Lorentzian polynomials

DefinitionHomogeneous polynomials of degree \(d\)#✓fully proved

For natural numbers \(n\) and \(d\), \(H^d_n\) denotes the space of homogeneous polynomials of degree \(d\) in \(\mathbb{R}[w_1, \dots, w_n]\), including the zero polynomial.

def H (n d : ℕ) : Type := homogeneousSubmodule (Fin n) ℝ d
Used by 7

AbbreviationHomogeneous polynomials with positive coefficients#✓fully proved

\(P^d_n\) is the set of homogeneous polynomials \(f\) of degree \(d\) in \(\mathbb{R}[w_1, \dots, w_n]\) such that the coefficient in \(f\) of every monomial of degree \(d\) is strictly positive.

abbrev P (n d : ℕ) : Set (H n d) :=
  { f | ∀ m : Fin n →₀ ℕ, weight 1 m = d → 0 < f.1.coeff m }
Uses 1 · Used by 1

DefinitionHessian of a quadratic form#✓fully proved

For a homogeneous polynomial \(f\) of degree \(2\) in \(\mathbb{R}[w_1, \dots, w_n]\), its Hessian is the real \(n \times n\) matrix \(\mathcal{H}_f\) of second partial derivatives of \(f\): the diagonal entry \((i, i)\) is twice the coefficient of \(w_i^2\) in \(f\), and for \(i \ne j\) the entry \((i, j)\) is the coefficient of \(w_i w_j\) in \(f\).

noncomputable def H.hessian {n : ℕ} (f : H n 2) : Matrix (Fin n) (Fin n) ℝ :=
  fun i j ↦
    if i = j then
      2 * f.1.coeff (single i 2)
    else
      f.1.coeff (single i 1 + single j 1)
Uses 1 · Used by 1

DefinitionStrictly Lorentzian polynomials#✓fully proved

For natural numbers \(n\) and \(d\), the set \(\mathring{L}^d_n\) of strictly Lorentzian polynomials of degree \(d\) in \(\mathbb{R}[w_1, \dots, w_n]\) is defined by recursion on \(d\). Let \(P^d_n\) be the set of homogeneous polynomials of degree \(d\) in which every monomial of degree \(d\) has a strictly positive coefficient, and let \(\partial_i\) be the partial derivative with respect to \(w_i\). Then \(\mathring{L}^0_n = P^0_n\) and \(\mathring{L}^1_n = P^1_n\); \(\mathring{L}^2_n\) is the set of \(f \in P^2_n\) whose Hessian matrix \((\partial_i \partial_j f)_{i,j}\) is invertible and has exactly one positive eigenvalue, counted with multiplicity; and for \(d \ge 3\), \(\mathring{L}^d_n\) is the set of \(f \in P^d_n\) such that \(\partial_i f \in \mathring{L}^{d-1}_n\) for every \(i\).

def Ŀ (n d : ℕ) : Set (H n d) := match d with
  | 0 => P n 0
  | 1 => P n 1
  | 2 => { f ∈ P n 2 | IsUnit f.hessian ∧ (f.hessian.charpoly.roots.filter (0 < ·)).card = 1 }
  | d + 3 => { f ∈ P n (d + 3) | ∀ i, ⟨f.1.pderiv i, f.2.pderiv⟩ ∈ Ŀ n (d + 2) }
Uses 3 · Used by 1

DefinitionM-convex sets#✓fully proved

A subset \(J \subseteq \mathbb{N}^n\) is M-convex if for all \(\alpha, \beta \in J\) and every index \(i\) with \(\alpha_i > \beta_i\) there is an index \(j\) with \(\alpha_j < \beta_j\) and \(\alpha - e_i + e_j \in J\), where \(e_1, \dots, e_n\) are the standard unit vectors of \(\mathbb{N}^n\).

def IsMConvex {n : ℕ} (J : Set (Fin n →₀ ℕ)) : Prop :=
  ∀ α ∈ J, ∀ β ∈ J, ∀ i, α i > β i → ∃ j, α j < β j ∧ α - single i 1 + single j 1 ∈ J
Used by 1

AbbreviationHomogeneous polynomials with M-convex support#✓fully proved

\(M^d_n\) is the set of homogeneous polynomials \(f\) of degree \(d\) in \(\mathbb{R}[w_1, \dots, w_n]\) whose coefficients are all nonnegative and whose support, the set of exponent vectors \(\alpha \in \mathbb{N}^n\) of the monomials with nonzero coefficient in \(f\), is M-convex.

abbrev M (n d : ℕ) : Set (H n d) :=
  { f | IsMConvex (f.1.support : Set (Fin n →₀ ℕ)) ∧ ∀ m, 0 ≤ f.1.coeff m }
Uses 2 · Used by 1

DefinitionStable polynomials#✓fully proved

A polynomial \(f \in \mathbb{R}[w_1, \dots, w_n]\) is stable if either \(f\) is the zero polynomial or \(f(z_1, \dots, z_n) \ne 0\) for all \(z_1, \dots, z_n\) in the open upper half plane \(\mathbb{H} = \{z \in \mathbb{C} : \operatorname{Im} z > 0\}\).

def IsStable {n : ℕ} (f : MvPolynomial (Fin n) ℝ) : Prop :=
  f = 0 ∨ ∀ z : Fin n → ℍ, aeval (((↑) : ℍ → ℂ) ∘ z) f ≠ 0
Used by 1

AbbreviationStable homogeneous polynomials with nonnegative coefficients#✓fully proved

\(S^d_n\) is the set of homogeneous polynomials \(f\) of degree \(d\) in \(\mathbb{R}[w_1, \dots, w_n]\) whose coefficients are all nonnegative and that are stable: either \(f\) is the zero polynomial or \(f\) has no zero in \(\mathbb{H}^n\), where \(\mathbb{H}\) is the open upper half plane in \(\mathbb{C}\).

abbrev S (n d : ℕ) : Set (H n d) := { f | IsStable f.1 ∧ ∀ m, 0 ≤ f.1.coeff m }
Uses 2 · Used by 1

DefinitionThe set of Lorentzian polynomials#✓fully proved

For natural numbers \(n\) and \(d\), the set \(L^d_n\) of Lorentzian polynomials is a set of homogeneous polynomials of degree \(d\) in \(\mathbb{R}[w_1, \dots, w_n]\), defined by recursion on \(d\). For \(d \le 2\), \(L^d_n = S^d_n\), the set of homogeneous stable polynomials of degree \(d\) with nonnegative coefficients. For \(d \ge 3\), \(L^d_n\) is the set of \(f \in M^d_n\) (the homogeneous polynomials of degree \(d\) with nonnegative coefficients and M-convex support) such that \(\partial_i f \in L^{d-1}_n\) for every \(i \in \{1, \dots, n\}\), where \(\partial_i\) is the partial derivative with respect to \(w_i\).

def L (n d : ℕ) : Set (H n d) := match d with
  | 0 => S n 0
  | 1 => S n 1
  | 2 => S n 2
  | d + 3 => { f ∈ M n (d + 3) | ∀ i, ⟨f.1.pderiv i, f.2.pderiv⟩ ∈ L n (d + 2) }
Uses 3 · Used by 1

TheoremLorentzian polynomials as the closure of strictly Lorentzian polynomials#●ready to prove

For all natural numbers \(n \ge 1\) and \(d\), the closure of the set \(\mathring{L}^d_n\) of strictly Lorentzian polynomials of degree \(d\) in \(n\) variables, taken in the space \(H^d_n\) of homogeneous polynomials of degree \(d\) in \(\mathbb{R}[w_1, \dots, w_n]\) with the Euclidean topology on coefficients, is the set \(L^d_n\) of Lorentzian polynomials of degree \(d\) in \(n\) variables.

The paper (Theorem 2.25) allows \(n = 0\), but the statement fails for \(n = 0\) and \(d = 2\): the \(0 \times 0\) Hessian of a polynomial cannot have a positive eigenvalue, so \(\mathring{L}^2_0\) is empty, while \(L^2_0 = S^2_0\) contains the zero polynomial, which is homogeneous of degree \(2\).

public theorem_wanted theorem_2_25 (n d : ℕ) (hn : 0 < n) : closure (Ŀ n d) = L n d
Sources
  • P. Brändén and J. Huh, Lorentzian polynomials, Annals of Math, 192 (3) 2020. Statement taken from https://github.com/ImperialCollegeLondon/AnnalsChallenge. Problem annals_lorentzian_polynomials of the lean-eval benchmark.
Uses 3
9 fully proved1 ready to prove

The dependency explorer needs JavaScript and its graph data.

Download the complete graph data.