Towards replacing detector simulation with heterogeneous GNNs in flavour physics analyses

TL;DR

This work tackles the looming computing bottleneck in LHCb detector simulations by introducing Rex, a fast detector-response emulator built on heterogeneous graph neural networks. Rex learns general detector-patterns across arbitrary multi-body decays, enabling interpolation to unseen channels and offering a drop-in, on-demand alternative to full simulation. The architectures embed physics directly into the graph structure, achieving accurate momentum smearing, PID, and vertexing outputs with substantial speed-ups (~10^5×) and modular deployment. The approach is demonstrated to work across diverse decay topologies and is poised for adaptation to other experiments facing similar simulation pressures, with future work focusing on acceptance modelling, broader variable coverage, and stabilization enhancements.

Abstract

Driven by the increasing volume of recorded data, the demand for simulation from experiments based at the Large Hadron Collider will rise sharply in the coming years. Addressing this demand solely with existing computationally intensive workflows is not feasible. This paper introduces a new fast simulation tool designed to address this demand at the LHCb experiment. This tool emulates the detector response to arbitrary multibody decay topologies at LHCb. Rather than memorising specific decay channels, the model learns generalisable patterns within the response, allowing it to interpolate to channels not present in the training data. Novel heterogeneous graph neural network architectures are employed that are designed to embed the physical characteristics of the task directly into the network structure. We demonstrate the performance of the tool across a range of decay topologies, showing the networks can correctly model the relationships between complex variables. The architectures and methods presented are generic and could readily be adapted to emulate workflows at other simulation-intensive particle physics experiments.

Figures (22)

• Figure 1: Overview of the stages of the full simulation and those of the approach of Rex.
• Figure 2: Overview of the structure of Rex.
• Figure 3: Example of the edge connection structure of a graph employed during a communication layer of the smearing and PID networks.
• Figure 4: Examples of pre-processing for a variable describing the level of momenta smearing, the ratio of the reconstructed momenta $P_{reco}$ to the true momenta $P_{true}$. The physical distribution of the variable (top), the distribution using shared pre-processing (left), and unique per-particle-type pre-processing (right) are all provided.
• Figure 5: Example edge structures within graphs passed through the vertexing network, the examples cover various 3- and 4-body reconstruction topologies. Faded nodes indicate particles missed; these are not included in the graph structures that encode the reconstruction topology.