Numerical FONLL #195

RoyStegeman · 2023-05-24T19:06:24Z

Implements numerical FONLL - supersedes #175

RoyStegeman · 2023-05-24T19:09:45Z

For now I just went with a quick-and-dirty implementation where the output are just the asy terms as implemented, this implementation is mainly for my own understanding. I get some qualitative agreement with apfel though in most of the kinematic domain at least ~10% off.

Before I proceed though, @alecandido @felixhekhorn, do you know if the definition of the FFN0 scheme in apfel is the same as what we need? And if so, what is the level of agreement I should find?

src/yadmark/benchmark/runner.py

src/yadism/observable_name.py

alecandido

Most of the files have small modifications, and the correct ones.

Now I'm on the phone, so I'm not really comfortable with the large files. But I'll try to finalize later today the review for the combiner and kernels. Apparently they look good as well.

src/yadism/input/compatibility.py

src/yadism/coefficient_functions/__init__.py

src/yadism/observable_name.py

src/yadism/coefficient_functions/__init__.py

Co-authored-by: Giacomo Magni <[email protected]>

RoyStegeman · 2023-10-09T14:33:02Z

@giacomomagni thanks for your comments, I hope I addressed all of them adequately.

Benchmark runners are quite dirty/contain old stuff, but this should be fine

It was quite dirty when I found it. However, since FONLL won't be benchmarked against apfel again (at least using this runner) I'd also be happy to undo the changes to the apfel_.py module introduced by this PR.

alecandido · 2023-10-09T17:14:44Z

It was quite dirty when I found it. However, since FONLL won't be benchmarked against apfel again (at least using this runner) I'd also be happy to undo the changes to the apfel_.py module introduced by this PR.

We could stop benchmarking FONLL, since it requires quite different operations. But it might be worth to benchmark instead FFNO, that was in APFEL, but not in yadism before this PR.

src/yadism/coefficient_functions/asy/kernels.py

Co-authored-by: Giacomo Magni <[email protected]>

tests/yadbox/test_export.py

andreab1997 · 2023-10-11T13:43:11Z

I realized something (tell me if you agree): when we ask yadism to do renormalization scale variations to the 7 subgrids of nFONLL for FONLL-B, it is using a different PTO for each of them (either 1 or 2). This means that, for some of them, we are doing actually NNLO ren sv instead of NLO (as we used to do in aFONLL and as we should do I believe). @RoyStegeman Can you confirm this?

CC: @felixhekhorn, @giacomomagni @alecandido

felixhekhorn · 2023-10-11T13:44:15Z

src/yadism/esf/esf.py

@@ -116,11 +116,12 @@ def compute_local(self):
        if self._computed:
            return
        cfc = cf.Combiner(self)
-        full_orders = [(o, 0, 0, 0) for o in self.orders]
        # prepare scale variations
        sv_manager = self.info.configs.managers["sv_manager"]
        if sv_manager is not None:
            full_orders = sv.build_orders(self.info.configs.theory["pto"])


as @andreab1997 pointed out: this should be "pto_evol". Since FONLL-B is a NLO scheme we also need to have NLO scale variations. Actually, this means that also the massive grids need to have that setting!

RoyStegeman · 2023-10-11T15:39:01Z

Thanks for pointing this out - I did not think about SV, but indeed if SV is NLO for all contributions (i.e. similar to FONLL-A) instead of the same order as FONLL-B for the central scale, then for SV the runcards will need to be updated to have PTO=1. If it were only FFN0 it would indeed have been possible to replace pto with pto_evol, but for massive this won't work.

So I suppose this should be fixed in pineko?

andreab1997 · 2023-10-12T08:54:47Z

Thanks for pointing this out - I did not think about SV, but indeed if SV is NLO for all contributions (i.e. similar to FONLL-A) instead of the same order as FONLL-B for the central scale, then for SV the runcards will need to be updated to have PTO=1. If it were only FFN0 it would indeed have been possible to replace pto with pto_evol, but for massive this won't work.

So I suppose this should be fixed in pineko?

I believe at this point we should just have a PTO (we can call it as you like) that we use for EKO and for the renormalization scale variations in yadism and that would be always 1 for FONLL-A and B and always 2 for FONLL-C and D. Then we need another PTO (that can be PTODIS maybe) that instead is used only by yadism and that is 2 always for FONLL-C but it is the usual 2,2,1,2,2,1,2 for FONLL-B. Then we would be only left with what PTODIS is used at the moment: as far as I understood it is used for the FFNS0. Could this instead be lifted to yadism?

RoyStegeman · 2023-10-12T09:23:28Z

Your proposal is exactly what yadism already does for the central scale, though I'm not exactly sure what you mean here:

Then we would be only left with what PTODIS is used at the moment: as far as I understood it is used for the FFNS0. Could this instead be lifted to yadism?

I'd have to look at how SV is implemented to see if it requires any changes, but from what I understood SV at NLO corresponds to both PTO=1 and PTODIS=1?

andreab1997 · 2023-10-12T10:57:25Z

Your proposal is exactly what yadism already does for the central scale, though I'm not exactly sure what you mean here:

Then we would be only left with what PTODIS is used at the moment: as far as I understood it is used for the FFNS0. Could this instead be lifted to yadism?

I'd have to look at how SV is implemented to see if it requires any changes, but from what I understood SV at NLO corresponds to both PTO=1 and PTODIS=1?

As far as I understood PTODIS is used for the asy parts in the case of FONLL-B. For SV it would be enough to ask for the sv kernel related to PTO=1 for each of the 7 theories, which in my mind could have been PTO since it has to be 1 already for the eko computation. So I would say SV should not use PTODIS at all. What I was saying is that if PTO we put equal to 1 always (for the reasons I said), then in yadism you need a different parameter that should be 2,2,1,2,2,1,2, according to the subtheory (which at the moment is instead PTO). Right?
As far as I understood then PTODIS is on top of all this discussion because being related to the asy components is yet another parameter which we may need to keep or to lift in yadism (if possible). But maybe I misunderstood the role of PTODIS

felixhekhorn · 2023-10-12T14:38:10Z

Just to be clear (and as @RoyStegeman already pointed out, it can be confusing): there are only two (2) couplings powers in the game - and no more. Unfortunately there names are ambiguously:

the (old) "PTO" in the theory card: since this is inherited from a NNPDF theory this refers to the PTO of evolution and in yadism this corresponds thus to pto_evol
the (new) "PTODIS" in the theory card: this we invented ourselves to exactly accommodate FONLL-B and it refers to pto inside yadism

the fallback for 2 is 1

2 should be used to compute coefficient functions
1 should be used to compute anything related to resummation, i.e. the logs in asy and $a_s$

alecandido · 2023-10-13T09:37:18Z

As @felixhekhorn said, PTODIS has been introduced to support FONLL-B (and all the skewed FONLL schemes), the whole of it, not just asy (indeed the fixed order in the asymptotic has to be the exact same as in the massive part, you can play just with the logs).

Concerning SV, I'm not sure I understand the point: they should be computed with the exact same settings of the central one, just varying the scale.
If central uses a non-trivial PTODIS, then SV have to do it as well.

felixhekhorn · 2023-10-13T11:01:32Z

Concerning SV, I'm not sure I understand the point: they should be computed with the exact same settings of the central one, just varying the scale.
If central uses a non-trivial PTODIS, then SV have to do it as well.

The question is to which order are you generating the logs? (inside the coeff fnc <=> inside the as expansion)

felixhekhorn

I didn't check everything super carefully, but all in all it looks good (and I trust in the benchmarks ;-) )

This reverts commit 2360c21.

niclaurenti · 2023-10-24T09:41:59Z

Accidentally pushed here...sorry

RoyStegeman marked this pull request as draft May 24, 2023 19:06

felixhekhorn added physics physics features benchmarks Benchmark (or infrastructure) related labels May 25, 2023

felixhekhorn mentioned this pull request May 25, 2023

Numerical FONLL #175

Closed

5 tasks

alecandido reviewed May 29, 2023

View reviewed changes

src/yadmark/benchmark/runner.py Outdated Show resolved Hide resolved

alecandido reviewed May 29, 2023

View reviewed changes

src/yadism/observable_name.py Show resolved Hide resolved

RoyStegeman mentioned this pull request May 30, 2023

Clean up inputcard keys #198

Closed

5 tasks

alecandido reviewed Jun 1, 2023

View reviewed changes

src/yadism/input/compatibility.py Outdated Show resolved Hide resolved

felixhekhorn reviewed Jun 1, 2023

View reviewed changes

src/yadism/coefficient_functions/__init__.py Outdated Show resolved Hide resolved

alecandido reviewed Jun 1, 2023

View reviewed changes

src/yadism/observable_name.py Outdated Show resolved Hide resolved

src/yadism/observable_name.py Outdated Show resolved Hide resolved

src/yadism/coefficient_functions/__init__.py Outdated Show resolved Hide resolved

src/yadism/coefficient_functions/__init__.py Show resolved Hide resolved

RoyStegeman added 19 commits June 5, 2023 14:09

try something for asy implementation

c4e193d

checkpoint

742a952

rebase

b428827

why doesn't apfel agree?

8735b2f

use nf instead of nl inputs for ffn0

cfc5910

add some docstrings

53fc25b

remove ipdb

bd1fd67

change sign asy

ecbe8d5

benchmark ffn0 for various nf

804e51d

expand xgrid ffn0 benchmark

fbec1f4

fix FFN0x in yadism

8b82e41

fix test

9ee1e6c

organize apfel_.py

2bc3745

update ffn0 benchmark

dd1cbb1

some cleanup

5fa33c3

remove asy observable in favor of using scheme

4a9dc2b

reinstate is_havylight flavor_family

fafa337

rename fonll to asy

1b4a4d6

generate_heavy_intrinsic_diff -> generate_heavy_intrinsic_asy

a77f66a

RoyStegeman and others added 4 commits October 9, 2023 14:41

Update src/yadism/coefficient_functions/__init__.py

80e61ad

Co-authored-by: Giacomo Magni <[email protected]>

update doctring asy.kernels module

5b61932

remove fonll from apfel runner

fdc153a

remove skip_heavylight from light.kernels.generate inputs

73aacec

giacomomagni reviewed Oct 9, 2023

View reviewed changes

src/yadism/coefficient_functions/asy/kernels.py Outdated Show resolved Hide resolved

src/yadism/coefficient_functions/asy/kernels.py Outdated Show resolved Hide resolved

Apply suggestions from code review

c60ecfc

Co-authored-by: Giacomo Magni <[email protected]>

felixhekhorn reviewed Oct 10, 2023

View reviewed changes

tests/yadbox/test_export.py Outdated Show resolved Hide resolved

remove FONLLParts=full from yadbox inputcard

bb150a1

felixhekhorn reviewed Oct 11, 2023

View reviewed changes

Update Adler

6c8834c

RoyStegeman and others added 3 commits October 15, 2023 14:18

change Adler interpolation function

aaca1c1

Use LeProHQ.Adler

560dea0

Merge branch 'master' into numerical_fonll

5df9d8a

giacomomagni mentioned this pull request Oct 20, 2023

adding new keys to theory template NNPDF/banana#51

Merged

felixhekhorn approved these changes Oct 24, 2023

View reviewed changes

niclaurenti added 2 commits October 24, 2023 11:39

Add hack nans function

2360c21

Revert "Add hack nans function"

abe93f0

This reverts commit 2360c21.

RoyStegeman merged commit a1937ab into master Nov 1, 2023
4 checks passed

RoyStegeman deleted the numerical_fonll branch November 1, 2023 13:47

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

Numerical FONLL #195

Numerical FONLL #195

RoyStegeman commented May 24, 2023 •

edited

Loading

RoyStegeman commented May 24, 2023

alecandido left a comment

RoyStegeman commented Oct 9, 2023

alecandido commented Oct 9, 2023

andreab1997 commented Oct 11, 2023

felixhekhorn Oct 11, 2023

RoyStegeman commented Oct 11, 2023 •

edited

Loading

andreab1997 commented Oct 12, 2023 •

edited

Loading

RoyStegeman commented Oct 12, 2023

andreab1997 commented Oct 12, 2023

felixhekhorn commented Oct 12, 2023

alecandido commented Oct 13, 2023

felixhekhorn commented Oct 13, 2023

felixhekhorn left a comment

niclaurenti commented Oct 24, 2023

Numerical FONLL #195

Numerical FONLL #195

Conversation

RoyStegeman commented May 24, 2023 • edited Loading

RoyStegeman commented May 24, 2023

alecandido left a comment

Choose a reason for hiding this comment

RoyStegeman commented Oct 9, 2023

alecandido commented Oct 9, 2023

andreab1997 commented Oct 11, 2023

felixhekhorn Oct 11, 2023

Choose a reason for hiding this comment

RoyStegeman commented Oct 11, 2023 • edited Loading

andreab1997 commented Oct 12, 2023 • edited Loading

RoyStegeman commented Oct 12, 2023

andreab1997 commented Oct 12, 2023

felixhekhorn commented Oct 12, 2023

alecandido commented Oct 13, 2023

felixhekhorn commented Oct 13, 2023

felixhekhorn left a comment

Choose a reason for hiding this comment

niclaurenti commented Oct 24, 2023

RoyStegeman commented May 24, 2023 •

edited

Loading

RoyStegeman commented Oct 11, 2023 •

edited

Loading

andreab1997 commented Oct 12, 2023 •

edited

Loading