Kostov's eigenpolynomials

Every monic complex polynomial of degree $n$ with a root at $-1$ is a Schur–Szegő composition of $n-1$ polynomials $(x+1)^{n-1}(x+a_j)$, with the $a_j$ unique up to order, and Kostov studied the map from its coefficients to the elementary symmetric functions of the $a_j$. The eigenvectors of the linear part of this map are given by polynomials with simple positive roots, and consecutive ones interlace.

Results

Fix $n\ge2$. Kostov's coefficient map sends the coefficients of a monic polynomial of degree $n$ with a root at $-1$ to the elementary symmetric functions of the parameters $a_j$ in its Schur–Szegő factorization [Kos10, §1, eq. (1)]. For $0\le k<n-1$ put $$ \lambda_k=\frac{n^k}{(n-1)(n-2)\cdots(n-k)} . $$

Theorem (eigenvalues). The linear part of the map has each $\lambda_k$ as an eigenvalue, with a one-dimensional eigenspace.

Theorem (reduced eigenpolynomials). For $0\le j\le n-3$ there is a reduced eigenpolynomial $Q_{n,j}$ giving an eigendirection for $\lambda_{j+1}$. It is monic of degree $j$, satisfies $X^jQ_{n,j}(1/X)=(-1)^jQ_{n,j}(X)$, and vanishes at $1$ when $j$ is odd. Its roots are real, simple and positive.

Theorem (Kostov). [Kos10, Thm. 3] For $1\le j\le n-4$ we have $Q_{n,j}\ll Q_{n,j+1}$.

Theorem (Kostov). [Kos10, Thm. 4] Let $1\le j\le n-5$, let $x_0<\dots<x_{j-1}$ be the roots of $Q_{n,j}$ and $z_0<\dots<z_{j+1}$ those of $Q_{n,j+2}$. For $0\le i<j/2$, $$ z_i<x_i<z_{i+1},\qquad z_{j-i}<x_{j-1-i}<z_{j+1-i}, $$ and for odd $j$ the middle roots satisfy $x_{(j-1)/2}=z_{(j+1)/2}=1$.

Proof

In suitable coordinates the eigenproblem becomes that of a Bernstein-type operator. The proofs then use a compactness argument for real zeros, Descartes' rule of signs, and a Wronskian argument.

References

  1. [Kos10] Vladimir P. Kostov. Interlacing properties and the Schur--Szegő composition. Functional Analysis and Other Mathematics, 3(1), 65-74, 2010.
    BibTeX
    @article{Kostov2010,
      author = {Kostov, Vladimir P.},
      title = {Interlacing properties and the {S}chur--{S}zeg{\H{o}} composition},
      year = {2010},
      journal = {Functional Analysis and Other Mathematics},
      volume = {3},
      number = {1},
      pages = {65--74},
      doi = {10.1007/s11853-010-0039-2},
      url = {https://doi.org/10.1007/s11853-010-0039-2}
    }
  2. [KS06] Vladimir Kostov & Boris Shapiro. On the Schur--Szegö composition of polynomials. Comptes Rendus. Mathématique, 343(2), 81-86, 2006.
    BibTeX
    @article{KostovShapiro2006,
      author = {Kostov, Vladimir and Shapiro, Boris},
      title = {On the {S}chur--{S}zeg{\"o} composition of polynomials},
      year = {2006},
      journal = {Comptes Rendus. Mathématique},
      volume = {343},
      number = {2},
      pages = {81--86},
      doi = {10.1016/j.crma.2006.06.007},
      url = {https://doi.org/10.1016/j.crma.2006.06.007}
    }

Definitions

  • The canonical reduced Bernstein eigenpolynomial

    RealRooted.Kostov.eigenpolynomialsource
    noncomputable def eigenpolynomial (n j : ℕ) : ℝ[X] :=
      if h : j + 3 ≤ n then Classical.choose (exists_reduced_eigenpolynomial n j h) else 0
  • Eigenvalue candidates for the Schur–Szegő coefficient map

    RealRooted.Kostov.eigenvaluesource
    def eigenvalue (n k : ℕ) : ℝ :=
      (n : ℝ) ^ k / ((n - 1).descFactorial k : ℕ)

Theorems

  • Every eigenvalue candidate has a nonzero eigenvector

    RealRooted.Kostov.linearPart_hasEigenvectorsource
    theorem linearPart_hasEigenvector (n : ℕ) (hn : 2 ≤ n) (k : ℕ) (hk : k < n - 1) :
        ∃ v : Fin (n - 1) → ℝ, v ≠ 0 ∧ linearPart n v = eigenvalue n k • v
  • Each candidate eigenspace is one-dimensional

    RealRooted.Kostov.linearPart_eigenspace_linesource
    theorem linearPart_eigenspace_line (n : ℕ) (hn : 2 ≤ n) (k : ℕ) (hk : k < n - 1)
        (u v : Fin (n - 1) → ℝ) (hu : u ≠ 0) (hu' : linearPart n u = eigenvalue n k • u)
        (hv' : linearPart n v = eigenvalue n k • v) : ∃ c : ℝ, v = c • u
  • A reduced eigenpolynomial gives its coefficient-map eigendirection

    RealRooted.Kostov.eigenpolynomial_hasEigenvectorsource
    theorem eigenpolynomial_hasEigenvector (n j : ℕ) (hj : j + 3 ≤ n) :
        ∃ v : Fin (n - 1) → ℝ, v ≠ 0 ∧
          monicPolynomial n v - monicPolynomial n 0 =
            X * (X + 1) ^ (n - j - 2) * eigenpolynomial n j ∧
          linearPart n v = eigenvalue n (j + 1) • v
  • The reduced eigenpolynomial is monic

    RealRooted.Kostov.eigenpolynomial_monicsource
    theorem eigenpolynomial_monic (n j : ℕ) (hj : j + 3 ≤ n) :
        (eigenpolynomial n j).Monic
  • The reduced eigenpolynomial has its indexed degree

    RealRooted.Kostov.natDegree_eigenpolynomialsource
    theorem natDegree_eigenpolynomial (n j : ℕ) (hj : j + 3 ≤ n) :
        (eigenpolynomial n j).natDegree = j
  • Kostov: eigenpolynomials have simple positive roots

    RealRooted.Kostov.eigenpolynomial_roots_simple_possource
    theorem eigenpolynomial_roots_simple_pos (n j : ℕ) (hj : j + 3 ≤ n) :
        (eigenpolynomial n j).roots.Nodup ∧ (eigenpolynomial n j).roots.card = j ∧
          ∀ r ∈ (eigenpolynomial n j).roots, 0 < r
  • The eigenpolynomials are real-rooted

    RealRooted.Kostov.eigenpolynomial_splitssource
    theorem eigenpolynomial_splits (n j : ℕ) (hj : j + 3 ≤ n) :
        (eigenpolynomial n j).Splits
  • Kostov, Theorem 3: consecutive eigenpolynomials interlaceMain result

    RealRooted.Kostov.eigenpolynomial_strictInterlsource
    theorem eigenpolynomial_strictInterl (n j : ℕ) (hj1 : 1 ≤ j) (hj : j + 4 ≤ n) :
        StrictInterl (eigenpolynomial n j) (eigenpolynomial n (j + 1))
  • Kostov, Theorem 4: the two-step root comparison

    RealRooted.Kostov.eigenpolynomial_two_stepsource
    theorem eigenpolynomial_two_step (n j : ℕ) (hj1 : 1 ≤ j) (hj : j + 5 ≤ n) :
        ∃ x z : ℕ → ℝ, StrictMonoOn x (Set.Iio j) ∧ StrictMonoOn z (Set.Iio (j + 2)) ∧
          (∀ i, i < j → (eigenpolynomial n j).IsRoot (x i)) ∧
          (∀ i, i < j + 2 → (eigenpolynomial n (j + 2)).IsRoot (z i)) ∧
          (∀ i, i < j / 2 →
            z i < x i ∧ x i < z (i + 1) ∧ z (j - i) < x (j - 1 - i) ∧
              x (j - 1 - i) < z (j + 1 - i)) ∧
          (Odd j → x ((j - 1) / 2) = 1 ∧ z ((j + 1) / 2) = 1)
  • The reduced eigenpolynomial is reciprocal up to sign

    RealRooted.Kostov.reverse_eigenpolynomialsource
    theorem reverse_eigenpolynomial (n j : ℕ) (hj : j + 3 ≤ n) :
        (eigenpolynomial n j).reverse =
          C ((-1 : ℝ) ^ j) * eigenpolynomial n j
  • Odd reduced eigenpolynomials vanish at one

    RealRooted.Kostov.eval_one_eigenpolynomial_of_oddsource
    theorem eval_one_eigenpolynomial_of_odd (n j : ℕ) (hj : j + 3 ≤ n)
        (hodd : Odd j) : (eigenpolynomial n j).eval 1 = 0

Lean source RealRooted/Challenges/Roots/KostovEigenpolynomials.lean at revision d6670742. Formalized 2026-10-12.