Meromorphic Szegő functions and asymptotic series for Verblunsky coefficients (Q856800)

Authors' introduction: This paper is concerned with the spectral theory of orthogonal polynomials on the unit circle (OPUC) in the case of particularly regular measures. Throughout, we will consider probability measures on \(\partial D=\{z\mid|z|=1\}\) of the form \[ d\mu =w(\theta)\frac{\partial\theta} {2\pi}+d\mu_s,\tag{1} \] where \(w\) obeys the Szegő condition, that is, \[ \int^{2\pi}_0\log\bigl(w(\theta) \bigr)\frac{d\theta}{2\pi}>-\infty.\tag{2} \] In that case, the Szegő function is defined by \[ D(x)=\exp\left(\int^{2\pi}_0 \frac{e^{i\theta}+ z}{e^{i\theta}-z}\log\bigl( w(\theta)\bigr)\frac{d\theta}{2\pi} \right). \tag{3} \] Not only does \(w\) determine \(D\), but \(D\) determines \(w\), since \(\lim_{r\uparrow 1}D(re^{i\theta})\equiv D(e^{i\theta})\) exists for a.e. \(\theta\) and \[ \omega(\theta)=\bigl|D(e^{i\theta})\bigr|^2. \tag{4} \] Indeed, \(D\) is the unique function analytic on \({\mathbf D}=\{z \mid|z|<1\}\) with \(D(0)>0\) and \(D\) nonvanishing on \({\mathbf D}\) so that (1.4) holds. Given \(d\mu\), we let \(\Phi_n\) be the monic orthogonal polynomial and \(\varphi_n=\Phi_n/\| \Phi_n\|_{L^2(d\mu)}\). The \(\Phi_n\)'s obey the Szegő recursion \[ \Phi_{n+1} (z)=z\Phi_n(z)-\overline\alpha_n\Phi^*_n(z).\tag{5} \] where, for \(P_n\) a polynomial of degree \(n\), \[ P^*_n(z)=z^n\overline{P_n(1/\overline z)}.\tag{6} \] The \(\alpha_n\) are called Verblunsky coefficients. They lie in \({\mathbf D}\), and \(\mu\mapsto\{a_n\}^\infty_{n=0}\) is a bijection of nontrivial measures on \(\partial{\mathbf D}\) and \({\mathbf D}^\infty\). Our goal here is to focus on the map and its inverse. Here is the background on our first main result: (A) Nevai and Totik proved that \(\limsup_{n\to \infty}|\alpha_n|^{1/n}\leq R^{-1}<1\) if and only if (a) \(d\mu\) obeys the Szegő condition and \(d\mu_s=0\); (b) \(D(z)^{-1}\) is analytic in \(\{z\mid|z|<R\}\). (B) Barrios, López and Saff proved that for \(R>1\). \[ \alpha_n=cR^{-n}+O\bigl(\bigl((1-\varepsilon)R^{-1}\bigr)^n\bigr)\tag{7} \] if and only if \(D(z)^{-1}\) is meromorphic in a circle of radius \(R(1+\delta)\) with a single, simple pole at \(z=R\). (C) Simon considered the functions \[ S(z)=-\sum^\infty_{j=0} \alpha_{j-1}z^j\tag{8} \] (with \(\alpha_{-1}=-1)\) and \[ r(z)=\overline {D(1/\xi)}D(z)^{-1},\tag{9} \] and proved that if \(\limsup_{n\to\infty} |\alpha_n|^{1/n}\leq R^{-1}<1\), then for some \(\delta>0\), \(r(z)-S(z)\) is analytic in \(\{z\mid 1-\delta<|z|<R^2\}\), so that \(S(z)\) and \(r(z)\), which will have singularities on \(|z|=R\) if \(\limsup_{n\to \infty}| \alpha_n|^{1/n}=R^{-1}\), must have the same singularities in \(\{z\mid R\leq|z|<R^2\}\). In [\textit{B. Simon}, Orthogonal polynomials on the unit circle. Part 1: Classical theory (2005; Zbl 1082.42020)], instead of \(S(z)\) as defined by (8), one has \(S(z)\) defined by \(S_{\text{book}}(z)=\sum^\infty_{j=0}\alpha_jz^j\), and the theorem is stated as analyticity of \(z^{-1}r(z)+S_{\text{book}} (z)\), equivalent to analyticity of \(r-S\). But, as we explain in \S4, (8) is the more natural object. Rather than \(1-\delta<|z|<R^2\), has \(R^{-1}<|z|<R^2\), but that is wrong since \(\overline{D(1/z)}\) can have poles at the Nevai Totik zeros. (D) Using Riemann-Hilbert, methods, Deift and Ostensson have extended the result on analyticity of \(r(z)-S(z)\) to \(\{z\mid 1-\delta<|z|<R^3\}\). (E) Barrios, López and Saff have proven that if \[ \alpha_n=cR^{-n}+O\biggl( \bigl((1+ \varepsilon)R\bigr)^{-n-nm^2}\biggr)\tag{10} \] then \(D(z)^{-1}\) is meromorphic in \(\{z\mid|z|<R^{2m-1}+\delta\}\) with poles precisely at \(z_k=R^{2 k-1}\), \(k=1,2,\dots,m\). In particular, if (10) holds for all \(n\), then \(D(z)^{-1}\) is entire meromorphic except for poles at \(R^{2k -1}\), \(k=1,2,\dots\). Our main goal in this paper is to give a complete analysis of what can be said about \(\alpha_n\) if \(D(z)^{-1}\) is meromorphic in some disk and, contrariwise, about \(D(z)^{-1}\) if \(\alpha_n\) has an asymptotic expansion as a sum of exponentials. We describe our precise results in \S4. Along the way, we found a direct, simple proof of the Deift-Ostensson result that is also simpler than the argument Simon used for his weaker result. So we will give this proof next, then analyze two simple examples, and return in \S4 to a general overview and sketch of the rest of the paper. Of course, included among the entire meromorphic functions are the rational functions, and there is prior literature on this case. Szabados considered the case \(D(z)^{-1}=1/q(z)\) for a polynomial \(q\), and Ismail and Ruedemann and Pakula discussed \(D(z)^{-1}=p(z)/q(z)\) for polynomials \(p\) and \(q\). They have some results on asymptotics of \(\Phi_n\) but no discussion of links to the \(\{\alpha_n\}^\infty_{n=1}\). As was completing this manuscript, I received the latest draft of a paper of Martínez-Finkelshtein, McLaughlin and Saff that has some overlap with this paper. After refereeing, I received a preprint of Golinskii and Zlatoš with an explicit formula for the Taylor coefficients of \(D(z)^{-1}/D(0)^{-1}\) in terms of the \(\alpha\)'s that may provide another proof of our results.