Asymptotic zero distribution of the polynomials Ξ ~ n \widetilde{\Xi}_{n} Abstract We consider the polynomials Ξ n \Xi_{n} introduced in [ 5 ] and studied in further details in [ 6 ] , which are expressed in terms of Eulerian polynomials of type B, and study the zero distribution of the rescaled family Ξ ~ n ​ ( x ) := Ξ n ​ ( x ) , n ≥ 2 . \widetilde{\Xi}_{n}(x):=\Xi_{n}(\sqrt{x}),\qquad n\geq 2. Writing the zeros of Ξ ~ n \widetilde{\Xi}_{n} in the interval ( 0 , 1 ) (0,1) as 0 < x n , 1 ≤ ⋯ ≤ x n , n − 1 < 1 01 , and all solutions are z_{k}=\frac{1}{2}\log\frac{1}{x}+\pi ik=:z_{0}+\pi ik,\quad k\in\mathbb{Z}, where z_{0}:=\frac{1}{2}\log\frac{1}{x}=-\frac{1}{2}\log x>0 is the unique pole of smallest modulus (real and positive). Each z_{k} is a simple pole of g . The residue of g at z_{0} is \operatorname{Res}(g,z_{0})=\frac{e^{z_{0}}}{(1-xe^{2z})^{\prime}(z_{0})}=\frac{e^{z_{0}}}{-2xe^{2z_{0}}}=\frac{e^{z_{0}}}{-2x\cdot(1/x)}=-\frac{e^{z_{0}}}{2}. Choose radii 0|z_{0}| implies (R/z_{0})^{-m}\to 0 . Dividing by m and letting m\to\infty yields \lim_{m\to\infty}\frac{1}{m}\,\frac{S_{m+1}(x)}{S_{m}(x)}=\lim_{m\to\infty}\frac{m+1}{m}\,\frac{1}{z_{0}}=\frac{1}{z_{0}}. Because \displaystyle z_{0}=\frac{1}{2}\log\frac{1}{x}=-\frac{1}{2}\log x, we obtain \displaystyle\frac{1}{z_{0}}=\frac{2}{\log(1/x)}=\frac{2}{-\log x}. This proves the claim. ∎ Lemma 1.2 . For every x\in(0,1) we have \lim_{m\to\infty}\frac{1}{m}\frac{B_{m+1}(x)}{B_{m}(x)}=\frac{2(1-x)}{-\log x}. (4) Proof. From [ 6 , Lemma 2.3, Eq.7] , we know \sum_{k=0}^{\infty}(2k+1)^{m}x^{k}=\frac{1}{(1-x)^{m+1}}B_{m}(x),\qquad|x|<1, Putting it in place, we have B_{m}(x)=(1-x)^{m+1}S_{m}(x) . Hence \frac{B_{m+1}(x)}{B_{m}(x)}=(1-x)\frac{S_{m+1}(x)}{S_{m}(x)}. Combining this with lemma 1.1 yields ( 4 ).. ∎ Lemma 1.3 . Let n\in\mathbb{N} and 00 we consider s(x+i\varepsilon) with \varepsilon tending to 0 from above. The Stieltjes inversion formula yields \rho(x)=-\frac{1}{\pi}\,\lim_{\varepsilon\rightarrow 0^{+}}\Im s(x+i\varepsilon),\qquad 00 and u(x)=\frac{\sqrt{x}-1}{\sqrt{x}+1}\in(-1,0). Writing q(x):=\frac{1-\sqrt{x}}{1+\sqrt{x}}\in(0,1),\qquad u(x)=-q(x), and using the branch of the logarithm continuous from the upper half-plane, we obtain \lim_{\varepsilon\rightarrow 0^{+}}\log u(x+i\varepsilon)=\log q(x)+i\pi, where \log on the right-hand side denotes the real natural logarithm of q(x)\in(0,1) . Hence \lim_{\varepsilon\rightarrow 0^{+}}\frac{1}{\log u(x+i\varepsilon)}=\frac{\log q(x)-i\pi}{\log^{2}q(x)+\pi^{2}}. Substituting this limit into ( 8 ) with z=x+i\varepsilon , and noting that all remaining terms in ( 8 ) converge to real limits as \varepsilon\rightarrow 0^{+} , a straightforward computation of the limit of the imaginary parts gives -\frac{1}{\pi}\,\lim_{\varepsilon\rightarrow 0^{+}}\Im s(x+i\varepsilon)=\frac{2}{\sqrt{x}\,(1-x)\bigl(\log^{2}\!\tfrac{1-\sqrt{x}}{1+\sqrt{x}}+\pi^{2}\bigr)}. Thus \mu has density \rho(x) on (0,1) given by ( 7 ) and vanishes off (0,1) , which completes the proof. ∎ Theorem 2.3 (Limiting zero distribution) . Let \widetilde{\Xi}_{n} be defined by ( 1 ), and let \mu_{n} be the empirical zero measure from ( 3 ). Let F_{n} denote the distribution function of \mu_{n} , i.e. F_{n}(x):=\mu_{n}((-\infty,x]),\qquad x\in\mathbb{R} Then F_{n}(x) converges pointwise on \mathbb{R} to the function F(x):=\begin{cases}0,&x\leq 0,\\[2.84526pt] \displaystyle\frac{2}{\pi}\arctan\!\left(\frac{1}{\pi}\log\frac{1+\sqrt{x}}{1-\sqrt{x}}\right),&0