- 积分
- 73
- 注册时间
- 2003-1-6
- 仿真币
-
- 最后登录
- 1970-1-1
|
发表于 2012-6-27 22:58:58
|
显示全部楼层
来自 美国
That's why we have the so-called a-posetriori error estimation which is still a popular subject in the academic circle.
Let assume that the equation we solve is the right one which reflects the true reality and the unknown exact solution is uex. We are going to use the finite element solution uh to approximate uex.
The quantity we are interested is some kind of functional in terms of F(uex - uh) which is unknown. If we play with mathematics, we can get the following:
lower bound <= F(uex - uh) <=upper bound
If we can get the lower and upper bounds, we know how far we are away from the true error. Currently, people we can do get the bounds for certain problems such as linear elliptic problem or parabolic problem. However, there is no commercial software that can do this type of error estimation.
Another way to do the error estimation is based on superconvergence invented by Zienkiewicz. By postprocessing the finite element solution, we can have a solution which is closer to the unknown exact solution than the original finite element solution. As far as I know, most commercial FE softwares have this capability. But we cannot get bounds in this case. |
评分
-
1
查看全部评分
-
|