# Differences between reconstruction- and generation-level variables in HEP data

I am working on a CMS - related project where the ROOT trees contain both reconstruction-level and generation-level particle variables (like mass). However, I don't know the basic difference between the two or when we prefer using one type of variable over the other?

Real data goes $$\text{Collisions} \to \text{Detector} \to \text{Trigger} \to \text{Reconstruction}$$

Simulated data goes $$\text{Event generator} \to \text{Simulated detector} \to \text{Trigger} \to \text{Reconstruction}$$

Generator-level is the information from the first step, before the events are passed through the simulated detector. This is also sometimes called "MC Truth" or "Truth level".

Reconstruction-level is the information from the last step, after the reconstruction algorithms have been run over the data from the simulated detector.

You can compare the two in order to quantify things like efficiencies and resolutions.

What follows is basically a guess based I my experience with several experiments (not one of which was at a collider; all fixed target and neutrino stuff).

I don't know the CMS data, but presumably if you have "generation level" quantities you are looking at Monte Carlo, yes?

If so this will be a kind of record keeping that is useful for both debugging the simulation and for providing functions to deduce the underlying physical distributions from the measured distributions (which include FSIs, multiple scattering, errors in clustering and so on). Most simulations picks some "true at vertex level" quantities, and then manipulate (and possibly weight) them to get the "as measured" quantities which appear as if they were data.

"Generation level" would be the former.

I will continue from dmckee's answer, again guessing you are working on a Monte Carlo generated set.

The monte carlo event generator , generates the event from basic quantum mechanical calculations, it is a method of getting at the integrals of the plethora of Feynman diagrams for proton proton scattering . It is the ideal event. The detector simulations are necessary to get at specific numbers, the energies and momenta of the particles generated at the generation vertex, as the detector degrades them, so as to compare with the real measurements. The final numbers are the reconstruction level variables, to be compared with the measurements from the detector.

This could be a nested definition: a gamma hitting the electromagnetic calorimeter treated as generated and the summed energy in the calorimeter as a reconstruction, because at each subdetector the simulation assumes primary interaction from theoretical formulas and generates what the reconstruction should look like.