The Floor Theorem for the Yang–Mills Hessian at the Berger Field: A Floor Law Across the Berger Spheres
The tones of a squashed sphere Strike a bell and it rings at a set of tones fixed by its shape. Squash the bell and the tones move. The second variation of the Yang–Mills energy plays the same role for a gauge field. Its eigenvalues say which directions lead downhill and which lead up, how fast nearby fields grow or decay, and what a lattice computation of the field has to reproduce. On round spheres these tones are classical. Bourguignon and Lawson worked out stability over the round sphere, and Laquer computed the index of the canonical connection over every symmetric space. Nobody, as far as I can find, had computed them exactly on a squashed three-sphere, or at a field that is not a Yang–Mills field. This paper does both at once, on one natural case. The field Take the round three-sphere and stretch it by √3 along one direction: the Berger metric g = (3, 1, 1). Its Levi-Civita connection, read as an SU(2) gauge field, is what I call the Berger field. It is the instantaneous state of a gauge field carried along the Ricci flow of the metric, and it is the background on which the lattice Yang–Mills forces of Reading the Curve were tested on the Poincaré homology sphere. It is not at rest: unlike the round case, it is not a Yang–Mills field. This paper gives those lattice computations an exact target. The result The floor formula. By Peter–Weyl the Hessian splits into finite sectors, one for each spin j/2. In every sector the field f (θ1 + iθ2) ⊗ (τ1 + iτ2), with f the top weight along the long axis, is an exact eigenvector with eigenvalue (j(j+2) − 14)/3, orthogonal to every gauge direction. Since j(j+2) is the j-th eigenvalue of the Laplacian on the unit three-sphere, the floor is the scalar tone, shifted down by 14 and divided by the stretch. The left-invariant sector, exactly. Its gauge-orthogonal spectrum is −14/3 (twice), 16 − 2√73, 22/3 (twice) and 16 + 2√73. So this sector has exactly three growth directions (on the Poincaré sphere these are all of them; on the full sphere, other sectors add more), and all three come from the field acting on itself. With the commutator dropped, the same sector is 4/3 (six times) and 12 (three times), all positive. On the Poincaré sphere. Glue the opposite faces of a dodecahedron with a tenth of a turn and you get Poincaré's homology sphere, S3/2I. On it only the sectors carrying Klein's icosahedral invariants survive: j = 0, 12, 20, 24, 30, … Below 142 the gauge-orthogonal spectrum consists of the six left-invariant values above and the spin-6 values from 154/3 upward. The floor gap. The first sector above the left-invariant one begins 168/3 above the growth mode. That is one third of 168, the first eigenvalue of the scalar Laplacian on the Poincaré sphere. The same law on every stretched sphere The number 3 was never special. Stretch the sphere by any amount t, so its geometry is (t, 1, 1), and the same circular twist is still an exact vibration of its field, with energy λt(j) = [ j(j+2) − 2(t−3)·j − 2(t−2)(t+4) ] / t. Proved for every t, symbolically in both j and t. At t = 3 it is the formula above. Certified as the floor at t = 4. The floor is j²/4 − 8, and the full low spectrum on the Poincaré sphere is classified below 136. Proved to be the floor of every sector, for every stretch from 2.75 to 10. A conserved charge (the turn about the long axis, counted on the field, the frame and the gauge index together) splits every sector into blocks of at most nine states, whatever the sector's size. The twist is alone in its block. Every other block is certified to lie above it: for j < 100 by interval certificates in which the stretch itself is an interval (48,762 block-slices, no failure), and for j ≥ 100 by a closed-form Gershgorin bound proved positive by interval bisection. In particular, at t = 3 the floor formula is the bottom of every sector, not only the ones that survive on the Poincaré sphere. Why it has this shape. In the Weitzenböck identity the twist is a lowest Landau level: the transverse raising operator kills it, its momentum along the fibre is κ = (2t − 4 − j)/√t, and its internal frame is the lowest eigenvector of the curvature term, at 12 − 8t (for t ≥ 12/7, so on the whole certified range). So λt(j) = κ² + 2(j + t) + (12 − 8t), exactly. A tipping point. Below t ≈ 2.698 the twist stops being the floor of the gauge-orthogonal spectrum. That value is found numerically, not certified. For the operator H + GG* the tipping point is exact: tA = 2.7311…, the root of an explicit polynomial in √t. Where each twist goes free. The twist in sector j costs nothing at a stretch given in closed form, tj = ½(√((j+2)² + 2(j+4)²) − (j+2)), and beyond it lowers the energy. The floor gap on the Poincaré sphere is (240 − 24t)/t. It equals 168/3 only at t = 3 and closes at t = 10. Two numerical experiments, labelled as such in the paper, say what this law is and is not: It describes the landscape, not yet a flow in time. The Berger field slides toward a flat field faster than any twist can grow. Its resting relative is stiff. The squashed meron is the one non-flat resting left-invariant field found. It keeps a single unstable direction across all the shapes searched, and nothing on it goes soft. How it is proved A Weitzenböck identity. H + GG* = Σ Xi*Xi + R0, proved once in the universal enveloping algebra in exact arithmetic, so it holds in every sector at the same time. R0 is a constant 9×9 matrix with smallest eigenvalue −12, and the connection term has ω2 = 76/3. Together they bound every sector from below. Certified inertia counts. For the four surviving sectors that matter (j = 0, 12, 20, 24), the shifted operator is assembled in 200-bit interval arithmetic and factored. Every source of error is bounded rigorously, with outward rounding: the enclosure, and the residual of the computed factorization, checked after the fact (total perturbation below 7 × 10−56). Sylvester's law of inertia then counts the eigenvalues below each threshold. So the floor value is the bottom of its sector, twice, and every other value lies at least 1/3 higher, less a certified margin of 10−53. Every pivot of every factorization is published with the code. Everything beyond. From j = 30 on, the Weitzenböck bound alone puts every sector above 142 (exactly, √320 − √(76/3) > √154). Every sector, every stretch in [2.75, 10]. The charge splitting and the Gershgorin quantity are exact identities, checked symbolically in t. The block certificates use the same interval method as the inertia counts, with the stretch as an interval. Checks The round sphere. At g = (1, 1, 1) the same construction gives −2, 1, 6, 7 and 10 in spins 0 to 2. That is the published fluctuation spectrum of the meron (Kumar, Lechtenfeld and Picanço Costa), with exactly one negative value, as Laquer's theorem requires. The lattice. The two lattice Yang–Mills forces of Reading the Curve, linearised at the Berger field on triangulations of the Poincaré sphere, resolve the left-invariant sector and the bottom of spin 6. For example, the growth mode −14/3 = −4.667 comes out as −4.475, −4.651 and −4.661 on the native force at n = 1, 2, 3. At n = 1, 2 the left-invariant modes are identified by their eigenvectors, not only their values. What is claimed, and what is not This is a statement about the classical linearised operator. It is not a statement about the quantum theory, and it is not a Yang–Mills mass gap. The growth directions are negative eigenvalues, the opposite of a gap. At a field that is not at rest, separating gauge directions from physical ones is a choice. The theorem is stated on the L2 (Coulomb) slice, and the full Hessian is given alongside it. The value (j(j+2) − 14)/3 is tied to t = 3; on other Berger spheres the floor is λt(j). The floor is proved for every stretch from 2.75 to 10. Below t ≈ 2.698 it is not the floor (numerically), and outside [2.75, 10] nothing is certified. What is new in version 2 Section 6, the floor across the Berger family: the floor law for every stretch (proved symbolically); the complete Floor Theorem at (4, 1, 1); the floor proved in every sector for every t in [2.75, 10], which also settles every sector at t = 3; why the floor law has its shape: the twist is a lowest Landau level; the tipping point t* ≈ 2.698 (numerical); the stretches where each twist goes free, in closed form; a new figure. Two clearly labelled numerical observations: the flow test, and the search for resting fields. The availability line now cites the concept DOI, so it always points to the latest version. The files in this record paper.pdf: the paper, 14 pages. floor-theorem-v2-latex.zip: the LaTeX source, the style file and the figure, with the figure's script. It compiles as is with pdflatex and reproduces the paper. floor-theorem-v2-code-and-data.zip: every script the theorem runs through, at its repository path, with the receipts it produced and the lattice spectra behind the table and the figure. The proof scripts need only Python with NumPy, SymPy and mpmath. Unzipped into an empty folder they rerun, and the receipts they write match the committed ones. See its README.md. MANIFEST.txt: the SHA-256 of every file and the commit the record was built from. The code The scripts live in GIGI, my open-source geometric engine (PolyForm Noncommercial 1.0.0): github.com/nurdymuny/gigi, tag floor-theorem-v2. The lattice spectra used in the checks come from the code and data of Reading the Curve, which carry the meshes and the forces. Companion papers Reading the Curve, Not Counting the Points, where the Berger field first appears and where the lattice forces checked here are built: doi:10.5281/zenodo.24328085. Three Ways Across a Turn: Chord, Arc and Tangent: doi:10.5281/zenodo.24375338. Related work by the author The Double Cover Principle: Every Circle Carries Two Circumferences: doi:10.5281/zenodo.18895462. The Geometry of Samene
Authors
- Bee Rosa Davis (ORCID: https://orcid.org/0009-0009-8034-4308)
Institutions
- Brown University (US)
Publication Details
- Journal
- Zenodo (CERN European Organization for Nuclear Research)
- Published
- 2026-11-20
- DOI
- https://doi.org/10.5281/zenodo.24386613
- Primary Topic
- Geometric Analysis and Curvature Flows
- Type
- preprint