[原创]我做的并行自适应MHD模式，两维vortex问题动画

zzuhuyan

[这个贴子最后由zzuhuyan在 2007/12/25 05:36pm 第 1 次编辑]

[watermark]大家讨论一下[/watermark]

guowuhui

不错，

merryfish

good job. It seems you use fully threaded tree (FTT) method to do the mesh refinement. Have you incorporated the load balancing algorithm into your code?

zzuhuyan

目前load balance通过调用paramesh内部模块来实现，我还没有详细研究这个问题，paramesh通过构造一个Peano-Hilbert 空间曲线把所有的block串联起来，然后把这条曲线分成n份，来实现load balance

merryfish

paramesh uses block structured adaptive mesh, and space filling curves to distribute the bolcks. it is easy to redistribute the blocks between different processes, but this kind of algorithm takes long time distribute the blocks.

merryfish

you can reference the FLASH code which using paramesh to do the block structured adaptive mesh refinement and coarsement.

merryfish

merryfish

merryfish

zzuhuyan

请问你上面的两个benchmark MHD problem，是用什么实现的？是否考虑磁场散度保持为零条件，如何在粗细网格差值过程中，保持此条件，另外判别加密放粗条件的公式是什么，多谢指教！

merryfish

The divergence free condition should be satisfied at all level grids, there are a lot of choice to control the divergence free of magnetic field. AMR criterion
should be chosen carefully, also you have a lot of choice depending on your problem.

zzuhuyan

这些好象都是实话，能给些指导意见么？
ps：你给的图好象是flash的结果

merryfish

please give more details about your code. i.e. which kind of method you use to control the divergence free of magnetic field, and amr criterion et al.
Then I will give you my ideas, actually if you work on the space weather forecast, the cartesian + block AMR are not the good choice. in the future you will have the difficulties in incorporating the CME initialization alogrithms, and so on.

zzuhuyan

我用的算法是两阶godunov类方法＋minimod limiter，控制磁场散度使用的是Constrained Transprot(CT)  方法，网格加密时候磁场差值根据 "li and li JCP 2004" 得到，目前可以保持散度在10^-12的量级！目前的后项误差估计公式是计算整个区域内的磁场旋度，速度散度和旋度的算术平均值，如果某个网格上的这些量大于平均值，它要加密，如果小于平均值的10分之一，那么它要放粗！现在看来这样选择的后项误差估计公式不能很好的跟踪激波和间断的位置，能否给一些建议和意见，或者参考文献，多谢！[br][br][以下内容由 zzuhuyan 在 2008年01月03日 08:42am 时添加] [br]
另外你说”actually if you work on the space weather forecast, the cartesian + block AMR are not the good choice“，请问你是如何考虑这个问题的，用球坐标下AMR或者非结构网格AMR？

merryfish

Yes, you can try other criterion such the density gradient, vortex, et al. Li&Li';s method is good enough to keep divergence free of magnetic field, we use this kind of method also. CT is easy to code, and it is a good method to keep divergence free of magnetic field. we also incorporate the CT method in our code.
The best choice for the CME initialization model in MHD code is using the field-aligned mesh.