Operator matrices preserving ordered PF vectors
A matrix of linear operators on polynomials preserves ordered Pólya frequency vectors if and only if every $2\times2$ submatrix does. For entries among $1$, $\theta=x\,d/dx$ and $x$, exactly thirteen $2\times2$ blocks do, which decides which operator recurrences of this kind have interlacing PF rows.
Results
A polynomial is PF if it has nonnegative coefficients and is zero or has only real, nonpositive roots. On this page $f\ll g$ also holds when $f$ or $g$ is zero, as in the zero-aware convention. A vector $(f_1,\dotsc,f_n)$ is an ordered PF vector if every $f_i$ is PF and $f_i\ll f_k$ for $i<k$.
An $m\times n$ operator matrix $M=(M_{ij})$ has real-linear operators on $\mathbb R[x]$ as entries and acts by $(Mf)_i=\sum_j M_{ij}f_j$. It preserves ordered PF vectors if $Mf$ is an ordered PF vector whenever $f$ is. A block $\begin{pmatrix}a&b\\c&d\end{pmatrix}$ preserves ordered PF pairs if, for all PF $f\ll g$, the polynomials $af+bg$ and $cf+dg$ are PF and $af+bg\ll cf+dg$.
Theorem (local to global). Let $m,n\ge2$. Then $M$ preserves ordered PF vectors if and only if every $2\times2$ submatrix on rows $i<k$ and columns $j<l$, adjacent or not, preserves ordered PF pairs. A $2\times2$ matrix preserves ordered PF vectors if and only if its four entries form a block preserving ordered PF pairs, and products of matrices that preserve ordered PF vectors preserve them.
Theorem (blocks over $\{1,\theta,x\}$). Let the entries be the identity $1$, the operator $\theta$ and multiplication by $x$. A $2\times2$ block over this alphabet preserves ordered PF pairs if and only if it is one of the following thirteen, written as (top row)/(bottom row):
- $(1,1)/(1,1)$, $(1,1)/(\theta,\theta)$, $(1,1)/(x,1)$, $(1,1)/(x,x)$;
- $(\theta,1)/(\theta,1)$, $(\theta,1)/(\theta,\theta)$, $(\theta,1)/(x,\theta)$;
- $(\theta,\theta)/(\theta,\theta)$, $(\theta,\theta)/(x,x)$;
- $(x,1)/(x,1)$, $(x,1)/(x,x)$, $(x,\theta)/(x,\theta)$, $(x,x)/(x,x)$.
Hence an $m\times n$ matrix over $\{1,\theta,x\}$ with $m,n\ge2$ preserves ordered PF vectors if and only if every $2\times2$ submatrix is one of the thirteen blocks. A one-row matrix does if and only if its entries weakly decrease from left to right in the order $1<\theta<x$, and a one-column matrix if and only if its entries weakly increase from top to bottom. For $m,n\ge2$, a matrix over $\{1,\theta\}$ does exactly when its entries equal to $1$ form an upper-right rectangle, possibly empty ($mn+1$ matrices), and a matrix over $\{\theta,x\}$ does exactly when all rows are equal, with $x$ columns followed by $\theta$ columns, or all columns are equal, with $\theta$ rows followed by $x$ rows ($m+n$ matrices).
Proposition. For PF polynomials $p$ and $q$:
- if $p\ll q$, then $p'+q'\ll q+xp'$;
- $p\ll q$ if and only if $q\ll xp$;
- if $p\ll q$, then $p+q\ll xp+q$.
Corollary (Euler-derivative lag recurrence). Let $P_{n+2}=\theta P_{n+1}+x\,\theta P_n$. The matrix $$ \begin{pmatrix}x\theta&\theta\\ x(\theta^2+\theta)&x\theta+\theta^2\end{pmatrix} $$ sends $(P_n,P_{n+1})$ to $(P_{n+2},P_{n+3})$ and preserves ordered PF vectors. Among the nine transitions between the lag factors $1$, $\theta$ and $x$, exactly the two alternating identity/Euler transitions preserve ordered PF vectors. If $P_0=1$ and $P_1=x$, then every $P_n$ is real-rooted.
For couplings of two first-order blocks $a+b\theta+cx+e\,x(N-\theta)$, a scalar criterion on the negative ray with a weight $w\ge0$ reduces to explicit inequalities. For the pair $(\theta,\,x(N-\theta))$ it holds exactly when $N^2\le4w$, and for the reversed pair, with $N\ge1$, exactly when $(N-1)^2\le4w$.
References
- Matrices preserving interlacing sequences, for entries that are multiplications by polynomials
- Operators preserving interlacing