What are the main algorithms the LHC particle detectors use to reconstruct decay pathways?

Question

I am just starting to look into how we understand the data from particle collisions.

My question is, what are the algorithms or ways that these detectors interpret the data? Are there standard approaches? Or if not, what are some good papers or places to look to get started in learning more about the implementation and/or details of how this works?

So far I haven't dug into any textbooks, but many articles on the web and this was somewhat helpful in pointing to where to look:

http://arstechnica.com/science/2010/03/all-about-particle-smashers-part-ii/

So from my understanding so far, there are a few different LCH "experiments", which are physical structures that are optimized to focus on specific aspects of data from a collision event. The detector measures all kinds of particle emissions and changes in electrical fields, and then seems to try to backtrack and figure out all the emission/decay events that might have taken place in that split second.

From my understanding so far, basically the computer programs used to compute these possible "decay pathways" must be using some standard algorithms or something, and must have built into them all possible particle emission pathways (like all possible Feynman diagrams if there is such a thing).

Are there any good resources or standard algorithms/approaches to understanding how particle detectors analyze their data?

This post imported from StackExchange Physics at 2014-08-12 09:35 (UCT), posted by SE-user Lance Pollard

Comments

i think cern has a page with references about algorithms and software used in various experiments, apart from that papers detailing the results of LHC experiments usually mention what kind of algorithms were used and what software (if any) (off the top of my head, papers about the higgs-like boson experiments had references about algorithms and software used)

This post imported from StackExchange Physics at 2014-08-12 09:35 (UCT), posted by SE-user Nikos M.

---

This is a huge topic. I literally took a semester course in grad school to get enough foundation to be ready to start learning when I took up research. and subsequently attended two different summer schools to learn a bit more. Sub-subjects include tracking, particle ID, jet identification, calorimetry and a huge body of work on beating the combinatorial explosion in the several places it rears its ugly head.

This post imported from StackExchange Physics at 2014-08-12 09:35 (UCT), posted by SE-user dmckee

---

@dmckee hardcore, sounds like a lot but still interesting. What was the name of the course so I can check out textbooks maybe then related to the topic?

This post imported from StackExchange Physics at 2014-08-12 09:35 (UCT), posted by SE-user Lance Pollard

---

It was a special class without a regular number and we mostly didn't use a text, but rather selected papers. We all had a copies of Perkins and of Leo from previous course work.

This post imported from StackExchange Physics at 2014-08-12 09:35 (UCT), posted by SE-user dmckee

---

You might be interested in this page. Basically the vast majority of the collision data is discarded immediately by custom hardware; the vast majority of what's left is discarded immediately by slower but more sophisticated software; and what passes those stages (a mere GB/s or so of data) is stored forever and analyzed later (probably multiple times, by different groups, for different purposes, using different techniques).

This post imported from StackExchange Physics at 2014-08-12 09:35 (UCT), posted by SE-user benrg

---

@benrg The selection of trigger conditions and the design of the electronics to allow you to set those conditions are both arts in and over themselves, but they are only the crudest part of the analysis. In order to make them fast they are limited to being minimally smart.

This post imported from StackExchange Physics at 2014-08-12 09:35 (UCT), posted by SE-user dmckee

---

I think an overview answer which gives a sense of the vastness of the field could be useful. I don't know nearly enough about detectors to write it, though...

This post imported from StackExchange Physics at 2014-08-12 09:35 (UCT), posted by SE-user David Z

---

Yeah that would be perfect! This is the closest I've found so far to some description of algorithms that might be being used (like the Monte-Carlo algorithm): quantumdiaries.org/2010/12/11/when-feynman-diagrams-fail.

This post imported from StackExchange Physics at 2014-08-12 09:35 (UCT), posted by SE-user Lance Pollard

---

In a sense the Monte Carlo (simulation) is not a reconstruction algorithm at all. It is a critical part of the analysis chain, but comes into play after you have figured out what kinds of things were going on in each event. I've written a about about how MCs are used elsewhere on the site. Here and here.

This post imported from StackExchange Physics at 2014-08-12 09:35 (UCT), posted by SE-user dmckee

Answer 1

The algorithms used are as many as the experimental setups times the detectors used in the setups. They are built to fit the detectors and not the other way around.

The common aspects are a few

1)charged particles interact with matter ionizing it and one builds detectors where the passage of an ionizing particle can be recorded. It can be a bubble chamber, a Time Projection Chamber, a vertex detector ( of which there exist various types).These are used in conjunction with strong magnetic fields and the bending of the tracks gives the momentum of the charged particle.

2)Neutral particles are either

a)photons, and the electromagnetic calorimeters measure them.

b) hadronic, i.e. interact with matter, and hadronic calorimeters are designed so as to measure the energy of these neutrals

c) weakly interacting, as neutrinos, which can only be detected by measuring all the energy and momenta in the event finding the missing energy and momentum.

In addition there are the muon detectors, charged tracks that go through meters of matter without interacting except electromagnetically and the outside detectors are designed to catch them.

The complexity of the LHC detectors requires these enormous collaborations of 3000 people working on one goal : getting physics data out of the system. Algorithms are a necessary part of this chain and are made to order using the basic physics concepts that drive the detectors.

As Curiousone says in order to understand the algorithms entering in the data reduction from these detectors a lot of elbow grease is needed. Certainly they are custom made.

This post imported from StackExchange Physics at 2014-08-12 09:35 (UCT), posted by SE-user anna v

Answer 2

Well, if you have the time... CERN has all the technical design reports for its detectors online at http://cds.cern.ch/. They are excellent reading material.

Start with a search for "ATLAS technical design report" and "CMS technical design report" and work your way trough the references in those documents. Once you understand the geometry of the detectors (not a small feat), you can start reading about "trigger algorithms" and "reconstruction algorithms". You may have to pick up a thing or two about particle matter interactions and the GEANT simulation software.

Little warning... it took me almost two years to read trough just the parts that were important to my work...

This post imported from StackExchange Physics at 2014-08-12 09:35 (UCT), posted by SE-user CuriousOne

Comments

Do you have a particularly good link from one of the searches? This looks pretty awesome, just the kind of thing I was looking for, some real data :)

This post imported from StackExchange Physics at 2014-08-12 09:35 (UCT), posted by SE-user Lance Pollard

---

If you google "cms algorithm" you are offered a list of algorithms . I picked "jet algorithms" and this is the link  https://twiki.cern.ch/twiki/bin/view/CMSPublic/WorkBookJetAnalysis .

You will see the complexity and continuous changes that we are talking about, just for this one algorithm.