On the best constant in the error bound for the \(H_0^1\)-projection into piecewise polynomial spaces (Q1266108)

The authors present optimal error bounds for approximation of a class of functions by piecewise polynomials of second and third degrees. The class of functions includes solutions of Dirichlet problem for elliptic differential equations in one space dimension. Consider functions \(u\in H^1_0\cap H^2\) on the interval \([0,1]\). For each partition \(\Delta\) of \([0,1]\), define the space \(S_{\Delta,N}\) to consist of polynomials of degree \(\leq N+1\) on each subinterval of \(\Delta\) and satisfying Dirichlet boundary conditions. Let \(P_{\Delta,N}u\) denote the \(H_0^1\)-projection of \(u\) into \(S_{\Delta,N}\) and consider the error estimates \[ \| u'-(P_{\Delta,N}u)'\| \leq C_N h \| u''\| , \] where \(h\) is the length of the largest subinterval of \(\Delta\). It is known that \(C_0=1/\pi\) is the best possible. The authors prove that the best possible value of \(C_1=1/(2\pi)\) and that the best possible value of \(C_2\) satisfies \(1/8.98954\leq C_2\leq 1/8.92338\). The proofs rely on comparisons between \(P_{\Delta,N}u\) and \(P_{\Delta,N}U\), where \(U\) is the Fourier series of \(u\). The authors also remark that these best estimates hold in two dimensions through use of tensor product spaces.