From what I understood from Ger's talk in Albuquerque (and maybe
I misunderstood) was that they've always excluded some short
baselines, they are just being much more aggressive now. I don't
know what they were using previously, but the current calibration
excludes baselines < 250 wavelengths. So I wouldn't say they're
just getting around to realizing this bias, but they are being
more conscience of it.
As a side note - 250 wavelengths seems really far out there (from
an MWA background anyway). What does that do to your calibration
when you're essentially chopping all baselines that _do_ see your
calibrator as a point source?
-Adam
On Fri, Dec 11, 2015 at 6:37 AM Chris Carilli <[email protected]
<mailto:[email protected]>> wrote:
funny. leaving out short spacings in self cal is standard
practice at the
VLA since the 1980's, for the obvious reason that the model
rarely
contains all the flux density seen on short baselines.
surprised they are
just getting around to this now.
do they perform the power spectral analysis on final image
cubes in 3D?
cc
> Bonjour Heratics,
>
> I talked to Saleem and Leon quite a bit in Paris
(supposedly, all the
> talks
> were filmed <http://ilp.upmc.fr/firstlight/program.php>,
but only the
> first
> two days have been posted and that doesn't include either
talk or mine).
> Some interesting tidbits and clarifications:
>
> - The key improvement over previous results is that they
now use only
> the long baselines to calibrate (especially the
direction-dependent
> calibration). Before using short baselines in the
direciton-dependent
> calibration caused a loss of all diffuse structure,
which got absorbed
> into
> their inferred beams. That calibration is done on every
10 seconds of
> data.
>
>
> - The LOFAR analysis strategy has been repeated
iterations of refining
> a
> source model, calibrating to it, and rerunning all their
analysis.
> They've
> done this ~10 times so far. They say it'll get a lot
faster now, since
> they
> just bought a big new analysis cluster. On the other
hand, the analysis
> scales with the amount of data used and this limit comes
from ~150
> hours
> out of the ~2000 they have in the can (about 4 petabytes).
>
>
> - The power spectrum error bars (e.g. on slide 25 of
Ger's talk) are 2
> sigma. The power spectra really are inconsistent with
noise at all k,
> though they only go up the a relatively modest k
corresponding to a 0.4
> MHz
> binned channel width.
>
>
> - Saleem showed variance statistic plots as a function
of frequency,
> both for Stokes I and Stokes V. Stokes V look at the
10ish mK level, I
> at
> the 50ish mK level. Those numbers are hard to interpret,
since the
> variance
> statistic (both data and theory) depend on the smoothing
scale in
> frequency. He also showed a "cross RMS" which came from
correlating
> neighboring channels (on roughly the 100 kHz scale),
which gave
> slightly
> better results (40ish mK). This is a harder statistic to
interpret,
> since
> the correlation of EoR structure between channels on
that scale is not
> 100%. It seems strange to me that they wouldn't look at
interleaved
> channels, since they have so much spectral resolution at
the outset,
> but
> they must have to bin too coarsely in frequency at some
intermediate
> stage
> for that to work.
>
>
> - Saleem said that LOFAR didn't see the wedge in their
final result,
> referring to the same slide as Ger's slide 24. Whatever
their
> systematics,
> they must mix k_perp modes. After his talk, Saleem
showed me some power
> spectra with less foreground subtraction that he said
had "wedge
> structure"
> in them. It was really hard to tell how "wedge-like"
they were, in part
> because they were small and over a limited k range, and
in part because
> they used those contours plots that are so hard to read.
>
> -Josh
>
> On Wed, Dec 9, 2015 at 11:36 PM, Saul Kohn
<[email protected] <mailto:[email protected]>> wrote:
>
>> This is published online (no login required; the
conference website), so
>> I
>> think it's kosher to share. Check out slides 23-25.
>> http://lwa.phys.unm.edu/abq2015/talks/deBruyn.pdf
>>
>> Saul
>>
>> On Thu, Dec 3, 2015 at 9:12 AM, Adrian Liu
<[email protected] <mailto:[email protected]>> wrote:
>>
>>> Absolutely, it’s complementary and valuable. It’s
definitely a
>>> result
>>> that’s not just line-of-sight, and relies on foreground
subtraction
>>> at low
>>> k’s. The foregrounds were removed using the GMCA
algorithm, with 6
>>> components, though (understandably) during the talk they
didn’t go
>>> into too
>>> much detail about it. (I had to talk to Ger afterwards to
get the
>>> number 6).
>>>
>>> Ger will unfortunately not be here on Friday, which is
why he had to
>>> present yesterday even though all the other EoR talks are
on Friday.
>>>
>>> > On Dec 3, 2015, at 6:50 AM, Chris Carilli
<[email protected] <mailto:[email protected]>> wrote:
>>> >
>>> > well, yes and no. the number may not be much different,
but, if they
>>> are
>>> > doing this in the image domain as well as line of
sight, it shows
>>> that
>>> > progress can be made in 3D. that is, if the result is
believable? I
>>> would
>>> > assume so, given they are presenting in public, but
need to see
>>> paper.
>>> and
>>> > maybe they are focusing on LoS PS? mantra: 'multiple
approaches and
>>> > experiments are good...'
>>> >
>>> > I will be talking about PAPER and HERA at New Mexico
Tech physics
>>> dept
>>> > today at 4PM, for those in ABQ who might be bored and
want to take
>>> the
>>> > drive...
>>> >
>>> > I will be in ABQ friday and will try to get some info
from Ger.
>>> >
>>> > cc
>>> >
>>> >
>>> >
>>> >> as i mentioned earlier, saleem said that the limits were
>>> "PAPER-like",
>>> so
>>> >> nothing revolutionarily new... he also mentioned
moving back to the
>>> power
>>> >> spectrum statistic, in lieu of the variance, since the
foregrounds
>>> were
>>> >> easier to deal with...
>>> >>
>>> >>> On 03.12.2015., at 14.30, Adam Beardsley
>>> <[email protected]
<mailto:[email protected]>>
>>> >>> wrote:
>>> >>>
>>> >>> mK
>>> >>>> On Thu, Dec 3, 2015 at 6:30 AM Jonathan Pober
>>> >>>> <[email protected]
<mailto:[email protected]>> wrote:
>>> >>>> 20 or 30 mK or mK^2?
>>> >>>>
>>> >>>>> On Wed, Dec 2, 2015 at 11:47 PM, Adrian Liu
<[email protected] <mailto:[email protected]>>
>>> >>>>> wrote:
>>> >>>>> Yes. They have both P(kperp,kpara) limits as well
as P(k) limits.
>>> No
>>> >>>>> mention of the variance method today.
>>> >>>>>
>>> >>>>>> On Dec 2, 2015, at 9:42 PM, danny jacobs
>>> <[email protected] <mailto:[email protected]>>
>>> >>>>>> wrote:
>>> >>>>>>
>>> >>>>>> But this is a power spectrum limit rather than
with the variance
>>> >>>>>> method?
>>> >>>>>>
>>> >>>>>>> On Wednesday, December 2, 2015, Saul Kohn
>>> <[email protected] <mailto:[email protected]>
>>> >
>>> >>>>>>> wrote:
>>> >>>>>>> Hi all,
>>> >>>>>>>
>>> >>>>>>> To follow this up, Ger made a LOFAR EoR project
overview
>>> >>>>>>> presentation (probably the same one as Dave was
mentioning) in
>>> >>>>>>> Albuquerque this afternoon.
>>> >>>>>>>
>>> >>>>>>> They were hazy on the details of power spectrum
estimation from
>>> >>>>>>> their images, but he quoted *preliminary* limits
of between 20
>>> and
>>> >>>>>>> 30mK for redshifts 7.5 to 10, k~0.05 Mpc^-1 using
155 hours of
>>> data.
>>> >>>>>>> Their estimate on systematics comes from Stokes
V, which was at
>>> >>>>>>> levels ~ 10mK, so intersecting with the
higher-power EoR
>>> models.
>>> On
>>> >>>>>>> the slide it quoted "Zaroubi et al. 2016", so I
guess we can
>>> expect
>>> >>>>>>> to see something official next year.
>>> >>>>>>>
>>> >>>>>>> Saul
>>> >>>>>>>
>>> >>>>>>>> On Wed, Nov 11, 2015 at 5:06 PM, DAVID DEBOER
>>> >>>>>>>> <[email protected]
<mailto:[email protected]>> wrote:
>>> >>>>>>>> To clarify a bit, I’m not sure that
Michiel’s
>>> statement
>>> applies
>>> >>>>>>>> to anything new, but rather that there was some
possibility
>>> that
>>> >>>>>>>> Ger had made a presentation somewhere regarding
a limit.
>>> >>>>>>>>
>>> >>>>>>>> Dave
>>> >>>>>>
>>> >>>>>>
>>> >>>>>> --
>>> >>>>>> Sent from Gmail Mobile
>>> >>
>>> >
>>> >
>>> >
>>>
>>>
>>>
>>
>
