Recasting a spell

For three successive Januaries now, since I started this blog in 2018, I posted a list of the things to look forward to, which for whatever reason didn't materialise and so were essentially repeated the next year. Given the state of the world right now some positive thinking seems to be needed more than ever, but it would be a bit of a joke to repeat the same mistake again. In particular, the measurement of the muon anomalous magnetic moment (which is apparently all I blog about) has still not been announced, and I'm led to wonder whether last year's controversies regarding the lattice QCD calculations have played a role in this, muddying the water.

Instead today I want to write a little about an effort that I have joined in the last couple of years, and really started to take seriously last year: recasting LHC searches. The LHC has gathered a huge amount of data and both main experiments (CMS and ATLAS) have published O(1000) papers. Many of these are studying Standard Model (SM) processes, but there are a large number with dedicated searches for New Physics models. Some of them contain deviations from the predictions of the Standard Model, although at present there is no clear and sufficiently significant deviation yet -- with the obvious exception of LHCb and the B-meson anomalies. Instead, we can use the data to constrain potential new theories.

The problem is that we can't expect the experiments to cover even a significant fraction of the cases of interest to us. For an important example, the simplest supersymmetric models have four 'neutalinos' (neutral fermions), two 'charginos' (charged fermions), plus scalar squarks and sleptons -- and two heavy neutral Higgs particles plus a charged Higgs; it is clearly impossible to list in a few papers limits on every possible combination of masses and couplings for these. So the experiments do their best: they take either the best-motivated or easiest to search for cases and try to give results that are as general as possible. But even then, supersymmetric models are just one example and it is possible that a given search channel (e.g. looking for pair production of heavy particles that then decay to jets plus some invisible particles as missing energy) could apply to many models, and it is impossible in a given paper to provide all possible interpretations.

This is where recasting comes in. The idea is to write a code that can simulate the response of the relevant LHC detector and the cuts used by the analysis and described in the paper. Then any theorist can simulate signal events for their model (using now well-established tools) and analyse them with this code, hopefully providing a reasonably accurate approximation of what the experiment would have seen. They can then determine whether the model (or really the particular choice of masses and coupling for that model) is allowed or ruled out by the particular search, without having to ask the experiments to do a dedicated analysis.

Ideally, recasting would be possible for every analysis -- not just searches for particular models, but also Standard Model searches (one example which I have been involved in is recasting the search for four top-quarks which was designed to observe the Standard Model process and measure its cross-section, but we could then use this to constrain new heavy particles that decay to pairs of tops and are produced in pairs). However this is a lot of work, because theorists do not have access to the simulation software used for the experiments' own projections (and it would probably be too computationally intensive to be really useful anyway) so the experiments cannot just hand over their simulation code. Instead there is then a lot of work to make approximations, which is why there is really some physics involved and it is not just a mechanical exercise. Sometimes the experiments provide pseudocode which include an implementation of the cuts made in the analysis, which helps understanding the paper (where there is sometimes some ambiguity) and often they provide supplementary material, but in general getting a good recast is a lot of work.

In recent years there has been a lot of effort by both experimentalists and theorists to make recasting easier and to meet in the middle. There are lots of workshops and common meetings, and lots of initiatives such as making raw data open access. On the theory side, while there are many people who write their own bespoke codes to study some model with some search, there are now several frameworks for grouping together recasts of many different analyses. The ones with which I am most familiar are CheckMATE, Rivet, ColliderBIT and MadAnalysis. These can all be used to check a given model against a (non-identical) list of analyses, but work in somewhat different ways -- at their core though they all involve simulating signals and detectors. There is therefore a little friendly competition and very useful cross-checking. Then there are other useful tools, in particular SModelS, which instead cleverly compares whatever model you give it to a list of simplified model results given by the LHC experiments, and uses those to come up with a much faster limit on the model (of course this loses in generality and can fall prey to the assumptions and whims of the particular simplified models results that are available, but is nonetheless very useful).

So the reason for the post today is the paper I was just involved in. It is a proceedings of a workshop where a group of people got together to recast a bunch of the latest LHC searches in the MadAnalysis framework. I didn't take part in the workshop, but did recast a search which was useful for a paper last year (if you are interested, it involves the signal in the inset picture)

so there are now 12 new reinterpretations of some of the latest LHC searches. This should be useful for anyone comparing their models to data. You can see by the number of authors involved how labour-intensive this is -- or check out the author list on last year's white paper; there are still many more searches from Run 2 of the LHC that are yet to be recast, so we still have our work cut out for some time to come before there is any new data!

If you are interested in the latest developments, there will be a forum next month.