K-divisibility of the K-functional and Calderón couples (Q791115)

The important Brudnyĭ-Krugljak ''K-divisibility'' theorem enables a great deal of the theory of real method interpolation spaces to be considerably simplified and extended. It can also be expected to be generally useful in other branches of analysis in view of the examples where the K-functional is related to other interesting quantities (moduli of continuity, singular values of operators, etc....). The theorem deserves restating here: Let \(\bar A=(A_ 0,A_ 1)\) be a compatible couple of Banach spaces. Suppose that for some \(a\in A_ 0+A_ 1, K(t,a;\bar A)\leq \sum^{\infty}_{n=1}\psi_ n(t)\) for all \(t>0\), where each \(\psi_ n\) is a positive concave function and \(\sum^{\infty}_{n=1}\psi_ n(1)<\infty.\) Then a can be decomposed into a sum \(a=\sum^{\infty}_{n=1}a_ n\) such that, for some absolute constant \(\gamma\) and each integer n, \(K(t,a;\bar A)\leq \gamma \psi_ n(t)\) for all \(t>0.\) The present paper provides an alternative proof of this result. It improves the original estimate for \(\gamma\) showing that \(\gamma\leq 8\). (More recently A. A. Dimitrier has apparently obtained that \(\gamma \leq(\sqrt{2}+1)^ 2\).) The proof is then adapted to the particular case of Banach lattices where it yields a new and quite general second theorem concerning monotonicity properties of interpolation spaces with respect to the K-functional. To give some idea of this result let us say that a normed intermediate space A with respect to the couple \(\bar A\) is an \(L^ q-K\) space if whenever \(a\in A\), \(b\in A_ 0+A_ 1\) and \((\int^{\infty}_{0}[K(t,b;\bar A)/K(t,a;A)]^ qdt/t)^{1/q}\leq 1\) then \(b\in A\) with \(\| b\|_ A\leq C\| a\|_ A.\) It is known, from the work of Calderón, Mityagin, Sedaev, Semenov and Sparr that for any compatible couples of weighted \(L^ p\) spaces \(\bar L=(L^{p_ 0}\!\!_{w_ 0},L^{p_ 1}\!\!_{w_ 1})\) the class of all interpolation spaces is exactly the class of all \(L^{\infty}-K\) spaces. We call any couple with this property a ''Calderón-Mityagin couple''. The second theorem contains the majority of all previously known results about Calderón-Mityagin couples. It also enables, for example, the determination of the best value of q such that all interpolation spaces with respect to the couple \((L^ p(R^ n),W^{1,p}(R^ n)\) are \(L^ q-K\) spaces \((W^{1,p}=Sobolev\) space). A rather nice feature of the second theorem which has emerged subsequently in joint work of the author and \textit{Per Nilsson} [e.g. in Springer Lecture Notes Math. 1070, 54-65 (1984)] is that the \(q=\infty\) case is best possible in a certain sense. A special case of this result states that, for any order continuous Banach lattice X of measurable functions with the Fatou property, the couple \((X_{w_ 0},X_{w_ 1})\) is a Calderón-Mityagin couple for all choices of weight functions \(w_ 0\), \(w_ 1\) if and only if X is a (weighted) \(L^ p\) space for some p.