Introduction

In the nnpdf project, data files used by the code may be grouped into two categories, theory and experiment. Experimental data and the information pertaining to the treatment of systematic errors are held in the CommonData files. FK tables, and CFACTOR files store the precomputed information for use when calculating theoretical predictions corresponding to information held in the equivalent CommonData. In this section the file formats and naming conventions for these files will be detailed, along with the directory structure employed by the nnpdf code.

For NNPDF4.0 and later fits, a considerably larger number of theory options will be explored than in previous determinations. The current theory documentation only refers to 4.0 and previous fits and is thus outdated. In NNPDF3.0 the main theory variations used were perturbative order, value of the strong coupling and the number of active flavours in the VFNS. For NNPDF3.1 and later, it has been necessary to accommodate variations in additional parameters, such as treatments of the heavy quark mass (pole vs MS-bar), scale variations, intrinsic charm, resummation effects etc. The book-keeping used to enable efficient variations of the theoretical treatment used in fits post-3.0 will therefore also be outlined here.

This section will begin by detailing the specifications for the file formats used by the code, first with the experimental data file formats and layouts in Experimental data files and secondly with the file formats used for theoretical predictions in Theory data files. Finally the organisation of these files within the nnpdf structure will be described in Organisation of data files.

Important definitions

In order to clarify the later description, here are a few important terminological points to note.

Dataset vs Experiment

When referring to a collection of data points two words are used in the nnpdf code which have specific meanings. Dataset refers to the result of a specific measurement, typically associated with a single experimental paper and corresponds to the DataSet class in the nnpdf code. Experiment refers to a collection of Datasets which might be associated by experimental cross-correlations. For example, the ATLAS 2010 R=0.4 inclusive jet measurement and the ATLAS 2011 high-mass Drell-Yan measurement are both examples of Datasets as used in the NNPDF3.0 analysis. Both of these datasets are grouped into the ATLAS Experiment as they have systematic uncertainties that are cross-correlated with each other. In this document, when using these terms in this sense, they will be italicised for clarity.

Dataset naming conventions

See dataset_naming_convention for a definition of how datasets should be named.