490
J. Apostolakis
and reactors and the all-particle offshoot MCNPX. Its models will be contrasted with
the capabilities of FLUKA and GEANT4, but it has not seen use in HEP detectors,
due to lack of electron/gamma models above 1 GeV, restrictive licensing and export
control.
In order to compare with measurements, the response of detectors to the energy
deposited by an event’s tracks must be estimated. One tool used for the detailed
estimation of the energy deposition in a gas detector is Garfield [26]. It generates
low-energy gammas and electrons, down to eV energy using HEED [27], uses
the electric field calculated externally, and transports electrons and ions under
the combined influence of its electromagnetic field and diffusion. The recent
Garfield++ rewrite [28] and extension extended its capabilities, added refined
electron transport and physics models for semiconductors, and enabled interfacing
with GEANT4 [29]. Its computational cost is 2–3 orders of magnitude larger, so it
is utilized sparingly: for studies and to generate an accurate parameterization of a
detector’s response [30] for use in large scale simulations.
11.2.2 Level of Simulation and Computation Time
The modeling of every physical interaction, from the initial particle’s energy down
to the interaction of eV scale photons and electrons—or even the interactions
of neutrons down to thermal energies—is possible. The computing cost of such
simulation is prohibitive for most practical applications, and simplifications are
required. Yet in some cases it is necessary to simulate down to very low energies,
for example in order to estimate the activation of materials by neutrons.
In complex detectors, such as in an LHC detector, the full simulation of each
event takes between 0.1 and a few minutes on modern computers, depending on the
type of event (minimum bias or t/t-bar) and the region simulated (rapidity coverage).
This limits the number of events that can be simulated.
In some applications the simulation effort can be reduced for many events: by
simulating first the particles that are involved in the trigger. Otherwise, one may
seek to limit the number of secondary particles generated or the total track length
simulated, or to simplify the treatment of the most frequent interactions.
Another alternative is fast simulation. This involves selecting only a fraction of
tracks for simulation, and approximating the detector and key physics interactions
in order to reduce the computation time per event by one, two or more orders of
magnitude. Fast simulation is a powerful tool for modern experiments, as it allows
speedy turnaround to address changing conditions or assumptions, and to explore
different model parameters at an affordable computing cost. It can be calibrated
using full simulation, data or both. However it is not capable to estimate resolutions
and correlations, and it can be harder to obtain systematic errors.
Recently ATLAS has created a hybrid simulation mode by selecting for detailed
simulation the conical regions of the detector around the most energetic primary
particles, and using fast simulation models for the remainder [31].
J. Apostolakis
and reactors and the all-particle offshoot MCNPX. Its models will be contrasted with
the capabilities of FLUKA and GEANT4, but it has not seen use in HEP detectors,
due to lack of electron/gamma models above 1 GeV, restrictive licensing and export
control.
In order to compare with measurements, the response of detectors to the energy
deposited by an event’s tracks must be estimated. One tool used for the detailed
estimation of the energy deposition in a gas detector is Garfield [26]. It generates
low-energy gammas and electrons, down to eV energy using HEED [27], uses
the electric field calculated externally, and transports electrons and ions under
the combined influence of its electromagnetic field and diffusion. The recent
Garfield++ rewrite [28] and extension extended its capabilities, added refined
electron transport and physics models for semiconductors, and enabled interfacing
with GEANT4 [29]. Its computational cost is 2–3 orders of magnitude larger, so it
is utilized sparingly: for studies and to generate an accurate parameterization of a
detector’s response [30] for use in large scale simulations.
11.2.2 Level of Simulation and Computation Time
The modeling of every physical interaction, from the initial particle’s energy down
to the interaction of eV scale photons and electrons—or even the interactions
of neutrons down to thermal energies—is possible. The computing cost of such
simulation is prohibitive for most practical applications, and simplifications are
required. Yet in some cases it is necessary to simulate down to very low energies,
for example in order to estimate the activation of materials by neutrons.
In complex detectors, such as in an LHC detector, the full simulation of each
event takes between 0.1 and a few minutes on modern computers, depending on the
type of event (minimum bias or t/t-bar) and the region simulated (rapidity coverage).
This limits the number of events that can be simulated.
In some applications the simulation effort can be reduced for many events: by
simulating first the particles that are involved in the trigger. Otherwise, one may
seek to limit the number of secondary particles generated or the total track length
simulated, or to simplify the treatment of the most frequent interactions.
Another alternative is fast simulation. This involves selecting only a fraction of
tracks for simulation, and approximating the detector and key physics interactions
in order to reduce the computation time per event by one, two or more orders of
magnitude. Fast simulation is a powerful tool for modern experiments, as it allows
speedy turnaround to address changing conditions or assumptions, and to explore
different model parameters at an affordable computing cost. It can be calibrated
using full simulation, data or both. However it is not capable to estimate resolutions
and correlations, and it can be harder to obtain systematic errors.
Recently ATLAS has created a hybrid simulation mode by selecting for detailed
simulation the conical regions of the detector around the most energetic primary
particles, and using fast simulation models for the remainder [31].