[Nek5000-users] Restarting simulation in same run

[nek5000-users at lists.mcs.anl.gov]

Hi Parveen

There is implementation of arnoldi time-stepper for spectra calculation 
with parpack for nek5000. I think I've already sent you some info about 
it. The code you can find at
ftp://ftp.mech.kth.se/pub/adam/Nek5000/toolbox/
There are sfd and arnoldi set-ups based on ext_cyl example form 
repository. The restart of the stepper you can find in the code. We do 
it by setting istep to 0 (forces solver to go through 
BDF1->BDF2->BDF3->BDF3) and resetting some vectors to zero.
Best regards
Adam

On 22/10/14 15:07, nek5000-users at lists.mcs.anl.gov wrote:
> Hello Paul
>
> I dont want to restart in the middle.
>
> Every time I want to start the time stepping from scratch. So I want 
> to know how to reset everything for a fresh simulation. Every time 
> ARPACK gives me a vector, I have to start with 
> BDF1->BDF2->BDF3->BDF3... until final time.
>
> I think accuracy of time stepper also matters since we are 
> approximating exact evolution operator exp(A*t) with the time stepper.
>
> I have read some papers on time stepping approach but the details are 
> not clear.
>
> Thanks
> praveen
>
> On Wed, Oct 22, 2014 at 5:06 PM, <nek5000-users at lists.mcs.anl.gov 
> <mailto:nek5000-users at lists.mcs.anl.gov>> wrote:
>
>     Hi Praveen,
>
>     I would guess you need to develop your own timestepper because
>     Nek uses BDFk & EXTk  to time march, where typically k=3.  At
>     start-up, it uses k=1 and k=2 to bootstrap.   If you start
>     timestepping
>     in the middle of the simulation then you need to overwrite the
>     existing coefficients so that you can restart the timestepper,
>     plus you need to store your initial condition (which presumably
>     amounts to overwriting vx,vy,vz, and pr).
>
>     There are others in the community who have done these kinds of
>     things (I have not), who might have some further comments.
>
>     Best,
>
>     Paul
>
>     PS ----- Are you trying to find eigenvalues associated with
>     the spatial operators ?   i.e., does the accuracy of the
>     timestepper matter?  If not, you can likely get away with
>     k=1, in which case you don't need much history in the backward
>     difference formula or extrapolation.  Pressure, however, is
>     a bit of a tricky business because we use splitting and there
>     is a residual O(dt^k) impact...
>
>
>
>
>
> _______________________________________________
> Nek5000-users mailing list
> Nek5000-users at lists.mcs.anl.gov
> https://lists.mcs.anl.gov/mailman/listinfo/nek5000-users

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.mcs.anl.gov/pipermail/nek5000-users/attachments/20141027/84085c19/attachment.html>