Hi Paul, Thanks for the quick answer. I have been using a fixed time-step. After a suggestion from Johan, I will try to overintegrate with lxd \approx 2*lx1, since I use a curved geometry. In addition, it might be useful to try with an even polynomial order? (Since my problems start as I change from N=5 to N=7, maybe N=8 is better?) George Quoting [email protected]:
Hi George,
Is your timestep size variable? or fixed?
You can fix it (highly recommended) by setting it to the value you want, but negative (e.g., param 12 = -.001 in the .rea file would imply that DT=.001 for all timesteps).
From what you describe, I am guessing that this is the issue. In particular, variable timestepping + projection has some difficulties.
With dealiasing and filtering, I never have problems with Torder=3, and I would recommend using this. I concur, however, with Johan that Torder=2 is more stable. (See the CRAS article with J. Mullen in 2001.)
I hope this helps - please let us know if problems persist.
Paul
On Wed, 7 Dec 2011, [email protected] wrote:
Dear All,
I have been conducting a direct numerical simulation of turbulent pipe flow using nek5000 at a moderately high Reynolds number (at present Re_tau=550). For this, I have to use a very fine grid in order to capture the flow physics accurately (for a length of 25R, I am using a total of 450 million grid points when assuming a polynomial order 7). However, I have been having problems with the stability of the code when using TORDER = 3, i.e. parameter 027. The way I conducted my simulation was as follows:
- I started with polynomial order 3, TORDER set to 3, and run for 4 flow through times, starting from random noise. At this point I had the projection parameters, i.e. P094 and P095 set to non-zero values.
- Afterwards, I increased the polynomial order to 5, with TORDER kept at 3, and ran for another 2 flow through times. Here, I set the projection parameters P094 and P095 to zero.
- Now, starting from the last field of the previous run, I changed to my target polynomial order 7, and keeping TORDER = 3. However, the codes explodes shortly after restarting. The error is: "failed in HMHOLTZ." I also tried a considerably lower time-step, but the problem persisted. Meanwhile, when I change TORDER to 2, the code runs fine and does not explode. The "physical" results are absolutely fine, and the statistics I get are in perfect agreement to our expectations.
I did not get the same problem while running a turbulent pipe flow at a lower Reynolds number (Re_tau=180) and with less grid points.
Probably also of importance, I did test with various ways of restarting, i.e. storing in fld or f files, using 4 or 8 byte accuracy, and storing multiple consecutive fields for the multi-step time scheme. None of these changes made significant impact. Also, I am using overintegration and filtering (0.01 in parameter P103); increasing the filtering (by a factor of 5) did also not help.
I am wondering if any faced a similar issue before, and if would have an insight on how to tackle it. Could my problem maybe be related to the restart-issue that Adam brought up recently?
Best regards George
---------------------------------------------------------------- This message was sent using IMP, the Internet Messaging Program. _______________________________________________ Nek5000-users mailing list [email protected] https://lists.mcs.anl.gov/mailman/listinfo/nek5000-users
_______________________________________________ Nek5000-users mailing list [email protected] https://lists.mcs.anl.gov/mailman/listinfo/nek5000-users
---------------------------------------------------------------- This message was sent using IMP, the Internet Messaging Program.