arxiv: 2603.21453 · v3 · submitted 2026-03-23 · 🧮 math.CA

Recognition: 2 theorem links

· Lean Theorem

Local Bernstein theory, and lower bounds for Lebesgue constants

Terence Tao

Authors on Pith no claims yet

Pith reviewed 2026-05-15 01:26 UTC · model grok-4.3

classification 🧮 math.CA

keywords Bernstein theoryLebesgue constantspolynomial interpolationErdős lower boundholomorphic functionsresidue theoremlocalization

0 comments

The pith

Localizing Bernstein bounds to rectangles lets Erdős lower bounds on Lebesgue constants apply to shorter intervals and in integral form.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that Bernstein-type inequalities continue to control the growth of certain holomorphic functions inside rectangles sitting above an interval I, provided the functions stay bounded and real on the base, grow at most exponentially on the top, and at most double-exponentially on the sides. These localized bounds are then used to restrict the classical Erdős lower bound on the Lebesgue constant of polynomial interpolation from the full interval [-1,1] down to arbitrary subintervals I. The same technique produces an asymptotically sharp integral lower bound on the Lebesgue function over I. A reader cares because the result sharpens the known limits on how small the Lebesgue constant can be when one only needs good approximation on a smaller piece of the domain.

Core claim

Classical global Bernstein theory establishes sharp control on entire functions of exponential type that are bounded and real-valued on the real axis. The paper localizes some of this theory to rectangular regions {x+iy: x in I, 0 ≤ y ≤ y0}, showing that Bernstein-type bounds with acceptable errors can continue to hold for functions holomorphic in such rectangles, bounded and real-valued on the lower edge, at most exponentially large on the upper edge, and at most double exponentially large on the vertical sides. As a consequence, the Erdős lower bound sup λ(x) ≥ (2/π) log n - O(1) localizes to shorter intervals I, and the integral lower bound ∫_I λ(x) dx ≥ (4|I|/π²) log n - o(log n) holds.

What carries the argument

Localized Bernstein bounds for holomorphic functions in rectangles with controlled exponential growth on three sides and real values on the base, obtained via weighted applications of the residue theorem.

If this is right

The pointwise lower bound sup_{x in I} λ(x) ≥ (2/π) log n - O(1) holds for any fixed subinterval I of [-1,1].
The integral lower bound ∫_I λ(x) dx ≥ (4|I|/π²) log n - o(log n) is asymptotically sharp.
Both results answer questions posed by Erdős and Turán on localization of Lebesgue-constant inequalities.
The same rectangle-based argument supplies local versions of other Bernstein-type estimates in approximation theory.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

The rectangle localization technique could be tested numerically on explicit Chebyshev or Legendre interpolation nodes to measure how quickly the o(log n) term vanishes for moderate n.
Similar growth-control arguments might extend the integral lower bound to weighted Lebesgue constants or to interpolation on arcs rather than intervals.
If the double-exponential side growth can be relaxed further, the method might apply to even narrower rectangles and produce sharper constants in local approximation.

Load-bearing premise

The functions under consideration remain holomorphic inside the rectangle, real-valued on the lower edge, at most exponentially large on the upper edge, and at most double-exponentially large on the vertical sides.

What would settle it

A concrete holomorphic function inside a rectangle that satisfies the stated growth conditions yet violates the predicted Bernstein bound, or a polynomial interpolation operator whose Lebesgue function integrates to less than (4|I|/π²) log n over some interval I of length |I|.

Figures

Figures reproduced from arXiv: 2603.21453 by Terence Tao.

**Figure 1.** Figure 1: The monic Chebyshev polynomial P(x) = 21−nTn(x) with n = 20. Note the local sinusoidal behavior in the interior of the interval [−1, 1]. Not displayed is the rapid (and non-sinusoidal) growth of P outside of this interval; see [PITH_FULL_IMAGE:figures/full_fig_p006_1.png] view at source ↗

**Figure 2.** Figure 2: The Lebesgue function λ(x) for the polynomial in [PITH_FULL_IMAGE:figures/full_fig_p008_2.png] view at source ↗

**Figure 3.** Figure 3: The logical dependencies between the main results of this paper involving trigonometric polynomials (or functions of global exponential type). The spacing here is chosen to be consistent with that in [PITH_FULL_IMAGE:figures/full_fig_p011_3.png] view at source ↗

**Figure 4.** Figure 4: The logical dependencies between the main results of this paper involving polynomials (or functions of local exponential type). The results in boxes are analogous to the corresponding results in [PITH_FULL_IMAGE:figures/full_fig_p012_4.png] view at source ↗

**Figure 5.** Figure 5: The square wave sgn(cos x), together with the Fej´er sum approximant (2.3) with M = 1, 2, 5, 10, 20. Note how the approximant, being the convolution of the square wave with a non-negative approximation to the identity (the Fej´er kernel), stays bounded by 1 in magnitude, avoiding the Gibbs phenomenon. (Image generated by Gemini.) where the Fourier expansion converges in L 2 ([0, 2π)), and m ranges over o… view at source ↗

**Figure 6.** Figure 6: A schematic depiction of the distribution of harmonic measure on ∂R+(I, y0) for a Brownian motion starting at x0 + iη. If in addition u is square-integrable on the real line, we recall the classical Littlewood–Paley L 2 identity (2.10) Z ∞ 0 Z R |∇u(x + iy)| 2 ydxdy = 1 2 Z R u(x) 2 dx where we write |∇u| 2 as shorthand for |∂xu| 2 + |∂yu| 2 . See for instance [28, §IV.1.2] for a proof. We will need some e… view at source ↗

**Figure 7.** Figure 7: The ellipse {cos(ir − θ) : 0 ≤ θ ≤ 2π} lies between the disks D(0,sinh r) and D(0, cosh r); we illustrate this here with r = 1. (Image generated by Gemini.) we see that the curve {cos((1 + ε)iy − θ) : 0 ≤ θ ≤ 2π} is an ellipse with semi-major axis cosh((1 + ε)y) and semi-minor axis sinh((1 + ε)y) centered at the origin; see [PITH_FULL_IMAGE:figures/full_fig_p022_7.png] view at source ↗

**Figure 8.** Figure 8: The arcsine distribution ρ = ρas, which is Lipschitz continuous and comparable to 1 in the bulk of [−1, 1], but develops singularities at the endpoints. This is superimposed with a normalized histogram of the Chebyshev nodes in [PITH_FULL_IMAGE:figures/full_fig_p024_8.png] view at source ↗

**Figure 9.** Figure 9: The logarithmic potential Uµ(x) for the polynomial in [PITH_FULL_IMAGE:figures/full_fig_p024_9.png] view at source ↗

**Figure 10.** Figure 10: The logarithmic potential Uµ(x + iη) for the polynomial in Figure 1 and η = 1/n, compared with the predicted value of α − πηρ(x) for x in the bulk of [−1, 1] and ρ = ρas. The non-zero value of η acts to damp the oscillations in the logarithmic potential caused by the discrete nature of the nodes; in particular, the potential is now smooth (and in fact harmonic in x + iη). In practice, these oscillations … view at source ↗

**Figure 11.** Figure 11: An illustration of Lemma 8.3 with k = 8, n = 20, and xk the Chebyshev nodes in Example 1.9. (Image generated by Gemini.) not directly use the theory of such weights here, although experts familiar with that theory will note other similarities (for instance that log |P(x)| will behave somewhat like a function of bounded mean oscillation). We first recall a simple lemma from [16, Lemma IV]. Lemma 8.3 (Lower… view at source ↗

**Figure 12.** Figure 12: An illustration of Corollary 8.4 with y = 0, n = 20, xk the Chebyshev nodes in Example 1.9, and k chosen so that xk ≤ x ≤ xk+1 for x ∈ [x1, xn]. (Image generated by Gemini.) Proof. First suppose that y = 0. From Lemma 8.3, (1.20), and the triangle inequality we have that |P(x)| 1 |x − xk||P′ (xk)| + 1 |x − xk+1||P′ (xk+1)| ≥ 1. Since |x − xk|, |x − xk+1| ≥ δ(x) and |P ′ (xk)|, |P ′ (xk+1)| ≥ pk, the c… view at source ↗

read the original abstract

Classical (or ``global'') Bernstein theory establishes sharp control on entire functions of exponential type that are bounded and real-valued on the real axis. We localize some of this theory to rectangular regions $\{ x+iy: x \in I, 0 \leq y \leq y_0 \}$, showing that Bernstein-type bounds with acceptable errors can continue to hold for functions holomorphic in such rectangles, bounded and real-valued on the lower edge of the rectangle, at most exponentially large on the upper edge, and at most double exponentially large on the vertical sides. As a consequence of these bounds, we are able to localize the Erd\H{o}s lower bound $\sup_{x \in [-1,1]} \lambda(x) \geq \frac{2}{\pi} \log n - O(1)$ on the Lebesgue constant of interpolation on $C([-1,1])$ to shorter intervals $I$ than $[-1,1]$, answering a question of Erd\H{o}s and Tur\'an. By using suitably weighted versions of the residue theorem, we also obtain the asymptotically sharp lower bound $\int_I \lambda(x)\ dx \geq \frac{4|I|}{\pi^2} \log n - o(\log n)$ for integral variants of such constants, answering a further question of Erd\H{o}s.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

Tao localizes Bernstein theory to rectangles with controlled growth and gets new lower bounds on Lebesgue constants that answer Erdős-Turán questions.

read the letter

The core advance is a localized version of Bernstein bounds that works for functions holomorphic in a rectangle over a shorter interval I, real and bounded on the bottom edge, at most exponential on the top, and at most double-exponential on the vertical sides. From this he recovers the Erdős-type sup lower bound on the Lebesgue constant restricted to I and, via weighted residues, the integral lower bound ∫_I λ(x) dx ≥ (4|I|/π²) log n - o(log n). Both are presented as direct answers to open questions rather than corollaries of the global theory. The growth conditions are chosen carefully so the residue theorem produces acceptable error terms without circularity. The argument builds straightforwardly from classical Bernstein estimates, and the o(log n) remainder looks plausible once the weights are in place. One minor soft spot is that the double-exponential allowance on the sides is fairly generous; it works for the proof but might exclude some natural examples where growth is only single-exponential. The estimates themselves appear tight enough that the main claims should survive checking. This is technical work aimed at people who care about Lebesgue constants, interpolation stability, and approximation on finite intervals. Anyone already following Erdős-Turán type problems or numerical analysis bounds will get immediate value from the localized statements. The paper is coherent on its own terms and formally grounded enough to deserve a serious referee rather than a desk rejection.

Referee Report

0 major / 2 minor

Summary. The paper localizes classical Bernstein theory for entire functions of exponential type to holomorphic functions in rectangles {x + iy : x ∈ I, 0 ≤ y ≤ y0} that are bounded and real-valued on the lower edge, at most exponentially large on the upper edge, and at most double-exponentially large on the vertical sides. Using weighted residue theorems, it derives localized versions of the Erdős lower bound sup_{x∈I} λ(x) ≥ (2/π) log n − O(1) for the Lebesgue constant of polynomial interpolation on C([-1,1]), and the integral lower bound ∫_I λ(x) dx ≥ (4|I|/π²) log n − o(log n).

Significance. If the localized Bernstein bounds and error estimates hold under the stated growth conditions, the work resolves questions of Erdős and Turán by extending sharp lower bounds on Lebesgue constants from the full interval [-1,1] to arbitrary subintervals I, with an asymptotically sharp integral form. The combination of growth-controlled localization and weighted residues provides a flexible framework that may apply to other problems in approximation theory on restricted domains.

minor comments (2)

[§1, §3] §1 and §3: The transition from the global Bernstein inequality to the localized version with explicit error terms (arising from the vertical-side growth) would benefit from a self-contained statement of the precise constant in the o(log n) remainder before the application to λ(x).
[§4] The notation for the weighted residue theorem (used to obtain the integral bound) should include a brief reminder of the weight function chosen, to make the passage from the pointwise bound to the integral form fully transparent without external references.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We are grateful to the referee for their summary of the manuscript and for recommending minor revision. The description provided accurately reflects the paper's contributions to localizing Bernstein theory and deriving lower bounds for Lebesgue constants on subintervals. As no specific major comments were raised in the report, we have no particular points to address.

Circularity Check

0 steps flagged

No significant circularity; derivation self-contained

full rationale

The paper derives localized Bernstein bounds for holomorphic functions in rectangles under explicit growth conditions (bounded real on bottom edge, exponential on top, double-exponential on sides) via standard complex analysis and the residue theorem, then obtains the localized Erdős lower bound and the integral form ∫_I λ(x) dx ≥ (4|I|/π²) log n - o(log n) as direct consequences. No step reduces the target lower bounds to fitted parameters, self-definitions, or load-bearing self-citations; the argument starts from classical global Bernstein theory and applies it locally without circular dependence on the claimed results.

Axiom & Free-Parameter Ledger

0 free parameters · 2 axioms · 0 invented entities

The work relies entirely on standard results from complex analysis and classical Bernstein theory without introducing new free parameters, ad-hoc axioms, or postulated entities.

axioms (2)

standard math Classical Bernstein theory for entire functions of exponential type bounded on the real axis
The localization is built directly on this established body of results.
standard math Residue theorem and contour integration in complex analysis
Invoked to obtain the integral form of the Lebesgue lower bound.

pith-pipeline@v0.9.0 · 5533 in / 1500 out tokens · 69694 ms · 2026-05-15T01:26:44.446745+00:00 · methodology

discussion (0)

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

IndisputableMonolith/Cost/FunctionalEquation.lean dAlembert_cosh_solution_aczel / Jcost uniqueness echoes

?

echoes
ECHOES: this paper passage has the same mathematical shape or conceptual pattern as the Recognition theorem, but is not a direct formal dependency.

Theorem 1.6 (Local Bernstein theory) … |f(x+iy)| ≤ A (1+O(e^{-πL/4y0})) cosh((1+O(1/(λ min(y0,L)))) λ y)
IndisputableMonolith/Foundation/RealityFromDistinction.lean reality_from_one_distinction unclear

?

unclear
Relation between the paper passage and the cited Recognition theorem.

Theorem 1.4(iii) (Duffin–Schaeffer) … |f(x+iy)| ≤ A cosh(λ y)

What do these tags mean?

matches: The paper's claim is directly supported by a theorem in the formal canon.
supports: The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends: The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses: The paper appears to rely on the theorem as machinery.
contradicts: The paper's claim conflicts with a theorem or certificate in the canon.
unclear: Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

33 extracted references · 33 canonical work pages

[1]

T. M. Apostol,Mathematical analysis, Addison-Wesley Publishing Co., Reading, Mass.-London-Don Mills, Ont., 1974, xvii+492 pp

work page 1974
[2]

Bernstein,Sur une propri´ et´ e des polyn´ omes, Communications de la Soci´ et´ e Math´ ematique de Charkow, (2), vol

S. Bernstein,Sur une propri´ et´ e des polyn´ omes, Communications de la Soci´ et´ e Math´ ematique de Charkow, (2), vol. 14 (1914), 1–6

work page 1914
[3]

Bernstein,Sur une propri´ et´ e des fonctions enti` eres, Comptes Rendus, vol

S. Bernstein,Sur une propri´ et´ e des fonctions enti` eres, Comptes Rendus, vol. 176 (1923), 1603–1605

work page 1923
[4]

Bernstein,Sur la limitation des valeurs d’un polynomeP n(x)de degr´ e n sur tout un segment par ses valeurs en(n+ 1)points du segment, Izv

S. Bernstein,Sur la limitation des valeurs d’un polynomeP n(x)de degr´ e n sur tout un segment par ses valeurs en(n+ 1)points du segment, Izv. Akad. Nauk. SSSR (1931), 1025-1050

work page 1931
[5]

Bloom,Erd˝ os problems,https://www.erdosproblems.com/

T. Bloom,Erd˝ os problems,https://www.erdosproblems.com/

work page
[6]

R. P. Boas,Some properties of functions of exponential type, Trans. Amer. Math. Soc.40(1936), no. 3, 287–302

work page 1936
[7]

Brutman,Lebesgue functions for polynomial interpolation – a survey, Ann

L. Brutman,Lebesgue functions for polynomial interpolation – a survey, Ann. Numer. Math.4(1997), no. 1-4, 111–127

work page 1997
[8]

Brutman, I

L. Brutman, I. Gopengauz, and D. Toledano,On the integral of the Lebesgue function induced by inter- polation at the Chebyshev nodes, Acta Math. Hungar.90(2001), no. 1-2, 11–28

work page 2001
[9]

J. G. van der Corput and H. Schaake,Ungleichungen f¨ ur Polynome und trigonometrische Polynome, Compos. Math.2(1935), 321–361

work page 1935
[10]

R. J. Duffin and A. C. Schaeffer,Some properties of functions of exponential type, Bull. Amer. Math. Soc.44(1938), no. 4, 236–240. 50

work page 1938
[11]

Erd˝ os,Problems and results on the theory of interpolation

P. Erd˝ os,Problems and results on the theory of interpolation. II, Acta Math. Acad. Sci. Hungar.12 (1961), 235-244

work page 1961
[12]

Erd˝ os,Problems and results on the convergence and divergence properties of the Lagrange interpolation polynomials and some extremal problems, Mathematica (Cluj),10(1968), 65–73

P. Erd˝ os,Problems and results on the convergence and divergence properties of the Lagrange interpolation polynomials and some extremal problems, Mathematica (Cluj),10(1968), 65–73

work page 1968
[13]

Erd˝ os,Open problems, in: Open Problems in Approximation Theory, B

P. Erd˝ os,Open problems, in: Open Problems in Approximation Theory, B. Bojanov ed. (June 1993), pp. 238–242, SCT Publishing

work page 1993
[14]

Erd˝ os and J

P. Erd˝ os and J. Szabados,On the integral of the Lebesgue function of interpolation, Acta Math. Acad. Sci. Hungar.32(1978), no. 1-2, 191–195

work page 1978
[15]

Erd˝ os, J

P. Erd˝ os, J. Szabados and P. Vertesi,On the integral of the Lebesgue function of interpolation. II, Acta Math. Acad. Sci. Hungar.68(1995), no. 1-2, 1–6

work page 1995
[16]

Erd˝ os and P

P. Erd˝ os and P. Tur´ an,On interpolation, III. Interpolatory theory of polynomials, Ann. of Math.41 (1940), 510–553

work page 1940
[17]

Erd˝ os and P

P. Erd˝ os and P. Tur´ an,An extremal problem in the theory of interpolation, Acta Math. Acad. Sci. Hungar.12(1961), 221-234

work page 1961
[18]

Faber,Ober die interpolatorische Darstellung stetiger Funktionen, Jahresb

G. Faber,Ober die interpolatorische Darstellung stetiger Funktionen, Jahresb. der Deutschen Math. Ver., 23(1914), 190–210

work page 1914
[19]

L. Fej´ er,Die Absch¨ atzung eines Polynoms in einem Intervalle, wenn Schranken f¨ ur seine Werte und ersten Ableitungswerte in einzelnen Punkten des Intervalle s gegeben sind, und ihre Anwendung auf die Konvergenz Hermitescher Interpolationsreihen, Math. Zeitschrift32(1930), 426–457

work page 1930
[20]

J. B. Garnett and D. E. Marshall,Harmonic measure, New Mathematical Monographs, vol. 2, Cambridge University Press, Cambridge, 2005

work page 2005
[21]

I. S. Gradshteyn and I. M. Ryzhik,Table of integrals, series, and products, Elsevier/Academic Press, Amsterdam, 2015, xlvi+1133 pp

work page 2015
[22]

H¨ ormander,Some inequalities for functions of exponential type, Math

L. H¨ ormander,Some inequalities for functions of exponential type, Math. Scand.3(1955), 21–27

work page 1955
[23]

A. I. Kamzolov,Approximation of functions on the sphereS n, Serdica84(1984), no. 1, 3–10

work page 1984
[24]

Levin,Distribution of zeros of entire functions, American Mathematical Society, Providence, RI, 1980

B. Levin,Distribution of zeros of entire functions, American Mathematical Society, Providence, RI, 1980

work page 1980
[25]

Muckenhoupt,Weighted norm inequalities for the Hardy maximal function, Trans

B. Muckenhoupt,Weighted norm inequalities for the Hardy maximal function, Trans. Amer. Math. Soc. 165(1972), 207–226

work page 1972
[26]

Runge, ¨Uber empirische Funktionen und die Interpolation zwischen ¨ aquidistanten Ordinaten, Zeitschrift f¨ ur Mathematik und Physik46(1901), 224–243

C. Runge, ¨Uber empirische Funktionen und die Interpolation zwischen ¨ aquidistanten Ordinaten, Zeitschrift f¨ ur Mathematik und Physik46(1901), 224–243

work page 1901
[27]

Sothanaphan,https://www.erdosproblems.com/forum/thread/1153#post-4438, Feb 24, 2026

N. Sothanaphan,https://www.erdosproblems.com/forum/thread/1153#post-4438, Feb 24, 2026

work page 2026
[28]

E. M. Stein,Singular integrals and differentiability properties of functions, Princeton Mathematical Series, No. 30, Princeton University Press, Princeton, N.J., 1970

work page 1970
[29]

Q. I. Rahman,Functions of exponential type, Trans. Amer. Math. Soc.135(1969), 295–309

work page 1969
[30]

Riesz,Eine trigonometrische Interpolationsformel und einige Ungleichungen f¨ ur Polynome, Jahresber

M. Riesz,Eine trigonometrische Interpolationsformel und einige Ungleichungen f¨ ur Polynome, Jahresber. Deutsch. Math.-Verein.23(1914), 354–368

work page 1914
[31]

Szeg¨ o,Schriften der K¨ onigsberger Gelehrten Gesellschaft, Naturwissenschaftliche Klasse, vol

G. Szeg¨ o,Schriften der K¨ onigsberger Gelehrten Gesellschaft, Naturwissenschaftliche Klasse, vol. 5 (1928), p. 69

work page 1928
[32]

Booklet produced for the conference ”Paul Erd˝ os and his mathematics”, Budapest, July 1999 (1999)

Various,Some of Paul’s favorite problems. Booklet produced for the conference ”Paul Erd˝ os and his mathematics”, Budapest, July 1999 (1999)

work page 1999
[33]

V´ ertesi,Optimal Lebesgue constants for Lagrange interpolation, SIAM J

P. V´ ertesi,Optimal Lebesgue constants for Lagrange interpolation, SIAM J. Numer. Anal.,27(1990), 1322–1331. UCLA Department of Mathematics, Los Angeles, CA 90095-1555. Email address:tao@math.ucla.edu 51

work page 1990