BIT 11 (197I), 884-388
A-STABLE
RUNGE-KUTTA
PROCESSES
F. H. CHIPMAN Abstract.
The concept of strong A-stability is ...
76 downloads
548 Views
174KB Size
Report
This content was uploaded by our users and we assume good faith they have the permission to share this book. If you own the copyright to this book and it is wrongfully on our website, we offer a simple DMCA procedure to remove your content from our site. Start by pressing the button below!
Report copyright / DMCA form
BIT 11 (197I), 884-388
A-STABLE
RUNGE-KUTTA
PROCESSES
F. H. CHIPMAN Abstract.
The concept of strong A-stability is defined. A class of strongly A-stable RungeKutta processes is introduced. It is also noted that several classes of implicit Runge-Kutta processes defined by Ehle [6] are A-stable. I. Introduction.
T h e v-stage R u n g e - K u t t a process for numerically solving the n-dimensional s y s t e m y' = f ( x , y ) , y(O)=Yo, is given b y (1.1)
Ym+I = Ym+ h W K ,
m = 1,2 . . . . .
where K =
with K i = f
xm+c th,ym+h~bi~K j~l
t , ]
and W = ( w t I . . . . . wv/), with I the n × n i d e n t i t y matrix. T o s t u d y t h e stability of process (1.1), we a p p l y it t o the scalar equation y ' = qy, R e (q)< 0. The resulting difference e q u a t i o n is
(1.2)
Ym+i = Y,~ + h W K
where (1.3)
DEYIiVITION 1.1. Process (1.1) is well defined if (1.3) has a unique solution for all q w i t h R e (q) < 0. Hence, for a well defined process, (1.2) m a y be w r i t t e n
(1.4)
Ym+l = (1 + q h W ( I - q h B ) - t e ) y m = E(qh)y~,
where e = ( 1 . . . 1) ~" a n d B = (b~s). Received July 28, 1971.
A-STABLE RUNGE-KUTTA PROCESSES
385
DEFINITION 1.2. Process (1.1) is A-stable, in the sense of Dahlquist [5], if IE(z)l < 1 for all complex z with negative real part. DEFINITION 1.3. Process (1.1) is strongly A-stable if it is A-stable and l i m ~ (~)_~_~E (z ) = O. If a process is not well defined, then E ( z ) is not defined at v or fewer points in the left half complex plane. Rewriting E ( z ) as a rational function det(I-zB) + z W adj ( I - z B ) e P(z) E(z)
=
=
det(I-zB)
- -
Q(z)'
we see t h a t an A-stable process is well defined if P ( z ) and Q(z) have no common zeros in the left half plane. In particular, any A-stable process with a Pad~ approximation E ( z ) to exp(z), is well defined, and thus all processes considered in this paper are well defined. 2. C l a s s e s of A - S t a b l e P r o c e s s e s .
Butcher's [2] v-stage processes of order 2v, based on Gaussian quadrature formulae, have been shown to be A-stable by several authors (see [6] and [7]). Although Butcher's methods based on Radau and Lobatto quadrature formulae [3] are not A-stable, Ehle [6] has constructed similar methods which he conjectured to be A-stable. The A-stability of these processes is established in [4]. Table 2.1 summarizes some known classes of A-stable processes. The w i and ct are weights and abscissae of the appropriate quadrature formula, and B = (hi1), V ~ - (CiJ-1), C = (ciJ/j), N = (1/i) and D = d i a g ( w f ) are all square matrices of dimension v. Table 2.1. A - S t a b l e R u n g e - K u t t a
Processes.
Class
Quadrature formula
B defined by
G (Butcher) IA (Ehle) IIA (Ehle) IIIA (Ehle) IIIB (Ehle) IIIc (Chipman)
Gaussia~ Radau, c1= 0 Radau, cv = 1 Lobatto Lobatto Lobatto
B = C V -1 B = D -1 ( V ~) -1 ( N - C)TD B = C V -1 B = C V -1 B = D - I ( V ~ ) -1 (N - C)TD
Equation (2.1)
Order ~V
2v--1 2v - 1 2v - 2 2v - 2 2v--2
The collocation methods of Wright [7] and Axelsson [1] based on Gaussian, Radau and Lobatto quadrature formulae are equivalent to class G, IIA and I I I B processes respectively.
386
F.H. CHIPMAN
I n addition to Ehle's class IIIA and I I I B processes, a third class can be defined, based on L o b a t t o quadrature formulae. This class, referred to as class I I I c, is defined in the following w a y : 1. Choose wt and ct as the weights and abscissae of a v-point L o b a t t o quadrature formula. 2. Choose bil=wl, i= 1,2 . . . . v. 3. Choose the remaining btj b y
(bl~ ... blv (2.1)
by2
! C12 C1-- W1 "~ •.,
c?-I I 1 •
=
Cv2 Cv Wl -~
bvv)
C2 ... .
c2v--2\
}
1 cv
-1
Cvv-2]
v-11
I n [4] it is established t h a t these processes are of order 2 v - 2, and are strongly A-stable since E(qh) in equation (1.2) becomes a second subdiagonal Pad6 approximation to exp (qh). Similarly it is noted t h a t processes in classes I~ and I I ~ are also strongly A-stable. Coefficients for several processes in class I I I c are given in the appendix. I t is the author's experience t h a t these methods can be used to solve stiff systems of ordinary differential equations at least as efficiently as methods now in use. A future paper will deal with the implementation of A-stable R u n g e - K u t t a processes.
Appendix.
Coefficients for class I I I c processes, v < 5.
v=2
1
1
2
2
1
1
~
~
1
1
0 1
v=3
1
1
1
6
3
g
1
5
1
12
12
1
2
1
1
2
1
0 1
A-STABLE RUNGE-KUTTA PROCESSES
v=4
12
12
1
1
1
12
12
lo-7
1--2
v=5
V~
387
5 - Vg
60
10
60
1
1 0 + 7]/5
12
60
1
5
5
1
12
12
12
12
5 + V5 6o
1
5
5
1
12
12
12
12
10
1
7
2
7
1
2--0
60
1-5
60
2-0
0
1
29
47 -- 105r
2 9 - 30r
3
1--r
20
180
315
180
140
2
1
7 (47+105r~
73
3
1
20
180 \
16
]
360
7
(47--105r~
180 \
16
]
160
1
29 + 3 0 r
47 + 105r
29
3
l+r
20
180
315
180
140
2
1
49
16
49
1
20
180
45
180
20
1
49
16
49
1
20
180
45
180
20
REFERENCES 1. O. Axelsson, A Class of A-stable Methods, BIT 9 (1969), 185-199. 2. J. C. Butcher, Implicit Runge-Kutta Integration Processes, Math. Gomp. 18 (1964), 50-64. 3. J. C. Butcher, Integration Processes Based on Radau Quadrature Formulas, Math. Comp. 18 (1964), 233-244. 4. F. H. Chipman, Numerical Solution of Initial Value Problems using A-stable RungeKutta Processes, Research Report CSRR 2042, Dept. of AACS, University of Waterloo.
388
F . H . CHIPMAN
5. G. D a h l q u i s t , A Special Stability Problem for Linear Multistep Methods, B I T 3 (1963), 27-43. 6. B. L. E h l e , On Padd Approximation to the Exponential Function and A-stable Methods .for the Numerical Solution of Initial Value Problem~, R e s e a r c h R e p o r t C S R R 2010, D e p t . A A C S , U n i v e r s i t y of W a t e r l o o . 7. K . W r i g h t , Some Relationships Between Implicit Runge-Kutta, Collocation and Laaczos T Methods, and their Stability Properties, B I T 10 (1970), 217-227.
FACULTY OF MATHEMATICS UNIVERSITY OF WATERLOO WATERLOO, ONTARIO CANADA
DEPT. OF MATHEMATICS ACADIA UNIVERSITY WOLI~VILLE, NOVA SCOTIA CANADA