Factorization scale is not computed correctly

felixhekhorn commented 2 years ago

The factorizations scales for NMC_NC_EM_P_F2 are bugged and I don't understand why
as we're in DIS Q2 = muF2
in the observable card we have for the last bin Q2: 99.03 - see https://github.com/NNPDF/runcards/blob/58c13d69787fc35cfd7f3ca9091f08ef1608ebf0/nnpdf31_proc/NMC_NC_EM_P_F2/observable.yaml#L837

and this is also reflected in the grid:

$ pineappl convolute data/grids/200-NMC_NC_EM_P_F2.pineappl.lz4 NNPDF40_nnlo_as_01180
LHAPDF 6.4.0 loading /usr/share/lhapdf/LHAPDF/NNPDF40_nnlo_as_01180/NNPDF40_nnlo_as_01180_0000.dat
NNPDF40_nnlo_as_01180 PDF set, member #0, version 1                                                      
bin     Q2            x           F2p      scale uncertainty
---+-----+-----+------+------+------------+--------+--------
0  0.16  0.16 0.0015 0.0015 1.7642563e-2 -105.85%  576.52%      
[...]
259 99.03 99.03  0.675  0.675 2.2627258e-2  -23.83%   39.32%

but this is not reflected by the subgrids command:

$ pineappl subgrids --muf data/grids/200-NMC_NC_EM_D_F2.pineappl.lz4
o  b  l   muf
-+---+--+-----
[...]
3 259 10 9.951

and this is also not reflected in Grid.axes https://github.com/N3PDF/pineappl/blob/725e4109c2b9f90cc883abd251c190409fe9f22c/pineappl/src/grid.rs#L1222 called via https://github.com/NNPDF/fktables/blob/7f34897005a90a6bcca38bf45a90b8dfb599dc99/fkutils/builder.py#L133 leading eventually to
```
$ cat data/operators_cards/NMC_NC_EM_P_F2.yaml
Q2grid:
```
26.07 [...]
99.03

and so finally resulting in

$ python build.py 
┌───────────────┐
│ Computing ... │
└───────────────┘
/media/FK/fktables/data/grids/200-NMC_NC_EM_D_F2.pineappl.lz4
+ /media/FK/fktables/data/operators/200-NMC_NC_EM_D_F2.tar
= /media/FK/fktables/data/fktables/200-NMC_NC_EM_D_F2.pineappl.lz4
[00:00:00] ░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░░      14/40040   - ETA: 00:00:00 
thread '<unnamed>' panicked at 'Couldn't find q2: 0.16 with xir: 1.0 and muf2_grid: [26.07, 26.36, 26.49, 26.52, 26.55, 26.6, 26.61, 26.63, 26.64, 34.74, 35.28, 35.32, 35.37, 35.4, 35.42, 35.43, 35.45, 35.5, 35.57, 44.94, 46.95, 46.98, 47.01, 47.03, 47.12, 47.16, 47.26, 63.04, 63.23, 63.48, 63.52, 63.53, 63.56, 63.65, 90.68, 96.35, 98.05, 98.82, 99.03]', /home/pineappl/pineappl/pineappl/src/grid.rs:1519:33
Traceback (most recent call last):
File "/media/FK/fktables/build.py", line 109, in <module>
b.build_fktables(pdf)
File "/media/FK/fktables/fkutils/builder.py", line 226, in build_fktables
pineko.convolute(grid, operator_path.resolve(), fktable_path, pdf)
File "/media/FK/pineko/src/pineko/__init__.py", line 116, in convolute
fktable = pineappl_grid.convolute_eko(operators, "evol", order_mask=order_mask_qcd)
File "/media/FK/env/lib/python3.10/site-packages/pineappl/grid.py", line 270, in convolute_eko
self.raw.convolute_eko(
pyo3_runtime.PanicException: Couldn't find q2: 0.16 with xir: 1.0 and muf2_grid: [26.07, 26.36, 26.49, 26.52, 26.55, 26.6, 26.61, 26.63, 26.64, 34.74, 35.28, 35.32, 35.37, 35.4, 35.42, 35.43, 35.45, 35.5, 35.57, 44.94, 46.95, 46.98, 47.01, 47.03, 47.12, 47.16, 47.26, 63.04, 63.23, 63.48, 63.52, 63.53, 63.56, 63.65, 90.68, 96.35, 98.05, 98.82, 99.03]

(of course the very same in proton and deuteron)

Any ideas on your side?

cschwan commented 2 years ago

@felixhekhorn subgrid --muf prints the unsquared factorization scale, so 9.951 * 9.951 = 99.022401 which is much closer to the desired value (the last digit might be off due to rounding). Having the unsquared values is nice to recognize prominent masses, such as the W- and Z-boson masses. However, having the squared options would be useful as well; I've opened https://github.com/N3PDF/pineappl/issues/102 for that.

felixhekhorn commented 2 years ago

@felixhekhorn subgrid --muf prints the unsquared factorization scale, so 9.951 * 9.951 = 99.022401 which is much closer to the desired value (the last digit might be off due to rounding). Should I also add --muf2 and --mur2 options to print the squared values?

ok, I see - yes, please, if that is not a big effort a --muf2 would be appreciated

felixhekhorn commented 2 years ago

still, the true problem remains: why does axes cut out some muf^2 values?

the biggest bin is indeed there Q2=99.03
but the low bins are all missing, i.e. Q2 < 26.07

cschwan commented 2 years ago

It seems that the EKO is incomplete. Here's the relevant piece of code:

https://github.com/N3PDF/pineappl/blob/ca86b939d74707e1197c4cfa8387f1fd3402921e/pineappl/src/grid.rs#L1512-L1524

These lines search eko_info for the proper factorization scale value, but this object comes from outside so there's something wrong at a higher level. But also this method needs proper error handling; see https://github.com/N3PDF/pineappl/issues/85.

felixhekhorn commented 2 years ago

It seems that the EKO is incomplete

true, but we use axes to determine which EKOs we actually need - so the problem is there

In [1]: import pineappl

In [2]: g = pineappl.grid.Grid.read("200-NMC_NC_EM_D_F2.pineappl.lz4")

In [3]: g.axes()
Out[3]: 
[...]
[26.07,
  26.36,
  26.49,
  26.52,
[...]

cschwan commented 2 years ago

Ah :smile: I see, so this comes full circle... I'll have a look at it!

cschwan commented 2 years ago

@felixhekhorn please try it again with commit e91a7df606e3cc36f52108545319c94a267456a9.

felixhekhorn commented 2 years ago

yes, seemed to have done the trick:

[0.16,                                                                                                  │   Compiling pyo3-macros v0.14.5
  0.17,                                                                                                  │   Compiling thiserror v1.0.30
  0.25,                                                                                                  │   Compiling ndarray v0.15.4
  0.26,                                                                                                  │   Compiling bincode v1.3.3
  0.35,                                                                                                  │   Compiling numpy v0.14.1
  0.36,                                                                                                  │   Compiling pineappl v0.5.0-beta.6 (/media/FK/pineappl/pineappl)
  0.45,                                                                                                  │   Compiling pineappl_py v0.5.0-beta.6 (/media/FK/pineappl/pineappl_py)
  0.46,                                                                                                  │    Finished release [optimized] target(s) in 18.96s
  0.6,

(aaa tmux mess - don't worry ;-) )

felixhekhorn commented 2 years ago

@AleCandido I'm rerunning the EKOs at the moment, though only one of them - the other one can be linked since they are identical

alecandido commented 2 years ago

Ok, thank you very much :D

NNPDF / pineappl

Factorization scale is not computed correctly #101