Francesco Giuli

From the beginning of my Ph.D., the focus of my research has mainly been the deep study of the internal structure of the proton. I have participated in several measurements of the Parton Distribution Functions (PDFs) of the proton both within and outside the ATLAS Collaboration and I became an expert, and later a developer, of the xFitter framework, an open-source tool commonly used in PDF studies. I was the convener of the ATLAS PDF Fit Forum and the principal contributor to the first PDF fit from a LHC collaboration which uses multiple data sets (even at 13 TeV, which are not used into the latest global PDF analyses) and assesses the impact of correlation of systematic uncertainties between various data sets. This is something which cannot be done by global fitters, since it requires a deep insight into the experimental details of the input measurements. This work provided crucial information to the whole PDF community, and they are already taken into account into the next version of global fits. Alongside my PDF studies, I have been actively involved in analyses which are particularly relevant for the extraction of SM fundamental parameters.

I worked on measuring the low mass Drell-Yan cross section down to mll = 7 GeV (first analysis in ATLAS). This analysis explores kinematic regions complementary to the ones accessible with the LHCb and HERA experiments, where novel effects in both perturbative and non-perturbative QCD (such as small-x or small-pT logarithmic enhancement) may become important. Currently, I am convening the the ATLAS W/Z Physics Group and leading the precision measurement of the production cross section of W boson in association with a c-quark, focusing on the muon decay final state. Given that this channel is dominated by strange-initiated production, these data can help in resolving the debate on the level of strangeness suppression at low Bjorken x. The Soft Muon Tagging (SMT) technique to identify the hadronic jet with a spatially-associated muon is exploited in this analysis. I am one of the major developers of this algorithm, which has also been used in a recent top-quark mass measurement, where semi-leptonic decays of b-hadrons produced in the top-quark decay chain are used. The results of this study are characterised by reduced and largely-independent systematic uncertainties compared to the conventional top-quark mass analyses conducted in the lepton + jets channel.

At present, I am contributing to the measurements of tt/Z cross-section ratio and inclusive W, Z cross-section with early Run 3 data. In these two analyses, I generated higher-order, state-of-the-art NNLO QCD + NLO EW predictions for different PDF sets. Furthermore, I am involved in the Z boson mass measurement, evaluating higher order QED and EW corrections to the Z boson line shape and I will be soon involved in muon calibration studies, retrieving an additional sagitta bias correction using the ``pseudo-mass'' asymmetry method.

I am also concentrating on improving the pT(Z) and pT(W) modelling using state-of-the-art N3LO+aN4LL calculations, performing fits to several ATLAS and non-ATLAS measurements to better constrain the non-perturbative corrections at very small transverse momenta (pT < 5GeV) introduced in the calculations. This effort could lead to substantial improvements in future mW determinations. I am also interested in finalising the measurement of the effective leptonic weak mixing angle using Z-boson decays. In the past, I generated theoretical predictions for the ATLAS W charge asymmetry measurement at 8 TeV and I investigated the potential impact of this determination. Since the uncertainties due to PDFs are the largest source for this measurement, further constraints might lead to sizeable decrement in the determination of sin2(thetaW).

I also had the opportunity to contribute to other tasks in ATLAS during this time. I took part to the development and installation of the Phase 1 upgrade of high-performance Resistive Plate Chambers triplets specifically designed for the barrel-to-forward transition region of the ATLAS detector (BIS78 project).

Moreover, in the last two years, I was heavily involved in trigger operations, being on-call expert for both on-line and menu High Level Trigger (HLT). Since very recently, I was the ATLAS Express Stream Coordinator, and I was responsible for defining the express stream menu which provides the data used in the initial Data Quality assessment and I am also monitoring of the actual express stream bandwidth to ensure that the corresponding targets are being met.

Additionally, I am exploring the possibility of running the ATLAS trigger software on GPUs or FPGAs. The development of the HL-LHC HLT software for usage on these new technologies may allow for faster processing than using CPUs, and could lead to further developments such as applying machine learning on tracking algorithms at the trigger level. These potential benefits are currently being studied by the community. In this context, I investigated the timing performance of commercially available accelerator cards in the context of the New Small Wheel HLT. I studied the performance obtained running three machine learning models (a Deep Neural Network for cluster position reconstruction and a Convoluted Neural Network and a Recurrent Neural Network for muon pattern recognition) on a commercially available Xilinx Alveo U50 and U250 and VCK5000 Versal development cards. I compared the inference time obtained between CPU, GPU and the above-mentioned cards using development tools and off-the-shelf hardware from companies.