More MC Studies

1
More MC Studies

• See a major difference between the number of
  background subtracted charm particles as a
  function of target slab between the data and the
  MC.
• The data has a much larger charm yield deficit in
  the upstream target segments compared to MC.
• This suggests a major simulation problem in the
  target region which could be source of lifetime
  inconsistency and too much MCS
• Other possibility is a problem with MCS
  simulation but this does not appear to be the
  case.
• There is a small problem with the beam spot, but
  fixing it does not cure the problem.
• Effect is present before Tarsil installed.
• Evidence that effect is due to SSD acceptance is
  weak

2
Motivation Whats wrong with our lifetimes?
Want to study differences between K? and KK
lifetimes which we can do statistically to 1.12
which is 3? better than CLEO
No f(t) K3?
KK 401?4.5fs
K3? with f(t)
But we are seeing ?5 inconsistencies in K3?
lifetime when split in Zprimary and P(K3?) after
MC correction
Varying kaonicity and elsig mass fit/sideband
3
Assigning slab number
This is the distribution of the primary vertex
for sideband subtracted silver mode charm. k?
k2? k3? kk? pk? are combined. We are in mcs
period 6 Rungt 9422
1
2
0
5/6
3
4
4
Charm slab populations
There is a significant fall off in the charm
yield as a function of slab number present in the
data The fall off in data is most severe for
short lived charm such as pk? and least severe
for k2?. The fall off in data for k?,k3?, kk?
is nearly identical as are their lifetimes. This
feature is not well matched by the MC. The fall
off is much less severe and the effect is nearly
independent of state lifetime. MC has too much
charm in tarsil.
DATA
MC
data blow up
mc blow up
5
What could cause this?

• Simple a priori explanations of an deficit in the
  upstream targets
• Improperly modeled charm kinematics (study E and
  Pt)
• Improperly modeled targets or beam profile
• Study x and y profiles
• SSD aperture acceptance problem
• Artificially restrict acceptance
• Look at high momentum
• Too little daughter absorption in MC
• Tarsil acceptance problem (See if effect exists
  pre-tarsil)
• MCS problem causing DCL cut failure or SSD
  reconstruction inefficiency (Study DCL
  distribution for slabs)

6
Dilution Effect?
Zsecondary for the various subtracted silver
charm species in mcs period 6.
k?
k3?
k2?
If the fall off with slab number is tied to
Zsecondary one expects the slope to be smeared
out for long lived states since they have a large
probability of decaying in the slab downstream of
their production slab.
pk?
kk?
7
Eliminating dilution
data
mc
We can eliminate the dilution by putting a cut on
Zsec - Zprim lt 5 mm which makes all species
short-lived . Indeed all 5 species have a
consistent slab charm populations with this cut
with suppression of the upstream slabs much
larger in data than in Monte Carlo. If daughter
absorption were a significant cause of the
data/mc discrepancy, we would expect a larger
upstream depopulation for k3pi than for kpi which
is not observed.
8
Acceptance effect?
If upstream fall off is due to SSD acceptance one
would expect to see a momentum dependence to slab
profile since low momentum implies low angles.
These distributions are for all charm but with a
dZ lt 5 mm cut to eliminate dilution.
Indeed both the MC and data samples show more
variation at lower charm state momentum which is
consistent with an acceptance effect.
data
MC
This is also confirmed by the behavior of slab 7
which indicates a much higher TR1 fraction at low
momentum in both MC and data. The MC seems to
display a larger P dependence than the data which
is strange. Finally, if one compares the low
momentum MC and data sample, there is still
significantly more variation in the data than in
the MC.
P lt 40
9
Asymmetry Study
The momentum asymmetry ? for D?K? is related to
the maximum opening angle between a track and the
D momentum. In the limit where the kaon and pion
masses are negligible and the D has low Pt, we
obtain these equations for the the maximum
asymmetry if the maximum opening angle is R/?Z
where ?Z is the distance from the decay secondary
and the 1st SSD station and R is the radius of
this station.
data MC
? lt 0.15
0.30lt? lt 0.15
? ?(ZSSD1 -Z2ndry)/E
? gt 0.30
There does appear to be fall off at high
asymmetry in data which is not as severe as that
in MC for ?gt 0.3 but it is not as dramatic as I
would have hoped.
10
Limiting the SSD fiducial
mc 8000
mc 8000 all 5 modes
The plot on the left gives the maximum transverse
distance from 0 for each of the charm secondaries
for a mc run at run 8000. TRAK(8?11,ipwc) was
used to extrapolate x and y to Z 4.63 which is
Z(SSD1). rmax (x2 y2)1/2. The plot on the
left shows the slab profile for all 5 modes (with
?z lt 5 mm) for all reconstructed simulated charm
and for where rmax lt 1 cm. This is for an MC run
at 8000 so the tarsil shouldnt be a factor. But
wait .. there is an rmax horror show to come!
11
A false disagreement in Rmax distributions!
We were seeing a big difference between data and
MC in the largest transverse position of a
secondary track at SSD station 1 Rmax is
sensitive to the charm kinematics, beam profile,
and SSD simulation. This turned out to be a red
herring. The trak(8-11) array is in M2
coordinates and there is a large transverse shift
between systems in data but not in MC!
all modes
data mc mc shift
K?
K2?
KK?
K3?
pK?
This discrepancy appeared to be in the opposite
direction of the z profile mismatch. The MC had a
larger fraction of charm from the upstream slabs.
We would thus expect a larger MC Rmax if the
secondaries subtend a similar cone as the data.
We put the same shift in the M2-GB coordinates in
MC as in data to get the red MC simulations and
now the match is fairly good. Problem found
thanks to Erik G!
12
But the shift does not fix the upstream deficit
area norm
peak norm
Here we compare the slab distribution for
background subtracted charm (all 5 modes but with
a dZ cut) to the data and a MC run with the ?0.8
cm rungeom offset present in the data and the
nominal MC without this shift . As expected this
offset does not appear to affect the MC results
or make them closer to the data. To further
emphasis the severity of the effect, we show the
same plot where the MC is normalized to slab 4
rather than the total yield.
13
But how do the charm kinematics compare anyway?
data mc
K? K2? K3?
KK? pK?
Both the average Ptsq and energy spectra of the 5
modes are harder in data than in the (pythia6) MC
we are running. The Pt spectra for the Ds seems
somewhat anomalous in data. These differences are
rather minor in our view and will tend to
compensate somewhat in producing an angular cone.
14
How about the target geometry and beam spot?
The x primary vertex distribution is relatively
well simulated. The y profile in MC is broader
than in data. The Z profile materials appear to b
in the correct position but the charm fraction is
way off of course
data mc
The Tr1 profile is strange in data.
mcs period 6
15
How about the track intercepts at the 4 ssd
stations?
Except for the slightly broader y distributions
the x and y intercept of background subtracted mc
charm secondaries are a nearly perfect match to
the data at each of the 4 SSD planes. This MC
uses the release beam spot but was run with the
?0.8 cm xy shift the M2 and SSD coordinate
system.
16
How about the quality of the SSD tracks?
k2?
k?
k3?
We are plotting the number of SSD tracks with no
missing hits for the 5 modes. Agreement between
data and mc seems poor.
kk?
pk?
17
How about the track/stub fraction
Generally the number of 5 chamber tracks in data
is a reasonable match to that in MC except for
the pk? mode where the MC produces a harder
spectrum than the data. Am surprised that the k3?
Ntrack distribution is not a better match to the
data given how well the average momentum matches.
data mc
k2?
k3?
k?
ltPgt
pk?
kk?
mode
N 5chm
18
Try a new mc with better y beam width ,SSD
offset,and mcs1
The new mc uses a Gaussian beam spot with a
y-width closer to that in the data. The x and y
profile of the 2ndary vertex are very close to
data. The x and y position of charm secondaries
extrapolated to SSD4 are now very close as well.
But the slab populations seem to be an even
poorer match than before. The new MC is run with
mcs1, the old had mcs5. Both MC are run with
modified rungeom file.
19
Does restricting the primary vtx to beam center
affect things?
The plot to the right has a cut requiring X,
Y lt 0.5 mm where X and Y are the primary vertex
positions. There is nearly no difference in the
fall off for events in the upstream target slabs.
20
How does restricting maximum R at slab 4 affect
things?
rs4mx is the largest radial value of the
intercept of each charm track with SSD station 4.
We compare data to two mc one with mcs5 and one
with mcs1. Both use Johns modified MCS file for
mcs period 6 and both use an adjusted Gaussian
ball for y beam. The rs4mx distributions for data
and either MC are nearly identical.
We are able to induce more of an upstream deficit
in data or MC by requiring rs4mx lt 1 cm. But in
all cases the data has a larger deficit than the
data.
21
Does the beam intensity affect deficit?
This is the CTDL instantaneous beam intensity
based on scalars. We plot the background
subtracted slab populations for low and high
intensity and see little difference.
22
What happens with extra SSD smearing
I ran a special MC based on the Cumalat MCS file
with mcs1 where I smeared tracks with a 25
micron Gaussian prior to digitizing them in the
SSD. The distortion in the DCL distribution is
evident relative to data and a MC with no extra
smear
Interestingly enough, the MC with additional
smears looks even less like the data. At higher
multiplicity there is actually a upstream slab
surplus!
23
Did the MCS1 affect the mass resolution?
Generally using the default mcs5 which includes
Moliere and elastic scattering causes an
overestimated width relative to the data. You
get a much closer match to the K3pi width versus
slab number by setting mcs1. This setting
simulates just MCS and uses a simple Gaussian
projected width of ? (.014/P) sqrt(x/x0) with
no log terms.
Generally using the default mcs5 which includes
Moliere and elastic scattering causes an
overestimated width relative to the data. You
get a much closer match to the K3pi width versus
slab number by setting mcs1. This setting
simulates just MCS and uses a simple Gaussian
projected width of ? (.014/P) sqrt(x/x0) with
no log terms.
BO1
up tarsil
BO2
BO3
dwn tarsil
tr1
BO4
The green mc shows a run with mcs5 using a
modified mcs.dat prepared by John Cumalat. The
magenta mc uses the JC mcs file but run with
mcs1 (simple Rossi)
24
A change in primary vtx multiplicity
This is the multiplicity of the primary vtx for
the 5 modes. The mcs5 mc seems to have a lower
primary vtx multiplicity than the mcs1 mc which
is very strange since it includes inelastic
scattering which includes jet secondaries and
this is turned off in the mcs1 mc.
Upon reading the code, we see that mcs5 is
required to get daughter absorption. This
probably explains why the mcs1 mc has a higher
Nprim
25
Charm secondary properties
We are comparing the averages of three quantities
for the 5 background subtracted modes. n12 is the
fraction of perfect 12 hit SSD tracks.
ltfracthigt is the average of the fraction of hits
of an SSD track
ltmomgt is the average track momentum. These
averages are plotted versus the slab number of
the primary vertex.
The new mc with a better beam spot match (and
mcs1) seems to reproduce the SSD properties of
the tracks better than the old but mysteriously
it is a worse match to track momentum.
26
Comparing charm properties
mcs 1 MC
The resolution and elsig dependence with slab are
a good match between data and MC. The primary
vertex dependence is probably due to the fact
that the mcs1 MC has no inelastic absorption so
it is higher in the upstream slab segments.
27
How stable is beam?
Beam center shifts are about 3 mm and quite
evident but small on the scale of the beam spot
size.
28
Does upstream depopulation occur before tarsil
installed?
MC run 8000
DATA mcs 4
These have an elsiggt7 cut rather than dz cut
(whoops!)
Again the fall off in data for the upstream slabs
in data seems much more severe than in MC even
w/o tarsil The Z profiles of data amd MC compare
well except TR1 seems thinner in MC. Note that
the upstream/downstream ratio is smaller in MCS4
than MCS 6 even though the slabs are closer to
the SSD! True both in data and MC. Beam energy
change effect??? IS this a big HINT?
mcs 6 mcs4
data
mcs 4
data mc
29
How much does Pchm change versus run?
We plot the background subtracted, average charm
momentum over the 5 modes versus run number.
There is indeed a significant difference between
mcs period 4 and 6. Is it large enough to account
for such a dramatic change in the slab profile?
mcs6
30
X and Y target profiles
The observed x and y distributions of the primary
vertex appear to be centered but distorted
Gaussians with y narrower than x. Without
envoking the beam profile, the MC user gets
generated Gaussian distributions with y broader
than x. The reconstructed widths are
significantly smaller than the generated widths.
But this reflects the finite size of the target.
31
Is the beam spot the same on all slabs?
This is the background subtracted charm x and y
primary vtx distribution in data (mcs 6) for the
various slabs. They are remarkably consistent as
we would expect. The points with the large error
bars are for slab 0.
32
Effects of beam spot
Ran a flat simulated beam spot to study beam
profile effects. This is the target profile
obtained in slab 3 for this MC. I am surprised
that there is little deviation from a uniform
accepted vtx distribution across the face of the
target except a slight effect at the corners. To
emphasize SSD acceptance effects we compare the
default MC to the flat MC slab profiles where we
have a dZ cut and a cut on the charm momentum of
lt 40 GeV. There is nearly no difference in the
slab profiles.
Plt40
I believe that this is the shape of the target
and tarsil used presently in McFOCUS. TR1 is 17
larger.
33
DCL versus slab

• The k3? DCL distribution is the least uniform
• Not much variation in DCL distribution for
  different slabs in data or in MC.
• Slab 4 slightly lower DCL than Slab 1
• Data has a peak at high DCL which is missing in
  MC.
• Its unlikely that MCS or confusion is problem

data
MC
data
DCL
34
New Microsim (from development)
BeO2
This distribution actually has an elsiggt7 cut
rather than dz cut (whoops!)
BeO1
BeO3
BeO4
Indeed the new microsim does a better job at
reproducing the dcl distribution versus slab
especially for BeO3 and BeO4 where confusion
should be the largest. Unfortunately the new
Microsim does not appear to affect the slab
population.
35
A surprising slab vrs Nprimary effect
Plot slab profile as a function of the primary
vertex multiplicity for data and MC
data
This distribution actually has an elsiggt7 cut
rather than dz cut (whoops!)
jcmcs micro mc
The profile depends only slightly on multiplicity
of the primary vertex in data.
In MC, the profile depends dramatically on
multiplicity of the primary vertex.
This has got to be a really big hint as to whats
going on since it is a dramatic difference
between the data and MC! But I havent a clue as
to what it means.
36
Success (?) with a re-written rgmcs.sf
I basically took all of the crap out of the
rgmcs.sf routine and made a version just based on
interaction or collision lengths. Seems to have
fixed the problem considerably when I use the
collision length. Need to check other
distributions however!
3.5ltNprm lt 6.5
Nprm lt 3.5
norm to peak bin
Norm to peak bin
6.5ltNprm
all Nprm
These distributions actually have an elsiggt7 cut
rather than dz cut (whoops!)
37
But does new rgmcs.sf get multiplicity dependence
correct?
K?
K2?
K3?
KK?
pK?
These distributions actually have an elsiggt7 cut
rather than dz cut (whoops!) This accounts for
the large lifetime dependence but data and mc are
handled the same.
38
Slab vrs Type with dZ corrected
With the dZ cut we are not getting nearly as good
agreement with the new mc as without it but
things are much better than in old mc.
K3?
K2?
K?
KK?
pK?
all modes
39
Slab vrs Nprim w/ correct dZ
Again, we see the rgmcs fix is not perfect with
the corrected dZ cut.
3.5ltNprm lt 6.5
Nprm lt 3.5
Some differences might persist since the MC is
not subjected to exact same editing cuts as data.
Perhaps the elsig gt7 cut which was mistakenly put
on several of the slides where agreement was
better.
6.5ltNprm
40
Correct dZ and Skim cut?
The data was originally skimmed with an elsig gt 5
cut. When this was applied to the MC with the
JWMCS elastic option, the agreement with the data
is much better than before.
3.5ltNprm lt 6.5
Nprm lt 3.5
all Nprm
6.5ltNprm
Both data and MC have an elsiggt 5 cut which was a
sub-skim requirement on data.
41
K3pi Width with new MC
The widths are very close to the data except for
the upstream Tarsil
42
K3? Prim and Secondary Vtx with new MC
Zprimary
Zsecondary
Agreement between the mc and data for K3? is
quite good except in the regions where the
material is not BeO . The mc is producing more
tarsil events and less in TR1. For this
comparison pains were taken to insure the same
elsig and PiD cuts in data as in MC. These cuts
are elsiggt5 , kaonicitygt 2 , picongt -2 and the
Run gt 9422. The picon cut is fairly tight.
43
Investigated many possible sources.

• Improperly modeled charm kinematics
• Simulations done with Pythia6 are very close in
  P(D) and Pt (D)
• Improperly modeled primary vertex properties
• Primary vertex distributions seem very good as
  well. A bug due to improper absorption code was
  found. We found the primary vertex was a big
  player.
• Improperly modeled targets or beam profile
• Only a slight problem was found with beam
  profile. Target positions are great match. A
  slight anomaly with TR1 was seen
• SSD aperture acceptance problem
• Major focus of the search and but very minor
  player
• SSD reconstruction inefficiency
• No indication of a problem. The transverse
  profiles in data and MC are a very good match.

44
More possible sources

• MCS problem causing DCL cut failure
• No evidence for this. DCL is well reproduced with
  new Microsim
• K3? width versus slab number is very well
  reproduced when bogus rgmcs.sf options are turned
  off.
• Tarsil problem
• Upstream deficit also appeared in mcs period 4.
  There are some oddities with the width of K3?
  when it verticizes in the upstream tarsil which
  bears investigation.
• Too little daughter absorption in MC
• This turned out to be much of it but I was fooled
  by a lack of multiplicity dependence.
• Too little primary vertex track absorption in MC
• This was a surprisingly large part of the problem
  and could bear further investigation. It probably
  softened most of the mode dependence.

45
But did all of work this fix the K3? lifetime?
MC fixes did not fix the lifetime problem. Since
we are able to get a good match between data and
MC for the BeO slabs, but a poor match to the
Tarsil and TR1, we performed fits with an
additional cut that the primary vertex is in one
of the 4 BeO slabs. This helped only slightly.
Four of the K3? split sample fits with f(t)
corrections are illustrated below with this BeO
cut. Data is from MCS period 6. Based on the
disagreement between the Z distributions between
MC and data, we tried 4 different cuts on the
maximum second vertex Z positions which are color
coded. A huge discrepancy ( ?22 fs!) is present
for Zpgt-3 and Pgt70. The sign depends on whether
or not the cut is upstream or downstream of the
2nd Tarsil. The Zs lt -0.5 sample seems to have
acquired a slope in the sub-sample number.
Zpgt -3 Plt70
Zpgt -3 Pgt70
Zplt -3 Plt70
Zplt -3 Pgt70
l/?gt11 Wkgt2
46
How big are the f(t) corrections?

• The first 4 points are for the sub-samples with
  no MC correction. The second 4 shows results
  after the f(t) corrections and are taken from the
  previous slide. The pattern on the last (Zpgt -3
  Pgt 70) sub-sample appears to be
• The Zslt 0 -0.5 points are under corrected by
  -22 fs. But the MC corrections are enormous.
• The Zslt 2 1 points are over corrected by 22
  fs. But the MC corrections are small.

47
Features of the Zp lt 2 Fit
zgt-3 pgt70
We show the mass distributions used in the 4th
sub sample (Zpgt -3 Pgt 70) fit. The fit uses 6
bins of 300 fs. We also show the f(t) corrections
used for each subsample fit. Finally we show the
yield versus t for the 4 subsamples. Log scales
are used throughout. A clear non-Gaussian tail is
present in the mass plot but the estimates from
sideband subtracted fits are consistent with
those from mass fits. A nearly exponential
lifetime is observed for these fits since f(t)
corrections are relatively mild.
??
48
Features of the Zp lt -0.5 Fit
zgt-3 pgt70
yield versus t
In this fit, the f(t) corrections are enormous
for both the 3rd and 4th subsample. The last time
bin in the 4th subsample has very low statistics
in the mass fit, but results from direct sideband
and mass fit are consistent.
f(t)
49
What more can be done?

• Do fits with separate f(t) for each slab.
• Correct any minute differences in K3? momenta
  spectrum between data and MC by rejection and try
  again.
• The (Zgt-3 Pgt70) sub-sample is probably very
  sensitive to any P discrepancy
• But average momenta in MC and data are close
  (84.62 data vrs 84.95 MC)
• Just completed but to no avail! (see next slide)
• Try to resolve problems with Tarsil and SSD
  simulation.
• Problems with pass number microrico agreement,
  fhi ratio, and K3? width when primary near
  tarsil.
• Its clear we need more people working on
  understanding the MC if we ever intend to publish
  measurements.

???
50
Correcting for differences in mom spec
rgmcsjw MC with phoc data
rgmcsjw MC data
There were slight differences in the K3? momentum
between the MC (with corrected absorption) and
the background subtracted data. We performed
rejection on the MC based on the ratio of data/mc
yields in each momentum bin. We also limited P to
range 30 ltPlt175 GeV. This new MC was used to
supply the f(t) correction and the fits were
performed again. Unfortunately very little change
was observed in the fitted values.
Zpgt -3 Plt70
Zplt -3 Pgt70
Zpgt -3 Pgt70
Zplt -3 Plt70
Corrected with phoc MC
51
A massive problem from -0.5ltZseclt2 ??
Given the dramatic lifetime difference one gets
as one varies the Zsec cut from -0.5 to 2, there
must be some terrible problem with the MC in this
region. Amazingly enough the MC (generated with
427 fs) and the data are in excellent agreement.
This distribution has the 4th split sample cuts
of Zpgt -3 and Pgt 70 and Wkgt2 l/?gt11.
Given the simulation looks fine in this region,
could the problem be with the fit?
52
Replay the MC through the Fit

• To check on possible biases, we replayed the MC
  back through the fit.
• This is a cheat since the same MC was used to
  evaluate the f(t) correction.
• We see two disturbing things
• The fit returns a significantly higher lifetime
  than the MC input lifetime.
• We get a nearly identical pattern among the 3
  split samples as in the data.

Not able to get this same pattern so far when we
use analyze MC data set which is independent of
the f(t) set