A kernel-derived orthogonal basis for spectral functions from Euclidean correlators

This paper proposes a systematic, prior-free framework that constructs an orthogonal functional basis from Euclidean correlators to extract robust constraints and global structures of spectral functions, serving as a complementary preprocessing tool for existing reconstruction techniques rather than a direct reconstruction method.

The Gist

Imagine you are trying to figure out what a complex, hidden machine looks like inside, but you can only see the shadows it casts on a wall.

In the world of quantum physics, scientists study particles and forces using a "shadow" called a Euclidean correlator. This is a mathematical curve they can calculate on supercomputers. However, the real information they want—the spectral function (which tells them how particles move, decay, and interact)—is hidden inside that curve.

The problem is that figuring out the machine from the shadow is like trying to guess the exact shape of a cloud just by looking at its silhouette. It's a "broken" puzzle: many different shapes could cast the same shadow, and tiny errors in the shadow make the guess wildly wrong. This is known as an ill-posed inverse problem.

The Paper's Big Idea: A New Way to Look at the Shadow

Most scientists try to solve this by guessing the shape of the machine and tweaking it until the shadow matches. This paper, however, suggests a different approach. Instead of trying to reconstruct the entire machine perfectly, the authors propose a tool to extract specific, reliable facts about the machine directly from the shadow.

They do this by creating a special set of "measurement tools" (called an orthogonal basis).

The Analogy: The "Sieve" and the "Smoothie"

Think of the spectral function (the hidden machine) as a smoothie made of many different fruits (frequencies).

• The Problem: You have a cup of this smoothie, but you can't see the fruits inside. You only have a blurry photo of the cup.
• The Old Way: You try to guess the recipe by saying, "I think it's 50% apple, 30% banana, and 20% strawberry," and then checking if the photo matches. If your guess is slightly off, the whole recipe fails.
• The New Way (This Paper): Instead of guessing the whole recipe, you use a set of special sieves.
  • Sieve #1 catches only the "heavy" fruits.
  • Sieve #2 catches only the "light" fruits.
  • Sieve #3 catches fruits in the middle.

By running the smoothie through these sieves, you don't get the whole picture back, but you get very precise numbers: "There are exactly 3 heavy fruits and 5 light fruits."

How It Works (The Magic Trick)

The authors realized that the mathematical formula connecting the shadow (Euclidean correlator) to the smoothie (spectral function) has a secret structure.

1. The "Sieves" are built from the Shadow itself: They take the shadow, do some math operations (like cutting it into pieces and stretching it), and turn those pieces into the "sieves" (basis functions).
2. They are "Orthogonal": This is a fancy math word meaning the sieves don't overlap. Sieve #1 doesn't catch what Sieve #2 catches. This makes the measurements independent and clean.
3. The Result: They can calculate the "weight" of the smoothie in each sieve directly from the shadow data, without needing to guess the recipe first.

What Did They Find?

They tested this method on three different "fake" smoothies (model spectral functions):

• The Smooth One: For a smooth, simple smoothie, their method worked incredibly well. It could reconstruct the whole smoothie almost perfectly using just a few sieves.
• The Bumpy One: For a smoothie with some bumps and peaks, it worked very well for the big picture and for measuring the "heavy" parts (which correspond to important physics like transport coefficients—how easily heat or electricity flows).
• The Chaotic One: For a smoothie with wild, rapid spikes, the method couldn't capture the tiny, jagged details. However, it was still very good at telling them the "average" weight and the heavy parts.

The Catch (Why It's Not a Magic Wand Yet)

The paper admits a major limitation: Precision.

To get the "sieve" numbers right, the input shadow (the Euclidean correlator) must be calculated with extreme precision. If the shadow has even a tiny bit of "noise" (statistical error), the math gets unstable, and the numbers blow up.

Think of it like trying to balance a house of cards on a shaking table. The method is brilliant, but if your data isn't perfect, the house collapses.

The Bottom Line

This paper doesn't offer a way to instantly see the hidden machine. Instead, it offers a powerful diagnostic tool.

• It's a "Pre-Processor": Before scientists try to reconstruct the whole image, they can use this tool to check: "Does my guess match the basic rules of the shadow?"
• It's a "Safety Net": It gives them hard, mathematical constraints that any good theory must obey.
• It's a "Low-Energy Lens": It is particularly good at measuring the slow, heavy movements of particles (transport), which are crucial for understanding things like the behavior of the early universe or super-hot matter in particle colliders.

In short, the authors built a new set of mathematical rulers that measure the hidden world directly from the shadows, helping physicists get a clearer, more reliable picture of the quantum universe, even if they can't see the whole picture at once.

Technical Summary

1. Problem Statement

In lattice Quantum Chromodynamics (QCD) and many-body physics, essential dynamical information (such as particle excitation spectra, decay rates, and transport coefficients) is encoded in spectral functions ($\rho$). However, these are not directly accessible; they must be inferred from Euclidean correlation functions ($\Pi_E$) computed in lattice simulations.

The relationship between the two is an integral transform with a smooth kernel:
$$\Pi_E(\tau) = \int_0^\infty d\omega \, K(\tau, \omega) \rho(\omega)$$
Inverting this equation to recover $\rho(\omega)$ from discrete, noisy lattice data is a classic ill-posed inverse problem. Existing reconstruction techniques (e.g., Maximum Entropy, Bayesian inference, Backus-Gilbert) often rely on strong priors, regularization assumptions, or specific ansätze, which can bias the results. There is a need for systematic, prior-free methods to extract robust constraints and global features of spectral functions without assuming their specific shape.

2. Methodology

The author proposes a systematic, prior-free framework that constructs an orthogonal functional basis directly from the kernel of the Euclidean correlator. The methodology proceeds in three main steps:

A. Derivation of Constraints and Basis Functions

Instead of attempting a direct inversion, the author differentiates the Euclidean correlator $\Pi_E(\tau)$ with respect to Euclidean time $\tau$ and integrates over restricted time intervals $[l_m, u_m]$.

• By performing $n$-th order differentiation and integration, a set of constraints $C_n^{(m)}$ is derived.
• These constraints relate to weighted averages of the spectral function divided by frequency ($\rho/\omega$):
  $$C_n^{(m)} = \int_0^\infty d\omega \, \frac{\rho(\omega)}{\omega} S_n^{(m)}(\omega)$$
• The weight functions $S_n^{(m)}(\omega)$ (the basis functions) are analytically determined solely by the kernel of the integral transform and the chosen integration limits. They decay exponentially at large frequencies, ensuring convergence.

B. Construction of an Orthogonal Basis

The raw basis functions $S_n^{(m)}$ are not orthogonal. To create a stable expansion, the author reorganizes them into an orthogonal basis $\{Q_i(\omega)\}$ using a Gram-Schmidt-like procedure (matrix diagonalization):

1. Compute the overlap matrix $X_{kl} = \int S_k^* S_l$.
2. Diagonalize $X$ to find eigenvalues $\lambda_i$ and eigenvectors.
3. Construct the orthogonal basis $Q_i$ via a transformation matrix $M$.

C. Spectral Expansion

The spectral function (specifically $\rho/\omega$) is approximated as a linear expansion in this orthogonal basis:
$$\left( \frac{\rho(\omega)}{\omega} \right)_{\text{approx}} = \sum_i c_i Q_i(\omega)$$
Crucially, the expansion coefficients $c_i$ are directly calculable from the lattice data (the constraints $C_n^{(m)}$) without introducing priors:
$$c_i = \sum_j M_{ji} C_j$$
This allows the spectral function to be reconstructed as a controlled expansion where the coefficients are fixed by the data.

3. Key Contributions

• Prior-Free Framework: The method extracts global constraints and structures from Euclidean correlators without assuming a specific spectral ansatz or using Bayesian priors.
• Kernel-Derived Orthogonal Basis: It introduces a novel mathematical tool: an orthogonal basis derived strictly from the properties of the Euclidean kernel, tailored to the specific physics of the correlator.
• Diagnostic Tool: The paper positions this method not as a standalone reconstruction algorithm for high-resolution details, but as a preprocessing or diagnostic tool to test spectral ansätze, generate physical constraints, and provide stable inputs for more complex reconstruction methods.
• Analytical Insight: It provides explicit analytical forms for the basis functions and constraints, demonstrating how differentiation and integration of the correlator map to specific features in the spectral domain.

4. Results

The author tested the framework on three distinct model spectral functions under idealized conditions (assuming perfect knowledge of the Euclidean correlator):

• Model 1 (Smooth, single peak): The approximation using only three basis functions ($n=0$) successfully reproduced the input spectral function with high accuracy. The low-energy transport coefficient ($\Gamma = \lim_{\omega \to 0} \rho/\omega$) was reproduced with better than 2% accuracy.
• Model 2 (Multi-peak structure): Using nine basis functions ($n \leq 2$), the method achieved a very successful approximation, outperforming several other methods (MaxEnt, SVD, Padé) compared in literature for this specific model. The transport coefficient was reproduced within 0.3% for $n \leq 2$.
• Model 3 (Rapid fluctuations): For a model with rapid oscillations, the method struggled to reproduce fine details even with 15 basis functions ($n \leq 4$). This is expected as the basis functions are smooth and cannot capture high-frequency oscillations. However, the transport coefficient was still reproduced within 8% using only $n=0$ constraints.

Numerical Challenges:
The study identified a significant numerical challenge: the eigenvalues of the overlap matrix decay exponentially. This leads to a situation where the expansion coefficients involve the cancellation of very large numbers ($O(10^2)$ to $O(10^3)$). Consequently, the method requires extremely high-precision input for the Euclidean correlator to avoid rounding errors, making direct application to noisy lattice data difficult without further stabilization.

5. Significance and Future Outlook

• Complementary Role: The framework serves as a robust "sanity check" or preprocessing step. It can validate whether a proposed spectral ansatz is consistent with the Euclidean data or provide a set of physically motivated constraints to feed into Bayesian or regularized inversion methods.
• Low-Energy Physics: The method is particularly effective at extracting low-energy quantities (transport coefficients), which are often the primary goal in lattice QCD studies of quark-gluon plasma and heavy quark diffusion.
• Future Work: The author notes that applying this to realistic lattice data (with finite statistics, lattice spacing artifacts, and limited time resolution) is the next critical step. Future research will focus on optimizing integration ranges and strategies to mitigate the numerical instability caused by vanishing eigenvalues.

In summary, Yamada provides a rigorous mathematical bridge between Euclidean correlators and spectral functions, offering a new way to extract global structural information and transport coefficients without relying on subjective priors, while highlighting the inherent precision requirements for its practical implementation.