Logarithmic joint spectral values and flattening-rank center polytopes
DefinitionAsymSpecConn_Supplement_logValueSetalgebraic-complexityasymptotic-spectrump2o-gran-per-chapterp2o-plan-paperp2o-v1tensor-rank
For a commutative semiring with preorder and a finite family , the logarithmic joint spectral value set is
A point is a star center of when and the whole closed line segment from every to lies in . The module also defines the closed probability simplex, the flattening-rank vector attached to a leg, its convex hull over a selected set of legs, and the tensor monomial .
These notions are the common interface for Theorems 3.11, 3.12, 4.1, 4.8, and 4.12.
Formalization Note Logarithms are only used on the paper's nonzero tensors, where spectral values are positive. Finite coordinate types allow both numbered families and graph-edge families.
Definition code
import Mathlib
import Definitions.Def_AsymSpecConn_Spectrum
set_option autoImplicit false
open scoped NNReal
namespace AsymSpecConn.Supplement
/-- The coordinatewise base-two logarithm of the joint spectral value set
`log Δ_F(b₁, …, bᵣ)` used in §§3.6--4.8 of Vrana and Zuiddam,
*The asymptotic spectrum of tensors is connected*, preprint (STOC 2027 submission,
version of November 2, 2026, sha256 85a0b2ac…), pp.27--36.
**Formalization Note** The source only applies this definition to nonzero tensors, where every
spectral value is positive. `Real.logb 2` is therefore evaluated on its mathematical domain. -/
noncomputable def logValueSet {S : Type*} [CommSemiring S] (P : Preorder S)
{J : Type*} [Fintype J] (b : J → S) : Set (J → ℝ) :=
Set.range (fun f : ↥(AsymSpecConn.Spectrum S P) =>
fun j => Real.logb 2 (((f.1 (b j) : ℝ≥0) : ℝ)))
/-- A point `c` is a star center of `C` when `c ∈ C` and every closed segment from a point
of `C` to `c` lies in `C` (p.12 and Theorems 3.11--4.12). -/
def StarCenter {J : Type*} [Fintype J] (C : Set (J → ℝ)) (c : J → ℝ) : Prop :=
c ∈ C ∧ ∀ y ∈ C, ∀ t ∈ Set.Icc (0 : ℝ) 1, (1 - t) • y + t • c ∈ C
/-- `C` is star-convex with respect to every point of `D`. -/
def StarConvexWith {J : Type*} [Fintype J] (C D : Set (J → ℝ)) : Prop :=
∀ c ∈ D, StarCenter C c
/-- The probability simplex on a finite type, including boundary points with zero weights. -/
def IsProbabilityWeight {I : Type*} [Fintype I] (theta : I → ℝ) : Prop :=
(∀ i, 0 ≤ theta i) ∧ ∑ i, theta i = 1
/-- The flattening-rank center belonging to a leg `i`. -/
noncomputable def flatteningCenter {S J : Type*} [Fintype J] {k : ℕ}
(R : Fin k → S → ℕ) (b : J → S) (i : Fin k) : J → ℝ :=
fun j => Real.logb 2 (R i (b j) : ℝ)
/-- The convex hull of the flattening-rank vectors indexed by `I`. -/
noncomputable def flatteningPolytope {S J : Type*} [Fintype J] {k : ℕ}
(R : Fin k → S → ℕ) (b : J → S) (I : Set (Fin k)) : Set (J → ℝ) :=
convexHull ℝ (flatteningCenter R b '' I)
/-- `b^u = ∏_j b_j^{u_j}` (p.30), including the zero multi-index, whose value is `1`. -/
def tensorMonomial {S : Type*} [CommMonoid S] {r : ℕ} (b : Fin r → S) (u : Fin r → ℕ) : S :=
∏ j, b j ^ u j
end AsymSpecConn.Supplement
Source
Vrana and Zuiddam, The asymptotic spectrum of tensors is connected, preprint (STOC 2027 submission, version of November 2, 2026, sha256 85a0b2ac698c902a02f2df11dd0671a5d9ca373aed822ae44b7e2640fc152c7d), §3.6 p.27 and §4.1 pp.29–30