Hi,

To merge this to 2.6.3, we have to keep such change minimal for the moment.

So I have had the bare minimum to the current status.
This is
a) detect fully identical expression --same string-- and merge them
b) remove couplings exactly equals to zero (no point to keep them obviously)

This is pretty minimal and quite save.
for loop_sm this does not do anything in the unitary gauge. (but make few simplification in the feynman gauge)
for the loop_qcd_qed_sm, they are 148 coupling equals to zero, so this makes some cool cleaning.

Now we can also not push this in 2.6.3 to be save. They are nothing crucial in such change and can easily wait months to be include.

Cheers,

Olivier


=== modified file 'models/import_ufo.py'
--- models/import_ufo.py 2018-06-01 08:46:21 +0000
+++ models/import_ufo.py 2018-06-13 15:01:47 +0000
@@ -14,7 +14,7 @@
 ################################################################################
 """ How to import a UFO model to the MG5 format """

-
+import collections
 import fractions
 import logging
 import os
@@ -530,8 +530,10 @@
                 self.add_CTinteraction(interaction_info, color_info)


-        for interaction in self.interactions:
+        for interaction in list(self.interactions):
             self.optimise_interaction(interaction)
+            if not interaction['couplings']:
+                self.interactions.remove(interaction)


         self.model.set('conserved_charge', self.conservecharge)
@@ -584,6 +586,32 @@

     def optimise_interaction(self, interaction):

+
+        #  Check if two couplings have exactly the same definition.
+        #  If so replace one by the other
+        if not hasattr(self, 'iden_couplings'):
+            coups = collections.defaultdict(list)
+            coups['0'].append('ZERO')
+            for coupling in self.ufomodel.all_couplings:
+                #if isinstance(coupling.value, str):
+                coups[str(coupling.value)].append( coupling.name)
+
+            self.iden_couplings = {}
+            for idens in [c for c in coups.values() if len(c)>1]:
+                for i in range(1, len(idens)):
+                    self.iden_couplings[idens[i]] = idens[0]
+
+        # apply the replacement by identical expression
+        for key, coup in list(interaction['couplings'].items()):
+            if coup in self.iden_couplings:
+                interaction['couplings'][key] = self.iden_couplings[coup]
+            if interaction['couplings'][key] == 'ZERO':
+                del interaction['couplings'][key]
+
+
+
+
+
         # we want to check if the same coupling is used for two lorentz strucutre
         # for the same color structure.
         to_lor = {}
@@ -1807,9 +1835,9 @@
         self.remove_couplings(self.del_coup)

         # modify interaction to avoid to have identical coupling with different lorentz
-        for interaction in self.get('interactions'):
+        for interaction in list(self.get('interactions')):
             self.optimise_interaction(interaction)
-
+
         # deal with parameters
         parameters = self.detect_special_parameters()
         self.fix_parameter_values(*parameters, simplify=rm_parameter,


On 13 Jun 2018, at 14:36, Valentin Hirschi <<email address hidden><mailto:<email address hidden>>> wrote:

Hi Olivier,

Sure, the numerical approach is one possibility and we discussed it a bit already in the previous emails of this thread.
We were then worried that we could hit false positive, and you are saying that we could reduce this risk to an insignificant number by doing it repeatedly with arbitrary assignment of the external parameters.
I suppose this could work indeed (I think it shouldn't be *too* hard to implement, but Olivier should confirm).

Of course, when using a restriction file when loading the model, we should only do this using the specified input parameters, as they may exhibit extra simplifications that are ok to use, at least for K=1.
But then we could also combine this some other ratios detected with your method Rik, as those would be safe to use anyway.

I don't think this is too crucial, so we should only go for simple solutions if there are some, or none if too complicated.

Cheers


On Wed, Jun 13, 2018 at 2:26 PM, Rikkert Frederix <<email address hidden><mailto:<email address hidden>>> wrote:
Guys,

Wouldn't it make more sense to check this numerically? I.e., assign some random numbers to all the terms, compute the ratio between all the couplings and repeat this a couple of times. If the ratio between couplings doesn't change up to machine precision for several choices of the random numbers consider them equal with the (numerical) proportionality factor given by the ratio?

Cheers,
Rik

--
https://code.launchpad.net/~maddevelopers/mg5amcnlo/2.6.3_optimizenlomodel/+merge/346864
You are reviewing the proposed merge of lp:~maddevelopers/mg5amcnlo/2.6.3_optimizenlomodel into lp:~mg5core2/mg5amcnlo/2.6.3.<http://2.6.0.3/>



--
Valentin