Matrix Stability Analysis
Matrix Stability Analysis#
In the matrix method of stability analysis, the numerical solution at every grid point at time step \(n+1\) is expressed in terms of the values at time step \(n\) as:
where \(\mathbf{u}^{n}=u_{m=0,1,\dotsc,M}^n\) and \(F\) is a matrix that represents the finite-difference operator.
Let us consider again the FTCS scheme for the diffusion equation.
We can write (41) as
which, if we take \(u_0^n=u_M^n=0\), can be cast in matrix form as:
From (45) we can define the recurrence relationship
which makes it clear that the stability of the scheme is related to the matrix \(F\), as \(|\mathbf{u}^{n+1}/\mathbf{u}^{n}|=|F|\), where \(|F|\) is a norm of \(F\).
A suitable norm of \(F\) is the spectral norm:
where \(\rho(F)\) is the spectral radius of \(F\).
For \(F\) in (45), each eigenvalue \(l_p\) and eigenvactor \(\mathbf{v}_p\) satisfies
If the eigenvectors form a complete, linearly independent set, they form a basis with which an arbitrary initial condition \(\mathbf{u}_0\) can be expressed as:
For most cases of common interest, \(F\) indeed has a complete set of eigenvectors
and we can write:
or
For the scheme to be stable, \(\mathbf{u}^{n+1}\) must remain bounded, which implies
The effec of a particular \(l_p\) in the numerical solution will be:
Amplifying, if \(|l_p|> 1\).
Neutral, if \(|l_p|= 1\).
Damping, if \(|l_p|< 1\).
Returning to the numerical solution of the diffusion equation (45), we write:
with
so that \(F=f(G)\) and thus \(l_p=f(g_p)\), where the \(g_p\) are the eigenvalues of G.
Since \(G\) is tridiagonal with structure \(a,b,c\), the \(g_p\) are
and the \(l_p\) are
which is the same as (42), leading to the same stability condition as the Von Neumann stability analysis method.