previous overview next
The Main Result
$
\def \EN {\quad \mbox{and} \quad}
$
Consider a one-dimensional uniform mesh with grid-spacing $dx$. Neighbouring
grid-points in the mesh are coupled by coefficients $a$ in forward direction
and by coefficients $b$ in backward direction. Setting up linear equations for
such a grid gives rise to a tri-diagonal matrix, which can be (re-)normalized
to obtain $1$'s on the main diagonal:
$$
\left[ \begin{array}{ccccccc} . & . & . & & & & \\
& -b & 1 & -a & & & \\
& & -b & 1 & -a & & \\
& & & -b & 1 & -a & \\
& & & & . & . & .
\end{array} \right]
$$
The system of equations can be solved by employing a Newton-Raphson MultiGrid
method. By employing the requirement that Properties of the tri-diagonal system
should be Persistent on any coarsened or refined grid, we find that the
Rule of Positive Coefficients is universally valid. And the discriminant
of the equations system must be positive:
$$
a \ge 0 \EN b \ge 0 \EN 1 - 4.a.b \ge 0
$$
In the limiting case of a immensely fine grid, the system of equations becomes
associated with a linear ODE of second order, called the Governing Equation:
$$
A \frac{d^2T}{dx^2} + B \frac{dT}{dx} + C.T = 0
$$
Where $x=$ coordinate, $A,B,C=$ constants and $T(x)=$ solution.
It is conjectured that the discriminant of the characteristic equation of
the ODE must be positive (or zero). Having investigated all special cases, this
turns out to be the only condition:
$$
B^2 - 4.A.C \ge 0
$$
Hence, oscillatory solutions can never be described by the governing ODE
of a persistent tri-diagonal system of equations.
The coefficients $a$ and $b$ can be expressed in the coefficients $A,B,C$ of
the governing ODE and the grid-spacing $dx$. The solution of the tri-diagonal
system is nothing else but a sampling on the grid of the Analytical Solution
belonging to the governing ODE.