2
$\begingroup$

I am working with finite elements using domain decomposition in 2D and one of the solutions I need to obtain is the co-normal derivative of the solution along a segment that is the intersection of two subdomains.

From trace theorems and Sobolev spaces, it is known that this derivative is in $H^{-1/2}$, and therefore convergence is also proven in that space. My question is: once I have obtained my discrete FEM solution for this derivative, how can compute its $H^{-1/2}$ norm?

Since it is defined as the dual space of $H^{1/2}$, I have not found any way of effectively computing the $H^{-1/2}$ norm of a function.

Someone mentioned to me that the use of wavelets could be a possible way but I haven't found any resource backing up that claim. I appreciate the help, thanks in advance.

$\endgroup$
  • $\begingroup$ The negative norm is non-computable, it is taking supreme over an infinite dimensional space. You can only approximate it using a mesh-dependent norm. $\endgroup$ – Shuhao Cao Mar 24 '15 at 15:50
  • $\begingroup$ First: are you sure that the trace theorem you are using is optimal? I know that the "basic" trace theorem loses a full derivative while the "sharp" trace theorem for the $H^k$ spaces actually only loses half a derivative. Second, all you can really do is use the definition: try a sequence of test functions of unit norm and conclude that the dual norm is at least the largest value you get from the test functions. $\endgroup$ – Ian Mar 24 '15 at 15:51
  • $\begingroup$ ShuhaoCao: Thank you for the suggestion. I have not used mesh dependent norms before. Ian: For the case we are considering, we are sure about the trace theorem. We have thought about trying with a sequence of functions, but with this approach we are always understimating the norm and we would like at least to overestimate it. $\endgroup$ – MBenedetto Mar 24 '15 at 16:01

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Browse other questions tagged or ask your own question.