No single polynomial inequality defines a real binary Gram locus
A binary tensor is an array with two positions along each of its directions. This note considers real tensors of Frobenius norm at most one. It studies the determinants of their one-mode Gram matrices and rules out describing all attainable tuples by one additional polynomial nonnegativity condition when there are at least three factors.
Open the original PDF · Download PDF · Source and support archive · BibTeX · Zenodo record
Overlap and earlier work
Alper Ferudun’s 2 October 2026 unrefereed preprint gives an overlapping, stronger theorem on the positive interior of the hull for every n ≥ 3. It is an earlier formal preprint than this 12 October note, and its multiplicity argument overlaps with the mechanism used here. This page makes no exclusive-firstness claim.
A theorem-bearing proof was submitted in GitHub PR170, created on 30 September 2026. The formal preprint remains dated 12 October 2026. Those records do not certify mutual independence of authorship, literal simultaneity or historical public visibility; current public visibility does not independently archive what was public on the earlier date.
The marginal polygon inequalities and real constructions come from Higuchi, Sudbery and Szulc, Bravyi, and Domanov, Stegeman and De Lathauwer. Sinn had given an earlier general boundary/interior obstruction. Seigal identified the convex hull and proposed the determinant-polynomial condition addressed here. The note does not claim a new general obstruction principle.
The result and its limits
Let di be the determinant of the two-row mode-i flattening’s Gram matrix. The tensors are real and range over the closed Frobenius unit ball, including zero determinants. Write Gn for their determinant tuples and Cn = [0, 1/4]n.
For every fixed n ≥ 3, no real polynomial P in those determinant coordinates satisfies
Gn = {d ∈ Cn : P(d) ≥ 0}.
The same conclusion holds when the domain is the known convex hull Hn, obtained by also imposing di ≤ ∑j ≠ idj. The assertion covers arbitrary polynomial degrees and real coefficients.
- For n = 1, the Gram locus is {0}. For n = 2, it is the diagonal segment; the single condition −(d1 − d2)2 ≥ 0 works inside the cube.
- The higher-factor proof uses exact faces where added determinants are zero. It does not establish Ferudun’s stronger all-positive higher-factor result.
- The conclusion concerns one additional weak inequality in determinant coordinates. It does not rule out several conditions, Boolean combinations or descriptions with auxiliary variables.
Why one condition fails
For three factors, the smaller Gram eigenvalues obey triangle inequalities, and an explicit real tensor realizes every admissible triple. In determinant coordinates, a polynomial has a reflection symmetry coming from the two roots of the same quadratic equation for each eigenvalue.
On a genuine boundary, a defining polynomial must change sign, so a boundary factor must occur an odd number of times. The reflection carries that same exact multiplicity to a surface inside the feasible region, where nonnegativity requires an even multiplicity. The contradiction rules out one defining polynomial. Exact zero-determinant slices then carry the obstruction to more factors.
A positive four-factor counterexample
The manuscript also gives the rational tuple
d = (2401/10000, 81/625, 81/625, 1/10000).
Every coordinate is positive. The tuple lies strictly inside the cube and the convex hull, and it satisfies Seigal’s proposed condition Q1 ≥ Q2. Nevertheless, it violates a necessary marginal polygon inequality and cannot arise from any real tensor of Frobenius norm at most one. Thus the four-factor counterexample is not confined to zero-determinant faces.
Original manuscript abstract
For a real tensor with n binary factors, consider the determinants of its one-mode Gram matrices as the tensor ranges over the closed Frobenius unit ball. For every n ≥ 3, this locus cannot be defined by one polynomial nonnegativity condition in the determinant coordinates, even after imposing the cube bounds or the known convex-hull inequalities. This answers Seigal's single-polynomial question negatively for n ≥ 4. The proof combines established qubit marginal polygon inequalities, an explicit real three-factor realization, a multiplicity obstruction, and exact zero-determinant slices. A strictly positive rational counterexample also refutes the proposed polynomial. Our theorem was recorded in a September 30, 2026 GitHub submission. Ferudun's October 2, 2026 preprint gives an overlapping stronger theorem; we credit it and record the chronology without claiming exclusive priority or certifying mutual independence.
Read the unchanged manuscript
Open PDF · Download PDF · Original PDF on Zenodo
Download source and support archive · Original archive on Zenodo
Citation and reuse
Kriebel, Alec (2026). No single polynomial inequality defines a real binary Gram locus. Version 1.0.0. Zenodo. 10.5281/zenodo.23318851. Download BibTeX.
The manuscript and the author’s original text, code, notes and results in the support archive are shared under CC BY 4.0. Cited third-party material retains its own rights.
Review status and contributions
This work is unrefereed and has not undergone conventional human peer review. AI tools were used extensively in solving, drafting, source comparison, reproduction and adversarial verification. Alec Kriebel is responsible for the contents; formal proof certification is not claimed.
Substantial mathematical and expository contributions can support collaboration. Authorship and acknowledgments would reflect the actual intellectual and writing contributions.