Consider the diffusion equation with a coefficient $\nu$: $$-\nabla \cdot (\nu \nabla u)=f$$ with Dirichlet boundary conditions $u = g_D$ in $\partial \Omega$.

If the coefficient would be constant, then the weak form (see step-74 from deal.ii library) is: $$\begin{align*} &\sum_{K\in {\mathbb T}_h} (\nabla v_h, \nu \nabla u_h)_K\\ &-\sum_{F \in F_h^i} \left\{ \left< [{v_h}], \nu \{ \nabla u_h\} \cdot \mathbf n \right>_F +\left<\{ \nabla v_h \}\cdot \mathbf n,\nu [{u_h}]\right>_F -\left< [{v_h}],\nu \sigma [{u_h}] \right>_F \right\}\\ &-\sum_{F \in F_h^b} \left\{ \left<v_h, \nu \nabla u_h\cdot \mathbf n \right>_F + \left< \nabla v_h \cdot \mathbf n , \nu u_h\right>_F - \left< v_h,\nu \sigma u_h\right>_F \right\}\\ &=(v_h, f)_\Omega - \sum_{F \in F_h^b} \left\{ \left< \nabla v_h \cdot \mathbf n, \nu g_D\right>_F - \left<v_h,\nu \sigma g_D\right>_F \right\}. \end{align*}$$

However, the authors wrote (before writing the weak form)

For simplicity, we assume that the diffusion coefficient $\nu$ is constant here. Note that if $\nu$ is discontinuous, we need to take this into account when computing jump terms on cell faces.

Does this mean that I only have to write $[\nu]$ when I have the over the interior faces $F_h^i$?

My goal is to obtain a weak formulation when $\nu$ is not only constant, but also a function of space, i.e. $\nu(x,y)$


1 Answer 1


I don't think so, your would violate the product rule.

enter image description here

  • $\begingroup$ Thanks @ConvexHull for your answer. So what do you think they mean when they say "you have to take this into account"? $\endgroup$ Aug 15, 2021 at 18:37
  • $\begingroup$ @bobinthebox In their derivation they put $\nu$ outside the brackets. If $\nu$ is a function of space, the weak form would miss some extra terms. $\endgroup$
    – ConvexHull
    Aug 15, 2021 at 18:40
  • $\begingroup$ The first summand they have is okay even if $\nu=\nu(x,y)$, but I have to try to write down all the steps from the beginning. Do you have some reference so that I can check if what I find is the correct weak form? $\endgroup$ Aug 15, 2021 at 18:50
  • $\begingroup$ I guess the generic weak form would be for $\hat u = u - g_D$: $(\nabla v_h, \nu \nabla \hat u)_{L^2(\Omega)} = (v_h, f)_{L^2(\Omega)}$, before problem specific FE spaces enter business. $\endgroup$
    – Dan Doe
    Jan 13 at 16:54

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.