Antares/KM3NeT group meeting - reconstruction

Europe/Amsterdam
Zoom

Zoom

Description

Zoom session link:
https://nikhef.zoom.us/j/96916253682?pwd=eGYxUGhTTDlDV2dwd2VrMTBqRnZDZz09

 
Meeting ID: 969 1625 3682
Passcode: 984867

 

The reconstruction meeting of March the 19th was dedicated in great part to an overview of the toy-MC methods present in aanet, presented by Jordan.

Jordan - Toy-MC overview presentation:

Motivation for toy MC is three-fold:

  1. Try to achieve complete understanding of the event topology from first principles,
  2. Isolate individual parts of the reconstruction problem,
  3. Allow for experimentation and fast evaluation.

So how do we do toy MC for a high-E neutrino-DIS event?
In principle, you need to:

  1. Know the neutrino interaction type,
  2. Understand and simulate prompt particles (types, energies, directions),
  3. Simulate the showering,
  4. Simulate the photon production and propagation,
  5. PMT simulation.

Instead of doing the above in full detail, four simplifications are made:

  1. Assume only EM showers and put all energy in one leading lepton within the nu-direction.
  2. Don't simulate the showering in detail; sample average, elongated EM profile.
  3. Assume that all produced photons are visible.
  4. Assume PMT is perfect; i.e. all impinging photons create an observable signal.

The Jpp arrival-time PDFs are defined in terms of PMT-dependent parameters.
To include the effects of elongation, we use the parameterization from Claudio Kopper's thesis.
This shifts the position of the peak in the emission profile and broadens the arrival-time PDFs for all PMTs.
By integrating the PDF, we obtain the expected number of photo-electrons observed by each of the PMTs.
These can be used to generate a random number of observed photo-electrons (by drawing from a Poisson-distribution) for each PMT.

The arrival times of the corresponding hits need to be drawn from the elongated PDFs.
There are two ways to do this.
One method is to draw directly from the N-step energy-elongated PDF.
This method is true to the underlying distribution, but is slow and less accurate (due to the need to discretize the distribution into N steps in energy).
The other way is to pick a random point along the shower's longitudinal profile and draw from the non-elongated arrival-time PDF.
This method is faster and more accurate.
In the limit of large N the two methods will yield the same result.

The drawing of the random arrival time itself is done based on an inversion of the arrival-time CDF, i.e. the arrival-time quantile function.
There is no easy analytical expression for this quantile function.
Therefore, the function is approximated numerically, using a binary search method.

On slide 11 we see that the MC nicely follows the analytical arrival-time PDF for a single PMT.
On an event-by-event basis the toy-MC also reproduces the general shape of the arrival time-distribution derived by more rigorous simulation software, such as sirene.

To make the toy MC more accurate, we can add multiple tracks, backgrounds and blurring at the expense of greater CPU needs.

----------------------- End of toy-MC overview presentation ---------------------------

Bouke:

Working on comparison between JShowerfit and Aashowerfit. Running jobs after encountering trouble with the batch processing which are now resolved.
Also finishing track reconstruction documentation action points. 

Aart: Things have been messy with the new releases of software recently. Notify me if there are any issues with the new aanet version.
 

Thijs:

Working on double bang reconstruction. Ran into some problems when looking at hits instead of mc hits - including background & pmt simulation. Looking into events which act strangely - giving a weird likelihood. Found a potential solution to the problem and taking time to understand it.

Brían: What about the Jacobians document?

Thijs: At the point now where values can be implemented and plots of the derivatives can be made. Can look at events from JSirene with a single electron, with/without shower elongation and look at the likelihood and its derivatives. 


Brían:

Similarly to Thijs, looking at real hits with background etc included, and investigating strange likelihoods. Printing out event and hit information and looking at the likelihood to understand what’s going on. Evaluating the PDF separately for the muon and show and checking the coordinate frame. Mostly evaluating background and trying to understand why. 

Jordan (besides toy-MC):

Likelihood plots are nice. In the reconstruction the minimiser seems to have trouble finding the minimum. Looking into the minimiser results vs. the true minimum values. 

Timon:

Looking at PDF values and how to get to first hits probabilities. 

Jordan: You can take a look at the likelihood document Aart wrote. 

 

There are minutes attached to this event. Show them.