If we discretize a parabolic pde to obtain the system of ODE's $\frac{\boldsymbol{B}}{\Delta t} \boldsymbol{u}_k = (\boldsymbol{K} + \frac{\boldsymbol{B}}{\Delta t}) \boldsymbol{u}_{k-1} + \boldsymbol{f}_k$ where $\boldsymbol{B}$ is the mass while $\boldsymbol{K}$ is the stiffness matrix, is there any condition on $\frac{\boldsymbol{B}}{\Delta t}$ such that this system is stable?

This system results from discretizing a parabolic PDE via forward Euler.

  • $\begingroup$ You are missing the $B$ matrix on the right hand side of your equation. $\endgroup$ Feb 7, 2020 at 4:51

1 Answer 1


Well, boy, do I have the resource for you :-) Take a look at lectures 26 and 27 at https://www.math.colostate.edu/~bangerth/videos.html

  • $\begingroup$ Hi @Wolfgang, thanks so much for your help. There are 2 things I'd like to ask: 1) In your notes, you write that the mass matrix $M \approx h^d I$ where $h$ is the discretization length in the spatial domain. Any references (i.e. books) for more details on this? 2) What about a CFL-type condition for FEM? Must there be a relationship between $\|M\|_2$ and $\Delta t$? Any references would be appreciated. $\endgroup$
    – secondrate
    Feb 8, 2020 at 15:54
  • $\begingroup$ @secondrate: About the relationship of the mass matrix and $h^dI$: Think of the relationship of any matrix $K$ to an operator $\cal K$ as $K_{ij}=(\varphi_i, {\cal K} \varphi_j)$. Then it's clear that the operator that $M$ corresponds to is simply the identity operator. The $h^d$ comes out of the integration: $\varphi_i$ and $\varphi_j$ only overlap on the set of cells adjacent to one vertex, and the total integration area/volume where the product is nonzero is then of order $h^d$ in $d$ dimensions. $\endgroup$ Feb 9, 2020 at 17:46
  • $\begingroup$ I don't understand your second question, though. Are you asking whether the 2-norm of $M$ somehow enters the CFL limit for $\Delta t$? The answer is no. $\endgroup$ Feb 9, 2020 at 17:48

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.