Skip to main content

Section 2.3 Convergence, Consistency and Stability of Finite Difference Methods

Unlike the ODEs case, for PDEs there is no general result about convergence and consistency of fixed-stepsize numerical methods except in the linear case. Hence, in this section we focus on linear parabolic or hyperbolic PDEs, namely PDEs of the type
\begin{equation*} u_t=Lu \end{equation*}
where \(L\) is a differential operator with respect to space coordinates.
Definition 2.3.1.
Consider a rectangle \(R=[0,T]\times[0,L]\) and an integer \(M>1\text{.}\) By a Finite Difference fixed-stepsize method on \(R\) with time stepsize \(h\) and space step-size \(k=L/M\) we mean a map
\begin{equation*} \Phi_{h}:\bR^{M+1}\to\bR^M. \end{equation*}
This map represents the advance of the numerical solution from time \(t\) to time \(t+h\text{.}\)
Consider the linead Advection PDE \(u_t+u_x=0\text{.}\) In this case \(L=-d/dx\text{.}\) The map corresponding to the FTBS method is
\begin{equation*} \Phi_h(t,u_1,u_2,\dots,u_M) = (a(t+h),u_2-\nu(u_2-u_1),\dots,u_M-\nu(u_M-u_{M-1})), \end{equation*}
where \(\nu=h/k\) and the function \(a(t)\) in the expression above is the extra boundary condition needed by the FTBS method, representing the values of the solution at \(x=0\text{.}\)

Using this map, the FTBS method can be written as
\begin{equation*} (u_{n+1,1},u_{n+1,2},\dots,u_{n+1,M}) = \Phi_h(t_n,u_{n,1},u_{n,2},\dots,u_{n,M}). \end{equation*}
Notice that \(\Phi_h\) does not depend on time since \(L\) does not.
Definition 2.3.3.
The Finite Difference method \(\Phi_h\) has order of accuracy\(\;p\) in the interval \([0,T]\) if, given a solution \(u\) of the linear PDE \(u_t=Lu\text{,}\) we have that
\begin{equation*} \|u(t+h)-\Phi_h u(t)\|=O(h^{p+1}) \end{equation*}
for \(h\to0\) and any \(t\in[0,T]\text{.}\) \(\Phi_h\) is consistent if it has order of accuracy \(p>0\text{.}\)
Consider the method
\begin{equation*} \Phi_h(u_{m,n}) = u_{m,n} + \frac{h}{2k}\left(u_{m,n+1}-u_{m,n-1}\right) \end{equation*}
with
\begin{equation*} k = h^{p/2}. \end{equation*}
Then \(\Phi_h\) has accuracy order \(p\) for each \(p\in[0,2]\)
Definition 2.3.5.
The Finite Difference method \(\Phi_h\) is convergent in the interval \([0,T]\) if, given a solution \(u\) of the linear PDE \(u_t=Lu\) with initial condition \(u(0)=u_0\text{,}\) we have that
\begin{equation*} \lim_{n\to\infty}\|u(t)-\Phi^n_{t/n}u_0\|=0. \end{equation*}
Definition 2.3.6.
The Finite Difference method \(\Phi_h\) is stable in the interval \([0,T]\) if, for some constant \(C>0\text{,}\) we have that
\begin{equation*} \sup_{nh\in[0,T]}\|\Phi^n_{h}\|\leq C \end{equation*}