Exactly, I’m attempting to try a higher order like 2 or 3 with a more robust machine. Since we’re talking about the order and mesh resolution, I have two questions that have been bothering me for a long time about the issue.
-
A coarser mesh is preferable with a high scheme order, for it means a lower time cost and shows the power of FR algorithms. Of course, measurements on anti-alising would be suggested to take to avoid the nonphysical oscillation for a coarse mesh. But how could we determine if a mesh is too coarse to get wrong calculation results. For instance, if a total number of 100,000 eles is required to get a correct result using FV with
order = 2
, then is it right that at least 25,000 eles is appropriate forp = 2
and 1,100 eles forp = 3
(Nele~1/p^2)? I was also aware of that the desired wall normal resolution is only needed to be achieved for a certain polynomial order. So as long as we keep the y+ value suitable for the certain polynomial order, the result under ILES should be guaranteed, right? -
In the common cases like cylinder and airfoil, it’s highly recommended generating a curved boundary mesh using high-order in Gmsh (Ref Bassi, 1997). But does that mean a high-order mesh is less attractive, considering the non-curved shape like a square or a plate, compared with a high-order polynomial order in FR scheme? The latter will subdivide the straight boundary and generate polynomials to achieve a high-order and accurate result.
Thanks for the previous kind and detailed replies. And as a complement to the Keyerror : Kernel mul has no providers
I met, it works by adding export PYFR_XSMM_LIBRARY_PATH=/path/to/compiled/libxsmm.so
.
Regards, Thatcher