SModelS v2.3: enabling global likelihood analyses

We thank the referees for their positive assessments and the constructive criticism, which made us improve our paper. We made an effort to properly address all points raised by the referees (and more) and hope that the paper is now ready for publication in SciPost Physics. The changes to the manuscript are listed below.

- As requested by both referees, we have added a statement at the beginning of section 2.2 (page 5) clarifying what we mean by "approximately uncorrelated" analyses. Moreover, we now mention in footnote 4 the possibility of overlaps of signal and control regions in different analyses. This now reads:

"By approximately uncorrelated we mean that signal regions do not overlap and inter-analyses correlations of systematic uncertainties (stemming, e.g., from luminosity measurements) can be neglected. (footnote: Overlaps of SRs of one analysis with the control regions of another analysis in the combination can in principle induce correlations of systematic uncertainties and therefore should also be checked. However, we generally expect the effect to be negligible compared to other uncertainties in SModelS.)"

- Related to the above, on page 6, 2nd paragraph, we added a remark regarding reference [17] as an example for an approach to explicitly testing (and quantifying) analyses correlations.

However, we do not wish to give explicit recommendations as to the validity of analyses combinations: in the present paper, we present the new functionalities of the new SModelS version, but, as mentioned in the text, it is the responsibility of the user to apply them sensibly. This holds in particular for the question which analyses they treat as combinable.

- In response to the remark by the 2nd referee, that “Regarding the example discussed in Sec. 2, I think it would be interesting to extend discussion with a pair of points marked as blue/red in Fig. 5, e.g. those at ~(700, 200) GeV.“ we have added the paragraph starting with “We like to stress here …” as next-to-last paragraph in section 2.2 (page 7), as well as two explicit examples in section 4 (new Figs. 7 and 8, and a paragraph discussing them on page 13).

- In response to the question by the 2nd referee about the “robustness” argument in section 4, we have added a new figure showing the effect of statistically combining the hadronic EW-ino searches from ATLAS and CMS, ATLAS-SUSY-2018-41 and CMS-SUS-21-002 on the expected reach (now Fig. 5) and revised the paragraph starting with “The assessment of the excluded parameter space can be improved …” accordingly. Moreover, we have slightly expanded the last paragraph on page 13 and the concluding paragraph of section 4 (on page 14) to clarify what we mean by more robust constraints.

- Still in response to the 2nd referee: khaki is the name of the colour in matplotlib; for better readability, we changed “khaki” to “khaki (light yellowish)” in the text and the caption of Fig. 6. The choice of colour gives a good contrast in the plot without being too dominant, so we do not wish to change it.

- The 2nd referee criticised the presentation of results in section 4 in terms of neutralino vs. chargino mass and suggested that we “consider presentation of a plot(s) in terms of M1/M2/mu parameters instead.” To our mind, the presentation in Figs. 3-6 in terms of neutralino vs. chargino mass is more appropriate for our discussion. But, we have added an appendix (Appendix D) with auxiliary plots that show our results in terms of M1/M2/mu parameters. We hope this is satisfactory to the referee.

Other changes:

- In addition to the changes triggered by the referees’ comments, we have corrected a number of typos and updated some references. Moreover, we corrected a mistake in the 3rd paragraph on page 6: "In the former case, the observed limit is weaker, in the latter case stronger than the expected limit." (“stronger” and “weaker” were interchanged).

- We also realised that 13 of our SLHA input files had wrong neutralino2 decays due to a bug in SOFTSUSY; we have taken these 13 points out of the scan data, which leaves us with 18544 scan points instead of 18557 points in version 1. This has no impact on any of the plots or conclusions presented.

- Our results for SModelS v2.1 and v2.3 w/o SR combination had been done with a sigmacut parameter of 1e-3 fb, but those for v2.3 with SR combination had been done with the default sigmacut=5e-3 value. It turned out that the results in higgsino LSP region are somewhat sensitive to this sigmacut parameter. We have therefore re-run with a sigmacut of 1e-3 fb. This affects the exclusion status of about 100 points mostly with higgsino LSP from SModelS v2.3 with SR combination. All numbers and plots (Figs. 2-6) are updated accordingly. We stress that, apart from the strip at m(chargino1)\approx m(neutralino1), there is no visible effect in the figures.

- Last but not least, we realised that users might expect the ATLAS-SUSY-2018-16 analysis (the Run 2 search for electroweak production with compressed mass spectra) to be implemented in the SModelS database, though it is not (because it is not compatible with the approximations made in SModelS). We therefore added a clarifying remark in section 3, at the end of the first paragraph of page 8. And we changed “the full suit of available electroweak-ino searches” to “a large variety of electroweak-ino searches” in the abstract to avoid confusion about this point.