I am looking for reference on how to implement nonlinear boundary conditions. Specifically, I am interested in implementing a radiation boundary condition for heat transfer with the FEM:

$-k \frac {\partial T} {\partial n} =F (T^{4} - T_{\infty}^{4})$

with $T$ temparature, $F$ is some factor, $k$ the thermal conductivity.

An ideal reference has the theory and a simple (1D/2D) example and possibly some code to try/verify. I have looked at Finite Element Analysis for Heat Transfer (Hou-Cheng Huang) but for the price the review was not favorable. Maybe some knows of some lecture nodes, a deal ii or fenics implementation I could look at?


This is not particularly difficult once you realize that the nonlinear boundary condition simply yields a nonlinear term in the weak formulation. Let's assume that you want to solve the steady state problem, i.e., $$ -k \Delta T = f \qquad \text{in} \ \Omega\\ -k \frac{\partial T}{\partial n} = F(T^4 - T_\infty^4) \qquad \text{on} \ \partial\Omega. $$ You convert this into the weak form by multiplying the equation by a test function and integrating by parts, which yields $$ (\nabla \varphi,k\nabla T)_\Omega -(\varphi,k\partial_n T)_{\partial\Omega} = (\varphi,f)_\Omega $$ Now realize that you can substitute the boundary condition in the second term on the left to obtain $$ (\nabla \varphi,k\nabla T)_\Omega +(\varphi,F T^4)_{\partial\Omega} = (\varphi,f)_\Omega + (\varphi,F T_\infty^4)_{\partial\Omega} $$ You then only have to implement a nonlinear solver for this problem -- but that works the same way as for all other nonlinear problems and you can find many references for this. (Among them are my video lectures on nonlinear problems.)

| cite | improve this answer | |
  • $\begingroup$ Thanks. I have a non-linear solver working just fine, also for linear Robin BCs. But what I miss to understand is how I should linearize the $(\varphi,F T^4)_{\partial\Omega}$ term. I have seen your video lectures (great!) but could not find anything about linearization of boundary terms. $\endgroup$ – user21 Jan 10 '18 at 15:10
  • 1
    $\begingroup$ It works exactly the same as for nonlinear domain terms. You write down the residual of the weak form (which now has a nonlinear boundary term, instead of a nonlinear domain term) and compute its derivative via the $\lim_{\varepsilon\rightarrow 0}$ process. $\endgroup$ – Wolfgang Bangerth Jan 10 '18 at 16:58
  • $\begingroup$ The comment lead me in the right direction. It works now. Thx. $\endgroup$ – user21 Jan 15 '18 at 6:41

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.