Questions

• How is presto’s sigma calculated?
  • As I understand it you take medians of chunks to estimate the
    amplitudes in sections of the FFT; how is the size of these chunks
    determined? Do they overlap at all?
  • How much of accelsearch’s time is spent computing these
    medians? Could we afford to use overlapping/running medians or other
    tricks, if they would help?

• How does rfifind work? Look for overstrong frequency/time
  subintegrations, plus take FFTs of each channel and look for birdies?
  • It seems to be horribly slow with (some) GUPPI data – is this a
    settings issue or an intrinsic problem?

• How is prepfold sigma calculated? Specifically, how is the
  background level estimated? Is it just an std of the datacube along
  the pulse phase direction, followed by an average along the other two
  dimensions?
  • When I do a search for known pulsars with poor prepfold sigma,
    the only examples I came up with were in beams that had five or more
    known pulsars (often bright). Any idea why this condition leads to an
    unreasonably low prepfold sigma? (Upon reflection, it’s possible that
    this is being caused by the bug you were emailing us about. I’m not
    really clear on what triggers it, so I’ll have to experiment.)

• Why are so many poor candidates generated?

• Why do those candidates end up with significantly higher presto sigmas compared to genuine pulsars?