Collinear test FAILED

Asked by Ulrich Haisch on 2017-09-12

Dear MadGraph5_aMC@NLO team

I am trying to generate a bbbar + missing energy signal sample at NLO in the 5FS.

The UFO I am using is here

https://www.dropbox.com/s/omlrsa98f8wp6sp/Pseudoscalar_2HDM.tgz?dl=0

The commands I am using are

import model Pseudoscalar_2HDM-bbMET_5FS

generate p p > j j xd xd~ [QCD]

output ppbbxx_5FS

launch

The output that I get reads as follows

INFO: Output of the failing test:
 Enter 0 to compute MC/MC(limit)
       1 to compute MC/ME(limit)
       2 to compute ME/ME(limit)
 Enter xi_i, y_ij to be used in coll/soft tests
  Enter -2 to generate them randomly
 Enter number of tests for soft and collinear limits
 A PDF is used, so alpha_s(MZ) is going to be modified
 Old value of alpha_s from param_card: 0.11799999999999999
  ****************************************

       NNPDFDriver version 1.0.3
   Grid: NNPDF23nlo_as_0119_qed_mem0.grid
  ****************************************
 New value of alpha_s from PDF nn23nlo: 0.11899999999999999
WARNING: the value of maxjetflavorspecified in the run_card ( 4) is inconsistent with the number of light flavours inthe model. Hence it will be set to: 5
 Give FKS configuration number ("0" loops over all)

 =================================================

 NEW FKS CONFIGURATION:
 FKS configuration number is 1
 FKS partons are: i= 7 j= 1
 with PDGs: i= 21 j= 2

 Enter graph number (iconfig), '0' loops over all graphs
Using random seed offsets: 1 , 4 , 0
  with seed 32
 Ranmar initialization seeds 10730 9410
tau_min 1 1 : 0.20000E+02 -- 0.64000E+03
tau_min 2 1 : 0.20000E+02 -- 0.64000E+03
tau_min 3 1 : 0.60000E+02 0.60000E+02 0.64000E+03
tau_min 4 1 : 0.60000E+02 0.60000E+02 0.64000E+03
tau_min 5 1 : 0.20000E+02 -- 0.64000E+03
tau_min 6 1 : 0.20000E+02 -- 0.64000E+03

 Soft tests done for (Born) config 1
 Failures: 0
     Soft test 1 PASSED. Fraction of failures: 0.00

 Collinear tests done for (Born) config 1
 Failures: 1
Collinear test 1 PASSED. Fraction of failures: 0.01

 =================================================

[...]

Soft tests done for (Born) config 1
 Failures: 0
     Soft test 4 PASSED. Fraction of failures: 0.00

 Collinear tests done for (Born) config 1
 Failures: 1
Collinear test 4 PASSED. Fraction of failures: 0.01

 =================================================

 NEW FKS CONFIGURATION:
 FKS configuration number is 5
 FKS partons are: i= 7 j= 1
 with PDGs: i= -2 j= 21

 Enter graph number (iconfig), '0' loops over all graphs

 Soft tests done for (Born) config 1
 Failures: 0
     Soft test 5 PASSED. Fraction of failures: 0.00

 Collinear tests done for (Born) config 1
 Failures: 30
Collinear test 5 FAILED. Fraction of failures: 0.30

 =================================================

 NEW FKS CONFIGURATION:
 FKS configuration number is 6
 FKS partons are: i= 6 j= 2
 with PDGs: i= 2 j= 21

 Enter graph number (iconfig), '0' loops over all graphs

 Soft tests done for (Born) config 1
 Failures: 0
     Soft test 6 PASSED. Fraction of failures: 0.00

 Collinear tests done for (Born) config 1
 Failures: 41
Collinear test 6 FAILED. Fraction of failures: 0.41
Error detected in "launch -f"
write debug file /Users/haisch/Documents/Codes/MG5_aMC_v2_6_0/ppbbxx_5FS/run_03_tag_1_debug.log
If you need help with this issue please contact us on https://answers.launchpad.net/mg5amcnlo
aMCatNLOError : Some tests failed, run cannot continue.
        Please check that widths of final state particles (e.g. top) have been set to 0 in the param_card.dat.

The process

p p > j j xd xd~

receives contribution from two spin-0 pseudoscalar particles, h3 and h4, that are a mixture of the normal
pseudoscalar A in 2HDMs and a extra singlet P. The mass eigenstates h3 and h4 both can decay to dark
matter, i.e. xd xd~

To understand what is going on I have also generated

generate p p > j j h3 [QCD]

generate p p > j j h4 [QCD]

and computed the cross section. In this case I do not get an error.

Furthermore,

generate p p > j j xd xd~ / h3 [QCD]

also works fine, while

generate p p > j j xd xd~ / h4 [QCD]

again fails the collinear tests

Like in the case

https://answers.launchpad.net/mg5amcnlo/+question/255905

the problems seems to be that in the case

generate p p > j j xd xd~

there are too many relevant scales in the problem (DM mass, h3 and h4
mass, etc.)

In fact by playing around with the input I managed to get rid of the error
concerning the collinear tests. Likewise, I am also able to get rid of the
error by increasing the value of

max_fail

in

SubProcesses/test_soft_col_limits.f

from 0.3 to 0.5

Now my questions:

1) Should I worry that for some parameter choices with largely separated scales close to 50%
of the points fail the collinear tests?

2) Besides changing the file test_soft_col_limits.f by hand is there another way to set the
parameter max_fail to a higher value? Or maybe even skip over the soft and collinear tests
altogether?

Apologies for the longish question and thanks in advance for an answer.

Best,

Uli

Question information

Language:
English Edit question
Status:
Solved
For:
MadGraph5_aMC@NLO Edit question
Assignee:
marco zaro Edit question
Solved by:
marco zaro
Solved:
2017-09-13
Last query:
2017-09-13
Last reply:
2017-09-13
Best marco zaro (marco-zaro) said : #1

Ciao Uli,
i managed to reproduce your problem, in particular there are failures in these two directories (you can check yourself with
grep FAILED P*/test_ME.log
inside SubProcesses): P0_uux_xdxdxbbx and P0_uxu_xdxdxbbx

Said this, these tests compute numerically the convergence of the real matrix-element to the form of the counterterm in the collinear configuration. There are some parameters in the code that regulate how close one gets to the limit in these chekcs, the tolerance of the tests and the maximum allowed fraction of failures. The latter is 0.3 (max_fail, declared around line 22 inside test_soft_col_limits.f or symmetry_fks_test_ME.f in older versions). The fraction of failures i have found (and you also found) is between 0.3 and 0.4, that is just above this threshold.
Having a look at the numerical behaviour of these tests i see nothing suspect, it may just be that the code is not getting close enough to the limit for the points which fail, thus i would just recommend to increase the max_fail to ~0.45 and have a try.

Let me know if it does not work or if you encounter any problem.

Cheers,

Marco

On 13 Sep 2017, at 10:29, Olivier Mattelaer <email address hidden> wrote:

> Question #657910 on MadGraph5_aMC@NLO changed:
> https://answers.launchpad.net/mg5amcnlo/+question/657910
>
> Assignee: None => marco zaro
>
> --
> You received this question notification because you are assigned to this
> question.

Ulrich Haisch (zoppo) said : #2
Ulrich Haisch (zoppo) said : #3

Thanks marco zaro, that solved my question.