Comparison between data and MC (IC79)

Here, I have shown several plots for the data and MC comparison for IC79 data. This comparison will be essential for the separation of EHE neutrino signals from backgrounds. Parameters are looked for that have reasonable agreement between data and background MC, but does different behavior between data and signal MC.

Used data

I have used EHE filter data whose condition is (portia) NPE is greater than one thousand.

The run number of 116180, 116200, 116210, 116220 were used for the observation data. All of the data is the part of the burn sample.

The JULIeT data (100k each) are used for signals and HE CORSIKA data with iron primary (~20k) are used for the backgrounds. The SPICE 1 tables are used for the generation.

Observational data check

The effective livetime is 1.078 days.

Time difference between events. Looks normal. The rate is 1.46 Hz.

Data and MC comparison

In following plots, the colors mean

• black: observation data
• blue: HE CORSIKA (iron, SIBYLL)
• red: EHE signal sum (GZK (Shigeru et al))
• red: EHE signal (GZK) from mu
• red: EHE signal (GZK) from mu
• magenta: EHE signal (GZK) from tau
• green: EHE signal (GZK) from nue
• dashed red: EHE signal (GZK) from numu
• dashed magenta: EHE signal (GZK) from nutau

Fundamental plots

NPE agreements between data and background MC is nice.	We see a difference in data and background MC probably due to conincidents events.	Distribution becomes clean when a condtion is put. NDOM > 100 && CoGZ > -200m && NPE > 10^3.5	NDOM.
CoGX	CoGY	CoGZ. A difference seen at deeper place.
Closest distance. Agreement is nice.	Closest distance for each contributions.

mpe, photorec, mue, etc

Energy (photorec). There is a tail for data.	Energy (mue)	dE/dX
Energy (photorec). The tail reduced a bit. NDOM > 100 && CoGZ > -200m && NPE > 10^3.5	Energy (mue) NDOM > 100 && CoGZ > -200m && NPE > 10^3.5	dE/dX NDOM > 100 && CoGZ > -200m && NPE > 10^3.5
ZA (Linefit)	ZA (spe single)	ZA (spe 4 iteration)	ZA (mpe)
Linefit speed Z

Cut values (mpe)

Ldir. This parameter also may be useful.	Ldir	Ndir.	Ndir
Log likelihood	Log likelihood	NDOF	NDOF

Ehe-star related parameter

Look at this page for ehe-star.

Area ratio	Area ratio from each components. nue give high area ratio.
Gradient. Althouhg mismatch in bg, bg and signal have different distributions.	Gradient from each components. nue give high area ratio.
Gradient. Althouhg mismatch in bg, bg and signal have different distributions.	Gradient from each components. nue give high area ratio.

cascade related parameter

Energy

Energy

Nch

Zenith angle Vs NPE

Mis-reconstruction may be problematic...

Zenith angle Vs NPE (data)	Zenith angle Vs NPE (CORSIKA)	Zenith angle Vs NPE (GZK)
Zenith angle (mpe) Vs Energy (data)	Zenith angle (mpe) Vs Energy (CORSIKA)	Zenith angle (mpe) Vs Energy (GZK)

Summary and future plan

• Agreement between data and MC is relatively reasonable at this point without cuts.
• Some cleaning needed. (need for removing more coincident events)
• Mis-reconstruction of high level reconstruction may be a problematic. (need check after some cleaning.)
• Some parameters seem to be useful for sig/bg separation. (closest distance, Ldir, area ratio, gradient and so on) The differences are based on that fact that neutrino can have a starting track (or/and cascades).
• Other intrinsic difference between muon bundles and single track should also be investigated such as difference of dE/dX at start point and at the end point. (Budles lose its energy faster.)
Sophisticated reconstruction leads to mis-reconstructions that should be avoided. So, we may not be able to use the results. Although, I like to improve energy informaiton with probably timming information (spread of hit time). With that information, the distance to a track can be estimated, so that it will lead better energy estimation.
• I will use Random Forest method (decision tree). The method tells us which parameter is useful for the separation.
• I can make a "supercuts" with the parameters by taking into account some varable dependence such as energy information, but the random forest method automatically takes into accounts such dependences, so it may be convenient and more powerful.