Optimal interval refinement near singularities (Q1263021)

Let f be a real analytic function on [-1,1] which can be continued analytically into \({\mathbb{C}}\setminus \{z| z=p+iy\), \(| y| \geq q\}^ p,\)q\(\in {\mathbb{R}}\), \(q>0\), where \(p+iq\) is singularity of f. It is well known the dependence between polynomial approximation of these functions and size of the ellipse of analyticity. The same is true for each partial interval \([x_{j-1},x_ j]\) of a partition \(\tau_ n\) of [-1,1], given by: \(-1=x_ 0<x_ 1<...<x_{j-1}<x_ j<...<x_ n=1.\) Let \(E_ k(\epsilon_ 0,\epsilon_ 1)\) be the ellipse with foci at \(\epsilon_ 0<\epsilon_ 1\) and \(k=a+b/e\), where \(0<b<a\) are the semiaxes and \(e=\sqrt{a^ 2-b^ 2}\). With respect to the partition \(\tau_ n\), define the ellipses \(E_{k_ j}(x_{j-1},x_ j)\), \(1\leq j\leq n\), by \(p+iq\in E_{k_ j}\) \((x_{j-1},x_ j)\), where \(k_ j\) are uniquely determined by \(\tau_ n\). If \(m(\tau_ n)=\min \{k_ 1,...,k_ n\}\), then for every \(p+iq\), \(q>0\) and \(n\in {\mathbb{N}}\), there exists a unique balanced partition \((k_ 1=...=k_ n=k)\) of [-1,1] with \(n+1\) nodes, and if \(\sigma_ n\) is an arbitrary partition of [-1,1] with \(n+1\) nodes then \(m(\sigma_ n)\leq m(\tau_ n)\); the equality holds only if \(\sigma_ n=\tau_ n\). One proves that for any \(n\in {\mathbb{N}}\), there exists a unique optimal partition, which solves a nonlinear difference equation. An asymptotic expression for it is given.