Finite range decompositions of positive-definite functions (Q2497938)

The article is devoted to the problem of obtaining some sufficient conditions for a positive-definite function to admit decomposition into a sum of positive-definite functions which are compactly supported within disks of increasing diameters \(L^n\). The consideration is carried out for the more general case of positive-definite bilinear forms \(f\to v(f,f)\) defined on \(C^\infty_0\). Definition. Let \(v\) be a translation invariant bilinear form. We say that \(v\) admits a translation invariant finite range decomposition if, for some \(L>1\), there exist positive-definite forms \(G_j\) such that \[ v=\sum_{j\geq 1}G_j; \tag{1} \] \[ G_j(f,g)= 0 \text{ if dist(supp\,\(f\),\;supp\,g})\geq L^j; \tag{2} \] \[ \text{for }j\in{\mathbb N},\,G_j \text{ is translation invariant}. \tag{3} \] The authors prove that such decompositions exist when \(v\) is dual to a bilinear form \(\varphi\to\int| B\varphi| ^2\), where \(B\) is a vector valued partial differential operator satisfying some regularity conditions. The main result on the existence of such decomposition concerns bilinear forms associated to the Green's function of a large class of elliptic partial differential operators. It also gives a decomposition if \(v\) is not translation invariant and then condition (3) is replaced by a kind of uniformity of convergence of \(\sum_{j\geq 1}G_j\). The authors obtain a more explicit form of this decomposition. Section 4 of the article is especially devoted to concrete examples based on the construction used to prove the main results on the existence in Section 3. The last section discusses the problem of smoothness of a finite range decomposition \(G=\sum_1^\infty G_j\), i.e., when the forms \(G_j\) have kernels \(g_j(x,y)\) such that \[ G_j(f,f)= \iint f(x)g_j (x,y)f(y)\,dx\,dy \] and, for \(j\geq 2\), these kernels \(g_j\) are \(C^\infty\). The authors perform some calculations for the Laplacian to construct a finite range decomposition with \(C^\infty\) smoothness. Let us also note that the article contains several references to some problems that are of theoretical as well as applied interest. Thus, the article should be interesting not only for specialists in harmonic analysis, but also in such areas as probability, partial differential equations, functional analysis, etc.