Morris flow

The flow $e^{tD_q}$ generated by the Morris operator preserves real stability of symmetric polynomials, and the normalized eigenfunctions of $D_q$ are real stable.

Results

Following Liu and Zhang [LZ26], we consider symmetric polynomials in $x_1,\dotsc,x_q$ and the differential operators $$ A_q=\sum_{i=1}^qx_i^2\partial_i^2,\qquad B_q=2\sum_{1\le i<j\le q}\frac{x_i^2\partial_i-x_j^2\partial_j}{x_i-x_j},\qquad D_q=B_q-A_q, $$ where $\partial_i=\partial/\partial x_i$. The operator $B_q$ maps symmetric polynomials to symmetric polynomials. A real polynomial is real stable if it has no zeros when all variables lie in the open upper half-plane; in particular it is nonzero (see real stability). We write $\operatorname{Sym}_{q,\le N}$ for the real space of symmetric polynomials in $q$ variables of total degree at most $N$, and $e^{tD_q}$ for the exponential of the restriction of $D_q$ to this space.

Theorem (Liu–Zhang). Let $t\ge0$ and $N\in\mathbb N$. If $P\in\operatorname{Sym}_{q,\le N}$ is real stable, then $e^{tD_q}P$ is real stable.

For a partition $\lambda$ of $N$ with at most $q$ parts, the elementary product is $E_\lambda=\prod_je_{\lambda'_j}$, where $\lambda'$ is the conjugate partition and $e_k$ is the $k$th elementary symmetric polynomial in $x_1,\dotsc,x_q$; it is real stable. Put $$ \varepsilon_\lambda=2\sum_{(i,j)\in[\lambda]}(q+1-i-j), $$ the sum over the cells of the Ferrers diagram of $\lambda$.

Proposition (normalized flow). The polynomial $e^{-t\varepsilon_\lambda}\,e^{tD_q}E_\lambda$ is real stable for every $t\ge0$ and has total degree at most $N$. A coefficientwise limit of this flow along $t=n\to\infty$ through the natural numbers that is nonzero and of total degree at most $N$ is real stable.

Theorem (Liu–Zhang). Let $\lambda$ be a partition of $N$ with at most $q$ parts such that $\lambda_i-\lambda_{i+1}\le1$ for $1\le i<q$. Let $V_\lambda$ be the span of the monomial symmetric polynomials $m_\mu$ for partitions $\mu\preceq\lambda$ in dominance order with at most $q$ parts. There is a unique $P\in V_\lambda$ with $D_qP=\varepsilon_\lambda P$ whose coefficient of $m_\lambda$ is $1$, and $P$ is real stable.

Proof

The flows $e^{-tA_q}$, a product of coordinate operators on the monomial basis, and $e^{tB_q}$, diagonal on the elementary basis, preserve real stability, and the Lie–Trotter product formula writes $e^{t(B_q-A_q)}$ as a limit of alternating products of the two. A nonzero coefficientwise limit of real stable polynomials of bounded degree is real stable. For the eigenfunction, $D_q$ is triangular in the monomial basis on the dominance downset of $\lambda$ with diagonal entries $\varepsilon_\mu$, and when $\lambda_i-\lambda_{i+1}\le1$ the eigenvalue $\varepsilon_\lambda$ is the largest, so the normalized flow from $E_\lambda$ converges to $P$.

References

  1. [LZ26] Feihu Liu & Zihao Zhang. Real-Rootedness and Gamma-Positivity for a Variation of the Morris Constant Term. arXiv e-prints, 2026.
    BibTeX
    @article{LiuZhang2026bx,
      author = {Feihu Liu and Zihao Zhang},
      title = {Real-{R}ootedness and {G}amma-{P}ositivity for a {V}ariation of the {M}orris {C}onstant {T}erm},
      year = {2026},
      eprint = {2609.15201v3},
      url = {https://arxiv.org/abs/2609.15201v3},
      journal = {arXiv e-prints}
    }

Definitions

  • The operator A_q = Σ xᵢ² ∂ᵢ²

    RealRooted.LiuZhang.morrisAsource
    def morrisA (q : ℕ) :
        MvPolynomial (Fin q) R →ₗ[R] MvPolynomial (Fin q) R :=
      ∑ i, (squaredPDeriv i).comp (pderiv i).toLinearMap
  • The operator B_q of the Morris flow

    RealRooted.LiuZhang.morrisBsource
    def morrisB (q : ℕ) :
        MvPolynomial (Fin q) R →ₗ[R] MvPolynomial (Fin q) R :=
      2 • ∑ i, ∑ j, if hij : i < j then
        squaredPDerivDividedDifference i j hij.ne else 0
  • The Morris operator D_q = B_q − A_q

    RealRooted.LiuZhang.morrisDsource
    def morrisD (q : ℕ) :
        MvPolynomial (Fin q) R →ₗ[R] MvPolynomial (Fin q) R :=
      morrisB q - morrisA q
  • The eigenvalue ε_λ of D_q

    RealRooted.LiuZhang.morrisEigenvaluesource
    def morrisEigenvalue (q : ℕ) {n : ℕ} (mu : n.Partition) : ℤ :=
      2 * (Finset.range q).sum fun i =>
        (Finset.range (partitionRow mu i)).sum fun j =>
          (q : ℤ) + 1 - ((i : ℕ) : ℤ) - 1 - ((j : ℕ) : ℤ) - 1
  • The flow exp(t D_q) on symmetric polynomials of bounded degree

    RealRooted.LiuZhang.morrisDExpsource
    def morrisDExp (q N : ℕ) (t : ℝ) (x : MorrisDegreeCoords q N) :
        MvPolynomial (Fin q) ℝ :=
      morrisDegreePolynomial q N
        (NormedSpace.exp (t • morrisDContinuous q N) x)
  • The normalized flow exp(−t ε_λ) exp(t D_q) E_λ

    RealRooted.LiuZhang.normalizedMorrisDExpsource
    def normalizedMorrisDExp {q N : ℕ} (lambda : N.Partition)
        (hlambda : lambda.parts.card ≤ q) (t : ℝ) :
        MvPolynomial (Fin q) ℝ :=
      MvPolynomial.C (Real.exp (-t * morrisEigenvalue q lambda)) *
        morrisDExp q N t (morrisElementaryPoint lambda hlambda)

Theorems

  • The flow exp(t D_q) preserves real stability of symmetric polynomials for t ≥ 0

    RealRooted.LiuZhang.MvRealStable.morrisDExpsource
    theorem MvRealStable.morrisDExp {q N : ℕ} {t : ℝ} (ht : 0 ≤ t)
        {x : MorrisDegreeCoords q N}
        (hx : MvRealStable (morrisDegreePolynomial q N x)) :
        MvRealStable (morrisDExp q N t x)
  • The flow preserves real stability of a symmetric polynomial of bounded degree

    RealRooted.LiuZhang.MvRealStable.morrisDExp_of_symmetricsource
    theorem MvRealStable.morrisDExp_of_symmetric {q N : ℕ} {t : ℝ}
        (ht : 0 ≤ t) {P : MvPolynomial (Fin q) ℝ} (hP : MvRealStable P)
        (hsym : P.IsSymmetric) (hdeg : P.totalDegree ≤ N) :
        MvRealStable (RealRooted.LiuZhang.morrisDExp q N t
          (morrisDegreeEquiv q N
            ⟨P, IsSymmetric.mem_morrisDegreeSubmodule hsym hdeg⟩))
  • The elementary product E_λ is real stable

    RealRooted.LiuZhang.MvRealStable.morrisElementarysource
    theorem MvRealStable.morrisElementary {q N : ℕ}
        (lambda : N.Partition) (hlambda : lambda.parts.card ≤ q) :
        MvRealStable (RealRooted.LiuZhang.morrisElementary
          (R := ℝ) q lambda)
  • The normalized Morris flow is real stable for t ≥ 0

    RealRooted.LiuZhang.MvRealStable.normalizedMorrisDExpsource
    theorem MvRealStable.normalizedMorrisDExp {q N : ℕ}
        (lambda : N.Partition) (hlambda : lambda.parts.card ≤ q)
        {t : ℝ} (ht : 0 ≤ t) :
        MvRealStable (normalizedMorrisDExp lambda hlambda t)
  • A nonzero coefficientwise limit of the normalized flow is real stable

    RealRooted.LiuZhang.MvRealStable.of_normalizedMorrisDExp_tendstosource
    theorem MvRealStable.of_normalizedMorrisDExp_tendsto {q N : ℕ}
        (lambda : N.Partition) (hlambda : lambda.parts.card ≤ q)
        {P : MvPolynomial (Fin q) ℝ} (hPdeg : P.totalDegree ≤ N)
        (hcoeff : ∀ m : Fin q →₀ ℕ,
          Tendsto (fun n : ℕ =>
            (RealRooted.LiuZhang.normalizedMorrisDExp
              lambda hlambda n).coeff m)
            atTop (𝓝 (P.coeff m))) (hP : P ≠ 0) : MvRealStable P
  • D_q has a unique eigenfunction in the dominance downset with top coefficient 1

    RealRooted.LiuZhang.existsUnique_morrisEigenfunctionsource
    theorem existsUnique_morrisEigenfunction {N q : ℕ}
        (lambda : N.Partition) (hlambda : lambda.parts.card ≤ q)
        (hadj : ∀ i, i + 1 < q →
          lambda.part i - lambda.part (i + 1) ≤ 1) :
        ∃! P : MorrisDownsetSpace q lambda,
          morrisD q (P : MvPolynomial (Fin q) ℝ) =
              (morrisEigenvalue q lambda : ℝ) •
                (P : MvPolynomial (Fin q) ℝ) ∧
          morrisDownsetTopCoeff lambda hlambda P = 1
  • Every normalized eigenfunction of D_q in the downset is real stableMain result

    RealRooted.LiuZhang.MvRealStable.morrisEigenfunctionsource
    theorem MvRealStable.morrisEigenfunction {N q : ℕ}
        (lambda : N.Partition) (hlambda : lambda.parts.card ≤ q)
        (hadj : ∀ i, i + 1 < q →
          lambda.part i - lambda.part (i + 1) ≤ 1)
        (P : MorrisDownsetSpace q lambda)
        (hP : morrisD q (P : MvPolynomial (Fin q) ℝ) =
              (morrisEigenvalue q lambda : ℝ) •
                (P : MvPolynomial (Fin q) ℝ) ∧
          morrisDownsetTopCoeff lambda hlambda P = 1) :
        MvRealStable (P : MvPolynomial (Fin q) ℝ)

Lean source RealRooted/Challenges/Stability/MorrisFlow.lean at revision d6670742. Formalized 2026-10-16.