Norms of Large Toeplitz Band Matrices (Q4942073)

Let \(T(b)=(b_{j-k})_{j,k=0}^\infty\) and \(T_n(b)=(b_{j-k})_{j,k=0}^n\) stand for the infinite and the \((n+1)\times (n+1)\) Toeplitz matrices generated by~\(b\). The authors establish tight estimates for the difference \(\|T(b)\|_p-\|T_n(b)\|_p\) and the difference of the norms of related operators (here \(\|\cdot\|_p\) is the operator norm associated with the \(\ell^p\)-norm, \(1\leqslant p<\infty\)). Let \(A_n=\sum_j\prod_kT_n(b_{jk})\), \(M_p=\lim \|A_n\|_p\) as \(n\to\infty\). The authors give tight two-sided estimates for the difference \(M_p-\|A_n\|_p\). It is known that if \(T(b)\) is a Hermitian Toeplitz band matrix and \(A_n=T_n(b)\), then \(M_2-\|A_n\|_2\) goes to zero with polynomial speed. The authors show that such a slow convergence rate is, ``in a sense, exceptional'' and prove that in the generic case \(M_p-\|A_n\|_p\) approaches zero with exponential speed. The obtained results give good error estimates when computing the norms of certain infinite matrices via their large truncations.