Reduction of bivariate determinantal representations

A real stable bivariate polynomial becomes, after an explicit linear change of coordinates, a homogeneous ternary polynomial that is hyperbolic at $(0,1,0)$. A determinantal form of this ternary polynomial, as provided by the Helton–Vinnikov theorem, gives a determinantal representation of the original polynomial.

Results

A real polynomial $p(z,w)$ is real stable if it has no zero with both coordinates in the upper half-plane. Let $H$ be the homogenization of $p$ in variables $(Y,T,Z)$ and let $c=H(0,1,1)$. For $c\ne0$ put $$ q=c^{-1}H(Y,T+X,T-X),\qquad q(0,1,0)=1. $$

Theorem (reduction). Let $p$ be real stable with $c\ne0$. Then $q$ is hyperbolic at $(0,1,0)$, and the substitution $Y=1$, $T=(z+w)/2$, $X=(z-w)/2$ returns $c^{-1}p$ from $q$.

Theorem (determinantal forms). Let $c\ne0$ and let $M,N$ be real $d\times d$ matrices with $q=\det(TI+XM+YN)$. Then $$ p=c\,\det\bigl(N+z\,(I+M)/2+w\,(I-M)/2\bigr). $$

Proposition (closure). For each $d$, the evaluations of $\det(TI-XU-V)$ at $X=0,1,\dotsc,d$, recorded as arrays of coefficients in $T$, form a closed set as $U,V$ range over symmetric real $d\times d$ matrices.

References

Definitions

  • The normalized ternary polynomial c⁻¹ H(Y, T + X, T - X)

    RealRooted.BivariateDetRep.normalizedExactReductionsource
    def normalizedExactReduction (p : MvPolynomial (Fin 2) ℝ) :
        MvPolynomial (Option (Fin 2)) ℝ :=
      MvPolynomial.C (homogenizationBoundaryValue p)⁻¹ *
        exactReduction (MvPolynomial.ordinaryHomogenization p p.totalDegree)
  • Evaluation arrays of symmetric real characteristic pencils

    RealRooted.symmetricPencilEvaluationRangesource
    def symmetricPencilEvaluationRange (d : ℕ) :
        Set (Fin (d + 1) → Fin (d + 1) → ℝ) :=
      {a | ∃ U V : Matrix (Fin d) (Fin d) ℝ,
        U.IsSymm ∧ V.IsSymm ∧ a = symmetricPencilEvaluation d U V}

Theorems

  • Back substitution recovers p up to the boundary constantRealRooted original

    RealRooted.BivariateDetRep.exactBackSubstitution_normalizedExactReductionsource
    theorem exactBackSubstitution_normalizedExactReduction
        (p : MvPolynomial (Fin 2) ℝ) :
        exactBackSubstitution (normalizedExactReduction p) =
          MvPolynomial.C (homogenizationBoundaryValue p)⁻¹ * p
  • A form q = det(T I + X M + Y N) gives p = c det(N + z (I+M)/2 + w (I-M)/2)Main resultRealRooted original

    RealRooted.BivariateDetRep.eq_C_mul_realDetPencil_of_normalizedExactReduction_eqsource
    theorem eq_C_mul_realDetPencil_of_normalizedExactReduction_eq
        {p : MvPolynomial (Fin 2) ℝ} (hc : homogenizationBoundaryValue p ≠ 0)
        {d : ℕ} (M N : Matrix (Fin d) (Fin d) ℝ)
        (hq : normalizedExactReduction p =
          realDetPencil (0 : Matrix (Fin d) (Fin d) ℝ)
            (ternaryDetPencilMatrices M N)) :
        p = MvPolynomial.C (homogenizationBoundaryValue p) *
          realDetPencil N (bivariateDetPencilMatrices M)
  • The set of evaluation arrays of symmetric characteristic pencils is closedRealRooted original

    RealRooted.isClosed_symmetricPencilEvaluationRangesource
    theorem isClosed_symmetricPencilEvaluationRange (d : ℕ) :
        IsClosed (symmetricPencilEvaluationRange d)

Lean source RealRooted/Challenges/Stability/BivariateDeterminantalReduction.lean at revision d6670742. Formalized 2026-10-29.