How can i simulate 10^12 particles simultaneously?

Particles, Track, Event, Run and Biasing
1

Good morning!
I want to simulate 10^12 particles at a run.However ,every times I execute
/run/beamOn 1000000000000, it will failed and print out below.

2021-01-18 16-47-26屏幕截图
2021-01-18 16-47-26屏幕截图912×420 41.7 KB

Could anyone give me a solution to realize my requirements?Thank you very much.

2

You can’t, not in a single run. The number of events is stored in a signed integer (G4int), with a maximum value of 2,147,483,647. To generate that many events, you should do five runs of maximum size in your job:

/run/beamOn 2000000000
/run/beamOn 2000000000
/run/beamOn 2000000000
/run/beamOn 2000000000
/run/beamOn 2000000000

Set up your analysis code so that it either accumulates across all the runs of a job, or includes the run number (which increments with each beamOn) in the output filename.

If you want to run your jobs in parallel, you will need to write your own code to initialize the random number seed uniquely for each job (in my experiment, we wrote a utility class to generate seeds from UUIDs; you could use the timestamp, job ID, or something).

1 Like
3

wouldn’t 1e12 particles be 500 runs with 2e9 particles each?
I would be interested in your solution to automate the merging of primitive scorers, if you use them :slight_smile:

4

@weller Yup, it would be 500 runs. But if the user is already prepared to wait for 10^12 events to finish, waiting for 500 runs to take the same amount of time isn’t a problem. Alternatively, they can launch the jobs in parallel on a compute farm and do the output merging afterward.

We don’t use primitive scorers in my experiment, unfortunately. We’re using SDs to collect hits, write those out to ROOT N-tuples, and do analysis offline. If the scorers produce simple histogram-like output, then merging them should be as easy as summing the histograms.

5

I am interested in that if run these five jobs in parallel, we need to change the random number seed. if we use below command:
/run/beamOn 2000000000
/run/beamOn 2000000000
/run/beamOn 2000000000
/run/beamOn 2000000000
/run/beamOn 2000000000
Dose the project automatically change the random number seed at each run?

6

Depends on the code. Where in your code do you seed the random number generator?

7

It does depend on the user’s application code to set unique random seeds at the start of the job. However, the random engine will not be reset at the start of the run by Geant4. So doing multiple /run/beamOn in a single job will give you unique events in each run. But doing separate jobs each with a single /run/beamOn will give you identical results, unless your user application sets a unique seed.

8

You may have a look to README of example TestEm4 : Radom numbers handling
See macro rndmSeed.mac

9

If you write a macro to /run/beamOn 500 times just add a /random/SetSeeds/ X Y and set incrementing seeds

1 Like