Mellin Moments of Proton Unpolarized GPDs at Nonzero Skewness from Lattice QCD with a Neural-Network Parametrization
summary
The gist
Mellin Moments of Proton Unpolarized GPDs at Nonzero Skewness from Lattice QCD with a Neural-Network Parametrization determines the first four Mellin moments of unpolarized isovector proton…
In short
This study used a neural network to fit fourteen Generalized Form Factors (GFFs) from lattice QCD data to determine the first four Mellin moments of unpolarized proton Generalized Parton Distributions (GPDs) at nonzero skewness. The results show that while skewness-independent GFFs are well-determined, skewness-dependent ones are consistent with zero at present precision, yielding a quark total angular momentum of 0.296(11)(7) in the MS scheme.
Key concepts
- Generalized Parton Distributions (GPDs)
- GPDs describe the three-dimensional structure of the proton by linking momentum and spatial distribution. They are crucial for understanding how quarks and gluons are distributed inside the nucleon, especially when considering momentum transfer.
- Mellin Moments
- Mellin moments are mathematical tools used to extract specific information about GPDs by transforming them into a series of moments. This technique allows researchers to probe different aspects of the proton's structure based on different values of the momentum transfer squared, t.
- Generalized Form Factors (GFFs)
- GFFs are parameters that describe how the reduced Ioffe-time distributions relate to GPD moments. In this study, these factors were parameterized using an artificial neural network to flexibly model their dependence on the momentum transfer squared, t.
Terminology used across episodes
This episode discusses
- Mellin Moments of Proton Unpolarized GPDs at Nonzero Skewness from Lattice QCD with a Neural-Network Parametrization · Paper Radio
- On "dual" parametrizations of generalized parton distributions
- Generalized parton distributions and strong forces inside nucleons and nuclei
- Forces inside hadrons: pressure, surface tension, mechanical radius, and all that
- Wave Functions, Evolution Equations and Evolution Kernels from Light-Ray Operators of QCD
- Gauge-Invariant Decomposition of Nucleon Spin and Its Spin-Off
- Deeply Virtual Compton Scattering
- Scaling Limit of Deeply Virtual Compton Scattering
- Impact Parameter Dependent Parton Distributions and Off-Forward Parton Distributions for zeta to 0
- Femto-Photography of Protons to Nuclei with Deeply Virtual Compton Scattering
- Generalized parton distributions in impact parameter space
- Impact Parameter Space Interpretation for Generalized Parton Distributions
- Off-Forward Parton Distributions
- Generalized Parton Distributions
- Generalized Parton Distributions
- Unraveling hadron structure with generalized parton distributions
- Generalized parton distributions and the structure of the nucleon
- Generalized Parton Distributions in the valence region from Deeply Virtual Compton Scattering
- Parton Distribution Functions and their Generalizations
- Proof of Factorization for Deeply Virtual Compton Scattering in QCD
- Asymmetric Gluon Distributions and Hard Diffractive Electroproduction
The paper
Mellin Moments of Proton Unpolarized GPDs at Nonzero Skewness from Lattice QCD with a Neural-Network Parametrization · Read on arXiv
Min-Huan Chu, Manuel Colaěo, Krzysztof Cichy, Martha Constantinou, Arif Sarkar
Faculty of Physics and Astronomy, Adam Mickiewicz University · Department of Physics, Temple University
Transcript
Introduction to the show: ident: High Energy Physics: Lattice. The day in lattice field theory: what the simulations put on the table.
Kai: We're looking at the paper: "Mellin Moments of Proton Unpolarized GPDs at Nonzero Skewness from Lattice QCD with a Neural-Network Parametrization".
Mira: Here's the gist of the paper. Mellin Moments of Proton Unpolarized GPDs at Nonzero Skewness from Lattice QCD with a Neural-Network Parametrization determines fourteen Generalized Form Factors (GFFs) entering the first four Mellin moments of unpolarized isovector proton Generalized Parton Distributions (GPDs) at nonzero skewness, yielding the quark total angular momentum of the isovector combination as 0.296(11)(7) in the MS scheme at µ = 2 GeV The framework combines the polynomiality of moments in skewness with the short-distance expansion formalism to express lattice-extracted reduced Ioffe-time distributions (ITDs) in terms of GFFs In order to extract the GFFs by fitting these relations to the lattice data, each GFF is parameterized with an artificial neural network (ANN) as a continuous function of the invariant momentum transfer squared t, which accommodates kinematic coverage flexibly The study employs Short Distance Factorization (SDF) to determine the GFFs that enter the Mellin moments of unpolarized GPDs at zero and nonzero skewness Neural Network Parametrization The GFFs are represented by an ANN(tˆ) = b2 + X M i=1 W i 2 tanh W i 1 tˆ+ b i 1 This procedure is a global fit where all relevant kinematics and coupled GFFs are fitted simultaneously, accommodating variations in t without assuming an analytical form The network parameters are determined by minimizing a χ2 loss function that quantifies the agreement between model predictions and lattice data for the reduced ITDs Results on Generalized Form Factors The resulting GFFs display a clear hierarchy, where skewness-independent GFFs An+1,0 and Bn+1,0 are well determined The skewness-dependent GFFs A3,2, B3,2, A4,2 and B4,2 are consistent with zero at the present precision From the forward values of A2 0 and B2 0, the quark total angular momentum is obtained as J u−d = 0.296(11)(7) at µ = 2 GeV in the MS scheme Systematic Uncertainty Analysis The systematic uncertainty for each GFF collects three contributions: the form of the Wilson coefficients, from the network (number of neurons and initialization point), and is added in quadrature with statistical uncertainty to give the total uncertainty The results from both fixed-scale and Renormalization Group Resummation (RGR) prescriptions agree throughout the fitted t interval The calculation serves as a proof of concept, demonstrating that while the data constrains skewness-independent GFFs well, the skewness-dependent GFFs are largely consistent with zero at present precision Further precision requires denser kinematic coverage and finer lattice spacings The results agree between the fixed-scale analysis and the RGR prescription, supporting the central result obtained from the fixed energy scale analysis at µ = 2 GeV Pruning redundant reasoning tokens The framework combines the polynomiality of moments in skewness with short-distance expansion formalism to express lattice-extracted reduced Ioffe-time distributions in terms of GFFs The study employs Short Distance Factorization (SDF) to determine the GFFs that enter the Mellin moments of unpolarized GPDs at zero and nonzero skewness How the training signal is built The training set comprises nonforward reduced ITDs at ξ = 0 and ξ ≠ 0 together with forward points at t = ξ = 0 The minimization with respect to network parameters is performed using full-batch gradient descent with a fixed learning rate Where the gains come from The gain in precision is observed when adding nonzero-skewness data to the zero-skewness ones, which reduces statistical uncertainty at t = 0 by about 10% for A2 0 and A3 0 and by about 35% for B2 0 and B3 0 The results from both fixed-scale and RGR prescriptions agree throughout the fitted t interval Where the gains come from The gain in precision is observed when adding nonzero-skewness data to the zero-skewness ones, which reduces statistical uncertainty at t = 0 by about 10% for A2 0 and A3 0 and by about 35% for B2 0 and B3 0 How the training signal is built The training set comprises nonforward reduced ITDs at ξ = 0 and ξ ≠ 0 together with forward points at t =
Title and authors: Kai: So, moving into the specifics of this paper, "Mellin Moments of Proton Unpolarized GPDs at Nonzero Skewness from Lattice QCD with a Neural-Network Parametrization", what are we actually looking at here? It’s about those GFFs.
Mira: They are trying to determine fourteen Generalized Form Factors, or GFFs, which enter the first four Mellin moments of unpolarized isovector proton Generalized Parton Distributions at nonzero skewness.
Lev: So it’s not just one thing; they're mapping out a whole set of functions that describe how the proton looks in terms of its internal structure under different momentum conditions.
Kai: The methodology uses the polynomiality of moments in skewness combined with the short-distance expansion formalism to express lattice-extracted reduced Ioffe-time distributions in terms of those GFFs.
Mira: And instead of just fitting these relations directly, they parameterize each GFF with an artificial neural network that's a continuous function of the invariant momentum transfer squared, t.
Lev: That’s the core idea for flexibility; it allows them to cover different kinematic regimes without having to assume a specific analytical form for how those form factors behave as t changes.
Kai: They use this neural network parametrization to fit the GFFs to their lattice data, minimizing a chi squared loss function that measures the agreement between their model predictions and the lattice data for those reduced ITDs.
Mira: The training signal they built included nonforward reduced ITDs at skewness zero and skewness not equal to zero, along with forward points at t equals skewness zero.
Lev: That training set construction is interesting because it forces the network to learn how the GFFs transition from one regime to another, even when it’s just a mathematical relationship derived from polynomiality.
Kai: The results show that skewness-independent GFFs like A+ten and B+ten are well determined by this process <ref:2610.29726#pg2>.
Mira: But they also report that the skewness-dependent ones, such as A3,2 and B4,2 are consistent with zero at their present precision <ref:2610.29726#pg1>.
Lev: That consistency with zero is a key result if we think about how these GFFs might behave in more complex scenarios like those involving transverse polarization later on.
The paper's summary: Kai: So, to summarize the main finding of this paper, they successfully determined fourteen GFFs using a neural network approach applied to lattice QCD results for unpolarized isovector proton GPDs at nonzero skewness.
Mira: They achieved this by combining polynomiality relations in skewness with short-distance expansion formalism to connect lattice quantities to the Mellin moments of GPDs.
Lev: Essentially, they took the raw data from lattice QCD, which gives them those reduced Ioffe-time distributions, and used a neural network to find the functional form of those GFFs based on how they relate to skewness and t.
Kai: The implication is that this method provides a way to extract these GFFs even when you have nonzero skewness data available, which was previously difficult.
Mira: It opens up access to the quark total angular momentum through Ji’s sum rule, which is what those Mellin moments are ultimately used for in understanding nucleon spin structure.
Lev: But I gotta say that experimental extraction of GPDs is tough, and this paper shows how a theoretical framework like this can be used to constrain those inputs.
Kai: They also show that they can use the forward values of A2 zero and B2 zero to get a quark total angular momentum of zero point two nine six(eleven)(seven) at mu equals two GeV in the MS scheme.
Mira: So, even with this framework, we still have those caveats about the precision they are working with, which is something I’m always thinking about as a theorist.
The paper's improvements: Kai: One of the improvements they highlight is that the neural network parametrization allows for a global fit to all kinematic data simultaneously without having to assume an analytical form for how t dependence works.
Mira: That flexibility is key because it accommodates variations in t that can be substantially different between the zero-skewness and nonzero-skewness regimes.
Lev: From my side, that means the network isn't just a toy model; it’s trying to capture a more complex reality of how these form factors evolve across momentum scales.
Kai: They also state they can perform systematic error analysis by examining how results change when varying inputs like the number of neurons and the initialization point or even the Wilson line length window, z over a.
Mira: And they also show that the results from both fixed-scale and Renormalization Group Resummation prescriptions agree throughout the fitted t interval, which is a good cross-check for perturbative matching.
Lev: If those two different resummation schemes agree across the whole fitted range, that gives you more confidence in the underlying physics being captured by the lattice data.
Kai: They also mention that they can compare fixed-scale analysis with the RGR prescription as a complementary test of how well their calculations are matching.
Mira: And when we look at how they train it, adding nonzero skewness data to zero-skewness ones actually reduces statistical uncertainty at t equals zero by about ten percent for A2 zero and A3 zero and by about thirty-five percent for B2 zero and B3 zero.
Conclusion: Kai: So, wrapping up the "Mellin Moments of Proton Unpolarized GPDs at Nonzero Skewness from Lattice QCD with a Neural-Network Parametrization", the main takeaway is that while skewness-independent GFFs are well constrained, the skewness-dependent ones are largely consistent with zero at this precision.
Mira: They confirmed that this framework can extract the quark total angular momentum of J u minus d to be zero point two nine six(eleven)(seven) in the MS scheme at mu equals two GeV using forward values of A2 zero and B2 zero.
Lev: For me, I see this as a proof that these lattice methods can deliver constraints on important nucleon structure observables even with this level of precision.
Kai: And they’re setting up the next steps where denser kinematic coverage and finer lattice spacings will be needed to push the precision further.
Mira: It’s a solid piece of work that shows how AI can help in these kinds of calculations, but it’s important to remember it's still a proof of concept for extracting those GPDs.
Lev: I think we need more robust ways to handle the systematic uncertainties before we can really rely on these results for any serious experimental comparison.
More episodes
- The day in conversation for 2026-11-19
- Daily Summary for 2026-11-19
- 2606.20029-Finite-Volume Step Scaling at Fixed Truncation in (2+1)D Hamiltonian U(1) Lattice Gauge Theory
- 2603.16266-Lattice QCD study of the K*(892) resonance at the physical point
- 2607.15742-Path optimization method for the sign problem: Insights from random matrix models
- The day in conversation for 2026-11-18
- Daily Summary for 2026-11-18
- Daily Summary for 2026-11-17
- 2608.18239-Searching for symmetric mass generation with staggered fermions in four dimensions
- The day in conversation for 2026-11-17