|
发表于 2011-11-18 22:14:15
|
显示全部楼层
回复 24# 通流 的帖子
CFD scheme has to be least 2nd order-accurate otherwise the Reynolds number is affected by numerical dissipations. Thus, if you use a 1st-order scheme, you don't even know what the Re is. Not good.
Why high-order schemes? The reason is simple. Suppose h = 1/N is grid spacing, then for a scheme of n-th order convergence, the error is of the order O(h^n) by definition. For 3D simulations, The total number of grid points is N^3, the roughly the total error is bounded by
N^3 * h^n = N^(3-n).
If n < 3, then as N goes to infinity, the total error is going to grow. However, if n > 3, the total error will decrease as N increases. n = 3 is the boarder line case, and that is why only n > 2 is qualified to be a high-order scheme. |
|