Optimization of direct methods of solving periodic boundary value problems (Q1413076)

The paper is concerned with direct methods for solving the periodic boundary value problems \[ Ax \equiv x^{(m)}(s)+B(x;s)=y(s) \quad (x \in X,\;y \in Y,\;s \in (-\infty,\infty)), \] \[ x^{(k)}(0)=x^{(k)}(2 \pi), \quad k=0, \dots, m-1, \] where \(X\) and \(Y\) are spaces of \(2\pi\)-periodic functions, \(B: X \to Y\) is a linear operator. The case of integro-differential operators \(B\) is included. Suppose \(A\) and \(A^{-1}\) are bounded. Let \(X_n=H_n^T \cap X, Y_n=H_n^T \cap Y\), where \(H_n^T\) is a subspace of all trigonometric polynomials of order not exceeding \(n\). The equation is approximated by its finite-dimensional version \[ x_n^{(m)}(s)+B_n(x_n;s)=y_n(s)\quad (x_n \in X_n,\;y_n \in Y_n,\;B_n:X_n \to Y_n), \] where the families \(\{ B_n \}\) and \(\{ y_n \}\) specify a solution method for the original problem. For various ways of choosing \(X\) and \(Y\), the author establishes error estimates and proves that several known finite-dimensional methods such as the Galerkin method, the method of subdomains, and method of the collocation are optimal (asymptotically optimal, optimal by order) over the family of all direct methods associated with the choice of \(H_n^T\) as a discretization tool.