When one considers the 1-D diffusion equation in cartesian coordinates

$$\frac{\partial u}{\partial t}=\chi\frac{\partial^2u}{\partial x^2},$$

one finds that the amplification factor for the Crank Nicolson scheme (with central differences in the spatial derivatives) is

$$A=\frac{1-2F\sin^2(k\Delta x/2)}{1+2F\sin^2(k\Delta x/2)},$$

where $F=\chi\Delta t/(\Delta x)^2$. Meaning that this method is unconditionally stable, i.e., the method is stable whatever $F$ is.

Now, let us consider the equation in cylindrical coordinates with a source term:

$$\frac{\partial u}{\partial t}=\chi\frac{1}{r}\frac{\partial}{\partial r}\left[r\frac{\partial u}{\partial r}\right]+S_{ext}$$

with boundary conditions $\partial_rT(r=0)=0$ and $T(r=1)=T_a$.

When $S_{ext}=0$, I get that the amplification factor is given by

$$A=\frac{1-\frac{F}{2}\left[4\sin^2\left(\frac{k\Delta r}{2}\right)-\frac{\sin(k\Delta r)}{q}\right]}{1+\frac{F}{2}\left[4\sin^2\left(\frac{k\Delta r}{2}\right)-\frac{\sin(k\Delta r)}{q}\right]}$$

where $q$ is the cell number. I am not sure if this is the correct result, namely it is strange to me the fact that it depends on $q$. Indeed, for small $q$ it is easy to see that $A$ may become larger than 1 for some of the wavenumbers $k$ and independently of $F$ .


1. Does this mean that the method may become unstable?

2. When onde considers $S_{ext}\neq0$ the same result for $A$ applies?

3. And what about the accuracy in the $S_{ext}\neq0$ case? Because I have found that, with increasing $S_{ext}$, the deviation from the theoretical result increases a lot before the steady-state is reached.

  • $\begingroup$ I am not sure what is your numerical scheme here. How do you deal with $r$ and $1/r$ in your scheme ? And what is $K_{ext}$ ? $\endgroup$ – Héhéhé Jul 19 '19 at 23:11
  • $\begingroup$ I used the Crank Nicolson scheme. Regarding the $1/r$ at $r=0$, I make use of the boundary condition $\partial_ rT(r=0)=0$. Should I insert the discretized equation in my question? About the $K_{ext}$, that was a typo: I should have written $S_{ext}$. $\endgroup$ – AJHC Jul 20 '19 at 8:47
  • $\begingroup$ Yeah I think you should insert it, because it is still not clear for me how you discretize $r$ and $1/r$. $\endgroup$ – Héhéhé Jul 20 '19 at 9:30

The Von Neumann stability analysis only works for linear PDEs with constant coefficients. It is not the case for your problem because of the $r$ and $1/r$ terms.

  • $\begingroup$ And do you know what type of stability analysis could be done in this case? $\endgroup$ – AJHC Jul 20 '19 at 10:02
  • $\begingroup$ I guess energy method could work, but I am not an expert. $\endgroup$ – Héhéhé Jul 20 '19 at 11:01

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.