Monte Carlo 09 presses on

2 November 2009

Accounting for dead material: the simulation team has added a railway wagon worth of metal to the Geant 4 simulated detector. Courtesy of Craig on Flickr.


Starting in June and ramping up through the summer, the Monte Carlo (MC) campaign for 2009 covers a lot of ground. “It’s a bit complicated because we expect three energies from the LHC,” says Borut Kersevan, Physics MC Production Coordinator. 

Rather than simply aiming for the 10 TeV run, as they did last year, the team has simulated around ten million events for proton energies as low as 900 GeV – that of the LHC’s very first collisions. They are currently working on 7 TeV collisions, to occur with the 3.5 TeV beams expected in early 2010. And, in keeping with the current outlook for summer 2010 and beyond, they intend to simulate more 10 TeV collisions. “For 10 TeV we will also recycle last year’s production,” says Borut.

Simulations start with event generation: colliding the protons. Making sure that ATLAS uses the best possible simulated collisions is the task of ATLAS Monte Carlo Conveners Judith Katzy and Claire Gwenlan, with aid from around thirty other specialists. “We have to make sure that whatever the generator outputs is okay,” says Judith. This includes checking the settings for the physics models and ensuring the correct functioning of the software.

“Of course, the reliable simulation of any LHC process also depends on how well we understand the structure of the proton,” says Claire. Since MC08, they have upgraded the parton distribution functions (PDFs), which describe the inside of a proton – the quarks and gluons that will be doing the actual colliding.

The MSTW (Martin-Stirling-Thorne-Watt) Collaboration has developed special modified Leading Order (mLO) PDFs for leading order event generators, such as Herwig and Pythia. The use of mLO PDFs should help produce outputs that resemble more accurate next-to-leading order results.

The new PDFs also affect the predictions of the generators, such as the number of particles produced in a collision. But with measured hadron-hadron collisions at the Tevatron and ISR, the team has some idea of what to expect from lower-energy collisions. The MC09 tune altered the parameter settings of the phenomenological models to describe these lower-energy events.

These events are then “measured” by a simulated ATLAS detector, using Geant4 to describe the material response of the detector. Subsequent ‘digitization’ simulates the electronic response of the ATLAS sub-detectors. Finally, that simulation is fine-tuned to account for detector mis-alignments and dead channels. “At this point you get a response that’s reasonably close to the detector itself,” Borut says.

These reconstructions are run at Tier-1 and Tier-2 sites around the world, following the course that real data will take through the Grid. And also like real data, the detector simulation files (known as HITS, as they are like detector “hits”) are written to tapes. “There’s about 400 million events, say half a petabyte, of these HITS sitting on tapes all around the world,” says Borut.

The detector simulation becomes more and more accurate as the simulation team continues to add new levels of detail, down to wires and screws. Borut reckons that they’ve added a “railway wagon of metal” to the simulated ATLAS since MC08, according to a Geant4 expert. In addition, teams are developing more and more advanced reconstruction algorithms.

To take advantage of such improvements, the Monte Carlo “data” are reprocessed a couple of times a year, and the most recent of these campaigns started in early October with release 15.3.1.6. “Almost surprisingly, we’re almost through with it, meaning ATLAS distributed computing is really getting better and better,” says Borut. The last reprocessing effort took three months.

In the more distant future, Borut is concerned about running high-luminosity simulations: “The LHC environment is going to be so complex that the simulation barely manages to run on the resources we have.” An upgrade plan that will bolster Grid hardware and streamline simulation software is currently underway.

But he is optimistic about the upcoming beam and first collisions. “We have a good ‘guestimate’ of how the detector will work,” he says, “but we won’t, of course, know until we have actual data.” 

And, in fact, many have taken advantage of the data available. “We’ve actually managed to already compare data to Monte Carlo with cosmics,” he says. “Okay, it is my personal perspective, but the predictions agree surprisingly well with the actual data, meaning we are on a good track.”

 

Katie McAlpine

ATLAS e-News