20.3. Stability and Convergence

• Stability: Determines whether errors will grow or decay with time. For example, in FDM for time-dependent problems, explicit methods require small time steps (CFL condition).

Stability in numerical methods refers to how errors behave as calculations progress. If errors grow over time, the solution becomes unreliable; if they decay, the solution is considered stable and more trustworthy. In the Finite Difference Method (FDM) particularly for time-dependent situations, a critical constraint must be followed: time steps must be small enough to ensure stability, often referred to as the CFL (Courant-Friedrichs-Lewy) condition. This condition prevents the propagation of numerical errors that can lead to divergences in the solution.

Think of stability like balancing on a tightrope. If you're wobbling, it becomes harder to maintain your balance (error growth). If you can keep your movements small and controlled, you stay balanced (error decay). In numerical methods, staying within the limits (CFL condition) helps keep your solution 'balanced', preventing it from going off track.

• Consistency: The discretized equations should approximate the original PDE accurately.

Consistency in the context of numerical methods means that as the grid is refined (i.e., the mesh is made finer), the discretized version of the partial differential equation (PDE) converges towards the true PDE. In simpler terms, if we were to take smaller and smaller steps in our calculations, our numerical approximation should become a more accurate representation of the actual mathematical model we are trying to solve. This helps in ensuring that the method is reliable and can produce meaningful results as the discretization improves.

Imagine you're trying to draw a circle. If you use only a few straight lines, it will look jagged and rough (not consistent). But if you use many smaller lines that increasingly approach the circle’s path, the shape becomes smoother and more accurate. Similarly, ensuring our numerical approximation approaches the actual PDE leads to more reliable results.

• Convergence: If a method is consistent and stable, then it converges to the correct solution.

Convergence is the property that determines whether the numerical solution approaches the true solution of the PDE as the mesh is refined or as computation progresses. For a numerical method to be considered effective, it must be both consistent (accurately approximating the PDE) and stable (keeping errors in check). If both conditions are met, it can be confidently said that the numerical method converges, meaning that by refining our discretization (making our grid finer), we will reach an increasingly accurate solution that approximates the actual answer to the PDE.

Think of convergence like tuning a musical instrument—like a guitar. When you pluck a string, the pitch might be slightly off at first. If you keep adjusting the strings (refining your method), eventually you reach the desired note (the correct solution). If the adjustments are stable and consistent, you will get closer to the perfect sound—you converge to the right pitch.

--