You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
At the moment we want to keep compatibility with the runcard format defined by #220
However, while reworking the internals for numerical FONLL I just defined the concept of the "evolution point", that (apart from the name) is a clean abstraction suitable for the runcard as well.
For this reason, I start here a list of desiderata for the next runcard update, without any idea about its schedule.
After #242 we should not need any longer, and parallelism in Python has poor support, better avoiding it, and apply at some different level (dispatch multiple jobs, use parallel primitives).
"""Number of cores used to parallelize integration."""
Max nf
Drop both MaxNfPdf and MaxNfAs because redundant with the matchings mechanism.
Actually, the first one is fully redundant, while the second is only redundant if equal to the first, but there is no evidence that it makes sense using it separately during an evolution run (and since EKO is a library, like APFEL, if you want just to evolve $\alpha_s$ you can pass your matchings to the Couplings class)
Runcards update
It will be up to the user to use the new nested runcards.
We will move the update() function to ekomark, for its internal uses.
The text was updated successfully, but these errors were encountered:
At the moment we want to keep compatibility with the runcard format defined by #220
However, while reworking the internals for numerical FONLL I just defined the concept of the "evolution point", that (apart from the name) is a clean abstraction suitable for the runcard as well.
For this reason, I start here a list of desiderata for the next runcard update, without any idea about its schedule.
And related:
runcards.update()
Evolution points
Move
num_flavs_init
:eko/src/eko/quantities/heavy_quarks.py
Lines 106 to 111 in 0224739
to
mu0
:eko/src/eko/io/runcards.py
Lines 95 to 96 in 0224739
Also the couplings reference should be an evolution point
(scale, num_flavs_ref)
:eko/src/eko/quantities/couplings.py
Lines 23 to 25 in 0224739
Intrinsic range
eko/src/eko/runner/parts.py
Lines 45 to 62 in 0224739
eko/src/eko/quantities/heavy_quarks.py
Lines 114 to 115 in 0224739
Integration cores
After #242 we should not need any longer, and parallelism in Python has poor support, better avoiding it, and apply at some different level (dispatch multiple jobs, use parallel primitives).
eko/src/eko/io/runcards.py
Lines 87 to 88 in 0224739
Max nf
Drop both
MaxNfPdf
andMaxNfAs
because redundant with the matchings mechanism.Actually, the first one is fully redundant, while the second is only redundant if equal to the first, but there is no evidence that it makes sense using it separately during an evolution run (and since EKO is a library, like APFEL, if you want just to evolve$\alpha_s$ you can pass your matchings to the
Couplings
class)Runcards update
It will be up to the user to use the new nested runcards.
We will move the
update()
function toekomark
, for its internal uses.The text was updated successfully, but these errors were encountered: