Log-concavity transform of the Boros–Moll rows

The log-concavity transform of a Boros–Moll row has simple negative roots that interlace those of the Narayana polynomial, and every Boros–Moll row is infinitely log-concave.

Results

The Boros–Moll coefficients are $$ d_i(n)=2^{-2n}\sum_{k=i}^{n}2^k\binom{2n-2k}{n-k}\binom{n+k}{n}\binom{k}{i}. $$ The log-concavity transform of a sequence $a$ is $L(a)_i=a_i^2-a_{i-1}a_{i+1}$ with $a_{-1}=0$. Put $M_n(x)=\sum_{i=0}^nL(d(n))_i\,x^i$ and let $N_n(x)=\sum_{i=0}^n\frac1{i+1}\binom ni\binom{n+1}ix^i$ be the Narayana polynomial. A polynomial is a Pólya frequency polynomial if it has nonnegative coefficients and only real roots, all nonpositive. A polynomial is infinitely log-concave if all iterates of $L$ applied to its coefficient sequence are nonnegative.

Theorem (Xie–Zhang). For $n\ge1$, $M_n$ has degree $n$ and $n$ simple negative roots. They strictly interlace the $n$ roots of $N_n$, with the roots of $M_n$ leftmost: if $a_0<\dots<a_{n-1}$ are the roots of $M_n$ and $b_0<\dots<b_{n-1}$ those of $N_n$, then $a_0<b_0<a_1<b_1<\dots<a_{n-1}<b_{n-1}$ [XZ26].

Theorem (Brändén). The log-concavity transform of a Pólya frequency polynomial is a Pólya frequency polynomial [Bra11]. Hence every Pólya frequency polynomial is infinitely log-concave.

Corollary. For $n\ge1$, the Boros–Moll row $\sum_i d_i(n)x^i$ is infinitely log-concave, as conjectured by Boros and Moll [BM99]. Its transform $M_n$ is a Pólya frequency polynomial.

Proof

For $n\ge36$ the interlacing reduces to the sign of $M_n$ at the zeros of $N_n$, which we bound using Gaussian quadrature at the zeros of Jacobi polynomials; the cases $1\le n\le35$ are checked by exact computation. Brändén's theorem follows from moment functionals that do not vanish on polynomials with roots in a half-plane, together with Laguerre-type approximations.

References

See also infinite log-concavity on symmetricfunctions.com.

  1. [XZ26] Matthew H. Y. Xie & Philip B. Zhang. Infinite log-concavity of the Boros--Moll sequences. arXiv e-prints, 2026.
    BibTeX
    @article{XieZhang2026InfiniteLogConcavity,
      author = {Matthew H. Y. Xie and Philip B. Zhang},
      title = {Infinite log-concavity of the {B}oros--{M}oll sequences},
      year = {2026},
      eprint = {2609.20653v1},
      url = {https://arxiv.org/abs/2609.20653v1},
      journal = {arXiv e-prints}
    }
  2. [Bra11] Petter Brändén. Iterated sequences and the geometry of zeros. Journal für die reine und angewandte Mathematik (Crelles Journal), 2011(658), 2011.
    BibTeX
    @article{Branden2011b,
      author = {Brändén, Petter},
      title = {Iterated sequences and the geometry of zeros},
      year = {2011},
      journal = {Journal für die reine und angewandte Mathematik (Crelles Journal)},
      volume = {2011},
      number = {658},
      doi = {10.1515/crelle.2011.063},
      url = {https://doi.org/10.1515/crelle.2011.063}
    }
  3. [BM99] George Boros & Victor H. Moll. An integral hidden in Gradshteyn and Ryzhik. Journal of Computational and Applied Mathematics, 106(2), 361-368, 1999.
    BibTeX
    @article{BorosMoll1999,
      author = {Boros, George and Moll, Victor H.},
      title = {An integral hidden in {G}radshteyn and {R}yzhik},
      year = {1999},
      journal = {Journal of Computational and Applied Mathematics},
      volume = {106},
      number = {2},
      pages = {361--368},
      doi = {10.1016/S0377-0427(99)00081-3},
      url = {https://doi.org/10.1016/S0377-0427(99)00081-3}
    }

Definitions

  • Boros–Moll coefficients

    RealRooted.BorosMoll.bmCoeffsource
    noncomputable def bmCoeff (n i : ℕ) : ℝ :=
      (1 / 4 : ℝ) ^ n *
        ∑ k ∈ Finset.Icc i n, (2 : ℝ) ^ k * ((2 * n - 2 * k).choose (n - k) : ℝ) *
          ((n + k).choose n : ℝ) * (k.choose i : ℝ)
  • Generating polynomial of the log-concavity transform of a Boros–Moll row

    RealRooted.BorosMoll.bmMsource
    noncomputable def bmM (n : ℕ) : ℝ[X] :=
      ∑ i ∈ Finset.range (n + 1), C (bmL n i) * X ^ i
  • Infinitely log-concave: every iterate of the transform is nonnegative

    RealRooted.Branden.IsInfinitelyLogConcavesource
    def IsInfinitelyLogConcave (p : ℝ[X]) : Prop :=
      ∀ k : ℕ, HasNonnegCoeffs ((RealRooted.Branden.logConcavityTransform^[k]) p)

Theorems

  • For n ≥ 1, M_n has degree n and n simple negative roots

    RealRooted.BorosMoll.bmM_rootssource
    theorem bmM_roots (n : ℕ) (hn : 1 ≤ n) :
        bmM n ≠ 0 ∧ (bmM n).natDegree = n ∧ (bmM n).Splits ∧ (bmM n).roots.Nodup ∧
          ∀ r ∈ (bmM n).roots, r < 0
  • The roots of M_n strictly interlace those of the Narayana polynomial N_nMain result

    RealRooted.BorosMoll.bmM_strictInterlaces_narayanaNsource
    theorem bmM_strictInterlaces_narayanaN (n : ℕ) (hn : 1 ≤ n) :
        StrictInterlacesSameDegree (bmM n) (narayanaN n) n
  • The log-concavity transform preserves Pólya frequency polynomials

    RealRooted.Branden.IsPFPolynomial.logConcavityTransformsource
    theorem IsPFPolynomial.logConcavityTransform {p : ℝ[X]}
        (hp : IsPFPolynomial p) :
        IsPFPolynomial (RealRooted.Branden.logConcavityTransform p)
  • Pólya frequency polynomials are infinitely log-concave

    RealRooted.Branden.IsPFPolynomial.isInfinitelyLogConcavesource
    theorem IsPFPolynomial.isInfinitelyLogConcave {p : ℝ[X]}
        (hp : IsPFPolynomial p) : IsInfinitelyLogConcave p
  • For n ≥ 1, the Boros–Moll row d(n) is infinitely log-concave

    RealRooted.BorosMoll.borosMollRow_isInfinitelyLogConcavesource
    theorem borosMollRow_isInfinitelyLogConcave (n : ℕ) (hn : 1 ≤ n) :
        RealRooted.Branden.IsInfinitelyLogConcave (borosMollRow n)

Lean source RealRooted/Challenges/Combinatorial/BorosMoll.lean at revision d6670742. Formalized 2026-10-08.