U.S.S.R. Comput.Maths.Math.Phys., Printed in Great Britain
Vo1.25,No.2,pp.101-X6,1985
0341-5553/85 $10.00+0.00 Pergomon Journals Ltd.
ANANALYSISOF THE MODIFIED EXCLUSION METHOD IN CERTAIN PROBLEMS*
V.V. EEPYAEB A quantitative estimate is given of the error of a modified exclusion method with an increased stability to rounding errors when solving symmetric grid equations generated by a second-order scalar elliptic operator.
In /l/, where the application of the modified exclusion method to solving certain grid equations was described, the basis of the algorithm's increased stability to rounding errors had a qualitative character. In this paper a quantitative estimate of the algorithm's error to a first approximation of perturbation theory when solving sysnsetric equations generated The results obtained here can henceforth by a second-order scalar elliptic operator is given. be used as the basis of the increased stability of the modified method when solving a wider class of problems, as shown by numerical experiments /2/. 1 Certain properties of second-order grid equations. Before estimating the error, we shall carry out a preliminary analysis. 1. Suppose the coefficients of the positive-definite matrix A of order n satisfy following conditions:
the
i-n
a,,>& Let us consider
O,,SO,
i,j=i,2
,...,
n,
j+i;
I&= &>O,
i=i,
2,.. .) rz.
(1.1)
I-1
the influence
of a special kind of error on the solution
of the equation
AU--F. We introduce
matrix
A’ of order n, for which aulnu’Sp.
(1.2) the following relations,hold:
a,p-0.
i,j==l.Z,...,n.
aG
(1.3) d,/d,‘
We formulate
and
the following
i=l,
2..
. , n.
statement.
If two symmetric positive-definite matrices A and A’ satisfy Theorem. (1.3), the following expression holds for the arbitrary vector U : zGUU’AU/
Proof. in the form
It is easy to prove
Ed,& i-1
(1.1)
U’A’V<‘B.
that the quadratic
U’AU-
conditions
(1.4) form for the matrix A can be represented
‘c’r,a,,(~i-,,)~. i-l I--r+‘
A similar representation is also possible for U’A’U. u'_-1U and WA’U are sums of non-negative quantities. from (1.5) and (1.3).
(1.5)
It follows from (1.5) and (1.1) that The correctness of Theorem 1 follows