betaFeedback

WorksMathematics

An explicit counterexample of degree 29440 to Smale's mean value problem with constant one

Abstract

For a complex polynomial p with p(0)=0 and p'(0)=1, let S(p,0) be the minimum of |p(c)/c| over the critical points c of p. Adamczewski and Colbrook recently gave an explicit polynomial of degree 29984 with S(p,0)>1, certified by interval arithmetic, built on the counterexample found by GPT-6 Astra. Within their family of polynomials, with re-chosen coefficients, we give an explicit p in Q(i)[z] of degree 29440 with 1.00000093069254310 < S(p,0) < 1.00000093069254311. Jatar and Ng have shown that Smale's constant sup_p S(p,0) exceeds 1.1018, through polynomials of unspecified degree; the point here is an explicit polynomial with a certified bound. The bound is certified twice: by a ball-arithmetic Krawczyk certificate covering all 29439 critical points, and, independently, by the Adamczewski-Colbrook verifier, which confirms 29418 of them and is inconclusive (insufficient working precision) on the other 21. The exact coefficients and both verification programs are provided as ancillary files.

Introduction

Let pp be a complex polynomial of degree n≥2n\ge2 and a∈Ca\in\mathbb{C} with p′(a)≠0p'(a)\ne0. Smale [5] proved that

S(p,a)=min⁡p′(c)=0∣p(c)−p(a)(c−a)p′(a)∣≤4S(p,a)=\min_{p'(c)=0}\left|\frac{p(c)-p(a)}{(c-a)p'(a)}\right|\le4

and asked whether the constant 4 can be replaced by 1, or even by (n−1)/n(n-1)/n. The constant-one version was disproved by GPT-6 Astra, with a formal Lean 4 proof [1]. Adamczewski and Colbrook [2] explained the underlying mechanism and gave an explicit polynomial p∈Q(i)[z]p\in\mathbb{Q}(i)[z] of degree 29 984 with S(p,0)>1.00000049595801S(p,0)>1.00000049595801, certified by interval arithmetic [3]. Independently, Jatar and Ng [4] reduced the determination of Smale’s mean value constant M=sup⁡S(p,0)M=\sup S(p,0), the supremum over normalized polynomials, to an extremal problem for admissible pairs of a univalent function and a finite Blaschke product, and showed that M>1.1018M>1.1018; their polynomials come from an approximation argument, and no explicit degree is given. Thus the value 1.00000093 . . . below is not a bound on MM; what is new here is only the degree of an explicit, certified example.

This note records a smaller explicit counterexample. It uses the same construction as [2 Appendix A]: the same lemniscate coordinate, model anchor, Blaschke parameter, correction centres and correction structure. Only the truncation level NN, the size of the high-degree tail and the coefficient tables are changed. The reduction in degree is modest (about 1.8%); no new mechanism is claimed.

The polynomial

We keep the notation of [2 Appendix A]: m=16m=16, A=1061603/250000A=1061603/250000, S=A+1/AS=A+1/A, q(z)=(z16+iA)/Sq(z)=(z^{16}+iA)/S, the model anchor aa, the Blaschke parameter rr and the truncated model hLh_L of [2 (A.5)]. We take

N=1840,L=N−2=1838,tail blocks t=48,N = 1840,\qquad L = N - 2 = 1838,\qquad\text{tail blocks }t = 48,

and set

P(z)=(z−a)hL(z)hL(a)+Δ0(z)+X(z)+H(z),P(z) = (z-a)\frac{h_L(z)}{h_L(a)}+\Delta_0(z)+X(z)+H(z),
Δ0=(z−a)2(KR0+∑j=03ϕj(z−μ)j),K=(z−ξ1)2(z−ξ2)2,μ=ξ1+ξ22,\Delta_0=(z-a)^2\left(KR_0+\sum_{j=0}^{3}\phi_j(z-\mu)^j\right),\qquad K=(z-\xi_1)^2(z-\xi_2)^2,\qquad\mu=\frac{\xi_1+\xi_2}{2},
X=(z−a)2∑ℓ=0127∑k=015xℓk(5z6)kq(z)ℓ,X=(z-a)^2\sum_{\ell=0}^{127}\sum_{k=0}^{15}x_{\ell k}\left(\frac{5z}{6}\right)^kq(z)^\ell,
H=ω(γq(z)N+q(z)N−t∑ℓ=0t−1∑k=015eℓk(5z6)kq(z)ℓ),H=\omega\left(\gamma q(z)^N+q(z)^{N-t}\sum_{\ell=0}^{t-1}\sum_{k=0}^{15}e_{\ell k}\left(\frac{5z}{6}\right)^kq(z)^\ell\right),

with R0R_0 the correction polynomial of [2] (its 1024 coefficients re-chosen), rational ω\omega with ∣ω∣=1\lvert\omega\rvert=1 exactly, rational γ>0\gamma>0, and a rational base point a^\hat{a}. The normalized polynomial is

p(x)=P(a^+x)−P(a^)P′(a^).p(x)=\frac{P(\hat{a}+x)-P(\hat{a})}{P'(\hat{a})}.

All data are exact elements of Q\mathbb{Q} or Q(i)\mathbb{Q}(i) and are listed in the ancillary file polynomial_N1840.json, which follows the format of [3] (field defining formula). In particular p∈Q(i)[z]p\in\mathbb{Q}(i)[z].

Only γqN\gamma q^N reaches degree 16N=29 44016N=29\,440; all other terms have degree at most 29 439, and the leading coefficient ωγ/SN\omega\gamma/S^N is nonzero. Hence deg⁡p=29 440\deg p=29\,440, and p(0)=0p(0)=0, p′(0)=1p'(0)=1 by construction.

Theorem 1. The polynomial p∈Q(i)[z]p\in\mathbb{Q}(i)[z] above has degree 29 440, its 29 439 critical points are simple, and

1.00000093069254310<S(p,0)<1.00000093069254311.1.00000093069254310 < S(p,0) < 1.00000093069254311.

In particular pp is a counterexample to Smale’s mean value conjecture with constant one.

Remark 2. Adamczewski and Colbrook note [2 §1.1] that perturbations εzn\varepsilon z^n of an explicit counterexample, together with Rouché’s theorem, give counterexamples in every larger degree. Applied to pp, the same argument would give counterexamples in every degree n≥29 440n\geq29\,440; we do not repeat it here, and Theorem 1 only concerns degree 29 440.

Verification

Since S(p,0)S(p,0) is unchanged by the affine normalization, it suffices to work with PP and a^\hat{a}: S(p,0)=min⁡P′(c)=0∣Q(c)∣S(p,0)=\min_{P'(c)=0}\lvert Q(c)\rvert with

Q(c)=P(c)−P(a^)(c−a^)P′(a^).Q(c)=\frac{P(c)-P(\hat{a})}{(c-\hat{a})P'(\hat{a})}.

Certificate 1 (ball arithmetic)

Certificate 1 (ball arithmetic). Arithmetic is done with complex balls (Arb, via python-flint) at 1280 bits, starting from the exact rationals of the specification. For each of 29 439 approximate critical points:

  1. three Newton steps for P′P' give a centre zz;

  1. on the square box DD of half-side ρ\rho about zz, an enclosure of P′′(D)P''(D) is obtained from P′′(z)P''(z) and a Cauchy bound sup⁡∣P′′′∣≤6B/(R−ρ2)3\sup\lvert P'''\rvert\leq6B/(R-\rho\sqrt{2})^3, where BB is a rigorous majorant of ∣P∣\lvert P\rvert on the disc of radius RR about zz (term-by-term triangle inequality);

  1. the Krawczyk operator K(D)=z−P′(z)/P′′(z)+(1−P′′(D)/P′′(z))(D−z)K(D)=z-P'(z)/P''(z)+(1-P''(D)/P''(z))(D-z) is shown to lie in the interior of DD, which proves that DD contains exactly one zero of P′P', and that it is simple;

(4) ∣Q∣|Q| is bounded below on all of DD by a second-order Taylor enclosure of PP with the same remainder bound, and a point counts as certified only if this lower bound exceeds 1. All 29 439 points pass. The boxes have half-side at most 7.4⋅10−407.4 \cdot10^{-40}, the refined centres differ from the supplied ones by at most 7.7⋅10−127.7 \cdot10^{-12}, and any two supplied centres are at distance at least 6.05⋅10−56.05 \cdot10^{-5}; hence the boxes are pairwise disjoint. They thus contain 29 439 =deg⁡P′=\deg P' distinct zeros of P′P', i.e. all critical points, each simple. The smallest certified value is

min⁡P′(c)=0∣Q(c)∣≥1.0000009306925431017438…\min_{P'(c)=0} |Q(c)| \ge1.0000009306925431017438\ldots

attained in a single box, which also gives the upper bound of Theorem 1.

Certificate 2 (Adamczewski–Colbrook verifier)

Certificate 2 (Adamczewski–Colbrook verifier). The program verify.cpp of [3] (outward-rounded interval bounds with MPFR-directed endpoints, Rouché test on critical disks) was recompiled with N_PARAM=1840, TAIL_BLOCKS_PARAM=48, the expected parameters in make_parameters.py being read from the specification; the specification passes all its exact-rational consistency checks. Run disk by disk on the 29 439 critical disks, it certifies ∣Q∣>1|Q| > 1 on 29 418 of them, with smallest lower bound 1.000000930692128…\ldots, on the same critical point as Certificate 1. On the remaining 21 disks (indices, in the order of critical_disks_N1840.csv, listed in the ancillary summary) its fixed working precision does not suffice to conclude; these are covered by Certificate 1 only.

[2]this note
NN18741840
degree29 98429 440
tail blocks848
certified lower bound for S(p,0)S(p,0)1.000000495958011.00000093069254310

Table 2.

Ancillary files

  • polynomial_N1840.json: exact specification (QQ and Q(i)Q(i) data).

  • centres_N1840.csv: the 29 439 supplied critical-point centres, sorted by increasing ∣Q∣|Q|.

  • verify_ball_N1840.py: Certificate 1 for any range of critical points (Python with python-flint and numpy).

  • certificate_N1840.log: summary of both certificates.

  • critical_disks_N1840.csv, make_parameters_N1840.py: input disks and the modified parameter script used to run the verifier of [3] (Apache License 2.0).

Status and acknowledgements

Theorem 1 is a computer-assisted result: its proof is the pair of certificates above, which anyone can rerun from the ancillary files. The construction is entirely that of Adamczewski and Colbrook [2], building on the counterexample found by GPT-6 Astra [1]; only parameters were changed. The computations were carried out with the assistance of Claude (Anthropic).

References

Meaux, France

Email address: [email protected]

References

  1. [1]T. Adamczewski, Mean-value-problem: Lean 4 disproof of Smale’s mean value conjecture (K = 1), GitHub repository, 2026.
  2. [2]T. Adamczewski and M. J. Colbrook, From boundary geometry to counterexamples in Smale’s mean value problem, arXiv:2610.36158, 2026.arxiv.org/abs/2610.36158
  3. [3]M. J. Colbrook, Interval_arithmetic_counterexample_Smale_mean_value, GitHub repository, 2026.
  4. [4]A. Jatar and T. W. Ng, Smale’s mean value constant and admissible univalent Blaschke pairs, arXiv:2610.21472, 2026.arxiv.org/abs/2610.21472
  5. [5]S. Smale, The fundamental theorem of algebra and complexity theory, Bull. Amer. Math. Soc. (N.S.) 4 (1981), 1–36.

Paper details

vlnc:W_wt76cvek
Version
v1 · latest
Published
Fresh — published within the last 30 days
License
CC BY 4.0
Usage

views · 12

downloads · 0

A view counts once per browser per day, a download each time the PDF is requested.

  • Mathematics

Anthropic Claude Opus

View statement
Looking for related papers…
Contents