曝光臺 注意防騙
網曝天貓店富美金盛家居專營店坑蒙拐騙欺詐消費者
To determine the coefficients, k1 and k2 are expanded in a Taylor series about (tn, xn),
giving:
k1 = hn fn
k2 = hn
fn + bk1
¶ f
¶x
n
+ ahn
¶ f
¶t
n
+ . . .
(6.9)
Hence:
xn+1 = xn + (g1 + g2)hn fn + g2bh2
n
¶ f
¶x
n
fn + g2ah2
n
¶ f
¶t
n
+ . . . (6.10)
In these equations, fn has been used as a shorthand notation for f (xn, tn). The expansion
of xn+1 can be compared with the Taylor series for the actual local solution
zn(t):
zn(tn+1) = zn(tn) + hnz˙n(tn) + h2
n
2
¨zn(tn) + . . .
= xn + hn fn + h2
n
2
¶ f
¶x
n
fn +
¶ f
¶t
n
+ . . . (6.11)
Matching the coefficients of the powers of hn yields three equations involving the
unknown coefficients:
g1 + g2 = 1
g2b = 1
2 (6.12)
g1a = 12
It is now possible to choose one coefficient as parameter, which results in a oneparameter
family of Runge-Kutta methods. For instance, if a is chosen as a parameter,
we get:
k1 = hn f (xn, tn)
k2 = hn f (xn + ak1, tn + ahn) (6.13)
xn+1 = xn +
1 −
1
2a
k1 + 1
2a
k2
Obvious choices of a are 12
, which yields a method closely related to the rectangle
rule, and 1, yielding a method related to the trapezoid rule. The method is of second
order, because no choice of a can eliminate the h3 terms, which were neglected in the
above equations.
Higher-order Runge-Kutta methods can be derived in a similar manner. Although
these higher-order methods provide an improvement of accuracy, this comes at the
cost of additional derivative evaluations. For a given overall accuracy in a time response
calculation, there is a trade-off between many small steps with a lower-order
method, or fewer steps, but more derivative evaluations with a higher-order method.
6.1. Numerical integration methods 79
ai bij gi
gi
0 35
384
5179
57600
15
15
0 0
3
10
3
40
9
40
500
1113
7571
16695
45
44
45 −56
15
32
9
125
192
393
640
89
19372
6561 −25360
2187
64448
6561 −212
729 −2187
6784 − 92097
339200
1 9017
3168 −355
33
46732
5247
49
176 − 5103
18656
11
84
187
2100
1 35
384 0 500
1113
125
192 −2187
6784
11
84 0 1
40
Table 6.1: Coefficients of the Dormand-Prince (4,5) pair
In order to reduce the number of function evaluations, mathematicians have developed
several algorithms that combine Runge-Kutta integration with an estimation
of the error in the computed function at each time-step; the error estimate can be used
to control the step-size automatically in order to meet a specified accuracy. Over the
years, several so-called Runge-Kutta pairs have been developed. These pairs combine
two Runge-Kutta methods, usually of adjacent orders p q, which share their a and
b coefficients and differ only in the g’s.
For p = 1, . . . , 4 it is possible to obtain a pth-order method with p function evaluations,
while for p = 5 or 6, p function evaluations will produce a (p − 1)st-order
method only [17]. However, for Runge-Kutta pairs the ‘extra’ function evaluation is
not wasted, as the difference between the two solutions provides a convenient error
estimate. Such pairs are defined by:
ki = hn f (xn +
i−1
å
j=1
bij ki , tn + ai hn) (6.14)
The new value xn+1 is obtained using a weighted combination of the six k’s. The two
Runge-Kutta methods that form a pair result in two solutions for xn+1:
xn+1 = xn +
Nå
i=1
gi ki
x
n+1 = xn +
Nå
i=1
g
i ki (6.15)
where N is the number of function evaluations. For 4th/5th-order pairs, N 6. In
practice, the lower-order solution x
n+1 is not actually computed; instead the error
estimate:
dn xn+1 − x
n+1 =
Nå
i=1
(gi − g
i ) (6.16)
is determined and used for step-size control.
80 Chapter 6. Analytical tools
The coefficients ai, bij, and gi can again be found by by expanding the k’s in Taylor
中國航空網 www.k6050.com
航空翻譯 www.aviation.cn
本文鏈接地址：
FDC 1.4 – A SIMULINK Toolbox for Flight Dynamics and Contro(41)
