# reducing nlo run time

Hello,

I am trying to run an NLO process. This takes an unsually large amount of time. This is probably because of the large number of diagrams. Is there any way to reduce the computation time? The generation is repeated for 5 iterations. Is there any way to prevent that?

Thanks in advance.

Regards,

Sayan.

## Question information

- Language:
- English Edit question

- Status:
- Solved

- Assignee:
- No assignee Edit question

- Solved by:
- Sayan Dasgupta

- Solved:
- 2020-12-03

- Last query:
- 2020-12-03

- Last reply:
- 2020-12-03

## This question was reopened

- 2020-12-03 by Sayan Dasgupta

The answer is likely no.

Not sure to understand what you mean by the five iteration but they are certainly a reason for that.

Olivier

Sayan Dasgupta (sayandg) said : | #2 |

Ok thanks Olivier.

Sayan Dasgupta (sayandg) said : | #3 |

Dear Olivier,

Sorry for bugging with this question again. I am trying to generate just 10000 events for the process

g g > e+ e- e+ e- $$ e+ e- / a [QCD]

having 56 diagrams. I am not parton showering. I am running the job in a computer having 40 cores and still it is not ending even after three days. Is this normal or there is some problem occuring inside?

But that's not NLO is it?

This should be LO (loop-induced but LO).

If this is indeed true, then

1) loop-induced processes are indeed very slow to integrate because

- this is a loop computation ( so ~ 100 times slower than a tree level computation with two additional jet-- so here a 2>6--). Additionally, some phase-space point can be unstable and need quadruple precision (another factor 100 times slower)

- phase-space integration is more tricky --since we lack some information about loop impact like internal threshold of the loop/...

2) your process should include contribution from

g g > Z Z

which has some integrable singularities. i.e. it is impossible to generate events for tthat process without cut (even if the cross-section is correctly computed and finite).

3) For this mode (loop induced), you have plenty of freedom to change many parameter to test if a setup is more optimal. All those parameters are explained in the loop-induced paper.

If you are really looking for NLO process.

Then you should not force gluon on initial state since this is not infra-red save.

(which means that the computation is not finite and therefore will have issue to converge)

Cheers,

Olivier

> On 3 Dec 2020, at 10:05, Sayan Dasgupta <email address hidden> wrote:

>

> Question #694252 on MadGraph5_aMC@NLO changed:

> https:/

>

> Status: Solved => Open

>

> Sayan Dasgupta is still having a problem:

> Dear Olivier,

>

> Sorry for bugging with this question again. I am trying to generate just

> 10000 events for the process

>

> g g > e+ e- e+ e- $$ e+ e- / a [QCD]

>

> having 56 diagrams. I am not parton showering. I am running the job in a

> computer having 40 cores and still it is not ending even after three

> days. Is this normal or there is some problem occuring inside?

>

> --

> You received this question notification because you are an answer

> contact for MadGraph5_aMC@NLO.

Sayan Dasgupta (sayandg) said : | #5 |

Thanks for your reply Olivier.

I am trying to generate the processes that should have three types of diagrams:

1. Tree level with light quarks in the initial state going to four leptons via two Z bosons. This one shouldn't have any gluon.

2. Loop level with gluons in the initial state going to four leptons via a light quark box and two Z bosons.

3. Loop level with gluons in the initial state going to four leptons via a Higgs and two Z bosons.

These are displayed in the paper 1803.09751. But I can't figure out the syntax to get these exact diagrams. I shall be grateful if you could help me out.

Regards,

Sayan.

Hi,

Separate 2 and 3 is not adivsed since those will interfere.

This is especially important if you are in the regime where both Z are onshell

The safest syntax for your loop are

generate g g > e+ e- e+ e- [QCD]

(and that's include all the loop diagram that you are interested in)

any other syntax needs to be justified since you remove physical diagram and therefore approximate the computation.

For example separating your diagram 2 and 3 is likely a bad idea since they can strongly interfere (apriori) on the phase-space where two Z are onshell.

Cheers,

Olivier

> On 3 Dec 2020, at 12:35, Sayan Dasgupta <email address hidden> wrote:

>

> Question #694252 on MadGraph5_aMC@NLO changed:

> https:/

>

> Status: Answered => Open

>

> Sayan Dasgupta is still having a problem:

> Thanks for your reply Olivier.

>

> I am trying to generate the processes that should have three types of

> diagrams:

>

> 1. Tree level with light quarks in the initial state going to four

> leptons via two Z bosons. This one shouldn't have any gluon.

>

> 2. Loop level with gluons in the initial state going to four leptons via

> a light quark box and two Z bosons.

>

> 3. Loop level with gluons in the initial state going to four leptons via

> a Higgs and two Z bosons.

>

> These are displayed in the paper 1803.09751. But I can't figure out the

> syntax to get these exact diagrams. I shall be grateful if you could

> help me out.

>

> Regards,

> Sayan.

>

> --

> You received this question notification because you are an answer

> contact for MadGraph5_aMC@NLO.

Sayan Dasgupta (sayandg) said : | #7 |

Hello,

I can merge 2 and 3 using the syntax you mentioned. But how can I add those to 1. Mg5 doesn't allow the addition of LO and NLO. Also if I don't add 1, the process will only remain loop induced and won't properly converge as you mentioned.

Regards,

Sayan.

Those are NOT NLO.

This is wrong to say that.

You can either argue that this is a LO process or a sub-piece of the NNLO

computation of p p > e+ e+ e- e-.

You can not claim that this those are NLO

> But how can I add those to 1.

We do not allow to generate events simultaneously for tree-level computation and loop-induced computation. You have to separate them and generate them separatly. (and merge the resulting events)

The reason for that is that the phase-space integration mechanism used for those two computations are too different.

> the process will only remain loop induced and won't properly converge as you mentioned.

1) Adding your first contribution will not impact the convergence (how could this be possible? they do not have the same initial/final state even in some IR limit)

2) The loop is itself finite, so you do not need to cancel the divrgent part via a counter-term that will compensate another divergencies from the Real Computation. (This is what happens in a NLO computation but you are not in a NLO setup)

3) The divergencies that you observe is related to the integration over the phase-space.

The integral is in this case integrable but you have a singularity if both Z are colinear with the beam.

One example of such integrable singularity is here

https:/

So we can compute the cross-section but not generate events (due to the singularity)

The solution is to put a cut (as explained in the loop-induced paper).

The same issue happens for tree-level computation, typically the issue is worse since the integral is in top of that infinite like for

e+ e- > e+ e-

The solution is the same, you have to apply cuts to avoid the divergency in order to generate events (and in this case to have a finite cross-sections)

Cheers,

Olivier

> On 3 Dec 2020, at 13:11, Sayan Dasgupta <email address hidden> wrote:

>

> Question #694252 on MadGraph5_aMC@NLO changed:

> https:/

>

> Status: Answered => Open

>

> Sayan Dasgupta is still having a problem:

> Hello,

>

> I can merge 2 and 3 using the syntax you mentioned. But how can I add

> those to 1. Mg5 doesn't allow the addition of LO and NLO. Also if I

> don't add 1, the process will only remain loop induced and won't

> properly converge as you mentioned.

>

> Regards,

> Sayan.

>

> --

> You received this question notification because you are an answer

> contact for MadGraph5_aMC@NLO.

Sayan Dasgupta (sayandg) said : | #9 |

Thanks Olivier, I probably got the point. I will try running again after providing some cuts.

Regards,

Sayan