loop_qcd_qed_sm model- associated production of higgs

Hi,

When running the loop computation, we always check if we can trust the value returned.
If the precision is not good enough then we use another method to compuate the loop and do the check again.

If all the reduction technnique available are failing then we move to quadruple precision which increase the precision by huge factor but which is 100 times slower and obviously we still check the precision.

If some events are still not precise enough, then that phase-space point is simply set to zero.
If too many events are set to zero then you see the warning that you faced.

Solution on that is likely not existing,
However, you can check that you have ninja and collier reduction method available.
You can also play with the parameter of the madloop_card to check
1. that all reduction method are indeed activated (irigi is often set to OFF)
2. you can play with the level of requested precision (or even bypass the precision check). Of course, this is your responsability...

Cheers,

Olivier

> On 21 Mar 2021, at 12:11, Antara Sarkar <email address hidden> wrote:
>
> New question #696165 on MadGraph5_aMC@NLO:
> https://answers.launchpad.net/mg5amcnlo/+question/696165
>
> Dear MG5 experts,
>
> Within model loop_qcd_qed_sm, I am trying to higgs and photon from initial gluons with the syntax:
> generate g g > h a [noborn=QED]
> I get all diagrams as expected. However generating events gives an error as follows:
>
> Generating 10000 events with run name run_01
> survey run_01
> INFO: compile directory
> initMadLoop -r -f
> Initializing MadLoop loop-induced matrix elements (this can take some time)...
> MadLoop initialization finished.
> compile Source Directory
> Using random number seed offset = 21
> INFO: Running Survey
> Creating Jobs
> Working on SubProcesses
> INFO: P0_gg_ha
> INFO: Idle: 1, Running: 1, Completed: 0 [ current time: 16h28 ]
> INFO: Idle: 0, Running: 1, Completed: 1 [ 2m 47s ]
> INFO: Idle: 1, Running: 0, Completed: 2 [ 3m 18s ]
> CRITICAL: Channel 1 has encountered a fraction of 0.301
> of numerically unstable loop matrix element computations
> (which could not be rescued using quadruple precision).
> The results can not be trusted.
> WARNING: Channel 1 has encountered a fraction of 0.301
> of numerically unstable loop matrix element computations
> (which could not be rescued using quadruple precision).
> The results can not be trusted.
> INFO: Idle: 0, Running: 0, Completed: 3 [ 3m 18s ]
> Channel 1 has encountered a fraction of 0.301
> of numerically unstable loop matrix element computations
> (which could not be rescued using quadruple precision).
> The results can not be trusted.
> Cluster Error detected. Do you want to clean the queue? ("c"=continue the run anyway) [y, n, c][60s to answer]
>> n
> Command "generate_events run_01" interrupted with error:
> Exception : Channel 1 has encountered a fraction of 0.301
> of numerically unstable loop matrix element computations
> (which could not be rescued using quadruple precision).
> The results can not be trusted.
> Please report this bug on https://bugs.launchpad.net/mg5amcnlo
> More information is found in '/home/antara/my_computation/MG5_aMC_v2_7_2/PROC_loop_qcd_qed_sm_7/run_01_tag_1_debug.log'.
> Please attach this file to your report.
> INFO:
> quit
> INFO:
> MG5_aMC>
>
>
> Could you tell why this is occuring and how to solve it? I do not find such an error if the initial partons are quarks. The process is simulated but with initial gluons I encounter this error.
>
> Best regards,
> Antara
>
>
> --
> You received this question notification because you are an answer
> contact for MadGraph5_aMC@NLO.

Thanks Olivier for the quick response.

This is the MadLoop param card :

! This file is for the user to set the different parameters of MadLoop.
! The name of the variable to define must start with the '#' sign and then
! the value should be put immediately on the next line.

!
#MLReductionLib
!6|7|1|3
! Default :: 6|7|1|3
! The tensor integral reduction library.The current choices are:
! 1 | CutTools
! 2 | PJFry++
! 3 | IREGI
! 4 | Golem95
! 5 | Samurai
! 6 | Ninja
! 7 | COLLIER
! One can use the combinations to reduce integral,e.g.
! 1|2|3 means first use CutTools, if it is not stable, use PJFry++,
! if it is still unstable, use IREGI. If it failed, use QP of CutTools.
! Notice that any reduction tool not avaialble on the system will be automatically
! skipped.

! When using quadruple precision with Ninja or CutTools, the reduction will
! always be done in quadruple precision, but the parameters below allow you to
! chose if you want to also recompute the *integrand* in quadruple precision.
! Doing so is slow but might improve the accuracy in some situation.
#UseQPIntegrandForCutTools
!.TRUE.
! Default :: .TRUE.
#UseQPIntegrandForNinja
!.TRUE.
! Default :: .TRUE.
!

! =================================================================================
! The parameters below set the parameters for IREGI
! =================================================================================

#IREGIMODE
!2
! Default :: 2
! IREGIMODE=0, IBP reduction
! IREGIMODE=1, PaVe reduction
! IREGIMODE=2, PaVe reduction with stablility improved by IBP reduction

#IREGIRECY
!.TRUE.
! Default :: .TRUE.
! Use RECYCLING OR NOT IN IREGI
!

! =================================================================================
! The parameters below set the stability checks of MadLoop at run time
! =================================================================================

! Decide in which mode to run MadLoop
!
! imode:| description
! 1 | Double precision, loops reduced with propagator in original order
! 2 | Double precision, loops reduced with propagator with reversed order
! 4 | Quadruple precision, loops reduced with propagator in original order
! 5 | Quadruple precision, loops reduced with propagator with reversed order
! -1 | Exhaustive automated numerical stability checks. See below for details.
!
! Due to the architecture of the program, you are better off
! rerunning the full PS point in quadruple precision than just a single loop
! because the two things would almost take the same time. So '-1' is always
! very recommended.
#CTModeRun
!-1
! Default :: -1
! In the negative mode -1, MadLoop first evaluates each PS points in modes 1 and 2,
! yielding results Res1 and Res2, and then check if:
! (Res1-Res2)/(2*(Res1+Res2)< MLStabThres
! If it is not the case, MadLoop evaluates again the PS point in modes 4 and 5,
! yielding results Res4 and Res5, and then check if:
! (Res4-Res5)/(2*(Res4+Res5)< MLStabThres
! If it is the case then the unstable phase-space point could be cured. If it is
! not the case, MadLoop outputs a warning.
! Notice that MLStabThres is used only when CTModeRun is negative.
#MLStabThres
!1.0d-3
! Default :: 1.0d-3
! You can add other evaluation method to check for the stability in DP and QP.
! Below you can chose if you want to use zero, one or two rotations of the PS point
! in QP.
#NRotations_DP
!0
! Default :: 0
#NRotations_QP
!0
! Default :: 0
#NRotations_QP
!0
! Default :: 0

! By default, MadLoop is allowed to slightly deform the Phase-Space point in input
! so to insure perfect onshellness of the external particles and perfect energy-momentum
! conservation. The deformation is minimal and such that it leaves the input PS point
! unchanged if it already satisfies the physical condiditions mentioned above.
! This integer values select what is the method to be employed preferably to restore this
! precision. It can take the following values:
!
! -1 :: No method is used for double precision computations, and method 2 will be used
! preferentially when quadruple precision (for which this precision improvement
! is mandatory, otherwise quadruple precision is pointless)
! 1 :: This methods imitates what is done in PSMC, namely
! a) Set the space-like momentum of the last external particle to be the
! opposite of the sum of the others (with a minus sign for the initial states).
! b) Rescale all final state space-like momenta by a fixed value x computed such
! that energy is conserved when particles are put exactly onshell. This value
! is determined numericaly via Ralph-Newton's method.
! c) Set all energies to have particles exactly onshell.
! 2 :: This method applies a shift to the energy and the x and y components of the first
! initial state momentum in order to restore exact energy momentum conservation after
! particles have been put exactly onshell via a shift of the z component of their
! momenta.
#ImprovePSPoint
!2
! Default :: 2

! =================================================================================
! The parameters below set two CutTools internal parameters accessible to the user.
! =================================================================================

! Choose here what library to chose for CutTools/TIR to compute the scalar loops of the
! master integral basis. The choices are as follows:
! (Does not apply for Golem95, where OneLOop is always used)
! 2 | OneLOop
! 3 | QCDLoop
#CTLoopLibrary
!2
! Default :: 2
! Choose here the stability threshold used within CutTools to decide when to go to
! higher precision.
#CTStabThres
!1.0d-2
! Default :: 1.0d-2

! =================================================================================
! The parameters below set the general behavior of MadLoop for the initialization
! =================================================================================

! Decide in which mode to run when performing MadLoop's initialization of
! the helicity (and possibly loop) filter. The possible modes are:
!
! Decide in which mode to run MadLoop
!
! imode:| description
! 1 | Double precision, loops reduced with propagator in original order
! 2 | Double precision, loops reduced with propagator with reversed order
! 4 | Quadruple precision, loops reduced with propagator in original order
! 5 | Quadruple precision, loops reduced with propagator with reversed order
!
#CTModeInit
!1
! Default :: 1

! CheckCycle sets on how many PS points trials the initialization filters must be
! obtained. As long as MadLoop does not find that many consecutive PS points for
! which the filters are the same, it will start over but only a maximum of
! MaxAttempts times.
#CheckCycle
!3
! Default :: 3
#MaxAttempts
!10
! Default :: 10

! Setting the threshold for deciding wether a numerical contribution is analytically
! zero or not.
#ZeroThres
!1.0d-9
! Default :: 1.0d-9

! Setting the on-shell threshold for deciding whether the invariant variables
! of external momenta are on-shell or not. It will only be used in constructing
! s-matrix in Golem95.
#OSThres
!1.0d-8
! Default :: 1.0d-8

! The setting below is recommended to be on as it allows to systematically used the
! first PS point thrown at ML5 to be used for making sure that the helicity filter
! read from HelFilter.dat is consistent as it might be no longer up to date with
! certain changes of the paramaters by the user.
#DoubleCheckHelicityFilter
!.TRUE.
! Default :: .TRUE.

! This decides whether to write out the helicity and loop filters to the files
! HelFilters.dat and LoopFilters.dat to save them for future runs. It usually
! preferable but sometimes not desired because of the need of threadlocks in the
! context of mpi parallelization. So it can be turned off here in such cases.
#WriteOutFilters
!.TRUE.
! Default :: .TRUE.

! Some loop contributions may be zero for some helicities which are however
! contributing. In order to save their computing time, you can chose here to try
! to filter them out. The gain is typically minimal, so it is turned off by default.
#UseLoopFilter
!.FALSE.
! Default :: .FALSE.

! The integer below set at which level the user wants to filter helicity configuration.
! Notice that this does not entail any approximation. It only offers the possibility of
! performing exact simplifications based on numerical checks. HelicityFilterLevel =
! 0 : No filtering at all. Not HelFilter.dat file will be written out and *all* helicity
! configurations will be computed.
! 1 : Analytically zero helicity configurations will be recognized as such by numerical
! comparisons (using the 'ZeroThres' param) and consistently skipped in further
! computations.
! 2 : Filters both helicity configuration which are analytically zero *and* those
! consistently identical (typically because of CP symmetry).
! (Will only effectively do it if process was generated in 'optimized_mode')
#HelicityFilterLevel
!2
! Default :: 2

! This decides whether consecutive consistency for the loop filtering setup is also
! required.
#LoopInitStartOver
!.FALSE.
! Default :: .FALSE.

! This decides wether consecutive consistency for the helicity filtering setup is also
! required. Better to set it to false as it can cause problems for unstable processes.
#HelInitStartOver
!.FALSE.
! Default :: .FALSE.

! =================================================================================
! The parameters below set the main parameters for COLLIER
! To edit more specific technical COLLIER parameters, modify directly the content
! of the subroutine 'INITCOLLIER' in the file 'MadLoopCommons.f'
! =================================================================================

! Decide if COLLIER must be computed multiple times to evaluate the UV pole residues
! (Withing a Monte-Carlo performed in MG5aMC, this is automatically disabled internally)
#COLLIERComputeUVpoles
!.TRUE.
! Default :: .TRUE.

! Decide if COLLIER must be computed multiple times to evaluate the IR pole residues
! (Withing a Monte-Carlo performed in MG5aMC, this is automatically disabled internally)
#COLLIERComputeIRpoles
!.TRUE.
! Default :: .TRUE.

! Decide if COLLIER must be computed multiple times to evaluate the IR pole residues
#COLLIERRequiredAccuracy
!1.0d-8
! Default :: 1.0d-8
! A value of -1.0d0 means that it will be automatically set from MLStabThres.
! The default value of 1.0d-8 corresponds to the value for which COLLIER's authors
! have optimized the library.

! Decide whether to use COLLIER's internal stability test or the loop-direction
! switch test instead.
#COLLIERUseInternalStabilityTest
!.TRUE.
! Default :: .TRUE.
! COLLIER's internal stability test is at no extra cost but not as reliable
! as the loop-direction switch test, which however doubles the reduction time.
  ! This parameter is only relevant when running MadLoop with CTModeRun=-1.
! If you find a large number of unstable points with COLLIER for complicated
! processes, set this parameter to .FALSE. to make sure the PS points flagged
! as unstable with COLLIER really are so.

! Set up to which N-loop to use the COLLIER global caching system.
#COLLIERGlobalCache
!-1
! Default :: -1
! -1 : Enable the global cache for all loops
! 0 : Disable the global cache alltogether
! N : Enable the global cache but only for up to N-loops

! Use the global cache when evaluating the poles as well (more memory consuming)
! During a Monte-Carlo it is typically not useful anyway, because the pole
! computation is automatically disabled for COLLIER, irrespectively of the value
! of the parameters COLLIERCompute<UV/IR>poles specified above.
#COLLIERUseCacheForPoles
!.FALSE.
! Default :: .FALSE.

! Choose which branch(es) of COLLIER have to be used
#COLLIERMode
!1
! Default :: 1
! COLLIERMode=1 : COLI branch
! COLLIERMode=2 : DD branch
! COLLIERMode=3 : Both DD and COLI branch compared

! Decide if COLLIER can output its information in a log directory.
#COLLIERCanOutput
!.FALSE.
! Default :: .FALSE.

/* End of param file */

I activated collier, ninja and irigi reduction methods. Is it alright?
Still I find the same error coming up.
Also, there are many options that deal with precision. Could you tell which parameter should I change the level of precision or even bypass precision.

Best regards,
Antara

I have never played with those parameters myself, so I'm as good as you as reading the card and try stuff.

If you do not care if your loop is 1e-1 or 1e-5 then you can probably set
CTModeRun
to 1 (or 2)
if you want to force quadruple precision for all point to be a bit better then, you can use 3 or 4.

If you just want to relax the precision on the matrix element, then you can set the precision of
MLStabThres to the value that you want

Do not forget to remove the "!" before the value when you edit the line otherwise your change will not be consider.

Cheers,

Olivier

Okay.. I shall check with that for the gluon initiated process and will let you know if it works. However, I was also trying to produce higgs and photon from quarks at loop level. The production process is simulated fine. But I wish to decay this higgs to b b~ making changes in the madspin card. This error arises:

INFO: detected model: loop_qcd_qed_sm. Loading...
Set group_subprocesses to Auto
Note that you need to regenerate all processes
set collier to /home/antara/my_computation/MG5_aMC_v2_7_2/HEPTools/lib
set lhapdf to /home/antara/my_computation/MG5_aMC_v2_7_2/HEPTools/lhapdf6/bin/lhapdf-config
set ninja to /home/antara/my_computation/MG5_aMC_v2_7_2/HEPTools/lib
set spinmode none # Use one of the madspin special mode
set max_weight_ps_point 400 # number of PS to estimate the maximum for each event
decay t > w+ b, w+ > all all
decay t~ > w- b~, w- > all all
decay w+ > all all
decay w- > all all
decay z > all all
decay h > b b~
launch
INFO: Will use seed 331013800
INFO: generate 10000 decay event for particle h
Command "generate_events run_01" interrupted with error:
NoDiagramException : No amplitudes generated from process Process: h > b b~. Please enter a valid process
quit
INFO: storing files of previous run
gzipping output file: unweighted_events.lhe
INFO: Done
INFO:
more information in /home/antara/my_computation/MG5_aMC_v2_7_2/PROC_loop_qcd_qed_sm_9/index.html
MG5_aMC>

Please look into it and tell how to solve it.

Best regards,
Antara

Thanks Olivier Mattelaer, that solved my question.