Manual:Random numbers

Offline Documentation: [Offline Category] [FAQ] [Howto] [Reference] [Manual]

Offline Documentation

This article is part of the Offline Documentation

Other pages...

Offline Category FAQ How Tos Reference Manual Getting Started Software Installation

See also...

General Help on this Wiki

Pseudo-random number streams are managed implicitly by nuwa.py.

Random Number Systems

We use CLHEP's random number system though the layer that Gaudi provides. Geant4 uses the same system. It is allowed to use ROOT's TRandom system but never set its seed from user code. The ROOT seed is set at the same time as the CLHEP one, and to the same value. All other random number system, including Pythons should not be used as doing so may lead to irreproducible results.

Managing the Random Seeds

The random stream is initialized once per execution cycle. The seed is built from the run number, execution cycle number, and hostid. Just after initialization some number of randoms are drawn ("burnt") as it was found there were otherwise correlations.

The user can achieve reproducibility by specifying the run number, execution cycle number and host id in

nuwa.py -R runnumber -N execcycle -H hostid 

The algorithm that manages this is DybAlg/DybShuffle. This algorithm saves a record of the seeds for each execution cycle in:

/Event/Random/RandomHeader

The object is at this location is a RandomHeader

Consuming Random Numbers

If your code needs to consume/use random numbers it should use Gaudi's random facilities. These are best accessed through Rndm::Numbers objects. When writing an algorithm make a Rndm::Numbers object for each distribution you need. Here is an example from GenTools/GtPositionerTool:

// declare inside class definition in the .h file:
private:
  Rndm::Numbers m_uni, m_gauss;

// initialize inside initialize()
IRndmGenSvc *rgs = 0;
if (service("RndmGenSvc",rgs,true).isFailure()) {
    fatal() << "Failed to get random service" << endreq;
    return StatusCode::FAILURE;        
}
if (m_uni.initialize(rgs, Rndm::Flat(0,1)).isFailure()) {
    fatal() << "Failed to initialize uniform random numbers" << endreq;
    return StatusCode::FAILURE;
}
if (m_gauss.initialize(rgs, Rndm::Gauss(0,1)).isFailure()) {
    fatal() << "Failed to initialize Gaussian random numbers" << endreq;
    return StatusCode::FAILURE;
}

// Consume:
x = m_gauss() * rms + mean;
t = m_uni() * (max-min) + min;

Offline Software Documentation: [Offline Categories] [FAQ] [Offline Manual Category]