{"id":"6e375c97-5531-4695-91c6-0499c4d31373","arxiv_id":"2506.12206","paper_version":1,"verdict":"CONDITIONAL","confidence":"MODERATE","novelty_score":8.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Random Kac polynomials have discriminant |Δ(f_n)| = n^{2n} e^{-D_* n(1+o(1))} with an explicit universal constant D_* ≈ 5.92947.","lead":"Random polynomials with independent coefficients have a discriminant that concentrates at a precise scale: its logarithm behaves like 2n log n minus a universal constant times n. The paper computes that constant explicitly, correcting an earlier heuristic that predicted a different leading behavior.","discovery_kind":"first_principles","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Internal factor-4 error in the conditional-variance computation (Claim 26, Eqs. 51–52) makes the stated D*≈5.929 incorrect; corrected constant is ≈1.34, so Theorem 1 as stated is false.","rationale":"After checking the internal reductions, the only theorem-critical quantity that determines D* is c* from Lemma 23, which is computed via the Gaussian Kac-Rice density ψ_n. That computation contains a clear algebraic error: the normalizer s_n(t) is defined with the factor 1/n, so the second derivative s''_n already contains the factor 4 from the exponential parametrization; the conditional variance of the t-derivative is its Schur complement with E|G'|^2 = s''/4. The paper instead uses s'' directly, inflating τ by a factor 4 and inserting log S rather than log(S/4). The numerical value D*≈5.929 in (10) matches the erroneous formula, so this is not a typo in the abstract. The reader's concern about the companion-paper lemmas [19, Lemma 4.3 etc.] is legitimate but secondary: even granting all imported bounds, the stated constant is wrong. The proof could likely be repaired, but the current Theorem 1 is not correct as stated.","tokens_in":31386,"tokens_out":35930,"duration_ms":333484,"concrete_test":"Differentiate (43) and verify the identities E|G'_n|^2 = s''_n/4 and E[G_n \\overline{G'_n}] = s'_n/2; then replace (52) by τ_n = (s''_n - s'^2_n/s_n)/4 and recompute the integral in Lemma 23. A quick numeric check at t=1 with Claim 27 limits (s=0.4323, s'=-0.2970, s''=0.3234) gives τ=0.0298, whereas the paper's S(1)=0.1193. If the corrected integral shifts D* from ≈5.929 to ≈1.34, Theorem 1's stated constant is false.","verdict_should_be":"REJECT","load_bearing_attack":"Section 5.3, Eq. (51)–(52): the conditional variance of G'_n(t) given G_n(t)=0 is computed as es_n = s''_n - (s'_n)^2/s_n. This is wrong by a factor 4. With G_n(t)=n^{-1/2}Σ γ_k e^{-tk/n} and G'_n(t)=-n^{-3/2}Σ k γ_k e^{-tk/n}, one has E|G'_n|^2 = n^{-3}Σ k^2 e^{-2tk/n} = s''_n/4 and E[G_n \\overline{G'_n}] = -n^{-2}Σ k e^{-2tk/n} = s'_n/2, so the correct Schur complement is τ_n = (s''_n - (s'_n)^2/s_n)/4. The paper's '4∫λ² - 4(∫λ)²/∫1' in the limit is therefore four times the true conditional variance. The limiting density in (46) should read Ψ(t) = (S(t)/(8a(t)))(log(S(t)/4)+1-γ), not (S/(2a))(log S+1-γ); this changes the universal constant D* in (10) from ≈5.92947 to ≈1.34 (using ∫(1/t²-1/sinh²t)dt=1). The error is internal, affects Theorem 1's explicit constant directly, and is independent of the [19] imports.","agreement_with_reader":"disagree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies random Kac polynomials f_n(z)=∑_{k=0}^n ξ_k z^k with i.i.d. mean-zero, variance-one sub-Gaussian coefficients and proves a law of large numbers for the logarithmic discriminant: (1/n)(log|Δ(f_n)|−2n log n) converges in probability to −D_* for an explicit universal constant D_*>0, equivalently |Δ(f_n)|=n^{2n}e^{−D_*n(1+o(1))}. The proof combines a symmetrized representation of the discriminant (Lemma 5 and Claim 6), a concentration result for the Mahler measure (Proposition 3), a lengthy reduction of the derivative sum over roots to a sum over a net (Section 4), Gaussian comparison via Berry–Esseen (Section 7), and a Kac–Rice computation of the limiting expectation (Section 5). The main theorem is reduced to Propositions 3 and 4.","tokens_in":31657,"tokens_out":12843,"duration_ms":146571,"significance":"If correct, the result would be a valuable and non-obvious universal law of large numbers for the discriminant of random Kac polynomials, with an explicit constant; it would also provide a quantitative justification for numerical observations in the irreducibility literature. The paper contains a clean reduction of the main theorem to two concentration statements, and the symmetrized discriminant representation is an attractive idea. However, the explicit constant computation in Section 5.3 contains a factor-four error that changes the value of D_*, and several load-bearing estimates are imported without proof from the authors' companion preprint. These issues are fixable, and the overall approach is defensible, so the paper merits a major revision rather than rejection.","major_comments":[{"comment":"The conditional variance of G'_n(t) given G_n(t)=0 is computed incorrectly. Since G_n(t)=n^{-1/2}∑ γ_k e^{-tk/n}, one has E|G'_n(t)|^2=s''_n(t)/4 and E[G_n(t)\\,\\overline{G'_n(t)}]=s'_n(t)/2, so the Schur complement is (s''_n-(s'_n)^2/s_n)/4, not s''_n-(s'_n)^2/s_n. Consequently Eq. (46) should read Ψ(t)=(1/t^2−1/sinh^2 t)(log(S(t)/4)+1−γ)/8, and both c_* in Eq. (9) and D_* in Eq. (10) must be recomputed. The numerical value D_*≈5.92947 stated in Section 1 is therefore not the value proved by the manuscript. This is a load-bearing error in the statement of Theorem 1, although the general structure of the proof may survive with a corrected constant.","section":"§5.3, Eqs. (51)–(52), Claim 26"},{"comment":"The concentration estimates that exclude roots where the derivative is small are imported from the companion preprint [19] without statements of the imported results. In particular, Claim 13 and Claim 15 are direct consequences of [19, Lemma 4.3], and Lemma 14 uses [19, Claim 3.5]; Theorem 11, used in Lemma 10, is [19, Theorem 1.3]. Because these estimates are essential for reducing the derivative sum to the net sum in Proposition 4, the manuscript is not self-contained and the referee cannot verify Proposition 4 from the present text alone. The authors should either state the needed results precisely, with proofs or precise references to statements in [19], or move the necessary portions of [19] into an appendix.","section":"§3–§4, Claims 13, 15, Lemma 14"}],"minor_comments":[{"comment":"Once the constant is corrected, the abstract and introduction should be updated so that the advertised numerical value matches the result actually proved.","section":"Abstract and Theorem 1"},{"comment":"The quantity \\tilde{s}_n(t) used in Eqs. (51)–(52) is not given a displayed definition; defining it explicitly as \\tilde{s}_n(t)=s''_n(t)-(s'_n(t))^2/s_n(t) before using it would improve readability.","section":"Notation around Eq. (47)"},{"comment":"The relation between Φ in Eq. (8) and S(t) in Eq. (46) is only implicit; writing Φ(t)=(1/t^2−1/sinh^2 t)\\log S(t) explicitly would help the reader track the constant computation.","section":"Eq. (8) and Claim 26"},{"comment":"The Erdős–Turán bound is quoted with a factor 2/π that is not explained; a reference to the exact form used would be helpful.","section":"Proof of Claim 16"},{"comment":"There are several typographical issues, including 'discrimiant' and the broken word 'POL YNOMIALS' in the title; these should be corrected in the final version.","section":"Throughout"}],"recommendation":"major_revision","confidential_remarks":"The factor-four error in Section 5.3 is serious because it directly invalidates the explicit constant in the main theorem, but the error is local and fixable, and the overall strategy appears sound. The heavy reliance on [19] is also a concern for self-containedness; if [19] is not yet published, the editors may wish to verify its status or require the authors to state the imported results. I do not recommend rejection, as the structural contributions are substantial and the constant can be recomputed."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"The main reduction in this paper is clean and the symmetrized Mahler-measure representation of log|Δ| is a real idea. Using reciprocal symmetry to equate the two derivative sums, then concentrating them on scale n via a net argument and Gaussian comparison, is a well-organized route from root statistics to a first-order law for the discriminant. That structure is new, and it replaces the linear-growth heuristic from [5, Section 4] with the correct n log n scale. Proposition 3, the Mahler-measure concentration, is also a solid standalone result.\n\nThe problem is the stated constant. The Schur complement in Claim 26 is off by a factor of 4. With G_n(t) = n^{-1/2} Σ γ_k e^{-tk/n}, the derivative has E|G'_n|^2 = s''_n/4 and E[G_n \\bar G'_n] = s'_n/2, so the conditional variance is (s''_n − (s'_n)^2/s_n)/4, not s''_n − (s'_n)^2/s_n. The paper's S(t) in Eq. (52) is therefore four times the true limit, and the log term should be log(S/4)+1−γ. The corrected limiting density changes D* from about 5.929 to about 1.34. I checked the scaling and the stress-test note is correct. This is a load-bearing error for the explicit constant, though the qualitative form of the LLN — |Δ| = n^{2n} e^{-D n(1+o(1))} — very likely survives with a different D.\n\nTwo smaller issues. The proof leans heavily on companion-paper results from [19] (Lemma 4.3, Theorem 1.3, Claim 3.5) that are not proved or fully stated here. This is a transparency problem: the concentration of the derivative sum depends on them, so a referee cannot fully check the paper without the companion. Also, the value of D* comes from a numerically evaluated integral, which deserves independent verification once the factor-4 issue is fixed.\n\nWho is this for? Random polynomial theorists and anyone using the discriminant as an algebraic invariant. The proof architecture is worth serious referee time, and the constant error looks fixable. I would send it to a knowledgeable referee and insist the Schur-complement computation be redone and the [19] imports be stated explicitly. Not a desk reject.","headline":"Genuinely useful proof architecture and a nice symmetrized Mahler-measure representation, but the explicit constant D* is wrong: a factor-4 slip in the conditional-variance computation changes D* from roughly 5.93 to about 1.34.","tokens_in":32247,"tokens_out":4146,"would_cite":false,"duration_ms":48553,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["60F05","30C15"],"pacs":[],"model":"deepseek-v4-flash","headline":"For random polynomials, the discriminant obeys a sharp law of large numbers: with high probability $|\\Delta(f_n)| = n^{2n}e^{-{\\sf D}_* n(1+o(1))}$ for an explicit universal constant ${\\sf D}_* \\approx 5.92947$.","keywords":["random polynomials","discriminant","law of large numbers","Kac polynomials","Mahler measure","Gaussian comparison","Kac-Rice formula","reciprocal symmetry"],"falsifier":"Run $f_n$ with standard Gaussian or Rademacher coefficients for $n$ large (say $10^4$ or beyond) and compute $(1/n)(\\log|\\Delta(f_n)| - 2n\\log n)$ over many samples: the law of large numbers predicts convergence to approximately $-5.92947$, so a stable deviation would refute the theorem. A sharper target is the annulus bound behind Proposition 4: for any sub-Gaussian law with $P(\\xi=0)=0$, the probability that $f_n$ has a root in $\\{1-\\log^3 n/n \\le |z| \\le 1\\}$ with $|f'_n(\\alpha)| \\le n^{5/4}/\\log^4 n$ is claimed to tend to zero (Claim 15); measuring this probability directly and finding it bounded away from zero would break the argument even if the final constant happened to agree.","tokens_in":2605,"feed_emoji":"🎲","tokens_out":2778,"duration_ms":157098,"temperature":0.7,"pith_summary":"This paper proves that the discriminant of a random polynomial — the algebraic product $A^{2n-2}\\prod_{i<j}(\\alpha_j-\\alpha_i)^2$ of all pairwise root differences — satisfies a law of large numbers: for independent, mean-zero, variance-one, sub-Gaussian coefficients, with high probability $|\\Delta(f_n)| = n^{2n}e^{-{\\sf D}_* n(1+o(1))}$, where ${\\sf D}_*\\approx 5.92947$ is an explicit universal constant. The $2n\\log n$ leading term is universal across all such coefficient distributions, and the result replaces the earlier heuristic that the log-discriminant grows only linearly. The discriminant matters because it is the invariant that detects double roots and squares, and for random integer polynomials its typical size and non-squareness underpin the conjecture that the Galois group is generically the full symmetric group. The proof centers on a symmetrized representation of $\\log|\\Delta|$ as a Mahler-measure term plus two derivative sums that are equal in law by reciprocal symmetry, each shown to concentrate by a net argument, Gaussian comparison, and an explicit Kac-Rice computation.","feed_headline":"n^{2n}e^{-5.93n}: the universal size of random discriminants","feed_subtitle":"This algebraic invariant, key to factoring random integer polynomials, follows a universal law of large numbers.","key_machinery":"The load-bearing identity is a symmetrized representation of the discriminant (Claim 2 and Lemma 5): for a degree-$n$ polynomial $P$ with simple roots and no root on the unit circle, $$\\log|\\$\\Delta$(P)| = \\sum_{|\\$\\alpha$|<1}\\log|P'(\\$\\alpha$)| + \\sum_{|\\$\\alpha$|>1}\\log\\left|\\frac{P'(\\$\\alpha$)}{\\$alpha^{{n-2}}$}\\right| + (n-2)\\$int_0^{1}$ \\log|P($e^{{2\\pi i\\theta}}$)|\\,d\\$\\theta$ .$$ The third term is the logarithmic Mahler measure. For the random polynomial $f_n$, the reciprocal polynomial $z^n f_n(z^{-1})$ has the same law as $f_n$, so the two derivative sums are equal in distribution (Claim 6); this distributional reciprocal symmetry is what cancels the non-universal contribution of roots far from the unit circle. Proposition 3 shows the Mahler measure concentrates at $-\\gamma/2$, and Proposition 4 shows each derivative sum concentrates at $c_*$; the latter is the technical bulk, carried by (i) removing the annulus away from the unit circle and the contribution of roots where $|f'_n|$ is atypically small, using companion-paper lemmas; (ii) a net of mesh $n^{-1-\\beta}$ in the near-circle annulus $\\{1-\\log^3 n/n \\le |z| \\le 1\\}$, where the linear approximation at net points predicts the roots; (iii) a Berry-Esseen Gaussian comparison showing the net sum is concentrated and close in law to the Gaussian coefficient case; and (iv) an exact Kac-Rice computation of the Gaussian mean, whose limiting density $\\Psi(t)$ carries the same radial prefactor $1/t^2 - 1/\\sinh^2 t$ as the expected root density of Kac polynomials. The constant assembles as $-{\\sf D}_* = -\\gamma/2 + 2c_*$, with $c_* = 1-\\gamma + \\int_0^\\infty \\Phi(t)\\,dt$.","core_discovery":"The paper's central discovery is that the logarithm of the discriminant of a random Kac polynomial concentrates on a scale far below its size, with a limiting value composed of exactly computable pieces. Theorem 1 states that $\\frac{1}{n}(\\log|\\Delta(f_n)| - 2n\\log n)$ converges in probability to $-{\\sf D}_*$, where ${\\sf D}_* = \\frac{\\gamma}{2} - 2(1-\\gamma) - 2\\int_0^\\infty \\Phi(t)\\,dt \\approx 5.92947$, with $\\gamma$ Euler's constant and $\\Phi$ the explicit function in (8). The constant assembles linearly: the Mahler measure of $f_n$ contributes $-\\gamma/2$ per degree (Proposition 3), and each of the two derivative sums — over roots with $|\\alpha|<1$ and their reciprocal partners with $|\\alpha|>1$ — contributes $c_* = 1-\\gamma + \\int_0^\\infty \\Phi(t)\\,dt$ (Proposition 4). The equality in law of the two sums (Claim 6) is what cancels the non-universal, distribution-dependent contribution of roots far from the unit circle, leaving a universal constant. The authors also record that the no-atom-at-zero assumption is inessential: the asymptotic holds except on the event, of asymptotic probability $P(\\xi=0)^2$, that the polynomial has a double root and the discriminant is exactly zero.","pith_inferences":["A testable extension the paper does not pursue: the rate of convergence to $-{\\sf D}_*$ is governed in the proof by the net error $n^{-\\beta/5}$, so coefficient laws with a near-atom at zero — legal sub-Gaussian laws with tiny $P(|\\xi|<\\varepsilon)$ — should show visibly slower approach to the limit, which one could check numerically.","The same symmetrized decomposition, Mahler measure plus two law-equal derivative sums, should transfer to other resultant-type quantities, such as the resultant of two independent random Kac polynomials or the discriminant of a polynomial with a planted root, yielding explicit exponential scales of the same shape.","Because the limiting density $\\Psi$ carries the Kac-polynomial radial root-density prefactor $1/t^2 - 1/\\sinh^2 t$, the constant ${\\sf D}_*$ is effectively an integral over root statistics; joint with the companion root-separation results, this suggests the next-order fluctuations of $\\log|\\Delta|$ should be expressible through the same Gaussian process and likely of order $n^{1/2}$ for Gaussian c"],"forward_implications":["With high probability $|\\Delta(f_n)| = n^{2n}e^{-{\\sf D}_* n(1+o(1))}$: the discriminant is enormous, and its logarithm is concentrated on a window of size $o(n)$ around $2n\\log n - {\\sf D}_* n$.","The $2n\\log n$ leading term is universal across all mean-zero, variance-one sub-Gaussian coefficient laws, correcting the linear-growth heuristic from earlier numerics on discriminants of random integer polynomials.","Relaxing $P(\\xi=0)=0$: except on the event, of asymptotic probability $P(\\xi=0)^2$, that $f_n$ has a double root — where $\\Delta = 0$ by definition — the same asymptotic holds.","The proof's Gaussian comparison also gives a template for the next order: the authors note that for Gaussian coefficients the variance of $\\log|\\Delta(f_n)|$ is expected to grow linearly, with asymptotically normal fluctuations.","The paper frames the theorem as a modest justification for numerical observations that discriminants of random integer polynomials are typically enormous and concentrated, the regime in which the Galois group is generically not the alternating group."],"supporting_citations":[{"why":"Companion-paper lemmas the argument leans on: Theorem 1.3 and Corollary 1.2 (the limiting series almost surely has no double zero; the double-root probability is $P(\\xi=0)^2$) and Lemma 4.3 (bounding the probability of a near-circle root with abnormally small derivative, used in Claims 13 and 15).","marker":"[19]"},{"why":"Universality of the minimum modulus of random trigonometric polynomials, used in Claim 7 to rule out roots on the unit circle and in the alternative proof of Proposition 3.","marker":"[9]"},{"why":"Numerical study of discriminants of random integer polynomials whose linear-growth heuristic the paper corrects and partially justifies.","marker":"[5]"},{"why":"Root clustering of Kac polynomials at scale $n^{-1}$ on the unit circle, used for the $2n\\log n$ leading term and to identify the radial prefactor of $\\Psi$.","marker":"[14]"},{"why":"Supplies the general Kac-Rice formula (Theorem 6.4) used to compute the Gaussian mean that yields $c_*$.","marker":"[2]"},{"why":"Berry-Esseen theorem for sums of independent random vectors (Corollary 17.2), on which the Gaussian comparison (Lemma 19) rests.","marker":"[7]"}],"fun_headline_variants":["Universal constant 5.93 for random polynomial discriminants","Random discriminants obey a universal law of large numbers","Universal size law for random discriminants","Random polynomial discriminants: a universal exponential law"],"cache_read_input_tokens":34304,"weakest_assumption_plain":"The load-bearing premise is that two lemmas from the companion paper hold for every sub-Gaussian coefficient law: a root very close to the unit circle is only rarely accompanied by an abnormally small derivative, and the limiting infinite power series almost surely has no double zero; the paper cites these lemmas rather than proving them, and if either fails for some admissible law, the concentration of the derivative sum and the value of ${\\sf D}_*$ collapse.","fun_headline_variants_meta":{"raw":{"variants":["Universal constant 5.93 for random polynomial discriminants","Random discriminants obey a universal law of large numbers","Universal size law for random discriminants","Random polynomial discriminants: a universal exponential law"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000891,"raw_usage":{"total_tokens":3880,"prompt_tokens":1020,"completion_tokens":2860,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":636,"completion_tokens_details":{"reasoning_tokens":2800}},"tokens_in":636,"tokens_out":2860,"duration_ms":25877,"temperature":1.0,"reasoning_tokens":2800,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-07T00:57:06.417449+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"Run $f_n$ with standard Gaussian or Rademacher coefficients for $n$ large (say $10^4$ or beyond) and compute $(1/n)(\\log|\\Delta(f_n)| - 2n\\log n)$ over many samples: the law of large numbers predicts convergence to approximately $-5.92947$, so a stable deviation would refute the theorem. A sharper target is the annulus bound behind Proposition 4: for any sub-Gaussian law with $P(\\xi=0)=0$, the probability that $f_n$ has a root in $\\{1-\\log^3 n/n \\le |z| \\le 1\\}$ with $|f'_n(\\alpha)| \\le n^{5/4}/\\log^4 n$ is claimed to tend to zero (Claim 15); measuring this probability directly and finding it bounded away from zero would break the argument even if the final constant happened to agree.","supporting_citations":[{"cited_title":"Limit law for root separation in random polynomials","cited_arxiv_id":"2505.02723","evidence_quote":"Companion-paper lemmas the argument leans on: Theorem 1.3 and Corollary 1.2 (the limiting series almost surely has no double zero; the double-root probability is $P(\\xi=0)^2$) and Lemma 4.3 (bounding the probability of a near-circle root with abnormally small derivative, used in Claims 13 and 15)."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Universality of the minimum modulus of random trigonometric polynomials, used in Claim 7 to rule out roots on the unit circle and in the alternative proof of Proposition 3."},{"cited_title":"Bary-Soroker and G","cited_arxiv_id":null,"evidence_quote":"Numerical study of discriminants of random integer polynomials whose linear-growth heuristic the paper corrects and partially justifies."},{"cited_title":"Ibragimov and O","cited_arxiv_id":null,"evidence_quote":"Root clustering of Kac polynomials at scale $n^{-1}$ on the unit circle, used for the $2n\\log n$ leading term and to identify the radial prefactor of $\\Psi$."},{"cited_title":"Aza ¨ ıs and M","cited_arxiv_id":null,"evidence_quote":"Supplies the general Kac-Rice formula (Theorem 6.4) used to compute the Gaussian mean that yields $c_*$."},{"cited_title":null,"cited_arxiv_id":null,"evidence_quote":"Berry-Esseen theorem for sums of independent random vectors (Corollary 17.2), on which the Gaussian comparison (Lemma 19) rests."}],"review_version":1}