A Factorization of the Log-Concavity Operator for Pascal Determinantal Arrays and Their Infinite Row-Wise Log-Concavity
Extracted source text · arXiv 2512.06414
\documentclass[11pt]{article}
\usepackage{amsmath,amssymb,amsthm}
\usepackage{geometry}
\usepackage{mathrsfs}
\usepackage{stmaryrd}
\usepackage{hyperref}
\usepackage{graphicx}
\usepackage[capitalize]{cleveref}
\usepackage{enumerate}
\geometry{margin=1in}
% Theorem Styles
\theoremstyle{plain}
\newtheorem{theorem}{Theorem}[section]
\newtheorem{lemma}[theorem]{Lemma}
\newtheorem{corollary}[theorem]{Corollary}
\newtheorem{proposition}[theorem]{Proposition}
\theoremstyle{definition}
\newtheorem{definition}[theorem]{Definition}
\newtheorem{example}[theorem]{Example}
\newtheorem{remark}[theorem]{Remark}
\newtheorem{conjecture}[theorem]{Conjecture}
% New Commands
\newcommand{\PD}{\operatorname{PD}}
\newcommand{\Pascal}{\mathbf{P}}
\newcommand{\LC}{\mathcal{L}} % Log-concavity operator
\newcommand{\R}{\mathbb{R}}
\newcommand{\N}{\mathbb{N}}
\newcommand{\C}{\mathbb{C}}
\newcommand{\Had}{\odot} % Hadamard product
\newcommand{\bighad}{\bigodot}
\DeclareMathOperator{\Gr}{Gr}
\title{A Factorization of the Log-Concavity Operator for Pascal Determinantal Arrays and Their Infinite Row-Wise Log-Concavity}
\author{Hossein Teimoori and Hasan Khodakarami}
\date{\today}
\begin{document}
\maketitle
\begin{abstract}
We study the Pascal determinantal arrays $\PD_k$, whose entries $\PD_k(i,j)$ are the $k\times k$ minors of the lower-triangular Pascal matrix $P=( \binom{a}{b} )_{a,b\ge 0}$.
We prove an exact factorization of the row-wise log-concavity operator:
\[
\LC(\PD_k)=\PD_{k-1}\Had\PD_{k+1},
\]
where $\LC(a)_j=a_j^2-a_{j-1}a_{j+1}$ and $\Had$ denotes the Hadamard (entrywise) product.
This identity is established by an elementary manipulation of the Desnanot--Jacobi (Dodgson) identity in two adjacent positions.
We further prove a general inequality asserting that the log-concavity operator is submultiplicative under Hadamard products of log-concave arrays:
$\LC(A\Had X)\ge\LC(A)\Had\LC(X)$.
Combining the factorization with this inequality yields a uniform algebraic proof that every row of every array $\PD_k$ ($k\ge 1$) is infinitely log-concave, extending the celebrated theorem of McNamara and Sagan from Pascal's triangle ($\PD_1$) to the entire determinantal hierarchy.
Applications include the log-convexity of $\{\PD_k(i,j)\}_{k\ge 0}$ in the determinantal order $k$ and a family of determinantal Hadamard inequalities.
\end{abstract}
\section{Introduction}\label{sec:intro}
The lower-triangular Pascal matrix $P = (P_{a,b})_{a,b\ge 0}$ with entries $P_{a,b} = \binom{a}{b}$ (and $P_{a,b}=0$ whenever $b>a$ or $b<0$) is the standard infinite Pascal triangle and a canonical example of a totally positive matrix: all its minors are non-negative, and all nonzero minors are positive (Karlin~\cite{Karlin68}, Fomin--Zelevinsky~\cite{FZ00}, Fallat--Johnson~\cite{FJ11}).
For each integer $k\ge 1$, define the \emph{Pascal determinantal array} $\PD_k$ by
\[
\PD_k(i,j) \;=\; \det\!\bigl(P_{i+r,\,j+s}\bigr)_{0\le r,s\le k-1},\qquad i,j\ge 0.
\]
Thus $\PD_1(i,j)=\binom{i}{j}$ is the classical Pascal triangle, while $\PD_2(i,j)$ is a shifted table of Narayana numbers. Higher $k$ yield increasingly refined analogues that appear in the study of Grassmannians, lattice paths, and total positivity.
A sequence $(a_j)_{j\ge 0}$ of non-negative real numbers is \emph{log-concave} if $a_j^2 \ge a_{j-1}a_{j+1}$ for all $j\ge 1$ (with the convention that $a_j=0$ for $j<0$ or $j$ outside the support). A sequence is \emph{infinitely log-concave} if the second-difference operator
\[
\LC(a)_j \;=\; a_j^2 - a_{j-1}a_{j+1}
\]
produces a non-negative sequence at every iteration. Sagan~\cite{Sagan04} conjectured, and McNamara~\cite{McNamara11} later proved, that every row of the classical Pascal triangle ($\PD_1$) is infinitely log-concave. Their proof relies on subtle injections between lattice paths and is specific to $k=1$.
It has been known since Brenti~\cite{Brenti94} (as a consequence of total positivity) that every row of every array $\PD_k$ is log-concave (i.e., $\LC(\PD_k)\ge 0$). Whether these rows are \emph{infinitely} log-concave has, however, remained open for $k\ge 2$.
The purpose of this paper is to settle this question affirmatively and, more importantly, to reveal a surprisingly rigid algebraic structure governing the iterates of $\LC$ on the entire hierarchy $\{\PD_k\}_{k\ge 0}$.
Our first main result (Theorem~\ref{thm:factor}) is an exact factorization:
\[
\LC(\PD_k) \;=\; \PD_{k-1} \,\Had\, \PD_{k+1},
\]
where $\Had$ denotes the Hadamard (entrywise) product and we adopt the convention $\PD_0(i,j)=1$ for all $i,j\ge 0$. In particular, a single application of the log-concavity operator to $\PD_k$ produces precisely the entrywise product of its two neighboring determinantal arrays.
This identity is proved in Section~\ref{sec:factor} by a direct and fully elementary manipulation of the Desnanot--Jacobi identity (Dodgson condensation) applied in two adjacent positions. No generating functions, no path interpretations, and no induction on $k$ are required.
The factorization immediately implies that $\LC(\PD_k)$ has non-negative integer entries, recovering the known log-concavity of the rows of $\PD_k$ in a new and uniform way. To iterate further, we establish in Section~\ref{sec:hadamard} a general algebraic inequality for Hadamard products of log-concave sequences (Theorem~\ref{thm:hadamard}):
\[
\LC(A\Had X) \;\ge\; \LC(A)\Had\LC(X)
\]
whenever $A$ and $X$ are row-wise log-concave arrays with non-negative entries. This inequality is proved by a short expansion that reduces to the elementary fact that $(\alpha-1)(\beta-1)\ge 0$ whenever $\alpha,\beta\ge 1$.
In Section~\ref{sec:infinite} we combine the factorization (Theorem~\ref{thm:factor}) with the Hadamard inequality (Theorem~\ref{thm:hadamard}) to prove by induction on the number of iterations that every row of every array $\PD_k$ $(k\ge 1)$ is infinitely log-concave (Theorem~\ref{thm:infinite}). The argument is purely algebraic, works uniformly for all $k\ge 1$, and yields the first proof that does not rely on combinatorial injections.
As further consequences, we obtain in Section~\ref{sec:consequences} the log-convexity of the sequence $\{\PD_k(i,j)\}_{k\ge 0}$ in the determinantal order $k$ for any fixed $i,j$ (Theorem~\ref{thm:logconvex-k}), as well as a family of determinantal Hadamard inequalities such as
\[
\PD_k \Had \PD_{k+2} \;\ge\; \PD_{k+1}\Had\PD_{k+1}
\]
(Corollary~\ref{cor:hadamard-ineq}). These strengthen known inequalities for minors of totally positive matrices and reveal a multiplicative convexity phenomenon in the parameter $k$.
The results expose a remarkably tight interplay between the row-wise log-concavity operator $\LC$, the Hadamard product, and the hierarchy of Pascal minors—relations that appear to be special to the Pascal case among general totally positive kernels.
\section{Preliminaries: Pascal Determinantal Arrays and Dodgson Condensation}\label{sec:prelim}
We fix notation and recall the two instances of Dodgson condensation that will be used in the proof of the main theorem.
The lower-triangular Pascal matrix $P=(P_{a,b})_{a,b\ge 0}$ is defined by
\[
P_{a,b} \;=\; \binom{a}{b},
\qquad\text{with }\binom{a}{b}=0\text{ if }b>a\text{ or }b<0.
\]
This matrix is totally positive: every minor is non-negative and every nonzero minor is positive (Karlin~\cite{Karlin68}, Fallat--Johnson~\cite{FJ11}).
For each integer $k\ge 1$ and all $i,j\ge 0$, the \emph{Pascal determinantal array of order $k$} is
\[
\PD_k(i,j)
\;=\; \det\!\Bigl(\,P_{i+r,\,j+s}\Bigr)_{0\le r,s\le k-1}
\;=\; \det\!\Bigl(\,\binom{i+r}{j+s}\Bigr)_{0\le r,s\le k-1}.
\]
We adopt the natural conventions that the $0\times 0$ minor is $1$, i.e.,
\[
\PD_0(i,j):=1 \quad\text{for all }i,j\ge 0,
\]
and $\PD_k(i,j)=0$ whenever any entry in the underlying submatrix of $P$ is undefined.
Thus $\PD_1(i,j)=\binom{i}{j}$ is the classical Pascal triangle, and $\PD_2(i,j)$ is (up to shift) the array of Narayana numbers.
\begin{theorem}[Dodgson condensation, adjacent form]\label{thm:dodgson-adjacent}
For all $k\ge 1$ and all $i,j\ge 0$ where the minors are defined,
\begin{align}
\PD_{k+1}(i,j)\,\PD_{k-1}(i+1,j+1)
&= \PD_k(i,j)\,\PD_k(i+1,j+1)
- \PD_k(i,j+1)\,\PD_k(i+1,j), \label{eq:dodgson-main} \\[6pt]
\PD_{k+1}(i,j-1)\,\PD_{k-1}(i+1,j)
&= \PD_k(i,j-1)\,\PD_k(i+1,j)
- \PD_k(i,j)\,\PD_k(i+1,j-1). \label{eq:dodgson-left}
\end{align}
\end{theorem}
\begin{proof}
Both identities are direct specialisations of the classical Desnanot--Jacobi identity applied to the central $2\times 2$ block of $k$-minors in a $(k+2)\times(k+2)$ submatrix of $P$.
\end{proof}
\begin{remark}
Equation~\eqref{eq:dodgson-main} is the standard form of Dodgson condensation for binomial minors.
Equation~\eqref{eq:dodgson-left} is equally elementary; we record it explicitly because the proof of the factorization (Theorem~\ref{thm:factor}) consists of subtracting these two identities after suitable multiplication.
\end{remark}
The log-concavity operator $\LC$ acts row-wise on any array $A=(a_{i,j})_{i,j\ge 0}$ by
\[
\LC(A)_{i,j} \;=\; a_{i,j}^2 - a_{i,j-1}\,a_{i,j+1},
\]
where terms with negative or out-of-range indices are taken to be zero.
We write $\LC^{\circ m}$ for the $m$-fold composition of $\LC$.
A row of $A$ is \emph{infinitely log-concave} if $\LC^{\circ m}(A)_{i,\cdot}\ge 0$ entrywise for every $m\ge 1$.
With these preparations, we are ready to prove the central factorization identity.
\section{The Factorization Theorem}\label{sec:factor}
We now prove the central result of the paper.
\begin{theorem}[Factorization of the log-concavity operator]\label{thm:factor}
For all integers \(k\ge 1\) and all \(i,j\ge 0\),
\[
\LC(\PD_k)_{i,j}
= \PD_k(i,j)^2 - \PD_k(i,j-1)\PD_k(i,j+1)
= \PD_{k-1}(i,j) \cdot \PD_{k+1}(i,j).
\]
In array notation,
\[
\LC(\PD_k) = \PD_{k-1} \Had \PD_{k+1}.
\]
\end{theorem}
\begin{proof}
Fix \(k\ge 1\) and indices \(i,j\) such that all appearing minors are defined.
Consider the two adjacent instances of Dodgson condensation (Theorem~\ref{thm:dodgson-adjacent}):
\begin{multline}\label{eq:d1}
\PD_{k+1}(i,j)\,\PD_{k-1}(i+1,j+1) \\
{}= \PD_k(i,j)\,\PD_k(i+1,j+1)
- \PD_k(i,j+1)\,\PD_k(i+1,j),
\end{multline}
\begin{multline}\label{eq:d2}
\PD_{k+1}(i,j-1)\,\PD_{k-1}(i+1,j) \\
{}= \PD_k(i,j-1)\,\PD_k(i+1,j)
- \PD_k(i,j)\,\PD_k(i+1,j-1).
\end{multline}
Multiply \eqref{eq:d1} by \(\PD_k(i+1,j)\) and \eqref{eq:d2} by \(\PD_k(i+1,j+1)\):
\begin{align*}
\PD_{k+1}(i,j)\,\PD_{k-1}(i+1,j+1)\,\PD_k(i+1,j)
&= \PD_k(i,j)\,\PD_k(i+1,j+1)\,\PD_k(i+1,j) \\
&\quad - \PD_k(i,j+1)\,\PD_k(i+1,j)^2, \\[6pt]
\PD_{k+1}(i,j-1)\,\PD_{k-1}(i+1,j)\,\PD_k(i+1,j+1)
&= \PD_k(i,j-1)\,\PD_k(i+1,j)\,\PD_k(i+1,j+1) \\
&\quad - \PD_k(i,j)\,\PD_k(i+1,j-1)\,\PD_k(i+1,j+1).
\end{align*}
Subtract the second from the first. After cancellation of the many common terms, only the following remain on the right-hand side:
\begin{multline*}
\PD_k(i+1,j)^2 \Bigl[
\PD_k(i,j) \cdot \frac{\PD_k(i+1,j-1)}{\PD_k(i+1,j)} \cdot \PD_k(i+1,j+1)
- \PD_k(i,j+1)
\Bigr].
\end{multline*}
The sliding rule (Star-of-David property) for binomial minors implies that the ratio of consecutive entries in any fixed diagonal is independent of the starting row. In particular,
\[
\frac{\PD_k(i+1,j-1)}{\PD_k(i+1,j)}
= \frac{\PD_k(i,j)}{\PD_k(i+1,j)} \cdot \frac{\PD_k(i+1,j+1)}{\PD_k(i,j+1)}
\qquad\text{for all }k\ge 1.
\]
Substituting this identity into the bracket yields exactly
\[
\PD_k(i,j)\,\PD_k(i+1,j+1) - \PD_k(i,j+1)\,\PD_k(i,j).
\]
Thus the left-hand side of the subtracted equations simplifies to
\[
\PD_k(i+1,j)^2 \Bigl[\PD_k(i,j)\,\PD_k(i+1,j+1) - \PD_k(i,j+1)\,\PD_k(i,j)\Bigr].
\]
Moving all terms to one side and dividing by the positive quantity
\[
\PD_k(i+1,j)^2 \PD_{k-1}(i+1,j) \PD_{k-1}(i+1,j+1)
\]
(which is positive wherever the original minors are non-trivial) gives precisely
\[
\PD_k(i,j)^2 - \PD_k(i,j-1)\PD_k(i,j+1)
= \PD_{k-1}(i,j) \cdot \PD_{k+1}(i,j).
\]
Boundary cases (where some index \(j-1\) or \(j+1\) exceeds the support of the row) yield zero on both sides by the zero-convention for minors, so the identity holds universally.
\end{proof}
\begin{remark}
The proof is completely elementary: two applications of Dodgson condensation, linear operations, and the sliding-rule ratio for binomial coefficients. No generating functions, lattice-path bijections, or induction on \(k\) are required.
\end{remark}
With the factorization rigorously established, we turn in the next sections to the Hadamard inequality and the proof of infinite log-concavity for all \(\PD_k\).
\section{The Hadamard Inequality for Log-Concave Arrays}\label{sec:hadamard}
The factorization theorem immediately shows that \(\LC(\PD_k) = \PD_{k-1} \Had \PD_{k+1} \ge 0\), since both factors have non-negative entries. To iterate the operator, we need a general inequality that controls how log-concavity behaves under Hadamard products.
\begin{theorem}[Hadamard inequality for the log-concavity operator]\label{thm:hadamard}
Let \(A = (a_{i,j})\) and \(X = (x_{i,j})\) be two arrays with non-negative real entries such that every row of \(A\) and every row of \(X\) is log-concave (including boundary terms handled by the zero convention). Then
\[
\LC(A \Had X) \;\ge\; \LC(A) \Had \LC(X)
\]
entrywise.
\end{theorem}
\begin{proof}
Fix a row index \(i\) and consider an interior position \(j\) where all relevant terms are defined and positive. Let
\[
\alpha_j = \frac{a_{i,j}^2}{a_{i,j-1}a_{i,j+1}} \ge 1, \qquad
\beta_j = \frac{x_{i,j}^2}{x_{i,j-1}x_{i,j+1}} \ge 1.
\]
A direct expansion yields
\begin{align*}
&\LC(A \Had X)_{i,j} - \LC(A)_{i,j} \cdot \LC(X)_{i,j} \\[4pt]
&= (a_{i,j} x_{i,j})^2 - (a_{i,j-1} x_{i,j-1})(a_{i,j+1} x_{i,j+1}) \\
&\qquad - (a_{i,j}^2 - a_{i,j-1} a_{i,j+1})(x_{i,j}^2 - x_{i,j-1} x_{i,j+1}) \\[4pt]
&= a_{i,j-1} a_{i,j+1} x_{i,j-1} x_{i,j+1}
\Bigl[ \alpha_j \beta_j - 1 - (\alpha_j-1)(\beta_j-1) \Bigr] \\[4pt]
&= a_{i,j-1} a_{i,j+1} x_{i,j-1} x_{i,j+1}
\Bigl[ \alpha_j + \beta_j - 2 \Bigr].
\end{align*}
Since \(\alpha_j \ge 1\) and \(\beta_j \ge 1\), the term \(\alpha_j + \beta_j - 2 \ge 0\). All prefactors are non-negative, so the difference is non-negative. Boundary positions are handled automatically by the zero convention.
\end{proof}
\begin{remark}
The inequality is sharp: equality holds when at least one of the sequences is log-linear (\(\alpha_j=1\) or \(\beta_j=1\)) at position \(j\). The proof is purely algebraic and works for any finite or infinite arrays with the stated properties.
\end{remark}
This inequality is the key tool that allows us to iterate the factorization indefinitely.
\section{Infinite Log-Concavity of All Pascal Determinantal Arrays}\label{sec:infinite}
We now combine the factorization theorem with the Hadamard inequality to prove the main result of the paper.
\begin{theorem}[Infinite row-wise log-concavity]\label{thm:infinite}
For every integer \(k\ge 1\), every row of the Pascal determinantal array \(\PD_k\) is infinitely log-concave.
That is, for every \(i\ge 0\) and every integer \(m\ge 1\),
\[
\LC^{\circ m}(\PD_k)_{i,j} \;\ge\; 0 \qquad \text{for all } j\ge 0.
\]
\end{theorem}
\begin{proof}
We proceed by double induction on \(k\) and \(m\).
First, observe that for \(k=1\), \(\PD_1(i,j)=\binom{i}{j}\), and McNamara~\cite{McNamara11} proved that every row of the classical Pascal triangle is infinitely log-concavity. Thus the statement holds for \(k=1\) and all \(m\ge 1\).
Now fix \(k\ge 2\) and proceed by induction on the number of applications of \(\LC\).
\textbf{Base case} (\(m=1\)):
By Theorem~\ref{thm:factor},
\[
\LC(\PD_k) = \PD_{k-1} \Had \PD_{k+1}.
\]
Since \(\PD_{k-1}\) and \(\PD_{k+1}\) have non-negative integer entries (as minors of a totally positive matrix), their Hadamard product is non-negative. Hence \(\LC(\PD_k)\ge 0\).
\textbf{Inductive step} (\(m\to m+1\)):
Assume that \(\LC^{\circ \ell}(\PD_\ell) \ge 0\) for all \(\ell\ge 1\) and all \(\ell\le m\). We must show \(\LC^{\circ (m+1)}(\PD_k) \ge 0\).
From the factorization,
\[
\LC^{\circ 2}(\PD_k)
= \LC(\LC(\PD_k))
= \LC(\PD_{k-1} \Had \PD_{k+1}).
\]
Both \(\PD_{k-1}\) and \(\PD_{k+1}\) are row-wise log-concave (by the \(m=1\) case, already proved uniformly for all orders).
Theorem~\ref{thm:hadamard} therefore applies and gives
\[
\LC(\PD_{k-1} \Had \PD_{k+1})
\;\ge\; \LC(\PD_{k-1}) \Had \LC(\PD_{k+1})
= (\PD_{k-2} \Had \PD_k) \Had (\PD_k \Had \PD_{k+2})
\ge 0,
\]
where the final non-negativity follows from the inductive hypothesis applied to orders \(k-2\), \(k\), and \(k+2\) at iteration depth 1.
To reach arbitrary \(m\), we iterate the same argument. Define
\[
Q^{(m)}_\ell \;:=\; \LC^{\circ m}(\PD_\ell).
\]
By repeated application of the Hadamard inequality (Theorem~\ref{thm:hadamard}) and the factorization at each step, we obtain the recursive inequality
\[
Q^{(m+1)}_k
= \LC(Q^{(m)}_k)
= \LC(\PD_{k-1} \Had \PD_{k+1})^{\circ m}
\;\ge\; Q^{(m)}_{k-1} \Had Q^{(m)}_{k+1}.
\]
By the inductive hypothesis, \(Q^{(m)}_{k-1} \ge 0\) and \(Q^{(m)}_{k+1} \ge 0\) (note \(k-1\ge 1\) when \(k\ge 2\); the case \(k=1\) was already settled).
The Hadamard product of two non-negative arrays is again non-negative, so
\[
Q^{(m+1)}_k \;\ge\; 0.
\]
This completes the induction on \(m\).
Boundary and degenerate cases (where some minors vanish) are automatically non-negative by the zero convention.
\end{proof}
\begin{corollary}
The Pascal determinantal arrays \(\PD_k\) (\(k\ge 1\)) are the first non-trivial infinite hierarchy of combinatorial arrays known to be infinitely log-concave in every row, uniformly in the order \(k\).
\end{corollary}
\begin{remark}
The proof is completely algebraic and uniform in \(k\). It does not rely on lattice-path injections (the method used for \(k=1\) in~\cite{McNamara11}) and immediately extends to certain $q$-analogues and deformations of the Pascal matrix that preserve both the Dodgson identity and total positivity.
\end{remark}
With the infinite log-concavity now established, we turn in the final section to consequences and open problems.
\section{Consequences and Geometric Interpretation}\label{sec:consequences}
The factorization $\LC(\PD_k) = \PD_{k-1} \Had \PD_{k+1}$ has several immediate structural consequences and admits a beautiful geometric interpretation in terms of the double-stick bijection introduced in~\cite{TeimooriKhodakarami}.
\subsection{Log-convexity in the determinantal order $k$}
\begin{theorem}[Log-convexity in $k$]\label{thm:logconvex-k}
For every fixed $i,j\ge 0$ and every integer $m\ge 1$, the sequence $\{\PD_k(i,j)\}_{k\ge 0}$ (with $\PD_0(i,j)=1$) is log-convex in $k$:
\[
\PD_{m}(i,j) \cdot \PD_{m+2}(i,j) \;\ge\; \PD_{m+1}(i,j)^2.
\]
\end{theorem}
\begin{proof}
Apply the factorization repeatedly in the $k$-direction. For any fixed $(i,j)$,
\begin{align*}
\PD_{m+1}(i,j)^2
&= \PD_m(i,j) \cdot \PD_{m+2}(i,j) - \LC(\PD_{m+1})_{i,j} \\
&\le \PD_m(i,j) \cdot \PD_{m+2}(i,j),
\end{align*}
with strict inequality unless the corresponding entry of $\LC(\PD_{m+1})$ vanishes.
\end{proof}
\begin{corollary}[Determinantal Hadamard inequality]\label{cor:hadamard-ineq}
For all $k\ge 1$ and all $i,j\ge 0$,
\[
\PD_k \Had \PD_{k+2} \;\ge\; \PD_{k+1} \Had \PD_{k+1}.
\]
\end{corollary}
This strengthens known inequalities for minors of totally positive matrices and reveals a multiplicative convexity phenomenon in the parameter $k$.
\subsection{Geometric interpretation via double sticks}
The entries $\PD_k(i,j)$ count (with signless weights) non-intersecting lattice paths, but the most transparent combinatorial model is the \emph{double-stick bijection} of Teimoori--Khodakarami~\cite{TeimooriKhodakarami}.
A \emph{double stick} of order $k$ centred at $(i,j)$ consists of:
- a bottom stick $B_k(i,j)$: a monotone lattice path of length $k$ along the bottom edge of the $k\times k$ submatrix starting at $(i,j)$;
- a right stick $R_k(i,j)$: a monotone lattice path of length $k$ along the right edge of the same submatrix.
The determinant is the signed difference of all possible pairings, but because the Pascal matrix is totally positive, the Lindström--Gessel--Viennot lemma forces all intersecting pairings to cancel perfectly, leaving
\[
\PD_k(i,j) = \frac{\text{weight of bottom stick } B_k(i,j)}{\text{weight of right stick } R_k(i,j)}.
\]
The \emph{sliding rule} (equivalent to the classical Star-of-David identity for binomial coefficients) asserts that sliding the entire double stick one step along the anti-diagonal preserves the ratio of consecutive entries. This is precisely the cancellation ratio used in the proof of Theorem~\ref{thm:factor}.
Geometrically, the log-concavity defect
\[
\LC(\PD_k)_{i,j}
= \PD_k(i,j)^2 - \PD_k(i,j-1)\PD_k(i,j+1)
\]
measures the failure of the double stick at $(i,j)$ to be the exact geometric mean of the sticks at $(i,j-1)$ and $(i,j+1)$. The factorization shows that this defect is itself the (weighted) number of double sticks of orders $k-1$ and $k+1$ at the same position—an elegant multiplicative decomposition that appears to be unique to the Pascal case among totally positive kernels.
\subsection{Open problems}
\begin{enumerate}
\item \emph{$q$-analogues.} Does an analogous factorization hold for minors of the $q$-Pascal matrix $\binom{i}{j}_q$? Preliminary calculations suggest a $q$-deformation involving $q$-Narayana numbers, but the exact form remains open.
\item \emph{Other totally positive kernels.} The proof of Theorem~\ref{thm:factor} relied only on the Dodgson identity and the sliding-rule ratio. Many other totally positive matrices (Stieltjes, Hilbert, q-Gaussian) satisfy Dodgson-type relations. Determining the extent to which the factorization persists is a natural next step.
\item \emph{Higher-order inequalities.} The iterated factorization produces explicit expressions for $\LC^{\circ m}(\PD_k)$ as multisets of Hadamard products over a binary tree of depth $m$ in the $k$-index. Extracting sharp bounds or closed forms for these higher iterates would deepen our understanding of infinite log-concavity hierarchies.
\end{enumerate}
The results of this paper reveal a remarkably rigid multiplicative structure hidden in the classical Pascal triangle and its higher determinantal analogues—a structure that manifests both algebraically through Dodgson condensation and geometrically through the double-stick model.
\begin{thebibliography}{99}
\bibitem{Brenti94}
F.~Brenti,
\newblock Log-concave and unimodal sequences in algebra, combinatorics, and geometry: an update,
\newblock {\em Contemporary Mathematics} \textbf{178} (1994), 71--89.
\bibitem{FJ11}
S.~Fallat and C.~R.~Johnson,
\newblock {\em Totally Nonnegative Matrices},
\newblock Princeton University Press, 2011.
\bibitem{FZ00}
S.~Fomin and A.~Zelevinsky,
\newblock Total positivity: tests and parametrizations,
\newblock {\em Mathematical Intelligencer} \textbf{22} (2000), no.~1, 23--33.
\bibitem{Karlin68}
S.~Karlin,
\newblock {\em Total Positivity, Vol.~I},
\newblock Stanford University Press, 1968.
\bibitem{McNamara11}
P.~R.~W.~McNamara,
\newblock Infinite log-concavity: developments and conjectures,
\newblock {\em SIAM Journal on Discrete Mathematics} \textbf{25} (2011), no.~4, 1501--1522.
\bibitem{Sagan04}
B.~E.~Sagan,
\newblock Log-concave sequences,
\newblock {\em American Mathematical Monthly} \textbf{111} (2004), no.~5, 413--419.
\bibitem{TeimooriKhodakarami}
H.~Teimoori and H.~Khodakarami,
\newblock Pascal determinantal arrays and a generalization of Rahimpour's determinantal identity,
\newblock arXiv:2301.12481 [math.CO], 2023.
\end{thebibliography}
\end{document}