We consider the finite difference method for the approximation

$\left\{\begin{matrix} -u''(x)+q(x)u(x)=f(x)\\ u'(a)=u'(b)=0 \end{matrix}\right.$

and let $K$ be the $(N+2) \times (N+2)$ matrix of the method. Let $v \in \mathbb{R}^{N+2}, v=\begin{pmatrix} v_0\\ v_1\\ \dots\\ \dots\\ \dots\\ v_{N+1} \end{pmatrix}$

If $(Ku)_i \leq 0 \forall i=0,1, \dots, N-1$ then $\max_{1 \leq i \leq N} \{ v_i \} \leq \max \{ v_0, v_{N+1},0\} (\star)$


We suppose that $(\star)$ does not hold.

Then $\exists n \in \{ 1,2, \dots, N\}$ such that $v_n= \max_{1 \leq i \leq N} v_i >0$, $v_n>v_0$ and $v_n > v_{N+1}$.

Now $(Ku)_n=u_{n-1}+(2+h^2 q(x_n))v_n-v_{n+1} \leq 0 \Rightarrow (2+h^2 q(x_n)) v_n \leq v_{n-1}+v_{n+1}$

but $v_n \geq v_{n-1}$ and $v_n \geq v_{n+1} \Rightarrow 2 v_n \geq v_{n-1}+v_{n+1}$

Thus, $(2+h^2 q(x_n))v_n \leq v_{n-1}+v_{n+1} \leq 2 v_n$

This can only hold if $q(x_n)=0$ and $v_n=v_{n-1}=v_{n+1} \Rightarrow \dots \Rightarrow v_i=v_n \forall i$, contradiction.

Could you explain me why it holds that $(Ku)_n=u_{n-1}+(2+h^2 q(x_n))v_n-v_{n+1}$ ?

  • $\begingroup$ do you mean that $v$ is an approximate solution to your problem? Do you try to prove some version of maximum principle for this equation? Why do you need finite differences for that? $\endgroup$ – Michael Medvinsky Nov 14 '15 at 21:39

You first have to verify that the property holds for the analytical solution. This property does not hold every time, for example $$u=\sin^3 x, \quad x\in[a,b]=[0,\pi]$$ satisfies the equation and the boundary condition for every $q(x)$ with $$f(x)=q(x)+3) \sin^3x-6 \sin x \cos ^2x$$ Note that $u(0)=0$ , $u(\pi)=0$ but $u(\pi/2)=1$

Consequently, the property cannot hold for discrete solution.

Finally, the maximum is not necessarily holds at the end points of the interval.


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.