X-Pipeline Review Meeting Monday 24 Nov 2008 09:00 Pacific / 12:00 Eastern
Agenda: Monday 24 Nov 2008 09:00 Pacific / 12:00 Eastern
- Code review: At line 706 of xtimefrequencymap.
Minutes:Attendance: Peter Raffai, Patrick Sutton, Jolien Creighton, Xavier Siemens.
- Re-run of GRBs with subset of injections, SG linear+circular low-frequency+high-frequency. Comparing with preliminary run. Close to opening the box. Main change was a fix to the normalization of the power spectrum. Overall scaling error but different for each IFO -- this means that the weighting of the different IFOs weren't the same -- different coherent combinations.
- No radical change in expected upper limits from previous revision.
- Some code reordering to improve code efficiency.
- Would like to open the box in a few weeks or so.
- Things that we had wanted to see:
- Data conditioning. Weren't keen on LPEF. Need to revisit xconditiondata.
- Overwhitening of data... Patrick wasn't keen on this big a change at this stage.
- DQ flags issue ... I think we converged on that.
- Which simulations to use for upper limit results. On-source or off-source? Entangled with DQ flag issue. This needs to be resolved.
- Continued reviewing xtimefrequencymap from line 706.
- 706--1192: different likelihood types.
- We will only review these: crossenergy, crossinc, plusenergy (aka hardconstraint), plusinc, nullenergy, nullinc (aka incoherentenergy but not what cWB calls incoherent energy), standard.
- Heads-up: Two others to be used in Virgo case: two more null energies just for H1+H2 h1h2nullinc and h1h2nullenergy. To be reviewed later.
- Why not compute e+ and ex vectors rather than f+ and fx and then normalize later (as is done)? It is possible that |fx| is close to zero... but then the normalization would cause problems?
- 823: problem with aligned detectors with cross energy. Normalization factor will give you nans. Just don't compute cross energy for H1H2 network. But it fx can get very small for non-aligned networks too.... For some sky positions and some frequencies this case can become pathological. For small but not underflowing numbers then things work out OK. Perhaps a check for range problems would be in order, but this is probably very unlikely to occur.
- Code is inefficient: recomputes plus and cross energies many times rather than just reusing.
- Line 1150: standard likelihood. For aligned H1 H2, the likelihood ratio is the plusenergy. Here, "standard likelihood" is taken to be the total energy, not the plusenergy. Technical document appears to be mum on the issue of an aligned network. Gap in documentation. Also, possibly a bug in code (well, code is correctly documented, but this is not what Patrick thinks it should be doing). This does affect the search: whether you use plusenergy or totalenergy for H1H2 GRBs!
- We reviewed plusenergy, crossenergy, nullenergy, standard, totalenergy.