Boundary Condition Treatment
How discretized PDEs incorporate the conditions at domain edges, where accuracy and stability are most easily lost.
Why boundaries are hard
The interior of a discretized domain is governed by the differential operator, but the edges must encode physical conditions: fixed values, prescribed fluxes, or open outflow. Boundaries are where a scheme most often loses accuracy or stability, because stencils run out of neighbors and the physics may impose constraints not naturally expressed on the grid.
The main types
- Dirichlet: the value of the solution is prescribed on the boundary
- Neumann: the normal derivative (flux) is prescribed
- Robin: a linear combination of value and derivative is prescribed
- Periodic: the domain wraps, so opposite edges are identified
Implementation techniques
Dirichlet conditions are often imposed by eliminating boundary unknowns or by overwriting boundary rows of the system matrix. Neumann and Robin conditions require approximating the derivative at the edge, commonly with a one-sided difference or, to preserve second-order accuracy, with a ghost cell: a fictitious point outside the domain whose value is set so the discrete derivative matches the prescribed condition. Ghost cells keep the interior stencil uniform, which simplifies code and preserves order.
Outflow and absorbing boundaries
For wave and advection problems on truncated domains, naive boundaries reflect outgoing waves back into the domain, corrupting the solution. Absorbing or non-reflecting boundary conditions, and perfectly matched layers, are designed to let waves leave with minimal reflection. Getting these right is essential in electromagnetic and plasma-wave simulation, where spurious reflections can masquerade as physical resonances.
Consistency with the interior scheme
A common pitfall is pairing a high-order interior scheme with a low-order boundary treatment, which degrades global accuracy. The boundary discretization should match the interior order, and for time-dependent problems it must also preserve stability, since a stable interior scheme can be destabilized by an ill-posed boundary closure. This coupling is why boundary treatment is analyzed jointly with the spatial discretization, not as an afterthought.