Code review comment for lp:~maddevelopers/mg5amcnlo/3.0.2

Valentin Hirschi (valentin-hirschi) wrote :

Hi Rik,

Ok and I guess truncating the history to keep only a fixed number of past
events (may not impact the quality of the fit so much if this number is
large enough) would be hard to implement?

On Wed, May 13, 2020 at 6:25 PM Rikkert Frederix <email address hidden>
wrote:

> Hi Valentin,
>
> PolyVirtual keeps all events (i.e., the random numbers that corresponds to
> the generated momenta) for which the virtual have been computed in memory
> to get the best possible polynomial fit for the virtuals. It also writes
> them all to disk so that they can be re-used in the next integration step.
>
> I had hoped that in practice there are few enough PS points for which the
> virtuals need to be computed (i.e., well below several million) that this
> would not become an issue for most processes (in particular since the
> virt-tricks should work slightly better with the PolyVirtuals than
> without). Apparently this is not good enough when running on a cluster
> requiring high accuracy.
>
> Best,
> Rikkert
>
> --
> https://code.launchpad.net/~maddevelopers/mg5amcnlo/3.0.2/+merge/381317
> Your team MadDevelopers is subscribed to branch
> lp:~maddevelopers/mg5amcnlo/3.0.2.
>

--
Valentin

« Back to merge proposal