shockbubble convergence (in time)

[Matthew Emmett]

Hi everyone,

I'm (finally) following up on a chat I had with Jed at the last HPC^3
conference...

I'd like to compare SDC time-stepping to the RK104 scheme in PyClaw
using the shockbubble example. I am able to run PyClaw with SDC and
it appears to be working, so now the real work begins. Before I get
too far into things, is there a particular starting time that is of
particular interest (where the solution is non-trivial) to anyone?

Also, is there a slick way of restarting a run from an output file
that I am missing?

Hope everyone is doing well!
Matt

[David Ketcheson]

This sounds interesting, Matt.  Restarting is trivial:

>>> claw.solution = Solution(3,format='petsc')

That's from the documentation here: http://numerics.kaust.edu.sa/pyclaw/classes.html#restarting-a-simulation.  The first argument is the frame number to start from.

No need for the format argument unless you're using PetClaw).

I don't think there is any special time to look at.  It is hard to say that one scheme is "better" than another for turbulent solutions of the Euler equations, since there's no solution to converge to.  Also, keep in mind that we haven't implemented characteristic decomposition in the WENO reconstruction.  As you know, this can cause problems if you take solver.weno_order higher than 5.  Any contributions along those lines would be great.  I hope you'll contribute the SDC integrator code, too.

-David

--
You received this message because you are subscribed to the Google Groups "claw-dev" group.
To post to this group, send email to claw...@googlegroups.com.
To unsubscribe from this group, send email to claw-dev+u...@googlegroups.com.
For more options, visit this group at http://groups.google.com/group/claw-dev?hl=en.

[Matthew Emmett]

Thanks David. I'll choose a few different starting times and see how it goes.

I will certainly contribute SDC and any tweaks to WENO that I do.

BTW, if there is another HPC^3 conference in the works and I can help
with organising in any way, I would be happy too. The previous
conferences were really great and I appreciate all the effort that
went into them.

Matt

[Matthew Emmett]

On Fri, Oct 12, 2012 at 8:12 AM, David Ketcheson
<david.k...@kaust.edu.sa> wrote:

> This sounds interesting, Matt. Restarting is trivial:
>
>>>> claw.solution = Solution(3,format='petsc')

For the shockbubble example, I also had to do:

>>> for state in claw.solution.states:
>>> auxinit(state)

Is this expected or am I not dumping the solution properly (so that
aux is dumped too)?

Matt

[David Ketcheson]

Hi Matt,

That is expected (but was undocumented; thanks for pointing it out).  You can add

    claw.write_aux_init = True

    claw.write_aux_always = True

to your script (where `claw` refers to the Controller) in order to output the aux variables at every step in a file called `fort.a####`.  Then the Solution initialization will automatically load the aux values too.  I've added this to the documentation here: http://numerics.kaust.edu.sa/pyclaw/classes.html#outputting-aux-values.

-David

Matt

[Aron Ahmadia]

Matt, 

Please contact Kyle and Andy, who are pushing on HPC3 2013

A

[Matthew Emmett]

Thanks -- the new documentation is perfect.

Matt

On Mon, Oct 15, 2012 at 9:50 AM, David Ketcheson