4
$\begingroup$

Consider the $1$-D heat equation: $$ u_t = a \Delta u = au_{xx} \\ u(0,t) = u(1,t) = 0 \\ u(x,0) = u_0(x) $$ where $a > 0$ is constant and $u_0$ is given. It is a classic result that the implicit finite difference method BTCS is unconditionally stable.

Now, for $u = u(r)$, consider the analogous problem in the unit disk: $$ u_t = a \Delta u = \dfrac{a}{r}\left(ru_r\right)_r = au_{rr} + \dfrac{a}{r}u_r \\ u(1,t) = 0 \\ u_r(0,t) = 0 \\ u(r,0) = u_0(r) $$

If we assume we have a uniform mesh $0 \leq j \leq N$, $0 \leq i \leq M$, $h=1/N$, $k = 1/M$ with $U_j^i = u(jh,ik)$ then the BTCS scheme is $$ \dfrac{U_0^{i+1} - U_0^i}{k} = 4a\dfrac{U_1^{i+1} - U_0^{i+1}}{h^2} \\ \dfrac{U_j^{i+1} - U_j^i}{k} = a\dfrac{U_{j+1}^{i+1} - U_j^{i+1} + U_{j-1}^{i+1}}{h^2} + a\dfrac{U_{j+1}^{i+1} - U_{j-1}^{i+1}}{2jh^2}, 1 \leq j \leq N $$

I assume this is also unconditionally stable, but we cannot apply von Neumann analysis because the coefficients are no longer constant (and have a singularity at $r=0$). Could anyone confirm that this is in fact unconditionally stable and provide a reference or an outline how to show it? Perhaps by using an energy method in a weighted $L^2$-norm?

$\endgroup$

1 Answer 1

4
$\begingroup$

When you do the von Neumann analysis, I guess you end up with an amplification factor $G=U^{i+1}/U^i$ which depends both on $d=ka/h^2$ and $j$ (and $\theta$). So for stability you have to find $d$ which satisfies $|G(d,j,\theta)|\leq 1$, for $0<\theta<2\pi$ and $0<j<N$.

The equation you are solving is not the same as the 1D equation, so it's not obvious if it will also be unconditionally stable.

$\endgroup$

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.