Accurate uncertainty propagation is crucial for parton distribution functions (PDFs), particularly given the high-precision data expected from the HL-LHC. Traditional non-Bayesian approaches often struggle with strong non-linear dependencies in the forward map, motivating the need for more reliable Bayesian inference methods. However, these methods come with significant computational costs.
An ideal PDF parametrisation should satisfy three key criteria: (i) it must respect theoretical constraints, such as small- and large-x scaling behaviour, sum rules, and integrability; (ii) it should be sufficiently flexible to explore the space of candidate PDFs within the set of continuous, differentiable functions; and (iii) it should allow for efficient fitting of model parameters. While much attention has been given to the first two properties, the third—expedience of fitting—has remained largely unoptimised in the literature.
The goal of this talk is to explore this third aspect, focusing on strategies to improve the efficiency of PDF fitting.
