ATLAS inclusive and sliced samples (eg W+jets)

[ChrisVer]

I have a small technical question, if anyone has ever worked with it...
What does inclusive and sliced samples mean (at a MC generator)? I have seen that the sliced samples are divided according to mass slices for the different processes, but the inclusive are not. Why would they give such samples seperated? I mean as I see it, wouldn't the sum of the sliced samples be the "inclusive" one?
Thanks.

 

[mfb]

I mean as I see it, wouldn't the sum of the sliced samples be the "inclusive" one?

Not necessarily. Imagine you want to describe some process in the range of 100 GeV to 4 TeV (pT, pT sum, invariant mass - doesn't matter here). To get a reasonable statistics at 4 TeV with an inclusive sample, you might need billions of events. Most of them are at low energy, where you have more events than you can ever hope to process: simulations of detector response are a significant fraction of the overall CPU hours used for high-energy physics.
Making multiple separate samples, each with appropriate size, reduces the number of events significantly. To get a single spectrum, you have to multiply them with some weight before you can add them.

 

[ChrisVer]

Not necessarily. Imagine you want to describe some process in the range of 100 GeV to 4 TeV (pT, pT sum, invariant mass - doesn't matter here). To get a reasonable statistics at 4 TeV with an inclusive sample, you might need billions of events

Agree on that, and somehow it helps me understand the inclusive samples. And I agree that at 4TeV you will need some billions of events too.

Most of them are at low energy, where you have more events than you can ever hope to process: simulations of detector response are a significant fraction of the overall CPU hours used for high-energy physics.

aha, nice... so getting the necessary events at high energies you will get several orders of magnitude more events at low ones too, making the generation of events extremely slow.

Making multiple separate samples, each with appropriate size, reduces the number of events significantly. To get a single spectrum, you have to multiply them with some weight before you can add them.

Yes I agree, the weight should take into account the different cross sections...
However if you add them in order to get the overall background you expect at let's say 0.1-4TeV , would you add appart from the sliced samples the inclusive one as well?

 

[mfb]

Agree on that, and somehow it helps me understand the inclusive samples. And I agree that at 4TeV you will need some billions of events too.

You don't need billions of events at high energy, but you need events everywhere.

To get 1000 around 4 TeV, you might need billions events in total. It is more effective to produce 10000 in the range of 4 TeV, 10000 in the range of 3 TeV and so on.
In terms of luminosity, the highest sample might correspond to something like 20+ times the data luminosity, the other samples can be lower as statistics is not so crucial there any more.

If the inclusive sample is independent, it can get added. It can improve the statistics in the low-energetic range.

 

[ChrisVer]

If the inclusive sample is independent, it can get added. It can improve the statistics in the low-energetic range.

Wouldn't that however count as a double-counting?

 

[mfb]

If the inclusive sample is independent

!

 

[RGevo]

A sliced sample is a sub sample of the inclusive, so they are not independent.

Combining samples requires a correct merging, where certain events would be vetoed according to a double counting criterion.