29.1. Foundations#

29.1.1. Local truncation error and Consistency#

Let

\[y_{n+1} = \Psi(t_{n+1}; y_n, ..., y_{n-k+1}; h, f(\cdot)) = y_n + h A(t_{n+1}; y_n, ..., y_{n-k+1}; h, f(\cdot)) \ ,\]

the numerical method for the solution of the initial value problem

\[\begin{split}\begin{aligned} \dot{y}(t) & = f(y(t),t) \\ y(0) & = y_0 \ . \end{aligned}\end{split}\]

The local truncation error is the error produced by one step of the method, i.e.

\[\delta_{n+k}^h := y_{n+k} - y(t_{n+k}) \ ,\]

being \(y_{n+k}\) the result of one step of the numerical method, and \(y(t_{n+k})\) the exact solution using the same initial conditions (assuming, there’s no error in previous steps). A numerical method is defined consistent if

\[\lim_{h \rightarrow 0} \frac{\delta^h_{n+k}}{h} = 0 \ .\]

The method has order \(p\) if

\[\delta^h_{n+k} = O(h^{p+1}) \quad , \quad \text{for } h \rightarrow 0 \ .\]

29.1.2. Global truncation error and Convergence#

The global truncation error \(e_n\) is the accumulation of the error up to time \(t_n\),

\[e_n^h := y_{n} - y(t_n) \ .\]

A numerical method is defined convergent if

\[\lim_{h \rightarrow 0} \max_n | e_{n} | = 0 \ .\]

29.1.2.1. Recurrence relation for One-step methods#

A one-step method reads

\[y_{n+1} = \Psi(t_{n+1}; y_n; h, f(\cdot)) = y_n + h A(t_{n+1}; y_n; h, f(\cdot)) \ ,\]

A recurrence relation exists between the global and the local truncation error

this implies some bound of the global error, see wiki