MadGraph event generation produces errors while running on the cluster, but not while running on my PC

Hi,

Not sure what I would be able to do for you here.
Would you be able to put a link to your model, such that I can try locally to see what the issue can be?
I'm more worried about some IO issue in this case actually. But it would be nice to run this process with full compiler security flag to be sure that noting special occurs.

Thanks,

Olivier

Hi,

thanks for your answer!

The model can be found under
https://github.com/LHC-DMWG/model-repository/tree/master/models/Pseudoscalar_2HDM

Is there an easy way to run the process with full compiler security flag for me?

Best,
Thomas

Thanks I will take a look.

> Is there an easy way to run the process with full compiler security flag
> for me?

Sure you can go to the Source directory.
You have a file make_opts
and inside a line
#FFLAGS+= -g -fbounds-check -ffpe-trap=invalid,zero,overflow,underflow,denormal -Wall

That you can uncomment or replace it by
FFLAGS+= -g -fcheck=all -ffpe-trap=invalid,zero,overflow,underflow,denormal -Wall
.
then in that directory you can do a "make clean" to be sure that everyting will be recompile.

Cheers,

Olivier

> On 19 Jun 2019, at 10:12, Thomas Hugle <email address hidden> wrote:
>
> Question #681396 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/681396
>
> Status: Answered => Open
>
> Thomas Hugle is still having a problem:
> Hi,
>
> thanks for your answer!
>
> The model can be found under
> https://github.com/LHC-DMWG/model-repository/tree/master/models/Pseudoscalar_2HDM
>
> Is there an easy way to run the process with full compiler security flag
> for me?
>
> Best,
> Thomas
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

As an update:

I ran the process with full compiler security flag and the errors appear to all be of the same type and happen at the same place in the code.
The error is a "Program received signal SIGFPE: Floating-point exception - erroneous arithmetic operation" and the full backtrace is:
Program received signal SIGFPE: Floating-point exception - erroneous arithmetic operation.

Backtrace for this error:
#0 0x2AE04EF56697
#1 0x2AE04EF56CDE
#2 0x2AE04F9E927F
#3 0xF75FAC in __reductiond_MOD_calcdpv1
#4 0xF7F971 in __reductiond_MOD_calcdred
#5 0xF90C63 in __reductiond_MOD_calcd
#6 0xE4C8BB in __reductiontn_MOD_calctn
#7 0xEA6452 in __collier_coefs_MOD_tn_main_checked_cll
#8 0xEB17F0 in __collier_coefs_MOD_tn_main_cll
#9 0xD6F6AC in __collier_tensors_MOD_tnten_main_checked_cll
#10 0xD72AA1 in __collier_tensors_MOD_tnten_main_cll
#11 0x467064 in ml5_0_0_1_collierloop_ at COLLIER_interface.f:362
#12 0x463F33 in ml5_0_0_1_tirloop_ at TIR_interface.f:204 (discriminator 1)
#13 0x485C67 in ml5_0_0_1_loop_4_ at CT_interface.f:1013 (discriminator 8)
#14 0x45B919 in ml5_0_0_1_loop_ct_calls_1_ at loop_CT_calls_1.f:114
#15 0x449385 in ml5_0_0_1_sloopmatrix_ at loop_matrix.f:1132
#16 0x44F9F4 in ml5_0_0_1_sloopmatrixhel_ at loop_matrix.f:2939 (discriminator 2)
#17 0x45025B in ml5_0_0_1_sloopmatrixhel_thres_ at loop_matrix.f:2981 (discriminator 2)
#18 0x44090D in smatrix1_ at matrix1.f:309 (discriminator 2)
#19 0x43F3FB in dsig1_ at auto_dsig1.f:125
#20 0x43C7C8 in dsigproc_ at auto_dsig.f:474 (discriminator 1)
#21 0x43E38E in dsig_ at auto_dsig.f:266
#22 0x10EBB8F in sample_full_ at dsample.f:137
#23 0x40BA0A in MAIN__ at driver.f:177
ERROR DETECTED

Maybe this helps in pinpointing the source of the error.

Best,
Thomas

Thanks,

However this points to an issue in collier where the only proposal that I can give is to unlink collier to see if this improves the situation. (I will put Valentin in cc if he has something smarter to tell)

One other idea is to reduce the flag to
 -g -fcheck=all -Wall

Since not allowing floating point exception might be too trash for external code like Collier.

Cheers,

Olivier

Thanks for the ideas on how to address the problem.
Unluckily neither of them freed me of the errors.

Reducing the flag to
-g -fcheck=all -Wall
I get the (assumingly) same error as in the beginning "STOP 1; ERROR DETECTED".

When disabling collier and running with full compiler security flag I also still get errors, but the backtrace changes a bit:
Program received signal SIGFPE: Floating-point exception - erroneous arithmetic operation.

Backtrace for this error:
#0 0x2AE86402A697
#1 0x2AE86402ACDE
#2 0x2AE864ABD27F
#3 0xD97885 in __discretesampler_MOD_ds_get_point_with_binid at DiscreteSampler.f:2049
#4 0xD9A05D in __discretesampler_MOD_ds_get_point_with_integerbinid at DiscreteSampler.f:1877
#5 0x43E2F9 in dsig_ at auto_dsig.f:330
#6 0xD8A34F in sample_full_ at dsample.f:137
#7 0x40B10A in MAIN__ at driver.f:177
ERROR DETECTED

Best,
Thomas

Here a bit more information on things that I tried:
Changing the compiler from gcc-4.8.5 to gcc-5.4.0 or gcc-7.2.0 did not help for the two Higgs doublet + pseudoscalar model, however it seems to have helped with another model (two Higgs doublet + scalar).

Digging a bit deeper, the error that triggers the breakdown of the simulation seems to be happening in the DiscreteSampler module (in principle this is already apparent from the full debug log in my very first post, but I did not fully realize it back then):
DiscreteSampler:: Error, no point could be picked with random variable 0.8378990292549133 using upper bound found of 0.0000000000000000.
(An error like this appears in the "run1_app.log" files of the SubProcesses that fail. Interestingly, not all SubProcesses fail.)

Also, before the error there are floating-point exceptions (cf. "run1_app.log" files of the SubProcesses):
Note: The following floating-point exceptions are signalling: IEEE_INVALID_FLAG IEEE_DIVIDE_BY_ZERO IEEE_UNDERFLOW_FLAG IEEE_DENORMAL
STOP 1

For completeness and since it might help understanding the problem, I will add the log files of a SubProcess that fails on the cluster but works on my PC.

Best,
Thomas

I leave out the model parameters in both files (they are identical to the full debug log posted above)

CLUSTER (giving an error):
=====================

 Collider parameters:
 --------------------

 Running at P P machine @ 13000.000000000000 GeV
 PDF set = lhapdf
 alpha_s(Mz)= 0.1300 running at 1 loops.
 alpha_s(Mz)= 0.1300 running at 1 loops.
 Renormalization scale set on event-by-event basis
 Factorization scale set on event-by-event basis

 getting user params
Enter number of events and max and min iterations:
 Number of events and iterations 4000 5 3
Enter desired fractional accuracy:
 Desired fractional accuracy: 2.0000000000000000E-002
Enter 0 for fixed, 2 for adjustable grid:
Suppress amplitude (0 no, 1 yes)?
 Using suppressed amplitude.
Exact helicity sum (0 yes, n = number/event)?
 Monte-Carlo over helicities

 ********************************************
 * You are using the DiscreteSampler module *
 * part of the MG5_aMC framework *
 * Author: Valentin Hirschi *
 ********************************************

Enter Configuration Number:
Running Configuration Number: 1
 Using dconfig= 14
 BW Setting 2 1 1 0
 Attempting mappinvarients 1 6
 Completed mapping 6
 about to integrate 10 4000 5 3 10 1
 Using non-zero grid deformation.
 10 dimensions 4000 events 10 invarients 5 iterations 1 config(s), (0.99)
 Using h-tuple random number sequence.
 Error opening grid
 Using Uniform Grid! 24
 Using uniform alpha 1.0000000000000000
 Grid defined OK
 Set CM energy to 13000.00
 Mapping Graph 1 to config 1
Setting grid 1 0.23669E-05 1
 Setting BW -2 2 91.187600000000003
 Setting PDF BW 9 3 125.00000000000000
 Transforming s_hat BW 9.2455621301775159E-005 4.8325680473372789E-009
   1 1 2 3 4 5 6 7 8 9 10
 Masses: 0.000E+00 0.000E+00 0.000E+00 0.000E+00 0.100E+02 0.100E+02
Using random seed offsets 1 : 1
  with seed 14
 Ranmar initialization seeds 18869 9388
  Particle 3 4 5 6
      Et > 10.0 10.0 0.0 0.0
       E > 0.0 0.0 0.0 0.0
     Eta < 2.5 2.5 -1.0 -1.0
   xqcut: 0.0 0.0 0.0 0.0
d R # 3 > -0.0 0.4 0.0 0.0
d R # 4 > -0.0 -0.0 0.0 0.0
d R # 5 > -0.0 -0.0 -0.0 0.0
s min # 3> 0.0 0.0 100.0 0.0
s min # 4> 0.0 0.0 100.0 0.0
s min # 5> 0.0 0.0 0.0 0.0
xqcutij # 3> 0.0 0.0 0.0 0.0
xqcutij # 4> 0.0 0.0 0.0 0.0
xqcutij # 5> 0.0 0.0 0.0 0.0
 Excluding BW -1 1
 Requiring BW -2 2
 Requiring BW -3 3
 alpha_s for scale 125.00588138446317 is 0.12380727518388480
  ==========================================================================================
 { }
 { ^[[32m ^[[0m }
 { ^[[32m ,, ^[[0m }
 { ^[[32m`7MMM. ,MMF' `7MM `7MMF' ^[[0m }
 { ^[[32m MMMb dPMM MM MM ^[[0m }
 { ^[[32m M YM ,M MM ,6"Yb. ,M""bMM MM ,pW"Wq. ,pW"Wq.`7MMpdMAo. ^[[0m }
 { ^[[32m M Mb M' MM 8) MM ,AP MM MM 6W' `Wb 6W' `Wb MM `Wb ^[[0m }
 { ^[[32m M YM.P' MM ,pm9MM 8MI MM MM , 8M M8 8M M8 MM M8 ^[[0m }
 { ^[[32m M `YM' MM 8M MM `Mb MM MM ,M YA. ,A9 YA. ,A9 MM ,AP ^[[0m }
 { ^[[32m.JML. `' .JMML.`Moo9^Yo.`Wbmd"MML..JMMmmmmMMM `Ybmd9' `Ybmd9' MMbmmd' ^[[0m }
 { ^[[32m MM ^[[0m }
 { ^[[32m .JMML. ^[[0m }
 { ^[[32m^[[0mv2.6.6 (2018-06-28), Ref: arXiv:1103.0621v2, arXiv:1405.0301^[[32m ^[[0m }
 { ^[[32m ^[[0m }
 { }
  ==========================================================================================
 ===============================================================
 INFO: MadLoop read these parameters from ../MadLoop5_resources/MadLoopParams.dat
 ===============================================================
  > MLReductionLib = 6|7|1
  > CTModeRun = -1
  > MLStabThres = 1.0000000000000000E-004
  > NRotations_DP = 1
  > NRotations_QP = 0
  > CTStabThres = 1.0000000000000000E-002
  > CTLoopLibrary = 2
  > CTModeInit = 1
  > CheckCycle = 4
  > MaxAttempts = 10
  > UseLoopFilter = F
  > HelicityFilterLevel = 1
  > ImprovePSPoint = 2
  > DoubleCheckHelicityFilter = F
  > LoopInitStartOver = F
  > HelInitStartOver = F
  > ZeroThres = 1.0000000000000001E-009
  > OSThres = 1.0000000000000000E-008
  > WriteOutFilters = F
  > UseQPIntegrandForNinja = T
  > UseQPIntegrandForCutTools = T
  > IREGIMODE = 2
  > IREGIRECY = T
  > COLLIERMode = 1
  > COLLIERRequiredAccuracy = 1.0000000000000000E-008
  > COLLIERCanOutput = F
  > COLLIERComputeUVpoles = T
  > COLLIERComputeIRpoles = T
  > COLLIERGlobalCache = -1
  > COLLIERUseCacheForPoles = F
  > COLLIERUseInternalStabilityTest = T
 ===============================================================
 ##INFO: For loop-induced processes it is preferable to always set the parameter LoopInitStartOver to True, so it is hard-set here to True.
 ##INFO: For loop-induced processes it is preferable to always set the parameter HelInitStartOver to True, so it is hard-set here to True.
 ##INFO: Due to the dynamic setting of the reference scale for contributions comparisons, it is preferable to set the parameter CheckCycle to a value larger than 4, so it is hard-set here to 5.
------------------------------------------------------------------------
| You are using CutTools - Version 1.9.3 |
| Authors: G. Ossola, C. Papadopoulos, R. Pittau |
| Published in JHEP 0803:042,2008 |
| http://www.ugr.es/~pittau/CutTools |
| |
| Compiler with 34 significant digits detetected |
 ----------------------------------------------------------------------

########################################################################
# #
# You are using OneLOop-3.6 #
# #
# for the evaluation of 1-loop scalar 1-, 2-, 3- and 4-point functions #
# #
# author: Andreas van Hameren <email address hidden> #
# date: 18-02-2015 #
# #
# Please cite #
# A. van Hameren, #

  +----------------------------------------------------------------+
  | |
  | Ninja - version 1.2.0 |
  | |
  | Author: Tiziano Peraro |
  | |
  | Based on: |
  | |
  | P. Mastrolia, E. Mirabella and T. Peraro, |
  | "Integrand reduction of one-loop scattering amplitudes |
  | through Laurent series expansion," |
  | JHEP 1206 (2012) 095 [arXiv:1203.0291 [hep-ph]]. |
  | |
  | T. Peraro, |
  | "Ninja: Automated Integrand Reduction via Laurent |
  | Expansion for One-Loop Amplitudes," |
  | Comput.Phys.Commun. 185 (2014) [arXiv:1403.1229 [hep-ph]] |
  | |
  +----------------------------------------------------------------+

Note: The following floating-point exceptions are signalling: IEEE_INVALID_FLAG IEEE_DIVIDE_BY_ZERO IEEE_UNDERFLOW_FLAG IEEE_DENORMAL
STOP 1
Thanks for using LHAPDF 6.1.6. Please make sure to cite the paper:
  Eur.Phys.J. C75 (2015) 3, 132 (http://arxiv.org/abs/1412.7420)
# Comput.Phys.Commun. 182 (2011) 2427-2438, arXiv:1007.4716 #
# A. van Hameren, C.G. Papadopoulos and R. Pittau, #
# JHEP 0909:106,2009, arXiv:0903.4665 #
# in publications with results obtained with the help of this program. #
# #
########################################################################
########################################################################
# #
# You are using OneLOop-3.6 #
# #
# for the evaluation of 1-loop scalar 1-, 2-, 3- and 4-point functions #
# #
# author: Andreas van Hameren <email address hidden> #
# date: 18-02-2015 #
# #
# Please cite #
# A. van Hameren, #
# Comput.Phys.Commun. 182 (2011) 2427-2438, arXiv:1007.4716 #
# A. van Hameren, C.G. Papadopoulos and R. Pittau, #
# JHEP 0909:106,2009, arXiv:0903.4665 #
# in publications with results obtained with the help of this program. #
# #
########################################################################
 RESET CUMULATIVE VARIABLE
 RESET CUMULATIVE VARIABLE

         *******************************************
         * C O L L I E R *
         * *
         * Complex One-Loop Library *
         * In Extended Regularizations *
         * *
         * by A.Denner, S.Dittmaier, L.Hofer *
         * *
         * version 1.2.3 *
         * *
         *******************************************

 Iteration 1 Mean: 0.3922E-16 Abs mean: 0.3922E-16 Fluctuation: 0.283E-17 0.398E-14 2.7%
  1 0.3922E-16 0.3922E-16 +- 0.2834E-17 4.57
 Writing out events 7.2648745727604389E-020 4.5698840605824955
 Relative summed weights:
  0.1000E+01 0.0000E+00
 Relative number of events:
  0.1000E+01 0.0000E+00
 Events:
        4000 0
 DiscreteSampler:: Error, no point could be picked with random variable 0.7042701840400696 using upper bound found of 0.0000000000000000.

ls status:
input_app.txt
run1_app.log

PC (works):
=========
 Collider parameters:
 --------------------

 Running at P P machine @ 13000.000000000000 GeV
 PDF set = lhapdf
 alpha_s(Mz)= 0.1300 running at 1 loops.
 alpha_s(Mz)= 0.1300 running at 1 loops.
 Renormalization scale set on event-by-event basis
 Factorization scale set on event-by-event basis

 getting user params
Enter number of events and max and min iterations:
 Number of events and iterations 2000 1 1
Enter desired fractional accuracy:
 Desired fractional accuracy: 2.0000000000000000E-002
Enter 0 for fixed, 2 for adjustable grid:
Suppress amplitude (0 no, 1 yes)?
 Using suppressed amplitude.
Exact helicity sum (0 yes, n = number/event)?
 Monte-Carlo over helicities

 ********************************************
 * You are using the DiscreteSampler module *
 * part of the MG5_aMC framework *
 * Author: Valentin Hirschi *
 ********************************************

Enter Configuration Number:
Running Configuration Number: 1
 Using dconfig= 14
 BW Setting 2 1 1 0
 Attempting mappinvarients 1 6
 Completed mapping 6
 about to integrate 10 2000 1 1 10 1
 Using unknown grid deformation: -2
 10 dimensions 2000 events 10 invarients 1 iterations 1 config(s), (0.99)
 Using h-tuple random number sequence.
 Error opening grid
 Using Uniform Grid! 24
 Using uniform alpha 1.0000000000000000
 Grid defined OK
 Set CM energy to 13000.00
 Mapping Graph 1 to config 1
Setting grid 1 0.23669E-05 1
 Setting BW -2 2 91.187600000000003
 Setting PDF BW 9 3 125.00000000000000
 Transforming s_hat BW 9.2455621301775159E-005 4.8325680473372789E-009
   1 1 2 3 4 5 6 7 8 9 10
 Masses: 0.000E+00 0.000E+00 0.000E+00 0.000E+00 0.100E+02 0.100E+02
 Got moffset 1
Using random seed offsets 1 : 1
  with seed 14
 Ranmar initialization seeds 18869 12545
  Particle 3 4 5 6
      Et > 10.0 10.0 0.0 0.0
       E > 0.0 0.0 0.0 0.0
     Eta < 2.5 2.5 -1.0 -1.0
   xqcut: 0.0 0.0 0.0 0.0
d R # 3 > -0.0 0.4 0.0 0.0
d R # 4 > -0.0 -0.0 0.0 0.0
d R # 5 > -0.0 -0.0 -0.0 0.0
s min # 3> 0.0 0.0 0.0 0.0
s min # 4> 0.0 0.0 0.0 0.0
s min # 5> 0.0 0.0 0.0 0.0
xqcutij # 3> 0.0 0.0 0.0 0.0
xqcutij # 4> 0.0 0.0 0.0 0.0
xqcutij # 5> 0.0 0.0 0.0 0.0
 Excluding BW -1 1
 Requiring BW -2 2
 Requiring BW -3 3
 alpha_s for scale 124.99908171426037 is 0.12380829279978629
  ==========================================================================================
 { }
 {   }
 {  ,,  }
 { `7MMM. ,MMF' `7MM `7MMF'  }
 {  MMMb dPMM MM MM  }
 {  M YM ,M MM ,6"Yb. ,M""bMM MM ,pW"Wq. ,pW"Wq.`7MMpdMAo.  }
 {  M Mb M' MM 8) MM ,AP MM MM 6W' `Wb 6W' `Wb MM `Wb  }
 {  M YM.P' MM ,pm9MM 8MI MM MM , 8M M8 8M M8 MM M8  }
 {  M `YM' MM 8M MM `Mb MM MM ,M YA. ,A9 YA. ,A9 MM ,AP  }
 { .JML. `' .JMML.`Moo9^Yo.`Wbmd"MML..JMMmmmmMMM `Ybmd9' `Ybmd9' MMbmmd'  }
 {  MM  }
 {  .JMML.  }
 { v2.6.5 (2018-02-03), Ref: arXiv:1103.0621v2, arXiv:1405.0301  }
 {   }
 { }
  ==========================================================================================
 ===============================================================
 INFO: MadLoop read these parameters from ../MadLoop5_resources/MadLoopParams.dat
 ===============================================================
  > MLReductionLib = 6|7|1
  > CTModeRun = -1
  > MLStabThres = 1.0000000000000000E-004
  > NRotations_DP = 1
  > NRotations_QP = 0
  > CTStabThres = 1.0000000000000000E-002
  > CTLoopLibrary = 2
  > CTModeInit = 1
  > CheckCycle = 4
  > MaxAttempts = 10
  > UseLoopFilter = F
  > HelicityFilterLevel = 1
  > ImprovePSPoint = 2
  > DoubleCheckHelicityFilter = F
  > LoopInitStartOver = F
  > HelInitStartOver = F
  > ZeroThres = 1.0000000000000001E-009
  > OSThres = 1.0000000000000000E-008
  > WriteOutFilters = F
  > UseQPIntegrandForNinja = T
  > UseQPIntegrandForCutTools = T
  > IREGIMODE = 2
  > IREGIRECY = T
  > COLLIERMode = 1
  > COLLIERRequiredAccuracy = 1.0000000000000000E-008
  > COLLIERCanOutput = F
  > COLLIERComputeUVpoles = T
  > COLLIERComputeIRpoles = T
  > COLLIERGlobalCache = -1
  > COLLIERUseCacheForPoles = F
  > COLLIERUseInternalStabilityTest = T
 ===============================================================
 ##INFO: For loop-induced processes it is preferable to always set the parameter LoopInitStartOver to True, so it is hard-set here to True.
 ##INFO: For loop-induced processes it is preferable to always set the parameter HelInitStartOver to True, so it is hard-set here to True.
 ##INFO: Due to the dynamic setting of the reference scale for contributions comparisons, it is preferable to set the parameter CheckCycle to a value larger than 4, so it is hard-set here to 5.

------------------------------------------------------------------------
| You are using CutTools - Version 1.9.3 |
| Authors: G. Ossola, C. Papadopoulos, R. Pittau |
| Published in JHEP 0803:042,2008 |
| http://www.ugr.es/~pittau/CutTools |
| |
| Compiler with 34 significant digits detetected |
 ----------------------------------------------------------------------

########################################################################
# #
# You are using OneLOop-3.6 #
# #
# for the evaluation of 1-loop scalar 1-, 2-, 3- and 4-point functions #
# #
# author: Andreas van Hameren <email address hidden> #
# date: 18-02-2015 #
# #
# Please cite #
# A. van Hameren, #

  +----------------------------------------------------------------+
  | |
  | Ninja - version 1.2.0 |
  | |
  | Author: Tiziano Peraro |
  | |
  | Based on: |
  | |
  | P. Mastrolia, E. Mirabella and T. Peraro, |
  | "Integrand reduction of one-loop scattering amplitudes |
  | through Laurent series expansion," |
  | JHEP 1206 (2012) 095 [arXiv:1203.0291 [hep-ph]]. |
  | |
  | T. Peraro, |
  | "Ninja: Automated Integrand Reduction via Laurent |
  | Expansion for One-Loop Amplitudes," |
  | Comput.Phys.Commun. 185 (2014) [arXiv:1403.1229 [hep-ph]] |
  | |
  +----------------------------------------------------------------+

Thanks for using LHAPDF 6.1.6. Please make sure to cite the paper:
  Eur.Phys.J. C75 (2015) 3, 132 (http://arxiv.org/abs/1412.7420)
# Comput.Phys.Commun. 182 (2011) 2427-2438, arXiv:1007.4716 #
# A. van Hameren, C.G. Papadopoulos and R. Pittau, #
# JHEP 0909:106,2009, arXiv:0903.4665 #
# in publications with results obtained with the help of this program. #
# #
########################################################################
########################################################################
# #
# You are using OneLOop-3.6 #
# #
# for the evaluation of 1-loop scalar 1-, 2-, 3- and 4-point functions #
# #
# author: Andreas van Hameren <email address hidden> #
# date: 18-02-2015 #
# #
# Please cite #
# A. van Hameren, #
# Comput.Phys.Commun. 182 (2011) 2427-2438, arXiv:1007.4716 #
# A. van Hameren, C.G. Papadopoulos and R. Pittau, #
# JHEP 0909:106,2009, arXiv:0903.4665 #
# in publications with results obtained with the help of this program. #
# #
########################################################################
 RESET CUMULATIVE VARIABLE
 RESET CUMULATIVE VARIABLE
 Iteration 1 Mean: 0.3474E-16 Abs mean: 0.3474E-16 Fluctuation: 0.301E-17 0.266E-14 2.2%
  1 0.3474E-16 0.3474E-16 +- 0.3009E-17 3.87
 Writing out events 7.9538527730590813E-019 3.8733910142143664
 Relative summed weights:
  0.1000E+01 0.0000E+00
 Relative number of events:
  0.1000E+01 0.0000E+00
 Events:
        2000 0
 Accuracy: 0.000 0.020 0.087 0.000
 Finished due to accuracy 0.0000000000000000 2.0000000000000000E-002

 -------------------------------------------------------------------------------
 Accumulated results: Integral = 0.3474E-16
                        Std dev = 0.3009E-17
                       Cross sec = 0.3474E-16
             Chi**2 per DoF. = 0.0000
 -------------------------------------------------------------------------------
-------------------------------------------------
---------------------------
 Results Last 1 iters: Integral = 0.3474E-16
                     Abs integral = 0.3474E-16
                          Std dev = 0.3009E-17
                  Chi**2 per DoF. = 0.0000
-------------------------------------------------
---------------------------
 Status 2.0000000000000000E-002 2 1

ls status:
ftn26
grid_information
input_app.txt
moffset.dat
results.dat
run1_app.log

Hi,

DiscreteSampler:: Error, no point could be picked with random variable 0.8378990292549133 using upper bound found of 0.0000000000000000.

This message is due to a consistency check of the code.
The code realises that something is wrong and stops.

One thing that you need to check is if the IEEE_DIVIDE_BY_ZERO is coming from your model.
One method is to put your param_card within the UFO model directory and renamed it restrict_xxx.dat
then do import model 2HDM_pseudoscalar-xxx
If you have a division by zero, it should crash and point you to the issue in your code.

Doing that will also remove from the model any coupling which are zero (or less than 1e-10).
Which can also help fo the other flag (or hurt you if you need one of such suppressed coupling)

Cheers,

Olivier

On 5 Jul 2019, at 11:47, Thomas Hugle <<email address hidden><mailto:<email address hidden>>> wrote:

Question #681396 on MadGraph5_aMC@NLO changed:
https://answers.launchpad.net/mg5amcnlo/+question/681396

Thomas Hugle gave more information on the question:
Here a bit more information on things that I tried:
Changing the compiler from gcc-4.8.5 to gcc-5.4.0 or gcc-7.2.0 did not help for the two Higgs doublet + pseudoscalar model, however it seems to have helped with another model (two Higgs doublet + scalar).

Digging a bit deeper, the error that triggers the breakdown of the simulation seems to be happening in the DiscreteSampler module (in principle this is already apparent from the full debug log in my very first post, but I did not fully realize it back then):
DiscreteSampler:: Error, no point could be picked with random variable 0.8378990292549133 using upper bound found of 0.0000000000000000.
(An error like this appears in the "run1_app.log" files of the SubProcesses that fail. Interestingly, not all SubProcesses fail.)

Also, before the error there are floating-point exceptions (cf. "run1_app.log" files of the SubProcesses):
Note: The following floating-point exceptions are signalling: IEEE_INVALID_FLAG IEEE_DIVIDE_BY_ZERO IEEE_UNDERFLOW_FLAG IEEE_DENORMAL
STOP 1

For completeness and since it might help understanding the problem, I
will add the log files of a SubProcess that fails on the cluster but
works on my PC.

Best,
Thomas

I leave out the model parameters in both files (they are identical to
the full debug log posted above)

CLUSTER (giving an error):
=====================

Collider parameters:
--------------------

Running at P P machine @ 13000.000000000000 GeV
PDF set = lhapdf
alpha_s(Mz)= 0.1300 running at 1 loops.
alpha_s(Mz)= 0.1300 running at 1 loops.
Renormalization scale set on event-by-event basis
Factorization scale set on event-by-event basis

getting user params
Enter number of events and max and min iterations:
Number of events and iterations 4000 5 3
Enter desired fractional accuracy:
Desired fractional accuracy: 2.0000000000000000E-002
Enter 0 for fixed, 2 for adjustable grid:
Suppress amplitude (0 no, 1 yes)?
Using suppressed amplitude.
Exact helicity sum (0 yes, n = number/event)?
Monte-Carlo over helicities

********************************************
* You are using the DiscreteSampler module *
* part of the MG5_aMC framework *
* Author: Valentin Hirschi *
********************************************

Enter Configuration Number:
Running Configuration Number: 1
Using dconfig= 14
BW Setting 2 1 1 0
Attempting mappinvarients 1 6
Completed mapping 6
about to integrate 10 4000 5 3 10 1
Using non-zero grid deformation.
10 dimensions 4000 events 10 invarients 5 iterations 1 config(s), (0.99)
Using h-tuple random number sequence.
Error opening grid
Using Uniform Grid! 24
Using uniform alpha 1.0000000000000000
Grid defined OK
Set CM energy to 13000.00
Mapping Graph 1 to config 1
Setting grid 1 0.23669E-05 1
Setting BW -2 2 91.187600000000003
Setting PDF BW 9 3 125.00000000000000
Transforming s_hat BW 9.2455621301775159E-005 4.8325680473372789E-009
  1 1 2 3 4 5 6 7 8 9 10
Masses: 0.000E+00 0.000E+00 0.000E+00 0.000E+00 0.100E+02 0.100E+02
Using random seed offsets 1 : 1
 with seed 14
Ranmar initialization seeds 18869 9388
 Particle 3 4 5 6
     Et > 10.0 10.0 0.0 0.0
      E > 0.0 0.0 0.0 0.0
    Eta < 2.5 2.5 -1.0 -1.0
  xqcut: 0.0 0.0 0.0 0.0
d R # 3 > -0.0 0.4 0.0 0.0
d R # 4 > -0.0 -0.0 0.0 0.0
d R # 5 > -0.0 -0.0 -0.0 0.0
s min # 3> 0.0 0.0 100.0 0.0
s min # 4> 0.0 0.0 100.0 0.0
s min # 5> 0.0 0.0 0.0 0.0
xqcutij # 3> 0.0 0.0 0.0 0.0
xqcutij # 4> 0.0 0.0 0.0 0.0
xqcutij # 5> 0.0 0.0 0.0 0.0
Excluding BW -1 1
Requiring BW -2 2
Requiring BW -3 3
alpha_s for scale 125.00588138446317 is 0.12380727518388480
 ==========================================================================================
{ }
{ ^[[32m ^[[0m }
{ ^[[32m ,, ^[[0m }
{ ^[[32m`7MMM. ,MMF' `7MM `7MMF' ^[[0m }
{ ^[[32m MMMb dPMM MM MM ^[[0m }
{ ^[[32m M YM ,M MM ,6"Yb. ,M""bMM MM ,pW"Wq. ,pW"Wq.`7MMpdMAo. ^[[0m }
{ ^[[32m M Mb M' MM 8) MM ,AP MM MM 6W' `Wb 6W' `Wb MM `Wb ^[[0m }
{ ^[[32m M YM.P' MM ,pm9MM 8MI MM MM , 8M M8 8M M8 MM M8 ^[[0m }
{ ^[[32m M `YM' MM 8M MM `Mb MM MM ,M YA. ,A9 YA. ,A9 MM ,AP ^[[0m }
{ ^[[32m.JML. `' .JMML.`Moo9^Yo.`Wbmd"MML..JMMmmmmMMM `Ybmd9' `Ybmd9' MMbmmd' ^[[0m }
{ ^[[32m MM ^[[0m }
{ ^[[32m .JMML. ^[[0m }
{ ^[[32m^[[0mv2.6.6 (2018-06-28), Ref: arXiv:1103.0621v2, arXiv:1405.0301^[[32m ^[[0m }
{ ^[[32m ^[[0m }
{ }
 ==========================================================================================
===============================================================
INFO: MadLoop read these parameters from ../MadLoop5_resources/MadLoopParams.dat
===============================================================
MLReductionLib = 6|7|1
CTModeRun = -1
MLStabThres = 1.0000000000000000E-004
NRotations_DP = 1
NRotations_QP = 0
CTStabThres = 1.0000000000000000E-002
CTLoopLibrary = 2
CTModeInit = 1
CheckCycle = 4
MaxAttempts = 10
UseLoopFilter = F
HelicityFilterLevel = 1
ImprovePSPoint = 2
DoubleCheckHelicityFilter = F
LoopInitStartOver = F
HelInitStartOver = F
ZeroThres = 1.0000000000000001E-009
OSThres = 1.0000000000000000E-008
WriteOutFilters = F
UseQPIntegrandForNinja = T
UseQPIntegrandForCutTools = T
IREGIMODE = 2
IREGIRECY = T
COLLIERMode = 1
COLLIERRequiredAccuracy = 1.0000000000000000E-008
COLLIERCanOutput = F
COLLIERComputeUVpoles = T
COLLIERComputeIRpoles = T
COLLIERGlobalCache = -1
COLLIERUseCacheForPoles = F
COLLIERUseInternalStabilityTest = T
===============================================================
##INFO: For loop-induced processes it is preferable to always set the parameter LoopInitStartOver to True, so it is hard-set here to True.
##INFO: For loop-induced processes it is preferable to always set the parameter HelInitStartOver to True, so it is hard-set here to True.
##INFO: Due to the dynamic setting of the reference scale for contributions comparisons, it is preferable to set the parameter CheckCycle to a value larger than 4, so it is hard-set here to 5.
------------------------------------------------------------------------
| You are using CutTools - Version 1.9.3 |
| Authors: G. Ossola, C. Papadopoulos, R. Pittau |
| Published in JHEP 0803:042,2008 |
| http://www.ugr.es/~pittau/CutTools |
| |
| Compiler with 34 significant digits detetected |
----------------------------------------------------------------------

########################################################################
# #
# You are using OneLOop-3.6 #
# #
# for the evaluation of 1-loop scalar 1-, 2-, 3- and 4-point functions #
# #
# author: Andreas van Hameren <email address hidden> #
# date: 18-02-2015 #
# #
# Please cite #
# A. van Hameren, #

 +----------------------------------------------------------------+
 | |
 | Ninja - version 1.2.0 |
 | |
 | Author: Tiziano Peraro |
 | |
 | Based on: |
 | |
 | P. Mastrolia, E. Mirabella and T. Peraro, |
 | "Integrand reduction of one-loop scattering amplitudes |
 | through Laurent series expansion," |
 | JHEP 1206 (2012) 095 [arXiv:1203.0291 [hep-ph]]. |
 | |
 | T. Peraro, |
 | "Ninja: Automated Integrand Reduction via Laurent |
 | Expansion for One-Loop Amplitudes," |
 | Comput.Phys.Commun. 185 (2014) [arXiv:1403.1229 [hep-ph]] |
 | |
 +----------------------------------------------------------------+

Note: The following floating-point exceptions are signalling: IEEE_INVALID_FLAG IEEE_DIVIDE_BY_ZERO IEEE_UNDERFLOW_FLAG IEEE_DENORMAL
STOP 1
Thanks for using LHAPDF 6.1.6. Please make sure to cite the paper:
 Eur.Phys.J. C75 (2015) 3, 132 (http://arxiv.org/abs/1412.7420)
# Comput.Phys.Commun. 182 (2011) 2427-2438, arXiv:1007.4716 #
# A. van Hameren, C.G. Papadopoulos and R. Pittau, #
# JHEP 0909:106,2009, arXiv:0903.4665 #
# in publications with results obtained with the help of this program. #
# #
########################################################################
########################################################################
# #
# You are using OneLOop-3.6 #
# #
# for the evaluation of 1-loop scalar 1-, 2-, 3- and 4-point functions #
# #
# author: Andreas van Hameren <email address hidden> #
# date: 18-02-2015 #
# #
# Please cite #
# A. van Hameren, #
# Comput.Phys.Commun. 182 (2011) 2427-2438, arXiv:1007.4716 #
# A. van Hameren, C.G. Papadopoulos and R. Pittau, #
# JHEP 0909:106,2009, arXiv:0903.4665 #
# in publications with results obtained with the help of this program. #
# #
########################################################################
RESET CUMULATIVE VARIABLE
RESET CUMULATIVE VARIABLE

        *******************************************
        * C O L L I E R *
        * *
        * Complex One-Loop Library *
        * In Extended Regularizations *
        * *
        * by A.Denner, S.Dittmaier, L.Hofer *
        * *
        * version 1.2.3 *
        * *
        *******************************************

Iteration 1 Mean: 0.3922E-16 Abs mean: 0.3922E-16 Fluctuation: 0.283E-17 0.398E-14 2.7%
 1 0.3922E-16 0.3922E-16 +- 0.2834E-17 4.57
Writing out events 7.2648745727604389E-020 4.5698840605824955
Relative summed weights:
 0.1000E+01 0.0000E+00
Relative number of events:
 0.1000E+01 0.0000E+00
Events:
       4000 0
DiscreteSampler:: Error, no point could be picked with random variable 0.7042701840400696 using upper bound found of 0.0000000000000000.

ls status:
input_app.txt
run1_app.log

PC (works):
=========
Collider parameters:
--------------------

Running at P P machine @ 13000.000000000000 GeV
PDF set = lhapdf
alpha_s(Mz)= 0.1300 running at 1 loops.
alpha_s(Mz)= 0.1300 running at 1 loops.
Renormalization scale set on event-by-event basis
Factorization scale set on event-by-event basis

getting user params
Enter number of events and max and min iterations:
Number of events and iterations 2000 1 1
Enter desired fractional accuracy:
Desired fractional accuracy: 2.0000000000000000E-002
Enter 0 for fixed, 2 for adjustable grid:
Suppress amplitude (0 no, 1 yes)?
Using suppressed amplitude.
Exact helicity sum (0 yes, n = number/event)?
Monte-Carlo over helicities

********************************************
* You are using the DiscreteSampler module *
* part of the MG5_aMC framework *
* Author: Valentin Hirschi *
********************************************

Enter Configuration Number:
Running Configuration Number: 1
Using dconfig= 14
BW Setting 2 1 1 0
Attempting mappinvarients 1 6
Completed mapping 6
about to integrate 10 2000 1 1 10 1
Using unknown grid deformation: -2
10 dimensions 2000 events 10 invarients 1 iterations 1 config(s), (0.99)
Using h-tuple random number sequence.
Error opening grid
Using Uniform Grid! 24
Using uniform alpha 1.0000000000000000
Grid defined OK
Set CM energy to 13000.00
Mapping Graph 1 to config 1
Setting grid 1 0.23669E-05 1
Setting BW -2 2 91.187600000000003
Setting PDF BW 9 3 125.00000000000000
Transforming s_hat BW 9.2455621301775159E-005 4.8325680473372789E-009
  1 1 2 3 4 5 6 7 8 9 10
Masses: 0.000E+00 0.000E+00 0.000E+00 0.000E+00 0.100E+02 0.100E+02
Got moffset 1
Using random seed offsets 1 : 1
 with seed 14
Ranmar initialization seeds 18869 12545
 Particle 3 4 5 6
     Et > 10.0 10.0 0.0 0.0
      E > 0.0 0.0 0.0 0.0
    Eta < 2.5 2.5 -1.0 -1.0
  xqcut: 0.0 0.0 0.0 0.0
d R # 3 > -0.0 0.4 0.0 0.0
d R # 4 > -0.0 -0.0 0.0 0.0
d R # 5 > -0.0 -0.0 -0.0 0.0
s min # 3> 0.0 0.0 0.0 0.0
s min # 4> 0.0 0.0 0.0 0.0
s min # 5> 0.0 0.0 0.0 0.0
xqcutij # 3> 0.0 0.0 0.0 0.0
xqcutij # 4> 0.0 0.0 0.0 0.0
xqcutij # 5> 0.0 0.0 0.0 0.0
Excluding BW -1 1
Requiring BW -2 2
Requiring BW -3 3
alpha_s for scale 124.99908171426037 is 0.12380829279978629
 ==========================================================================================
{ }
{   }
{  ,,  }
{ `7MMM. ,MMF' `7MM `7MMF'  }
{  MMMb dPMM MM MM  }
{  M YM ,M MM ,6"Yb. ,M""bMM MM ,pW"Wq. ,pW"Wq.`7MMpdMAo.  }
{  M Mb M' MM 8) MM ,AP MM MM 6W' `Wb 6W' `Wb MM `Wb  }
{  M YM.P' MM ,pm9MM 8MI MM MM , 8M M8 8M M8 MM M8  }
{  M `YM' MM 8M MM `Mb MM MM ,M YA. ,A9 YA. ,A9 MM ,AP  }
{ .JML. `' .JMML.`Moo9^Yo.`Wbmd"MML..JMMmmmmMMM `Ybmd9' `Ybmd9' MMbmmd'  }
{  MM  }
{  .JMML.  }
{ v2.6.5 (2018-02-03), Ref: arXiv:1103.0621v2, arXiv:1405.0301  }
{   }
{ }
 ==========================================================================================
===============================================================
INFO: MadLoop read these parameters from ../MadLoop5_resources/MadLoopParams.dat
===============================================================
MLReductionLib = 6|7|1
CTModeRun = -1
MLStabThres = 1.0000000000000000E-004
NRotations_DP = 1
NRotations_QP = 0
CTStabThres = 1.0000000000000000E-002
CTLoopLibrary = 2
CTModeInit = 1
CheckCycle = 4
MaxAttempts = 10
UseLoopFilter = F
HelicityFilterLevel = 1
ImprovePSPoint = 2
DoubleCheckHelicityFilter = F
LoopInitStartOver = F
HelInitStartOver = F
ZeroThres = 1.0000000000000001E-009
OSThres = 1.0000000000000000E-008
WriteOutFilters = F
UseQPIntegrandForNinja = T
UseQPIntegrandForCutTools = T
IREGIMODE = 2
IREGIRECY = T
COLLIERMode = 1
COLLIERRequiredAccuracy = 1.0000000000000000E-008
COLLIERCanOutput = F
COLLIERComputeUVpoles = T
COLLIERComputeIRpoles = T
COLLIERGlobalCache = -1
COLLIERUseCacheForPoles = F
COLLIERUseInternalStabilityTest = T
===============================================================
##INFO: For loop-induced processes it is preferable to always set the parameter LoopInitStartOver to True, so it is hard-set here to True.
##INFO: For loop-induced processes it is preferable to always set the parameter HelInitStartOver to True, so it is hard-set here to True.
##INFO: Due to the dynamic setting of the reference scale for contributions comparisons, it is preferable to set the parameter CheckCycle to a value larger than 4, so it is hard-set here to 5.

------------------------------------------------------------------------
| You are using CutTools - Version 1.9.3 |
| Authors: G. Ossola, C. Papadopoulos, R. Pittau |
| Published in JHEP 0803:042,2008 |
| http://www.ugr.es/~pittau/CutTools |
| |
| Compiler with 34 significant digits detetected |
----------------------------------------------------------------------

########################################################################
# #
# You are using OneLOop-3.6 #
# #
# for the evaluation of 1-loop scalar 1-, 2-, 3- and 4-point functions #
# #
# author: Andreas van Hameren <email address hidden> #
# date: 18-02-2015 #
# #
# Please cite #
# A. van Hameren, #

 +----------------------------------------------------------------+
 | |
 | Ninja - version 1.2.0 |
 | |
 | Author: Tiziano Peraro |
 | |
 | Based on: |
 | |
 | P. Mastrolia, E. Mirabella and T. Peraro, |
 | "Integrand reduction of one-loop scattering amplitudes |
 | through Laurent series expansion," |
 | JHEP 1206 (2012) 095 [arXiv:1203.0291 [hep-ph]]. |
 | |
 | T. Peraro, |
 | "Ninja: Automated Integrand Reduction via Laurent |
 | Expansion for One-Loop Amplitudes," |
 | Comput.Phys.Commun. 185 (2014) [arXiv:1403.1229 [hep-ph]] |
 | |
 +----------------------------------------------------------------+

Thanks for using LHAPDF 6.1.6. Please make sure to cite the paper:
 Eur.Phys.J. C75 (2015) 3, 132 (http://arxiv.org/abs/1412.7420)
# Comput.Phys.Commun. 182 (2011) 2427-2438, arXiv:1007.4716 #
# A. van Hameren, C.G. Papadopoulos and R. Pittau, #
# JHEP 0909:106,2009, arXiv:0903.4665 #
# in publications with results obtained with the help of this program. #
# #
########################################################################
########################################################################
# #
# You are using OneLOop-3.6 #
# #
# for the evaluation of 1-loop scalar 1-, 2-, 3- and 4-point functions #
# #
# author: Andreas van Hameren <email address hidden> #
# date: 18-02-2015 #
# #
# Please cite #
# A. van Hameren, #
# Comput.Phys.Commun. 182 (2011) 2427-2438, arXiv:1007.4716 #
# A. van Hameren, C.G. Papadopoulos and R. Pittau, #
# JHEP 0909:106,2009, arXiv:0903.4665 #
# in publications with results obtained with the help of this program. #
# #
########################################################################
RESET CUMULATIVE VARIABLE
RESET CUMULATIVE VARIABLE
Iteration 1 Mean: 0.3474E-16 Abs mean: 0.3474E-16 Fluctuation: 0.301E-17 0.266E-14 2.2%
 1 0.3474E-16 0.3474E-16 +- 0.3009E-17 3.87
Writing out events 7.9538527730590813E-019 3.8733910142143664
Relative summed weights:
 0.1000E+01 0.0000E+00
Relative number of events:
 0.1000E+01 0.0000E+00
Events:
       2000 0
Accuracy: 0.000 0.020 0.087 0.000
Finished due to accuracy 0.0000000000000000 2.0000000000000000E-002

-------------------------------------------------------------------------------
Accumulated results: Integral = 0.3474E-16
                       Std dev = 0.3009E-17
                      Cross sec = 0.3474E-16
            Chi**2 per DoF. = 0.0000
-------------------------------------------------------------------------------
-------------------------------------------------
---------------------------
Results Last 1 iters: Integral = 0.3474E-16
                    Abs integral = 0.3474E-16
                         Std dev = 0.3009E-17
                 Chi**2 per DoF. = 0.0000
-------------------------------------------------
---------------------------
Status 2.0000000000000000E-002 2 1

ls status:
ftn26
grid_information
input_app.txt
moffset.dat
results.dat
run1_app.log

--
You received this question notification because you are an answer
contact for MadGraph5_aMC@NLO.

Hey,

the problem is solved :)

The issue was that I had set (under input/mg5_configuration.txt) run_mode = 0 (single machine), which is automatically switched to run_mode=2 (multicore) with one core for loop induced processes as the one I am dealing with. However, this seems to have broken the discrete sampling routine.

Directly working with run_mode=2 and nb_core=4 got rid of the errors and the simulation appears to work as intended.

Best,
Thomas

Thanks, for your support!

You are not using run_mode=1 on the cluster?
Do I understand correctly that you want to run on 4 core of a single machine of the cluster (via manual submision of the job?)

On your PC, what was the option used? run_mode=0 or run_mode=2?

Cheers,

Olivier

Yes, I am not using run_mode=1 on the cluster. Since I am running a parameter scan, it seemed simpler to me to manually submit my jobs because didn't manage to figure out how to change my run parameters when using the cluster mode. Is there a simple way to do so?
My impression was that the cluster is well suited to calculate many events for one parameter set but not that well suited to calculate less events for many different parameter points. Of course this impression might be wrong.

Yes, I assume the cluster gives me 4 cores of a single node (via manual submission).

On my PC I did not set the run_mode option explicitly, but I assume it defaults to run_mode=2 since my computer used all of its cores.

Best,
Thomas

Thanks for the info, I will check on my laptop if I can reproduce issue with run_mode=0 then.

> Since I am running a parameter scan, it seemed simpler to me to manually submit my jobs because didn't manage to figure out how to change my run parameters when using the cluster mode. Is there a simple way to do so?

in the param_card you can set any value to
scan:[val1, val2, val3]

You have more advance syntax explained here:
https://answers.launchpad.net/mg5amcnlo/+faq/2735

> My impression was that the cluster is well suited to calculate many events for one parameter set but not that well suited to calculate less events for many different parameter points. Of course this impression might be wrong.

Not completely wrong, since you can not run within the same directory two benchmark simultaneously.
The various benchmark are done sequentially. This being said the computation of loop-induced process are very slow and therefore can easily take advantages of the cluster mode.

Now, they are nothing wrong to run like you did. I was just surprised.

Cheers,

Olivier

> On 5 Jul 2019, at 15:13, Thomas Hugle <email address hidden> wrote:
>
> Question #681396 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/681396
>
> Thomas Hugle posted a new comment:
> Yes, I am not using run_mode=1 on the cluster. Since I am running a parameter scan, it seemed simpler to me to manually submit my jobs because didn't manage to figure out how to change my run parameters when using the cluster mode. Is there a simple way to do so?
> My impression was that the cluster is well suited to calculate many events for one parameter set but not that well suited to calculate less events for many different parameter points. Of course this impression might be wrong.
>
> Yes, I assume the cluster gives me 4 cores of a single node (via manual
> submission).
>
> On my PC I did not set the run_mode option explicitly, but I assume it
> defaults to run_mode=2 since my computer used all of its cores.
>
> Best,
> Thomas
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Thanks for the additional information and the link.

Best,
Thomas