Simulated and Reconstructed Events with CMS and DELPHES

We provide the dataset used for fast simulation studies.

The dataset consists of preprocessed CMS Open Data for the ttbar,  gg->H->4l, QCD 470, and QCD 600 processes. Each file includes a branch with stable input particles before detector simulation and a branch with the reconstructed particle flow candidates. A 1 GeV pT cut is applied to the particle flow candidates, while a 0.25 GeV cut is applied to the stable input particles. Additionally, a |η| < 2.7 cut is applied to both collections.

Our model was trained on the first 2,800,000 events from the `qcd470_ttbar_train_cms.root` file, with the remaining 35,000 events used for validation.

Each test file contains 100,000 events.

For comparison, we provide detector simulations obtained with DELPHES 3.5 for the same set of truth-level events. We used the CMS card with pileup, including an average of 6.35 pileup vertices per event. To match the full simulation, we disabled the smearing of the truth primary vertex in Delphes, while the pileup vertices were smeared using the default values.