Optimised Distillation for Nucleon Correlation Functions

arXiv:2611.12474 · hep-lat · Submitted 2026-11-09 · Read on arXiv

Listen

Radio episode about this paper

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: "Optimised Distillation for Nucleon Correlation Functions".

Mira: Here's the gist of the paper. The calculation and analysis of nucleon correlation functions is essential for understanding the observed spectra and structure of these ubiquitous hadrons <ref:2611.12474#pg2>.

Kai: By modulating the contribution of each Laplacian eigenmode to the correlator via distillation profiles, we build optimised operators which improve the overlap with the nucleon ground state <ref:2611.12474#pg3>.

Mira: The analysis of Euclidean two-point correlation functions in lattice QCD is complicated by a windowing problem where early times are dominated by excited states and late times suffer from signal loss to noise <ref:2611.12474#pg3>.

Kai: Distillation addresses this by working within the subspace spanned by the low modes of the 3D gauge-covariant Laplacian <ref:2611.12474#pg3>.

Paper summary: Kai: So we're looking at this paper today called "Optimised Distillation for Nucleon Correlation Functions". Basically, they're dealing with these nucleon correlation functions in lattice QCD to figure out the structure of these particles.

Mira: And what they’re tackling is a big problem with how we look at those functions. At early times, you get all these excited states dominating the signal, and at late times, you just lose the ground state signal to noise exponentially > Kai Right.

Lev: Exactly. That windowing problem makes it really hard to know where in time to even start extracting the ground state energy reliably > Mira And this paper claims they can solve that by using distillation profiles to build better operators that overlap more with the actual nucleon ground state > Kai

Lev: It sounds like they're trying to filter out all that noise by focusing only on the lowest modes of a certain Laplacian operator, which is a way of doing things in lattice QCD > Mira But what makes this specific distillation profile better than just using the standard low modes? > Kai

Mira: The paper suggests using Gaussian profiles to modulate the weight of each eigenmode, so it puts more emphasis on the lower modes and suppresses the higher ones, instead of giving them equal weight as in standard distillation > Lev That’s a key technical detail; they’re essentially constructing a variational smearing basis by using multiple widths for those Gaussian profiles > Kai

Lev: So, the method involves solving what they call a generalized eigenvalue problem on each timeslice relative to a reference timeslice, which defines these principle correlators and eigenvectors for their optimal linear combinations > Mira That sounds like a lot of calculation happening right there on the lattice grid at every single time step > Kai

Lev: And they use elementals to replace standard interpolating operators, which are tailored to capture both the color and Dirac structure of the nucleon at rest, which is how you set up those initial operators > Mira But they also mention this construction doesn't require any additional inversions, it’s all done at the contraction level > Kai

Lev: That reduces the computational overhead significantly for what they are trying to achieve > Mira So, when they tested this on a nucleon at rest, what did they find about how it behaves compared to conventional methods? > Kai

Mira: They found that these optimized distillation profiles recreated excited state behavior similar to how you get results with conventional correlator calculations where you tune the source smearing for the best overlap with the ground state > Lev And they noted that this optimal weighting puts a higher weight on lower eigenmodes, and it drops to zero as those modes approach the cut-off > Kai

Paper summary: Lev: The result is an excited state suppression comparable to what you get with optimized Wuppertal smearing in other studies > Mira But when they introduce finite momentum, that degree of excited state suppression actually decreases as the momentum increases > Kai

Lev: That’s interesting because it means the method has a dependence on the momentum being used, which is something we have to keep in mind when we move this to real hardware > Mira So, what does this mean for figuring out nucleon matrix elements, like those for the isovector axial charge? > Kai

Lev: For matrix elements, they use those eigenvectors from their generalized eigenvalue problem to rotate the three-point function into this optimized basis > Mira And in that case, they saw a considerable reduction in excited state contamination when using these new operators for the isovector axial charge > Kai

Lev: It seems like this approach gives us cleaner data for those specific quantities, which is what we need to see to make real physical predictions > Mira They also mentioned that distillation profiles are compatible with existing correlator construction methods, so you can combine them with other operator bases to build even larger variational bases > Kai

Lev: Looking ahead, the team plans to investigate other forward nucleon matrix elements and extend this analysis to finite momentum transfer for form factor calculations > Mira That’s a good next step because form factors are crucial for describing how hadrons interact at different energy scales > Kai

Lev: It sounds like they're taking a technique that works well in the theory space and figuring out exactly how to make it computationally feasible on actual hardware, which is where I come in > Mira So, to wrap up this paper "Optimised Distillation for Nucleon Correlation Functions", what’s the main point we should remember? > Kai

Mira: The main thing is that by modulating the contribution of each Laplacian eigenmode with these Gaussian distillation profiles, they build operators that significantly improve the overlap with the nucleon ground state > Lev It moves us away from having to manually guess how many eigenvectors to keep based on some arbitrary rule and instead lets the math tell us which modes are important > Kai

Lev: And for experimentalists or theorists who need these nucleon correlation functions, this means cleaner signals, especially when you're looking at those early times where excited states usually mess things up > Mira This paper shows a way to optimize the noise reduction in a way that is systematically controlled by the Gaussian width parameter > Kai

Lev: It’s about making sure that when we extract these physical properties, we aren't just getting contaminated results from higher energy states than we accounted for > Mira So, this work gives us a more robust tool for extracting nucleon information from lattice QCD simulations > Kai

Conclusion: Kai: So we're looking at the paper "Optimised Distillation for Nucleon Correlation Functions" from arXiv. The core idea is they found a better way to build operators for nucleon correlation functions in lattice QCD.

Mira: Right, so they used these distillation profiles, which are basically ways to weigh different modes of a mathematical operator, to get much cleaner signals from the simulation. It’s about optimizing how you look at the data without having to guess manually how many modes are good enough.

Lev: From an error correction standpoint, that's smart because it means we don't have to spend all our computational budget trying every possible combination of operators; we just focus on what the math tells us is most relevant.

Kai: Exactly. They show this method works for a nucleon at rest, and it recreates behavior similar to how you tune source smearing in traditional methods, but they control the excited state contamination much better.

Mira: The key is that the weighting drops off nicely as you get to higher modes near the cut-off, which is a very clean way to suppress those unwanted excited states without losing too much signal. It's a systematic approach.

Lev: And for matrix elements, like those for the isovector axial charge, they used these eigenvectors to rotate the three-point function into this optimized basis and saw a big reduction in that contamination. That’s what we want to see when trying to extract physical properties from simulations.

Kai: It suggests that this isn't just some tweak; it’s a fundamental way to structure our variational basis for these calculations, and it holds up when they test it with finite momentum too, though the suppression changes depending on the momentum.

Mira: That dependence on momentum is something we need to watch closely because it tells us how sensitive this specific optimization is to the physics we're measuring at different scales. It’s a useful tool for now, but we have more ground to cover in terms of what other matrix elements this technique unlocks.

J. A. Crawford, J. Finkenrath, L. Barca

Department of Physics, Bergische Universität Wuppertal · Aix Marseille University, Université de Toulon, CNRS, CPT, Marseille · John von Neumann-Institut für Computing NIC, Deutsches Elektronen-Synchrotron DESY

hep-lat

Submitted: 2026-11-09

Updated: 2026-11-09

Comments: 9 pages, 4 figures, Contribution to the 43rd International Symposium on Lattice Field Theory (Lattice 2026)

License: http://creativecommons.org/licenses/by/4.0/

Importance score: 73/100

The gist: The calculation and analysis of nucleon correlation functions is essential for understanding the observed spectra and structure of these ubiquitous hadrons <ref:2611.12474#pg2>.

Key concepts

Euclidean two-point correlation functions
These are calculations performed in lattice QCD that help determine the structure and properties of nucleons. They are analyzed over time to extract information about the nucleon's ground state, but they suffer from noise issues at different time scales.
Laplacian eigenmodes
The Laplacian operator is used to decompose the system's behavior into different modes, similar to how a complex wave is broken down into simple sine waves. The paper focuses on using only the lowest energy modes of this decomposition to build better nucleon operators.
Distillation profiles
These are weighting functions applied during operator construction. Instead of choosing a fixed number of basis states, distillation profiles dynamically adjust the contribution of each Laplacian eigenmode, giving more weight to the most relevant low-energy modes.

Terminology

Summary

The calculation and analysis of nucleon correlation functions is essential for understanding the observed spectra and structure of these ubiquitous hadrons <ref:2611. p. 0

Where the paper says it

20, 42119 Wuppertal, Germany 𝑏Aix Marseille Univ, Université de Toulon, CNRS, CPT, Marseille, France 𝑐 John von Neumann-Institut für Computing NIC, Deutsches Elektronen-Synchrotron DESY, Germany E-mail: crawford@uni-wuppertal. de The calculation and analysis of nucleon correlation functions is essential for understanding the observed spectra and structure of these ubiquitous hadrons.

12474#pg2>. By modulating the contribution of each Laplacian eigenmode to the correlator via distillation profiles, we build optimised operators which improve the overlap with the nucleon ground state <ref:2611. p. 0
Where the paper says it

This leads to a common challenge in identifying a reasonable window on which to extract the ground state energy. By modulating the contribution of each Laplacian eigenmode to the correlator via distillation profiles, we build optimised operators which improve the overlap with the nucleon ground state.

12474#pg3>.

Introduction and Problem Statement

The analysis of Euclidean two-point correlation functions in lattice QCD is complicated by a windowing problem where early times are dominated by excited states and late times suffer from signal loss to noise <ref:2611.12474#pg3>. Distillation addresses this by working within the subspace spanned by the low modes of the 3D gauge-covariant Laplacian <ref:2611. p. 1

Where the paper says it

Distillation [1] is a powerful framework in which to compute hadronic correlation functions. Working within the subspace spanned by the low modes of the 3D gauge-covariant Laplacian, ∇ 2 𝑥 𝑦 [𝑡] = 6𝛿𝑥 𝑦 − ∑︁ 3 𝑗=1 𝑈𝑗(𝑥, 𝑡)𝛿𝑥+ ˆ𝑗,𝑦 + 𝑈 † (𝑥 − ˆ𝑗, 𝑡)𝛿𝑥− ˆ𝑗,𝑦 (1) where the eigenvectors on each timeslice 𝑡 are −∇2 𝑥 𝑦 [𝑡]𝑣 𝑗 [𝒚, 𝑡] = 𝜆 𝑗 [𝑡]𝑣 𝑗 [𝒙, 𝑡].

12474#pg3>. The use of distillation profiles negates the need to predetermine how many eigenvectors should be kept to achieve optimal overlap with the state of interest <ref:2611. p. 1
Where the paper says it

Outside of volume-scaling arguments, it is not clear beforehand how many eigenvectors should be kept to achieve the optimal overlap with the state of interest. The use of distillation profiles [2–4] negates this issue by reweighting each eigenmode such that it contributes optimally to the state of interest.

12474#pg4>.

Methodology for Operator Construction

The construction utilizes perambulators, which project propagators into the space spanned by the lowest modes of the Laplacian <ref:2611.12474#pg4>. Elementals replace interpolating operators capturing color and Dirac structure, further modulated by Gaussian profiles to place higher weight on lower eigenmodes <ref:2611. p. 2

Where the paper says it

Elementals, Φ, replace interpolating operators that capture the colour and Dirac structure of the nucleon at rest, Φ 𝑖 𝑗𝑘 𝛼𝛽 [𝑡] = ∑︁ 𝒙 𝜖 𝑎𝑏𝑐𝑣 𝑖 𝑎 [𝒙, 𝑡]𝑣 𝑗 𝑏 [𝒙, 𝑡]𝑣 𝑘 𝑐 [𝒙, 𝑡] [𝐶𝛾5] 𝛼𝛽, (4) where 𝜖 𝑎𝑏𝑐 is 3D Levi-Civita symbol, with 𝜖 123 = +1 and 𝐶 = 𝑖𝛾2𝛾4 is the charge-conjugation matrix. In the implementation of distillation profiles, we further modulate the relative weight of each eigenmode via a Gaussian profile, 𝑣 𝑖 [𝑡] → exp − 𝜆𝑖[𝑡] 2 2𝜎2 𝑣 𝑖 [𝑡], (5) where 𝜎 controls the width of the Gaussian profile.

12474#pg4>. A generalized eigenvalue problem (GEVP) is solved on each timeslice to define the principle correlators and eigenvectors that form the optimal linear combinations of the variational basis <ref:2611. p. 2
Where the paper says it

Importantly, this does not require any additional inversions and is performed entirely at the contraction level. We solve a generalised eigenvalue problem (GEVP) on each timeslice 𝑡 relative to a reference timeslice 𝑡0, 𝐶(𝑡)𝑣 (𝑢) (𝑡, 𝑡0) = 𝜆 (𝑢) (𝑡, 𝑡0)𝐶(𝑡0)𝑣 (𝑢) (𝑡, 𝑡0), (6) where the eigenvalues 𝜆 (𝑢) (𝑡, 𝑡0) = 𝑒 −𝐸𝑢 (𝑡−𝑡0 ) , with 𝐸𝑢 being the energy of state 𝑢, define the principle correlators and the eigenvectors 𝑣 (𝑢) (𝑡, 𝑡0) define the optimal linear combinations of the variational basis of distillation profiles.

12474#pg4>.

Performance in Nucleon Correlation Functions

For a nucleon at rest, optimised distillation profiles recreated excited state behaviour similar to conventional correlator calculations where source smearing was tuned for optimal overlap with the nucleon ground state <ref:2611.12474#pg6>. The optimal weighting places higher weight to lower eigenmodes, decreasing to zero as modes approach the cut-off, and this results in an excited state suppression comparable to optimised Wuppertal smearing <ref:2611. p. 4

Where the paper says it

from the GEVP, the optimised nucleon correlator is constructed from 𝐶(𝑡 − 𝑡0) = 𝑣 (0)† 𝑖 (𝑡, 𝑡0)𝐶𝑖 𝑗(𝑡)𝑣 (0) 𝑗 (𝑡, 𝑡0), (9) where 𝑣 (0) (𝑡, 𝑡0) is the eigenvector corresponding to the lowest eigenvalue. Compared to the flat profile, the optimal weighting places higher weight to lower eigenmodes, decreasing to zero as the modes approach the cut-off.

12474#pg6>. When finite momentum is introduced, the degree of excited state suppression decreases with increasing momentum <ref:2611. p. 4
Where the paper says it

(10) As the momentum increases, the onset of statistical noise occurs earlier in Euclidean time. We find that the degree of excited state suppression decreases with increasing momentum.

12474#pg6>.

Matrix Element Extraction and Conclusion

For nucleon matrix elements, eigenvectors from the GEVP are used to rotate the three-point function to the optimised basis <ref:2611.12474#pg8>. In the case of isovector axial charge, a considerable reduction in excited state contamination was found by using these optimised operators <ref:2611. p. 6

Where the paper says it

In the calculation of nucleon matrix elements, the eigenvectors obtained from the GEVP of the two-point correlator were used to rotate the three-point function to the optimised basis. In the case of the nucleon isovector axial charge 𝑔 𝑢−𝑑 𝐴 , we found a considerable reduction in excited state contamination by using these optimised operators.

12474#pg9>. The implementation of distillation profiles is compatible with existing correlator construction methods and can be combined with other operator bases to construct even larger variational bases <ref:2611. p. 6
Where the paper says it

In the case of the nucleon isovector axial charge 𝑔 𝑢−𝑑 𝐴 , we found a considerable reduction in excited state contamination by using these optimised operators. The implementation of distillation profiles is compatible with existing correlator construction methods, and can be combined with other operator bases to construct even larger variational bases.

12474#pg9>. Future work plans include investigating other forward nucleon matrix elements and extending the analysis to finite momentum transfer for form factor calculations <ref:2611. p. 7
Where the paper says it

Optimised Distillation for Nucleon Correlation Functions J. Crawford elements and extend our analysis to finite momentum transfer, reducing excited state contamination for form factor calculations.

12474#pg9>.

Green — the paper says it, and the page is printed. Red — no passage carries it, so treat that sentence as unverified. The same check runs on your own text: check a summary against its source, $5.

Improvements for AI systems

  1. Optimised operators can be constructed using distillation profiles to improve overlap with the nucleon ground state, which reduces excited state contamination in two- and three-point correlation functions. This allows for more reliable extraction of ground state energy and matrix elements from lattice QCD calculations.

  2. The use of Gaussian profiles modulating each eigenmode's weight allows the construction of a variational smearing basis without requiring additional inversions, as described by the GEVP to optimise the weighting of each retained eigenmode. This provides a cost-effective means for generating a variational basis for nucleon correlation functions.

  3. For finite momentum nucleons, the analysis shows that the degree of excited state suppression decreases with increasing momentum, indicating how distillation profiles interact with momentum smearing techniques to manage noise in boosted systems. This suggests an improvement in signal quality for form factor calculations at higher momenta.

  4. The method allows for the construction of nucleon matrix elements by taking a ratio of three-point correlation functions to two-point correlation functions, where By increasing the overlap with the ground state, A0, and decreasing the overlap with the first excited state, A1, we effectively suppress these contributions from these transition matrix elements. This directly improves the precision of extracted values like g u-d A.

Abstract

The calculation and analysis of nucleon correlation functions is essential for understanding the observed spectra and structure of these ubiquitous hadrons. Two challenges arise in their analysis: at early Euclidean times, they are dominated by excited state contributions and at later Euclidean times, the signal-to-noise ratio degrades exponentially. This leads to a common challenge in identifying a reasonable window on which to extract the ground state energy. By modulating the contribution of each Laplacian eigenmode to the correlator via distillation profiles, we build optimised operators which improve the overlap with the nucleon ground state. In these proceedings, we present our progress on constructing optimised distillation profiles for nucleons and their excited state suppression in two- and three-point correlation functions.

Sources

Related papers