Hermite-Jensen limits and $d$ log-concavity of $q$-multinomials

Basic Information

• Paper ID: 2511.02628
• Title: Hermite-Jensen limits and $d$ log-concavity of $q$-multinomials
• Author: Ken Ono (University of Virginia)
• Classification: math.NT (Number Theory), math.CO (Combinatorics)
• Publication Date: November 4, 2025
• Paper Link: https://arxiv.org/abs/2511.02628

Abstract

This paper investigates log-concavity and higher-order Turán inequalities for Gaussian $q$-binomial coefficients and $q$-multinomial coefficients. Although these coefficients are not globally log-concave, the author proves that within infinite families with bounded aspect ratios, these stronger inequalities hold uniformly in a central window $|m-\mu| < C\sigma$, where $\mu$ and $\sigma$ are the mean and standard deviation of the normalized distribution, respectively. These results arise from the asymptotic behavior of normalized Jensen polynomials, which can be approximated by Hermite polynomials.

Research Background and Motivation

1. Research Problem

This paper investigates whether coefficient sequences of $q$-binomial coefficients and $q$-multinomial coefficients satisfy log-concavity and stronger $d$-th order log-concavity (also called Turán inequalities).

2. Problem Significance

• Historical Background: The unimodality of $q$-binomial coefficients is a classical problem in combinatorics. Cayley conjectured this in 1856, Sylvester proved it using invariant theory in 1878, O'Hara provided a constructive combinatorial proof in 1990, and Pak and Panova proved strict unimodality in 2013.
• Theoretical Significance: Log-concavity is a stronger property than unimodality, while Turán inequalities are higher-order generalizations of log-concavity. These properties are closely related to polynomial hyperbolicity.
• Applied Value: These inequalities have broad applications in combinatorics, probability theory, and number theory, particularly in partition theory and symmetric function theory.

3. Limitations of Existing Methods

• $q$-binomial coefficients are not globally log-concave. For example, the coefficient sequence $(1,1,2,1,1)$ of $\binom{4}{2}_q = 1 + q + 2q^2 + q^3 + q^4$ violates log-concavity at $k=1$: $1^2 < 1 \cdot 2$.
• Previous research has focused primarily on unimodality and strict unimodality, with less investigation of log-concavity and higher-order Turán inequalities.

4. Research Motivation

Since global log-concavity fails, the natural question is: In what range do these inequalities hold? This paper proves that within "balanced" parameters and a central window, these stronger inequalities hold uniformly.

Core Contributions

1. Central Window Theorem: Proves that for families of $q$-binomial coefficients with bounded aspect ratio $\lambda \in (0,1)$, $d$-th order log-concavity holds uniformly in the central window $|m-\mu_{a,b}| \leq C\sigma_{a,b}$.
2. Hermite-Jensen Limits: Establishes asymptotic relationships between normalized Jensen polynomials and Hermite polynomials: $J^{d,m}_{a,b}(X) = H_d(X) + O_{d,\lambda,C}((a+b)^{-1/2})$
3. Generalization to $q$-Multinomials: Extends results to general $q$-multinomial coefficients $\binom{n}{n_1,\ldots,n_r}_q$.
4. Real Root Results: Proves that Jensen polynomials in the central window possess real roots (all zeros are real).
5. Unified Framework: Provides a unified analytical framework based on probabilistic methods, cumulant expansions, and Hermite polynomial generating functions.

Methodology Details

Task Definition

Input: $q$-binomial coefficients $\binom{a+b}{a}_q = \frac{(q;q)_{a+b}}{(q;q)_a(q;q)_b} = \sum_{k=0}^{ab} c_{a,b}(k)q^k$

Objective: Determine where the coefficient sequence $\{c_{a,b}(k)\}$ satisfies $d$-th order log-concavity inequalities: $(L^r c_{a,b})(k) \geq 0, \quad 1 \leq r \leq d$ where the operator $L$ is defined as $(La)_k := a_k^2 - a_{k-1}a_{k+1}$.

Constraints:

• Parameters $a,b \to \infty$ with aspect ratio $a/(a+b) \to \lambda \in (0,1)$
• Index $m$ lies in the central window $W_{a,b}(C) := \{m: |m-\mu_{a,b}| \leq C\sigma_{a,b}\}$

Model Architecture

1. Probabilistic Framework

Normalize coefficients into a probability distribution: $p_{a,b}(k) := \frac{c_{a,b}(k)}{\binom{a+b}{a}}$

Define a random variable $K$ such that $\Pr[K=k] = p_{a,b}(k)$, and compute its statistical parameters:

• Mean: $\mu_{a,b} = E[K] = \frac{ab}{2}$ (by symmetry)
• Variance: $\sigma_{a,b}^2 = \text{Var}(K) = \frac{ab(a+b+1)}{12}$ (computed via logarithmic differentiation)
• Normalized Parameter: $\delta_{a,b} := \frac{1}{\sqrt{2\sigma_{a,b}}}$

2. Jensen Polynomials

For degree $d$ and index $m$, define the Jensen polynomial: $J_{d,m}(X;u) := \sum_{j=0}^d \binom{d}{j} u_{m+j} X^j$

Normalized Jensen polynomial: $J^{d,m}_{a,b}(X) := \frac{\delta_{a,b}^{-d}}{p_{a,b}(m)} J_{d,m}(\delta_{a,b}X - 1; p_{a,b})$

3. Hermite Polynomials

Use physicist-style Hermite polynomials, defined by the generating function: $\sum_{d=0}^\infty H_d(X)\frac{t^d}{d!} = e^{-t^2+Xt}$

First few Hermite polynomials:

• $H_1(X) = X$
• $H_2(X) = X^2 - 2$
• $H_3(X) = X^3 - 6X$

Technical Innovations

1. Cumulant Expansion Method (Lemma 6)

Establishes a quadratic logarithmic ratio model within the central window: $\log\frac{p_{a,b}(m+j)}{p_{a,b}(m)} = A_{a,b}(m)j - \delta_{a,b}^2 j^2 + R_{a,b}(m,j)$

Key observations:

• Third cumulant $\kappa_3 = 0$ (by symmetry)
• Fourth cumulant $\kappa_4/\sigma^4 = O((a+b)^{-1})$
• Use Petrov's characteristic function method for local limit analysis

Technical Details: For the probability generating function $G(q) = \sum_k p_{a,b}(k)q^k$, set $q = e^{t/\sigma}$, then $\Lambda(t) := \log G(e^{t/\sigma}) = \log E[e^{t(K-\mu)/\sigma}]$

Cumulant expansion: $\Lambda(t) = \frac{t^2}{2} + \frac{\kappa_3}{6\sigma^3}t^3 + \frac{\kappa_4}{24\sigma^4}t^4 + O\left(\frac{|t|^5}{\sigma^5}\right)$

2. Hermite Assembly Lemma (Lemma 7)

Proves that the quadratic model leads to Hermite limits. For coefficients satisfying $\log\frac{w_j}{w_0} = Aj - \delta^2 j^2 + R_j$ we have $\frac{\delta^{-d}}{w_0}\sum_{j=0}^d \binom{d}{j} w_j(\delta X - 1)^j = H_d(X) + O_d((a+b)^{-1/2})$

Key Technique: Use binomial identities $\sum_{j=r}^d \binom{d}{j}\binom{j}{r} y^{j-r} = \binom{d}{r}(1+y)^{d-r}$ matched with the Hermite generating function $e^{-t^2+Xt}$.

3. Real Roots to Turán Inequalities Conversion (Lemma 9)

Utilize the Craven-Csordas theorem: For a real sequence $(a_k)$, the following are equivalent:

• For all $m$, the Jensen polynomial $J_{r+1,m}(X)$ has only real roots
• The $r$-th order Turán inequality $(L^r a)_k \geq 0$ holds for all $k$

4. Hurwitz Continuity (Lemma 8)

If a polynomial sequence $P_n(X)$ has coefficients converging to a polynomial $P(X)$ with only real roots, then for sufficiently large $n$, $P_n$ also has only real roots.

Proof Strategy: Use Rouché's theorem in small neighborhoods of each root to show $P_n$ has exactly one real root in each neighborhood.

Distinction from Existing Methods

1. From Global to Local: Rather than pursuing global log-concavity, establishes uniform results within the central window.
2. Probabilistic Perspective: Transforms combinatorial problems into asymptotic analysis of probability distributions.
3. Special Function Connection: Establishes profound connections with Hermite polynomials, previously unexplored in $q$-binomial research.
4. Unified Framework: Provides unified treatment from $q$-binomials to $q$-multinomials.

Experimental Setup

Numerical Verification

The paper provides two detailed numerical examples to verify theoretical results.

Example 1: $q$-Binomial Case

Parameters: $(a,b) = (50,50)$

• $\mu_{50,50} = 1250$
• $\sigma_{50,50} \approx 145.057$
• $\delta_{50,50} \approx 0.004874$

Computational Results: At $m = 1250$ (center),

• $J^{1,1250}_{50,50}(X) = 0.999977X + 0.004787 \approx H_1(X) = X$
• $J^{2,1250}_{50,50}(X) = 0.999907X^2 + 0.028721X - 1.963914 \approx H_2(X) = X^2-2$
• $J^{3,1250}_{50,50}(X) = 0.999790X^3 + 0.071796X^2 - 5.890518X - 0.083596 \approx H_3(X) = X^3-6X$

Example 2: $q$-Multinomial Case

Parameters: $(n_1,n_2,n_3) = (90,90,90)$, $n=270$, $\lambda_i = 1/3$

• $\mu = 12150$
• $\sigma \approx 605.413$
• $\delta \approx 0.001168$

Computational Results: At $m = 12150$,

• $J^{1,12150}(X) = 0.999998X + 0.000873 \approx H_1(X)$
• $J^{2,12150}(X) = 0.999995X^2 + 0.005237X - 1.494557 \approx H_2(X)$
• $J^{3,12150}(X) = 0.999991X^3 + 0.013092X^2 - 4.483363X - 0.011740 \approx H_3(X)$

Evaluation Metrics

1. Coefficient Approximation Error: Measures differences between coefficients of $J^{d,m}_{a,b}(X)$ and $H_d(X)$
2. Convergence Rate: Verifies error is $O((a+b)^{-1/2})$ or $O(n^{-1/2})$
3. Real Root Verification: Checks whether all zeros of Jensen polynomials are real
4. Turán Inequality Verification: Directly computes signs of $(L^r c_{a,b})(k)$

Experimental Results

Main Results

Theorem 1 ($q$-Binomials): Fix $d \geq 1$ and $\lambda \in (0,1)$. Assume $a,b \to +\infty$ with $a/(a+b) \to \lambda$. Then for each $C > 0$, uniformly for integers $m$ satisfying $|m-\mu_{a,b}| \leq C\sigma_{a,b}$, we have $J^{d,m}_{a,b}(X) = H_d(X) + O_{d,\lambda,C}((a+b)^{-1/2})$

Corollary 2 (Log-Concavity): Under the conditions of Theorem 1, there exists a constant $N = N(d,\lambda,C)$ such that for all $a,b \geq N$ with $a/(a+b) \in (\lambda-1/N, \lambda+1/N)$:

1. For each $m \in W_{a,b}$, the normalized Jensen polynomial $J^{d,m}_{a,b}(X)$ has only real roots
2. For each $1 \leq r \leq d$ and $k \in W_{a,b}$, $(L^r c_{a,b})(k) \geq 0$

Theorem 3 ($q$-Multinomials): Fix $d \geq 1$ and $r \geq 2$. Let $n = \sum_{i=1}^r n_i$ with ratios $n_i/n \to \lambda_i \in (\epsilon, 1-\epsilon)$. Then for each constant $C > 0$, uniformly for integers $m$ with $|m-\mu| \leq C\sigma$, as $n \to \infty$ we have $J^{d,m}(X) = H_d(X) + O_{d,r,\lambda,C}(n^{-1/2})$

Numerical Verification Results

From the provided examples, we observe:

1. High-Precision Approximation: For $d=1,2,3$, the leading coefficients of normalized Jensen polynomials differ from Hermite polynomial coefficients by less than $0.03\%$
2. Error Magnitude: For $(a,b)=(50,50)$, $(a+b)^{-1/2} = 0.1$, observed errors align with theoretical predictions
3. Central Window Effect: Approximation is best near the distribution center

Experimental Findings

1. Window Width Flexibility: The constant $C$ can be chosen arbitrarily, meaning inequalities can be guaranteed in central windows of any width
2. Importance of Balanced Parameters: The aspect ratio $\lambda$ must stay away from 0 and 1 (i.e., parameters must be "balanced") to ensure results
3. Feasibility of Higher-Order Generalizations: The method applies not only to log-concavity ($d=1$) but also to arbitrary-order Turán inequalities

Related Work

1. Unimodality Research

• Cayley (1856) and Sylvester (1878): First proposed and proved unimodality of $q$-binomial coefficients
• O'Hara (1990): Provided constructive combinatorial proof via explicit bijections using row/column moves on Ferrers diagrams
• Pak & Panova (2013): Proved strict unimodality using positivity of Kronecker coefficients in symmetric group representation theory

2. Log-Concavity Research

• Classical Theory: Log-concavity implies unimodality, but the converse is false
• This Paper's Contribution: First systematic study of local log-concavity for $q$-binomials

3. Jensen Polynomial Theory

• Griffin, Ono, Rolen & Zagier (2019): Applied Jensen polynomials to Riemann zeta function and other sequences, related to Riemann Hypothesis
• This Paper's Innovation: Applied Jensen polynomial methods to combinatorial objects, establishing connections with Hermite polynomials

4. Turán Inequalities

• Craven & Csordas (1989): Established equivalence between Jensen polynomial real roots and Turán inequalities
• This Paper's Generalization: Applied this theory to central windows of $q$-multinomial coefficients

Advantages of This Paper

1. Unified Framework: Provides unified treatment from $q$-binomials to $q$-multinomials
2. Quantitative Results: Gives explicit convergence rate $O((a+b)^{-1/2})$
3. Probabilistic Perspective: Introduces cumulant analysis, making the problem more transparent
4. Practicality: Central window results cover the most important regions for applications

Conclusions and Discussion

Main Conclusions

1. Central Window Principle: Although $q$-binomial and $q$-multinomial coefficients are not globally log-concave, they satisfy arbitrary-order Turán inequalities in the central window $|m-\mu| < C\sigma$.
2. Hermite Limit Theorem: Normalized Jensen polynomials converge to Hermite polynomials at rate $O((a+b)^{-1/2})$, providing a powerful asymptotic characterization.
3. Universality: Results hold uniformly for any fixed $d$ and any bounded aspect ratio $\lambda$ away from boundaries.

Limitations

1. Parameter Restrictions:
   • Requires aspect ratio $\lambda \in (0,1)$ away from 0 and 1, i.e., parameters must be "balanced"
   • Results do not apply to extremely unbalanced cases (e.g., $a \ll b$ or $a \gg b$)
2. Window Restrictions:
   • Results hold only in the central window, not for tail regions $|m-\mu| \gg \sigma$
   • Although window width can be any constant $C$, it remains $O(\sigma)$ scale
3. Asymptotic Nature:
   • Requires $a,b$ or $n$ sufficiently large
   • Explicit values of constant $N(d,\lambda,C)$ are not provided
4. Computational Complexity:
   • Computing cumulants and Jensen polynomials may be complex in practice
   • Constants may be large for very large $d$

Future Directions

1. Boundary Cases: Study behavior when $\lambda$ approaches 0 or 1
2. Tail Analysis: Explore whether similar results hold far from center
3. Effective Constants: Provide explicit bounds for $N(d,\lambda,C)$
4. Other $q$-Analogues: Generalize methods to other $q$-special functions
5. Computational Algorithms: Develop efficient algorithms for computing Jensen polynomials and verifying Turán inequalities

In-Depth Evaluation

Strengths

1. Methodological Innovation

• Interdisciplinary Synthesis: Skillfully combines combinatorics, probability theory, special function theory, and complex analysis
• Probabilistic Perspective: Transforms combinatorial problems into asymptotic analysis of probability distributions with clear reasoning
• Hermite Connection: Discovers profound connections between $q$-binomial coefficients and Hermite polynomials, a completely novel viewpoint

2. Technical Rigor

• Complete Proof Chain: From cumulant calculations to Hermite assembly, each step is supported by detailed lemmas
• Error Control: Explicitly provides convergence rate $O((a+b)^{-1/2})$
• Uniformity: Proves uniform validity throughout the central window, not merely pointwise convergence

3. Result Depth

• From Unimodality to Higher-Order Turán: Significantly advances research from unimodality to log-concavity to higher-order inequalities
• Unified Framework: Unified treatment of $q$-binomials and $q$-multinomials demonstrates method universality
• Quantitative Characterization: Provides not only qualitative results but also quantitative convergence rates

4. Presentation Clarity

• Clear Structure: Motivation, main results, technical lemmas, and proofs are well-organized
• Rich Examples: Concrete numerical examples aid understanding
• Self-Contained: All key lemmas include complete proofs

Weaknesses

1. Method Limitations

• Balanced Parameter Assumption: Cannot handle unbalanced cases that may arise in applications
• Asymptotic Nature: Requires sufficiently large parameters; lacks characterization for small parameters
• Central Window Restriction: Tail behavior remains unexplored

2. Computational Complexity

• Practical Issues: Although theoretically elegant, computing cumulants and Jensen polynomials may be complex
• Unknown Constants: Specific values of $N(d,\lambda,C)$ not provided, affecting practical application

3. Comparison with Existing Results

• Stronger Unimodality: Pak-Panova's strict unimodality is in some sense stronger than this paper's central window log-concavity (global vs. local)
• Complementary: Results complement rather than completely replace existing unimodality results

4. Generalization Potential

• Other $q$-Analogues: Unclear whether methods apply to other $q$-special functions (e.g., $q$-hypergeometric functions)
• Non-Symmetric Cases: Treatment of cases without symmetry (where $\kappa_3 \neq 0$) not discussed

Impact

1. Contribution to Field

• Theoretical Advance: Significantly advances understanding of $q$-binomial coefficient properties
• Methodology: Provides new tools for studying log-concavity of combinatorial sequences
• Bridge Role: Connects combinatorics, probability theory, and special function theory

2. Practical Value

• Partition Theory: Helps understand statistical properties of partitions
• Representation Theory: Related to Kronecker coefficients in symmetric group representation theory
• Numerical Methods: Provides theoretical guidance for numerically verifying log-concavity

3. Reproducibility

• Complete Theory: All proofs are complete and independently verifiable
• Numerical Examples: Provided examples can be reproduced with computer algebra systems
• Clear Methods: Technical approach is clear and generalizable to similar problems

Applicable Scenarios

1. Theoretical Research

• Study asymptotic properties of $q$-analogues
• Explore log-concavity of combinatorial sequences
• Analyze statistical behavior of partition functions

2. Practical Applications

• Statistical Physics: $q$-binomial coefficients have applications in statistical mechanics
• Coding Theory: Related to parameters of certain error-correcting codes
• Algorithm Analysis: Analyze average-case complexity of partition-related algorithms

3. Educational Value

• Demonstrates applications of probabilistic methods in combinatorics
• Illustrates practical utility of special function theory
• Exemplifies asymptotic analysis techniques

References

Key references cited in this paper include:

1. Cayley (1856) and Sylvester (1878): Pioneering work on $q$-binomial unimodality
2. O'Hara (1990): Constructive combinatorial proof
3. Pak & Panova (2013): Strict unimodality
4. Craven & Csordas (1989): Jensen polynomials and Turán inequalities
5. Griffin, Ono, Rolen & Zagier (2019): Jensen polynomials and Riemann Hypothesis
6. Petrov (1975): Theory of sums of independent random variables, used for local limit theorems

Summary

This is a high-quality number theory/combinatorics paper that provides new perspectives and profound results for studying log-concavity of $q$-binomial and $q$-multinomial coefficients by introducing probabilistic methods and Hermite polynomials. Although results are restricted to central windows and balanced parameters, they provide strong quantitative characterizations under these conditions. The paper demonstrates technical rigor, clear exposition, and significant contributions to the field, meriting further research and generalization.