PHYSTAT Seminar: Statistical Uncertainty quantification for Parton Density Distributions (PDFs)

Wednesday 28 May 2025, 15:30 → 17:30 Europe/Zurich

This virtual Seminar with two speakers discusses the statistical uncertainty quantification for the Parton Density Distributions (PDFs) inside protons or heavier nuclei. The PDFs are a fundamental ingredient for predicting the rates of important processes such as Higgs production in pp collisions at the LHC. 

Agenda:

3.30 - 4 pm	Pavel Nadolsky (SMU)	Introduction and CTEQ view
4 - 4.15 pm	Questions to Pavel
4.15 - 4.45 pm	Maria Ubiali (Cambridge)	NNPDF view
4.45 - 5 pm	Questions to Maria
5 - 5.30 pm	General discussion and conclusion

 

Please note: today is an intro to the topic by two representative speakers and we plan to have a 0.5-1 day mini-workshop  linked to (a day before) the next PDF4LHC meeting in October at CERN more details to come.

 

Short description of the parton distribution functions:

When 2 high energy protons collide, the actual interact is between a quark or gluon in one proton with similar components in the other. For a proton with momentum p, each type of constituent can carry a fraction x of the momentum, where x lies in the range zero to 1. Then f(x) is a function describing the distribution in x for that particular constituent.

The other relevant constituent of the proton is the gluon, which has zero charge. There are then 9 functions f_i(x) of relevance. The quarks, anti-quarks and gluon are collectively known as ‘partons’, and the f_i(x) are the PDFs

The details of the reactions that take place depend on the type and the x-values of the colliding partons in the two initial protons. For example, if the x-values of the two partons are large, a lot of energy will be available for the production of new particles. In general, predictions consist of the rate of a process, as well as differential distributions of relevant variables such as masses and angles.

The f-distributions are determined (with some uncertainties, both statistical and systematic) from fits to a large amount of various data from particle physics collisions. The main methods are:

1: Choosing parametric forms for the various f’s;  or

2: Using neural networks as a non-parametric approach.

The number of data points used is in the thousands, and parametric methods use tens of parameters.

In this seminar we discuss some statistical issues that come up in the determination of these fits.

GMT20250528-133234_Recording_3840x2160.mp4

Mindmap_PDFseminar250528.pdf

nadolsky_phystat.pdf

Ubiali.pdf