{"id":"bd3e7e07-f393-4880-8c1e-3e84c936a3d1","arxiv_id":"2608.10072","paper_version":1,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":1,"one_line_summary":"Recursion coefficients for high-degree asymmetric polynomial random matrix models are computed efficiently via a moment recursion, with large-n asymptotics reproducing Freud's conjecture and transition regions mapped in quartic and double-scaled SYK models.","lead":"This paper develops a faster way to compute the recursion coefficients that control how quantum states spread in random matrix models with high-degree or asymmetric potentials. The method is applied to the double-scaled SYK model, a common stand-in for quantum gravity, revealing transition regions where the coefficients turn irregular.","discovery_kind":"new_method","skeptic_critique":{"model":"deepseek-v4-flash","headline":"Eq. (3.13)-(3.14) rest on the unproven continuum assumption (3.5); neither the one-cut restriction nor validity for all asymmetric polynomial potentials is established.","rationale":"The reader's weakest_assumption correctly identifies the unproven convergence assumption in Eq. (3.5) as the soft spot of the central claim. My reading confirms that the derivation of Eq. (3.13)-(3.14) is algebraically sound given that assumption, but the assumption itself is neither proved nor derived, and it imports a one-cut, hydrodynamic picture whose domain of validity is not stated. The numerical evidence in the paper covers a quartic potential and a truncated symmetric DSSYK potential; neither tests the asymmetric high-degree case in which both w_d and w_{d-1} are nonzero and the one-cut condition may fail. The DSSYK truncation sensitivity, explicitly flagged by the authors, further weakens the paper's ability to use that model as evidence for the general formula. These points reinforce the reader's conditional verdict rather than overturning it: the central formula is plausible and partially verified, but the universal claim is not yet settled. No change to the verdict is warranted.","tokens_in":18598,"tokens_out":15309,"duration_ms":146386,"concrete_test":"Choose a genuinely asymmetric sextic potential, e.g. V(lambda)=lambda^6 + w_5 lambda^5 + w_1 lambda with w_5=1 and w_1=2. For a fixed moderate N (say N=20), compute R_n and S_n by the moment recursion (Eq. 3.3) plus the recursive algorithm (2.23) in high-precision arithmetic for n up to at least 10^4, and test whether R_n/R_infinity(n) and S_n/S_infinity approach 1 (Eq. 3.13). Additionally, to isolate assumption (3.5) from the algebraic step, test the continuum limit directly: for x=10 and x=100, compute R_{floor(xN)}/R(x) and S_{floor(xN)}/S(x) at N=100, 200, 400 and check convergence to 1. If either test fails, Eq. (3.13)-(3.14) is not valid for general asymmetric polynomial potentials, and the claimed universality is restricted to the tested cases.","verdict_should_be":"UNCHANGED","load_bearing_attack":"Equation (3.14) is obtained by evaluating the continuum string equations (2.19) under the assumption (3.5): R_n/R(x)->1 and S_n/S(x)->1 with x=n/N, plus R(x) unbounded and S(x) bounded. This assumption is stated in Section 3.2 without proof or derivation. It is not an automatic consequence of the large-N limit used to derive (2.19): that limit fixes x and sends N to infinity, whereas the desired n-to-infinity asymptotic at fixed N (or the x-to-infinity limit) is a different order of limits. Moreover, the continuum string equations (2.18)-(2.19) rest on a single-interval (one-cut) ansatz for the eigenvalue density; for general polynomial potentials admitting multi-cut equilibrium measures, there is no justification that the same R(x), S(x) describe the discrete recursion coefficients. The paper does not delimit the validity of (3.5), and the numerical checks cover only the one-cut quartic (d=4, with one set of parameters) and the symmetric truncated DSSYK (d=18, w_{d-1}=0). In the latter case the authors themselves note (footnote 2, Section 5) that the recurrence coefficients are highly sensitive to the truncation order and that the asymptotic behavior is fixed by the highest retained term, so this test does not exercise the S_n asymptotics at all. Consequently, the central claim that (3.14) holds for general normalizable polynomial measures is not established; it is conditional on an assumption that has only been tested in narrow, partly self-selected examples.","agreement_with_reader":"agree"},"referee_report":{"model":"deepseek-v4-flash","summary":"The paper studies recursion coefficients (equivalently Lanczos coefficients) of orthogonal polynomials associated with random matrix ensembles with polynomial potentials. It proposes a moment recursion method for computing these coefficients, derives large-n asymptotic formulas for general polynomial potentials — claiming R_n ~ n^{2/d}[N w_d Γ(d+1)/(Γ(d/2)Γ(d/2+1))]^{-2/d} and S_n → -w_{d-1}/(d w_d) — and tests these formulas against an asymmetric quartic potential and a truncated Chebyshev model of the double-scaled SYK model. It also connects gradient catastrophes of the continuum recursion functions to 'chaotic transition regions' in the discrete coefficients and computes spread complexity in both models.","tokens_in":18840,"tokens_out":7895,"duration_ms":72166,"significance":"If the asymptotic formulas are correct, they give a simple, parameter-free characterization of the leading large-n growth of Lanczos coefficients for arbitrary polynomial weights, and the R_n result recovers Freud's conjecture under the normalization N w_d = 1. The moment recursion method is a practical technical contribution, and the paper is transparent about the truncation sensitivity in the DSSYK section. The numerical checks for the quartic model across several N values are credible. However, the central asymptotic derivation relies on an unproven convergence assumption, so the paper's main claim is not fully established as stated.","major_comments":[{"comment":"The asymptotic formulas (3.13)–(3.14) rest on the assumption that R_n/R(x)→1 and S_n/S(x)→1 with x=n/N, together with R(x) diverging and S(x) bounded as x→∞. This assumption is asserted without proof or derivation, and it is not an automatic consequence of the large-N limit used to obtain the continuum string equations (2.19): that limit fixes x and sends N to infinity, whereas (3.5) concerns n→∞ for a fixed value of the ratio x, or equivalently an x→∞ limit. The paper should either prove (or cite a proof of) this convergence for normalizable polynomial potentials, or explicitly present (3.13)–(3.14) as conjectural. As written, the central claim is conditional on an assumption that has only been checked numerically in a narrow set of examples.","section":"Section 3.2, Eq. (3.5)"},{"comment":"The continuum string equations used in the derivation assume a single-interval support for the spectral density (the one-cut ansatz). The paper claims the asymptotic results hold for general polynomial potentials, but it does not state whether the one-cut condition is assumed. For potentials whose equilibrium measure is multi-cut, the derivation of (2.19) is not valid, and no argument is given that (3.13)–(3.14) nevertheless survive. The numerical evidence covers only a one-cut quartic and a symmetric truncated polynomial, so the general claim is not tested outside the one-cut regime.","section":"Section 2.2, Eq. (2.19); Section 3.2"},{"comment":"The DSSYK analysis truncates the infinite Chebyshev expansion at d=18, and the authors note that the recursion coefficients are highly sensitive to the truncation order, with the large-n asymptotics determined by the highest retained monomial. Consequently, the verification in Fig. 6 tests the truncated polynomial, not the actual DSSYK weight, and the DSSYK section does not provide evidence for the validity of (3.13) for the non-polynomial DSSYK potential. The manuscript should clarify the precise scope of the DSSYK claim, either by adding a truncation-convergence study or by explicitly restricting the claim to the truncated model.","section":"Section 5, Eq. (5.3) and footnote 2"}],"minor_comments":[{"comment":"When w_{d-1}=0, one has S_∞=0, and the expression lim S_n/S_∞ = 1 is undefined; the condition S_∞≠0, or an alternative formulation of the S_n limit, should be stated.","section":"Eq. (3.14)"},{"comment":"The remark that for 0<x<1 Eq. (2.19) also describes the average Lanczos coefficients upon reversing the coordinate x→1−x is stated without derivation; since this claim is used again in Section 5, a brief explanation or reference would be helpful.","section":"Section 2.2, after Eq. (2.19)"},{"comment":"Labels such as 'β=√−γ2' are opaque because γ2 is negative; writing β=√(−γ2) would be clearer.","section":"Figure 2"},{"comment":"The moment recursion divides by d w_d, so the method requires w_d≠0; this should be stated explicitly.","section":"Section 3.1, Eq. (3.3)"},{"comment":"The phrase 'general polynomial potentials' is used, but the paper treats polynomial potentials of finite degree d with leading coefficient w_d and the normalizability condition (2.2); the scope should be qualified accordingly.","section":"Abstract and Introduction"}],"recommendation":"major_revision","confidential_remarks":"The central asymptotic result for R_n is essentially Freud's conjecture for exponential weights, and the mathematical literature (e.g., Lubinsky, Mhaskar, and Saff) may provide a rigorous justification that the authors could cite or adapt. The S_n asymptotics are also standard from equilibrium measure theory. In the revision, the authors should either connect their derivation to these known results or clearly mark the continuum convergence assumption as a conjecture. The DSSYK truncation issue further limits the scope of the claimed application."},"author_rebuttal":null,"desk_editor":{"model":"deepseek-v4-flash","letter":"Two things worth knowing. The paper's real product is the large-n asymptotic for Lanczos/recursion coefficients of polynomial matrix models with asymmetric potentials, Eqs. (3.13)-(3.14). That formula is clean, plausible, and reduces to Freud's conjecture when N w_d = 1, which is a good external sanity check. The second product is the observation that gradient catastrophes in the continuum recursion functions R(x), S(x) track where the discrete coefficients enter chaotic transition regions. That is illustrated for the asymmetric quartic and for truncated DSSYK, and the spread complexity follow-up is honest: transition regions do not qualitatively change the complexity. The moment recursion is not deep - integration by parts plus the classical recursive algorithm - but it is practical for high-degree potentials, and the paper deserves credit for spelling it out. Main soft spot: the asymptotic derivation is conditional. Eq. (3.5) is stated as an assumption, and it is doing the work: the continuum equations (2.19) come from a large-N limit at fixed x = n/N, while the desired statement is n going to infinity at fixed N, or at least along a different order of limits. The paper does not prove that the discrete coefficients converge to those R(x), S(x) functions. Relatedly, (2.19) comes from a one-cut ansatz, so the general claim for all normalizable polynomial potentials in (3.14) is not actually established for multi-cut cases. I do not think this is fatal - the result is likely correct and the quartic numerics support it - but it should be labeled as a conjecture or backed by a reference to the Freud-equation literature, not presented as a derivation. The DSSYK section is the weakest part, exactly as the reader says. The potential is truncated at d = 18, and footnote 2 admits the recursion coefficients are highly sensitive to truncation order, with asymptotics fixed by the highest retained term. So the multiple transition regions at various q are properties of the truncated model until tested against increasing d or the full potential. The authors flag this, so it is not dishonest, but it limits the phenomenological claim. No circularity problem: no fitted parameters, and Freud's conjecture is an external benchmark. The citation pattern is fine; citing the author's own prior work for the continuum equations is appropriate because that is where those equations live. Bottom line: this is a useful methods paper for the Krylov/random-matrix subfield. It deserves a serious referee. I would send it out and ask the authors to soften the general claim, add a convergence argument or pointer, and add a d-dependence check for DSSYK. Not desk-reject material.","headline":"Useful large-n asymptotics for recursion coefficients of asymmetric polynomial potentials, but the central formula rests on an explicit convergence assumption and the DSSYK transition regions are truncation-sensitive.","tokens_in":799,"tokens_out":1729,"would_cite":true,"duration_ms":50526,"reading_group":"maybe","serious_thinker":"yes","would_accept_peer_review":true},"rs_alignment":null,"lean_confirmation":null,"pith_extraction":{"msc":["42C05","15B52","60B20"],"pacs":[],"model":"deepseek-v4-flash","headline":"The leading large-n growth of Lanczos coefficients in any polynomial random matrix model is fixed by the highest-degree monomial of the potential, while the next monomial fixes their limiting value.","keywords":["orthogonal polynomials","recursion coefficients","Lanczos coefficients","random matrix theory","Freud conjecture","spread complexity","double-scaled SYK model","moment recursion method"],"falsifier":"For the asymmetric quartic potential $V(\\lambda)=\\lambda^4+w_3\\lambda^3+w_2\\lambda^2+w_1\\lambda$, compute $R_n$ and $S_n$ by high-precision moment recursion out to $n/N\\sim 10^3$ and test whether $R_n/\\sqrt{n/(12N)}$ tends to 1 and $S_n$ tends to $-w_3/4$; a stable deviation from either limit would falsify the universal asymptotics.","tokens_in":18282,"feed_emoji":"📈","tokens_out":13665,"duration_ms":114436,"temperature":0.7,"pith_summary":"The paper derives the large-$n$ behavior of the recursion coefficients—equivalently the Lanczos coefficients of Krylov dynamics—for orthogonal polynomials associated with a general polynomial random matrix potential $V(\\lambda)=\\sum_{m=0}^d w_m\\lambda^m$, including asymmetric potentials. It claims that $R_n$ grows like $n^{2/d}$, with a prefactor fixed entirely by the degree $d$ and the top coefficient $w_d$, while $S_n$ tends to $-w_{d-1}/(d\\,w_d)$; setting $Nw_d=1$ reproduces Freud's conjecture. The paper also develops a moment recursion method that computes the coefficients efficiently for high-degree potentials by iterating a closed relation among the moments, avoiding the cumbersome discrete string equations. Numerical checks on an asymmetric quartic potential and on the double-scaled SYK model confirm the formulas, and the paper finds that transition regions in the coefficients do not qualitatively change spread complexity, whereas a two-branch structure produces early-time oscillations.","feed_headline":"One law governs Lanczos coefficients in polynomial matrix models","feed_subtitle":"The top potential term sets growth; asymmetry only sets the final value. Recovers Freud's conjecture.","key_machinery":"The load-bearing objects are the continuum string equations (2.19), algebraic equations for the continuum recursion functions $R(x)$ and $S(x)$ obtained as the large-$N$ limit of the discrete string equations; under the convergence assumption $R_n/R(x)\\to 1$ and $S_n/S(x)\\to 1$ with $x=n/N$, the dominant large-$x$ terms give the asymptotic formulas. The second mechanism is the moment recursion (3.3), a total-derivative identity that expresses every higher moment of a polynomial weight in terms of lower moments, which combined with the recursive algorithm (2.23) turns the computation of recursion coefficients for high-degree potentials into a numerically tractable iteration.","core_discovery":"The central discovery is that the large-$n$ limit of the recursion coefficients is universal and simple: for any normalizable polynomial potential of degree $d$, $R_n \\sim n^{2/d}\\left[N w_d \\Gamma(d+1)/(\\Gamma(d/2)\\Gamma(d/2+1))\\right]^{-2/d}$ and $S_n\\to -w_{d-1}/(d\\,w_d)$. The leading growth is controlled solely by the highest monomial, and the next-highest monomial fixes the limiting value of $S_n$. These formulas follow from the continuum string equations by keeping the dominant terms as $x=n/N\\to\\infty$ under the assumption that the discrete coefficients converge to the continuum recursion functions. For $Nw_d=1$, the $R_n$ formula reduces to Freud's conjecture, and in the DSSYK model the same framework locates multiple transition regions in $R_n$ while the recursion function stays accurate in the smooth intervals between them.","pith_inferences":["Beyond the paper: because the asymptotic formula depends only on the top-degree monomial, the large-$n$ behavior of Lanczos coefficients may be set by the spectral edge of the limiting eigenvalue density rather than by the full shape of the potential.","Beyond the paper: the moment recursion (3.3) could be applied to non-polynomial weights by truncating a high-order polynomial approximation of the potential, with the truncation degree controlling the asymptotic exponent—a sensitivity the paper notes for the DSSYK model.","Beyond the paper: the coincidence between gradient catastrophes and transition regions yields a testable prediction for other potentials: the onset of chaotic-looking oscillations in recursion coefficients should coincide with a turning point of the corresponding continuum string equation."],"forward_implications":["In any polynomial matrix model of degree $d$, leading Lanczos growth has the universal exponent $2/d$; lower-degree terms and potential asymmetry do not affect the leading growth.","Under $Nw_d=1$, the leading formula is exactly Freud's conjecture, so the paper unifies classical symmetric-potential results and extends them to asymmetric potentials.","The continuum recursion functions remain accurate approximations on smooth intervals, so the parameter space of the models can be classified by the number of gradient catastrophes of the recursion functions.","Transition regions in the recursion coefficients do not by themselves change the qualitative behavior of spread complexity, whereas a two-branch structure of the coefficients introduces early-time oscillations followed by monotonic growth."],"supporting_citations":[{"why":"Supplies the continuum string equations used to extract the large-x behavior.","marker":"[11]"},{"why":"Earlier work identifying recursion coefficients with Lanczos coefficients and deriving the continuum string equations; the basis for the asymptotic derivation.","marker":"[12]"},{"why":"Freud's conjecture, the classical statement whose leading form is recovered under Nw_d=1.","marker":"[14]"},{"why":"The recursive algorithm that turns moments into recursion coefficients, used together with the moment recursion.","marker":"[91]"},{"why":"Proof of Freud's conjecture for exponential weights, establishing the result that the Nw_d=1 formula reproduces.","marker":"[95]"},{"why":"Exact Chebyshev expansion of the double-scaled SYK potential used to set up the high-degree numerical model.","marker":"[98]"},{"why":"Provides the average Lanczos coefficients of the DSSYK model that the recursion functions reproduce for 0<x<1.","marker":"[99]"},{"why":"Constructs tridiagonal Hamiltonians modeling the DSSYK density of states, the comparison object for the recursion coefficients.","marker":"[100]"}],"fun_headline_variants":["Universal formula for Lanczos coefficients in matrix models","Asymmetry only sets final value; growth is universal","Freud's conjecture recovered in polynomial matrix models","One law for Lanczos coefficients: top monomial rules","Universality: growth from top monomial, offset from asymmetry"],"cache_read_input_tokens":3200,"weakest_assumption_plain":"The derivation assumes that the discrete recursion coefficients settle smoothly onto the continuum recursion functions as $n$ and $N$ grow with $x=n/N$ fixed; if that convergence fails, the asymptotic formulas are not justified.","fun_headline_variants_meta":{"raw":{"variants":["Universal formula for Lanczos coefficients in matrix models","Asymmetry only sets final value; growth is universal","Freud's conjecture recovered in polynomial matrix models","One law for Lanczos coefficients: top monomial rules","Universality: growth from top monomial, offset from asymmetry"]},"model":"deepseek-v4-flash","effort":"low","cost_usd":0.000837,"raw_usage":{"total_tokens":3656,"prompt_tokens":955,"completion_tokens":2701,"prompt_tokens_details":{"cached_tokens":384},"prompt_cache_hit_tokens":384,"prompt_cache_miss_tokens":571,"completion_tokens_details":{"reasoning_tokens":2623}},"tokens_in":571,"tokens_out":2701,"duration_ms":20256,"temperature":1.0,"reasoning_tokens":2623,"cache_read_input_tokens":384,"cache_creation_input_tokens":0},"cache_creation_input_tokens":0},"created_at":"2026-08-14T04:14:55.538512+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":"For the asymmetric quartic potential $V(\\lambda)=\\lambda^4+w_3\\lambda^3+w_2\\lambda^2+w_1\\lambda$, compute $R_n$ and $S_n$ by high-precision moment recursion out to $n/N\\sim 10^3$ and test whether $R_n/\\sqrt{n/(12N)}$ tends to 1 and $S_n$ tends to $-w_3/4$; a stable deviation from either limit would falsify the universal asymptotics.","supporting_citations":[],"review_version":1}