Projection technique's parameters
Hi Neks, My question is about the parameters which manage the projection method in the NEK5000: p93, p94 and p95. How can the correct values be selected for these parameters to obtain the better performance depending on the nature of simulation: laminar / turbulent, convection / diffusion, time step? Thank you for any suggestion, Andrew
Hi Andrew, The projection routines are currently being overhauled, but the general strategy is: Number of projections controlled by p93 and mxprev in SIZE. p94 > 0: start projecting velocity/temp after step p94 p95 > 0: start projecting pressure after step 95 I typically set p94 and p95 to 5 I set mxprev and p93 to 20, but sometimes to a higher value. I used to occasionally to to much higher than 20 (say, 80), but in a recent rewrite and effort to make projection more robust I slowed the projection step down with too much work, so another rewrite is due soon. Paul PS - note that certain combinations of run parameters don't support projection or potentially blow up. Set p94=0 in this case and retry. There was a bug that was fixed about 6 weeks back---it may have cured some of the cases that users were running into. ________________________________ From: [email protected] [[email protected]] on behalf of [email protected] [[email protected]] Sent: Monday, June 27, 2016 3:17 AM To: [email protected] Subject: [Nek5000-users] Projection technique's parameters Hi Neks, My question is about the parameters which manage the projection method in the NEK5000: p93, p94 and p95. How can the correct values be selected for these parameters to obtain the better performance depending on the nature of simulation: laminar / turbulent, convection / diffusion, time step? Thank you for any suggestion, Andrew
Paul, is there a paper/report we can have access to that goes in the details of the implementation, please? Also, are the typical values based on your personal experience? Thanks, Marco On Jun 27, 2016, at 8:55 AM, [email protected]<mailto:[email protected]> wrote: Hi Andrew, The projection routines are currently being overhauled, but the general strategy is: Number of projections controlled by p93 and mxprev in SIZE. p94 > 0: start projecting velocity/temp after step p94 p95 > 0: start projecting pressure after step 95 I typically set p94 and p95 to 5 I set mxprev and p93 to 20, but sometimes to a higher value. I used to occasionally to to much higher than 20 (say, 80), but in a recent rewrite and effort to make projection more robust I slowed the projection step down with too much work, so another rewrite is due soon. Paul PS - note that certain combinations of run parameters don't support projection or potentially blow up. Set p94=0 in this case and retry. There was a bug that was fixed about 6 weeks back---it may have cured some of the cases that users were running into. ________________________________ From: [email protected]<mailto:[email protected]> [[email protected]<mailto:[email protected]>] on behalf of [email protected]<mailto:[email protected]> [[email protected]<mailto:[email protected]>] Sent: Monday, June 27, 2016 3:17 AM To: [email protected]<mailto:[email protected]> Subject: [Nek5000-users] Projection technique's parameters Hi Neks, My question is about the parameters which manage the projection method in the NEK5000: p93, p94 and p95. How can the correct values be selected for these parameters to obtain the better performance depending on the nature of simulation: laminar / turbulent, convection / diffusion, time step? Thank you for any suggestion, Andrew _______________________________________________ Nek5000-users mailing list [email protected]<mailto:[email protected]> https://lists.mcs.anl.gov/mailman/listinfo/nek5000-users
Hi Marco, Here is the reference: 1. Fischer, P.F., "Projection techniques for iterative solution of Ax=b with successive right-hand sides." <http://www.mcs.anl.gov/~fischer/pubhtml/project.pdf> Comp. Meth. in Appl. Mech 163 (1998). There is also a 1993 ICASE Report, but it's mostly a subset of this paper. Paul PS - these typical values are based on my experience... but, it's easy to try a run for a few hundred steps and see which parameters make things run faster. ________________________________ From: [email protected] [[email protected]] on behalf of [email protected] [[email protected]] Sent: Monday, June 27, 2016 8:01 AM To: [email protected] Subject: Re: [Nek5000-users] Projection technique's parameters Paul, is there a paper/report we can have access to that goes in the details of the implementation, please? Also, are the typical values based on your personal experience? Thanks, Marco On Jun 27, 2016, at 8:55 AM, [email protected]<mailto:[email protected]> wrote: Hi Andrew, The projection routines are currently being overhauled, but the general strategy is: Number of projections controlled by p93 and mxprev in SIZE. p94 > 0: start projecting velocity/temp after step p94 p95 > 0: start projecting pressure after step 95 I typically set p94 and p95 to 5 I set mxprev and p93 to 20, but sometimes to a higher value. I used to occasionally to to much higher than 20 (say, 80), but in a recent rewrite and effort to make projection more robust I slowed the projection step down with too much work, so another rewrite is due soon. Paul PS - note that certain combinations of run parameters don't support projection or potentially blow up. Set p94=0 in this case and retry. There was a bug that was fixed about 6 weeks back---it may have cured some of the cases that users were running into. ________________________________ From: [email protected]<mailto:[email protected]> [[email protected]<mailto:[email protected]>] on behalf of [email protected]<mailto:[email protected]> [[email protected]<mailto:[email protected]>] Sent: Monday, June 27, 2016 3:17 AM To: [email protected]<mailto:[email protected]> Subject: [Nek5000-users] Projection technique's parameters Hi Neks, My question is about the parameters which manage the projection method in the NEK5000: p93, p94 and p95. How can the correct values be selected for these parameters to obtain the better performance depending on the nature of simulation: laminar / turbulent, convection / diffusion, time step? Thank you for any suggestion, Andrew _______________________________________________ Nek5000-users mailing list [email protected]<mailto:[email protected]> https://lists.mcs.anl.gov/mailman/listinfo/nek5000-users
participants (1)
-
nek5000-users＠lists.mcs.anl.gov