I'm working with error estimates for Poisson's equation of the form

$$\mathcal{E}^2_T = h_T^2\|-\Delta u - f\|_{L^2(T)} + \sum_{e\in \partial T} h_e\|n\cdot \nabla u\|_{L^2(e)}$$

where $T$ is an element and $h_T, h_e$ are representative sizes of the element volume and face, respectively.

For skewed elements, I'm unclear on what is used for $h_T$ in practice. I can imagine several options (radius of the largest circumscribable sphere/circle, $d$th root of the volume $|T|^{1/d}$, max edge length) for $h_T$ (though I believe $h_e$ is always just given to be the size of the face?).

Is there a preferred choice of for (slightly) skewed elements?

  • $\begingroup$ The cell residual must be $\Delta u +f$, not $\Delta u -f$. $\endgroup$ – Wolfgang Bangerth Dec 10 '13 at 10:26

You need to trace where this term comes from in the derivation of these error estimates. In your case, you get this from an interpolation estimate of the form $\|z-I_h z\|_{L^2(e)} \le h^{?} \|\nabla^2 z\|_{L^2(T)}$, if I recall correctly. You'll need to look up what exactly the estimate is, what it involves, and where the transformation from reference to real cell/edge enters here. You will then find that $h$ is related to the norm of the mapping from reference to real cell. What it is here, I don't know, but it can be found out.

  • $\begingroup$ Thanks - it appears to be ${\rm diam} T$, which matches the circumscribed circle diameter idea. $\endgroup$ – Jesse Chan Dec 11 '13 at 3:19

Your Answer

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

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