New features in comparison to LHAPDF

[cschwan]

We should consider supporting some new features w.r.t. LHAPDF:

• [ ] alphas uncertainty estimation: in LHAPDF it's unclear how to estimate alphas uncertainties and we're only given different PDF sets with different alphas values
• [ ] uncertainty estimation for PDF sets with MHOU

[scarlehoff]

• [ ] Nuclear PDFs

[cschwan]

• [ ] Nuclear PDFs

Aren't they already supported in LHAPDF (that is if they have the correct Particle field)?

[scarlehoff]

tbh I don't remember right now what were the issues with the way LHAPDF does nuclear pdfs wrt the way we want to do nNNPDF, but there were some limitations.

[cschwan]

@Radonirinaunimi do you by chance remember what those were?

[Radonirinaunimi]

The main issue is the inability to interpolate between different $A$ values. The way it works ATM is that one has different PDF set for different nucleus/nucleon and hence need to load them separately.

I'd very much argue that being able to interpolate in 3-dimension $(x, Q^2, A)$ instead of just 2 would be a great additional feature. (This will for instance simplify incredibly the integration with n3fit and validphys.)

[alecandido]

We should consider supporting some new features w.r.t. LHAPDF:

Actually, there is already a list for them, they are existing issues:

• nuclear is #9 (but with two-dimensional interpolation, higher dimensional requires further troubles, i.e. implement the interpolator)
• $\alpha_s$ I'm not sure how the uncertainty should be estimated at all, so I'd like to have the theoretical answer before coding it in partons
• MHOU: well, what we said is that a MHOU set should be used as any other set, the MHOU is baked in (unless you use MC scales like approach)

However, you can see that there are many other features to be implemented: #11, #3 + the non-physical ones. Feel free to add more, and to contribute to their implementation with PRs :)

[Radonirinaunimi]

nuclear is https://github.com/AleCandido/partons/issues/9 (but with two-dimensional interpolation, higher dimensional requires further troubles, i.e. implement the interpolator)

I have some vague recollection (potentially wrong) that you wanted to implement an $N$-dimensional interpolation routine. What happens with that idea?

[scarlehoff]

It is in progress. Right now only for N=1 though.

[Radonirinaunimi]

It is in progress. Right now only for N=1 though.

Ah, ok! Good to know that it is on the plan.

[alecandido]

I have some vague recollection (potentially wrong) that you wanted to implement an

So, we are reproducing LHAPDF interpolation. There is the other repo for interpolation, and it will host N-dim as well, but doing N-dim cubic interpolation might be complex, so (for the time being) the goal is to do 1 (done by @scarlehoff in https://github.com/AleCandido/ndinterp/pull/2) and 2 (already available in LHAPDF). If you want to start working out the math and algorithm for N-dim, so much the better. You could also open a PR in ndinterp with just md notes about the math. But we will implement also 1- and 2-dimensional separately, just to reduce the differences with LHAPDF.