[Users] meeting minutes for 2017-02-13

[Erik Schnetter]

On Mon, Feb 13, 2017 at 12:09 PM, Roland Haas <rhaas at illinois.edu> wrote:
> Present: Eloisa, Frank, Gwyneth, Ian, Yosef, Roland, Steve,
> Charalampos, Steve
>
> CarpetRegrid2:
> * regrid_every is the first condition checked, it only ever does
>   anything if the iteraton is a multiple of it
> * movement threshold is an additional condition that prevents grid
>   movement until the grids have moved a minimum distance
> * there is not well known optimal regridding frequency. One
>   consideration would be to regrid when all timelevels are "in sync" ie
>   with the same frequency that the coarsest level steps so that there
>   is no interpolation in time. Another competing consideration is that
>   the black holes should not have moved too far which gives an upper
>   limit. A reason to use regrid_every is also if the positions are not
>   known at every timestep for example if AHFinderDirect is used. With
>   PunctureTracker is used then the information is known at each time.
> * the dominant effect is to regrid frequently enough to track the black
>   holes moving
> * Courant factor is ratio between time step and space step on the
>   coarsest grid, to compute the courant factor on the finest grid
>   involves the time refinement factor

To be clear, the CFL factor on finer grids need to take both the
spatial and temporal refinement factors into account.

> * often one sees that the time refinement factors are set 1,1,2,4,8,...
>   instead of 1,2,4,8 which is the default. This is due to an
>   instability in the shift evolution equations. The group in Jena found
>   this to be important (there is a paper by Bernd Bruegmann et al),
>   there is a paper by Erik Schnetter that gives reason for this. It is
>   likely still required for purely Cartesian grids, for Llama
>   (curvilinear) grids one can get away without doing so

One can set the driver conditions in the evolution and gauge equations
such that this reduction is not necessary. I don't know whether this
has been implemented.

> * metric_timelevels is set to 3 for AHFinderDirect setting the same for
>   lapse an shift does not do harm but there is not directly obvious
>   reason why one would need it

QuasiLocalModes might look at lapse and shift to calculate the full
four-metric and its time derivative.

-erik

> * computing the energy carried of by gravitational waves is done by
>   computing the WeylScalar on a sphere and then do the integral
>   yourself. There are a number of analysis packages that help doing so
>   eg SimulationTools (google find it) by Ian Hinder for Mathematica or
>   possibly pyCactusET. The list of tools is available here:
> https://docs.einsteintoolkit.org/et-docs/Analysis_and_post-processing
> * some of GW extraction is shown in the gallery:
> http://einsteintoolkit.org/about/gallery/gw150914 and Ian will send instructions
> * if seeing NaNs when running on many nodes or other errors this is a
>   bug and should at most lead to an error message. static_tov uses a
>   very small grid and we would suggest using a higher resolution grid.
>   A gut feeling would say that anything above 27 MPI ranks is
>   difficult. The issue seems to happen for 40 MPI ranks.
>
> Website:
> * comments from Ian via email
> * comments from Roland, who will incorporate them into the website
>   repo: https://knarrff@bitbucket.org/einsteintoolkit/www.git
>
> MHD workshop:
> * https://docs.einsteintoolkit.org/et-docs/2017_MHD_Workshop has the
>   wiki page
> * Roland will write a summary in an email
>
> ET fails to build right now
> * there is an unsupported OMP pragma in MoL, seems like a compiler
>   specific issue
> * Roland suggests to check if all compilers on our "usual" clusters
>   support "OMP simd" yet
>
> GetComponents:
> * https://trac.einsteintoolkit.org/ticket/1916 contains a
>   parallelization for GetComponents please give it a try to see if you
>   can make it break
>
> Yours,
> Roland
>
> --
> My email is as private as my paper mail. I therefore support encrypting
> and signing email messages. Get my PGP key from http://keys.gnupg.net.
>
> _______________________________________________
> Users mailing list
> Users at einsteintoolkit.org
> http://lists.einsteintoolkit.org/mailman/listinfo/users
>



-- 
Erik Schnetter <schnetter at cct.lsu.edu>
http://www.perimeterinstitute.ca/personal/eschnetter/