Hermite–Biehler and Hurwitz stability criteria

By the Hermite–Biehler theorem, $g\ll f$ exactly when $f+ig$ has no zeros in the open upper half-plane. Consequently a real polynomial is Hurwitz stable if and only if its even and odd parts interlace, and if and only if its Hurwitz matrix is totally nonnegative.

Results

A real polynomial is Hurwitz stable if it has nonnegative coefficients and no zeros with positive real part. Let $p=\sum_k a_kx^k$ with $a_k=0$ for $k<0$. The Hurwitz matrix of $p$ is $(a_{2j-i})_{i,j\ge0}$.

Theorem (Hermite–Biehler). Let $f$ and $g$ be real polynomials with positive leading coefficients. Then $g\ll f$ if and only if $f+ig$ has no zeros with positive imaginary part.

Corollary (odd and even parts). For nonzero real polynomials $p$ and $q$, the polynomial $q(x^2)+x\,p(x^2)$ is Hurwitz stable if and only if $p$ and $q$ have nonnegative coefficients and $p\ll q$.

Theorem (Hurwitz). A nonzero real polynomial is Hurwitz stable if and only if its Hurwitz matrix is totally nonnegative.

Theorem (Routh–Hurwitz). A real polynomial $p$ with positive leading coefficient has all zeros in the open left half-plane if and only if the leading principal minors of its Hurwitz matrix of orders $1,\dots,\deg p$ are positive.

Proposition (coefficient signs). Let $p$ be a real polynomial with positive leading coefficient. If $p$ has no zeros with positive real part, its coefficients are nonnegative. If all its zeros have negative real part, then $a_0,\dots,a_{\deg p}$ are positive.

Proposition (Lace form). If $q(x^2)+x\,p(x^2)$ is Hurwitz stable, then the two-row Lace matrix of $(q,p)$ is totally nonnegative. Its even rows are shifts of the coefficient sequence of $q$ and its odd rows are shifts of that of $p$. In particular this holds whenever $p\ll q$ and $p,q$ have nonnegative coefficients.

Use

The corollary translates between Hurwitz stability of one polynomial and interlacing of two polynomials in $x^2$. Through the Lace form, interlacing pairs with nonnegative coefficients give totally nonnegative matrices.

References

  1. [Hol03] Olga Holtz. Hermite--Biehler, Routh--Hurwitz, and total positivity. Linear Algebra and its Applications, 372, 105-110, 2003.
    BibTeX
    @article{Holtz2003,
      author = {Holtz, Olga},
      title = {Hermite--{B}iehler, {R}outh--{H}urwitz, and total positivity},
      year = {2003},
      journal = {Linear Algebra and its Applications},
      volume = {372},
      pages = {105--110},
      publisher = {Elsevier BV},
      doi = {10.1016/s0024-3795(03)00501-9},
      url = {http://dx.doi.org/10.1016/S0024-3795(03)00501-9},
      issn = {0024-3795}
    }

Definitions

  • The Hermite–Biehler combination f + ig

    RealRooted.hermiteBiehlerPolynomialsource
    def hermiteBiehlerPolynomial (f g : ℝ[X]) : ℂ[X] :=
      complexify f + C Complex.I * complexify g
  • No zeros in the open upper half-plane

    RealRooted.IsUpperHalfPlaneStablesource
    def IsUpperHalfPlaneStable (p : ℂ[X]) : Prop :=
      ∀ z : ℂ, 0 < z.im → p.eval z ≠ 0
  • Hurwitz stable: nonnegative coefficients and no zeros with positive real part

    RealRooted.IsHurwitzStablesource
    def IsHurwitzStable (p : ℝ[X]) : Prop :=
      HasNonnegCoeffs p ∧ IsRightHalfPlaneStable (complexify p)
  • The odd/even polynomial q(x²) + x p(x²)

    RealRooted.oddEvenPolynomialsource
    def oddEvenPolynomial (p q : ℝ[X]) : ℝ[X] :=
      q.comp (X ^ 2) + X * p.comp (X ^ 2)
  • The classical Hurwitz matrix

    Matrix.hurwitzsource
    def hurwitz (c : ℕ → R) : Matrix ℕ ℕ R := fun i j =>
      if i ≤ 2 * j then c (2 * j - i) else 0

Theorems

  • Hermite–Biehler: g ≪ f implies f + ig has no zeros with positive imaginary part

    RealRooted.hermiteBiehlerForwardPossource
    theorem hermiteBiehlerForwardPos {f g : ℝ[X]}
        (hf : HasPosLeadingCoeff f) (hg : HasPosLeadingCoeff g) (h : StrictInterl g f) :
        IsUpperHalfPlaneStable (hermiteBiehlerPolynomial f g)
  • Hermite–Biehler, converse: f + ig without such zeros implies g ≪ f

    RealRooted.strictInterl_of_upperHalfPlaneStable_hermiteBiehlersource
    theorem strictInterl_of_upperHalfPlaneStable_hermiteBiehler
        {A D : ℝ[X]}
        (hA : HasPosLeadingCoeff A) (hD : HasPosLeadingCoeff D)
        (hHB : IsUpperHalfPlaneStable (hermiteBiehlerPolynomial A D)) :
        StrictInterl D A
  • q(x²) + x p(x²) is Hurwitz stable iff p ≪ q with nonnegative coefficientsMain result

    RealRooted.isHurwitzStable_oddEvenPolynomial_iffsource
    theorem isHurwitzStable_oddEvenPolynomial_iff {p q : ℝ[X]} (hp : p ≠ 0) (hq : q ≠ 0) :
        IsHurwitzStable (oddEvenPolynomial p q) ↔
          HasNonnegCoeffs p ∧ HasNonnegCoeffs q ∧ StrictInterl p q
  • Hurwitz criterion: stable iff the Hurwitz matrix is totally nonnegative

    RealRooted.isHurwitzStable_iff_hurwitz_isTotallyNonnegsource
    theorem isHurwitzStable_iff_hurwitz_isTotallyNonneg {p : ℝ[X]} (hp : p ≠ 0) :
        IsHurwitzStable p ↔ (Matrix.hurwitz p.coeff).IsTotallyNonneg
  • No zeros with positive real part implies nonnegative coefficients

    RealRooted.hasNonnegCoeffs_of_isRightHalfPlaneStablesource
    theorem hasNonnegCoeffs_of_isRightHalfPlaneStable {p : ℝ[X]}
        (hp : HasPosLeadingCoeff p) (h : IsRightHalfPlaneStable (complexify p)) :
        HasNonnegCoeffs p
  • All zeros with negative real part implies positive coefficients

    RealRooted.IsStrictlyHurwitzStable.coeff_possource
    theorem IsStrictlyHurwitzStable.coeff_pos {p : ℝ[X]} (h : IsStrictlyHurwitzStable p)
        (hp : HasPosLeadingCoeff p) {k : ℕ} (hk : k ≤ p.natDegree) :
        0 < p.coeff k
  • If q(x²) + x p(x²) is Hurwitz stable, the Lace matrix of (q, p) is totally nonnegative

    RealRooted.fullyInterlacingPair_of_isHurwitzStable_oddEvenPolynomialsource
    theorem fullyInterlacingPair_of_isHurwitzStable_oddEvenPolynomial {p q : ℝ[X]}
        (h : IsHurwitzStable (oddEvenPolynomial p q)) : FullyInterlacingPair q.coeff p.coeff
  • If p ≪ q have nonnegative coefficients, the Lace matrix of (q, p) is totally nonnegative

    RealRooted.StrictInterl.fullyInterlacingPairsource
    theorem StrictInterl.fullyInterlacingPair {p q : ℝ[X]} (hp : HasNonnegCoeffs p)
        (hq : HasNonnegCoeffs q) (hpq : StrictInterl p q) : FullyInterlacingPair q.coeff p.coeff
  • Routh–Hurwitz: all zeros in the open left half-plane iff leading minors are positive

    Matrix.strictlyHurwitzStable_iff_hurwitzLeadingPrincipal_det_possource
    theorem strictlyHurwitzStable_iff_hurwitzLeadingPrincipal_det_pos
        {p : ℝ[X]} (hlead : HasPosLeadingCoeff p) :
        IsStrictlyHurwitzStable p ↔
          ∀ n, n ≤ p.natDegree →
            0 < (hurwitzLeadingPrincipal p.coeff n).det

Lean source RealRooted/Challenges/Roots/HermiteBiehlerHurwitz.lean at revision d6670742. Formalized 2026-07-10.