Why Monte Carlo Simulations? What do they tell us?

[Notorious QED]

I am an undergraduate working in my university's high energy physics lab. I'm still new to particle physics, so I have to learn/understand a bunch of things on the fly.

We are currently searching for a new state theorized to occur in a specific decay process. I know we have to work with data only on the initial and final states, since we know nothing about intermediate states (they are not physical observables). I'm not too sure about the reconstruction process, but I will likely understand that better once I understand the following questions...

My questions are regarding about the purpose of Monte Carlo simulations.

1. Why is it necessary to run these simulations before analyzing the real data?
2. What exactly does a Monte Carlo simulation tell us?
3. How might a simulation help in searching for a new state?

I know these questions are fairly broad, so apologies in advance. If I didn't provide enough context or just generally didn't provide enough information, let me know and I'll try my best to make things clear. I looked through other threads about similar topics but didn't find them to be too satisfying. I'd like to get a handle on this as soon as possible, so any insight would be beneficial!

 

[BvU]

Hi,

Not an expert, just a 'user'. I quote an introduction paragraph from a section in my PhD thesis, just for fun (30 years old!).

Basically MC is an integration method. You can generate all kinds of distributions more or less effectively.

And programming MCs is fun. Do at least a simple one yourself to gain understanding. In real experiments MC is a team effort building on preceding experiments' codes.

 

[mfb]

Why is it necessary to run these simulations before analyzing the real data?

For various reasons, and different analyses will use it for different aspects.

You have to define the selection criteria in your search. You want to maximize your sensitivity to potential new particles. How do you do that? You cannot use the data you want to analyze to optimize your selection - you could bias yourself in choosing parameters that randomly highlight some statistical fluctuation. You can take some fraction of your data to tune the selection and the remaining part for the main analysis, but then you reduce the amount of data available for the analysis, and it also leads to other issues. In some analyses, you can obfuscate your data in such a way that you can use it to tune your selection without biasing the result, but if you search for new particles I don't see how that would work.

Your analysis won't have 100% efficiency for the new particle. To set limits (or to measure the cross section), you have to know the efficiency. Typically you need MC in some way for that.

Your analysis will have various background contributions. Typically you need MC in some way to estimate their size and shape.

Various other applications are possible, but it depends on the analysis.

 

[ChrisVer]

Why is it necessary to run these simulations before analyzing the real data?

Also one more reason is time itself- MC simulations can take a long time to be produced and that's something you don't want; in fact you want to be able and plug in the data in your analysis once it's stable and see how well things work. So by the time you get the data, it's good to have the MCs, or you'll be delayed.

What exactly does a Monte Carlo simulation tell us?

A simulation gives you what you expect to see in data given that your theory (out of which you make the simulation) is true.
One example from the Higgs discovery is the following plot, which shows the distribution of the 4-lepton invariant mass in MC (background of what you expect) and Data (what you actually observe).

How might a simulation help in searching for a new state?

I'd say it's impossible to make any kind of search without simulation. I mean, how do you know what you expect if not by simulating it? (maybe there is an alternative way, but it's not very good by itself). Without knowing your expectation, you can't measure deviations from it to call it a new discovery.
There are ways to estimate background from data, but I am not so sure how well applicable they are without MC validations.

 

[mfb]

I'd say it's impossible to make any kind of search without simulation.

Things like the Z, J/Psi and so on are easy to find without simulations because they are huge peaks above a small background. The first hadrons were found before computers could be used.Monte Carlo is used to design the detectors as well. How do you determine how many layers your tracking detector needs? Well, you simulate it with 4, 5 and 6 layers, and see how the performance looks in each case.

What is the readout rate your detector components will need? Simulate it.

What is the expected fluence (for radiation damage) your detector parts will receive? Simulate it.

How do you identify your objects in the reconstruction? Simulate them to figure out what works best.

 

[Vanadium 50]

I'd say it's impossible to make any kind of search without simulation

See https://arxiv.org/abs/1112.5154 Phys. Rev. Lett. 108 (2012) 152001

 

[mfb]

See https://arxiv.org/abs/1112.5154 Phys. Rev. Lett. 108 (2012) 152001

The muon and photon selections were developed using MC, and the selection criteria for this analysis were optimized based on MC as well ("These thresholds are chosen in order to optimize signal significance in the χ b (1P,2P) peaks.").

It would be possible to do those things without MC, but it would make the analyses significantly worse.

 

[Notorious QED]

Thank you very much, all of you! This all really helped, and I genuinely appreciate the responses :)

 

[Vanadium 50]

The muon and photon selections were developed using MC

To some degree, although the muon in particular can easily be set from the J/psi peak. But how far back do you go? This paper doesn't cite a MC paper, but it cites a paper that cites an MC paper. How many generations back do you go?

the selection criteria for this analysis were optimized based on MC as well

That's not what the paper says. It says that they optimized the significance of the other peaks. In data. Then they looked for the third one.

 

[mfb]

That's not what the paper says. It says that they optimized the significance of the other peaks. In data. Then they looked for the third one.

Ah, got the particle names wrong.

To some degree, although the muon in particular can easily be set from the J/psi peak. But how far back do you go? This paper doesn't cite a MC paper, but it cites a paper that cites an MC paper. How many generations back do you go?

That was my point. While you can discover particles without any MC input, it is not done, as the analyses get much better with MC. Every analysis uses results based on MC in some way. Some more direct, some less direct.