A coercive polynomial is nonnegative iff it is nonnegative where its gradient vanishes
ProvedSPQPoly.coercive_nonneg_iff_criticalp2o-gran-per-chapterp2o-plan-paperp2o-v1polynomial-optimizationreal-algebraic-geometrysums-of-squares
Let be a real polynomial in variables that is coercive: for every sequence in with . Then
This reduces the nonnegativity of the polynomial of Theorem 3.2 to its coercivity, claim (3.2.1), and its nonnegativity at critical points, claim (3.2.2).
Differences from the source:
- encoding: coercive: f(xₖ) → +∞ along every sequence with ‖xₖ‖ → +∞ → Tendsto (fun x => eval x p) (cocompact (Fin n → ℝ)) atTop; reason: in finite dimension the cocompact filter is generated by complements of bounded sets, so the two definitions coincide
- encoding: ∇p(x) = 0 → ∀ i, eval x (pderiv i p) = 0; reason: the formal partial derivatives of a polynomial are its partial derivatives
Preamble
import Mathlib open MvPolynomial Filter
Formal statement
namespace SPQPoly
/-- **A coercive polynomial is nonnegative iff it is nonnegative where its gradient vanishes.**
Ahmadi, Dibek & Hall, *Sums of Separable and Quadratic Polynomials*, arXiv:2105.04766v1, Theorem 3.2, proof, p.10:
"It follows that a coercive polynomial is nonnegative if and only if it is nonnegative at the
points where its gradient vanishes." Coercive is defined there: "a function f : Rn → R is
coercive if for every sequence {xk} with {||xk||} → +∞, we have {f(xk)} → +∞."
**Formalization Note.** `ℝⁿ` is `Fin n → ℝ`. Coercivity is
`Tendsto (fun x => eval x p) (cocompact _) atTop`; in finite dimension the cocompact filter is the
filter of complements of bounded sets, so this is the sequence definition of the page. The
gradient vanishing at `x` is `∀ i, eval x (pderiv i p) = 0`. -/
theorem coercive_nonneg_iff_critical {n : ℕ} (p : MvPolynomial (Fin n) ℝ)
(hp : Tendsto (fun x : Fin n → ℝ => eval x p) (cocompact (Fin n → ℝ)) atTop) :
(∀ x : Fin n → ℝ, 0 ≤ eval x p) ↔
(∀ x : Fin n → ℝ, (∀ i : Fin n, eval x (pderiv i p) = 0) → 0 ≤ eval x p) := by sorry
end SPQPoly
Source
Ahmadi, Dibek & Hall, Sums of Separable and Quadratic Polynomials, arXiv:2105.04766v1, Theorem 3.2, proof, p.10
Human review
Confirmed by the mission captain (proposal self-audit).
Confirmed by the moderator at approval.