SciPost Submission Page
Reducing negative weights in Monte Carlo event generation with Sherpa
by Katharina Danziger, Stefan Höche, Frank Siegert
Submission summary
| Authors (as registered SciPost users): | Stefan Höche · Frank Siegert |
| Submission information | |
|---|---|
| Preprint Link: | https://arxiv.org/abs/2110.15211v1 (pdf) |
| Date submitted: | 2021-11-05 11:31 |
| Submitted by: | Siegert, Frank |
| Submitted to: | SciPost Physics |
| Ontological classification | |
|---|---|
| Academic field: | Physics |
| Specialties: |
|
| Approaches: | Computational, Phenomenological |
Abstract
An increase in theoretical precision of Monte Carlo event generators is typically accompanied by an increased need for computational resources. One major obstacle are negative weighted events, which appear in Monte Carlo simulations with higher perturbative accuracy. While they can be handled somewhat easily in fixed-order calculations, they are a major concern for particle level event simulations. In this article, the origin of negative weights in the S-MC@NLO method is reviewed and mechanisms to reduce the negative weight fraction in simulations with the Sherpa event generator are presented, with a focus on V+jets and tt+jets simulations.
Current status:
Reports on this Submission
Strengths
1. Timely
2. Very clear presentation of an investigation into very relevant method
Weaknesses
1. While the approach reduces the number of negative weight, it fails to reproduce the results of the unmodified approach
Report
This report is very timely in presenting a study of three investigations into
reducing the occurrence of events with negative weight in the high-precision
fixed order and parton shower mergings possible in Sherpa. The presentation
is extremely clear and concise, and the paper is very well written.
The paper presents three unrelated approaches, and compares both the degree
to which the modifications reproduce the correct result, and the impact on
the frequency of negative weight events. One of the conclusions of the paper
is that (page 13) "In the high pT region, \epsilon can be lowered only by
defining the local K-factor using the lowest multiplicity process at NLO". I
have a question to the approach which would need to be investigated before
publication:
While it is correct that only the approach of the local K-factor has a
significant impact on the frequency of negative weight events, this approach
is also singled out by significant discrepancies in the distributions
obtained with both the unmodified approach and the other two approaches to
reducing the impact of negative weight events. For example, in figure 5a, the
very large pT region of interest, the method of the local K-factor
systematically undershoots the other approaches in all the last 16 bins, by
more than 3 standard deviations in several bins, and by more than two
standard deviations in 12 of the last bins. This surely is not just a
statistical fluctuation, but rather a systematic effect which would need
investigation. A similar systematic effect for the same method is seen in
figure 6. Systematic deviations are seen in the low HT-region in figure
3. However, these deviations are not discussed or compared with other
intrinsic uncertainty to the calculation. How do these differences compare to
even simple scale variations? There seems little point avoiding negative
weights at the cost of accuracy, since the negative weights arise only when
requiring higher accuracy on the predictions from the shower.
I am looking forward to seeing the result of this investigation.
Requested changes
1. Investigate whether the deviations discussed in the report are indeed systematic
2. Investigate the size of the deviations compared to the intrinsic uncertainty of the calculations