Question about random seed reproducibility when detector XML is modified

[Dr-Stubborn]

I would like to ask whether the following behavior is expected in ddsim.

I am using a particle gun with a fixed photon energy and a fixed polar angle, while the azimuthal angle is generated uniformly between 0 and 360 degrees. The command I use is:

ddsim --random.seed 100 \
    --enableGun \
    --gun.distribution uniform \
    --gun.energy "100*GeV" \
    --gun.particle gamma \
    --gun.thetaMin "88.03*degree" \
    --gun.thetaMax "88.03*degree" \
    --gun.phiMin "0.0*degree" \
    --gun.phiMax "360.0*degree" \
    --numberOfEvents 10 \
    --outputFile ALLEGRO_sim.root \
    --compactFile $K4GEO/FCCee/ALLEGRO/compact/ALLEGRO_o1_v03/ALLEGRO_o1_v03.xml

With the same XML file and the same random seed, the results are reproducible.

However, if I modify the detector XML while keeping exactly the same command and the same random seed, I find that the events are no longer identical. In particular, from event 2 onward, the generated particle directions already become different. In my tests, this happens for example when I change the magnetic field strength in the XML, or when I remove the field entry.

What I compare is the particle-gun-generated photon direction reported in the ddsim output.

My expectation was the following: since the gun configuration is the same, and since the incident particle information should be determined by the random seed, I expected the generated gun kinematics to remain identical even if the detector geometry changes.

So my question is:

Is it expected that changing the detector XML can already change the particle-gun-generated events, even when the same random seed is used?

More specifically, is the intended behavior that modifications of the detector description can change the sequence of random-number calls already during event 1, so that the random-engine state is different by the time event 2 starts?

I understand that this may be normal if the same random engine is shared globally across the simulation, but I would like to confirm whether this is indeed the intended design in DD4hep / ddsim.

Thank you very much for your help.

[andresailer]

Yes, this behaviour is expected. Since the change in geometry will change the number of calls to the random engine the state at start of the second event will be different.

But you can enable the "EventSeeder" for ddsim, which will give each event a reproducible random seed, so at least the initial state for the particle should be the same.

DD4hep/DDG4/python/DDSim/Helper/Random.py

Line 25 in 1829429

self.enableEventSeed = False

[Dr-Stubborn]

Hi, @andresailer . Thank you for helping me.

If I understand correctly, after setting enableEventSeed = True, the random seed for each event should be generated from the initialSeed, eventID, and runID, and this event seed should then determine the initial state of the incoming particle.

However, in my test, after enabling this option, I used the same initialSeed with two different XML files and generated 2 events in each case. When I compare the two setups, the corresponding newSeed values for each event are the same, but the flight direction of the particle gun in the second event still changes.

So I am wondering whether I have misunderstood how EventSeeder works, or whether there could be another source of randomness affecting the gun direction.

[andresailer]

SignalHandler    INFO  ++ Re-apply signal handlers
GenerationInit   INFO  +++ Initializing event 1. Within run:0 event 1.
Gun              INFO  Shoot [0] [20.000 , 20.000] GeV mu- pos:(0.000 0.000 0.000)[mm] dir:( 0.000  0.000  1.000)
Gun              INFO  Particle [0] mu-          Mom:20.000 GeV vertex:( 0.000  0.000  0.000)[mm] direction:( 0.391  0.905  0.165)
Gun              INFO  +-> Interaction [0] [20.000 , 20.000] GeV mu- pos:(0.000 0.000 0.000)[mm]
Gun              INFO  +++   +-> ID:  0 mu-          status:00000002 PDG:    13 Vtx:(+0.00e+00,+0.00e+00,+0.00e+00)[mm] time: +0.00e+00 [ns] #Dau:  0 #Par:0      
PrimaryHandler   INFO  +++++ G4PrimaryVertex at (+0.00e+00,+0.00e+00,+0.00e+00) [mm] +0.00e+00 [ns]
EventSeeder1     INFO  At beginEvent: eventID=0, runID=0 initialSeed=1337, newSeed=4059355708
ParticleHandler  INFO  +++ Event 0 Begin event action. Access event related information.
### Birks coefficients used in run time
   G4_POLYSTYRENE     0.07943 mm/MeV     0.00841958 g/cm^2/MeV  massFactor=  504.48535518201 effCharge= 18.500592148089
Geant4Output2EDM4hep INFO  +++ Saving EDM4hep event 0 run 0.
GenerationInit   INFO  +++ Initializing event 2. Within run:0 event 2.
Gun              INFO  Shoot [1] [20.000 , 20.000] GeV mu- pos:(0.000 0.000 0.000)[mm] dir:( 0.000  0.000  1.000)
Gun              INFO  Particle [0] mu-          Mom:20.000 GeV vertex:( 0.000  0.000  0.000)[mm] direction:(-0.418  0.253  0.872)
Gun              INFO  +-> Interaction [0] [20.000 , 20.000] GeV mu- pos:(0.000 0.000 0.000)[mm]
Gun              INFO  +++   +-> ID:  0 mu-          status:00000002 PDG:    13 Vtx:(+0.00e+00,+0.00e+00,+0.00e+00)[mm] time: +0.00e+00 [ns] #Dau:  0 #Par:0      
PrimaryHandler   INFO  +++++ G4PrimaryVertex at (+0.00e+00,+0.00e+00,+0.00e+00) [mm] +0.00e+00 [ns]
EventSeeder1     INFO  At beginEvent: eventID=1, runID=0 initialSeed=1337, newSeed=3016717421
ParticleHandler  INFO  +++ Event 1 Begin event action. Access event related information.
Geant4Output2EDM4hep INFO  +++ Saving EDM4hep event 1 run 0.

The EventSeeder comes in after the primary particle is generated. That's not going to work.
I'll have a look after Easter if this can be rectified.