On \(L^p\)-discrepancy of signed measures (Q5956457)

In this deep paper the authors continue their ongoing research on discrepancy theory for signed measures. The setting of the paper is the set of all real valued functions \(g\) with \[ \|g\|_{L^P(d\tau)} =\left(\int |g|^p d\tau\right)^{1/p} < \infty, \] where \(\tau\) is a positive unit Borel measure on \(I=[-1,1]\) and \(1\leq p<\infty\); \(\tau\) is assumed to be absolutely continuous, i.e., \(d\tau(x)=w(x) dx\) with \(w\in L^1(dx)\). For \(x\in I\) the interval \([x,1]\) is denoted by \(I_x\) and for a signed finite Borel measure on \(I\) the \(L^p( d\tau)\) discrepancy of \(\sigma\) w.r.t. \(\tau\) is defined by \[ D^p[\sigma, d\tau]=\left(\int_{-1}^1 |\sigma(I_x)|^p d\tau(x)\right)^{1/p}. \] It is outside the scope of a (short) review to reproduce the main results here (quite a number of intermediate formulae are needed to define everything), but the application on weak-star limits of counting measures leads to: 1. Let the weight \(w(x)\) satisfy the Szegő condition \[ \int_{-1}^1 {\log{w(x)}\over \sqrt{1-x^2}} dx>-\infty, \] then \[ D^1[\mu-\nu_n], d\mu]\leq c{(\log{n})^2\over n} (n\geq 2; c=c(w)>0\text{ a constant}). \] (here \(\mu\) is the equilibrium measure \( d\mu(x)= dx/(\pi\sqrt{1-x^2}), x\in I\)). 2. Let \(\nu_n\) be the zero-counting measure of the Pollaczek polynomials \(P_n(x;a,b)\), then for \(n\geq 2, 1\leq p<\infty\): \[ D^p[\mu-\nu_n, d\mu]\leq c\left({1\over n}\right)^{(p+1)/(2p)}(\log{n})^2 \] with some constant \(c=c(a,b,p)>0\). 3. Let \(\nu_n\) be as in the previous result, then for \(n\geq 2, 1<p<\infty\): \[ D^p[\mu-\nu_n, dx]\leq c\left({1\over n}\right)^{(p+2)/(2p)}(\log{n})^2, \] where \(c=c(a,b,p)>0\). These bounds are essentially smaller than the upper bounds found using the uniform discrepancy.