O n l i n e T a l k : 12-Nov-2020
Making Monte-Carlo event generators fit for the HL-LHC era
We are being challenged by the increasing need for precision in MC predictions, to keep up with the ten-fold increase in experimental data expected from the High-Luminosity LHC, and to prepare for the growing interest in high-multiplicity final states. Projections of the CPU budget available for MC simulations show that new ideas are indeed required. Leaving aside detector simulations, the main bottleneck for simulating non-trivial processes by far is the evaluation of fixed-order matrix elements.
In this talk I discuss two potential avenues to tackle this problem. One is to reduce the number of evaluations required by making the Monte-Carlo phase-space sampling more efficient, and I will review a method recently proposed based on Neural Importance Sampling. Secondly, one can increase the throughput of matrix element calculations, and I will discuss our effort to deliver a practical implementation that makes use of the massively parallel computing power of accelerator cards such as GPUs.