The su1n1narizecl approxiination inethocl^593
leading to additive schemes. The describing sche1nes of this type will be
studied in more detail in Section 10, but a great deal of work still needs
to be done in adopting those ideas. It is worth mentioning here their main
pee uliari tie;o;:
- the passage frmn the jth layer tot.he (j+ 1 )th layer can be perforn1ed
through the use of a sequence of the usual (two-point, three-point,
etc.) schemes; - the error of approximation provided by an acldi ti ve scheme is ta.ken
to be a sun1 of the residuals of all auxiliary schemes, that is, any
such scheme generates a summarized approximation.
In this regard, we should take into account that auxiliary schemes are
not obliged to approxinrnte the original proble1n; the approximation here
is ensured by su1nrnarizing all the residuals obtained.
In Chapter 2 we ca1ne across the necessity of generalizing the notion
of approxin1ation in the real ;o;ituations when a difference scheme cannot
provide on the grid w h local approximations with a desired order in the
nonn of the space C, but it does the san1e in one of the negative nonns,
that is, in a certain sen;o;e of summarizing.
Likewise, it may happen that a sche1ne on the grid W 7 cannot provide
local approxin1ations in t, but at the final stage the approxin1ation will be
achieved once we bring together the residuals over several time layers. The
notion of su1nn1arized approximation needs certain clarification. It seems
worthwhile giving sin1ple examples.
Example 1 The Cauchy problem, being the most familiar one, comes
first:
du
-+aH=O dt ) t > 0 , ·u ( 0) = tt 0.
Common practice involves for solving it the difference scheme
(3)
yj+l/2 - yJ.
-----+ a1YJ = 0'
T
j = 0, 1, 2 .... '
yi+l - yj+l/2.
------+ Cl2y1+1/2 = 0'
T
which consists of two explicit schemes with residuals l/! 1 and ~; 2 , respectively.
Within more compact notations
yi+1/2 = z1+l/2 _ii,