New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Apfel++ benchmarks #181
Apfel++ benchmarks #181
Conversation
@Radonirinaunimi Yadmark is running, Note that while adding this I noticed that we still have to fix all the couplings for polarised SF in case of CC and NC |
Thanks a lot @giacomomagni! I can indeed take it from here. |
Co-authored-by: Giacomo Magni <39065935+giacomomagni@users.noreply.github.com>
self.run( | ||
self.theory_updates_zm(pto), | ||
self.obs_updates_pol(), | ||
["ToyLH_polarized"], |
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Using ToyLH_polarized
significantly worsen the agreement; at LO the difference can be up to 12%. This also happens when using that set to benchmark against Apfel.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
it seems that using ToyLH_polarized
enehanced the difference in g1
at NNLO, which was barely visible before.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
can you break the difference down by channel? (using genpdf) how about using the good old
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
we can try, the difference should originate from the Non Singlet coefficient.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
here you have. At NNLO using -p toy_pol
and using a finer Q grid (nQ=1000
) for Apfel++:
gonly_pol
:
──────────
g1_total
──────────
x Q2 yadism yadism_error APFEL++ percent_error
0 0.000012 10.000000 1.567862e-04 7.992130e-12 1.567876e-04 -0.000894
1 0.000049 10.000000 4.635223e-04 6.373798e-12 4.635245e-04 -0.000470
2 0.000204 10.000000 1.072596e-03 1.879000e-11 1.072594e-03 0.000148
3 0.000853 10.000000 1.955818e-03 3.772849e-11 1.955810e-03 0.000395
4 0.003562 10.000000 2.601327e-03 7.953587e-11 2.601333e-03 -0.000248
5 0.014874 10.000000 1.670687e-03 1.505371e-10 1.670663e-03 0.001465
6 0.062102 10.000000 -3.127464e-03 2.028302e-10 -3.127201e-03 0.008424
7 0.194737 10.000000 -6.261686e-03 1.752376e-10 -6.261719e-03 -0.000526
8 0.336842 10.000000 -3.545204e-03 4.788348e-11 -3.545221e-03 -0.000486
9 0.478947 10.000000 -1.226541e-03 2.452783e-11 -1.226448e-03 0.007615
10 0.621053 10.000000 -2.511471e-04 4.300323e-12 -2.512205e-04 -0.029240
11 0.763158 10.000000 -2.113743e-05 3.250476e-13 -2.090429e-05 1.115271
12 0.905263 10.000000 -1.089370e-07 6.238962e-15 -8.873470e-08 22.767103
13 0.001000 3.000000 1.905295e-03 4.628793e-11 1.905296e-03 -0.000049
14 0.001000 7.388431 2.258469e-03 5.386868e-11 2.258471e-03 -0.000051
15 0.001000 18.196305 1.738741e-03 4.083879e-11 1.738742e-03 -0.000071
16 0.001000 44.814047 1.572657e-03 3.648971e-11 1.572659e-03 -0.000134
17 0.001000 110.368500 1.329227e-03 3.050560e-11 1.329231e-03 -0.000236
18 0.001000 271.816686 1.147919e-03 2.608454e-11 1.147925e-03 -0.000490
19 0.001000 669.432950 1.011382e-03 2.277513e-11 1.011394e-03 -0.001116
20 0.001000 1648.686404 9.156301e-04 2.044882e-11 9.156547e-04 -0.002681
21 0.001000 4060.401956 8.726769e-04 1.934155e-11 8.727295e-04 -0.006031
22 0.001000 10000.000000 8.932759e-04 1.965916e-11 8.933740e-04 -0.010984
lonly_pol
ie light quarks only:
──────────
g1_total
──────────
x Q2 yadism yadism_error APFEL++ percent_error
0 0.000012 10.000000 -0.000291 -2.251389e-11 -0.000287 1.305583
1 0.000049 10.000000 -0.000344 -4.627033e-11 -0.000340 1.095596
2 0.000204 10.000000 -0.000373 -1.629789e-11 -0.000373 0.057539
3 0.000853 10.000000 -0.000240 4.838951e-12 -0.000254 -5.772452
4 0.003562 10.000000 0.001481 5.130205e-11 0.001418 4.454185
5 0.014874 10.000000 0.012064 1.868573e-10 0.011870 1.632602
6 0.062102 10.000000 0.050398 7.749967e-10 0.049997 0.800746
7 0.194737 10.000000 0.109347 2.073501e-09 0.108995 0.322261
8 0.336842 10.000000 0.126687 5.465003e-09 0.126531 0.123587
9 0.478947 10.000000 0.110273 4.325416e-09 0.110178 0.086218
10 0.621053 10.000000 0.072825 2.807101e-09 0.072871 -0.063313
11 0.763158 10.000000 0.031264 8.198026e-10 0.031163 0.323404
12 0.905263 10.000000 0.004238 9.329867e-11 0.004395 -3.578833
13 0.001000 3.000000 0.000064 4.590512e-12 0.000041 56.256089
14 0.001000 7.388431 -0.000302 9.068681e-12 -0.000322 -6.256600
15 0.001000 18.196305 -0.000009 6.725702e-12 -0.000024 -61.740713
16 0.001000 44.814047 0.000060 4.396884e-12 0.000046 30.844171
17 0.001000 110.368500 0.000175 3.627017e-12 0.000164 6.923428
18 0.001000 271.816686 0.000250 3.047234e-12 0.000241 3.877718
19 0.001000 669.432950 0.000290 2.564404e-12 0.000283 2.767274
20 0.001000 1648.686404 0.000275 2.045430e-12 0.000269 2.471179
21 0.001000 4060.401956 0.000128 1.240913e-12 0.000122 4.640097
22 0.001000 10000.000000 -0.000245 -3.842776e-14 -0.000250 -1.908960
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
so was the problem nQ
?
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
so was the problem
nQ
?
Yes! I am just now investigating if nQ=1000
is too much or it is enough to use nQ=200
.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
so was the problem
nQ
?
But only partly, just to be clear the problem is still there but there is not much we can do.
We are implementing the analytical relation for NS g1 = F3
while Aplel++ is not. In the bench you see the difference, which is small but not fully invisible.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
But this is an analytical relation and I am not sure we'll ever do anything about this.
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
Also note that now with nQ=1000
, the agreement for the full NNLO is now ~per-mille level (except for when the absolute results are ~zero).
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
please fix pre-commit as much as you can
116ada7
to
4d15c43
Compare
As for the pre-commit, most of the pydocstyle fails are not related to this PR and would better be addressed in a different PR. As for black & isort, they modify all the pylint labels as follows: from .external import ( # pylint:disable=import-error,import-outside-toplevel
apfel_utils,
) Do we want to introduce these? |
agreed
if they want so, yes |
This PR is to introduce APFEL++ in yadmark.
This will be needed to benchmark polarized SF.