Choose timezone
Your profile timezone:
Powered by Indico
v3.3.2-pre
See the updated slides PDF for minimal post-meeting annotations. In more detail:
- Louie raised whether the Scatter::addPoint(const Point&) method is needed. We decided no: make the argument non-const and no need for an overload (at present)
- Chris G to double-check that the disappearing entries from the rank files in the coverage-script MR are consistent with non-published records disppearing due to the more refined search criteria with the new API
- barcode() functions are a little controversial, given that pushing back on improperly normalised differential histograms has been physical good practice. But we shouldn't make the operation unnecessarily hard in the few cases where it is the right thing to do. AB to add some provocative warnings, etc. to the Doxygen.
- The setCrossSection div-by-zero patch seems fine, but has a non-obvious clash with the slightly different histo-object juggling code in the master branch. Chris G to check.
- Leif raised that 0/0 divisions in the histogram divide() methods lead to annoying NaNs in the output. This isn't necessarily wrong, but it would be useful for analyses to be able to specify custom values to be used in such cases. AB to investigate and maybe implement in YODA with forwarding to Rivet. Do we want such an overload to be specific to 0/0, or also to other NaN-generating operations? finite/0 should probably return +-inf without being "fixed"
- The CMS min bias analysis with ratio plots missing from HepData is still partially blocked. Frank S to raise the issue privately and try to move things along.
- Rivet's default {xs,xserr} = {0,0} is a bit annoying with current tools, since it can't be rescaled. Chris G proposed {1,0} or {1,1}; AB isn't keen since it treats 1 pb as a magic number. FS suggested -1 instead, which is "better" but would require a special abs() call in the rivet-merge scaling code, and it is still possible for negative cross-sections to be valid MC outputs. AB would prefer that rivet-merge have an overwrite mode (maybe with {nan,nan} defaults?) rather than having to do scaling relative to an arbitrary default. AB thinks we used to have a variety of scaling syntaxes in Professor or yodamerge, e.g. "x2.0" or "=2.0" for relative and overwrite behaviours respectively: Chris G to explore.
- Christian B summarised the feedback from the verry useful Rivet HI workshop in April. Main issues were HepMC (for which there's a dedicated meeting on 7 June), comparing data and non-MC models (naively outside our scope, but CB may pursue), and encouraging HI experiments to use no-data pre-publication Rivet analyses as a communication route with theorists asked to supply predictions. CB is giving a Rivet tutorial at the RHIC user meeting on 8 June: we encourage his to record (he doesn't need to listen to himself!) and put it on the Rivet YouTube channel for You Know Who.