A coefficient normalization for positive definite forms
ORCID: 0009-0001-9320-500X · DOI and Zenodo record
Original source and verification ZIP · Open PDF · Download PDF · BibTeX
Original manuscript abstract
Every positive definite real homogeneous form admits linear coordinates in which its symmetric ordered coefficients have diagonal entries equal to one and all mixed entries strictly between zero and one. This answers Brandes's question in Problem 11 of the 2019 Oberwolfach report on analytic number theory, including its absolute-value inequality. The proof chooses a basis close to a direction minimizing the form on a Euclidean sphere. The desired coefficient inequalities then follow from a positive quadratic term in a polynomial difference.
Overview
A positive definite form is a homogeneous polynomial that is positive at every nonzero real vector. This note shows that suitable linear coordinates give it a simple coefficient pattern: the diagonal symmetric ordered entries are one, and every mixed entry is positive and smaller than one.
The coordinate vectors can be nearly parallel. Their independence is enough for the theorem; no orthogonality or quantitative conditioning bound is imposed. The coefficient convention is equally important: ordinary collected monomial coefficients carry additional multinomial factors.
The exact theorem
For every m ≥ 1, let p: ℝm → ℝ be a positive definite homogeneous form of positive degree D. Such a degree is even and D ≥ 2. Classical polarization gives the unique symmetric D-linear form A with p(x) = A(x,…,x).
There is a real basis z1,…,zm such that
p(zi) = 1 for every i,
0 < A(zj₁,…,zjᴅ) < 1 whenever the indices are not all equal.
For pure tuples the ordered entry equals one. Consequently, for every tuple the absolute value of the entry is at most the geometric mean of its corresponding diagonal values, with equality precisely for pure tuples. This is Brandes's requested absolute-value bound in the symmetric ordered-coefficient convention.
The change of variables is in GLm(ℝ). For m = 1 there are no mixed tuples.
Which coefficients are normalized?
The ordered expansion sums over all D-tuples of indices:
p(t) = ∑j₁,…,jᴅ A(ej₁,…,ejᴅ) tj₁⋯tjᴅ.
When equal monomials are collected, the coefficient of tα is the ordered entry multiplied by D!/(α1!⋯αm!). The theorem bounds the entry before this factor. It does not say that every ordinary monomial coefficient lies between zero and one.
For example, p(x,y) = x4 + 12x2y2 + y4 is positive definite. Its original symmetric ordered entry A(e1,e1,e2,e2) is 12/6 = 2. The result supplies suitable new coordinates; it asserts no bound in every starting basis.
Why clustering a basis works
Choose a unit vector v minimizing p on an auxiliary Euclidean sphere. After temporarily dividing by its positive minimum, p(v) = 1 and p(x) ≥ ‖x‖D. Tangential first and second derivatives give
A(vD−1,u) = 0,
Q(u,u) ≥ ‖u‖2/(D−1), where Q(a,b) = A(vD−2,a,b) and u ⟂ v.
Take an orthonormal basis w1,…,wm−1 of the tangent space, set wm = 0, and use xi = v + εwi. These vectors remain independent for every nonzero ε, even though they cluster near v.
For a tuple J, compare the product of its diagonal values with the Dth power of its multilinear entry. The resulting polynomial gap has quadratic term
(D/2) ε2 ∑r<s Q(wjᵣ−wjₛ, wjᵣ−wjₛ).
Every mixed tuple has a nonzero tangential difference, making this term strictly positive. The multilinear numerator separately stays positive by continuity from one. There are only finitely many tuples, so one sufficiently small positive ε makes all the required inequalities hold simultaneously.
Undo the temporary common scaling, then replace each column by zi = xi/p(xi)1/D using the original p. This positive rescaling preserves independence, gives unit diagonal values, and leaves every mixed ordered entry strictly between zero and one.
The source package includes an independent proof via positive definite bilinear slices and Cauchy–Schwarz, and a standard-library Python verifier with exact rational certificates. These finite checks supplement the analytic proof and do not formally verify the universal theorem.
Scope
The theorem is an existence result over the real general linear group. The constructed basis can be close to singular. Openness permits a rational unnormalized basis, but the final unit-diagonal rescaling need not preserve rationality.
No extension to merely nonnegative forms, unimodular integral changes, controlled conditioning, or an effective uniform choice of ε is asserted. The analytic proof has no computational assumption.
The question and precursor
Problem 11, attributed to Brandes in the 2019 Oberwolfach Analytic Number Theory Problem Session, asks for a linear change making each ordered coefficient's absolute value at most the geometric mean of its corresponding diagonal entries. The note makes the canonical symmetric polarization convention explicit.
Brandes's Forms representing forms: the definite case is a direct precursor. The available arXiv v2 author version includes the quadratic matrix-entry Cauchy–Schwarz normalization in Lemma 2.1. The publisher's final typeset edition was not inspected for this comparison; final-edition equivalence and pagination-specific theorem identity are not certified.
The bounded literature assessment does not establish first priority or present worldwide openness.
AI use and review status
Generative AI supplied the initial candidate argument and assisted internal checks, literature research, and writing. This is an unrefereed preprint. Automated reviews are not independent external human peer review or formal proof-assistant certification.
Original files and rights
The downloads above retain the original bytes and native ZIP-then-PDF order. Zenodo originals: source-and-verification.zip · paper.pdf.
The manuscript, prose and audit reports are licensed under CC BY 4.0; authored code is MIT licensed. The original package preserves this split, and cited third-party works retain their own rights.
Collaboration
Independent verification, substantive extensions, and development toward a journal submission would be welcome. Collaboration and authorship would reflect actual contributions.
Related literature
- H. L. Montgomery: Problem Session, in Analytic Number Theory. Oberwolfach Reports 16 (2019), no. 4, Report 50, pp. 3179–3183; Problem 11, attributed to Brandes, p. 3182.
- J. Brandes: Forms representing forms: the definite case. Journal of the London Mathematical Society (2) 92 (2015), no. 2, 393–410. Author version v2, including the quadratic precursor in Lemma 2.1.
Citation
Alec Kriebel. A coefficient normalization for positive definite forms. Zenodo, version 1.0, 23 September 2026. doi:10.5281/zenodo.22982894.