Convergence analysis of a regularized degenerate kernel method for Fredholm integral equations of the first kind (Q911240)

The author considers ill-posed problems of the form \((*)\quad \int^{1}_{0}k(s,t)x(t)dt=g(s)\) (in operator form \(Kx=g)\). Tikhonov's method of regularization consists of replacing (*) by the well-posed problem \(\alpha x(\alpha)+K^*Kx(\alpha)=K^*g,\) where \(K^*\) is the adjoint of K and \(\alpha\) is a small positive parameter. In addition, the author assumes that \(\| g-g^{\delta}\| \leq \delta\) and studies the regularized solution of \((+)\) \(\alpha x^{\delta}(\alpha)+K^*Kx(\alpha)=K^*g^{\delta}\). Put \(\tilde K=K^*K\) and replace \(\tilde K\) by the approximate degenerate kernel \(\tilde k_ n(s,t)=\sum^{n}_{j=1}w^ n_ jk_ j(s)k_ j(t)\quad with\quad k_ j(s)=k(u^ n_ j,s),\) where \(w^ n_ j\geq 0\) are quadrature weights and \(u^ n_ j\) are quadrature nodes. The approximate form of \((+)\) is then \(\alpha x_ n^{\delta}(\alpha)+\tilde K_ nx_ n^{\delta}(\alpha)=K^*g^{\delta}\). The author proves convergence theorems of the form \(x_ n^{\delta}(\alpha)\to x\) in \(L^ 2\) with best possible error estimates. Convergence theorems in \(C^ 0\) are also given without optimal error estimates.