18.2. Consistency, Stability, and Convergence

5.2.1 Consistency

A numerical method is consistent if the local truncation error (LTE) tends to zero as the step size ℎ → 0.

• Local Truncation Error (LTE): $\tau_n = \frac{y(x_{n+1}) - y_n - h f(x_n, y_n)}{h}$

• A method is consistent if: $\lim_{h \to 0} \tau_n = 0$

In numerical methods, consistency is an important property that relates to the accuracy of the method. A numerical method is considered consistent if the local truncation error (LTE), which measures how far off a numerical approximation is from the true solution, approaches zero when the step size (h) approaches zero.

The LTE can be calculated using the formula given. When we take the limit of the LTE as the step size reduces, if this limit results in zero, it confirms that our method improves its accuracy as we make smaller steps.

Think of trying to draw a straight line. If you draw with a thick pen, your line might wobble and appear jagged, representing a larger error (or truncation error). However, if you switch to a finer pen and draw with smaller increments, your line gets smoother and more accurate—the 'truncation error' decreases as your method becomes more refined.

5.2.2 Stability

Stability refers to how the method responds to errors. A method is stable if small perturbations in initial conditions or intermediate steps do not grow exponentially.

• For linear methods, stability is often tested using the test equation: $y' = \lambda y$ with $\lambda \in \mathbb{C}$.

• The numerical solution becomes: $y_{n+1} = R(h\lambda) y_n$ where $R(h\lambda)$ is the stability function.

• Absolute Stability: The method is absolutely stable if: $|R(h\lambda)| \leq 1$

• Stability Region: The set of all $h\lambda$ for which the inequality holds.

Example: Euler’s Method For Euler’s method applied to $y' = \lambda y$, $y_{n+1} = y_n + h\lambda y_n = (1 + h\lambda) y_n \Rightarrow R(h\lambda) = 1 + h\lambda$ Stability region: $|1 + h\lambda| \leq 1$

Stability in numerical methods is critical as it determines whether errors, due to initial conditions or during calculations, will remain manageable or spiral out of control. A method is defined as stable if even small errors can’t cause the results to grow uncontrollably. This concept can be illustrated using a test equation in linear methods, where we can analyze the stability function (R(hλ)) to see how perturbations affect our calculations.

For example, in Euler's method, if you input a small error in the initial condition, we check the stability condition. If the computed values remain bounded (|R(hλ)| ≤ 1), the method is stable. In contrast, if it exceeds 1, even slight changes result in significant divergence from the solution, showcasing instability.

Imagine you’re balancing a stick on your finger. If it starts to tilt slightly, you have to make small and calculated adjustments to keep it upright. If you overcompensate, the stick can easily fall down, similar to how errors can exacerbate if the numerical method is unstable. However, if you're precise and respond correctly to slight tilts, you can keep the stick balanced—this is like having a stable numerical method.

5.2.3 Convergence

A numerical method is convergent if the solution obtained from it tends to the exact solution as the number of steps increases (i.e., ℎ → 0).

Lax Equivalence Theorem: For a consistent numerical method for a well-posed problem, stability is necessary and sufficient for convergence.

So:

• Consistency + Stability ⇒ Convergence

Convergence ensures that as we improve our method by reducing the step size, the results increasingly resemble the true solution of the differential equation we're trying to solve. The concept of convergence links closely to the Lax Equivalence Theorem, which asserts that for any consistent numerical method applied to a well-defined problem, if the method is stable, it will naturally converge to the exact solution. This means that stability and consistency are both prerequisites that lead to accurate and reliable results in numerical analysis.

Think of it like tuning a guitar. As you make small adjustments to the tuning pegs to change the pitch of the strings, you get closer to the perfect tone. If your tuning method is consistent (you follow the same approach each time) and stable (small turns don’t lead to drastic wrong pitches), eventually, you’ll converge on the exact note. Similarly, a numerical algorithm should ensure both stability and consistency to hone in on the accurate solution.

--