Large-Momentum Effective Theory's Asymptotic Extrapolation vs the Inverse Problem

This paper defends Large-Momentum Effective Theory (LaMET) against recent claims that it suffers from an unquantifiable inverse problem, arguing that physics-guided systematic extrapolation remains the most reliable method for estimating uncertainties in parton distribution calculations even when lattice data precision is suboptimal.

The Gist

Imagine you are trying to figure out the exact recipe for a delicious, complex soup (the Parton Distribution Function, or PDF) that makes up a proton. You can't taste the soup directly because it's locked inside a tiny, invisible pot. Instead, you have to look at the steam rising from the pot (lattice QCD data) and try to guess the ingredients based on how that steam behaves.

This paper is a debate between two groups of physicists about how to read that steam to get the recipe right.

The Two Competing Methods

1. The "Short-Distance" Method (SDF): The "Blind Taste Test"
Imagine you can only stick your nose very close to the pot's lid (a very short distance, about 0.2–0.3 fm). You get a whiff of the steam, but it's faint and blurry.

• The Problem: Because you only have a tiny sniff, you have to guess what the rest of the soup tastes like. You might say, "Well, it smells like basil, so I'll guess the whole soup is basil." Or, "It smells like garlic, so maybe it's garlic soup."
• The Risk: This is called an Inverse Problem. You are trying to work backward from a tiny, incomplete clue to a whole picture. There are infinite ways to guess the recipe that fit your tiny sniff. The result is a lot of guessing, and the "error bars" (how sure you are) are huge because you're just making educated guesses based on a model.

2. The "LaMET" Method: The "Long-Range Telescope"
Now, imagine you have a telescope that lets you see the steam rising all the way up into the sky (a longer distance, up to 1.0 fm or more).

• The Advantage: As the steam rises, it doesn't just disappear randomly; it follows the laws of physics. It fades away in a very specific, predictable pattern (exponential decay).
• The Strategy: You measure the steam where you can see it clearly. Then, you use the laws of physics (the "telescope rules") to predict exactly how the steam behaves in the part of the sky you can't quite see yet. You aren't guessing; you are extrapolating based on a solid rule.
• The Result: This is a Forward Problem. You start with the data and the rules, and you calculate the recipe step-by-step. It's much more reliable.

The Controversy: "Is the Telescope Broken?"

A recent paper (Ref. [1]) argued: "Hey, the steam gets really foggy and hard to see at the very top of the telescope (the 'sub-asymptotic' region). Because the data is noisy there, we can't trust the prediction. Maybe we should just treat this like the 'Blind Taste Test' (Inverse Problem) and use fancy math to guess the recipe, even if the math is shaky."

The Authors of This Paper Say: "No, that's a bad idea."

Here is their argument, broken down with analogies:

1. The "Foggy Top" isn't a dead end; it's just a challenge.

The critics say the data gets too noisy at long distances to trust. The authors agree the data can be noisy, but they argue that physics gives us a safety net.

• Analogy: Imagine you are walking in the fog. You can see the path clearly for 10 meters. After that, it's foggy. A critic says, "We can't see the path, so we should just guess where it goes!"
• The Authors' Reply: "No, we know the path is a straight line (physics). Even if the fog is thick, we know the path must continue straight. We can estimate the error of our guess based on how thick the fog is. We don't need to throw away the straight-line rule and start guessing randomly."

2. The "Guessing Game" (Inverse Problem) is dangerous.

The critics suggest using "data-driven" math (like Gaussian Processes) to fill in the gaps without strict physical rules.

• Analogy: This is like asking a computer to draw a picture of a cat based on a blurry photo. If you don't tell the computer "cats have fur and whiskers," it might draw a cat with wheels or a cat made of soup.
• The Authors' Reply: Without the strict rules of physics (the "cat must have fur" rule), the math can produce results that look okay but are physically impossible. This leads to unnecessarily huge errors because the computer is allowed to imagine wild possibilities that nature never actually does.

3. The "Recipe" is already there; we just need to read it carefully.

The authors show that even with "noisy" data, if you use the correct physical rules (exponential decay), the final recipe (the PDF) stays very stable.

• Analogy: If you are trying to hear a song through a wall, and the sound gets quieter, you don't need to invent a new song. You just know the music is fading out. If you try to "reconstruct" the song using only the loudest part and guess the rest, you might hear a completely different tune. But if you know the song fades out exponentially, you can hear the whole tune correctly, even with the static.

The Bottom Line

The paper concludes that LaMET is the superior method.

• The Critics' View: "The data at the edge is too messy, so we can't trust the calculation. Let's treat it like a guessing game."
• The Authors' View: "The data might be messy, but physics is not. By using the known laws of how particles behave (exponential decay), we can reliably predict the rest of the data. This gives us a much more accurate recipe with honest, controlled error bars. Turning this into a 'guessing game' (Inverse Problem) just makes the errors bigger and less trustworthy."

In short: Don't throw away the rulebook just because the page is a little smudged. Use the rulebook to figure out what's under the smudge.

Technical Summary

Here is a detailed technical summary of the paper "LaMET's Asymptotic Extrapolation vs. Inverse Problem" by Chen et al.

1. Problem Statement

The paper addresses a critical debate regarding the Large-Momentum Effective Theory (LaMET) framework used in Lattice QCD to calculate parton distribution functions (PDFs).

• The Controversy: A recent study (Ref. [1], arXiv:2504.17706) raised concerns that lattice data in the "sub-asymptotic" region ($0.7 \lesssim z \lesssim 1.0$fm) lacks sufficient precision to support a controlled extrapolation to infinite distance ($z \to \infty$). The authors of Ref. [1] argued that if the data is too noisy, the Fourier Transform (FT) required to obtain quasi-PDFs becomes an ill-posed Inverse Problem (IP), similar to Short-Distance Factorization (SDF) approaches (e.g., pseudo-PDFs), where uncertainties cannot be reliably quantified without arbitrary mathematical conditioning.
• The Core Question: Is LaMET fundamentally an ill-posed inverse problem due to data limitations, or is it a well-posed Forward Problem (FP) where uncertainties can be systematically controlled via physics-guided asymptotic extrapolation?

2. Methodology and Theoretical Framework

The authors employ a multi-faceted approach combining Effective Field Theory (EFT) arguments, dispersion relations, and numerical re-analysis of lattice data.

• LaMET as a Forward Problem: The authors reiterate that LaMET is an EFT expansion. It calculates PDFs at a fixed momentum fraction $x$ by expanding light-cone observables in terms of quasi-distributions (calculated at large momentum $P^z$). Unlike SDF, which relies on fitting limited short-distance correlations to a model, LaMET utilizes matrix elements at all $z$ (conceptually) to define momentum.
• Asymptotic Extrapolation vs. Inverse Problem:
  • Asymptotic Extrapolation: The authors argue that lattice data in the sub-asymptotic region can be extrapolated to $z \to \infty$ using physics-guided constraints. Specifically, color confinement dictates that spatial correlation functions decay exponentially at large distances. This decay is governed by a mass scale (related to heavy-light mesons) that is independent of the hadron's momentum.
  • Inverse Problem (IP) Critique: They define a true IP (like in SDF) as a situation where data is too sparse to determine a continuous function without arbitrary model assumptions. They argue that treating LaMET as an IP ignores the rigorous physical constraints (exponential decay) available.
• Numerical Re-analysis:
  • The authors re-analyze data from Ref. [1] (which used a "ratio scheme" and Gaussian/RBF kernels) and compare it with high-precision data from Ref. [111] (using a "hybrid scheme").
  • They test various extrapolation models: Exact Exponential, Radial Basis Function (RBF), and log-RBF within a Gaussian Process Regression (GPR) framework.
  • They derive an upper bound for the Fourier Transform model uncertainty based on the maximum value of the correlation function in the extrapolated region (Eq. 12).

3. Key Contributions

A. Reframing LaMET as a Forward Problem

The paper establishes that LaMET is not an inverse problem in the mathematical sense of ill-posedness.

• Physical Constraints: The exponential decay of correlation functions at large $z$ is a rigorous consequence of QCD (dispersion relations). This provides a "prior" that makes the extrapolation a well-posed problem.
• Systematic Improvement: Unlike SDF, where the accessible $z$ range is limited to $\sim 0.3$ fm (where perturbation theory breaks down), LaMET allows access to $z \sim 1.0$ fm and beyond, where the exponential tail dominates. This allows for a systematic expansion where errors can be reduced by improving data precision or including subleading terms in the EFT.

B. Critique of Data-Driven IP Methods

The authors demonstrate that data-driven methods (like GPR with RBF kernels) used in Ref. [1] are problematic:

• Violation of Physical Conditions: GPR methods often fail to satisfy the physical requirement that the correlation function must decay exponentially. They can produce unphysical oscillations or growth in the tail region, leading to artificially large uncertainties.
• Arbitrariness: The choice of hyperparameters in GPR is often ad hoc and lacks a clear physical basis, making the resulting error estimates unreliable and overly conservative.

C. Quantification of Uncertainty Bounds

A major technical contribution is the derivation of a rigorous upper bound for the FT uncertainty (Eq. 12):
$$\delta f(x, P, \lambda_L) < \frac{4N_x |h(z, P; \lambda_L)|_{\text{max}}}{\pi x}$$
This bound shows that the uncertainty is controlled by the magnitude of the correlation function at the truncation point ($z_L$). Because the function decays exponentially, this contribution is naturally suppressed, unlike in power-law decays (SDF) where the uncertainty would be much larger.

D. Re-analysis of Ref. [1] Data

• The authors show that the large discrepancies observed in Ref. [1] were largely due to the use of the ratio scheme (which removes the exponential decay feature) and insufficient statistics in the sub-asymptotic region.
• When applying the same extrapolation models to high-precision data (Ref. [111]), the variations between different models are negligible at moderate $x$, and the results are well-bounded by the theoretical error estimate.

4. Results

• Controlled FT: The Fourier Transform in LaMET is under control. The model uncertainty arising from the extrapolation of the long-range tail is bounded and systematically reducible.
• Superiority of Asymptotic Extrapolation: Physics-guided asymptotic extrapolation provides more reliable and realistic error estimates than data-driven IP methods. The latter often yield "unnecessarily conservative" errors because they lack physical constraints.
• Data Quality: While current lattice data in the sub-asymptotic region ($z \sim 0.7-1.0$ fm) can be noisy, it is not impossible to achieve the precision required for controlled extrapolation. Techniques like kinematically enhanced interpolating operators and increased statistics can resolve these issues.
• Rejection of IP Equivalence: The paper conclusively argues that LaMET and SDF are fundamentally different. SDF is an IP due to the lack of long-distance data; LaMET is an FP because the long-distance behavior is physically constrained and calculable.

5. Significance

• Validation of LaMET: The paper defends the LaMET framework as the most robust method for calculating PDFs, GPDs, and TMDs from first principles, resolving concerns that it suffers from the same fundamental limitations as other approaches.
• Methodological Guidance: It provides a clear roadmap for future lattice calculations:
  1. Prioritize high-precision data in the sub-asymptotic region ($z \sim 0.5 - 1.2$ fm).
  2. Use physics-motivated asymptotic forms (exponential decay) for extrapolation rather than purely mathematical smoothing.
  3. Quantify uncertainties using the derived upper bounds rather than relying on the spread of unphysical models.
• Impact on Phenomenology: By establishing that LaMET errors are controllable and quantifiable, the paper strengthens the reliability of lattice QCD inputs for global PDF fits and high-energy physics phenomenology, ensuring that theoretical uncertainties are not underestimated or overestimated due to methodological confusion.

In summary, the authors argue that the "inverse problem" in LaMET is a mischaracterization. With proper physical constraints and sufficient data precision, LaMET remains a systematic, forward-problem expansion capable of delivering precise, error-controlled parton distributions.