite polynomials have only real roots. The answer is governed by one helper polynomial made from the coefficients, with precise thresholds and interlacing for two standard normalizations.","discovery_kind":"extension","skeptic_critique":{"model":"deepseek-v4-flash","headline":"The Appell-normalization classification depends on the asymptotic (5.6), imported from the author's unpublished [7]; without a proof or independent verification of that limit's uniformity, the non-real-zero steps in Theorem 1.3 and Corollary 5.1 are not established.","rationale":"After reading the manuscript, I agree with the reader's weakest-assumption analysis. The internal machinery (generalized Hermite polynomials, interlacing, induction on K-N_nr) is coherent, and the smaller gaps — the omitted proof of Lemma 2.4, the terse 'n0 >= n1+1' step in the proof of Theorem 1.3, and the unproved induction cases — appear repairable. The genuinely load-bearing input is (5.6), because it is the sole bridge from the characteristic polynomial P to non-real zeros of q_n in the Appell normalization. A failure of that asymptotics, especially of its uniformity, would invalidate the existence and exact-count statements in Theorem 1.3 and the uniform threshold in Corollary 5.1. The paper's reliance on two self-authored preprints ([7], [8]) makes this a correctness risk rather than a proven flaw; the concern is precisely the one the reader identified. My verdict is unchanged: CONDITIONAL, pending verification of the imported asymptotic.","tokens_in":16635,"tokens_out":27287,"duration_ms":219105,"concrete_test":"Re-derive (5.6) from the exact generating function (5.3), sum_n q_n(x) z^n = e^{xz-z^2/4} R(z), via a saddle-point (or Mehler-Heine type) coefficient estimate, without invoking [7, Theorem 1.1]; the derivation must produce the stated limit together with local uniform convergence on compact subsets of C\\{0}. If this derivation succeeds, the imported asymptotic is validated and the concern is resolved; if it cannot be obtained, the proof of Theorem 1.3 is left with an unsupported load-bearing step.","verdict_should_be":"UNCHANGED","load_bearing_attack":"The central proof of Theorem 1.3 turns on the asymptotic (5.6): lim_n (z/(n+1))^n n! q_n((n+1)/z) = z^K e^{-z^2/4} P(1/z), uniformly on compact sets. This is the only mechanism that converts a non-real zero of P into non-real zeros of q_n: in the base case N_nr=K it supplies an n0 with at least K non-real zeros, in the induction step it supplies n0 with at least N_nr non-real zeros, and in the 'only if' direction of Part (1) it rules out all-real q_n when P has a non-real zero. Corollary 5.1 uses the same limit, via Lemma 2.4, to make the threshold uniform in theta. The manuscript does not prove (5.6); it cites [7, Theorem 1.1], a self-authored preprint whose conditions and proof are not included or verified here. If (5.6) fails or holds only pointwise rather than locally uniformly, Hurwitz's theorem cannot be applied and both Theorem 1.3 and Corollary 5.1 lose their foundation. The asymptotic is plausibly derivable from the exact generating function (5.3) by a saddle-point argument, so the gap may be repairable, but as written it is load-bearing.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies the number of real zeros of finite linear combinations of K+1 consecutive Hermite polynomials in two normalizations: the standard normalization and the Appell normalization H_n/(2^n n!). For both cases it introduces the auxiliary polynomial P(x)=∑γ_j x^{K−j} and shows that its zeros determine the real-rootedness of q_n=Σγ_j H_{n−j}. In the standard case (Theorem 1.2), the paper proves that real-rootedness of P implies real-rootedness of q_n for all n≥K, with interlacing, and that if P has nonreal zeros a threshold depending only on those nonreal zeros suffices. In the Appell case (Theorem 1.3), it gives a complete classification: q_n is real-rooted for all n if and only if P has no nonreal zeros, and when P has N_nr nonreal zeros, after a threshold n0 the polynomial q_n has exactly n−N_nr real zeros and N_nr nonreal zeros. The paper also proves Turán-type inequalities, asymptotic zero behavior via Mehler–Heine-type limits, and a conjecture on the signs of the real zeros when P is nonreal-rooted. The main tools are a backward-shift operator Λ f=2xf−f′, a generalization of Hermite polynomials, Obreshkov's interlacing theorem, and a complex asymptotic for the Appell-normalized combinations.","tokens_in":16949,"tokens_out":12370,"duration_ms":100529,"significance":"If the central claims hold, this is a substantial contribution to the zero-location theory of linear combinations of classical orthogonal polynomials. The identification of P as the governing object is natural and cleanly separates the two normalizations: in the standard normalization real-rootedness of P is sufficient, while in the Appell normalization it is also necessary. The interlacing results and the Turán-type inequalities are of independent interest, and the asymptotics in Corollary 5.3 give a fairly complete picture of the zero distribution. The strategy via the backward-shift operator and generalized Hermite polynomials is elegant and yields new structural results, including the connection to multiple Hermite polynomials in Remark 1. The paper is not circular and introduces no fitted parameters; P is an input polynomial. The main reservations are that two load-bearing inputs are imported from the author's own preprints ([7] and [8]) and that a key lemma and some induction cases are stated without proof; these limit the paper's self-containedness but appear repairable.","major_comments":[{"comment":"The asymptotic (5.6), lim_n (z/(n+1))^n n! q_n((n+1)/z) = z^K e^{-z^2/4} P(1/z), uniformly on compact sets, is the sole mechanism in the proof of Theorem 1.3 that turns nonreal zeros of P into nonreal zeros of q_n. It is imported from the author's preprint [7, Theorem 1.1] without a proof or a statement of its hypotheses. The same asymptotic is used in Corollary 5.1 and Corollary 5.3. Since this is load-bearing, please include a proof or a precise, verifiable statement of the required uniformity; without this the main classification is not established within the manuscript.","section":"§5, Eq. (5.6)"},{"comment":"Lemma 2.4 is stated with its proof omitted ('the proof is similar to the usual proof for the Hurwitz's Theorem ... and it is omitted'). The uniformity of the integer n* with respect to the real parameter θ is essential for Corollary 5.1, and the lemma is not a standard textbook statement. Please supply a proof or a reference that actually contains this uniformity assertion; a Rouché-type argument may work, but it must be written out.","section":"Lemma 2.4"},{"comment":"The induction on K−N_nr in the proof of Theorem 1.3 begins with the base case N_nr=K and then states 'Assume next that K − N_nr > 1.' This excludes the case K−N_nr=1, i.e., P having exactly one real zero. That case is needed for the induction itself and for part (1) when K=1. The closing sentence 'If N_nr=0, then all the zeros of q_n has to be real for n≥0, because q_n=q′_{n+1}' does not by itself prove real-rootedness for all n. The induction step should be formulated for K−N_nr≥1 and should spell out how the case K−N_nr=1 reduces to the base case.","section":"Proof of Theorem 1.3"},{"comment":"In the proof of Corollary 4.1, after treating K−2m=0 and K−2m=1, the statement 'The cases K − 2m ≥ 2 can be proved similarly' delegates the iterative interlacing argument to the reader. Since this is the advertised improvement over Corollary 1.1, the iterative use of Lemmas 2.2 and 2.3 should be written out explicitly rather than left to a 'similarly'.","section":"Proof of Corollary 4.1"}],"minor_comments":[{"comment":"The phrase 'tipe II multiple Hermite polynomials' contains a typo; it should be 'type II'.","section":"Introduction"},{"comment":"In the definition of φ{l}, the phrase 'que sequence obtained by removing' should be 'the sequence obtained by removing'.","section":"§3, Eq. (3.7)"},{"comment":"After equation (5.1), the symbol H_n is used for the normalized polynomial, which is also the standard notation for the unnormalized Hermite polynomial used in Sections 3 and 4. This overloads notation; consider writing ~H_n for the normalized polynomial throughout.","section":"§5, Eq. (5.1)"},{"comment":"The symbol N_− is used for the number of negative real zeros of P, but this is not defined before its first use in the statement of Corollary 5.3. Please define it explicitly.","section":"Corollary 5.3"},{"comment":"In the proof of Corollary 5.3, 'asimptotic' should be 'asymptotic'.","section":"Corollary 5.3, proof"}],"recommendation":"major_revision","confidential_remarks":"The manuscript's central theorems depend on two self-authored preprints: the asymptotic (5.6) from [7] and Corollary 1.1 (plus Lemma 3.1) from [8]. The referee did not independently verify [7] and [8]. If the journal expects the published paper to be self-contained for load-bearing results, the author should be asked to include proofs or precise statements of these inputs. The omitted induction cases and the unproved Lemma 2.4 are also worth attention, but they appear repairable within the scope of a revision."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Quick take: this is a real contribution, not a desk reject. The generalized Hermite construction in Section 3 is the strongest part: the backward shift ΛH_n = H_{n+1} plus the φ,ψ sequences gives a clean framework where interlacing and Turán inequalities fall out naturally. The standard-normalization results, Theorem 1.2 and Corollary 4.1, are convincing and improve the threshold from [8] so that it depends only on the non-real zeros of P. The Appell classification in Theorem 1.3 is genuinely new if it holds.\n\nSoft spots, roughly in order of severity.\n\n(5.6) is load-bearing and imported from the author's unpublished [7]. It is the only mechanism that converts non-real zeros of P into non-real zeros of q_n, and Corollary 5.1 also depends on it through Lemma 2.4. The generating function (5.3) is present, so a referee can reasonably ask for a proof of (5.6) in an appendix or a citation to a published source. I do not think the asymptotic is false, but as written the central Appell results rest on unverified external support. The stress-test's concern lands; I would call it repairable rather than fatal.\n\nLemma 2.4 is stated with its proof omitted. It is not entirely trivial because the n* must be uniform in θ. It may well be true, but it needs a proof in the paper.\n\nThe proof of Theorem 1.3 has structural gaps. The induction on K − N_nr skips the case K − N_nr = 1, and the 'if' direction for N_nr = 0 is dispatched with the line 'because q_n = q'_{n+1}', which by itself is a non sequitur. The claim is true — eventual real-rootedness from (5.6) plus the monotonicity Z_nr(n+1) ≥ Z_nr(n) gives all n — but the written proof does not say that. Repairable, not fatal.\n\nCorollary 4.1 says 'the cases K − 2m ≥ 2 can be proved similarly' and Corollary 5.5 says 'if θ < 0, the proof is similar.' Those are exactly the places a referee will want expansion.\n\nCitation pattern: heavy reliance on [7] and [8], both self-authored preprints. Self-citation is not automatically a problem, but here the unproved asymptotic is the hinge, so the dependence should be made explicit.\n\nThis paper is for people working on zeros of orthogonal polynomial combinations, Appell sequences, and real-rootedness. It deserves a serious referee. My verdict is conditional: the Appell results can become solid after the (5.6) gap is closed and the induction is cleaned up; otherwise a major revision is needed.","headline":"Genuinely new results on zeros of Hermite combinations, but the Appell half leans on an unproved imported asymptotic and a couple of terse induction steps; worth a serious referee.","tokens_in":17431,"tokens_out":10870,"would_cite":true,"duration_ms":96713,"reading_group":"yes","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["42C05","26C10","33C45"],"pacs":[],"model":"deepseek-v4-flash","headline":"The auxiliary polynomial $P(x)=\\sum_{j=0}^K\\gamma_j x^{K-j}$ controls real-rootedness of all consecutive Hermite combinations, with exact zero counts in the Appell normalization.","keywords":["zeros of polynomials","Hermite polynomials","Appell polynomials","real-rootedness","interlacing of zeros","backward shift operator","multiple Hermite polynomials","Brenke-type asymptotics"],"falsifier":"For the Appell normalization with $K=2$, $\\gamma_0=1$, $\\gamma_1=0$, $\\gamma_2=1$, so $P(x)=x^2+1$ has two non-real zeros, Theorem 1.3 predicts that once $q_n=\\tilde H_n+\\tilde H_{n-2}$ first has exactly two non-real zeros, all later $q_n$ also have exactly two non-real zeros and $n-2$ real simple zeros. Computing the zeros for $n=3,4,\\ldots,100$ and checking that the count stabilizes at two non-real zeros would settle the claim; any later $n$ with a different count is a counterexample.","tokens_in":16387,"feed_emoji":"","tokens_out":13180,"duration_ms":94827,"temperature":0.7,"pith_summary":"The paper studies polynomials $q_n(x)=\\sum_{j=0}^K\\gamma_j\\tilde H_{n-j}(x)$ formed from $K+1$ consecutive Hermite polynomials, in two normalizations: the standard one ($\\tilde H_n=H_n$) and the one making $q_n$ an Appell sequence ($\\tilde H_n=H_n/(2^nn!)$, so $q_n'=q_{n-1}$). Its central claim is that in both cases the real-rootedness of every $q_n$ is controlled by the single finite polynomial $P(x)=\\sum_{j=0}^K\\gamma_jx^{K-j}$. If $P$ has only real zeros, then in both normalizations all $q_n$ are real-rooted with simple zeros, and the zeros of consecutive polynomials alternate. If $P$ has $N_{nr}$ non-real zeros, then in the Appell normalization $q_n$ has exactly $n-N_{nr}$ real zeros and $N_{nr}$ non-real zeros once $n$ is large enough, and in the standard normalization a threshold depending only on those non-real zeros guarantees the same real-rootedness. This reduces an infinite family of questions to one finite polynomial, and it makes the dependence on normalization explicit.","feed_headline":"One auxiliary polynomial dictates the real zeros of Hermite sums","feed_subtitle":"The real-rootedness of every consecutive Hermite combination reduces to the zeros of one polynomial P.","key_machinery":"The load-bearing mechanism is the backward shift $\\Lambda f(x)=2xf(x)-f'(x)$, which satisfies $\\Lambda H_n=H_{n+1}$ and is a real-zero-increasing operator, together with the auxiliary polynomial $P(x)=\\sum_{j=0}^K\\gamma_j x^{K-j}$, whose zeros are the single finite source of all real-rootedness information. The paper encodes $q_n$ in a family of generalized Hermite polynomials $h_{\\varphi,\\psi}$ built from $\\Lambda$; when $\\psi_i=0$ these generalize multiple Hermite polynomials and their interlacing properties are proved directly. For the Appell normalization the engine is the generating-function limit $$\\lim_n \\left(\\frac{z}{n+1}\\right)^n n!\\,q_n\\!\\left(\\frac{n+1}{z}\\right)=z^K $e^{{-z^2/4}}$P(1/z),$$ uniform on compact sets, which transfers non-real zeros of $P$ into non-real zeros of $q_n$ and gives the exact count in Theorem 1.3.","core_discovery":"The paper establishes Theorem 1.2 and Theorem 1.3. In the standard normalization, whenever $P(x)=\\sum_{j=0}^K\\gamma_j x^{K-j}$ has only real zeros, every $q_n$ with $n\\ge K$ has only real and simple zeros, and the zeros of $q_{n+1}$ strictly interlace those of $q_n$; if $P$ has non-real zeros, there is a positive integer $n_0$, depending only on $K$ and on the non-real zeros of $P$, such that the same is true for all $n\\ge n_0$. In the Appell normalization, the description is exact: $q_n$ is real-rooted for all $n\\ge 0$ if and only if $P$ has no non-real zeros, and if $N_{nr}>0$ there is a smallest $n_0$ such that $q_{n_0}$ has exactly $N_{nr}$ non-real zeros, with $q_n$ having exactly $n-N_{nr}$ real zeros and $N_{nr}$ non-real zeros precisely for $n\\ge n_0$; in that range all zeros are simple and the real zeros of consecutive polynomials interlace.","pith_inferences":["Because the proof uses only the backward-shift identity and a Brenke-type generating function, the same 'control polynomial $P$' structure should hold for other Appell chains built from classical orthogonal polynomials; testing it on monic Laguerre combinations would be a direct extension.","The exact-count statement suggests a computational certificate: for fixed coefficients, checking the zeros of one finite polynomial $P$ and one small-degree $q_{n_0}$ decides real-rootedness of the whole sequence.","The paper proves convergence of the non-real zeros to the scaled zeros of $P$ but does not give rates; studying their fluctuations as $n$ grows is a natural next step.","The paper leaves an explicit conjecture on the number of positive and negative zeros when all zeros of $P$ are non-real; the conjecture's parity-dependent counts are numerically checkable and, if true, would complete the sign-count description."],"forward_implications":["In the standard normalization, $P$ having only real zeros implies that all $q_n$ with $n\\ge K$ are real-rooted, simple, and interlaced degree-by-degree.","In the standard normalization with non-real zeros of $P$, the threshold $n_0$ depends only on $K$ and on those non-real zeros, so a single finite check certifies the whole family.","In the Appell normalization, the deficit in real zeros is constant for large $n$: $q_n$ has exactly $n-N_{nr}$ real zeros and $N_{nr}$ non-real zeros.","Both normalizations give a Turan-type inequality $q_{n-1}(x)^2-q_n(x)q_{n-2}(x)>0$ for all real $x$, from a threshold onwards (or for all $n\\ge 2$ when all zeros of $P$ are real in the Appell case).","Rescaled zeros have explicit limits: central real zeros follow the trigonometric Mehler-Heine scaling, the extreme real and non-real zeros approach the corresponding zeros of $P$, and the counting measure converges to the semicircle law."],"supporting_citations":[{"why":"Supplies the Brenke-type asymptotic (5.6), uniform on compact sets, which transfers non-real zeros of P to the Appell-normalized q_n.","marker":"[7]"},{"why":"Supplies Corollary 1.1, the uniform real-rootedness threshold for standard Hermite combinations, and the key lemmas on interlacing used in Section 4.","marker":"[8]"},{"why":"Establishes the earlier result that real zeros of the coefficient polynomial imply real-rootedness, which Theorem 1.2 recovers and extends.","marker":"[12]"},{"why":"Provides the theorem on zeros of linear combinations of orthogonal polynomials used in the zero-asymptotics corollary.","marker":"[2]"},{"why":"Supplies the Mehler-Heine formulas used to identify the central-zero scaling limits.","marker":"[17]"},{"why":"Supplies the backward-shift identity and the generating function for Hermite polynomials used throughout.","marker":"[14]"},{"why":"Provides the bound on Hermite zeros used to separate extreme zeros in the asymptotic analysis.","marker":"[22]"}],"fun_headline_variants":["One polynomial P controls real zeros of Hermite combinations","Auxiliary polynomial dictates real-rootedness of Hermite sums","Real zeros of Hermite combinations hinge on one polynomial","Zeros of Hermite linear combinations reduce to zeros of P","Interlacing and real zeros of Hermite sums follow from P"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The exact real/non-real zero count in the Appell normalization depends on the uniform Brenke-type asymptotic imported from a companion preprint; if that asymptotic fails to be uniform, the conclusion that non-real zeros of $P$ reappear as exactly $N_{nr}$ non-real zeros of $q_n$ does not follow.","fun_headline_variants_meta":{"raw":{"variants":["One polynomial P controls real zeros of Hermite combinations","Auxiliary polynomial dictates real-rootedness of Hermite sums","Real zeros of Hermite combinations hinge on one polynomial","Zeros of Hermite linear combinations reduce to zeros of P","Interlacing and real zeros of Hermite sums follow from P"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000284,"raw_usage":{"total_tokens":1697,"prompt_tokens":990,"completion_tokens":707,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":606,"completion_tokens_details":{"reasoning_tokens":623}},"tokens_in":606,"tokens_out":707,"duration_ms":6778,"temperature":1.0,"reasoning_tokens":623,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T15:21:26.994476+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For the Appell normalization with $K=2$, $\\gamma_0=1$, $\\gamma_1=0$, $\\gamma_2=1$, so $P(x)=x^2+1$ has two non-real zeros, Theorem 1.3 predicts that once $q_n=\\tilde H_n+\\tilde H_{n-2}$ first has exactly two non-real zeros, all later $q_n$ also have exactly two non-real zeros and $n-2$ real simple zeros. Computing the zeros for $n=3,4,\\ldots,100$ and checking that the count stabilizes at two non-real zeros would settle the claim; any later $n$ with a different count is a counterexample.","supporting_citations":[{"cited_title":"Iserles and E","cited_arxiv_id":null,"evidence_quote":"Establishes the earlier result that real zeros of the coefficient polynomial imply real-rootedness, which Theorem 1.2 recovers and extends."},{"cited_title":"Beardon, K.A","cited_arxiv_id":null,"evidence_quote":"Provides the theorem on zeros of linear combinations of orthogonal polynomials used in the zero-asymptotics corollary."},{"cited_title":"Olver, D.W","cited_arxiv_id":null,"evidence_quote":"Supplies the Mehler-Heine formulas used to identify the central-zero scaling limits."},{"cited_title":"Koekoek, P","cited_arxiv_id":null,"evidence_quote":"Supplies the backward-shift identity and the generating function for Hermite polynomials used throughout."},{"cited_title":"Szeg¨ o, Orthogonal Polynomials","cited_arxiv_id":null,"evidence_quote":"Provides the bound on Hermite zeros used to separate extreme zeros in the asymptotic analysis."}],"review_version":1}