Karlsruhe 2024 – wissenschaftliches Programm

Bereiche | Tage | Auswahl | Suche | Aktualisierungen | Downloads | Hilfe

T: Fachverband Teilchenphysik

T 67: Data, AI, Computing 5 (normalising flows)

T 67.4: Vortrag

Mittwoch, 6. März 2024, 16:45–17:00, Geb. 30.33: MTI

Using neural networks to calculate bounce actions — •Fabio Campello¹, Georg Weiglein², and Thomas Biekötter³ — ¹UHH, Hamburg, Deutschland — ²DESY, Hamburg, Deutschland — ³KIT, Karlsruhe, Deutschland

Computing the decay rate of a meta-stable state is a well-known problem with relevance in various areas of physics. The decay rate is dominated by an exponential factor B, called the bounce action. Determining the bounce action for a given potential and meta-stable vacuum involves solving a set of partial differential equations. Numerically solving these equations can be challenging, especially in instances where the meta-stable vacuum is nearly degenerate to the deeper vacuum, referred to as the thin-wall limit. There are several dedicated solvers available for this problem, however finding bounce actions in potentials of many variables still remains a challenge.

It is also established that neural networks can be used to solve any differential equation with fixed boundary conditions, as neural networks are general function approximators. We use a neural network to solve the partial differential equation for finding the tunneling path. Using a custom tensorflow operation for the loss function enables us to make use of the full capabilities of modern GPUs.

Subsequently, we apply this approach to analyze vacuum stability in both the Minimal Supersymmetric extension of the Standard Model (MSSM) and Next-to-MSSM (NMSSM), where we successfully determine bounce actions for the tree-level potential including all Higgs fields and 3rd generation sfermions, for a total of 22 scalar fields.

Keywords: Neural networks; Vacuum stability; Bounce action; Supersymmetry