pineko 0.1.1

Combine PineAPPL grids and EKOs into FK tables

pineko = PineAPPL + eko

pineko converts

• interpolation grids for theory predictions ('grids' for short) in the form of PineAPPL grids, together with
• Evolution Kernel Operators (EKO) generated by eko

into fast-kernel (FK) tables. The collection of all FK tables constitute the theory predictions for a PDF fit and therefore is often simply called 'theory'.

pineko replaces APFELcomb, which was used up to NNPDF4.0.

Prerequisites

Generating a 'theory', as defined above, requires several files which are described next.

pineko.toml

You need to provide a pineko.toml, that provides all necessary paths to the input and output folders. [DEBUG: Look at the debug example in this repo [1].]

ymldb

You need all files of the ymldb [2]. [DEBUG: Look at the respective load.sh script to load from dom.] This defines the mapping from datasets to FK tables.

Theory Runcards

You need to provide the necessary theory runcards named with their respective theory ID inside the <paths.theory_cards> folder [3].

Default Operator Card

You need to provide a default operator card for eko [4]. [DEBUG: Look at the respective load.sh script to load from dom.]

Grids

pineko does NOT compute grids, which are instead expected input to pineko. There are typically two ways to obtain grids: computing them from scratch with runcards or reusing existing ones.

Generate new Grids with rr

You need to run rr with a given theory runcard and put the generated grid file with the same name inside the <paths.grids>/<theory_id> folder. The name has to match the ymldb which is the case by default.

Inherit Grids from Existing Theory

You can reuse the grids from a different theory by running pineko theory inherit-grids SOURCE_THEORY_ID TARGET_THEORY_ID DATASET1 DATASET2 .... The relation between the source theory and the target theory is non-trivial [5].

Running pineko

Running pineko consists of two steps - each of them being potentially computationally expensive: computing the EKO and convoluting the EKO with the grid.

Computing the EKO

Generating new EKOs

This is a two step process:

1. Generate the necessary operator cards with pineko theory opcards THEORY_ID DATASET1 DATASET2 ...
2. Generate the actual EKOs with pineko theory ekos THEORY_ID DATASET1 DATASET2 ...

Inherit EKOs from Existing Theory

You can reuse the EKOs from a different theory by running pineko theory inherit-ekos SOURCE_THEORY_ID TARGET_THEORY_ID DATASET1 DATASET2 .... The relation between the source theory and the target theory is non-trivial [6].

Generating the FK Table

You need to have the EKOs computed in the previous step. Then you can convolute the EKOs with the grids by running pineko theory fks THEORY_ID DATASET1 DATASET2 ...

---

[1] Actually, instead we should provide a concise description here - but let's wait to be stable first

[2] this is to be replaced by the new CommonData format

[3] this is to be replaced by a binding to the true theory DB

[4] I'm thinking how to improve this, because how could we provide a study on the interpolation accuracy? at the moment there just equal

[5] examples being SV, different evolution settings, etc.

[6] examples being SV, different DIS settings, etc.