1549301742-The_Theory_of_Difference_Schemes__Samarskii

(jair2018) #1
Heat conduction equation with constant coefficients 311

So, not only the order of approximation, but also the stability of
scheme (16) depend on the pararneter (}.
While investigating stability we dealt actually only with two time
layers t j, t j +i and the step T = t j +i - t j. All the tricks and turns remain
unchanged when the grid w 7 becomes non-equidistant, that is, the step
TJ+i = tj+i - tj depends on the number of the layer. In this situation the
parameter(} may depend on the number j + 1 of the layer, say(}= (JJ+^1 ,
which serves to motivate the presence of another condition (} > (Jt+^1 =
~ - h^2 /(4rj+i) in place of (23). For instance, for the scheme of accuracy
O(h^4 + rf+ 1 ) we may accept (JJ+^1 = ~ - h^2 /(12rj+ 1 ). The condition
(} > (Jt+^1 is sufficient for the stability of the scheme with weights in the
case when the grid w 7 is non-equidistant.


  1. Stability with respect to the right-hand side. As a rnatter of fact, if
    (} > 0, condition (23) given by


1 h^2
4T '

is sufficient for the stability of scheme (16) with respect to the right-hand
side as well. It is interesting to consider problem (16b) and look for its
solution in the form
N-l
(26) fJ = 2= tk x (k),
k=l

by representing the right-hand side <pin terms of the eigenfunctions {X (k)}:


N-l N-l
(27) <p= L'PkX(k), II 'Pll^2 = 2= 'P~.
k=l k=l

Substituting (26) and (27) into (16b) and recalling that A,y(k) = -,\kX(k),
we derive the series
N-1
L { Tkt (l +(}TAJ+ Ak Tk - 'Pk } x (k) = 0,
k=!

from which it follows that the expression in the curly brackets vanishes as
far as an orthogonal system of eigenfunctions is concerned:


(28)

l-(l-(J)TAk
1 + (JTAk
Free download pdf