scientific article; zbMATH DE number 19653
[For the entire collection see Zbl 0732.00028.] Let \(X = G/K\) be a compact homogeneous space, \(D\) a closed densely defined \(G\)-invariant operator on \(L^ 2(X)\). An equation \(Df + \partial f/\partial t = 0\) where \(f = f(x,t)\), \(x\in X\), \(t\in\mathbb{R}\), is called an evolution equation. The usual heat equation is the case where \(D\) is the Laplace-Beltrami operator for a \(G\)-invariant Riemannian metric on \(X\). The evolution equation is said to be a discrete observable at \(x_ 0\in X\) if, for every time \(t_ 0\) and every neighbourhood \(V\) of \(x_ 0\), there exists a sequence \(x_ 1,x_ 2,\dots\) in \(V\) such that the solution of this equation matching the observed values \(f(x_ i,t_ 0)\) is unique (we drop here some details). Let us express \(L^ 2(X)\) an an increasing union of \(G\)-invariant finite dimensional subspaces \(E_ r(X)\), \(r=1,2,\dots\). The authors consider two approximations \(f_ r\) and \(\tilde f_ r\) to \(f\). The first one is the orthogonal projection of \(f\) on \(E_ r(X)\) for each \(t\). The second one is defined by: \(\tilde f_ r(x_ k,t_ 0)=f(x_ k,t_ 0)\), \(k=1,2,\dots\), \(\dim E_ r(X)\). Then the total approximation error \(f-\tilde f_ r\) splits as the sum of the ``head error \(f_ r-\tilde f_ r\) and the ``tail error \(f-f_ r\). It turns out that the head error depends linearly on the tail error. The main results of the paper are: asymptotic estimates of the head error at time \(t\) as \(r\to\infty\) and sufficient conditions (on the spectrum of \(D\), the smoothness of the initial data function, etc.) for the head error at time \(t\) to tend to zero as \(r\to\infty\). These results are sharpened when \(X\) is a compact symmetric space. The case of the heat equation on the sphere \(S^ n=SO(n+1)/SO(n)\) is considered in detail.
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 Q3979146)
