Add HHModelPinvVHH model.

This PR adds a variant of the HH model with support for the VHH process. The overall rate is now composed of r * [GGF + VBF + VHH] with a corresponding rate parameter r_vhh to control the standalone rate of VHH. The additions to the model are mostly a copy of what is already there for VBF with a few things left to do:

• Verify that the VHH formula is indeed similar to VBF. Right now, it's really just a plain copy.
• Update the cross section values of the VHH samples.
• Check if the label naming scheme is applicable. Maybe it's worth to rename C3 to kl to be fully consistent with the GGF and VBF samples.
• Update tasks and plots for additional r_vhh parameter.

When merged, the new model should perhaps live next to the existing one to avoid interference with the current GGF+VBF combination efforts. In the long term, this combined model might be favorable though.

@lowang @capalmer @lcadamur

added model label

changed the description

added 1 commit

• dfce6efe - Adapt tasks and plots for additional r_vhh parameter.

Compare with previous version

changed the description

marked the checklist item Update tasks and plots for additional r_vhh parameter. as completed

added 1 commit

• 0863a484 - update xsecs for VHH process and follow naming conventions

Compare with previous version

marked the checklist item Verify that the VHH formula is indeed similar to VBF. Right now, it's really just a plain copy. as completed

marked the checklist item Update the cross section values of the VHH samples. as completed

marked the checklist item Check if the label naming scheme is applicable. Maybe it's worth to rename C3 to kl to be fully consistent with the GGF and VBF samples. as completed

Looks like this converged quite quickly.

@lowang Can you try to create a limit scan with this branch? E.g.

law run PlotUpperLimits \
  --version dev1_model \
  --hh-model HHModelPinvVHH.model_default \
  --datacards <your_cards_here> \
  --pois r_vhh \
  --scan-parameters C2V,-10,10,21 \
  --show-parameters kl,CV \
  --workers <cores>

If you are on a Mac and have iterm with imgcat enabled, you can also add --view-cmd imgcat which will show the plot in the terminal after the tasks are done.

Hi Marcel,

I run the command on our test datacard and it seems running smoothly without errors.

limits__poi_r_vhh__scan_C2V_-10.0_10.0_n21__params_r1.0_r_qqhh1.0_r_gghh1.0_kl1.0_kt1.0_CV1.0.pdf

Fyi, I ran a quick comparison with the bbzz4l datacards (which do only depend on r_gghh) to check if the results remain the same. So indeed, datacards without the samples that are matched by the model and assigned to the respective r_* parameter are still evaluated correctly (the curves are identical).

model_check.pdf

@lowang Good to hear!

Did you disable the uncertainty band in your plot?

You can also add --y-log for better visualization.

added 1 commit

• 530cffee - Fix multi model tasks.

Compare with previous version

Yeah it seems the r is too large that y-axis overshadowed the uncertainty band. limits__poi_r_vhh__scan_C2V_-10.0_10.0_n21__params_r1.0_r_qqhh1.0_r_gghh1.0_kl1.0_kt1.0_CV1.0__log.pdf

Ah that rings a bell. The smooth graph area interpolation fails in ROOT when the point errors are outside the canvas. You can set --y-max to e.g. 1e7 to verify that the bands are at least extracted correctly.

No, I would rather think that combine did not save the uncertainty values in the output files. Can you send me the output of the MergeUpperLimits task (a single npz file) ?

I checked your output,

and combine can't estimate the other quantiles (see attachment), so I think we can leave it like that for now.

@lcadamur Would you mind having a look at the model file? The VHH-related parts are really just plain copies of the VBF code. Thank you!