29.2. Families of integration methods#
Methods for the first-order IVP
29.2.1. One-step methods (I): Euler, Crank-Nicolson#
29.2.2. Multi-step methods#
Adams-Bashforth and Adams-Moulton. The integration of the first-order equation (29.1) between \(t_n\) and \(t_{n+1}\) reads
AB and AM methods replace the function \(\mathbf{f}(t, \mathbf{y}(t))\) with a polynomial Lagrangian approximation \(\mathbf{p}_s(t) \simeq \mathbf{f}(t)\) built with values of \(\mathbf{f}(t_{n-k}, \mathbf{y}(t_{n-k}) )\), with \(k_{AB} = 0:s-1\) (AB are explicit methods) and \(k_{AM} = -1:s-1\) (AM are implicit methods).
Once the original function is replaced by its polynomial approximation, the integration can be performed exactly.
Backward Differentiation Formulas. BDF methods exploits the derivative of the polynomial approximation \(\mathbf{q}_s(t) \simeq \mathbf{y}(t)\). The value of the derivative \(\dot{\mathbf{q}}_s(t)\) evaluated in \((t_{n+1}, \mathbf{y}_{n+1})\) can be written as a linear combinations of the \(\{ \mathbf{y}_{n-k} \}_{k=-1:s-1}\) values. BDF immediately follows from the comparison
29.2.2.1. Adams-Bashforth (AB)#
For the derivation of the low-order AB methods, see Lagrange interpolation: Applications: Adams-Bashforth
29.2.2.2. Adams-Moulton (AM)#
For the derivation of the low-order AB methods, see Lagrange interpolation: Applications: Adams-Moulton
29.2.2.3. Backward Differentiation Formulas (BDF)#
For the derivation of the low-order AB methods, see Lagrange interpolation: Applications: BDF
29.2.3. One-step methods (II): Runge-Kutta (RK)#
Method.
with
Rationale
Starting from a \(n\)-order Taylor series
and matching it with a linear combination of evaluations of the first-order derivative,
with \(t_{n+1} = t_n + h\).
Butcher tableau
The coefficients of a RK method can be organized in Butcher tableau
Constraints
29.2.3.1. Example: RK2#
RK2 methods include mid-point, Heun and Ralston methods.
Derivation.
Starting from a \(2\)-order Taylor series, and matching it with a linear combination of evaluations of the first-order derivative.
with \(y'(t) = f(t, y(t))\), so that \(y''(t) = f_t(t,y(t)) + y'(t) f_y(t,y(t))\).
RK approximation.
with
Expanding in Taylor series \(k_2\),
inserting in the RK2 scheme
and matching coefficients of Taylor expansions, the coefficients of the RK method are computed as the solution of a underdetermined system of 3 equations with 4 unknowns
or leaving \(\alpha = c_2 = a_{21}\) as a free parameter,
This system has infinite solutions. Different methods correspond to different choices among these solutions:
mid-point method, \(\alpha = \frac{1}{2}\)
Heun method, \(\alpha = 1\)
Ralston’s method (minimum truncation error todo prove it!), \(\alpha = \frac{2}{3}\)