CFX Pre/Double Precision

Scott Nordsen · July 21, 2009, 15:31

I am working on a case with small finite volumes (~1e-18 m^3).

To run in fluent I used double precision solver since single precision had issues with the mesh.

Now, I am trying to run the same case in CFX.

I see how to run a double precision simulation with cfx5solve.

Question is how do I setup the case using cfx5pre?
-double is not an option.
When I read msh file into cfx5pre I get mesh errors as I did with fluent single precision.

The only difference is fluent is one tool (pre/solve/post) that I can start as dp from the start.

Thanks

Scott

ghorrocks · July 21, 2009, 19:57

Hi,

The fact that you have small finite volumes should not cause grief (my control volumes in my current work are 1e-20 m^3!), providing all the control volumes are a similar order of magnitude in size. There is no loss in precision of 1e-18 versus 1e0 as the significand has the same number of bits representing its size.

The problem is when you have a large range in sizes. What is the largest and smallest finite volume in your model?

Glenn Horrocks

Scott Nordsen · July 22, 2009, 10:02

Have a geom with a far field ~20D away down to the geometry that has edges ~0.1".
D~10 ft

So my cells range from 10m^3 to 1e-18 m^3. So as you would say I have a problem.

ckleanth · July 22, 2009, 12:12

an easy way of refining could be to run it first steady state with a coarse mesh and use a few adaptive mesh steps but I'm not sure how you will manage to get low volume aspect ratio.. I suppose your best bet is if you can create a nice hexa mesh so it wont distort during mesh adaptation

ghorrocks · July 22, 2009, 19:45

Hi,

Cell volumes ranging over 19 orders of magnitude is not a good idea. That is what your problem is. If your fine cells really need to be that fine then you are going to have to refine your coarse cells. Yes, that probably means you need a supercomputer but hey, this is CFD and why CFD needs supercomputers.

Glenn Horrocks

July 21, 2009, 15:31	CFX Pre/Double Precision	#1
Scott Nordsen New Member Join Date: Jun 2009 Posts: 13 Rep Power: 17	I am working on a case with small finite volumes (~1e-18 m^3). To run in fluent I used double precision solver since single precision had issues with the mesh. Now, I am trying to run the same case in CFX. I see how to run a double precision simulation with cfx5solve. Question is how do I setup the case using cfx5pre? -double is not an option. When I read msh file into cfx5pre I get mesh errors as I did with fluent single precision. The only difference is fluent is one tool (pre/solve/post) that I can start as dp from the start. Thanks Scott

July 22, 2009, 12:12		#4
ckleanth Senior Member George Join Date: Mar 2009 Location: Birmingham, UK Posts: 257 Rep Power: 18	an easy way of refining could be to run it first steady state with a coarse mesh and use a few adaptive mesh steps but I'm not sure how you will manage to get low volume aspect ratio.. I suppose your best bet is if you can create a nice hexa mesh so it wont distort during mesh adaptation __________________ Top 4 tips 1. Knowledge is everything and Ignorance is dangerous. 2. Understand your limitations and try to eliminate them. 3. Get yerself a bike and hoon the chuffer. You will soon learn why dogs like to hang their heads out the car window. 4. Please before asking any questions on how to run simulations in CFX, go though all the tutorials

Similar Threads
Thread	Thread Starter	Forum	Replies	Last Post
Pros and Cons for CFX, CFdesign, COMSOL	Val	Main CFD Forum	3	June 10, 2011 03:20
double precision cfx	pat	CFX	5	November 11, 2008 07:45
what's wrong about my code for 2d burgers equation	morxio	Main CFD Forum	3	April 27, 2007 11:38
CFX 64bit single-double precision	Tom	CFX	2	December 19, 2006 16:39
Double precision CFX	Vanessa	CFX	8	August 13, 2006 05:34

July 21, 2009, 19:57		#2
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,857 Rep Power: 144	Hi, The fact that you have small finite volumes should not cause grief (my control volumes in my current work are 1e-20 m^3!), providing all the control volumes are a similar order of magnitude in size. There is no loss in precision of 1e-18 versus 1e0 as the significand has the same number of bits representing its size. The problem is when you have a large range in sizes. What is the largest and smallest finite volume in your model? Glenn Horrocks

July 22, 2009, 10:02		#3
Scott Nordsen New Member Join Date: Jun 2009 Posts: 13 Rep Power: 17	Have a geom with a far field ~20D away down to the geometry that has edges ~0.1". D~10 ft So my cells range from 10m^3 to 1e-18 m^3. So as you would say I have a problem.

July 22, 2009, 19:45		#5
ghorrocks Super Moderator Glenn Horrocks Join Date: Mar 2009 Location: Sydney, Australia Posts: 17,857 Rep Power: 144	Hi, Cell volumes ranging over 19 orders of magnitude is not a good idea. That is what your problem is. If your fine cells really need to be that fine then you are going to have to refine your coarse cells. Yes, that probably means you need a supercomputer but hey, this is CFD and why CFD needs supercomputers. Glenn Horrocks