\(L^\infty\) estimation of the LDG method for 1-D singularly perturbed convection-diffusion problems (Q378992)

The authors consider the one-dimensional singular perturbed problem \[ -\varepsilon u^{\prime\prime}(x)+a(x)u^{\prime}(x)+b(x)u(x)=f(x)\tag{1} \] \[ x\in\,I=(0,1),\text{\;}u(0)=u(1)=0, \] where \(0<\varepsilon\ll1\) is the perturbation parameter, \(a(x)\geq\alpha>0\), \(a(x)\), \(b(x)\) and \(f(x)\) are sufficiently smooth on \(I\) and satisfy the conditions \(b(x)-a^{\prime}(x)/2\gg c_{0}>0\) or \(b(x)\equiv0\), \(a(x)\equiv\mathrm{const}\). Here, \(\alpha\) and \(c_{0}\) are constants independent of \(\varepsilon\). The authors examine pointwise error estimates of the local discontinuous Galerkin (LDG) method for problem (1). On a Shishkin-type mesh several uniform \(L^{\infty}\) error estimates for the LDG approximation for the solution and its derivative are established. As the total number of elements \(N\) tends to infinity, the error estimates for the solution and its derivative behave like \(O\left(\sqrt{\ln N}\left(\frac{\ln N}{N}\right)^{k+1/2}\right)\) and \(O\left(\left(\frac{\ln N}{N}\right)^{k+1/2}\right)\), correspondingly. Numerical experiments show that these estimates converge faster than \(O\left(\left(\frac{\ln N}{N}\right)^{k+1}\right)\).