Bernstein’s Conjecture in Modern Mathematics

• Bernstein’s Conjecture is a term encompassing diverse rigidity and extremal principles in geometric analysis, approximation theory, and interpolation.
• It addresses key problems such as the affine rigidity of minimal surfaces in PDEs and the existence of asymptotic Bernstein constants in polynomial approximation.
• The conjecture also extends to weighted interpolation, localized derivative inequalities, and reciprocity laws, underscoring its broad mathematical significance.

Searching arXiv for recent and foundational uses of “Bernstein’s conjecture” across mathematics. “Bernstein’s conjecture” is not a single canonical statement in modern mathematics. The label appears across geometric analysis, approximation theory, interpolation, number theory, and singularity theory, usually for a problem that either originates in work of S. N. Bernstein or adopts a characteristic Bernstein-type rigidity or extremal principle. In the sources considered here, the term encompasses the higher-dimensional minimal-surface problem, asymptotic questions about Bernstein constants in polynomial approximation, density problems for weighted polynomial approximation, equioscillation problems for optimal interpolation nodes, localized derivative inequalities, and several historically or terminologically adjacent statements (Lewintan, 2019, Gal, 2015, Poltoratski, 2011, Révész et al., 30 Jun 2026).

1. Terminological scope

The common feature behind these usages is not a single formal conjecture but a recurrent pattern: a sharp structural principle is known in a classical setting, and a “Bernstein conjecture” asks whether an analogous statement persists under higher dimension, extra constraints, weights, localization, or other structural modifications. Several cited papers make this terminological plurality explicit. Komornik and Loreti formulate a “Bernstein type inequality” rather than a conjecture bearing a fixed historical name (Komornik et al., 2010). Gal poses an “Open Question 1” about a convex Bernstein constant (Gal, 2015). Lemmermeyer writes of “Bernstein’s reciprocity law,” retrospectively connecting Bernstein’s 1904 work with Artin reciprocity (Lemmermeyer, 2012).

Context	Core statement	Status in cited work
Minimal surfaces	Entire solutions of the minimal-surface equation should be affine	True for $n\le 7$, false for $n\ge 8$ (Lewintan, 2019)
Convex approximation	Existence of $$\lim_{n\to\infty} n^\lambda E_n^{(+2)}(|x|^\lambda)$$	Open (Gal, 2015)
Weighted approximation	Characterize weights $W$ for polynomial density in $C_W$	Solved by a sequence criterion (Poltoratski, 2011)
Weighted interpolation	Extremal nodes minimize the Lebesgue constant iff interval maxima equioscillate	Proved for $e^{-t}$ on $[0,\infty)$ (Révész et al., 30 Jun 2026)
Localized Bernstein inequality	Gap-separated sums of truncated cosine powers satisfy a derivative bound	Proved for $M=1,2$; open for $M\ge 3$ (Komornik et al., 2010)

This multiplicity has a substantive mathematical reason. In each setting, “Bernstein” marks either a rigidity phenomenon—entire solutions must be affine—or a sharp extremal principle—optimal approximation or interpolation is characterized by equioscillation, positivity, or exact asymptotics.

2. Minimal surfaces and the Bernstein property

In PDE and geometric analysis, “Bernstein’s conjecture” most classically refers to the higher-dimensional extension of Bernstein’s theorem for nonparametric minimal surfaces. The prototype is the minimal-surface equation in two variables,

$$(1+u_y^2)u_{xx} - 2u_xu_yu_{xy} + (1+u_x^2)u_{yy} = 0,$$

for which Bernstein’s original theorem states that every entire $n\ge 8$0 solution on $n\ge 8$1 is affine (Lewintan, 2019). In modern terminology, a PDE has the Bernstein property if every entire $n\ge 8$2 solution is affine linear. The higher-dimensional minimal-surface equation,

$n\ge 8$3

was shown to have the Bernstein property for $n\ge 8$4, but this fails for $n\ge 8$5, where entire non-linear solutions exist (Lewintan, 2019).

That dimension threshold became one of the canonical rigidity/flexibility transitions in nonlinear elliptic theory. The survey on Bernstein-type theorems emphasizes that the same question propagates to minimal-surface-type equations, $n\ge 8$6-harmonic and $n\ge 8$7-harmonic equations, maximal surface equations, and higher-codimension systems (Lewintan, 2019). The picture is not uniform. For minimal surface systems in higher codimension, even in dimension two there are many non-linear entire solutions, for example those arising from holomorphic maps $n\ge 8$8, so the codimension-one Bernstein theorem does not simply extend (Lewintan, 2019).

Recent work on splitting-type linear-growth variational problems shows a further Bernstein-type refinement. For energies of the form

$n\ge 8$9

with suitable linear-growth conditions, every entire $$\lim_{n\to\infty} n^\lambda E_n^{(+2)}(|x|^\lambda)$$0 solution of the Euler–Lagrange equation is affine unless it is “unbalanced” in a precise asymptotic sense; under a quantitative $$\lim_{n\to\infty} n^\lambda E_n^{(+2)}(|x|^\lambda)$$1-balanced hypothesis, entire solutions must be affine (Bildhauer et al., 2023). The same paper gives an explicit non-affine entire solution

$$\lim_{n\to\infty} n^\lambda E_n^{(+2)}(|x|^\lambda)$$2

showing that without such balance assumptions Bernstein-type rigidity fails in the splitting case (Bildhauer et al., 2023). This suggests that, outside the isotropic minimal-surface setting, rigidity is often controlled by asymptotic anisotropy rather than by dimension alone.

3. Bernstein constants in approximation theory

A second major use of the name arises in approximation theory through Bernstein’s asymptotic constants. For unrestricted polynomial approximation on $$\lim_{n\to\infty} n^\lambda E_n^{(+2)}(|x|^\lambda)$$3, Bernstein’s classical result states that for $$\lim_{n\to\infty} n^\lambda E_n^{(+2)}(|x|^\lambda)$$4, $$\lim_{n\to\infty} n^\lambda E_n^{(+2)}(|x|^\lambda)$$5 not an even integer, the limit

$$\lim_{n\to\infty} n^\lambda E_n^{(+2)}(|x|^\lambda)$$6

exists and is finite and positive, where $$\lim_{n\to\infty} n^\lambda E_n^{(+2)}(|x|^\lambda)$$7 is the best uniform approximation error by polynomials of degree at most $$\lim_{n\to\infty} n^\lambda E_n^{(+2)}(|x|^\lambda)$$8 (Gal, 2015). The exact values of these constants are largely unknown; the terminology “Bernstein’s constant” refers to these limiting quantities.

Gal studies the convex analogue of this problem. For

$$\lim_{n\to\infty} n^\lambda E_n^{(+2)}(|x|^\lambda)$$9

the central question is whether, for $W$0,

$W$1

exists and is finite (Gal, 2015). The function $W$2 is convex for $W$3, so the problem is a shape-preserving counterpart of Bernstein’s classical asymptotic limit. The note proves no existence theorem for this convex Bernstein constant; it formulates the question and records that the normalized errors $W$4 are bounded, using the known equivalence

$W$5

for convex $W$6 on $W$7 (Gal, 2015).

The same note also formulates an entire-function analogue on $W$8, asking whether convex polynomial approximation on expanding intervals converges to best convex approximation by entire functions of exponential type (Gal, 2015). In this branch of the subject, “Bernstein’s conjecture” therefore denotes not a solved theorem but an open asymptotic existence problem: whether the convex restriction preserves not just the rate $W$9 but a full limit constant.

4. Weighted approximation, density, and optimal interpolation

A different classical “Bernstein problem” asks for which weights $C_W$0 algebraic polynomials are dense in the weighted uniform space

$C_W$1

Poltoratski gives a necessary and sufficient condition for non-density in terms of a balanced zero-density sequence $C_W$2 and its characteristic sequence $C_W$3: polynomials are not dense in $C_W$4 if and only if

$C_W$5

for some such $C_W$6 (Poltoratski, 2011). The same framework, through Bakan’s theorem, yields a criterion for polynomial density in $C_W$7 for finite positive measures $C_W$8 (Poltoratski, 2011). Here the Bernstein problem is fully solved, and its solution is encoded by sparse support geometry and a Cauchy-transform/entire-function criterion rather than by approximation asymptotics.

Another approximation-theoretic use concerns optimal interpolation nodes. For Lagrange interpolation on an interval, Bernstein conjectured that the interpolation operator norm is minimized exactly for node systems whose intervalwise Lebesgue maxima equioscillate, and Erdős added a “sandwich property” comparing extremal and non-extremal node systems. Révész and Szokol prove the Bernstein and Erdős conjectures for exponentially weighted polynomials on the halfline $C_W$9 with weight $e^{-t}$0 (Révész et al., 30 Jun 2026). In their setting, if

$e^{-t}$1

then the unique extremal node system $e^{-t}$2 satisfies

$e^{-t}$3

and for any other node system $e^{-t}$4,

$e^{-t}$5

(Révész et al., 30 Jun 2026).

The technical novelty of that result is that it treats a weighted, unbounded-domain problem in which the derivative matrices used in earlier proofs can be singular because the last interval maximum may occur at an endpoint (Révész et al., 30 Jun 2026). The proof replaces the classical nonsingularity-based method by a decomposition into regular and degenerate regions, a properness argument for the difference map

$e^{-t}$6

and a global homeomorphism theorem, which implies uniqueness of the equioscillating minimax system (Révész et al., 30 Jun 2026).

5. Bernstein-type inequalities and convexity inequalities

Komornik and Loreti formulate a localized analogue of Bernstein’s derivative inequality for trigonometric polynomials. Starting from the classical Bernstein–Fejér–Riesz inequality

$e^{-t}$7

for trigonometric polynomials $e^{-t}$8 of order $e^{-t}$9, they introduce the truncated cosine block

$[0,\infty)$0

and conjecture that for any finite sum

$[0,\infty)$1

with gap condition

$[0,\infty)$2

one has

$[0,\infty)$3

(Komornik et al., 2010). The conjecture is proved exactly for $[0,\infty)$4, proved in full generality for $[0,\infty)$5, and left open for $[0,\infty)$6 (Komornik et al., 2010). Its reformulation through the kernel

$[0,\infty)$7

shows that the problem is equivalent to positivity of the quadratic form

$[0,\infty)$8

so the conjecture becomes a positive-semidefiniteness question for a structured Gram-type matrix (Komornik et al., 2010).

A different Bernstein-type inequality appears in the theory of positive linear operators. Raşa’s conjecture states that for Bernstein basis polynomials

$[0,\infty)$9

every convex $M=1,2$0 satisfies

$M=1,2$1

for all $M=1,2$2 (Abel, 2016). The paper gives an elementary proof by representing the left-hand side as a sum of discrete second differences $M=1,2$3 multiplied by nonnegative derivatives of a generating polynomial evaluated at $M=1,2$4; convexity gives $M=1,2$5, and analogous results are obtained for the Mirakyan–Favard–Szász and Baskakov operators (Abel, 2016). Although this is not the minimal-surface Bernstein conjecture, it exemplifies the same pattern: a sharp convexity-preserving inequality associated with Bernstein’s name or Bernstein operators.

6. Historical extensions, reciprocity, and singularities

The name also appears in arithmetic and singularity-theoretic contexts. Lemmermeyer shows that what may be called “Bernstein’s reciprocity law” is, in modern terms, a Kummer-theoretic formulation of unramified Artin reciprocity. In the cyclic case, if $M=1,2$6 has order $M=1,2$7, $M=1,2$8 contains $M=1,2$9, and $M\ge 3$0 is the Hilbert class field, then there exists $M\ge 3$1 with

$M\ge 3$2

such that for prime ideals $M\ge 3$3,

$M\ge 3$4

(Lemmermeyer, 2012). When $M\ge 3$5 does not contain the necessary roots of unity, the corrected statement passes to $M\ge 3$6 and expresses the class of $M\ge 3$7 through residue symbols in $M\ge 3$8 (Lemmermeyer, 2012). Historically, this is not a standard modern “conjecture,” but it is a retrospective Bernstein-labelled precursor of Artin reciprocity.

In singularity theory, the relevant object is the Bernstein–Sato polynomial. For irreducible plane curve singularities with two Puiseux pairs and distinct algebraic monodromy eigenvalues, Yano’s conjecture predicts the generic set of $M\ge 3$9-exponents from Puiseux data alone. Artal Bartolo, Cassou-Noguès, Luengo, and Melle-Hernández show that the generic reduced Bernstein polynomial has roots given by the explicit combinatorial set $$(1+u_y^2)u_{xx} - 2u_xu_yu_{xy} + (1+u_x^2)u_{yy} = 0,$$0, and that the common roots across the whole equisingularity class are precisely $$(1+u_y^2)u_{xx} - 2u_xu_yu_{xy} + (1+u_x^2)u_{yy} = 0,$$1 (Bartolo et al., 2016). In this setting, the conjectural content is not Bernstein’s own but concerns the roots of a Bernstein polynomial; the result clarifies which roots are topologically forced and which depend on analytic moduli (Bartolo et al., 2016).

A nearby but terminologically distinct development occurs in Schoenberg theory. The probabilistic proof of Schoenberg’s theorem does not formulate a “Bernstein conjecture,” but it proves that if $$(1+u_y^2)u_{xx} - 2u_xu_yu_{xy} + (1+u_x^2)u_{yy} = 0,$$2 is continuous negative definite in all odd dimensions, then $$(1+u_y^2)u_{xx} - 2u_xu_yu_{xy} + (1+u_x^2)u_{yy} = 0,$$3 is a Bernstein function, equivalently the Laplace exponent of a subordinator, and the corresponding Lévy processes are subordinated Brownian motions (Kühn et al., 2018). This belongs to the wider Bernstein-function tradition rather than to a named Bernstein conjecture, but it illustrates the breadth of the Bernstein vocabulary across analysis.

Taken together, these usages show that “Bernstein’s conjecture” functions less as a single theorem-title than as a family name for rigidity, equioscillation, asymptotic-constant, and positivity problems. In geometric analysis it marks the affine rigidity of entire minimal graphs (Lewintan, 2019); in approximation it governs asymptotic constants, weighted density, and extremal interpolation (Gal, 2015, Poltoratski, 2011, Révész et al., 30 Jun 2026); in harmonic analysis it appears as a localized derivative-energy inequality (Komornik et al., 2010); and in adjacent traditions it attaches to reciprocity laws, Bernstein operators, and Bernstein–Sato roots (Lemmermeyer, 2012, Abel, 2016, Bartolo et al., 2016). The phrase is therefore best understood contextually: its meaning is determined by the branch of mathematics in which the Bernstein-type principle is being invoked.