In many problems, boundary conditions are specified in terms of the derivative of the function rather than the function itself. For example, consider the following boundary conditions:
This corresponds to the scenario discussed in “Applying Finite Difference Method to Bounday-Value Problems 1” where and . Additionaly, the constants are: and .
An alternative approach introduces grid points and outside the interval . As a result, the discretized differential equation at the boundaries becomes
and for the interior points:
To approximate the boundary condition more accurately, we use
which gives
Substituting into (4-1) yields
which is
Similarly, by eliminating from
we obtain
Using (4-4), (4-5) and (4-3), we arrive at the following system of equations for the unknowns :
The coefficient matrix of this system is
According to “A Page in My Notebook [5]”, if for , the matrix is nonsingular. However, if all are zero, the matrix becomes singular. This singularity signifies the nonuniqueness of solutions to the boundary-value problem. Specifically, when and is a solution, any shift , for any constant , is also a solution.
Another type of boundary condition that leads to nonunique solutions is when
In this case, and are both unknown, but only one of them needs to be included in the system of equations. Additionaly, if the solution is extended periodically outside of , we can assume that , leading to the equation:
Similarly,
By combining (4-6) and (4-7) with the equation for the interior:
we form a system of linear equations with unknowns . The corresponding coefficient matrix is
The matrix remains nonsingular if all are positive, but becomes singular if all are zero. This singularity, once again, signifies the nonuniqueness of the solution.
Exercise-1 Explain “To approximate the boundary condition more accurately, we use ”
Exercise-2 Show that “when and is a solution, any shift , for any constant , is also a solution.”