28.01.2013 Views

Zienkiewicz O.C., Taylor R.L. Vol. 3. The finite - tiera.ru

Zienkiewicz O.C., Taylor R.L. Vol. 3. The finite - tiera.ru

Zienkiewicz O.C., Taylor R.L. Vol. 3. The finite - tiera.ru

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

As we proved earlier, the Galerkin spatial approximation is justi®ed when the<br />

characteristic±Galerkin procedure is used. We can thus write the approximation<br />

ˆ N ~ f …2:92†<br />

and use the weighting N T in the integrated residual expression. Thus we obtain<br />

M… ~ f n ‡ 1 ÿ ~ f n †ˆÿ t‰…C ~ f n ‡ K ~ f n ‡ f n †ÿ t…K u ~ f n ‡ f n s †Š …2:93†<br />

in explicit form without higher-order derivatives and source terms. In the above<br />

equation<br />

…<br />

M ˆ N T …<br />

N d C ˆ N T @<br />

…U<br />

@x<br />

iN† d<br />

i<br />

… T<br />

@N<br />

K ˆ k<br />

@xi @N<br />

…<br />

d f ˆ N<br />

@xi T …2:94†<br />

Q d ‡ b:t:<br />

and Ku and f n s come from the new term introduced by the discretization along the<br />

characteristics. After integration by parts, the expression of Ku and fs is<br />

Ku ˆÿ 1<br />

…<br />

2<br />

fs ˆÿ 1<br />

…<br />

2<br />

@<br />

@x i<br />

@<br />

@x i<br />

…UiN T † @<br />

…U<br />

@x<br />

iN† d …2:95†<br />

i<br />

…U iN T †Q d ‡ b:t: …2:96†<br />

where b.t. stands for integrals along region boundaries. Note that the higher-order<br />

derivatives are not included in the above equation.<br />

<strong>The</strong> approximation is valid for any scalar convected quantity even if that is the<br />

velocity component Ui itself, as is the case with momentum-conservation equations.<br />

For this reason we have elaborated above the full details of the spatial approximation<br />

as the matrices will be repeatedly used.<br />

It is of interest that the explicit form of Eq. (2.93) is only conditionally stable. For<br />

one-dimensional problems, the stability condition is given as (neglecting the e€ect of<br />

sources)<br />

t 4 tcrit ˆ h<br />

…2:97†<br />

jUj<br />

for linear elements.<br />

In two-dimensional problems the criteria time step may be computed as 62;63<br />

t crit ˆ<br />

t t<br />

t ‡ t<br />

Characteristic-based methods 41<br />

…2:98†<br />

where t is given by Eq. (2.97) and t ˆ h 2 =2k is the di€usive limit for the critical<br />

one-dimensional time step.<br />

Further, with t ˆ t crit the steady-state solution results in an (almost) identical<br />

di€usion change to that obtained by using the optimal streamline upwinding<br />

procedures discussed in Part I of this chapter. Thus if steady-state solutions are the<br />

main objective of the computation such a value of t should be used in connection<br />

with the K u term.

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!