Big PR about using uT as fit axis, and developments to fit the W boson width (partially based on the other changes) by cippy · Pull Request #678 · WMass/WRemnants

cippy · 2026-03-25T11:09:04Z

This PR is supposed to supersede the previous PR #643

It does many things, but doesn't modify the default analysis behaviour. A short executive summary is the following

extension of the ABCD method to work with the uT axis
developments to fit the W width, either standalone or simultaneously with mass
customisation of some systematic uncertainties related to the recoil (optional)
several updates in plotting scripts

… analysis bins of specific axes (e.g. utAngleSign) which have identically zero yield in the A-Ax-B-Bx regions (cherry picked from commit 773c1d9)

… analysis bins of specific axes (e.g. utAngleSign) which have identically zero yield in the A-Ax-B-Bx regions

…current version of Rabbit, it will be probably reverted when the definitions of poi/noi/nuisanceNotConstrained are fixed

…ion on uT>0 with an hardcoded factor (normalization)

…ream_cherrypick Merge from WMass/main

…th a pT dependent shape

…ream_cherrypick Merge from WMass/main

…herrypick Merge from my branch

…ream_cherrypick Merging from WMass/Main

…ream_devel_150326 Merging from WMass/main

kdlong · 2026-03-25T12:40:06Z

scripts/rabbit/setupRabbit.py

+                if args.fakeTransferAxis in args.fakerateAxes
+                else ""
+            ),
+            fakeTransferCorrFileName=args.fakeTransferCorrFileName,


What does this actually do? It reads another file that used a looser selection, or something?

fakeTransferCorrFileName contains the correction needed for the prediction of the fakes with negative uT based on the one from high uT. It is needed because right now the low mT region for negative uT has no events because of kinematics, so the ABCD method cannot predict the background in the high mT region there. The correction is basically the ratio of the data yields between the two uT regions as obtained from the secondary vertices control region, subtracting the small prompt component. Then we also propagate uncertainties based on non closures between control and signal region, and data and QCD MC. It is essentially what Ruben presented during the workshop in Pisa.

kdlong · 2026-03-25T12:40:50Z

scripts/rabbit/setupRabbit.py

-            systAxes=["width"],
-            systNameReplace=[["2p09053GeV", "Down"], ["2p09173GeV", "Up"]],
-            passToFakes=passSystToFakes,
+        widthVarTag = ""


Maybe the width related stuff should go in rabbit_helpers since it's gotten so long

Yes, I can try to move there as many things as possible

kdlong · 2026-03-25T12:43:21Z

scripts/rabbit/setupRabbit.py

+                    labelsByAxis=["varTF", "downUpVar"],
+                )
+
+                ## syst for transfer factor difference between data and QCD MC in control region


Delete commented out stuff (here and other places)

This particular item is one thing we are still finalising, but I guess I can remove it for now.
Other places with commented code were in some cases non added by this PR, so I am not sure what to do with them.

Done for now, it will be added back in the near future

kdlong · 2026-03-25T12:43:52Z

scripts/rabbit/setupRabbit.py

+                    actionArgs=dict(
+                        altHistName="fakeCorr_closQCDsignal",
+                        varIdxs=[],
+                        correctionFile=f"{common.data_dir}/fakesWmass/{args.fakeTransferCorrFileName}.pkl.lz4",


Can this not be computed on the fly? Where does it come from?

Right now it cannot, because it makes use of the control region, so it is supposed to be produced in advance (the files will need to be stored in wremnants-data).
However, we are considering modifying the procedure so to derive the correction directly from the signal region using the QCD MC, rather than from data from the control region, to avoid depending too much on the latter. However, systematic uncertainties would still need to be produced using the control region where everything is validated, so they should be available as external inputs anyway.

bendavid · 2026-03-25T12:48:13Z

What actually needed to be changed to fit the W width? This was in principle already working.

cippy · 2026-03-25T14:08:05Z

What actually needed to be changed to fit the W width? This was in principle already working.

I made the choice of the prefit variation more flexible, through options. Then I also added new options to perform the same studies as we did for mW, such as decorrelation by variable or fitting the width difference between charges.
There were also some minor things related to calling addSystematic, some arguments were propagated as if one was measuring the mass rather than the width.

cippy · 2026-03-26T11:24:31Z

Last commit moved the definition of the mass and width systs from setupRabbit.py to rabbit_helpers.py, also fixing a couple of bugs/inconsistencies in their definition.

…ream_devel_150326 Merging from WMass/main

davidwalter2 · 2026-03-27T12:35:11Z

scripts/histmakers/mw_with_mu_eta_pt.py

There is a lot of stuff added, new columns and histograms even if one runs without any argument. Can those things that are only needed for tests be put into a block to be disabled by default e.g. using "--auxiliaryHistograms" or a new flag like "--addHistsForMETStudies" or so

Yes, good point, I'll do that.

davidwalter2 · 2026-03-27T12:45:26Z

wremnants/postprocessing/rabbit_helpers.py

@@ -36,14 +133,14 @@ def add_mass_diff_variations(
    )
    # mass difference by swapping the +50MeV with the -50MeV variations for half of the bins
    args = ["massShift", f"massShift{label}50MeVUp", f"massShift{label}50MeVDown"]
-    if mass_diff_var == "charge":
+    if any(mass_diff_var == var for var in ["charge", "utAngleSign"]):


if mass_diff_var in ["charge", "utAngleSign"]:

I don't know why I did like that, silly me :)

davidwalter2 · 2026-03-27T12:47:44Z

wremnants/production/include/recoil_tools.hpp

@@ -40,10 +40,13 @@ TVector2 get_z_mom(const float pt1, const float phi1, const float pt2,
  TVector2 l1 = TVector2();
  l1.SetMagPhi(pt1, phi1);

-  TVector2 l2 = TVector2();
-  l2.SetMagPhi(pt2, phi2);
+  // TVector2 l2 = TVector2();


remove comments

davidwalter2 · 2026-03-27T12:49:23Z

wremnants/utilities/styles/styles.py

@@ -166,6 +176,8 @@ def translate_html_to_latex(n):
    "MET_pt": {"label": r"$\mathit{p}_{\mathrm{T}}^{miss}$", "unit": "GeV"},
    "MET": {"label": r"$\mathit{p}_{\mathrm{T}}^{miss}$", "unit": "GeV"},
    "met": {"label": r"$\mathit{p}_{\mathrm{T}}^{miss}$", "unit": "GeV"},
+    # "mt": {"label": r"$\mathit{m}_{T}^{\mu,MET}$", "unit": "GeV"},


davidwalter2

Review of the non-notebook, non-analysisTools code changes. The uT-axis ABCD extension and W width fitting additions are clearly useful. A few comments below.

davidwalter2 · 2026-03-27T13:23:45Z

wremnants/postprocessing/histselections.py

+            trTensorPath = (
+                f"{data_dir}/fakesWmass/{self.fakeTransferCorrFileName}.pkl.lz4"
+            )
+            logger.warning(f"Loaded transfer tensor for fakes: {trTensorPath}")


This should be logger.info rather than logger.warning — loading a file successfully is not a warning situation. Same for line 479 (self.fakerate_integration_axes = ...), which is purely informational.

davidwalter2 · 2026-03-27T13:23:45Z

wremnants/postprocessing/histselections.py

+
+                else:
+                    logger.debug(
+                        f"All ABCD values are zeros! Returning Fake estimate based on other bin, with an HARDCODED norm. factor"


The comment says "HARDCODED norm. factor" but no explicit normalization factor is actually applied here — the fallback uses the complementary bin's smoothed spectrum multiplied by self.fakeTransferTensor. Could you clarify whether a normalization is intentionally missing, or update the comment to accurately describe what is being done?

I think it was an old message when we were only scaling the normalization of the prediction taken from the other bin, while now we propagate an actual shape (also with a different normalization) with some uncertainties. I'll change the message

davidwalter2 · 2026-03-27T13:23:45Z

wremnants/production/include/utils.hpp

+
+  // use scale=1.1 and smear=0.2 for 10% larger mean value and 20% resolution
+  // smearing
+  std::seed_seq seq{std::size_t(run), std::size_t(lumi), std::size_t(event)};


Each call to get_scaled_smeared_variable within the same event uses identical seeds (run, lumi, event), so smearMET_pt and smearMET_phi will draw from the same RNG state and be 100% correlated. Consider adding a call-site tag (e.g. a string hash or integer constant) to the seed_seq to ensure independence between the three smear/scale variations.

Thank you, I didn't realise it

davidwalter2 · 2026-03-27T13:23:45Z

wremnants/production/recoil_tools.py

@@ -1460,7 +1463,8 @@ def add_recoil_unc_W(
        return df

    def setup_recoil_Z_unc(self):
-        if not self.dataset.name in self.datasets_to_apply or not self.storeHists:
+        # if not self.dataset.name in self.datasets_to_apply or not self.storeHists:


Left-over commented code. If the self.storeHists guard is no longer needed, please remove the commented-out line rather than leaving it.

davidwalter2 · 2026-03-27T13:23:45Z

scripts/histmakers/mw_with_mu_eta_pt.py

+    # to define dedicated systematics for scale/smearing of met_pt
+    df = df.Define(
+        "scaleMET_pt",
+        "wrem::get_scaled_smeared_variable(run, luminosityBlock, event, MET_corr_rec_pt, 1.01, 0.0)",


The scale/smear values (1% scale, 5% smear for pt and phi) look like temporary test values, especially since in setupRabbit.py the phi smear is further rescaled by 0.4 with a # for test comment. What are the physics-motivated values for these? A TODO referencing the derivation would help.

I am still doing some studies to estimate reasonable values to use. I ended up with these based on the shape variations, but I actually believe they are still overestimated, because the uncertainty applied to MC should reflect the residual data/MC difference after calibration, and this is probably of order 1% or less (otherwise extracting mW from mT should be hopeless).

davidwalter2 · 2026-03-27T13:23:45Z

scripts/rabbit/setupRabbit.py

+            "eta-sign",
+            "eta-range",
+        ],
+        help="For use with --noi widthDiffW, select the variable to define the different mass differences",


Copy-paste typo in the help string: "define the different mass differences" should be "define the different width differences".

davidwalter2 · 2026-03-27T13:23:45Z

scripts/rabbit/setupRabbit.py

+                "expNoLumi",
+                "expNoCalib",
+            ],
+            scale=0.4,  # from 5% -> scale * 5% for test (see histmaker)


The scale=0.1 and scale=0.5 are commented out for scaleMET_pt and smearMET_pt, while smearMET_phi has an active scale=0.4 with a # for test comment. This inconsistency suggests the custom recoil syst block is not yet finalized. Are these ready to merge, or should this remain clearly marked as experimental?

I can probably keep the values I tested so far as the baseline, even though they might be optimised further.

davidwalter2 · 2026-03-27T13:23:45Z

wremnants/postprocessing/syst_tools.py

+    variation_size=0.1,
+    keepConstantAxisBin={},
+    fakeselector=None,
+    *args,


Having *args between keyword arguments with defaults (keepConstantAxisBin, fakeselector) and **kwargs is fragile — any extra positional argument would be silently captured by *args and ignored. Unless forwarding positional args to fakeselector.get_hist is intentional (which it appears to be from line 990), please add a comment clarifying this, or restructure to avoid the ambiguity.

…ream_devel_150326 Merge from WMass/main

cippy · 2026-03-28T17:36:49Z

Ok, this should be ready, assuming I didn't forget to implement any single comment that I received.
CI looks good, its the usual small difference in he BSM workflow.

For the seeding of the smearing, I could have implemented a more general solution to automatically increment the counter for the seed generation after each function call, but it was simpler to pass an integer directly, since the function is currently called only twice (actually three times, but the first one doesn't use the random number generator) and the manual setting allows the user to use the same number if ever needed.

cippy and others added 30 commits October 17, 2025 15:44

generalize eta labels better for plot_decorr_params

fc992c4

plotting and testing

d330024

Protection on the fakes estimation: fake yield is set to zero for the…

09c630f

… analysis bins of specific axes (e.g. utAngleSign) which have identically zero yield in the A-Ax-B-Bx regions (cherry picked from commit 773c1d9)

Protection on the fakes estimation: fake yield is set to zero for the…

8c5d50d

… analysis bins of specific axes (e.g. utAngleSign) which have identically zero yield in the A-Ax-B-Bx regions

small changes for plotting

e5ed649

consider the polVar as noi - workaround to better interface with the …

6085af8

…current version of Rabbit, it will be probably reverted when the definitions of poi/noi/nuisanceNotConstrained are fixed

fake prediction for uT<0 is now evaluated by scaling the ABCD predict…

5f8ce1f

…ion on uT>0 with an hardcoded factor (normalization)

label change for the utAngleSign axis

c9354b7

small fix

e433ca3

Merge branch 'main' of github.com:WMass/WRemnants into main_mergeUpst…

4bc98b5

…ream_cherrypick Merge from WMass/main

moved reference commit of rabbit

c79eca9

remove Zmumu OOA for theory agnostic with pol var

4c95789

fakes in uT<0 are evaluated by scaling the ABCD prediction on uT>0 wi…

488f6cc

…th a pT dependent shape

decorrelate FakeParam syst in uT bins

9293126

new syst for eta nonclosure in ut<0 bin

3f96a4c

updates for veto (no default change)

fec8ace

merge from rubenforti/signUtAnalysis

e7c51cc

updates

65201f6

updates for fakes vs uT

f1a26ce

updates

ea0dfe2

set noi=False for polynomial variations

2fc7884

fixes and updates to fit uT^mu

811a7b0

foo

ba08ba7

fixes

4d1e7b5

fixes

ea60c34

Merge branch 'main' of github.com:WMass/WRemnants into main_mergeUpst…

60299aa

…ream_cherrypick Merge from WMass/main

set default W width variation to original 0.6 MeV

1cd1a13

updates

98d150a

Merge branch 'main_mergeUpstream_testRuben' into main_mergeUpstream_c…

8a3041a

…herrypick Merge from my branch

restore default dictionary for datasets

4369f3a

cippy added 6 commits March 11, 2026 19:38

black to reformat

732cb46

fix wrong sample name for BSM studies

84791af

Merge branch 'main' of github.com:WMass/WRemnants into main_mergeUpst…

a9f232a

…ream_cherrypick Merging from WMass/Main

updates to test custom recoil syst

4a73d61

updates

b31ead9

Merge branch 'main' of github.com:WMass/WRemnants into main_mergeUpst…

be9cac5

…ream_devel_150326 Merging from WMass/main

cippy mentioned this pull request Mar 25, 2026

Updates for fakes with u_T^\mu bins and other fixes #643

Closed

kdlong reviewed Mar 25, 2026

View reviewed changes

moving definition of mass and width systs to rabbit_helpers

aed50a9

cippy added 5 commits March 26, 2026 16:55

Merge branch 'main' of github.com:WMass/WRemnants into main_mergeUpst…

3f50a31

…ream_devel_150326 Merging from WMass/main

fixing scripts/rabbit/setupRabbit.py

642e8f1

fix

c2a1062

fix

5d7d9de

Merge branch 'main' of github.com:WMass/WRemnants into main_mergeUpst…

b425570

…ream_devel_150326 Merging from WMass/main

davidwalter2 reviewed Mar 27, 2026

View reviewed changes

davidwalter2 mentioned this pull request Mar 27, 2026

fixing bugs and augment --presel option of setupRabbit.py #679

Merged

cippy added 4 commits March 27, 2026 14:52

couple of fixes

0c62ddb

implement comments from PR

c410b2a

make some utility histograms optional

fd8be87

Merge branch 'main' of github.com:WMass/WRemnants into main_mergeUpst…

25c14b4

…ream_devel_150326 Merge from WMass/main

davidwalter2 merged commit fe7c6a9 into WMass:main Mar 31, 2026
26 checks passed

davidwalter2 mentioned this pull request Mar 31, 2026

Small updates for signUt analysis #674

Open

Conversation

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

Uh oh!

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!

bendavid commented Mar 25, 2026

Uh oh!

cippy commented Mar 25, 2026

Uh oh!

cippy commented Mar 26, 2026

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!

davidwalter2 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!

cippy commented Mar 28, 2026

Uh oh!

Uh oh!

Reviewers

Assignees

cippy commented Mar 25, 2026 •

edited

Loading