Bernstein's Theorem (Polynomials)¶

Bernstein's polynomial theorem sharply bounds a complex polynomial's unit-circle derivative norm by its degree times its value norm.

Version: v1 · 2026-10-03 · History
Domain-specific #: 13012
Domain group: Formal Sciences
Origin domain: Mathematics
Subdomains: Polynomial Inequalities, Approximation Theory → Mathematics
Aliases: Bernstein's inequality for algebraic polynomials

Core Idea¶

For a complex algebraic polynomial $P$ of degree $n\geq1$, write $\|P\|_{\mathbb T}=\max_{|z|=1}|P(z)|$. Bernstein's polynomial inequality states

\[ \|P'\|_{\mathbb T}\leq n\,\|P\|_{\mathbb T}. \]

The degree bounds how steeply a polynomial can change on the unit circle relative to its maximum magnitude there. The factor $n$ is sharp: $P(z)=\alpha z^n$, $\alpha\ne0$, satisfies $\|P\|_{\mathbb T}=|\alpha|$ and $\|P'\|_{\mathbb T}=n|\alpha|$. In the unrestricted class, equality for nonzero degree-$n$ polynomials is the monomial extremal case. Ankeny and Rivlin's original 1955 research invokes Bernstein's theorem and the closely related derivative bound before deriving a restricted-zero growth improvement.^[1] The constant-polynomial $n=0$ case is trivial and is not an extremal degree argument.

A zero-location refinement is conditional, not part of the unrestricted conclusion. If $P$ has no zeros in the open unit disk, the Erdős–Lax bound improves the derivative factor from $n$ to $n/2$; zeros on the boundary are permitted.^[1]^[2] An n/(1+k) variant and inverse-approximation claim require separate hypotheses and sourcing, so they are not included here.

Structural Signature¶

Carrier: a one-variable complex algebraic polynomial with degree $n$, not a probability distribution or a general analytic function.
Common normed boundary: maxima of $P$ and $P'$ are measured on the same unit circle.
Degree control: $n$ multiplies the value norm; increasing degree enlarges the universal allowance.
Sharpness witness: the monomial $z^n$ realizes the unrestricted constant.
Optional restriction: excluding zeros from the open disk changes the admissible class and permits $n/2$, without altering the baseline theorem.

Sig role-phrases: degree-bounded polynomial; common boundary supremum; derivative norm; sharp monomial; conditional zero-exclusion refinement.

What It Is Not¶

The live Bernstein Inequalities entry concerns exponential concentration of random-variable sums, not polynomial derivatives. Bernstein's Theorem (Approximation Theory) is a converse-to-Jackson smoothness result based on approximation rate; its related vocabulary is not proof of identical identity. Bernstein Polynomial names a basis/approximation construction. A trigonometric polynomial has a related bound $\|T'\|_\infty\leq n\|T\|_\infty$, but its variable is periodic angle and its degree is Fourier frequency, so it is a related theorem rather than this entry's algebraic carrier.^[3] The classical real-interval Markov inequality has a different domain and degree dependence. A random use of the surname does not license merging these results.

Scope of Application¶

The theorem applies to complex polynomial magnitude and derivative on $|z|=1$; via maximum-modulus considerations, authors often phrase it on the closed unit disk as well. Its sharpness is a universal statement over all polynomials of a given degree. The $n/2$ refinement demands a zero-location test before use. A polynomial with a zero at the origin does not qualify even if its graph looks small on most of the disk. A radius rescaling is possible, but this entry fixes the unit circle to keep norm and constant unambiguous.

Clarity¶

The diagnostic has four steps. State the polynomial and exact degree. Compute or bound $\|P\|_{\mathbb T}$. Compare the boundary maximum of $P'$ with $n\|P\|_{\mathbb T}$. Only then ask whether all zeros avoid the open disk and whether $n/2$ applies. This order prevents a hidden change of class: a stronger constant is not a contradiction of the first theorem, and its failure for an unrestricted polynomial is not a counterexample.

Manages Complexity¶

The norm inequality replaces pointwise derivative calculations around an entire circle with one degree-scaled ceiling. It is an inverse-type control: finite algebraic degree limits oscillation/variation relative to size. It does not claim that every degree-$n$ polynomial actually reaches the ceiling. The monomial proves that no smaller universal coefficient works; it does not make every polynomial extremal. Ankeny and Rivlin use the restricted-zero theorem as a lever for a stronger growth estimate outside the unit circle, demonstrating how an additional structural hypothesis changes what can be proved.^[1]

Abstract Reasoning¶

There are two quantified statements. For all complex degree-$n$ polynomials, the coefficient $n$ is sufficient and best possible. For all polynomials in the smaller zero-free-in-the-open-disk subclass, $n/2$ is sufficient; $1+z^n$ shows that $n/2$ can be attained in that class. Tightening the premise while strengthening the conclusion is legitimate, but the conclusion cannot be carried back to the original larger class.

Knowledge Transfer¶

The mathematical technique of tying derivative size to a finite complexity parameter appears in other inequalities, including the trigonometric case discussed by Queffélec and Zarouf.^[3] That thematic transfer does not collapse carriers, norms, or proofs. In particular, the live approximation-converse theorem asks what a rate of approximating a function implies about smoothness, not how large one polynomial's derivative can be on $\mathbb T$.

Examples¶

Unrestricted sharpness, directly calculated. Let $P(z)=z^3$. On $|z|=1$, $|P(z)|=1$, while $P'(z)=3z^2$ gives $|P'(z)|=3$. Thus $3=3\cdot1$, equality in the degree-three bound. Mapped back: algebraic carrier, common boundary norm, derivative, and sharpness witness all have concrete occupants. The three zeros at $0$ mean the zero-free $3/2$ refinement cannot be used. This is our calculation of the source's extremal family, not an empirical example.^[1]

Restricted-class sharpness, differently occupied roles. Let $Q(z)=1+z^3$. Its three roots satisfy $z^3=-1$, hence lie on $|z|=1$, with none in the open disk. On the unit circle $\|Q\|_{\mathbb T}=2$ and $\|Q'\|_{\mathbb T}=3$. The refined ceiling is $(3/2)\cdot2=3$, again equality, while the unrestricted ceiling would be $3\cdot2=6$. Mapped back: the same norm and derivative roles are present, but zero location places $Q$ in a smaller class and changes the meaningful best coefficient. Boundary roots are allowed; silently requiring no roots on the closed disk would wrongly exclude this case.^[1]^[2]

Failed transfer. Apply the restricted $n/2$ formula to $P(z)=z^3$: it would demand $3\leq1.5$, false. This is not a failure of Erdős–Lax; $P$ has an interior zero and fails its premise. Mapped back: the zero-exclusion role is a genuine hypothesis, not decorative background.

Structural Tensions¶

There is no intrinsic two-sided design tension in the theorem. The unrestricted and restricted-zero bounds are nested propositions, not competing practices with opposed costs. Restricting a class gives a sharper conclusion but fewer eligible polynomials; that is a logical scope relation, not by itself a normative tradeoff. For any use, the salient diagnostic is whether zero location has actually been established before choosing $n$ or $n/2$.

Structural–Framed Character¶

The inequality is highly structural and minimally evaluative: a polynomial either meets the typed hypotheses or it does not, and the bound is true independent of whether an analyst values a small derivative. Human practice matters in choosing a notation, normalizing to the unit circle, and applying the result, not in making the proposition true. Its institutional history explains the names Bernstein and Erdős–Lax but is not a hidden premise. Its vocabulary travels to trigonometric polynomials, approximation, and even probability, but the shared surname and “inequality” do not establish the same identity. Importing the algebraic theorem into a new setting requires a valid mapping of polynomial degree, derivative and norm; simply recognizing a similar-looking bound is weaker. Its character: a sharp, typed mathematical relation with a conditional refinement, not a general motto that more complexity permits more change.

Structural Core vs. Domain Accent¶

The skeletal relation is a finite degree parameter bounding derivative norm by value norm, with an extremizer showing sharpness and a smaller admissible class permitting a stronger bound. The mechanism is inseparable from complex algebraic polynomials on a specified boundary and from the location of zeros for the refinement. The surname and historical route are accents; the zero-free condition is not an accent when invoking $n/2$. This named theorem fails the prime bar because stripping away polynomial degree, complex derivative and boundary norm leaves only the vague phrase “complexity controls change,” which does not determine any valid constant or test. Live Formal Theorem is the strict genus of this proved result; related Bernstein-named entries remain different results or namesake collisions.

This entry is a kind of Formal theorem.

Strict subsumption → live Formal Theorem. The live approximation-theory theorem, Bernstein polynomial, and probabilistic Bernstein inequalities are disambiguation neighbors only. A broader inequality parent would require a rigorous genus comparison; retrieval similarity is not enough.

Relationships to Other Abstractions¶

Current abstraction Bernstein's Theorem (Polynomials) Domain-specific

Parents (1) — more general patterns this builds on

Bernstein's Theorem (Polynomials) is a kind of Formal theorem Domain-specific

Bernstein's polynomial derivative inequality is a proved formal theorem.

Hierarchy paths (2) — routes to 2 parentless roots

Bernstein's Theorem (Polynomials) → Formal theorem → Formal System → Formalization → Representation → Abstraction

Show alternative path (1)

Neighborhood in Abstraction Space¶

Bernstein's Theorem (Polynomials) sits in a moderately populated region (49^th percentile for distinctiveness): it has near-neighbors but no dense thicket of look-alikes.

Family — Field Theory & Polynomial Structures (25 abstractions)

Nearest neighbors

Computed from structural-signature embeddings · 2026-10-08

Not to Be Confused With¶

Probability Bernstein inequalities: tail bounds involving variance and bounded summands.
Bernstein's approximation converse: approximation rate entails regularity of a target function.
Bernstein polynomial: a basis or approximation operator.
Trigonometric Bernstein inequality: analogous degree factor for trigonometric polynomials, with a periodic variable.^[3]
Unqualified $n/2$: false for $P=z^n$; it requires the zero-free-open-disk premise.

A redirect from “Bernstein's inequality (mathematical analysis)” (pageid 1091519/Q6449702) resolves to this polynomial-theorem page (pageid 47699360/Q25345912), but those identifiers do not establish a second encyclopedia identity or collapse distinct Bernstein inequalities.

References¶

[1] N. C. Ankeny and T. J. Rivlin, “On a Theorem of S. Bernstein,” Pacific Journal of Mathematics 5.6 (1955), 849–852, especially p.849 Theorems A and B and the discussion of the zero-free condition. Original research; the PDF's OCR obscures some displayed formulae, but the published page image is available. https://msp.org/pjm/1955/5-6/pjm-v5-n6-p01-s.pdf registry ↩a ↩b ↩c ↩d ↩e

[2] P. D. Lax, “Proof of a Conjecture of P. Erdős on the Derivative of a Polynomial,” Bulletin of the American Mathematical Society 50.8 (1944), 509–513, DOI: 10.1090/S0002-9904-1944-08177-9. Original proof; publisher full text was inaccessible during this author check, so the precise bound is also verified through Ankeny–Rivlin's original restatement. https://doi.org/10.1090/S0002-9904-1944-08177-9 registry ↩a ↩b

[3] Hervé Queffélec and Rachid Zarouf, “On Bernstein's Inequality for Polynomials,” Analysis and Mathematical Physics 9 (2019), arXiv:1903.10801, on the trigonometric-polynomial inequality and extensions. https://arxiv.org/abs/1903.10801 registry ↩a ↩b ↩c