Limits of zeros of recursive families
The zeros of $f_n=\sum_k\alpha_k\beta_k^n$ accumulate where two dominant bases have equal modulus or where the dominant coefficient vanishes. For polynomials defined by a linear recurrence, a nonreal accumulation point shows that the late members are not real-rooted, so this is a standard tool for proving that real-rootedness fails.
Results
A point $z_0\in\mathbb C$ is a limit of zeros of functions $(F_n)_{n\ge0}$ if every ball around $z_0$ contains a zero of $F_n$ for all sufficiently large $n$ (the liminf notion of Sokal). In the next three theorems let $f_n(z)=\sum_{k\in s}\alpha_k(z)\,\beta_k(z)^n$ with $s$ a finite index set.
Theorem (two dominant bases). Let $\alpha_k,\beta_k$ be analytic at $z_0$. Suppose there are distinct $i,j\in s$ with $\alpha_i(z_0)\ne0$, $\alpha_j(z_0)\ne0$, $\beta_j(z_0)\ne0$ and $|\beta_i(z_0)|=|\beta_j(z_0)|$, that $|\beta_k(z_0)|<|\beta_j(z_0)|$ for every other $k\in s$, and that $\beta_i/\beta_j$ is not constant near $z_0$. Then $z_0$ is a limit of zeros of $(f_n)$.
Theorem (unique dominant base). Let $\alpha_k,\beta_k$ be analytic at $z_0$. Suppose $\beta_i(z_0)\ne0$ and $|\beta_k(z_0)|<|\beta_i(z_0)|$ for every $k\ne i$, and that $\alpha_i(z_0)=0$ while $\alpha_i$ does not vanish identically near $z_0$. Then $z_0$ is a limit of zeros of $(f_n)$.
Theorem (necessary condition). Let $\alpha_k,\beta_k$ be continuous at $z_0$. If $z_0$ is a limit of zeros of $(f_n)$, then at $z_0$ one of the following holds: two distinct bases $\beta_i,\beta_j$ have equal modulus, at least that of every $\beta_k$; or there is $i$ with $\alpha_i(z_0)=0$ and $|\beta_k(z_0)|<|\beta_i(z_0)|$ for all $k\ne i$; or every $\beta_k(z_0)$ vanishes.
Theorem (order-two recurrences). Let $W_{n+2}=A\,W_{n+1}+B\,W_n$ with $W_0=U$, $W_1=V$, and suppose $U,V$ and functions $u,v$ are analytic at $z_0$, where on a ball around $z_0$ we have $u+v=A$, $uv=-B$ and $u\ne v$.
- If $v(z_0)\ne0$, $|u(z_0)|=|v(z_0)|$, $u/v$ is not constant near $z_0$, and $V-vU$ and $uU-V$ do not vanish at $z_0$, then $z_0$ is a limit of zeros of $(W_n)$.
- If $|u(z_0)|\ne|v(z_0)|$ and the coefficient of the root of larger modulus does not vanish at $z_0$ ($V-vU$ if $|v(z_0)|<|u(z_0)|$, $uU-V$ if $|u(z_0)|<|v(z_0)|$), then $z_0$ is not a limit of zeros of $(W_n)$.
- If real polynomials $W_n$ are nonzero for all large $n$ and $z_0$ is a nonreal limit of their zeros, then $W_n$ is not real-rooted for all sufficiently large $n$.
Theorem (Wang–Zhang). Let $a,b,c,d>0$ and $W_0=1$, $W_1=x$, $$W_{n+2}=(a x+b)\,W_{n+1}+(c x+d)\,W_n.$$ Then every $W_n$ is real-rooted if and only if $ad\le bc$.
Use
For $ad>bc$, the order-two theorem gives a nonreal limit of zeros of the Wang–Zhang recurrence, so $W_n$ is not real-rooted for all large $n$.
References
- [Sok04, Theorems 1.5(b), 1.5(a) and 1.5], for the first three theorems. They generalize the Beraha–Kahane–Weiss theorem.
- [WZ17, Theorem 2.1], for affine three-term recurrences.
- [Sok04, Theorems 1.5(b), 1.5(a) and 1.5] Chromatic roots are dense in the whole complex plane. Combinatorics, Probability and Computing, 13(2), 221-261, 2004.
BibTeX
@article{Sokal2004ChromaticRoots, author = {Alan D. Sokal}, title = {Chromatic roots are dense in the whole complex plane}, year = {2004}, journal = {Combinatorics, Probability and Computing}, volume = {13}, number = {2}, pages = {221--261}, doi = {10.1017/S0963548303006023}, url = {https://doi.org/10.1017/S0963548303006023}, eprint = {cond-mat/0012369} } - [WZ17, Theorem 2.1] Root geometry of polynomial sequences III: Type $(1,1)$ with positive coefficients. arXiv e-prints, 2017.
BibTeX
@article{WangZhang2017RootGeometryIII, author = {David G. L. Wang and Jiarui Zhang}, title = {Root geometry of polynomial sequences {{I}{I}{I}}: {T}ype $(1,1)$ with positive coefficients}, year = {2017}, eprint = {1712.06105v1}, url = {https://arxiv.org/abs/1712.06105v1}, journal = {arXiv e-prints} }