Message: Re: Count rate of 1Mhz or more Not Logged In (login)
 Next-in-Thread Next-in-Thread
 Next-in-Forum Next-in-Forum

Idea Re: Count rate of 1Mhz or more 

Forum: Run Management
Re: Question Count rate of 1Mhz or more (Cejo George)
Date: 01 Aug, 2013
From: Zach Hartwig <Zach Hartwig>

Hi Cejo,

> My program executes 1000 events in 1 second but my requirement is much
> higher, 1 million events per second as in real detectors.

While you may have your reasons, I'm surprised that you require simulation to have the same time characteristics as "real" life. May I ask why?

A few thoughts that might help although I can't be sure. Geant4 does not have any sense of continuous time: although each event keeps track of a global time (starting at 0 when the event begins by default), each event is completely independent including time. Simulating time-dependent detector effects (such as pulse pile up, for example) requires other methods rather than simply increasing the event rate. Also, if simulating the response function of your NaI detector is your goal then one can simply run events until the statistical error in each bin of your histogrammed response function is acceptable for your use case. Then the histogram can be scaled if you desire an absolute match to an experimental response function with a fixed number of total counts.

> Is there any method to increase the speed?

1 million events / second will be impossible for a single processor. That being said, here are three ideas:

0. You can parallelize your simulation, either with the new multithreaded methods released as Geant4.10.0 beta or by parallelizing the simulation yourself with MPI or TOP-C (See examples in $G4INSTALL/examples/extended/parallel/)

1. NaI has a scintillation response of 38000 photons/MeV. You can reduce this by a large factor provided your photon statistics are acceptable and then scale the response back at the end (if necessary).

2. Depending on your geometry, you can trying implementing variance reduction methods such as source biasing (artificially limiting the particle you shoot and correcting for it in your final results) or geometry biasing (assigning importance regions to volumes in your geometry to focus on the most important regions). See $G4INSTALL/examples/extended/biasing. Note that one must be extremely careful when using any variance reduction to ensure that the computed biased results are consistent with the analog (non-biased) results.

Hope this helps.

~Zach

 Add Message Add Message
to: "Re: Count rate of 1Mhz or more"

 Subscribe Subscribe

This site runs SLAC HyperNews version 1.11-slac-98, derived from the original HyperNews


[ Geant 4 Home | Geant 4 HyperNews | Search | Request New Forum | Feedback ]