Surface-based geometry modelling in VecGeom


For describing detector geometry, LHC experiments have used the constructive solid geometry (CSG) approach successfully during the past decade to undertake detailed simulation of 10-20 billion proton-proton collisions each year. These rely on detector models with deep volume hierarchies composed of around 10 thousand volumes, which describe the millions of physical elements. The VecGeom project aims at developing a high-performance library for geometry modelling, describing geometry in terms of 3D solid primitives, such as boxes, tubes, or cones.

Using the existing CSG approach to undertake simulations on GPUs has resulted in poor performance, due to thread divergence and poor work balancing. This student project will be part of the new approach for geometry navigation in particle transport, which creates a ‘flatter’ surface-focused description (a “bounded-surface” model), and brings existing geometry descriptions onto it while looking to better leverage the massive potential of GPUs. This model will be based on a small number of simpler data structures, surface types and constructs, in order to ensure that it is performant on current GPUs and other future accelerators. It must provide the same navigation functionality as the existing geometry modeller.

Task ideas

The student will be expected to work in close collaboration with the team developing this new geometry modeller for use on GPUs, writing code that uses simple data structures and is portable between CPU and GPUs. The concrete student project should consider planning based on the following tasks:


Desirable Skills

Expected results

Evaluation Task

Interested students please contact Andrei for a warm-up and evaluation task.


Additional Information

Corresponding Project

Participating Organizations