Time-like Anomalous dimensions #200

t7phy · 2023-01-20T01:22:38Z

Added the LO Anomalous dimensions as starters, hence the PR is currently set to draft

…like_unpolarized

alecandido

Thanks @t7phy for this PR!

In general, it looks good to me, I just left some comments to improve the style, by making it uniform to the rest of the code.
Concerning EKO itself, you can look at the other modules to have a reference (I guess you already done):
https://github.com/NNPDF/eko/blob/time_like_unpolarized/src/ekore/anomalous_dimensions/unpolarized/space_like/as1.py
while the official reference we follow for the docstrings is:
https://www.sphinx-doc.org/en/master/usage/extensions/example_numpy.html

Having uniform docstrings is not top-priority, but we try to stick to a single convention in order to minimize future changes (and possibly to make it easier to automate them).

src/ekore/anomalous_dimensions/unpolarized/time_like/as1.py

alecandido · 2023-01-20T08:16:57Z

@t7phy just another tiny advice: commits comments are useful to get a quick summary of changes done, and to rapidly identify which are the responsible comments, in case you're looking for something specific.

Writing great commit messages is terribly difficult and annoying, so lots of people are doing this poorly, and everyone is sloppy sometimes.
But if you can, try to be as descriptive as you can, keeping it brief (a short sentence for each commit is definitely enough).

felixhekhorn · 2023-01-23T09:32:02Z

src/ekore/anomalous_dimensions/unpolarized/time_like/__init__.py

@@ -1,21 +0,0 @@
-r"""The unpolarized, time-like Altarelli-Parisi splitting kernels.


this file you need for sure since it is the entry point to your implementation - so instead of deleting it you should adjust it (take a look to the space-like counterpart)

yes indeed, I had only removed it so I could add the actual functions once I am done with the as2 and as3 modules.

but you don't need to delete the file because of that, no? and for sure we start with LO first and once we established that we step forward ...

the file is still there, I had removed the functions which stated it is not implemented, to make way for the actual functions implementing the splitting functions. I will add the functions asap.

felixhekhorn · 2023-01-23T09:34:30Z

src/ekore/anomalous_dimensions/unpolarized/time_like/as1.py

+        LO quark-quark anomalous dimension $\gamma_{qq}^{(0)}(N)$ 
+
+    """
+    return constants.CF * (-3.0 + (4.0 * s1) - 2.0 / (N * (N + 1.0)))


this is the same as

eko/src/ekore/anomalous_dimensions/unpolarized/space_like/as1.py

Line 10 in 7a74ddb

def gamma_ns(N, s1):

so you should not reimplement, but just redirect (same of course for all the others); only gamma_singlet changes if I understood correctly (so that one you keep)

well actually, only the 2 functions, namely gamma_ns and gamma_gg remain the same. gamma_gq and gamma_qg do differ by factors of nf. I think it would be better and clearer if we have the _ns and _gg functions clearly specified and not redirect to the space-like folder because it provides a cleaner and clearer separation between TL and SL and also allows for proper citation of references with functions.

Even fine ... are there relations beyond LO between TL and SL?

no, its only for the LO afaik, for NLO and NNLO, they vary, so there will be different functions there anyway

Then I'm also fine with duplicating two lines ... just remember to add a unit test that actually checks that

src/ekore/anomalous_dimensions/unpolarized/time_like/as1.py

felixhekhorn · 2023-02-01T10:06:22Z

If you want to build on top of Caching harmonic sums #202 consider to rebase this PR else you can not test your code
in any case I would suggest you strongly to first do the LO benchmark before going up in order

felixhekhorn · 2023-02-01T10:07:57Z

src/ekore/anomalous_dimensions/unpolarized/time_like/as1.py

+    return result
+
+@nb.njit(cache=True)
+def gamma_gg(N, nf, cache, is_singlet=None):


Suggested change

def gamma_gg(N, nf, cache, is_singlet=None):

def gamma_gg(N, nf, cache):

you know you are in the singlet case here - so no need for the argument (this is of course also true for all other functions)

Yes, I know that, however in #202, I have modified the cache.get to take is_singlet as an argument as well. Since that function is called in gamma_qq and gamma_gg for the s1, I just set the is_singlet to None to ensure no missing arguments are there.

Indeed I will have to rebase this branch, and that's why I have been trying to complete the #202 to have it fully working.

Yes, I know that, however in #202, I have modified the cache.get to take is_singlet as an argument as well. Since that function is called in gamma_qq and gamma_gg for the s1, I just set the is_singlet to None to ensure no missing arguments are there.

mh? how you mean? yes, you need to pass the argument to get, but this does not mean it has to be an argument for gg - on the contrary, gg enforces the argument to be True (forever) - or am I missing something?

Indeed I will have to rebase this branch, and that's why I have been trying to complete the #202 to have it fully working.

I don't expect #202 to be closed any time soon ... so you should keep #202 working (as it was) and gradually increase the code basis whenever you make a new step

mh? how you mean? yes, you need to pass the argument to get, but this does not mean it has to be an argument for gg - on the contrary, gg enforces the argument to be True (forever) - or am I missing something?

It does not enforce to to be True. It sets it to be None, and then when the gamma_singlet is defined, it is then set to True, as it should be. I did this as a way to keep the code consistent across all orders. A little more than needed, I agree, but cleaner imho.

I still don't see this point (and I forgot to ask during the meeting) - @giacomomagni may I ask you for a comment?

well, its just a matter of code consistency across perturbative orders. At LO, in principal, I could hard code the cache.get to have is_singlet = True but, this way, the code is consistent such that every splitting function that uses cache, needs an argument is_singlet, and this is specified in the init file making it very consistent.

Sorry I'm a bit confused... I need to check #202. But I believe a function called gamma_gg can only be singlet,
so the argument should not be need otherwise it's really confusing. Instead is_singlet should be enforced in all the harmonics called inside this function.

felixhekhorn · 2023-02-01T10:11:41Z

Also please remember to install pre-commit ... (the longer you delay the more mess it will create 🙃 )

t7phy · 2023-02-01T10:25:04Z

Also please remember to install pre-commit ... (the longer you delay the more mess it will create upside_down_face )

I did install it, but there seems to be some installation issues with a few of the hooks, and I couldn't find a solution to fix them quickly. I will try to figure it out.

felixhekhorn · 2023-02-09T15:46:20Z

src/ekore/anomalous_dimensions/polarized/time_like/__init__.py

@@ -0,0 +1,21 @@
+r"""The polarized, time-like Altarelli-Parisi splitting kernels.


this file is deleted in master (likewise for OME)

felixhekhorn · 2023-02-09T15:52:28Z

src/ekore/anomalous_dimensions/unpolarized/time_like/as1.py

+        :math:`\gamma_{qq}^{(0)}(N)`
+
+    """
+    s1 = w1.S1(N)  # c.get(c.S1, cache, N, is_singlet)


If you wish you can immediately step into #202 , i.e. base this branch on top of the other

I have eliminated the conflicts from this branch for the moment, perhaps I should eliminate them from #202 as well before this step

giacomomagni

Hi @t7phy I left some minor comments, mainly about doc strings.
Looks quite good!

I suggest you, to make a decision for the harmonics, otherwise it's confusing:

either you stick to the old notation and you aim to close this before Caching harmonic sums #202.
either you base this on Caching harmonic sums #202, having in mind that that has to be merged before closing this. (PS I'll try to give an hand on the other one, if I have time :) )

giacomomagni · 2023-02-09T16:38:41Z

src/ekore/anomalous_dimensions/unpolarized/time_like/__init__.py

+
+    See Also
+    --------
+    ekore.anomalous_dimensions.unpolarized.time_like.as1.gamma_ns : :math:`\gamma_{ns}^{(0)}(N)`


I know we have this see also section in the corresponding unpolarised space like, but I would remove it everywhere
and not repeat it. It was added at the beginning of the project, I believe. I find it not useful anymore...

giacomomagni · 2023-02-09T16:41:12Z

src/ekore/anomalous_dimensions/unpolarized/time_like/as1.py

@@ -0,0 +1,175 @@
+"""The unpolarized LO Altarelli-Parisi splitting kernels."""


Suggested change

"""The unpolarized LO Altarelli-Parisi splitting kernels."""

"""The unpolarized |LO| Altarelli-Parisi splitting kernels."""

same for |NLO| and |NNLO| in all the docstring.

giacomomagni · 2023-02-09T16:45:59Z

src/ekore/anomalous_dimensions/unpolarized/time_like/as1.py

+    return result
+
+@nb.njit(cache=True)
+def gamma_gg(N, nf, cache, is_singlet=None):


Sorry I'm a bit confused... I need to check #202. But I believe a function called gamma_gg can only be singlet,
so the argument should not be need otherwise it's really confusing. Instead is_singlet should be enforced in all the harmonics called inside this function.

giacomomagni · 2023-02-09T16:48:11Z

src/ekore/anomalous_dimensions/unpolarized/time_like/as2.py

+        m1tpN = 1
+    elif is_singlet == False:
+        m1tpN = -1
+    else:


I think this case with npp(-1,N) should not appear. A quantity is either continued with m1tpN=1 or m1tpN=-1.

giacomomagni · 2023-02-09T16:50:23Z

src/ekore/anomalous_dimensions/unpolarized/time_like/as2.py

+
+
+@nb.njit(cache=True)
+def gamma_qqs(N, nf):


I know this is not consistent everywhere, but we usually call this gamma_ps or gamma_qq_ps.
@felixhekhorn we should make these name more uniform.

giacomomagni · 2023-02-09T16:53:03Z

src/ekore/anomalous_dimensions/unpolarized/time_like/as3.py

+@nb.njit(cache=True)
+def gamma_nsp(N, nf, cache, is_singlet=None):
+    r"""Compute the NNLO non-singlet positive anomalous dimension.
+


Remember to add the reference to the source code / publication in all these functions.

alecandido

Mainly tiny comments, but here benchmarks are very reduced (maybe only APFEL available? Consider adding at least APFEL++ then) and no tests provided.

I got this is not yet depending on #202, so, if possible, I would push to test and document a working LO, and only push for NLO and NNLO in further PRs.
It's not very useful to implement a large bunch of code without any intermediate test, and if we keep for long as a branch we (i.e. the branch author) will pay over and over the price of merging/rebasing on master.

alecandido · 2023-02-10T14:55:54Z

src/eko/evolution_operator/grid.py

@@ -106,6 +106,8 @@ def __init__(
        if order == (1, 0) and method != "iterate-exact":
            logger.warning("Evolution: In LO we use the exact solution always!")

+        logger.info(("Time" if configs.time_like else "Space") + "-like evolution")


As I proposed for polarized, let's make this machine-friendly:

Suggested change

logger.info(("Time" if configs.time_like else "Space") + "-like evolution")

logger.info(dict(evol_kind="time" if configs.time_like else "space"))

or, definitely simpler, not sure if better:

Suggested change

logger.info(("Time" if configs.time_like else "Space") + "-like evolution")

logger.info(dict(time_like=configs.time_like))

@felixhekhorn ?

let's follow #195 - so the latter

alecandido · 2023-02-10T14:56:20Z

src/eko/evolution_operator/operator_matrix_element.py

@@ -36,7 +36,6 @@ def build_ome(A, matching_order, a_s, backward_method):
    -------
    ome : numpy.ndarray
        matching operator matrix
-


Keep these lines, they are part of our docstrings format.

alecandido · 2023-02-10T14:58:33Z

src/ekore/anomalous_dimensions/unpolarized/time_like/__init__.py

+
+    See Also
+    --------
+    ekore.anomalous_dimensions.unpolarized.time_like.as1.gamma_ns : :math:`\gamma_{ns}^{(0)}(N)`


alecandido · 2023-02-10T15:00:27Z

src/ekore/anomalous_dimensions/unpolarized/time_like/as1.py

+def gamma_qq(N, cache, is_singlet=None):
+    r"""Compute the LO quark-quark anomalous dimension.
+
+    Implements Eqn. (B.3) from :cite:`Mitov:2006wy`.


Suggested change

Implements Eqn. (B.3) from :cite:`Mitov:2006wy`.

Implements :eqref:`B.3` from :cite:`Mitov:2006wy`.

see:

eko/doc/source/shared/roles.rst

Lines 76 to 77 in 13197aa

.. role:: eqref

:class: eqref

eko/src/eko/msbar_masses.py

Line 223 in 13197aa

:cite:`Liu:2015fxa` :eqref:`5`.

alecandido · 2023-02-10T15:03:02Z

tests/ekore/anomalous_dimensions/unpolarized/time_like/test_init.py

@@ -2,9 +2,10 @@

 import ekore.anomalous_dimensions.unpolarized.time_like as ad_ut

+# tests to be added later


Better now than later

alecandido · 2023-03-25T15:31:15Z

src/ekore/anomalous_dimensions/unpolarized/time_like/as2.py

+import numpy as np
+
+# from eko.constants import zeta2
+from numpy import power as npp


Python has a built-in power operator, i.e. **. Numba supports it, as much as numpy.power, so you will get the same code.

Please, favor built-ins over third-party extension functions (including NumPy).

alecandido · 2023-03-25T15:32:29Z

src/ekore/anomalous_dimensions/unpolarized/time_like/as2.py

+
+
+@nb.njit(cache=True)
+def gamma_nsp(N, nf, cache=None, is_singlet=None):


Suggested change

def gamma_nsp(N, nf, cache=None, is_singlet=None):

def gamma_nsp(N, nf, cache=None, is_singlet=False):

is_singlet has to be a boolean, you're also declaring as such in the docstring. Refrain from assign a value of a different type.

alecandido · 2023-03-25T15:33:47Z

src/ekore/anomalous_dimensions/unpolarized/time_like/as2.py

+    s2 = w2.S2(N)  # c.get(c.S2, cache, N, is_singlet)
+    sm2 = w2.Sm2(N, s2, True)  # c.get(c.Sm2, cache, N, is_singlet)
+
+    m1tpN = 1


Since it's 1, you can also drop it. I hope it will be optimized away, but at least it would uselessly increase compile-time.

alecandido · 2023-03-25T15:34:56Z

src/ekore/anomalous_dimensions/unpolarized/time_like/as2.py

+                    + 1152 * zeta2 * npp(N, 2)
+                    + 1248 * zeta2 * npp(N, 3)
+                    - 1296 * zeta2 * npp(N, 4)
+                    - 792 * zeta2 * npp(N, 5)
+                    + 876 * zeta2 * npp(N, 6)
+                    - 1368 * zeta2 * npp(N, 7)
+                    - 2340 * zeta2 * npp(N, 8)
+                    + 120 * zeta2 * npp(N, 9)
+                    + 1476 * zeta2 * npp(N, 10)
+                    + 792 * zeta2 * npp(N, 11)
+                    + 132 * zeta2 * npp(N, 12)


Collect things to make them readable and more efficient. Mathematica can help you to do this, if you need.

alecandido · 2023-03-25T15:37:07Z

src/ekore/anomalous_dimensions/unpolarized/time_like/as3.py

+
+
+@nb.njit(cache=True)
+def gamma_singlet(N, nf, cache, is_singlet=True):


Why do you need an is_singlet argument? It is written in the name of the function that is a singlet.

This is an old, not used branch, I have left it still just for reference. This PR should also be closed. The branch tlu_ad_v2 is the place with actual implementation that is ready.

alecandido · 2023-03-25T15:38:15Z

src/ekore/anomalous_dimensions/unpolarized/time_like/as3.py

+    is_singlet : boolean
+        True for singlet, False for non-singlet, None otherwise


If it were actually three-valued, you should have written:

Suggested change

is_singlet : boolean

True for singlet, False for non-singlet, None otherwise

is_singlet : boolean or None

True for singlet, False for non-singlet, None otherwise

But, as @giacomomagni wrote above, None makes no sense, and, as I wrote above, here even False makes no sense: it is a singlet quantity.

alecandido · 2023-03-25T15:49:49Z

@t7phy, re #200 (comment), please create a new PR for the new one:
https://github.com/NNPDF/eko/compare/tlu_ad_v2
Without a PR is difficult to track progresses for us, and to help you, looking at the code status. The PR interface provides a useful interface to discuss code and track conversations, that's why we use it a lot :)

As soon as you will have done, I'll close this (you can do it yourself, if you wish).

(This commit 7307c6e happened 10 days ago, and missing another PR I thought this was the branch.)

t7phy · 2023-03-25T23:15:35Z

closed in favor of #232

t7phy added 6 commits January 14, 2023 13:54

fc

d434cd7

Merge remote-tracking branch 'origin/split-math-in-module' into time_…

5a85aec

…like_unpolarized

leading order

8df49c0

leading order

78a0739

leading order

6d31c8b

leading order

86614dc

t7phy requested review from felixhekhorn and enocera January 20, 2023 01:22

alecandido reviewed Jan 20, 2023

View reviewed changes

t7phy added 2 commits January 20, 2023 10:58

formatting changes for consistency

d995a0e

formatting changes for consistency 2

7f0ee3f

felixhekhorn reviewed Jan 23, 2023

View reviewed changes

src/ekore/anomalous_dimensions/unpolarized/time_like/as1.py Outdated Show resolved Hide resolved

felixhekhorn added enhancement New feature or request physics new physics features labels Jan 23, 2023

t7phy added 2 commits January 27, 2023 03:55

NLO ADs, citations and formatting

5ccbcd5

code formatting

3016d63

alecandido force-pushed the split-math-in-module branch from 4c72a96 to 07aa331 Compare January 30, 2023 16:09

Base automatically changed from split-math-in-module to master January 30, 2023 16:25

t7phy added 3 commits January 30, 2023 20:46

code cosnsistency

569c615

update import

0b93753

consistency with new harmonic sums cache

fe0a49a

felixhekhorn reviewed Feb 1, 2023

View reviewed changes

Update pre-commit hooks

98b3e24

felixhekhorn mentioned this pull request Feb 8, 2023

Caching harmonic sums #202

Closed

1 task

t7phy added 7 commits February 9, 2023 02:28

time-like NNLO ADs

9558e22

pylint test

53356ae

pylint test 2

561648e

time_like ADs init file

2d8bede

LO benchmark

6bdd487

merge master to current this branch

b4e8bb9

removal of tests with NotImplementedError

feb2dce

felixhekhorn reviewed Feb 9, 2023

View reviewed changes

giacomomagni reviewed Feb 9, 2023

View reviewed changes

alecandido reviewed Feb 10, 2023

View reviewed changes

backup commit

7307c6e

alecandido reviewed Mar 25, 2023

View reviewed changes

t7phy closed this Mar 25, 2023

t7phy reopened this Mar 25, 2023

t7phy closed this Mar 25, 2023

giacomomagni deleted the time_like_unpolarized branch March 30, 2023 12:24

		@@ -1,21 +0,0 @@
		r"""The unpolarized, time-like Altarelli-Parisi splitting kernels.

	def gamma_gg(N, nf, cache, is_singlet=None):
	def gamma_gg(N, nf, cache):

		@@ -0,0 +1,21 @@
		r"""The polarized, time-like Altarelli-Parisi splitting kernels.

		@@ -0,0 +1,175 @@
		"""The unpolarized LO Altarelli-Parisi splitting kernels."""

	"""The unpolarized LO Altarelli-Parisi splitting kernels."""
	"""The unpolarized \|LO\| Altarelli-Parisi splitting kernels."""

	logger.info(("Time" if configs.time_like else "Space") + "-like evolution")
	logger.info(dict(evol_kind="time" if configs.time_like else "space"))

	logger.info(("Time" if configs.time_like else "Space") + "-like evolution")
	logger.info(dict(time_like=configs.time_like))

	Implements Eqn. (B.3) from :cite:`Mitov:2006wy`.
	Implements :eqref:`B.3` from :cite:`Mitov:2006wy`.

		@@ -2,9 +2,10 @@

		import ekore.anomalous_dimensions.unpolarized.time_like as ad_ut

		# tests to be added later



		@nb.njit(cache=True)
		def gamma_nsp(N, nf, cache=None, is_singlet=None):



		@nb.njit(cache=True)
		def gamma_singlet(N, nf, cache, is_singlet=True):

		is_singlet : boolean
		True for singlet, False for non-singlet, None otherwise

Time-like Anomalous dimensions #200

Time-like Anomalous dimensions #200

Conversation

t7phy commented Jan 20, 2023

alecandido left a comment

Choose a reason for hiding this comment

alecandido commented Jan 20, 2023

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

t7phy Jan 23, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

felixhekhorn commented Feb 1, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

felixhekhorn commented Feb 1, 2023

t7phy commented Feb 1, 2023

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

giacomomagni left a comment • edited Loading

Choose a reason for hiding this comment

giacomomagni Feb 9, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

giacomomagni Feb 9, 2023 • edited Loading

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

alecandido left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

alecandido commented Mar 25, 2023 • edited Loading

t7phy commented Mar 25, 2023

t7phy Jan 23, 2023 •

edited

Loading

felixhekhorn commented Feb 1, 2023 •

edited

Loading

giacomomagni left a comment •

edited

Loading

giacomomagni Feb 9, 2023 •

edited

Loading

giacomomagni Feb 9, 2023 •

edited

Loading

alecandido commented Mar 25, 2023 •

edited

Loading