In finite element textbooks there are error estimate inequalities like $e<~h^p$ or $e<~N^{-p/d}$, where h is the FE length scale, d the # of dimensions, p the FE order, N the # of DOFs. Are these only valid for linear PDEs? If not, are these valid for all well defined PDEs with good regularity?


1 Answer 1


What the exact form of these error estimates is depends on a number of factors:

  • What norm you measure the error in. When you state $e < h^p$, I assume you measure the error $e$ as the $H^1$ seminorm of the error $u-u_h$.
  • The regularity of the solution. If the solution is not sufficiently regular, then you cannot expect the full convergence rate.
  • The equation you consider. The estimate may be true for the Laplace equation, but not for first-order equations such as the advection equation. For nonlinear equations, you also often run into technical difficulties that lead to reduced convergence rates and/or the fact that our mathematical techniques are not good enough to actually prove anything.
  • The discretization you use.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.