SciPost Submission Page
FASTColor -- Full-color Amplitude Surrogate Toolkit for QCD
by Javier Mariño Villadamigo, Rikkert Frederix, Tilman Plehn, Timea Vitos, Ramon Winterhalder
Submission summary
| Authors (as registered SciPost users): | Javier Mariño Villadamigo · Tilman Plehn · Ramon Winterhalder |
| Submission information | |
|---|---|
| Preprint Link: | https://arxiv.org/abs/2509.07068v1 (pdf) |
| Date submitted: | Sept. 16, 2025, 10:19 a.m. |
| Submitted by: | Javier Mariño Villadamigo |
| Submitted to: | SciPost Physics |
| Ontological classification | |
|---|---|
| Academic field: | Physics |
| Specialties: |
|
| Approaches: | Theoretical, Computational |
Abstract
High-multiplicity events remain a bottleneck for LHC simulations due to their computational cost. We present a ML-surrogate approach to accelerate matrix element reweighting from leading-color (LC) to full-color (FC) accuracy, building on recent advancements in LC event generation. Comparing a variety of modern network architectures for representative QCD processes, we achieve speed-up of around a factor two over the current LC-to-FC baseline. We also show how transformers learn and exploit underlying symmetries, to improve generalization. Given the gained trust in trained networks and developments in learned uncertainties, the LC-to-FC approach will eventually benefit further from not needing a final classic unweighting step.
Author indications on fulfilling journal expectations
- Provide a novel and synergetic link between different research areas.
- Open a new pathway in an existing or a new research direction, with clear potential for multi-pronged follow-up work
- Detail a groundbreaking theoretical/experimental/computational discovery
- Present a breakthrough on a previously-identified and long-standing research stumbling block
Current status:
Reports on this Submission
Report
Both the generation of particle collision processes as well as the simulation of their interaction with detectors is both crucially important and resource intensive. This work investiages how leading-color (LC) simulations can be upscaled to full-color (FC) precision using a machine-learning based reweighting scheme.
The paper is very well written, and carries out numerical studies that demonstrate the gain (up to a factor of 4-5 in the best case) over previously established techniques.
Below are listed two minor points that might be addressed, but overall this paper would be a good publication for SciPost Physics.
-> Rigth now, the production of the training sample (for which both LC and FC are required) and the neural network training are treated as 'free'. Can the necessary compute be quantified and put into relation to the expected number of samples that will be 'upsampled'
-> Relatedly, a discussion how reuseable vs problem specific these reweighting networks are envisioned? Is this currently e.g. a separate network per process, or is there some synergy (realised or conceivable)
Recommendation
Ask for minor revision