Feature/baryons bhm fedeli by nikosarcevic · Pull Request #1277 · LSSTDESC/CCL

nikosarcevic · 2026-01-25T00:05:04Z

This PR adds the Fedeli 2014 baryon halo model to CCL and integrates it into the baryons module.

Included are:

a full fedeli14 baryon halo–model implementation under
pyccl/baryons/fedeli14_bhm/, including:
- numerics/utilities (numerics.py)
- mass fraction parametrizations (mas_fractions.py)
- halo profiles (halo_profiles.py)
- profile interpolation (profile_interpolation.py)
- halo-model power spectrum calculation (power_spectra.py(
- These components are wrapped in a central BaryonHaloModel class (baryon_halo_model.py)
a public-facing baryons API in
pyccl/baryons/fedeli14.py, exposing fedeli14 as a multiplicative baryonic
boost that returns baryon-modified power spectra
Advanced users can directly instantiate and customize the
BaryonHaloModel to experiment with halo ingredients and internal model
choices.
Example notebooks (under fedeli14_bhm/examples/) demonstrating:
- direct use of BaryonHaloModel
- comparison with other baryonic prescriptions in CCL.

I also include unit tests covering each fedeli14 component/script.

Note that:

implementation follows the existing CCL baryons architecture
linting is locally healthy
all tests are passing (only testing the added test pytest pyccl/tests/test_baryons_fedeli14*)

this pr is built off of current ccl master

this closes #1276

coveralls · 2026-01-25T03:01:34Z

Coverage Report for CI Build 24458542966

Coverage increased (+0.3%) to 97.783%

Details

Coverage increased (+0.3%) from the base build.
Patch coverage: 4 uncovered changes across 3 files (1047 of 1051 lines covered, 99.62%).
No coverage regressions found.

Uncovered Changes

File	Changed	Covered	%
pyccl/baryons/fedeli14_bhm/halo_profiles.py	140	138	98.57%
pyccl/baryons/fedeli14_bhm/baryon_halo_model.py	174	173	99.43%
pyccl/baryons/fedeli14_bhm/mass_fractions.py	99	98	98.99%

Coverage Regressions

No coverage regressions found.

Coverage Stats


Relevant Lines:	7847
Covered Lines:	7673
Line Coverage:	97.78%
Coverage Strength:	4.89 hits per line

💛 - Coveralls

carlosggarcia

Thanks for the work, @nikosarcevic. This is great!

This is a massive PR so I think it might be better if we do multiple iterations over a few files. I've started with the API ones. When we settle on those, I will go into the files with the actual implementation.

A few comments:

The CI is failing because the coverage level of the code has gone down. You need to write more unit tests to test all new additions to the code; i.e. as many lines as possible, which is what coveralls checks.
I think the policy is to put the notebooks in the CCLX repo. So you should move them from this PR.
If you come up with a way of making this PR shorter, that will make our lives easier and the lives of those who have to look into this in the future!

carlosggarcia · 2026-02-08T22:19:59Z

+    The operation applied is:
+
+    .. math::
+        P_{\rm out}(k,a) = P_{\rm in}(k,a)\,B_{\rm fedeli}(k,a).


Change B -> f, for consistency with the other modules.

done in c5812c5

(sorry about CI failing, am trying to fix it, i tihnk it is osx-arm64 standard nightmare suddenly)

carlosggarcia · 2026-02-08T22:24:25Z

+        - If ``renormalize_large_scales=True``, the boost is rescaled (per
+          scale factor) so that its mean over ``k <= k_renorm_max`` is unity.
+        - For ``a < a_min`` (currently hard-coded to 0.1 in the application
+          step), the boost is forced to unity.


Is this also enforced in Fedeli+2014 or is this an addition?

this is an addition in the wrapper

fedeli4 builds the baryonic halo model so that the large-scale limit is physically correct (e.g. the 2halo term recovers linaer theory and the k->0 limits recover the correct mass fractions), but it does not prescribe an explicit renormalization of a boost ratio to unity over a finite low-k range

the `renormalize_large_scales`` option is basically a numerical safeguard to ensure the ratio f(k,a) -> 1 on large scales when dividing by a chosen reference spectrum, given finite grids and interpolation effects.
Elisa and I had this problem on large scales and this is a solution I found.
in a similar fashion, forcing the boost to unity for a < 0.1 is bascially a defensive cutoff to avoid extrapolating the model outside its intended regime and is not also explicitly imposed in fedeli14

carlosggarcia · 2026-02-08T22:33:02Z

+                "star_x_delta", "star_alpha", "density_mmin", "density_mmax",
+            }:
+                val = float(val)
+            if key in {"n_m"}:


carlosggarcia · 2026-02-08T22:43:52Z

+            k: float | FloatArray,
+            a: float | FloatArray
+    ) -> float | FloatArray:
+        r"""Evaluate the Fedeli14 boost factor ``B(k,a)``.


Same as above B(k,a) -> f(k,a)

carlosggarcia · 2026-02-08T22:50:37Z

+        for i, aval in enumerate(a_use[:, 0]):
+            out[i, :] = bhm.boost(k=k_1d, a=float(aval), pk_ref=self.pk_ref)
+
+        # We need to enforce B(k->0)=1 by renormalizing on large scales.


carlosggarcia · 2026-02-09T02:33:22Z

+    per-a."""
+    b = BaryonsFedeli14(renormalize_large_scales=False)
+
+    calls: list[tuple[float, np.ndarray, str]] = []


debugging line?

tests now updated in 2ea9b55

carlosggarcia · 2026-02-09T02:37:46Z

@@ -0,0 +1,312 @@
+"""Unit tests for `pyccl.baryons.fedeli14`."""
+


Two general comments:

I think it's clever the way you use monkeypatch. I'd try to define a single DummyBHM and use the same for all tests. That would shorten the file and make it a bit easier to read, so that I don't have to re-think in each test what this dummy class is doing.

Although I guess you will test the output of the actual Fidelly14 boost computation. I think there is value if you add a single test that actually test that you get the right boost, without DummyBHM class.

done!

2ea9b55

68971d8

carlosggarcia · 2026-02-09T02:43:03Z

+
+    # sanity: grids unchanged
+    a0, lk0, _ = pk.get_spline_arrays()
+    assert np.allclose(a_arr, a0)


I think it's more explicit to use assert a0 == pytest.approx(a_arr, rel=1e-5). Numpy can change the relative/absolute tolerances and break all our tests.

good point! done in c4837d4

carlosggarcia · 2026-02-09T02:43:58Z

+    a0, lk0, _ = pk.get_spline_arrays()
+    assert np.allclose(a_arr, a0)
+    assert np.allclose(k_arr, np.exp(lk0))
+


All the lines below are a copy of the above?

This is no longer duplicated after the latest test cleanup. I kept one explicit grid-preservation check in test_incl_bary_eff_applies_unity_for_early_times_and_outside_k_support, and changed it to pytest.approx(..., rel=1e-5) as suggested.

carlosggarcia · 2026-02-09T02:45:11Z

+    assert np.allclose(pk_out[1:, :], 20.0)   # boosted by 10 for a>=0.1
+
+
+def test_incl_bary_eff_fixes_absurd_log_flag_data(


Not needed. Remove.

nikosarcevic · 2026-02-09T13:53:01Z

@carlosggarcia thanks for the feedback.
I have appended more tests to increase coverage

also fixed typing in 1638b4e

nikosarcevic · 2026-04-15T15:03:09Z

hey @carlosggarcia you I have addressed all the comments except moving the exmaples outside. I woudl like to keep the examples until yu are satisfied with the src code because I use those to check if the changes are correct in terms of what comes out in the plots. After you are ok with the PR, i will take the examples outside and add them to CCLX. Woud lthat owrk for you? I can also not commit them so you do not have the overhead.

nikosarcevic requested review from chrgeorgiou and elisachisari January 25, 2026 00:05

nikosarcevic added the baryons label Jan 25, 2026

nikosarcevic force-pushed the feature/baryons_bhm_fedeli branch from a07e69b to 423171e Compare January 25, 2026 03:17

nikosarcevic added the API label Jan 26, 2026

carlosggarcia requested changes Feb 9, 2026

View reviewed changes

nikosarcevic added 14 commits April 15, 2026 07:15

Add Fedeli14 baryon halo model module

00a6686

Add unit tests for Fedeli14 BHM

c650218

added more tests to fedeli14 pk to increase coverage

10605fb

added more tests to baryon halo model

71752f1

Fix Fedeli14 halo profile scalar return types

041a4bf

updated tests for halo profiles

4aa676b

udpated tets for profile interp

fa509ab

added tests for numerics

b11acc0

added tets for fedeli main top api module

57724ef

fixed linting in tests

69de3a4

changed boost B symbol to f in main api fedeli to keep CCL convention

087b776

added inline comment to explain norm per a

de91bf5

improved explanation in include barony effect docstring/behaviour

f706b6d

fixed if elif in fedeli class

c944831

nikosarcevic force-pushed the feature/baryons_bhm_fedeli branch from a38fce6 to c944831 Compare April 15, 2026 11:25

nikosarcevic added 5 commits April 15, 2026 08:29

updated fedeli14 as per reviewer comments

13831e9

removed stale test for fedeli14; updated docstring in test

3ccdc67

further udpated test_baryons_fedeli14 per carlos comments

2ea9b55

appended another test to test actual fedeli

68971d8

Use explicit pytest tolerances in Fedeli14 tests

c4837d4

		@@ -0,0 +1,312 @@
		"""Unit tests for `pyccl.baryons.fedeli14`."""

		assert np.allclose(pk_out[1:, :], 20.0) # boosted by 10 for a>=0.1


		def test_incl_bary_eff_fixes_absurd_log_flag_data(

Conversation

nikosarcevic commented Jan 25, 2026

Uh oh!

coveralls commented Jan 25, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Coverage Report for CI Build 24458542966

Coverage increased (+0.3%) to 97.783%

Details

Uncovered Changes

Coverage Regressions

Coverage Stats

💛 - Coveralls

Uh oh!

carlosggarcia left a comment

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

nikosarcevic commented Feb 9, 2026

Uh oh!

nikosarcevic commented Apr 15, 2026 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

Reviewers

Assignees

Labels

Projects

Milestone

Development

Uh oh!

3 participants

coveralls commented Jan 25, 2026 •

edited

Loading

nikosarcevic commented Apr 15, 2026 •

edited

Loading