The asymptotic form of the Hermite-Padé approximations for a system of Mittag-Leffler functions (Q2262919)

Simultaneous rational approximation with common denominator has been receiving quite some attention over the last 35 years. The author studies what has become known as Hermite-Padé approximation using \textit{German polynomials} (the name arising from the seminal paper by \textit{K. Mahler} [``Perfect systems'', Compos. Math. 19, 95--166 (1968; Zbl 0168.31303)] -- written during his stay at the Rijksuniversiteit Groningen in 1934/1935). \vskip0.3cm Let \(n,m_1,\ldots,m_k\) be fixed arbitrary non-negative numbers; \(m=\sum_{i=1}^k\,m_i,\;n_j=n+m-m_j\;(1\leq j\leq k)\). For any set of \(k\) functions \(\{f_j(z)\}_{j=1}^k\) of functions holomorphic at the origin, there exist polynomials \(Q_m,\,P_{n_j}^j\) with \[ \text{deg}\,Q_m\leq m,\;\text{deg}\,P_{n_j}^j\leq n_j, \] such that \[ R_{n,m}^j(z)=Q_m(z)f_j(z)-P_{n_j}^j(z)=A_jz^{n++m+1}+{\mathcal O}(z^{m+n+2}). \] (for \(k=1\) the \text {ordinary} Padé approximant is recovered) The main result of the paper uses the Mittag-Leffler functions \[ F_{\gamma}^j(z)={}_1F_1(1,\gamma;\lambda_j z)=\sum_{p=0}^{\infty}\,{\lambda_j^p\over (\gamma)_p}\,z^p\;(1\leq j\leq k), \] (\((\gamma)_0=1,\;(\gamma)_p=\gamma (\gamma+1)\cdots (\gamma+p-1)\text{ for }p\geq 1\)) and is stated in Theorem 2.2. Let \(n=m_1=\cdots=m_k\) and \(\pi_{kn,kn}^j(z;F_{\gamma}^j)\) be the Hermite-Padé approximants for the system of Mittag-Leffler functions \(F_{\gamma}^j(z)\), where \(\lambda_j=ib_j\;(1\leq j\leq k)\) with the \(b_j\) pairwise different non-zero real numbers. Then, we have, for any complex \(z,\,|z|\leq L\) and \(j\in\{1,2,\ldots,k\}\) for \(n\rightarrow\infty\) \[ \begin{align*}{ F_{\gamma}^j(z)-\pi_{kn,kn}^j(z;F_{\gamma}^j)=&(-i)^{(k+1)n+3}\text{sgn}\,(b_j){z^{kn+n+1}\over (\gamma)_{kn+n}} e^{\lambda_j z} e^{z\sum_{k=1}^p\,\lambda_p/(k+1}\cr &\times \sum_p^{\ast}\,(-1)^{pn}\sqrt{-{2\pi\over S_n''(x_p)}}e^{nS_n(x_p)}e^{-izx_p}\left({x_p\over b_j}\right)^{\gamma-1}(1+{\cal O}(1/n)),\cr }\end{align*} \] where the starred summation is rerstricted to the values of \(p\) from \(\{1,2,\ldots,k\}\) for which \(x_p\) lies in the interval with end points \(0\) and \(b_j\). Here \(\varphi(x)=x(x-b_1)\cdots (x-b_k)\), a polynomial having zeros at the points \(b^{\ast},b_1^{\ast},\cdots,b_k^{\ast}\) (in increasing order) and \(x_1,x_2,\ldots,x_k\) are the zeros of \(\varphi'(x)\), enumerated in increasing order (\(\varphi'\) does not have other zeros and in particularly \(S''(x_p)=\varphi''(x_p)/\varphi(x_p)<0\)). The other result given uses \(\gamma=1\) in the Mittag-Leffler functions, leading to the set \(\{e^{\lambda_j z}\}_{j=1}^k\); then \(\gamma_{kn+n}=(kn+n)!\) and the term with \(({x_p/ b_j})\) is equal to \(1\).