A multi-step method computes w i + 1 w_{i+1} w i + 1 from several previous approximations, not just w i w_i w i .
An m m m -step method for the IVP y ′ ( t ) = f ( t , y ) y'(t) = f(t, y) y ′ ( t ) = f ( t , y ) has a difference equation
for i = m − 1 , m , … , N − 1 i = m-1, m, \ldots, N-1 i = m − 1 , m , … , N − 1 , with constants a 0 , … , a m − 1 a_0, \ldots, a_{m-1} a 0 , … , a m − 1 and b 0 , … , b m b_0, \ldots, b_m b 0 , … , b m , and h = b − a N h = \dfrac{b-a}{N} h = N b − a .
Starting values w 0 = α w_0 = \alpha w 0 = α and w 1 , … , w m − 1 w_1, \ldots, w_{m-1} w 1 , … , w m − 1 come from a one-step method such as Runge-Kutta .
Types
Explicit Method
b m = 0 b_m = 0 b m = 0 . w i + 1 w_{i+1} w i + 1 is given directly by previously determined values.
Implicit Method
b m ≠ 0 b_m \neq 0 b m = 0 . w i + 1 w_{i+1} w i + 1 appears on both sides and must be solved for.
Adams Methods
Integrating y ′ = f ( t , y ) y' = f(t, y) y ′ = f ( t , y ) over [ t i , t i + 1 ] [t_i, t_{i+1}] [ t i , t i + 1 ]
f ( t , y ( t ) ) f(t, y(t)) f ( t , y ( t )) is replaced by a polynomial through the m m m most recent points, then integrated.
Below, f k = f ( t k , w k ) f_k = f(t_k, w_k) f k = f ( t k , w k ) and μ i \mu_i μ i lies in the interval spanned by the points used. τ i + 1 ( h ) \tau_{i+1}(h) τ i + 1 ( h ) is the local truncation error per unit step. Substitute the exact values y ( t k ) y(t_k) y ( t k ) into the method and divide the remaining residual by h h h . It measures the error introduced at step i + 1 i+1 i + 1 when all preceding values are exact.
Adams-Bashforth Explicit Methods
2-step, with τ i + 1 ( h ) = 5 12 y ′ ′ ′ ( μ i ) h 2 \tau_{i+1}(h) = \dfrac{5}{12} y'''(\mu_i) h^2 τ i + 1 ( h ) = 12 5 y ′′′ ( μ i ) h 2
3-step, with τ i + 1 ( h ) = 3 8 y ( 4 ) ( μ i ) h 3 \tau_{i+1}(h) = \dfrac{3}{8} y^{(4)}(\mu_i) h^3 τ i + 1 ( h ) = 8 3 y ( 4 ) ( μ i ) h 3
4-step, with τ i + 1 ( h ) = 251 720 y ( 5 ) ( μ i ) h 4 \tau_{i+1}(h) = \dfrac{251}{720} y^{(5)}(\mu_i) h^4 τ i + 1 ( h ) = 720 251 y ( 5 ) ( μ i ) h 4
5-step, with τ i + 1 ( h ) = 95 288 y ( 6 ) ( μ i ) h 5 \tau_{i+1}(h) = \dfrac{95}{288} y^{(6)}(\mu_i) h^5 τ i + 1 ( h ) = 288 95 y ( 6 ) ( μ i ) h 5
Adams-Moulton Implicit Methods
2-step, with τ i + 1 ( h ) = − 1 24 y ( 4 ) ( μ i ) h 3 \tau_{i+1}(h) = -\dfrac{1}{24} y^{(4)}(\mu_i) h^3 τ i + 1 ( h ) = − 24 1 y ( 4 ) ( μ i ) h 3
3-step, with τ i + 1 ( h ) = − 19 720 y ( 5 ) ( μ i ) h 4 \tau_{i+1}(h) = -\dfrac{19}{720} y^{(5)}(\mu_i) h^4 τ i + 1 ( h ) = − 720 19 y ( 5 ) ( μ i ) h 4
4-step, with τ i + 1 ( h ) = − 3 160 y ( 6 ) ( μ i ) h 5 \tau_{i+1}(h) = -\dfrac{3}{160} y^{(6)}(\mu_i) h^5 τ i + 1 ( h ) = − 160 3 y ( 6 ) ( μ i ) h 5
The m m m -step Adams-Bashforth and ( m − 1 ) (m-1) ( m − 1 ) -step Adams-Moulton methods both have error O ( h m ) O(h^m) O ( h m ) . The implicit one is more accurate.
Predictor-Corrector Method
Combine an explicit and an implicit method, since implicit ones rarely solve for w i + 1 w_{i+1} w i + 1 in closed form.
Predict
Compute w i + 1 ( p ) w_{i+1}^{(p)} w i + 1 ( p ) with an explicit method.
Correct
Substitute w i + 1 ( p ) w_{i+1}^{(p)} w i + 1 ( p ) into the right side of an implicit method to get w i + 1 w_{i+1} w i + 1 .
Standard choice: Runge-Kutta for starting values, Adams-Bashforth 4-step as predictor, Adams-Moulton 3-step as corrector.
Worked Examples
Example 1
Apply Adams-Bashforth 2-step to y ′ ( t ) = y − t 2 + 1 y'(t) = y - t^2 + 1 y ′ ( t ) = y − t 2 + 1 , 0 ≤ t ≤ 2 0 \leq t \leq 2 0 ≤ t ≤ 2 , y ( 0 ) = 0.5 y(0) = 0.5 y ( 0 ) = 0.5 , with h = 0.2 h = 0.2 h = 0.2 . Exact solution y ( t ) = ( t + 1 ) 2 − 0.5 e t y(t) = (t+1)^2 - 0.5 e^t y ( t ) = ( t + 1 ) 2 − 0.5 e t .
Solution Reveal Hide w 0 = 0.5 w_0 = 0.5 w 0 = 0.5 is given. Get w 1 w_1 w 1 with 1 step of order 4 Runge-Kutta .
k 1 = 0.2 f ( 0 , 0.5 ) = 0.3 k 2 = 0.2 f ( 0.1 , 0.5 + 0.15 ) = 0.328 k 3 = 0.2 f ( 0.1 , 0.5 + 0.164 ) = 0.3308 k 4 = 0.2 f ( 0.2 , 0.5 + 0.3308 ) = 0.35816 w 1 = 0.5 + 1 6 ( k 1 + 2 k 2 + 2 k 3 + k 4 ) = 0.829293 \begin{aligned}
k_1 &= 0.2\, f(0,\ 0.5) = 0.3 \\
k_2 &= 0.2\, f(0.1,\ 0.5 + 0.15) = 0.328 \\
k_3 &= 0.2\, f(0.1,\ 0.5 + 0.164) = 0.3308 \\
k_4 &= 0.2\, f(0.2,\ 0.5 + 0.3308) = 0.35816 \\
w_1 &= 0.5 + \tfrac{1}{6}\left( k_1 + 2k_2 + 2k_3 + k_4 \right) = 0.829293
\end{aligned} k 1 k 2 k 3 k 4 w 1 = 0.2 f ( 0 , 0.5 ) = 0.3 = 0.2 f ( 0.1 , 0.5 + 0.15 ) = 0.328 = 0.2 f ( 0.1 , 0.5 + 0.164 ) = 0.3308 = 0.2 f ( 0.2 , 0.5 + 0.3308 ) = 0.35816 = 0.5 + 6 1 ( k 1 + 2 k 2 + 2 k 3 + k 4 ) = 0.829293 Step to t 2 = 0.4 t_2 = 0.4 t 2 = 0.4 by Adams-Bashforth
f 0 = w 0 − 0.0 2 + 1 = 1.5 f 1 = w 1 − 0.2 2 + 1 = 1.789293 w 2 = w 1 + 0.2 2 [ 3 f 1 − f 0 ] = 0.829293 + 0.1 ( 3.867879 ) = 1.216081 \begin{aligned}
f_0 &= w_0 - 0.0^2 + 1 = 1.5 \\
f_1 &= w_1 - 0.2^2 + 1 = 1.789293 \\
w_2 &= w_1 + \frac{0.2}{2}\left[ 3 f_1 - f_0 \right] = 0.829293 + 0.1\,(3.867879) = 1.216081
\end{aligned} f 0 f 1 w 2 = w 0 − 0. 0 2 + 1 = 1.5 = w 1 − 0. 2 2 + 1 = 1.789293 = w 1 + 2 0.2 [ 3 f 1 − f 0 ] = 0.829293 + 0.1 ( 3.867879 ) = 1.216081 Exact y ( 0.4 ) ≈ 1.214088 y(0.4) \approx 1.214088 y ( 0.4 ) ≈ 1.214088 .
Example 2
Use the Adams-Bashforth 2-step predictor and the Adams-Moulton 2-step corrector to find w 2 w_2 w 2 for y ′ = y y' = y y ′ = y , y ( 0 ) = 1 y(0) = 1 y ( 0 ) = 1 , h = 0.1 h = 0.1 h = 0.1 , with w 1 = 1.105171 w_1 = 1.105171 w 1 = 1.105171 .
Solution Reveal Hide f 0 = 1 , f 1 = 1.105171 f_0 = 1, \quad f_1 = 1.105171 f 0 = 1 , f 1 = 1.105171 Predict
w 2 ( p ) = w 1 + 0.1 2 [ 3 ( 1.105171 ) − 1 ] = 1.220947 w_2^{(p)} = w_1 + \frac{0.1}{2}\left[3(1.105171) - 1\right] = 1.220947 w 2 ( p ) = w 1 + 2 0.1 [ 3 ( 1.105171 ) − 1 ] = 1.220947 Correct
w 2 = w 1 + 0.1 12 [ 5 ( 1.220947 ) + 8 ( 1.105171 ) − 1 ] = 1.221389 w_2 = w_1 + \frac{0.1}{12}\left[5(1.220947) + 8(1.105171) - 1\right] = 1.221389 w 2 = w 1 + 12 0.1 [ 5 ( 1.220947 ) + 8 ( 1.105171 ) − 1 ] = 1.221389 The exact value is e 0.2 = 1.221403 e^{0.2} = 1.221403 e 0.2 = 1.221403 . The corrector reduces the error from 4.6 × 10 − 4 4.6 \times 10^{-4} 4.6 × 1 0 − 4 to 1.4 × 10 − 5 1.4 \times 10^{-5} 1.4 × 1 0 − 5 .