Madgraph + Pythia with ISR

[dirkzerwas]

Check duplicate issues.

• Checked for duplicates

Goal

Running MG + Pythia without ISR is ok, but of ISR is turned on:

• accepted/trials is not 100%
• error message is "cannot set mass"
• severity increases with sqrts
• leads to a modification of the cross section by a factor acc/trials
• can lead to abort

Issue opened with Pythia: https://gitlab.com/Pythia8/releases/-/issues/564

Operating System and Version

all

compiler

all

The version of the key4hep stack

all

Package Version

all

Reproducer

k4GeneratorsConfig ../examples/FermionProduction.yaml
cd Run-Cards/Madgraph/Muon350
./Run_Muon350_ISR.dat

Additional context

No response

[dirkzerwas]

Marco Zaro pointed to an exchange with Pythia/Torbjorn in 2020 where the precision of the LHE output lead to a code change in ProcessContainer.cxx

The final state is summed and a consistency check performed. On the attached file the first event fails, while the second one passes.
The sum of the 4 vectors in the final state is E,P = (0.0000001 GeV, 0, 0, 5 MeV). Since E<P for the photon it seems like there is an inconsistency. Summing the initial state e-, e+ gives the same vector, so the problem seems to be with the madgraph output.

2Events.txt

[dirkzerwas]

The correction will be available in Madgraph v 3.5.9 and following (bug fixes in Template/SubProcesses/LO/unweight.f and madgraph/various/lhe_parser.py

[dirkzerwas]

The boost is calculated in Templates/LO/SubProcesses/unwght.f
The beta (and gamma) of the boost is calculated on a single particle, either the electron or the positron. If we calculate both betas, they are not equal (see attached output "betad1" and "betad2")
"Problem detected" means after boost Ee- or Ee+ > Ebeam
In several cases the use of the other beta leads to a consistent system Ee- or Ee+ <= Ebeam
"Boosted particle" lists the 4 vector after boost for the two betas in the output file

run1_app.log

[dirkzerwas]

A couple more things:
the shat is calculated in Templates/LO/SubProcesses/genps.f as x1 * x2 stot. this seems to imply a slight mismatch for electrons as shat should have the massive terms me^2 and as P^2 is not E^2 there is a small correction.

Starting point:

• 410 times out of 1250 events the correction (fallback) is needed

The idea:

• use the e-e+ system after radiation with 4 vectors p1= (x1Ebeam, 0, 0, sqrt(x12Ebeam2-me**2)) p2= (x2Ebeam, 0, 0, -sqrt(x22Ebeam2-me**2)) instead as p1+p2 to determine gamma = Esyst/msyst
• this leads to some improvement: only 25%

Further improvement:

• correction not based on the 4-vectors, but only on x1 and x2 as gamma = sqrt(x1/x2)*Eb1/(Eb1+Eb2) +sqrt(x2/x1)*Eb2/(Eb1+Eb2)
• this reduces the number of fallbacks to 40 of 1250 (significant improvement)

Final improvement:

• correcton the gamma by the ratio of the "overimprovement" if the boost leads to EafterBoost > Ebeam

[dirkzerwas]

An example event is here:
ExampleEventMG.txt
with a detailed comparison of the use of the three ways of calculating the beta.

A less verbose output is here:
run_01_log.txt
betaNorm is the "usual" beta used
betaSyst is the "system calculated beta"
comparing the energies with the expected energie, we are at the limit of double precision calculations.