Experimenters Bet on New Software
by Mike Perricon
"God does not play dice!" --Albert Einstein
"All the evidence shows that God was actually quite a gambler, and the universe is a great casino." --Stephen HawkingWhether or not God is a gambler, mortal physicists eagerly declare their devotion to Monte Carlo.
They're referring to "Monte Carlo data," spun from a mathematical casino of the mind, with a metaphorical roulette wheel generating random numbers to create a critical physics experimental tool: the simulation.
"In the end, claiming a discovery comes down to having something that looks more like a simulation of new physics than it looks like the Standard Model that we know and understand," said John Womersley, one of the leaders of the software effort for the DZero experiment. "That simulation must be something we know and trust."
Monte Carlo data creates simulations for a range of possible outcomes that the detectors will see in particle collisions, using random numbers at each turning point along the way to nudge the process in different possible directions. While no single result is definitive on its own, the array of simulated outcomes can show patterns that point to new discoveries-- or to more of the same old stuff. That distinction, based on what's termed Monte Carlo simulated data, can represent the difference between success and failure in the search for new physics.
To develop the reliability that must be implicit in their simulations for Run II, Womersley and the DZero software developers are issuing the Monte Carlo Challenge: a huge number of simulated "events," or particle collisions, consisting mostly of conventional background but also sprinkled with what Womersley calls "interesting stuff."
The Challenge simulations will be distributed to physicists throughout the 400-member collaboration beginning in October. A Monte Carlo Physics Workshop is planned for 2000, when collaboration members can present the results of their gleanings and compete for (non-simulated) prizes.
"We hope to use the Challenge as a way of restarting the physics effort, which has been a little dormant over the last couple of years because we haven't been taking new data," Womersley said. "Getting people interested in these simulated events is a way to demonstrate whether this whole vast amount of software works together, and it's a way to prepare to get the best physics out of it right from the beginning of the next run."
Womersley's description of a "vast amount of software" isn't an overstatement. About a quarter of the DZero collaboration, some 100 members, have been working on developing new software for what will essentially be a new detector for Run II. In addition to relying on energy measurements in its huge calorimeter, DZero will also make tracking measurements (as CDF historically has done) with the addition of a magnetic field and silicon vertex detector close to the point of the collisions between protons and antiprotons.
New capabilities mean new software to record and analyze new data, and DZero's old Fortran-based software wasn't up to the challenge. Womersley estimated that of a million lines of Fortran code, only one-fourth to one-third would be useful for Run II. That opened the door to what he termed a "radical" solution: starting all over.
A particle collision experiment needs at least three huge software efforts to understand the physics it produces: triggering, or sorting interesting events from uninteresting ones as they happen; reconstruction, or envisioning the collision by examining what's left; and simulation, or predicting how an interesting event will look. There's also a great need for on-line software, furnishing the control-room displays for monitoring the detector, providing controls for high voltage and detector parameters, and insuring data is of good quality by insuring the detector is operating as well as possible.
"It's an unseen vital detector component that's part of the upgrade efforts for both CDF and DZero," Womersley said. "If we mess it up, we mess up the experiment just as surely as if a component of the detector didn't perform as it should."
The software renovation came just as a new style of programming, called "object-oriented" software, was maturing. The products of a particle collision can be seen as objects, with properties to be derived and worked with; the object-oriented C++ language allows the linking of information about that object with the operations that can be performed on it. Object-oriented programming also fits naturally with graphics systems that can display objects in three dimensions.
Younger physicists are heavily represented through the DZero software effort, partly because younger people generally have more time to devote to the project than do senior physicists with additional responsibilities. But DZero was conscious from the beginning that using C++ would tilt the age balance. Younger people haven't had to shift their way of thinking: they learned C++ in college, where Fortran is now regarded as out of date.
But working with software and data requires infrastructure. The Computing Division's Joint Projects Group developed the basic software toolkits needed at both CDF and DZero. The Computing Division is also responsible for the infrastructure needed for networking, storing and processing data. Each detector has required about $9 million in computing hardware, with the Computing Division purchasing, installing and maintaining the equipment.
Instead of recording data on tape at the experiment, the event data will be sent by fiber optic cable to the Feynman computing center. There it will be processed by a "farm," a large number of relatively inexpensive PC's with a Linux operating system, and filed and retrieved by a tape robot system. Experimenters then will have access to an array of classifications: for example, they can request all the events recorded, or only those taken on a certain day or with a certain trigger.
All the software must be up and running--and viewed as reliable--for the opening moment of Collider Run II of the Tevatron in mid-2000. Now is the time to spin the wheel, establish the odds, and learn what to expect. The last thing physicists want to do is gamble with the results to come.
"If you reject 99.99 per cent of all collisions and keep the remaining interesting ones, you'd better make sure that new physics fits your definition of what's worth keeping," Womersley said. "Or else you'll look pretty stupid if something shows up later at LHC that we could have found here if we'd had a smarter trigger or a better reconstruction program. That's everybody's nightmare."
It's time to place your bets.
|last modified 9/3/1999 email Fermilab|