|
发表于 2012-9-11 10:23:08
|
显示全部楼层
回复 62# wdlxmzd 的帖子
Before I address your queries, allow me the review how we get here.
The discussion started with the issue of the CFL number in the context of the LBE, of which you made some authoritative, yet most preposterous, comments. I decided to offer my opinion. So, now you are nitpicking the "errors" and "mistakes" in my previous work. OK, let's see what mistakes I have made.
1) The MRT, MRT1, MRT2, etc. etc. (1.1 vs. 1.5)
To demonstrate the effects of ALL the relaxation rates, we carefully choose a set of values of the relaxation rates.
a) MRT: we use the "optimal" values of relaxation rates except for one -- s_q, which is determined by the so-called "magic combination" of Ginzburg [cf. Eq. (11) in the paper, Luo et al., Phys. Rev. E 83(5):056710 (May 2011)] in order to enforce accurate boundary conditions. Because of the relationship s_q(s_\nu), when s_\nu (= 1/tau) approaching 2 or 0, s_q will go to 0 and 2, respectively.
We show that when s_\nu > 1.96 or so, 1/s_q becomes so large that both q_x and q_y become "quasi-conservative" -- they are so sluggish that they interfere the hydrodynamic modes (density and momenta) severely. In LBE, their direct effect is to degrade the numerical stability.
b) MRT1: To prove the above point, we choose MRT1 with s_q fixed at 1.9. Of course, the stability improves when s_\nu > 1.96, but deteriorated when s_\nu < 1.96 (when compare to MRT case).
c) MRT2 and MRT3: to show the effect due to both s_e and s_\epsilon, we choose one of them fixed at 1.8, and the other fixed at the "optimal" value. This shows which relaxation rate may have a more severe effect on the stability.
d) MRT4: This is closest to the TRT model -- only s_\epsilon is different. It shows that the stability of MRT4 is rather similar to TRT, and it is only slightly better than TRT. Another reason to chose this case is that s_e affects the vorticity field more severely.
If we only care about the stability and nothing else, we will use the set of "optimal" values of relaxation rates.
Thus, we do have theory: kinetic theory (mode-mode coupling etc.) and linear analysis. The cases we chose to show is only to demonstrate how the things work.
We also conclude that the TRT model is perhaps the best if one considers together accuracy, stability, and computational efficiency, and in that order.
If this is not theory, then I can call it COMMON SENSE, especially after all the analysis has been done and published long ago. Of course, this will not convince Ms./Mr. WDLXMZD -- who thinks we have no theory thus can only chose some parameters at random (1.5 vs. 1.1?). I only hope that the by-standers of this discussion can appreciate and understand the rationale behind our tests -- we did the analysis first then conducted the tests, NOT the other way around. That is the
reason why we can chose only 5 sets of "typical" value to represent a 4-dimensional parameter space, which is huge.
2) Eq. (8e) in the paper: q_{x,y}^(1) = 0.
In the formula provided by WDLXMZD, q_{x,y}^(1) = O(u^2). To me, this is negligible -- I only kept the first order terms O(u) and neglect all the asymptotically smaller terms. My mistake, if it is a mistake, is my negligence to explain or qualify this -- it is not a SIMPLE mistake, as someone may conjecture.
Since WDLXMZD has paid such a keen attention to my work, which is an honor indeed, please explain to our audience what difference it makes. Does it affect the results in the paper in anyway? If yes, how and why?
Now, it is your time to provide your insights -- please explain why "LBM的稳定性等问题远比传统方法复杂,有太多的因素在影响,所以CFL数就不能 在LBM中具有其在传统算法中的地位。" To this day you have NOT given a reasonable explanation based on solid analysis, although you give numerous assertions without any substantiation. What are you talking about, may I ask?! |
|