7 Feb
2013
7 Feb
'13
8:20 p.m.
Hi all,
I have constructed a new jackknife that splits the first 53 vs. last 53
half-scans in a scanset. This is intended to test whether we could
have an instrumental signal that's repeatable from scanset to scanset,
but changes from one half-scan to another. An example would be a
thermal oscillation that dies down as you get farther from the partial
load curve.
The pairmaps are still in progress, but with a coadd through Oct. 2010
I find this jackknife fails. The EE and BB power in the differenced
map is a good match for the EE and BB power at low ell in our jack0
coadds. I also find a significant mismatch in the T map between the
two splits. You can find the note here:
http://bicep0.caltech.edu/~spuder/analysis_logbook/analysis/20130207_scanja…
I still need to complete the pairmaps and coadd for the two-year data
set, and repeat the same analysis for a signal+noise sim. The
preliminary results look promising, though, that this could be what
we've been looking for.
- Walt
7 Feb
7 Feb
8:50 p.m.
Hi Walt,
I agree with your conclusion that we need to subject a signal + noise sim
to this jackknife to really judge what is going on. However, I'm looking at
your Figure 2a, and I don't think I'm seeing what you describe. Those
little lobes look pretty dim, and certainly not brighter than the low ell
lobes in some of the the simulated BB jackknife spectra, especially the 1st
half/2nd half jack.
For instance, look at the B_pure, 1st/2nd half jack for a few of the s+n
realizations, espeically sigIa and sigII:
http://bicep0.caltech.edu/~spuder/analysis_logbook/analysis/20130121_ellipt…
Is what you're seeing not consistent with something like that? Also, a lot
of our excess BB is located in the second and third science bins, not just
the first. Maybe I'm misunderstanding.
(Unfortunately my EE jackknifes are not visible because my color scales are
matched to the signal spectra.)
-Chris
On Thu, Feb 7, 2013 at 7:20 PM, Walt Ogburn <ogburn(a)stanford.edu> wrote:
Hi all,
I have constructed a new jackknife that splits the first 53 vs. last 53
half-scans in a scanset. This is intended to test whether we could
have an instrumental signal that's repeatable from scanset to scanset,
but changes from one half-scan to another. An example would be a
thermal oscillation that dies down as you get farther from the partial
load curve.
The pairmaps are still in progress, but with a coadd through Oct. 2010
I find this jackknife fails. The EE and BB power in the differenced
map is a good match for the EE and BB power at low ell in our jack0
coadds. I also find a significant mismatch in the T map between the
two splits. You can find the note here:
http://bicep0.caltech.edu/~spuder/analysis_logbook/analysis/20130207_scanja…
I still need to complete the pairmaps and coadd for the two-year data
set, and repeat the same analysis for a signal+noise sim. The
preliminary results look promising, though, that this could be what
we've been looking for.
- Walt
_______________________________________________
Bicep2-list mailing list
Bicep2-list(a)lists.fas.harvard.edu
https://lists.fas.harvard.edu/mailman/listinfo/bicep2-list
--
**********************************************************************
Christopher Sheehy - Graduate Student - University of Chicago
University of Minnesota,
Room 220 Tate, 116 Church Street S.E. Minneapolis, MN, 55455
Tel: 612-625-1802 Fax: 612-624-4578 email: csheehy(a)uchicago.edu
**********************************************************************
9:41 p.m.
New subject: Analysis note: half-scan jackknife and possible culprit
Thanks, Chris. I agree that the plots I've posted are just a start,
and we need to do a more careful comparison (as well as finishing the
coadds in my note!).
However, I think there are a few things in my posting that are already
clearly fishy:
1) the bumps at low ell in EE. I haven't seen comparable in any sims.
If you have them and can show them in a different color stretch, that
would be interesting.
2) the T jack failure. I don't see comparably bad in any other
jackknifes.
3) the correlation between the BB bumps in jackknife and non-jack. I
don't see that in the others you point to.
4) the TB correlation that matches between jackknife and non-jack.
- Walt
On Thu, 7 Feb 2013 19:50:53 -0600
Chris Sheehy <csheehy(a)uchicago.edu> wrote:
Hi Walt,
I agree with your conclusion that we need to subject a signal + noise
sim to this jackknife to really judge what is going on. However, I'm
looking at your Figure 2a, and I don't think I'm seeing what you
describe. Those little lobes look pretty dim, and certainly not
brighter than the low ell lobes in some of the the simulated BB
jackknife spectra, especially the 1st half/2nd half jack.
For instance, look at the B_pure, 1st/2nd half jack for a few of the
s+n realizations, espeically sigIa and sigII:
http://bicep0.caltech.edu/~spuder/analysis_logbook/analysis/20130121_ellipt…
Is what you're seeing not consistent with something like that? Also,
a lot of our excess BB is located in the second and third science
bins, not just the first. Maybe I'm misunderstanding.
(Unfortunately my EE jackknifes are not visible because my color
scales are matched to the signal spectra.)
-Chris
On Thu, Feb 7, 2013 at 7:20 PM, Walt Ogburn <ogburn(a)stanford.edu>
wrote:
Hi all,
I have constructed a new jackknife that splits the first 53 vs.
last 53 half-scans in a scanset. This is intended to test whether
we could have an instrumental signal that's repeatable from scanset
to scanset, but changes from one half-scan to another. An example
would be a thermal oscillation that dies down as you get farther
from the partial load curve.
The pairmaps are still in progress, but with a coadd through Oct.
2010 I find this jackknife fails. The EE and BB power in the
differenced map is a good match for the EE and BB power at low ell
in our jack0 coadds. I also find a significant mismatch in the T
map between the two splits. You can find the note here:
http://bicep0.caltech.edu/~spuder/analysis_logbook/analysis/20130207_scanja…
I still need to complete the pairmaps and coadd for the two-year
data set, and repeat the same analysis for a signal+noise sim. The
preliminary results look promising, though, that this could be what
we've been looking for.
- Walt
_______________________________________________
Bicep2-list mailing list
Bicep2-list(a)lists.fas.harvard.edu
https://lists.fas.harvard.edu/mailman/listinfo/bicep2-list
8 Feb
8 Feb
12:14 a.m.
New subject: Analysis note: half-scan jackknife and possible culprit
1) the bumps at low ell in EE. I haven't seen
comparable in any
sims.
If you have them and can show them in a different color stretch, that
would be interesting.
2) the T jack failure. I don't see comparably bad in any other
jackknifes.
3) the correlation between the BB bumps in jackknife and non-jack. I
don't see that in the others you point to.
4) the TB correlation that matches between jackknife and non-jack.
I agree with all these points. Also , is the sign of the jackknife
consistent with the assumption that the first half of the scan set
is contaminated? In other words, "1st half-2nd half" - since I noticed
the extra low ell features in EE and BB have the same sign in Jk0 and
your new JK. If so that is addition evidence for thermal transients.
In Figure 2, the jackknife level is similar to jk0 at low ell but
is still perhaps a bit lower at high ell. Regardless, I believe this
could be the jackknife failure we've been looking for.
- Walt
On Thu, 7 Feb 2013 19:50:53 -0600
Chris Sheehy <csheehy(a)uchicago.edu> wrote:
Hi Walt,
I agree with your conclusion that we need to subject a signal +
noise
sim to this jackknife to really judge what is going on. However,
I'm
looking at your Figure 2a, and I don't think I'm seeing what you
describe. Those little lobes look pretty dim, and certainly not
brighter than the low ell lobes in some of the the simulated BB
jackknife spectra, especially the 1st half/2nd half jack.
For instance, look at the B_pure, 1st/2nd half jack for a few of
the
s+n realizations, espeically sigIa and sigII:
http://bicep0.caltech.edu/~spuder/analysis_logbook/analysis/20130121_ellipt…
Is what you're seeing not consistent with something like that?
Also,
a lot of our excess BB is located in the second and third science
bins, not just the first. Maybe I'm misunderstanding.
(Unfortunately my EE jackknifes are not visible because my color
scales are matched to the signal spectra.)
-Chris
On Thu, Feb 7, 2013 at 7:20 PM, Walt Ogburn <ogburn(a)stanford.edu>
wrote:
_______________________________________________
Bicep2-list mailing list
Bicep2-list(a)lists.fas.harvard.edu
https://lists.fas.harvard.edu/mailman/listinfo/bicep2-list
Hi all,
I have constructed a new jackknife that splits the first 53 vs.
last 53 half-scans in a scanset. This is intended to test
whether
we could have an instrumental signal that's repeatable from
scanset
to scanset, but changes from one half-scan to another. An
example
would be a thermal oscillation that dies down as you get farther
from the partial load curve.
The pairmaps are still in progress, but with a coadd through Oct.
2010 I find this jackknife fails. The EE and BB power in the
differenced map is a good match for the EE and BB power at low
ell
in our jack0 coadds. I also find a significant mismatch in the T
map between the two splits. You can find the note here:
http://bicep0.caltech.edu/~spuder/analysis_logbook/analysis/20130207_scanja…
I still need to complete the pairmaps and coadd for the two-year
data set, and repeat the same analysis for a signal+noise sim.
The
preliminary results look promising, though, that this could be
what
we've been looking for.
- Walt
_______________________________________________
Bicep2-list mailing list
Bicep2-list(a)lists.fas.harvard.edu
https://lists.fas.harvard.edu/mailman/listinfo/bicep2-list
2:35 a.m.
New subject: Analysis note: half-scan jackknife and possible culprit
Also , is the sign of the jackknife
consistent with the assumption that the first half of the scan set
is contaminated? In other words, "1st half-2nd half" - since I
noticed the extra low ell features in EE and BB have the same sign in
Jk0 and your new JK.
Jamie and I were talking about this earlier. I believe that, by the
construction of the jackknifes, none of the speclin plots allow us to
determine which half is contaminated (not even TB). We could try to
tell the difference by making separate BB plots for the early and late
splits. However, even this is a bit suspect, since (as John points
out) we're still doing ground subtraction across all scans in the
scanset.
- Walt
9:54 a.m.
New subject: Analysis note: half-scan jackknife and possible culprit
Without forming new coadds, can you just show "B" instead of "BB"?
or "E" instead of "EE" in spec(lin)?
Chao-Lin
----- Original Message -----
Also , is the sign of the jackknife
consistent with the assumption that the first half of the scan set
is contaminated? In other words, "1st half-2nd half" - since I
noticed the extra low ell features in EE and BB have the same sign
in
Jk0 and your new JK.
Jamie and I were talking about this earlier. I believe that, by the
construction of the jackknifes, none of the speclin plots allow us to
determine which half is contaminated (not even TB). We could try to
tell the difference by making separate BB plots for the early and
late
splits. However, even this is a bit suspect, since (as John points
out) we're still doing ground subtraction across all scans in the
scanset.
- Walt
1:43 p.m.
New subject: Analysis note: half-scan jackknife and possible culprit
I've requested this option too. I think the imaginary part might be
shown on the conjugate half-plane.
On 2/9/13 3:54 AM, Chao-Lin Kuo wrote:
Without forming new coadds, can you just show
"B" instead of "BB"?
or "E" instead of "EE" in spec(lin)?
Chao-Lin
----- Original Message -----
_______________________________________________
Bicep2-list mailing list
Bicep2-list(a)lists.fas.harvard.edu
https://lists.fas.harvard.edu/mailman/listinfo/bicep2-list
--
___________________________________________________________________
John Kovac jmkovac(a)cfa.harvard.edu
Assistant Professor, Astronomy and Physics, Harvard University
160 Concord Ave rm 310, Cambridge MA 02138, 617-496-0611
CURRENTLY AT SOUTH POLE STATION, ANTARCTICA
Also , is
the sign of the jackknife
consistent with the assumption that the first half of the scan set
is contaminated? In other words, "1st half-2nd half" - since I
noticed the extra low ell features in EE and BB have the same sign
in
Jk0 and your new JK.
Jamie and I were talking about this earlier. I believe that,
by the
construction of the jackknifes, none of the speclin plots allow us to
determine which half is contaminated (not even TB). We could try to
tell the difference by making separate BB plots for the early and
late
splits. However, even this is a bit suspect, since (as John points
out) we're still doing ground subtraction across all scans in the
scanset.
- Walt
2:14 a.m.
Hi Walt,
Okay, for point 1, I will remake the EE spectra using your color scale.
For points 3-4, I agree. It is striking how much the correlation between
jack and non-jack mimics the observed B2 jack/non-jack correlation for some
of the other jacks. It doesn't look like this correlation is predicted by
any sims I've run.
I also agree with point 2. The T jack map fails far worse than any of the
jacks we have constructed up until now, even those jacks formed from a
small subset of tags.
-Chris
On Thu, Feb 7, 2013 at 8:41 PM, Walt Ogburn <ogburn(a)stanford.edu> wrote:
Thanks, Chris. I agree that the plots I've
posted are just a start,
and we need to do a more careful comparison (as well as finishing the
coadds in my note!).
However, I think there are a few things in my posting that are already
clearly fishy:
1) the bumps at low ell in EE. I haven't seen comparable in any sims.
If you have them and can show them in a different color stretch, that
would be interesting.
2) the T jack failure. I don't see comparably bad in any other
jackknifes.
3) the correlation between the BB bumps in jackknife and non-jack. I
don't see that in the others you point to.
4) the TB correlation that matches between jackknife and non-jack.
- Walt
On Thu, 7 Feb 2013 19:50:53 -0600
Chris Sheehy <csheehy(a)uchicago.edu> wrote:
--
**********************************************************************
Christopher Sheehy - Graduate Student - University of Chicago
University of Minnesota,
Room 220 Tate, 116 Church Street S.E. Minneapolis, MN, 55455
Tel: 612-625-1802 Fax: 612-624-4578 email: csheehy(a)uchicago.edu
**********************************************************************
Hi Walt,
I agree with your conclusion that we need to subject a signal + noise
sim to this jackknife to really judge what is going on. However, I'm
looking at your Figure 2a, and I don't think I'm seeing what you
describe. Those little lobes look pretty dim, and certainly not
brighter than the low ell lobes in some of the the simulated BB
jackknife spectra, especially the 1st half/2nd half jack.
For instance, look at the B_pure, 1st/2nd half jack for a few of the
s+n realizations, espeically sigIa and sigII:
http://bicep0.caltech.edu/~spuder/analysis_logbook/analysis/20130121_ellipt…
Is what you're seeing not consistent with something like that? Also,
a lot of our excess BB is located in the second and third science
bins, not just the first. Maybe I'm misunderstanding.
(Unfortunately my EE jackknifes are not visible because my color
scales are matched to the signal spectra.)
-Chris
On Thu, Feb 7, 2013 at 7:20 PM, Walt Ogburn <ogburn(a)stanford.edu>
wrote:
> Hi all,
>
> I have constructed a new jackknife that splits the first 53 vs.
> last 53 half-scans in a scanset. This is intended to test whether
> we could have an instrumental signal that's repeatable from scanset
> to scanset, but changes from one half-scan to another. An example
> would be a thermal oscillation that dies down as you get farther
> from the partial load curve.
>
> The pairmaps are still in progress, but with a coadd through Oct.
> 2010 I find this jackknife fails. The EE and BB power in the
> differenced map is a good match for the EE and BB power at low ell
> in our jack0 coadds. I also find a significant mismatch in the T
> map between the two splits. You can find the note here:
>
>
>
http://bicep0.caltech.edu/~spuder/analysis_logbook/analysis/20130207_scanja…
I still need to complete the pairmaps and coadd for the two-year
data set, and repeat the same analysis for a signal+noise sim. The
preliminary results look promising, though, that this could be what
we've been looking for.
- Walt
_______________________________________________
Bicep2-list mailing list
Bicep2-list(a)lists.fas.harvard.edu
https://lists.fas.harvard.edu/mailman/listinfo/bicep2-list
12:32 a.m.
New subject: Analysis note: half-scan jackknife and possible culprit
In Chris' pager
There is some hint of jackknife failure at low ell for dk jack in these
for both B(pure) and E. At a much lower level than "Walt's" jack but it
is there.
Maybe I am stretching but this can point to something thermal ...
7 Feb
7 Feb
10:21 p.m.
New subject: Analysis note: half-scan jackknife and possible culprit
Hi Walt,
Sounds like a good lead. How is your jack treating ground subtraction?
Is the ground template still being formed over the entire scanset (i.e.
both halves of your jack) and then subtracted from each of the two
halves? Or is the template being formed within each half? Either way,
the interaction of ground subtraction with splitting the data in this
way is not so clear to me so I'm eager to see how the sim responds.
John
On 2/8/13 2:20 PM, Walt Ogburn wrote:
Hi all,
I have constructed a new jackknife that splits the first 53 vs. last 53
half-scans in a scanset. This is intended to test whether we could
have an instrumental signal that's repeatable from scanset to scanset,
but changes from one half-scan to another. An example would be a
thermal oscillation that dies down as you get farther from the partial
load curve.
The pairmaps are still in progress, but with a coadd through Oct. 2010
I find this jackknife fails. The EE and BB power in the differenced
map is a good match for the EE and BB power at low ell in our jack0
coadds. I also find a significant mismatch in the T map between the
two splits. You can find the note here:
http://bicep0.caltech.edu/~spuder/analysis_logbook/analysis/20130207_scanja…
I still need to complete the pairmaps and coadd for the two-year data
set, and repeat the same analysis for a signal+noise sim. The
preliminary results look promising, though, that this could be what
we've been looking for.
- Walt
_______________________________________________
Bicep2-list mailing list
Bicep2-list(a)lists.fas.harvard.edu
https://lists.fas.harvard.edu/mailman/listinfo/bicep2-list
--
___________________________________________________________________
John Kovac jmkovac(a)cfa.harvard.edu
Assistant Professor, Astronomy and Physics, Harvard University
160 Concord Ave rm 310, Cambridge MA 02138, 617-496-0611
CURRENTLY AT SOUTH POLE STATION, ANTARCTICA
11:51 p.m.
New subject: Analysis note: half-scan jackknife and possible culprit
Hi John,
It's a good point, and I've also been wondering about interactions with
ground subtraction. Currently, the template is formed in the standard
way: left- and right-going half-scans are treated separately, but the
early and late halves of my jackknife are handled together. I believe
that must introduce a correlation between the splits that will show up
in the difference map. As far as I know, any such correlation should
still average down when coadding over tags.
- Walt
8 Feb
8 Feb
2:38 a.m.
New subject: Analysis note: half-scan jackknife and possible culprit
I'm eager to see the sims. I think a feature of our az-fixed scanset
strategy is that the p3 and gs will be filtering different modes on the
sky for the first half vs. the second half of each scanset, so the sims
will tell us what degree of agreement to expect. I haven't been able to
see your posting yet, though--perhaps what you are seeing clearly goes
beyond the expected signal differences. It sounds very interesting--the
right kind of thing to be pushing on.
If there is contamination in the first part of each scanset that is
absent in the second, clearly cuts provide a quick path forward. If
there is an issue of linear drift across the scanset, then it may be
worth revisiting an idea I've mentioned where the ground template at
each az point is p1 (offset and time derivative) rather than constant
across the scanset.
John
On 2/8/13 5:51 PM, Walt Ogburn wrote:
Hi John,
It's a good point, and I've also been wondering about interactions with
ground subtraction. Currently, the template is formed in the standard
way: left- and right-going half-scans are treated separately, but the
early and late halves of my jackknife are handled together. I believe
that must introduce a correlation between the splits that will show up
in the difference map. As far as I know, any such correlation should
still average down when coadding over tags.
- Walt
--
___________________________________________________________________
John Kovac jmkovac(a)cfa.harvard.edu
Assistant Professor, Astronomy and Physics, Harvard University
160 Concord Ave rm 310, Cambridge MA 02138, 617-496-0611
CURRENTLY AT SOUTH POLE STATION, ANTARCTICA
5:53 p.m.
If there is contamination in the first part of each
scanset that is
absent in the second, clearly cuts provide a quick path forward. If
there is an issue of linear drift across the scanset, then it may be
worth revisiting an idea I've mentioned where the ground template at
each az point is p1 (offset and time derivative) rather than constant
across the scanset.
I spent a little bit of time investigating evidence for temperature leakage in the first
half/second half maps (both individually and jacked). I didn't turn up much. Put
more precisely, regressing Q(first half) and Q(second half) against T yield roughly the
same coefficient. Regressing (Q(first half) - Q(second half)) against T yields a
coefficient that's about an order of magnitude lower than the first case. This
disfavors the zeroth-order leakage, but one could imagine more complicated leakage
patterns.
Leakage that would be fixed by a linear drift model might be equally well mitigated by
performing a relgain deprojection per half-scan.
Randol
On 2/8/13 5:51 PM, Walt Ogburn wrote:
Hi John,
It's a good point, and I've also been wondering about interactions with
ground subtraction. Currently, the template is formed in the standard
way: left- and right-going half-scans are treated separately, but the
early and late halves of my jackknife are handled together. I believe
that must introduce a correlation between the splits that will show up
in the difference map. As far as I know, any such correlation should
still average down when coadding over tags.
- Walt
--
___________________________________________________________________
John Kovac jmkovac(a)cfa.harvard.edu
Assistant Professor, Astronomy and Physics, Harvard University
160 Concord Ave rm 310, Cambridge MA 02138, 617-496-0611
CURRENTLY AT SOUTH POLE STATION, ANTARCTICA
_______________________________________________
Bicep2-list mailing list
Bicep2-list(a)lists.fas.harvard.edu
https://lists.fas.harvard.edu/mailman/listinfo/bicep2-list
4117
days inactive
4117
days old
bicep2-list@lists.fas.harvard.edu
12 comments
5 participants
participants (5)
-
Chao-Lin Kuo -
Chris Sheehy -
John Kovac -
Randol Aikin -
Walt Ogburn