Reducing Negative Weights in MC@NLO by Improved Implementation of Born Spreading

Abstract

In this thesis, we present the efforts towards finding an improved implementation of the newly proposed Born Spreading method for reduction of negative S weights in the MC@NLO matching prescription. The improved implementations are tested on various LHC processes with MadGraph5_aMC@NLO (MG5_aMC) and compared to default MG5_aMC, 2×2×1 folding and default Born Spreading. The runtime in each step and fraction of negative S weights were recorded for each method to support a detailed discussion of their performance under considerations of both runtime and fraction of negative weights reduced. It was found that improved Born Spreading outperforms default Born Spreading convincingly, often achieving a greater reduction of negative S weights with shorter runtime. Moreover, improved Born Spreading is also comparable to 2 × 2 × 1 folding in terms of reduction of negative S weights, but with much shorter runtime.

Popular Abstract

In the past century, physics is amongst many disciplines that have seen tremendously rapid development. In particular, physicists have identified the fundamental constituents of matter and proposed a theory that describes the dominating interactions between them. This theory, known as the Standard Model is arguably physics’ best candidate for a theory of everything.

However, in addition to being powerful and ambitious, the Standard Model is also extremely complex. This leads to great difficulty in linking theory with experiments. In fact, most results from accelerator experiments are completely incomprehensible, in the sense that one cannot possibly ascertain what processes produced such results by the means of simple data analysis. To bridge this gap, computer simulations were developed. These simulations, known as event generators, have since become an integral tool for experimental particle physics.

In recent years, improving the efficiencies of event generators has been rising quickly on physicists’ priority lists. This is due to the pending upgrades to major accelerator experiments around the world, which leads to greater demand of computational resources. Unfortunately, unlike optimising an arbitrary computer program, developers of event generators will often find their hands tied by the laws of physics, since some obstacles are of physical origin. One of such obstacles is the presence of negative weights.

By negative weights, we refer to events that are associated with a negative contribution (weight) to the result. They are essential for obtaining the correct result, but lead to results with greater uncertainty. Since precision is an requirement as well, presence of negative weights means that more events must be generated, consuming more computational resources. Therefore, reduction of negative weights is critical for improving efficiencies of event generators.

Traditionally, the problem is tackled by folding, a method that trades time for accuracy. Typically, using this method represents at least two times more computing time compared to default. Since our current problem is exactly the lack of computational resources, relying on such a method might not be the best idea.

Fortunately, a good alternative was presented recently. This new method, known as Born Spreading, is much less demanding and gives results comparable to minimal folding in most cases. Most importantly, it’s full potential is yet to be explored.

My work is exactly to contribute to this exploration with the hope of finding an improved implementation of Born Spreading. Under a general consideration (runtime and effectiveness), it is not difficult to imagine that such an improved implementation can be considered comparable or even superior to folding. Therefore, if such an implementation is found, a discussion for including Born Spreading in the default program could be considered, representing a truly exciting opportunity to make a tiny contribution to a real problem in physics.

• author: Che, Yuxiao
• supervisor: Rikkert Frederix
• organization: Department of Physics; Particle and nuclear physics
• course: FYSK04 20241
• year: 2024
• type: M2 - Bachelor Degree
• subject: Physics and Astronomy
• keywords: Event Generators, Negative Weights, MC@NLO, MadGraph5_aMC@NLO
• language: English
• id: 9157443
• date added to LUP: 2024-06-11 09:06:06
• date last changed: 2024-06-11 09:06:06