Multivariate Szász growth bound
A polynomial with constant term $1$ whose Jensen-multiplier images are stable has coefficients and polydisc growth bounded in terms of its linear and quadratic coefficients alone. A uniform version shows that limits of stable polynomials grow at most like $e^{Cr^2}$, which is the analytic input to the Borcea–Brändén classification.
Results
Throughout, $f\in\mathbb C[z_i:i\in\sigma]$ is a polynomial in finitely many variables. We write $a(\alpha)$ for the coefficient of $z^\alpha$ in $f$, $|\alpha|=\sum_i\alpha_i$, and $e_i$ for the $i$th unit exponent. A polynomial is stable if it has no zeros when all variables lie in the open upper half-plane (see real stability).
For $\alpha,\beta\in\mathbb N^\sigma$ the generalized Jensen multiplier is $$ J(\alpha,\beta)=\prod_i\frac{(\beta_i)_{\alpha_i}}{\beta_i^{\alpha_i}}, $$ where $(b)_a=b(b-1)\dotsm(b-a+1)$ is the falling factorial, so that $J(\alpha,\beta)=0$ unless $\alpha\le\beta$. The operator $J_\beta$ sends $z^\alpha$ to $J(\alpha,\beta)\,z^\alpha$. Put $$ A=\Bigl(3\Bigl(\sum_i|a(e_i)|\Bigr)^2+2\sum_{i,j}|a(e_i+e_j)|\Bigr)^{1/2},\qquad C=2e^2A^2. $$
Theorem (Szász). If $\xi_1,\dotsc,\xi_k\in\mathbb C$ have $\operatorname{Im}\xi_j\le0$, then $$ \sum_j|\xi_j|^2\le3\Bigl|\sum_j\xi_j\Bigr|^2+2\Bigl|\sum_{i<j}\xi_i\xi_j\Bigr|. $$
In the next two statements assume that $a(0)=1$ and that $J_\beta f$ has no zeros in the open upper half-plane for every $\beta\in\mathbb N^\sigma$. This holds for stable $f$ [BB09, Lemma 5.2].
Corollary (coefficients and growth). For every $\beta$, $$ |a(\beta)|\le|\beta|^{-|\beta|/2}\Bigl(\prod_i\frac{\beta_i^{\beta_i}}{\beta_i!}\Bigr)A^{|\beta|}, $$ and if $r\ge0$ and $|z_i|\le r$ for all $i$, then $$ |f(z)|\le2^{|\sigma|}\,\frac{\sqrt{2e^2-e}}{e-1}\;e^{Cr^2}. $$
Theorem (uniform growth bound; Borcea–Brändén). Let $a\colon\mathbb N^\sigma\to\mathbb C$ and let $\kappa_m(\alpha)\in\mathbb R$ satisfy $\kappa_m(\alpha)\to1$ as $m\to\infty$ for every $\alpha$. Suppose that for each $m$ the polynomial $f_m$ with coefficients $\kappa_m(\alpha)\,a(\alpha)$ is stable or zero. Then there are real constants $B,C$ and an index $m_0$ such that $$ |f_m(z)|\le B\,e^{Cr^2}\qquad(m\ge m_0,\ r\ge0,\ |z_i|\le r\text{ for all }i). $$
Proof
Setting all variables equal to $t$ in $J_\beta f$ gives $\prod_j(1+\xi_jt)$ with $\operatorname{Im}\xi_j\le0$ and leading coefficient $(\beta!/\beta^\beta)\,a(\beta)$, so the Szász estimate and the arithmetic–geometric mean inequality give the coefficient bound, and summing over all monomials gives the growth bound. The uniform bound follows by induction on the number of variables in the support of $a$ and on its minimal total degree, writing $f_m=f_m|_{z_i=0}+z_i\int_0^1\partial_if_m(z_it,\cdot)\,dt$. This induction replaces a step in [BB09, Thm. 5.6] that assumes partial differentiation lowers the largest degree among the minimal support elements by one, which fails for a general support.
References
- [BB09] The Lee–Yang and Pólya–Schur programs. I. Linear operators preserving stability. Inventiones mathematicae, 177(3), 541-569, 2009.
BibTeX
@article{BorceaBranden2009I, doi = {10.1007/s00222-009-0189-3}, url2 = {https://doi.org/10.1007/s00222-009-0189-3}, year = {2009}, month = mar, publisher = {Springer Science and Business Media {LLC}}, volume = {177}, number = {3}, pages = {541--569}, author = {Julius Borcea and Petter Br{\"{a}}nd{\'{e}}n}, title = {The {L}ee--{Y}ang and {P}{\'{o}}lya--{S}chur programs. {I}. {L}inear operators preserving stability}, journal = {Inventiones mathematicae} }