Reference: Elements of Vibration Analysis, by L. Meirovitch, McGraw-Hill, New York, 1986.
To develop the second-order Runge-Kutta method, we assume an approxi
mation having the expression
xk =x) +e. e292 0k
where c; and ¢2 are constants and
= TH) ga = THLxtk) + 229] (12.75)
Note that x(k) + 23g merely represents the argument of the function f in the
expression{org3,in which 3 isaconstant. Theconstantsc;,¢2,anda, are determined
by insisting that Eqs. (12.72) and (12.74) agree through terms of second order in 7:
From Eqs. (12.75) we can write the Taylor series expansion
92 = THX) + 2291] = TEA) + TF)
01,2, (12.74)
- [ra +a + | 02.76)
so that, using the fist of Eqs. (12.75) and Ea, (12.76, Eqs. (12.74) become
x(k + 1) = x(k) + ex T/(K) + ar[nw + anne ne |
(12.77),
caungtrmtrouphcnndorderia inns (272)n{127 soc a
r) = stk) +(e + €2) TPH) + e272 T MH)
ate
(12.78)
so that there are two equations and three unknowns. This implies that Eqs. (12.78)do
not have a unique solution, so that one of the constants can be chosen arbitrarily,
provided the choice corresponding to 2 = is excluded. One satisfactory choice is
cy = 1/2, which yields
anak 2
12.79)
Inserting Eqs. (12.79) into Eqs. (12.74), in conjunction with Eqs. (12.75) and (12.76),
we obtain a computational algorithm defining the second-order Runge-Kutta method
in the form
atk #1) = xk) + Hoy 93) K=O1 2.0 (12.80)
where
a= Tk) 92=Tfxk) +m] k= 01,2, (1281)
land we note that the choice (12.78) leads to a symmetric form for the algorithm.
Following the same procedure, we can derive higher-order Runge-Kutta
2 ‘approximations. The derivations become increasingly complex, however, so that they
are omitted. Instead, we present simply the results. The most widely used isthe fourth-
forder Runge-Kutta method, defined by the algorithm
stk + 1) = (8) + Man +242 +299 ta) = 0,12... (12.82)where
= THK) gs = T/[x(K) +05] ga = T/Lx(K) + 0502]
Thh) +02] k= 0.1.2, (12.83)
In the vibration of single- and multi-degree-of-reedom systems, Eq, (12.66) is a
vector equation instead ofascalar equation. Adopting the vector notion, instead ofthe
‘matrix notation, the state equations for a system of order m can be written as follows:
90 = Ty0] (0284)
where y and f are m-dimensional vectors. Then, the fourth-order Runge-Kutta
method can be defined by the algorithm
Mk +1) = 908) + Has + es + 2g ae) KOT... (1285)
os
where
B= TH) ge = THy(k) + 05—,] gs = THLy(h) + 0522)
ee = THM +e) k= 01,2, (1286)
are m-dimensional vectors,
“The fourth-order Runge-Kutta method involves four evaluations ofthe vector f
for each integration step. So that it requires 2 large amount of computer time. The
methodisentremely accurate, however sothatitrequiresfewerstepsforadesiredlevel
of accuracy than other methods. This makes ita favorite in numerical integration
of nonlinear differential equations.