scientific article; zbMATH DE number 4051361

From MaRDI portal
Publication:3787799





Consider a function f(x), \(x=(x_ 1,...,x_ n)\), defined in the unit cube K \(n=\{x|\) \(0\leq x_ i\leq 1\); \(i=1,2,...,n\}\). We select in K n a net - arbitrary points \(x^{(1)}\), \(...,x^{(N)}\)- and approximate \(f\quad *=\sup \{f(x)| \quad x\in K^ n\}\) by \(f^*_ N=\max \{f(x^{(k)}| \quad 1\leq k\leq N\}.\) The main peculiarity of the paper is the consideration of different Lipschitz constants for different variables: \(f(x)\in Lip\) means that \[ | f(y)-f(x)| \leq \sum^{n}_{i=1}L_ i| y_ i-x_ i| \] for all x and y in K n. For a fixed net denote \(d_{\rho}=\sup \{(f^*-f^*_ N)| \quad f\in Lip\}.\) Theorem 1. For an arbitrary net: \(d_{\rho}\geq c_{\rho}\), where \[ c_{\rho}= \max (s!L_{i_ 1}...L_{i_ s}/N)^{1/s}; \] the maximum is over \(1\leq i_ 1<\) \(...<i_ s\leq n\); \(s=1,2,...,n.\) Theorem 2. For an arbitrary \(P_{\tau}\)-net: \(d_{\rho}\leq Ac_{\rho}\), where \(A=A(n,\tau)\) depends neither on N, nor on \(L_ 1\), \(...,L_ n.\) An optimal net \((d_{\rho}=\min)\) for a specified set of Lipschitz constants is at the same time a very poor choice for certain other sets of these constants (e.g. if several \(L_ i=0)\). Since in real problems we do not know the least \(L_ 1\), \(...,L_ n\) for f(x), optimal nets are useless. On the contrary, using \(P_{\tau}\)-nets we can do without estimates of \(L_ 1\), \(...,L_ n:\) as N increases, \(d_{\rho}\to 0\) and the order of convergence (defined by \(c_{\rho})\) is in all cases the best.











This page was built for publication:

Report a bug (only for logged in users!)Click here to report a bug for this page (MaRDI item Q3787799)