D_s* -> D_s γ
from D_s^* D_s @ 4160 MeV

D_s tag side

More complete tagging plots than what I presented last week. I've superimposed the K_s π π⁰ fit background on the MC determined background. BG in the signal region is above the fit by ~150 events, which is roughly how far the signal overshoots. The overshoot is from the charm part of the backgroud (as compared to continuum).

Is this common to K_s modes? K_sK and K_sK_sπ seem to have similar peaks, although their sharper (double gaussian) peaks aren't affected as much. The two K_sKππ modes don't seem to have a background peak near the signal region.

Status: I'm adding K_s info to my ntuples anyway (I currently don't preserve the information after making cuts), so I'll wait until those are done before looking into it more (unless someone has a bright idea).

D_s SL side: φ optimization

I've looked at signal/background counts and S²/(S+B) for a variety of cuts (φ mass, hit fraction, RICH PID on/off, dE/dx σ).
For simplicity, I only count in the signal D_s mass region.

Lowest frequency: M_φ (5 cuts)
2nd lowest frequency: RICH on/off toggle (2 cuts)
2nd highest frequency: Hit fraction (.3, .5, Tim, .1)
Highest frequency: dE/dx (# σ: 2.0, 2.5, 3.0, 4.5, 6.0)

Higher dE/dx better? Do we need PID at all?

Also note: different D_s modes have very different levels of BG (particularly neutrals).

Final M_Ds after electron cuts (f w/ RICH, TQ) and φ cuts (3 σ dE/dx, effectively no HF, z₀ and r₀ cuts at 5 cm/5 mm). Same w/ 6 σ dE/dx.

Neutrino missing mass with "naive" cuts, without the γ and with the γ

D_s tag side

I've gone back and formalized my D_s tag fitting procedure (although not my final D_s cuts). I now use a crystal ball shape on the appropriate modes (π π π and modes with neutrals).

The procedure:

• Fit to the MC truth shape, using a narrow gaussian + wide gaussian/crystal ball shape. The mean for both shapes is set to be the same, with all other parameters left to float (5 or 7 parameters).
• Correct the area for a systematic undercounting of events with these shapes (+1% correction to fits).
• Fit to the reconstructed shape, fixing N_wide/N_narrow and σ_wide/σ_narrow. At present, I also fix σ_narrow, although it looks like I can let it float.

Results from the reconstructed fits (10x sample, D_s⁺):

Mode	Fit estimate	True counts	# σ
KsK	39693.6 +/- 228.711	38670	4.47546
KKπ	119085 +/- 444.066	119291	-0.464916
KsKπ0	17656.8 +/- 263.991	15448	8.36704
KsKsπ	7842.23 +/- 145.61	7412	2.95469
KKππ0	44502.4 +/- 401.243	44324	0.444508
KsK+ππ	13176.8 +/- 220.204	12987	0.861939
KsK-ππ	23184.2 +/- 208.209	23035	0.716536
πππ	34757.6 +/- 419.867	34651	0.253968
πη	18666 +/- 247.57	19324	-2.65796
ππ0η	447.117 +/- 116.557	596	-1.27734
πη',η'→ππη	11673.3 +/- 125.393	11782	-0.866951
ππ0η',η'→ππη	255.409 +/- 29.319	217	1.31003
πη',η'→ργ	28745.8 +/- 393.813	27665	2.74441
Total	359,686 +/- 1,008	355,402	4.25

Note: I should have at least a ~6% statistical error across the full sample due to the numerator (semileptonic counts).
Note 2: My 1% difference in fit/true is roughly the same size as the correction that I make to force the MC fit counts to match the MC true counts.

Future D_s tagging plans:

• Make sure that the - sign D_s fits are reasonable (independent cross-check).
• Make sure the procedure works (fits converge and give reasonable results) on data-sized samples.
• Check a charged and neutral mode in D⁺ MC and data to make sure that my procedure holds up in real events.

Ignore this

D_s tag side

Using new fitting technique on MC only plots.
Red = Gauss + Crystal Ball. Otherwise, double gaussian.
I scale the fit estimate up by 1.00969 due to systematic underestimation for my chosen shapes.

Mode	Fit estimate	True counts	# σ
KsK	38757.1 +/- 195.921	38645	0.571918
KKπ	118881 +/- 343.133	119158	-0.806379
KsKπ0	15434.8 +/- 123.646	15415	0.160346
KsKsπ	7382.04 +/- 85.5056	7406	-0.280169
KKππ0	44377.3 +/- 209.654	44235	0.678834
KsK+ππ	12896.9 +/- 113.02	12975	-0.690772
KsK-ππ	22852.8 +/- 150.444	23020	-1.1114
πππ	34726.7 +/- 185.463	34570	0.845139
πη	19300.4 +/- 138.271	19238	0.451122
ππ0η	546.539 +/- 23.4078	592	-1.94212
πη',η'→ππη	11683.2 +/- 107.575	11767	-0.779306
ππ0η',η'→ππη	238.161 +/- 18.0388	217	1.17311
πη',η'→ργ	27777.7 +/- 165.866	27625	0.920358

Using new fitting technique on charm generic plots (no cont).
Red = Gauss + Crystal Ball. Otherwise, double gaussian.
I still scale the fit estimate up by 1.00969.

Mode	Fit estimate	True counts	# σ
KsK	39198.7 +/- 215.312	38670	2.45534
KKπ	117541 +/- 438.318	119291	-3.99275
KsKπ0	16630.7 +/- 261.26	15448	4.52695
KsKsπ	7723.68 +/- 143.484	7412	2.17222
KKππ0	42878.6 +/- 396.859	44324	-3.64199
KsK+ππ	12743.9 +/- 216.619	12987	-1.12218
KsK-ππ	23191.4 +/- 220.077	23035	0.710449
πππ	35174.2 +/- 276.428	34651	1.8926
πη	18697.3 +/- 157.536	19324	-3.9781
ππ0η	538.955 +/- 62.0912	596	-0.918732
πη',η'→ππη	11631 +/- 116.831	11782	-1.29239
ππ0η',η'→ππη	259.991 +/- 26.1085	217	1.64664
πη',η'→ργ	27494.6 +/- 258.807	27665	-0.658482
Total	353,704 +/- 875.545	355,402	-1.969

D_s tag side I currently fit the D_s mass spectrum after semileptonic cuts (the SL side) using the MC histograms pulled from the D_s tags (i.e. make D_s cuts, then use the MC truth tagged shape).

In the final analysis, I plan to use an analytic shape to fit my D_s tags, then use that same shape to fit my SL side. The hope is that this is more flexible to any differences in the MC/data (e.g. mass resolution) while still getting a systematic error cancelation.

• Step 1: Extend "Peter-like" cuts to all 13 modes.
• Step 2: Fit using a double gaussian + linear.

# sigma: 1.07044
# sigma: -5.65049
# sigma: 1.03628
# sigma: 1.44001
# sigma: -10.5182
# sigma: -0.360476
# sigma: -0.199277
# sigma: -5.62274
# sigma: -3.54609
# sigma: -213.959
# sigma: -1.88557
# sigma: 6.66927

D_s tag side

The current difference between my D_s tags and Peter's is ~.15%.
Comparison plots of my D_s⁺ tags to Peter's D_s tags.

The differences (in KKππ⁰, the worst offender):

• My charge-summed plots and Peter's charge-summed plots were defined differently (~3%)
• I was using the K fit for the track momentum instead of the π fit. This caused a problem with the 100 MeV π fit, which should reject soft D* daughter π (my π were coming out to ~120 MeV instead of ~80 MeV) (~5%)
• Our recoil mass cuts were a little bit different:
  Mine:
  • Veto all events with M_recoil < 2099 MeV (true should be ~2112 MeV).
  • Veto on track momenta, KKπ mass, etc.
  • Choose best candidate for each charge by minimizing |M_recoil - 2112 MeV|
  Peter's:
  • Veto all events with M_recoil < 2051 MeV.
  • Veto on track momenta, KKπ mass, etc.
  • Choose best candidate for each charge by minimizing |M_recoil - 2112 MeV|
  • Veto any best candidate with M_recoil < 2099 MeV.

SL side

I fit to the D_s mass for all events that pass a φ and e cut, using three different methods (10x sample):

• Sum all modes & fit to total with MC shape + linear BG
• Sum all neutral modes (w/ γ) and charged-only modes (e.g. KKπ) & fit each to MC shape + linear BG
• Fit each mode to MC shape + linear BG

• Each fit has 3 parameters. The D_s mass MC shape is taken from true D_s on the tag side (not just semileptonic events).
• Multiple candidates allowed -- only relevant for neutral modes due to extra tracks cut

Results of fit for each mode (10x sample):

Count SL side counts:

• After trying to get a systematic procedure/BG estimate for the cut and count method of getting the φ e ν BR, I migrated toward using a fit on the D_s mass.
• I made a simple minded fit using the crystal ball and putting all modes into one plot, allowing multiple candidates, and choosing a best candidate by lowest U (or by best φ, which still allows multiple candidates for neutrals).

  Results in 10x sample are close for an obviously flawed signal shape -- 3057.3 +/- 86.2 when 3108 true, and 2797.5 +/- 79.9 when 2882 true, for multiple candidates and best φ, respectively.

• My future method (in progress) is to use the true D_s mass shape from MC (taken from the tag side), and fit that + a linear BG for each of my 13 modes.

• I (blindly) looked at a variety of cuts (e.g. pull mass, # σ, etc.) on the π⁰ -- no luck in reproducing Peter's plots
• Peter finally had some time & was able to produce an event list for me!
  • Big issue: His event list and my event list differ in raw counts by about 2% (his higher), but our plots are different at the 15% level (mine higher) -- something's strange... Revisit assumptions:
    • Are the plots Peter gave me the same as the plots in his CBX?
      Yes. Looking at his CBX shows the plots to be identical to the comparable Roofit ones he gave me.
    • Are the plots I'm using for comparison the plots that are from his CBX?
      No. The CBX gives the D_s⁺ and D_s^- plots individually, but I'm using his summed plot to compare against my summed plots.
    • I initially tested the summed plot in the modes K_sK and KKπ. Did I miss something in those modes?
      No. I had these modes pegged correctly.
    • If summing + and - plots works in K_sK and KKπ, it should be good all the time, right?
      No. After all, a D_s⁺ will become a D_s^- just by swapping out a π⁺ and a π^-
  • Aside: it looks like there are some E₉/E₂₅ differences in our showers

Future Plans:

• Make sure Peter & my KKππ⁰ + and - plots compare well.
• Finish SL fits with MC truth D_s mass shape + linear function

Note: BR(D_s → φ e ν) = N_SL+tag / (N_tag * ε_SL)

Outstanding analysis issues:

• D_s Tag
  
  • Peter did 6 D_s modes, I have 13. Make the other 7 modes more "Peter-like" for consistency (and, of course, so that I can more easily adopt his systematics)
    • Recoil mass cut instead of M_bc (both effectively momentum cuts on the D_s
    • See if any of Peter's sytematics need to be redone/adapted for the new modes
    • Remove intermediate particles in the other 7 modes to smooth out the background for the # of tags fit
  • **Understand my peaking background in D_s tags (from true D_s, other modes). This also involves taking a second look at my D_s truth tagging.
  • **Figure out my π⁰ difference relative to Peter
  • Add continuum, radiative return, and τ to my tagging
  • Fit using Peter's functions to match his actual number of tags -- also, count up the statistical and systematic errors
• ε_SL
  
  • Check (correlated w/ SL) systematics when have final cuts
• D_s SL side
  
  • **Get an estimate for the level of background (passing non-φ e ν)
  • Get a systematics procedure
  • Optimize cuts (φ/K, best candidate D_s)
  • Add in η e ν

Requested loose φ/γ comparison
I've implemented some very simple γ cuts (my looser shower quality + an energy cut from 105 MeV to 177 MeV) and looked at what effect this would have on my analysis. Note that my efficiency for getting the correct photon is closer to 70% than to the table's 87%. No MM* cut is made (perhaps this was the point?).
Note: This is 20x the half sample (Koloina would have ~1,000 events in this sample)

Cuts	Passing φ e ν events	Charm, not φ e ν	Relative fraction of events
Loose φ, no γ (my standard)	2,815	84 [2.9%]	100% (defined)
Loose φ, has γ (my standard)	2,445	57 [2.3%]	87%
Normal φ, no γ (my standard)	1,901	39 [2.0%]	68%
Normal φ, has γ (my standard)	1,652	26 [1.5%]	59%

Determining amount of background

There are generally two sources of background that pass all cuts:

• D_s semileptonic to some other mode (often η e ν). This is ~1/3 of the estimated MC background
• D(_s)(*)D(_s)(*) to semileptonic (e.g. K e ν) or even leptonic modes (e.g. τ ν) with random combinations for the D_s and sometimes the φ

The extent to which I care:
      I expect ~300 events in the full sample (~150 in the half sample), which gives a statistical error of ~6%. With my existing φ and extra track cuts, my background accounts for about ~3% of the observed events.

Bottom line:
      I don't really have a feel for how much I can trust my Monte Carlo on the random BG combinations (depends on the non-SL/L side of D decays), but I only need to be confident to within a factor of ~2.

Note:
      BG shouldn't (and doesn't appear to) peak in D_s mass. Right now, I'm making the φ cuts, e cuts, and most of the D_s cuts, then looking at the D_s mass (which I normally cut on at 3 σ). My goal is to make the non-SL background stand out from the SL bad-D_s combinations.

Possible explanation: e cuts should be the same, but I should have a higher efficiency with D_s, φ, and the D_s* γ.

Investigation into φ+e:
I have looser cuts on the φ in three main areas:

• My φ mass cut is 30 MeV (overkill) instead of 10 MeV
• My track quality cuts are looser for the φ's daughter kaon only (10 cm/mm instead of 5 cm/mm)
• My minimum hit fraction cut is .1 (no real HF cut) instead of .5

Hit fraction cut:
Here's the efficiency of the hit fraction cut, by φ momentum.
This does not include BR(φ → KK), so the maximum possible efficiency is 49.1%

pφ	Basic cuts, HF > 0.5	Basic cuts, Tim's HF	Basic cuts, no HF cut
0 GeV - 0.2 GeV (1,000)	5.9% (77.6%)	6.9% (90.8%)	7.6%
0.2 GeV - 0.4 GeV (6,514)	15.3% (83.4%)	17.2% (93.7%)	18.4%
0.4 GeV - 0.6 GeV (12,109)	29.4% (90.1%)	31.6% (96.8%)	32.6%
0.6 GeV - 0.8 GeV (9,732)	36.4% (93.7%)	38.2% (98.3%)	38.9%
0.8 GeV - 1.0 GeV (2,889)	39.4% (95.5%)	40.8% (98.9%)	41.3%
1.0 GeV+ (51)	31.4% (88.9%)	35.3% (100.0%)	35.3%
All p integrated (32,295)	28.9% (91.5%)	30.8% (97.5%)	31.6%

Track quality cut:
Here is the impact of the looser TQ cuts (10 cm/mm) when I don't have any hit fraction applied (there is a correlation between the HF cut and the TQ cut -- looser TQ only buys you ~2% after you make an HF cut)

pφ	10 cm/mm TQ & 10 MeV mass	5 cm/mm TQ & 10 MeV mass
0 GeV - 0.2 GeV (1,000)	9.7%	7.6% (78.3%)
0.2 GeV - 0.4 GeV (6,514)	20.3%	18.4% (90.6%)
0.4 GeV - 0.6 GeV (12,109)	35.8%	32.6% (91.1%)
0.6 GeV - 0.8 GeV (9,732)	41.6%	38.9% (93.4%)
0.8 GeV - 1.0 GeV (2,889)	43.5%	41.3% (94.9%)
1.0 GeV+ (51)	37.3%	35.3% (94.6%)
All p integrated (32,295)	34.3%	31.6% (92.3%)

φ mass cut
Here's the impact of an overly wide (30 MeV) φ mass cut to the normal φ mass cut (10 MeV).
I'm using the loose TQ and no hit fraction.

pφ	30 MeV mass	10 MeV mass
0 GeV - 0.2 GeV (1,000)	10.6%	9.7% (91.5%)
0.2 GeV - 0.4 GeV (6,514)	21.9%	20.3% (92.7%)
0.4 GeV - 0.6 GeV (12,109)	38.1%	35.8% (94.0%)
0.6 GeV - 0.8 GeV (9,732)	44.3%	41.6% (93.9%)
0.8 GeV - 1.0 GeV (2,889)	46.0%	43.5% (94.6%)
1.0 GeV+ (51)	43.1%	37.3% (86.5%)
All p integrated (32,295)	36.5%	34.3% (94.0%)

Overall impact of all the cuts on ε_SL
Rather than the φ efficiency, I use the semileptonic efficiency for this comparison (including the e, which is soft for high p_φ).
Errors are about 0.5% on my values and 0.3% on Koloina's quoted values (except the first and last rows).

εSL	Justin	Justin-faking-Koloina's cuts	Koloina quoted (independent sample)
0 GeV - 0.2 GeV	9.2% (100.0%)	5.2% (56.5%)	2.5% (27.7%)
0.2 GeV - 0.4 GeV	17.0% (100.0%)	11.8% (69.6%)	7.9% (46.8%)
0.4 GeV - 0.6 GeV	27.6% (100.0%)	21.2% (76.9%)	18.8% (68.1%)
0.6 GeV - 0.8 GeV	29.7% (100.0%)	24.5% (82.7%)	25.7% (86.7%)
0.8 GeV - 1.0 GeV	27.8% (100.0%)	23.6% (85.0%)	25.7% (92.5%)
1.0 GeV+	19.6% (100.0%)	13.7% (70.0%)	25.7% (131.3%)
All p integrated	25.5% (100.0%)	20.0% (78.5%)	17.7% (69.4%)

Summary of efficiency differences, Justin/Koloina

• Difference in finding the D_s*'s daughter γ: 59%
• Differnece in φ cuts: 69%
• Difference in D_s modes: 89%*

Comparison with Jon: I haven't done much recently. I'm putting this aside for now. We have a meeting Tuesday; perhaps this will be discussed.

Over all data (d39-d41 and d47,d48), I expect 296 φeν using Peter's D_s and 315 φeν using my D_s. About 5% of observed events are expected to be background from existing MC.

Cuts	(NS)1/2	NB
Peter	17.2	13
Justin	17.8	15

Investigating φ e ν BG
      D_s mass with good e, φ
      D_s mass with good e, φ, and D_s

A long story in one plot: My D_s faking Peter (blue lines) vs. Peter's D_s (dots). The best candidate is chosen (using recoil mass) for each D_s charge.

Cuts involved: DTagging cuts

Note: Peter only has one π⁰ mode, and this mode also has the highest number of multiple candidates (although KKπ and 3π also have several multiple candidates).

Current Status: I tried several different cut combinations, to no avail. Peter has been very helpful so far, but he hasn't responded since Saturday (prepping for CLEO meeting?). I'll bug him again next week.

In vs. Out redux -- η e ν

Sample	Passing SL+tag	Expected BR	Num σ
Data-size 1	261	2.66%	+0.81
Data-size 2	247	2.52%	-0.06
Data-size 3	231	2.36%	-1.06
Data-size 4	235	2.40%	-0.81
Data-size 5	276	2.82%	+1.81
Data-size 6	232	2.37%	-1.00
Data-size 7	249	2.54%	+0.06
Data-size 8	273	2.79%	+1.62
Data-size 9	254	2.59%	+0.38
Data-size 10	244	2.49%	-0.25
Blind d39	872	2.41%	-1.48
Blind d40	2,101	2.58%	+0.89
Blind d41	2,006	2.56%	+0.18
Total	4,979	2.54%	N/A

Overall reduced Χ² for the 20 samples: 1.13

These BRs are using:

• N_DsDs* = 5,643,806 (assume 1/20th per sample)
• N^{gen, MC}_{η e ν + tag} = 285,198
• N^{obs, MC}_{η e ν + tag} = 4,951
• 92% of passing events are signal φ e ν

Currently trying to replicate Peter Onyisi's D_s plots from CBX07-14.
I see invariant mass peaks, but no quantitative comparison yet :(

Comparison with Jon

η e ν

Current Work: Since my replication of Jon's analysis isn't terribly far along, I've just looked at the absolute issue: Why are 25% of my η e ν events BG?

BG events: There are many different types of BG events that pass my cuts. Several of these have π⁰ showers faking η showers, so I first made a splitoff and π⁰ rejection cut.
About 1/3 of the remainder are D_s → η' e ν, where the η' → π⁰ π⁰ η (no explicit rejection on extra π⁰s). Overall, ~55% of the remainder are D_s D_s* at all. The rest are primarily D⁰ semileptonic modes, where a D_s tag happens to form.

η Cut	η e ν events	Non-η e ν events	S2 / (S+B) (x103
Pullmass < 3.0	5,136	1,682	3.9
(PM < 3.0) + splitoff	4,885	1,485	3.7
(PM < 3.0) + SO + π0 rejection	4,246	687	3.7
(PM < 3.0) + SO + π0 + (MM2 < 0.5 GeV)2	4,032	211	3.8

MM² plot

In vs. Out redux -- φ e ν

Sample	Passing SL+tag	Expected BR
Data-size 1	182	2.22%
Data-size 2	168	2.04%
Data-size 3	140	1.70%
Data-size 4	168	2.04%
Data-size 5	176	2.14%
Data-size 6	161	1.96%
Data-size 7	162	1.97%
Data-size 8	169	2.06%
Blind d39	612	2.016%
Blind d40	1,356	1.986%
Blind d41	1,358	2.067%
Total	3,326	2.024%

These BRs are using:

• N_DsDs* = 5,643,806 (assume 1/20th per sample)
• N^{gen, MC}_{φ e ν + tag} = 227,124
• N^{obs, MC}_{φ e ν + tag} = 3,306
• 95.3% of passing events are signal φ e ν

Cross-check, Jon

Ds Tag Mode	Justin-faking Jon, recon	Jon recon	Justin-faking Jon, MC true	Jon fit tags
Ks K	103,100	100,603	26,705	29,138
K K π	572,000	595,918	85,015	93,229
Ks K π0	205,200	164,000	10,211	11,805
Ks Ks π	57,440	38,230	5,360	5,046
K K π π0	1,114,000	1,190,244	33,662	38,342
Ks K+ π π	299,300	224,505	10,478	10,229
Ks K- π π	193,000	156,093	17,688	17,988
π π π	263,200	281,210	30,999	33,307
π η	46,780	46,331	11,231	13,463
π π0 η	497,100	444,666	37,441	51,041
π η'	32,510	31,772	7,258	9,532
π π0 η'	117,400	111,743	8,155	12,181
π η' -> ρ γ	222,900	197,975	19,969	23,073

MC consistency check

SL Mode	D39 BR	D40 BR	DECAY.DEC
φ e ν	2.03% [± 0.01%]	2.00% [± 0.01%]	2.02%
η e ν	2.52% [± 0.01%]	2.53% [± 0.01%]	2.53%

Note: There were 1.04 M D_sD_s* in D39 and 2.34 M in D40. There were about 42 K φ e ν in D39 and 94 K in D40.

In vs. Out Test

This yields a branching ratio for φ e ν of (2.02 ± 0.16)% based on 164 * 0.96 = 157 blind, reconstructed events. DECAY.DEC value is 2.020%.
I also get a branching ratio for η e ν of (2.96 ± 0.17)% based on 386 * 0.744 = 287 blind, reconstructed events. DECAY.DEC value is 2.530%.
Note: For η e ν, the ratio of reconstructed events in each sample doesn't match ratio of sample sizes by ~14% -- this accounts for essentially all of the surplus.

Cross-check, Jon

Ds Tag Mode	Justin MC true tags	Jon fit tags
Ks K	56,861	58,276
K K π	170,540	186,458
Ks K π0	22,863	23,610
Ks Ks π	14,004	10,092
K K π π0	67,455	76,684
Ks K+ π π	24,787	20,458
Ks K- π π	42,966	35,976
π π π	62,035	66,614
π η	22,491	26,926
π π0 η	74,920	102,082
π η'	14,473	19,064
π π0 η'	16,336	24,362
π η' -> ρ γ	40,588	46,146
Total	630,319	696,748

Full generic sample, tag+γ+SL
Using my approximation of Jon's cuts (D39 + D40 only):

SL Mode	Justin true	Jon true	Justin BG	Jon BG
φ e ν	963	1,306	34	62
η e ν	1,747	2,190	1,292	72

Aside: Cross-check with Koloina's shows similar issue. There is some electron oddness (high fakes) in both. Looking into it...

Comparison with results of other groups

Group Status Summary:

Group/cuts	Scaled MC events	Data events
Syracuse	45.6	43.5 ± 6.7
Rochester	65.3	67
Mine, no γ	162	--
Mine, 70%-80% εγ	113-130	--
Mine, 70%-80% εγ, tight K and 10 MeV φ	89-102	--

Cross-check: η e ν

φ/η consistency

Event cuts	Ratio
φ e ν: (My loose K/φ w/70% γ)/(Jon's)	1.73
φ e ν: (My best guess K/φ w/70% γ)/(Jon's)	1.53
η e ν: (My η w/70% γ)/(Jon's)	1.55

In vs. Out

Blind/MC consistency check:

• In blind data-sized MC sample made from dataset 40, 164 events pass all cuts.
• From dataset 39 MC (a 3.6x current data sample), estimate that 96% of passing events are signal.
• In dataset 39 MC, have 586 signal events passing.
• This means that we should have
  BR^meas./BR^decay.dec = 164 * 0.96 * 3.6 / 586 = 0.971 ± 0.076

Event quality cut

• Extra tracks
• Correct charges in event (D_s vs. electron)
• No multi-use of tracks (rejection of a track used in both the D_s and a φ's daughter kaon)

Previously:

Cuts	φ e ν	DsDs*	Charm	Cont	Charm - DsDs*	DsDs* - φ e ν
Pass e cuts	3,247	14,232	45,830	2,359	31,598	10,985
Pass φ cuts	1,267	3,279	4,972	88	1,693	2,012
Pass event quality cuts (no multi-use of tracks, charges match, no extra tracks)	588	604	618	0	14	16

Now:

Cuts	φ e ν	DsDs*	Charm	Cont	Charm - DsDs*	DsDs* - φ e ν
Pass e cuts	3,247	14,232	45,830	2,359	31,598	10,985
Pass φ cuts	1,267	3,279	4,972	88	1,693	2,012
No extra tracks	872	1,926	2,253	12	327	1,054
Proper charges	852	1,885	2,136	7	251	1,033
No multi-use (Ds)	588	604	618	0	14	16
No e/K multi-use	586	600	612	0	12	14

These multi-use events come almost entirely from KKπ (mode 1) and KKππ⁰ (mode 4).
The tracks are lowish momentum kaons (~200-400 MeV) that are not usually from the semileptonic φ (they are from the D_s ~2/3 of the time, with about 2/3 of the remainder being incorrect in both the D_s and the φ).

This should mean that a tighter φ cut will make some of these events get cut at the φ cut stage instead of the multi-use cut.

With 10 MeV γ cut:

Cuts	φ e ν	DsDs*	Charm	Charm - DsDs*	DsDs* - φ e ν
Pass e cuts	3,247	14,232	45,830	31,598	10,985
Pass φ cuts	1,085	2,714	3,510	796	1,629
No extra tracks	719	1,578	1,736	158	859
Proper charges	703	1,549	1,678	129	846
No multi-use (Ds)	486	492	498	6	6
No e/K multi-use	485	491	496	5	6

Looked into background and the cause for QQId-wrong particles in semileptonic events

Background Investigation:

Cuts	φ e ν	DsDs*	Charm	Cont	Charm - DsDs*	DsDs* - φ e ν
All events			7,142,659*	2,068,865*
Has Ds w/ 1920 < mDs < 2015 1980 < Mbc < 2100	6,178	248,348	1,023,371	262,459	775,023	242,170
Pass 3σ mDs	5,001	219,574	681,262	173,038	461,688	214,573
Pass 2005 < Mbc < 2075	4,547	206,938	518,788	117,746	311,850	202,391
Pass e cuts	3,247	14,232	45,830	2,359	31,598	10,985
Pass φ cuts	1,267	3,279	4,972	88	1,693	2,012
Pass event quality cuts (no multi-use of tracks, charges match, no extra tracks)	588	604	618	0	14	16

* from generated information, charm should be 9.25M and continuum should be 4.02M

Looking into QQIded-wrong γ in semileptonic events

QQId wrong particle	Events	%
All events	4,909	100%
Electron	0	0%
Ds	262	5.3%
φ	362	7.4% (4.6% w/ tight TQ)
γ	4,832	98.4%
γ (splitoff O.K.)	1,824	37%

Checked particle and parent QQId info for γ

Checked γ multiplicity in reconstructed events

Reconstruct without γ

U plot for ν+γ (since no γ reconstruction)

Type	# combinations	# events
All reconstructed	1591	1535
SL, correct QQIds	1450	1436*
SL, mis-QQIds	99	51
Non-SL	42	45

* 1386 unique; 38 with at least one mis-QQId; 12 with multiple correct QQId combinations (these go away with hf > 0.5 cut on kaons)

To add:

• Reasons for misQQIds
• Event mult. over all 13 modes

BR(D_s → heν) = (# tag + SL) / (#tag * ε_SL)

Currently looking at the (#tag + SL) part.

The Problem?: Often get extra photons in valid events: Missing mass plot (or U=E_miss - p_miss plot)

Intermediate Solution: Plot E_γ and fit.

Final Solution?:

1. Estimate the number of events where we only get a false photon in a valid event (don't trust the MC unless forced to). In MC, 6.9% of signal events pass with all particles tagged correctly (blue); 2.9% pass when missing at least one particle (red not blue).
2. Use these events for additional statistics.

Analysis summary

Ultimate Goal: Measure BR(D_s → heν) where h ⊂ (φ, η, η').

Primary Challenge: ν is unobservable.

Other Challenge: D_s mesons occur predominantly in D_s D_s*.

General Procedure:

• Target D_s D_s* events, where D_s* → D_s γ
• Isolate events by reconstructing all particles except the ν and show its existence via missing mass or equivalent (U = E_miss - p_miss)¹
• Get a BR by counting tags (defined as everything except the semileptonic D_s: here a D_s + γ) and seeing how often candidate events occur when such a tag exists:
  BR(D_s → heν) = (# tag + SL) / (#tag * ε_SL)
• Overall, need to:
  • Count # D_s + γ tags
  • Measure ε_SL (using MC only)
  • Count # tag + SL events

Counting D_s + γ tags:

• Choose several D_s modes to reconstruct for the tagging side:
  • Want a large ε * BR
  • Background is non-trivial for most modes, so most statistical power comes from a handful of available D_s modes
  • Originally used 8 D_s modes to intermediate particles (e.g. φ π, K* K, η ρ).
  • Now use 13 D_s modes to (mostly) final particles (e.g. KK π, η π π⁰)
    • Original modes were generally cleaner, but not hugely so. The same statistical power is probably kept for the primary modes (e.g. KKπ) because the background is so low. Additionally, some modes like φ π vs. f₀ π were not really distinguishable anyway. Other modes, like the ρ modes had such weak intermediate particle cuts that background wasn't substantially reduced, and it was simpler to allow non-resonant π π⁰ to be treated as tags along with the ρ

  ---

  
  (2/14/08)

  D_s semileptonic tags

  Changed D_s mass bin size from 10 MeV to 5 MeV.
  20x sample fit/expected peaking BG (1st iteration): 1.0195
  20x sample fit/expected signal (1st iteration): 1.01395
  20x sample fit/expected peaking BG (2nd iteration): 0.986421
  

  Tags vs. D_s mass with 5 MeV bins

  ---

  
  (2/7/08)

  D_s semileptonic tags

  Results of iterated fits (use first fit signal estimate to get better peaking BG estimate):

  File #	MC tagged signal (center)	Estimated, MC shape(center)	Estimated, MC shape iterated (center)	MC tagged signal (all)	Estimated, MC shape (all)	Estimated, MC shape iterated (all)
  1	890	941.189 +/- 97.4627	962.483 +/- 97.5407	916	956.047 +/- 130.686	997.928 +/- 130.801
  2	959	911.32 +/- 99.0369	933.594 +/- 99.1153	979	915.49 +/- 132.519	959.335 +/- 132.635
  3	871	1148.93 +/- 97.8826	1173.61 +/- 97.9706	902	1220.35 +/- 129.043	1268.68 +/- 129.175
  4	921	973.769 +/- 98.9356	996.283 +/- 99.0167	944	984.966 +/- 131.205	1029.05 +/- 131.326
  5	937	944.583 +/- 101.582	965.483 +/- 101.654	969	865.309 +/- 135.542	906.636 +/- 135.652
  6	968	1087.49 +/- 100.7	1111.49 +/- 100.785	991	1085.33 +/- 132.732	1132.4 +/- 132.859
  7	927	947.992 +/- 100.285	968.476 +/- 100.359	954	981.336 +/- 132.182	1021.58 +/- 132.292
  8	969	966.406 +/- 100.044	986.64 +/- 100.117	1006	890.905 +/- 132.48	930.657 +/- 132.589
  9	871	696.619 +/- 95.9886	712.196 +/- 96.0467	897	708.634 +/- 128.465	739.329 +/- 128.551
  10	949	928.467 +/- 99.7403	950.562 +/- 99.8177	980	889.885 +/- 132.685	933.348 +/- 132.802
  11	922	974.944 +/- 99.0731	995.334 +/- 99.1465	944	931.347 +/- 131.751	971.286 +/- 131.861
  12	936	784.568 +/- 99.1069	802.024 +/- 99.1702	960	744.516 +/- 132.795	778.913 +/- 132.888
  13	918	880.631 +/- 97.9478	898.745 +/- 98.0114	945	926.175 +/- 130.777	962.027 +/- 130.872
  14	907	880.27 +/- 99.7006	900.963 +/- 99.7754	930	878.651 +/- 132.318	919.465 +/- 132.43
  15	940	1086.18 +/- 101.801	1109.88 +/- 101.885	959	1114.3 +/- 132.905	1160.75 +/- 133.031
  16	914	973.522 +/- 101.78	997.386 +/- 101.864	937	946.152 +/- 134.859	992.823 +/- 134.982
  17	919	979.718 +/- 100.436	998.934 +/- 100.503	950	955.906 +/- 132.715	993.672 +/- 132.816
  18	951	975.983 +/- 99.8187	998.542 +/- 99.8975	988	1001 +/- 132.437	1045.33 +/- 132.554
  19	941	930.06 +/- 99.1743	949.599 +/- 99.2428	959	942.326 +/- 131.76	980.886 +/- 131.864
  20	910	833.367 +/- 96.8454	853.105 +/- 96.916	944	824.036 +/- 130.125	862.897 +/- 130.23
  Χ2/20	N/A	0.998	1.142	N/A	0.803	0.877
  20x sample	18520	18764.7 +/- 444.372	19184.1 +/- 444.706	19054	18585.3 +/- 590.263	19409.6 +/- 590.761

  Over 20x sample, first estimate for peaking BG is 1.88% high; second estimate after repeating fit is 1.44% low.

  ---

  
  (1/31/08)

  D_s semileptonic tags

  Signal and peaking BG by D_s mass.

  Fit to 20x sample

  .

  Data size sample #	MC tagged signal (center)	Signal using MC shape (center)	MC tagged signal (all)	Signal using MC shape (all)
  5	937	955.399 +/- 101.619	969	886.696 +/- 135.599
  6	968	1167.98 +/- 100.985	991	1243.25 +/- 133.157
  7	927	1013.62 +/- 100.522	954	1110.21 +/- 132.534
  8	969	1052.89 +/- 100.356	1006	1060.8 +/- 132.945
  9	871	749.394 +/- 96.1856	897	812.62 +/- 128.756
  18520	18760.3 +/- 444.368	19054	18576.5 +/- 590.258

  Working on:

  • Making bin widths parameters
  • (?) Iterate fit one more time (feed signal counts back in & find peaking BG again)
  • Extend to other modes (this is just π π⁰ η')
  • Double-check past work
  • Write summary of past work & results (CBX)

  Systematic ideas:

  • Use (data R/L)/(MC R/L) ratio as estimate for systematic
  • Use data R only to get weight, then use data L only

  ---

  
  (1/30/08)

  D_s semileptonic tags

  File sum plots

  ---

  
  (1/24/08)

  D_s semileptonic tags

  Current Goal(s):

  • Finalize procedure to remove peaking background from bad D_s with good γ
  • Get systematic from the peaking BG procedure and any other tagging systematics
  • Apply peaking BG subtraction procedure to all modes (currently testing in π π⁰, η' w/ η' → π π η since here peaking BG is roughly equal in magnitude to signal);

  Distribution of peaking BG/signal in 20x MC sample (ignore black line).

  Procedures:

  • Fit E_γ for every 10 MeV D_s mass slice
  • Estimate peaking BG by using low D_s mass sideband only.
  • Estimate peaking BG by using high D_s mass sideband only.
  • Estimate peaking BG by using an average of the low & high D_s mass sidebands.
  • Estimate peaking BG by assuming the MC has the correct relative amounts in each D_s mass bin and getting the overall weight from the sidebands.
  • Use zero peaking BG, fitting only the signal shape.

  Data size sample #	MC tagged signal	Signal using low sideband	Signal using high sideband	Signal using average sideband	Signal using MC shape	Signal assuming no peaking BG
  5	937	1070.75 ± 102.019	1247.57 ± 102.628	1141.35 ± 102.262	1064.29 ± 101.994	1868.06 ± 0
  6	968	1199.12 ± 101.096	1105.76 ± 100.765	1161.75 ± 100.964	1222.02 ± 101.177	1618.17 ± 0
  7	927	1002.82 ± 100.486	1131.56 ± 100.952	1054.25 ± 100.672	1077.72 ± 100.755	1548.43 ± 0
  8	969	969.068 ± 100.049	1231.97 ± 100.999	1073.96 ± 100.427	1115.7 ± 100.582	1576.87 ± 0
  9	871	656.695 ± 95.8351	981.392 ± 97.0517	786.126 ± 96.3191	807.095 ± 96.4012	1230.57 ± 0

  ---

  
  (1/17/08)

  D_s semileptonic tags

  Quick overview:

  • Goal: Find branching ratio (form factors?) for D_s → φ/η(/η'?) e ν
  • Procedure: Look in D_s D_s* events, where D_s* → D_s γ, one D_s goes to a tagged mode, and the other D_s has an electron and a φ/η (with zero missing mass or equivalent).
  • Tagged Modes:
    • 5 "Clean" Modes: K_s K, K K π, π π π, π η, π &eta' w/ η' → π π η
    • 3 "Medium" Modes:: K_s K_s π, K_s K⁺ π π, K_s K^- π π
    • 5 "Dirty" Modes:: K_s K π⁰, K K π π⁰, π π⁰ η, π π⁰, η' w/ η' → π π η, η' π w/ η' → &rho &gamma
  • Tags: Use D_s + γ as the tag, fitting charm+cont+signal histogram to a plot of E_γ (have also done recoil mass).
  • Semileptonic (full event): Find presence of electron and φ/η, cut on missing mass (or U), and fit to a plot of E_γ.
  Current semileptonic (full event) issues:
  • Perhaps drop missing mass/U cut altogether (i.e. there's effectively no background that fakes a tag + electron + φ/η)? Looks like efficiency in MC goes slightly up without added background, at least for φ studies.
  • "Background" is mostly non-D_s* daughter showers in real events (variety of causes). Some fraction of these events don't have a real shower found, so they might be used to improve statistics.
  Current tagging issues:
  • Problem: When looking at systematic errors from the histogram fit by considering other background shapes, I found that some of the non-clean modes had peaking background from bad D_s + real γ (typically from stealing a π or π⁰ from the other side D_s).
  • Solution: Assume that the MC has a reasonable shape for the peaking background, find the amount of peaking background from the D_s mass sidebands, then include this as a fixed amount in the fit.
  • Problem: Naive error calculations on the bad D_s sideband fits didn't give the expected errors. In a typical data-sized sample:

    Background	σN/N	Expected Error (1/√N or √(S+B)/S)
    Charm	6.8%	~1%
    Continuum	13.2%	~1%
    Peaking BG	13.2%	11.0%

  • Solution:
    • Note that the # of entries doesn't include underflow/overflow! (Corrected in above table)
    • Watch the correlations between each fit -- Charm & continuum have a correlation coefficient of -.97!
    • If I fix the continuum part of the fit, the errors come out as naively expected (1.5% for charm, 10% for signal).
    • Double-check by making 3 bin histogram, explicitly fitting the outer bins with charm only, fixing the charm weight, and fitting the middle with peaking background (σ_N/N vs. expected: 0.88% vs. 0.89% for outer bins, 10.35% vs. 10.33% for middle).
  Present Work:
  • Use flat assumption for peaking BG with D_s mass or something more sophisticated. Have sideband/center scaling of 1.17 assuming flat, 0.895 from counting.

  ---

  
  (11/15/07)

  D_s semileptonic tags

  Linear fit example

  ---

  
  (11/1/07)

  D_s semileptonic tags

  MC data-size sample fits
  New data fit (now 823 signal + 712 peaking BG).

  MC ratio of peaking BG to Signal is ~95% (double-check). Data ratio of peaking BG to signal is ~(86.5+/-3)%.

  Current mysteries:

  • Why are my errors still so high (in MC and in data)? Expect ~6% in signal and ~1% in charm BG; have ~12% in signal, ~6% in charm BG.
  • Different distribution of peaking BG by D_s mass in MC vs. data?

  ---

  
  (10/25/07)

  D_s semileptonic tags

  	Signal			Peaking BG
  Procedure	Truth	Fit Result	Ratio	Truth	Fit Result	Ratio
  (Original) Peaking part in charm BG shape	1091	1270	+16%	N/A	N/A	N/A
  Peaking BG determined from sideband (sample 1)	1122	1300	+16%	1203	1265	+5%
  Peaking BG determined from sideband (sample 2)	992	1090	-9%	1297	1122	-13%

  In data (previously 0 within errors; now 816 signal + 692 peaking BG).

  ---

  
  (10/11/07)

  D_s semileptonic tags

  E_γ for sideband region (fit with signal shape and Bad D_s shape).

  Signal shape vs. Bad D_s shape

  ---

  
  (10/4/07)

  D_s semileptonic tags

  A brief history of the D_s+γ tagging problem:
  

  • I needed an estimate for the systematic on using histograms as BG function for recoil mass vs. E_γ, so compared to polynomail function
  • The 5 "Clean" modes (K_sK, KKπ, πππ, πη, and πη') didn't have much of a difference in the # of signal events estimated using a polynomial vs. histogram BGs (except maybe πππ).
  • Several of the 8 "Unclean" modes had a large difference in the # of signal events found from the histogram BGs vs. the polynomial BGs. When there was a significant difference, the histogram method was the accurate one.
  • I identified the problem to be real E_γ pairing with bad D_s. The bad D_s tended to steal a π or π⁰ from the other side D_s.
  • My method so far has been to take 10 MeV slices of the D_s mass, fit the resultant E_γ spectrum with a histogram BG, and plot as a function of mass to determine the "true" signal.

  Systematic estimated by mode (# polynomial/# histogram - 100%)
  Ds Mode	Eγ	Recoil mass
  Ks K	2.73%	-1.87%
  K K π	-0.38%	-2.52%
  π π π	14.6%	-0.27%
  π η	1.1%	2.17%
  π η' (π π η)	0.36%	9.45%
  Ks K π0	82.0%	7.4%
  Ks Ks π	-17.4%	10.2%
  K K π π0	90.1%	-7.7%
  Ks K+ π π	16.3%	27.6%
  Ks K- π π	13.8%	3.7%
  π π0 η	-41.2%	-47.9%
  π π0 η' (π π η)	-5.8%	88.5%
  π η' (ρ γ)	-0.12%	30.2%

  
  For comparison, the worst statistical error in E_γ is K_s K_s π with 13%. "Clean" modes have statistical errors of ~3-6% in E_γ, or ~4-7% in recoil mass (with recoil mass being suspiciously high for several modes -- ~1.4σ after weighted average).

  Slice method results.

  ---

  
  (9/27/07)

  D_s semileptonic tags

  Slice D_s mass and compare bad D_s to signal events.

  ---

  
  (9/20/07)

  D_s semileptonic tags

  D_s mass and E_γ comparisons

  ---

  
  (9/13/07)

  D_s semileptonic tags

  Investigating good γ, bad D_s events:
  
  Correct mistake with good D_s, bad γ events
  D_s mass and M_bc

  ---

  
  (8/30/07)

  D_s semileptonic tags

  Tagging systematic studies:
  Fit with histogram and 6-degree polynomial, finding difference in number of events found:
  

  Mode	γ energy	Recoil mass
  Ks K	2.73%	1.87%
  K K π	0.38%	2.52%
  π π π	14.6%	0.27%
  π η	5.6%	6.26%
  π η' (π π η)	5.44%	8.53%

  Peaking BG discovery
  Peaking BG information

  ---

  
  (8/16/07)

  D_s semileptonic tags

  Results of different tagging procedures:
  D_s + γ tags

  Plots for different tagging procedures:
  Recoil Mass and polynomial background plots

  ---

  
  (7/24/07)

  D_s semileptonic tags

  Event Comparison

  ---

  
  (7/19/07)

  D_s semileptonic tags

  Focus on finalizing tagging:

  • Changed D_s mass cut from 1940 MeV - 1995 MeV to a 3 σ mass cut
  • Reviewed Recoil Mass as potential double-check for the # of tags -- still need to do fit
  • Tried making an M_bc or M_bc* cut to reduce background on tagging.

  Best Candidate U plot

  ---

  
  (7/12/07)

  D_s semileptonic tags

  • Tried different backgrounds for tags -- best were linear and N_charm*charm + N_cont*cont (best = low uncertainty and accurate)
  • Comparison of the tags from all 3 groups.
  • Tried different backgrounds for E_γ fit -- background histogram was still the best, although linear BG is good safety check.

  In vs. Out test:
  

  SL Mode	Decay.dec	MC true	Fit
  φ e ν	2.02%	2.06% (0.10%)	(2.07 ± 0.10)%
  η e ν	2.53%	2.50% (0.10%)	(2.64 ± 0.18)%

  In data:
  

  SL Mode	Justin	John (Roch.)	Koloina (Syr.)
  φ e ν	(2.42 ± 0.35)%	(2.263 ± 0.287)%	(2.51 ± 0.30)%
  η e ν	(2.26 ± 0.51)%	(2.045 ± 0.242)%	(2.440 ± 0.367)%
  η' e ν	???	(0.603 ± 0.248)%	(0.566 ± 0.283)%

  ---

  
  (7/5/07)

  D_s semileptonic tags

  Found two significant problems from last week:

  • FSR semileptonic events were being treated as non-semileptonic events (background).
  • There was a "stray" U cut of ± 50 MeV on my MC truth info, even for the no-U cut procedure.

  FSR Impact:
  

  • BR(φ e ν + X) / BR(φ e ν) = 1.15
  • BR(η e ν + X) / BR(η e ν) = 1.18
  DECAY.DEC branching ratios now reproduced with MC information:
  
  • BR(φ e &nu) = (2.014 ± 0.007)% [Input: 2.02%]
  • BR(η e &nu) = (2.531 ± 0.007)% [Input: 2.53%]

  U cut Impact:
  Efficiency difference:
  ε_{φ e ν, no U cut} / ε_{φ e ν, 50 MeV U cut} = 1.18
  ε_{η e ν, no U cut} / ε_{η e ν, 50 MeV U cut} = 1.34
  ε_{η' e ν, no U cut} / ε_{η' e ν, 50 MeV U cut} = 1.24
  I was using the no U efficiency on the 50 MeV U cut MC sample

  Overall impact:

  • FSR fix closes about 60% of the gap in my MC generated BR vs. MC reconstructed BR (prelim. check)
  • FSR problem should have a data impact since there was peaking background in my fit function.
  • U cut fix won't change the generated/reconstructed gap since the efficiency changes for both. Using no U cut makes the gap worse, since generated goes up while reconstructed value stays roughly the same (checking this.
  • U cut fix will change data value, since effective efficiency on data plots should have been lower. I'm now leaning toward no U cut in the final analysis, however.

  Current work/near-term plans:

  • Changing tag fitting from histogram background to function background
  • Reviewing effects of FSR/U cut fixes.
  • Writing up "note"

  ---

  
  (6/28/07)

  D_s semileptonic tags

  MC mock run through:
  

  • Reweighted charm/continuum to account for 5x vs. 20x samples (made low energy "hump" in E_γ background less prominent)
  • Used 1/3 of the MC to make "background" functions (includes right event, wrong particle -- usually D_s* γ);
    Tested on data sized sample made from other 2/3 of MC

  	φ e νCALC	φ e νMC TRUE	η e νCALC	η e νMC TRUE
  NTags (~3%)	17,577	~17.9K	17,577	~17.9K
  εSL (~4%)	--	15.8%	---	18.7%
  NAll Event	50.7 ± 8.9	46	72.8 ± 10.2	60
  BR(Ds → heν)	(1.83 ± 0.32)%	2.02%	(2.21 ± 0.31)%	2.53%

  Ran over data using background function from all MC:

  	φ e ν	η e ν
  NTags (~3%)	16,103 ± 365	16,103 ± 365
  εSL (~4%)	15.8%	18.7%
  NAll Event	32.2 ± 8.2	38.6 ± 10.8
  BR(Ds → heν)	(1.27 ± 0.32)%	(1.28 ± 0.28)%

  • Definitely inconsistent with Rochester & Syracuse (~2%/2.4%)
  Used all the MC to make "background" functions and tested on same sample:

  	φ e νCALC	φ e νMC TRUE	η e νCALC	η e νMC TRUE
  NTags (~3%)	17,952	~17.9K	17,952	~17.9K
  εSL (~4%)	--	15.8%	---	18.7%
  NAll Event	41.2 ± 10.0	46	47.6 ± 11.7	60
  BR(Ds -> heν)	(1.45 ± 0.35)%	2.02%	(1.42 ± 0.25)%	2.53%

  Two major concerns:

  • Fit function not getting the ideal value (and not getting the right number of background + signal from fit -- can be off by ~10%)
  • Number of MC "true" events lower than expected from input BR and prior efficiency (Note: means efficiency might be lower, which makes data BR worse!)

  Recent update:

  • Changed χ² fit to LL fit -- # signal + background from fit now very close to histogram (within 2-3%)
  • Made 8x data sized samples...

  ---

  
  (6/21/07)

  D_s semileptonic tags

  Summary of procedure

  D_s tagging:
  D_s "clean" modes in MC D_s "clean" mode fit to data 41

  U fit:
  φ e ν U plot over all MC types (charm, cont, radret, tau).

  ---

  
  (6/07/07)

  D_s semileptonic tags

  Recent work:
  • Converted from MN pds files to Cornell EventStore files:
    • Now have both ISR and No ISR files (pds were No ISR only)
    • No difference in E_γ (my tagging parameter) between ISR and No ISR.
    • Cornell does not have multi-body included while MN does (appears insignificant, but observable).
  • Looked into peaking background under E_γ:
    • Effect of D_s mass cut
    • Effect over different types of modes
    • Zoom in on a "bad" mode: ρ η
  • Started looking at D⁰ fitting

  Near future:

  • See what differences exist in other parameters (U, M_bc) between ISR and No ISR cases.
  • Do deeper checks with D⁰ fitting and compare to data
  • Get more complete D_s fits using histogram shape and convoluted functions

  ---

  
  (5/31/07)

  D_s semileptonic tags

  Recent work:
  • Comparing Photon energy resolutions (Note: Pete gets 5.62 +/- 0.06 MeV for both)
  • Big file with all E_γ plots for comparisons
  • Background for non-D_s D_s* and for D_s D_s* events
  • My recoil mass plots, emulating Pete's procedure

  ---

  
  (5/24/07)

  D_s semileptonic tags

  Recent work:
  • Solved MC mystery
  • Now have good E_γ plots
  • Fixed lab frame/beam CM frame issue
  • Added η' e ν
  • Went through η e ν U plots with Chul Su-style modes
  • Loosened D_s mass and M_bc pre-selection cuts to 1920-2015 MeV and 1980-2100 MeV, respectively
  • Fixed issue with E_γ boost (was using PDG D_s* mass instead of reconstructed)
  • Some D_s modes not in MC, or not tagged as such despite being in DECAY.DEC: φ π π π, K*⁰ K*⁺ π⁰
  • E_γ resolution difference with Pete: My σ = 7 MeV; Pete's = 5.5 MeV
    • Different samples?
    • runOnNewPass2 vs. setup_analysis?
  • Looked over E_γ plots by mode
  • Looked into different tagging using lorentz invariants directly
  • Looked into Recoil Mass inconsistencies; possibly still too tight D_s mass
  Near future:
  • D⁰* -> D⁰ γ tagging count, using histogram shape instead of function
  • Build infrastructure to run at Cornell

  ---

  
  (4/12/07)

  D_s semileptonic tags

  Current issue, probably with MC tagging.

  ---

  
  (2/8/07)

  D_s semileptonic

  Mode	# original MC events	# new DTag MC events (1/2 sample -> x2)
  φ π	68	90
  K* K	91	106
  Ks K	73	66

  ---

  
  (2/1/07)

  D_s semileptonic

  E_γ OR cuts

  ---

  
  (12/18/06)

  D_s semileptonic

  Boost while fixing p and E

  ---

  
  (12/14/06)

  D_s semileptonic

  E_γ fits:
  Background
  Everything
  Signal
  

  Procedure:

  • Fit background "region" (below 110 MeV and above 160 MeV) to f(x) = y₀ + A * xⁿ
  • Fixed those background parameters and added fit for gaussian + spread function (spread not convoluted)

  Results:

  Hypothesis	# in fit	# in MC	Fraction off
  Ds*	107,192	104,245	+2.8%
  Swap	96,017	104,245	-7.9%
  Average	101,605	104,245	-2.5%

  Notes:

  • Background peaks slightly (and broadly) in signal region. This has been isolated as being due to true γ + "bad" D_s (only have mass cut of 1940 MeV to 1995 MeV) {see below}
  • Because of the above point, the peak tends to fit slightly high relative to the signal peak.
  • Because of the non-convolution, the spread doesn't extend to the entire signal region.

  Peaking background:

  • Subset of background with MC good D_s and "bad" γ (not from D_s* or not real γ) {solid} and MC good γ with "bad" D_s {dashed} for the swap hypothesis and the D_s* hypothesis. Note that this is a relatively small subset of all background (see, e.g., the y-axis scale).
  • A plot of the D_s mass to see if there's a peak indicating D_s mesons with alternate tag modes. It doesn't look like there are alternate tag modes not accounted for, but there might be a soft peak there.
  • The total background with (pink) and without (dotted red) MC determined D_s* daughter γ.

  Next tagging questions:

  • Would a histogram fit work better than a function fit? (Try in D⁰* -- can look at data and E_γ shape should be similar)?
  • Is Pete's method better than ~10% (systematics probably lower)?

  ---

  
  (12/12/06)

  D_s semileptonic

  Full E_γ plot
  Signal part of E_γ plot

  ---

  
  (12/07/06)

  D_s semileptonic

  Example "complicated" E_γ fit (only bottom converged)
  Signal portion of E_γ fit. Total fit events: 17,700. Total MC count: 8,100.

  ---

  
  (12/05/06)

  D_s semileptonic

  Trying to get # tags from different variables (recoil mass, M_bc, E_γ): Current tag fit attempts

  ---

  
  (11/30/06)

  D_s semileptonic

  M_bc plots

  ---

  
  (11/28/06)

  D_s semileptonic

  U vs. cut variables

  ---

  
  (11/16/06)

  D_s semileptonic

  Initial attempt to duplicate Pete's results

  ---

  
  (11/14/06)

  D_s semileptonic

  Testing hybrid variables

  ---

  
  (11/9/06)

  D_s semileptonic

  Current fit procedure results

  Recoil Mass vs. various D_s* parameters

  Candidates per event

  ---

  
  (11/7/06)

  D_s semileptonic

  Looking into # (D_s+γ) tags for denominator of BR and # (D_s+γ+semileptonic) events for numerator (efficiency corrected)

  Getting # (D_s+γ) tags

  ---

  
  (10/30/06)

  D_s semileptonic

  D_s* exploration plots

  ---

  
  (10/26/06)

  D_s semileptonic

  D_s* exploration plots

  ---

  
  (10/24/06)

  D_s semileptonic

  Overview

  ---

  
  (10/17/06)

  D_s semileptonic

  Efficiency Review

  D_s Effect of D_s cuts

  Ds Mode	ε * BR	εJustin	εBrian
  φ π	17.9%	36.5%	36.7%
  φ ρ	4.1%	8.3%	8.9%
  η π	17.2%	43.8%	43.6%
  η ρ	7.0%	17.9%	15.7%
  η' π	4.8%	28.0%	27.2%
  η' ρ	1.6%	9.6%	9.1%
  K* K	26.7%	40.1%	37.9%
  Ks K	37.6%	54.6%	47.2%

  Total ε * BR = 3.44%

  γ

  Cut	εγ
  None	92.9%
  Not hot	91.6%
  No matched track	90.2%
  In barrel or endcap	87.5%
  E9/E25 OK or E9/E25 > 0.78	84.0% (81.2%)
  97 MeV < Eγ < 182 MeV	79.6%
  Splitoff OK	74.3%
  No π0 rejection	55.2%

  e
  Electron efficiencies

  pe	φ+e+ν	η+e+ν
  <200 MeV	12.6%	3.9%
  200-300 MeV	19.8%	7.4%
  300-600 MeV	61.0%	49.5%
  600+ MeV	6.6%	39.1%

  φ
  Phi cuts, momentum dist. and efficiencies

  ---

  
  (10/10/06)

  D_s semileptonic

  Chul Su comparison: ε_tag+γ+semi / ε_tag+γ

  Semilep Ds mode	Chul Su's figures	My typical cuts (loose)	Standard cuts
  φ+e+ν	21.0% ± 0.4%	28.8% ± 0.8%	20.2% ± 0.9%
  η+e+ν	45.9% ± 0.6%	44.2% ± 0.8%	41.0% ± 0.7%

  φ and η efficiencies, by momentum.

  electron efficiencies, by momentum (unconfirmed).

  ---

  
  (10/3/06)

  D_s semileptonic

  Chul Su semileptonic efficiencies

  Semilep Ds mode	εtag+γ+semi / εtag+γ
  φ+e+ν	21.0% ± 0.4%
  η+e+ν	45.9% ± 0.6%

  My prelim semileptonic efficiencies

  Tagged Ds mode	(φ+e+ν) εtag+γ+semi / εtag+γ	(η+e+ν) εtag+γ+semi / εtag+γ
  φ π	28.0%	35.7%
  φ ρ	20.0%	31.8%
  η π	23.5%	45.1%
  η ρ	--	81.5%
  η' π	29.1%	20.4%
  η' ρ	21.2%	36.4%
  K* K	30.9%	28.3%
  Ks K	29.6%	34.1%

  ---

  
  (9/19/06)

  D_s semileptonic current status

  Two types of examples where kaons miss TQ cuts: Decay in flight (top), and slightly off track reconstruction (side)

  Passing signal events with an MC failure (Red - Blue)
  • Efficiency for a signal event to pass with an MC correct: 6.90%
  • Efficiency for a signal event to pass without any events MC correct: 2.92%
  • Efficiency for a signal event to pass with an MC correct and without at least one MC correct: 3.1% (45% of those with MC correct)

  Particle	Fraction of the time missed	Fraction of the time as the only miss
  e	0%	0%
  φ	31.1%	14.4%
  Ds	12.1%	3.7%
  γ	80.6%	59.2%

  Attempt to find a parameter that separates a passing signal with correct reconstruction vs. passing signal with incorrect reconstruction.

  Near-term goals

  • Fix skimming/efficiency bug!!!
  • Check E₉/E₂₅ for Red/Blue separation
  • Redo efficiency breakdown (including "13 modes"), compare with Chul Su, and isolate any differences
  • Recheck optimal efficiencies

  ---

  
  (8/23/06)

  D_s semileptonic current status

  U plot integrated across all modes Cuts:
  • φ (not in D_s)
  • within 30 MeV
  • Kaon |σ_dE/dx| < 3.0
  • η (not in D_s)
  • Showers Good*
  • E_γ > 30 MeV
  • E_γ^barrel > 50 MeV
  • |σ_η| < 3.0
  • γ (not in D_s)
  • Shower Good*
  • 105 MeV < E_γ < 177 MeV {kinematics: (141 ± 27) MeV}
  • e
  • f with RICH > 0.9
  • D_s
  • 2005 MeV < M_bc < 2075 MeV {kinematics: 2040 MeV / (2047 ± 26) MeV}
  • -130 MeV < ΔE < -30 MeV {kinematics: -70.5 MeV / (-77 ± 27) MeV}
  • 1940 MeV < M_Ds < 1995 MeV {1968.5 MeV}
  • Daughter φ within 10 MeV
  • Daughter K* (neutral or charged) within 75 MeV
  • Daughter ρ within 150 MeV
  • Daughter π⁰ within 3 σ plus standard and splitoff
  • Daughter η within 3 σ plus standard and splitoff
  • Daughter K_s within 3 σ
  • Daughter η' within 10 MeV (π π η) or 30 MeV (ρ γ)
  • Underlying tracks:
    • |&sigma_dE/dx| < 3.5
    • |z₀| < 5 cm from beamspot
    • |r₀| < 5 mm from beamspot
    • Hit fraction < 0.5
    • Χ² < 1000
    • |cos θ| < 0.93
  * - Good showers are defined as:
  • Not hot
  • E₉/E₂₅ O.K.
  • Not matched track
  • Location O.K. (endcap or barrel)

  ---

  
  (8/23/06)

  D_s semileptonic

  post-FSR/Brem and alternate signal mode fixes (using φ e ν)

  Ds Tag Mode	Total Background	Possibly confusing background (within 50 MeV of U=0)	Signal events	εAll good	εEvent pass
  φ π	5	2	66	9.7%	15.5%
  φ ρ	1	0	17	2.4%	5.7%
  η π	3	0	26	10.0%	14.6%
  η ρ	0	0	54	5.1%	8.1%
  η' π	1	0	19	2.7%	3.2%
  η' ρ	0	0	11	1.2%	2.9%
  K* K	? (24)	1 (14)	91	15.3%1	22.5%
  Ks K	? (9)	0 (3)	72	18%	22.5%

  Current status shown in a U = E_miss - p_miss plot.

  ---

  
  (8/09/06)

  D_s semileptonic

  FSR/Brem. and "signal background" events

  Ds Tag Mode	FSR/Brem	signal background	true background	total background events
  φ π	16	3	3	22
  φ ρ	2	10	1	13
  η π	4	0	2	6
  η ρ	12	9	0	21
  η' π	2	0	1	3
  η' ρ	3	4	0	7
  K* K	21	16	2	39
  Ks K2	8	1	1	10
  Sum	68	43	10	121
  Percentage of total	56%	36%	8%	100%

  ---

  
  (7/26/06)

  D_s semileptonic

  Chul Su cut ε, isolation ε, sequential σK 97.0% 97.0% z0 95.7% 94.5% r0 93.8% 91.6% chi2 99.9% 91.6% hit fraction 87.5% 83.6% base cuts 99.7% 83.6% mass * 79.4%

  My cut ε, isolation ε, sequential σK 97.0% 97.0% z0 96.9% 95.5% r0 96.1% 94.1% chi2 97.0% 92.6% hit fraction 100% 92.6% base cuts 99.7% 92.6% mass * 92.6%

  ---

  
  (7/19/06)

  D_s semileptonic

  Current total U plots (adding all modes)
  Poor man's kinematic fit doesn't seem to improve much: No kin. fit followed by kin. fit on D_s using pdg mass constraint
  

  Current issues:

  • Some peaking background accounted for, but not all (~1/2 ?)
  • Slight difference in ε with Chul Su: 21.0% (φ e ν) and 46% (η e ν) vs. 30% (φ e ν) and 32% (η e ν) for semileptonic effici