78 Basic Concepts of the Theory of Difference Schemes
corresponding to different values of the parameter his of great interest and
refers to a difference scheme.
The main goal of any approximate method is to solve an original
(continuous) problem with a prescribed accuracy E > 0 in a finite number
of operations. In order to clarify whether it is possible in principle to
approximate a solution u of problem (35)-(36) by a solution yh of problem
(37) with any prescribed accuracy E > 0 depending on the step h(E), we
follow established practice. This is concerned with further cornparison of
Yh with u(x) in the space of grid functions }fh. Let uh be a value of the
function u(x) on the grid wh, so that uh E Hh. The error zh = Yh - uh of
the difference scheme (37) needs more a detailed exploration for a complete
and rigorous treatment.
The condition for zh can be derived upon substituting Yh = zh + ·uh
into (37). Through such an analysis the problem of the same type arises as
problem (37):
(38)
where 1/Jh and vh are residuals equal to 1/Jh ='Ph -Lh uh and vh = xh -lh uh.
The right-hand sides 1/Jh and vh of problem (38) are called the error of
approximation of equation (35) by the difference equation (37) and the error
of approximation of condition (36) by the difference condition lh Yh = xh on
a solution of problem (35)-(36) or, briefly, 1/Jh is the error of approximation
for the equation Lh Yh = 'Ph on a solution u(x) to equation (35) and vh
is the error of approximation for the condition lh Yh = xh on a solution of
problern (35)-(36).
The accurate account of the error zh of a scheme as well as of the
errors of approximations 1/Jh and vh is carried out in suitable norms II· ll(h),
II · ll(h) and II · llc 3 h) on the space of grid functions.
We say that a solution of the difference problem (37) converges to a
solution of problem (35)-(36) (scheme (37) is said to be convergent) if
or
II zh ll(h) =II P(I h I) II, where P(I h I) --+ 0 as I h I--+ 0.
The difference scheme (37) is said to be convergent with the rate 0(1 h In)
or is of the nth accuracy order (of accuracy 0( I h In)) if for all sufficiently
small I h I < h 0 the inequality