SciPost Submission Page
Exploring phase space with Neural Importance Sampling
by Enrico Bothmann, Timo Janßen, Max Knobbe, Tobias Schmale, Steffen Schumann
- Published as SciPost Phys. 8, 069 (2020)
|As Contributors:||Timo Janßen · Steffen Schumann|
|Arxiv Link:||https://arxiv.org/abs/2001.05478v3 (pdf)|
|Date submitted:||2020-03-27 01:00|
|Submitted by:||Schumann, Steffen|
|Submitted to:||SciPost Physics|
|Approaches:||Theoretical, Computational, Phenomenological|
We present a novel approach for the integration of scattering cross sections and the generation of partonic event samples in high-energy physics. We propose an importance sampling technique capable of overcoming typical deficiencies of existing approaches by incorporating neural networks. The method guarantees full phase space coverage and the exact reproduction of the desired target distribution, in our case given by the squared transition matrix element. We study the performance of the algorithm for a few representative examples, including top-quark pair production and gluon scattering into three- and four-gluon final states.
Ontology / TopicsSee full Ontology or Topics database.
Published as SciPost Phys. 8, 069 (2020)
Author comments upon resubmission
We hope that the paper in its present form can be published in SciPost.
List of changes
* added reference to Byckling, Kajantie Nucl. Phys. B 9 (1969) 568
* updated references
* introduced variance V_N at first appearance after Eq. (3)
* updated discussion of figures of merit, i.e. V_N and e_UW
* We have inverted the ordering of the sampler criteria (i) and (ii) to better frame the coverage argument, slightly reworded, footnote added
* added reference to the physics examples being presented later on to the introduction
* clarified relation between loss function and figures of merit, thereby adapting the notation for improved clarity
* extended the discussion of the event weight distribution for the top-quark decay example, i.e. Fig. 2.
* reworded description of Eq. (11) a little, in particular adding a reference to Eq. (7)
* updated captions of results tables, making explicit that we show the *MC integral* E_N as
an *estimator* for the physical widths/cross sections
* updated Fig 1a to illustrate that increasing the number of data points in the training has a diminishing
return of investment in terms of asymptotic phase space coverage for the non-surjective NN; also
added some minor clarifications in the main text
* Updated wording in the conclusion in regards of more complex multi-channel and high-multiplicity integration problems
Submission & Refereeing History
You are currently on this page
Reports on this Submission
Report 1 by Tilman Plehn on 2020-3-27 (Invited Report)
- Cite as: Tilman Plehn, Report on arXiv:2001.05478v3, delivered 2020-03-27, doi: 10.21468/SciPost.Report.1596
Thank you for taking into account my comments. I disagree with the referencing aspect, but then I am also not the author of the paper...