On multivariate polynomial \(L^ 2\)-approximation to zero (Q1059242)

This paper generalizes, by passing to many variables, some classical results of approximation theory in \(L^ 2\). This generalization refers to the estimation of the coefficients \(a_ n\) and \(a_{n-1}\) of any polynomial \(Q_ n(x)=a_ 0+a_ 1x+...+a_{n-1}x^{n-1}+a_ nx^ n.\) So, if we denote \(Q^ r_ n(x)=\sum_{| k| \leq n}a_ kx^ k\) where \(Q^ r_ n\) is the polynomial of total degree n in r variables, \(k=(k_ 1,k_ 2,...,k_ r)\in Z^ r_+\), \(x=(x_ 1,x_ 2,...,x_ r)\in R^ r,| k| =k_ 1+k_ 2+...+k_ r\), \(a_ k\in R\), then one shows that for \(| k| =n\) one has \(| a_ k| \leq 2^{-\bar r/2}(\prod^{r}_{q=1}h_{k_ q,k_ q}(2k_ q+1)^{1/2})\| Q^ r_ n\|\) and \(\sum_{| k| =n}a_ k\leq 2^ n(\sum_{| k| =n}2^{-\bar r}\prod^{r}_{q=1}(2k_ q+1)\left( \begin{matrix} 2k_ q\\ k_ q\end{matrix} \right)^ 2)^{1/2}\) where \(\bar r\) is the number of components \(k_ q\neq 0\) of k; \(h_{n,n}=2^{-n}\left( \begin{matrix} 2n\\ n\end{matrix} \right)\) and \(\| Q^ r_ n\| =\| Q^ r_ n\|_ 2=(\int_{I^ r}(Q^ r_ n(x))^ 2dx)^{1/2}\), \(I^ r=[-1,1]^ r\). One also gives some estimations for \(a_ k\) with \(| k| =n-1\) and \(\sum_{| k| =n-1}a_ k\) and one precises the equality cases.