Sparse modeling study of extracting charmonium spectral functions from lattice QCD at finite temperature

This study demonstrates that sparse modeling can successfully reconstruct charmonium resonance peaks from lattice QCD data at finite temperatures, yielding results qualitatively consistent with the maximum entropy method, though it struggles to resolve transport peaks without additional assumptions.

The Gist

The Big Picture: Listening to the "Ghost" of a Particle

Imagine you are in a dark room, and someone is playing a musical instrument, but you can't see them. All you have is a recording of the sound waves hitting the wall (the Euclidean-time correlation function). Your goal is to figure out exactly what instrument is being played and what notes it is hitting (the spectral function).

In the world of particle physics, specifically inside a Quark-Gluon Plasma (QGP)—a super-hot soup of particles created in heavy-ion collisions—scientists want to know how heavy particles called charmonium (a pair of heavy quarks stuck together) behave. Do they stay stuck together like a solid rock, or do they melt apart like ice in a hot pan?

The problem is that the "recording" (the data from computer simulations called Lattice QCD) is very noisy and incomplete. It's like trying to guess the song from a recording that has been played through a wall, with static, and only for a few seconds. Reconstructing the original song from this bad recording is a classic "ill-posed" problem: there are too many possible songs that could fit the noise.

The New Tool: Sparse Modeling (SpM)

For years, scientists have used a method called the Maximum Entropy Method (MEM) to guess the song. It's like assuming the song is as "boring" or "smooth" as possible unless the data forces it to be complex.

This paper introduces a new detective tool called Sparse Modeling (SpM).

The Analogy of the Sparse Solution:
Imagine you are trying to guess a secret code.

• Old Method (MEM): Assumes the code is a long, random string of numbers. It tries to smooth out the noise but often misses sharp, distinct features.
• New Method (SpM): Assumes the code is sparse. This means the code is mostly zeros, with only a few important numbers standing out. It's like looking for a few specific needles in a haystack, rather than trying to describe every single piece of hay.

The authors argue that the "song" of the charmonium particle is likely simple: it's mostly silence (zero), with a few sharp, distinct notes (resonance peaks) representing the particle's existence. SpM looks for that simplicity.

The Experiment: Testing the Detective

Before using SpM on real physics data, the authors had to test if it actually works. They did this in two ways:

1. The "Mock Data" Test (The Fake Recording)
They created a fake, perfect "song" (a known spectral function) and then added static noise to it, just like real experiments have.

• The Good News: When the "song" had a clear, sharp note (a resonance peak), SpM could find it very well. It successfully reconstructed the shape of the particle.
• The Bad News: When the "song" had a very specific, tricky feature called a transport peak (which happens when particles are moving freely in the hot soup), SpM struggled. It couldn't resolve this feature clearly without making extra guesses. It's like trying to hear a whisper in a hurricane; the method needs more than just "simplicity" to hear that specific sound.

2. The "No-Song" Test
They also tested if SpM would "hallucinate" a song when there wasn't one. They fed it data that had no peaks at all.

• The Result: SpM didn't make up fake peaks. It correctly said, "I don't see any distinct notes here." This is crucial because it means the method is honest and doesn't invent physics that isn't there.

The Real Deal: Analyzing Lattice QCD Data

After passing the tests, they applied SpM to real data from supercomputer simulations of charmonium at two temperatures:

1. Cooler than the melting point ($0.73 T_c$): The particles should be stuck together.
2. Hotter than the melting point ($1.46 T_c$): The particles should be melting or melted.

What they found:

• The Melting: At the cooler temperature, they saw a clear "peak" (the particle is alive). At the hotter temperature, the peak got wider and shifted, indicating the particle is struggling to hold together or has melted. This matches what other methods (MEM) found, giving them confidence that SpM is working.
• The Missing Whisper: Just like in the mock tests, they still couldn't clearly see the "transport peak" (the signal of free-moving particles) in the hot data. The method is great at finding the "rock" (the bound particle) but struggles to find the "fog" (the transport signal) without extra help.

The Conclusion: A New, Honest Lens

The paper concludes that Sparse Modeling is a powerful new tool for understanding the hot soup of the early universe.

• Why it's great: It relies on fewer assumptions than older methods. It doesn't force the data to be smooth; it just looks for the simplest explanation (the fewest "notes").
• The Verdict: It successfully confirms that charmonium particles melt as the temperature rises. It gives us a clearer picture of the "resonance" (the particle itself).
• The Limitation: It still needs a little help to hear the "transport" signals (how particles move freely). To hear those, scientists might need to combine SpM with a few extra assumptions about how that movement looks.

In summary: The authors built a new, sharper pair of glasses (SpM) to look at the blurry, noisy photos of the universe's hottest moments. The glasses are excellent at spotting the heavy particles and confirming they melt, but they still have a little trouble seeing the fine details of how those particles move once they are free.

Technical Summary

1. Problem Statement

In lattice Quantum Chromodynamics (QCD), meson spectral functions $\rho(\omega)$, which encode real-time dynamics and transport properties of hot and dense matter, cannot be computed directly. Instead, they must be inferred from Euclidean-time correlation functions $G(\tau)$ via an integral equation (Lehmann representation). This reconstruction is a classic ill-posed inverse problem due to:

• Limited Data: The number of temporal data points ($N_\tau$) is small.
• Statistical Noise: Monte Carlo simulations introduce noise that is amplified during inversion.
• Non-uniqueness: Many different spectral functions can reproduce the same noisy correlation data.

Traditional methods like the Maximum Entropy Method (MEM) rely on specific prior assumptions (e.g., default models) which can bias results. The authors aim to investigate Sparse Modeling (SpM) as an alternative framework that relies primarily on the assumption of sparsity (that the solution has few non-zero components in a specific basis) rather than complex priors, to extract charmonium spectral functions at finite temperatures ($T < T_c$ and $T > T_c$).

2. Methodology: Sparse Modeling (SpM)

The study employs SpM within the Intermediate Representation (IR) basis, a technique originally developed for condensed matter physics.

• Formalism:
  • The integral equation $G(\tau) = \int K(\tau, \omega)\rho(\omega)d\omega$ is discretized into a matrix equation $\vec{G} = K\vec{\rho}$.
  • The kernel $K$ is decomposed using Singular Value Decomposition (SVD): $K = USV^T$.
  • The problem is transformed into the IR basis ($\vec{G}' = U^T\vec{G}$, $\vec{\rho}' = V^T\vec{\rho}$), where the singular values $s_l$ decay exponentially.
  • Dimensionality Reduction: Components corresponding to very small singular values (where noise dominates) are truncated.
  • L1 Regularization (LASSO): To enforce sparsity, an L1-norm penalty term is added to the $\chi^2$ cost function:
    $$F(\vec{\rho}' | \vec{G}', \lambda) = \frac{1}{2}||\vec{G}' - S\vec{\rho}'||_2^2 + \lambda||\vec{\rho}'||_1$$
    This forces most components of the solution to be exactly zero, leaving only the significant features.
• Covariance Handling: Unlike previous SpM applications in lattice QCD, this study explicitly incorporates the covariance matrix of the correlation functions (arising from correlations between different Euclidean times) by transforming the kernel and data using the Cholesky decomposition of the inverse covariance matrix.
• Kernel Adaptation:
  • $T < T_c$: Uses a kernel $K^{(1)}$ designed to extract $\rho(\omega)/\omega^2$, ensuring $\rho(0)=0$ (no transport peak) and avoiding artificial singularities at low frequency.
  • $T > T_c$: Uses a kernel $K^{(0)}$ to extract $\rho(\omega)$ directly, allowing for the possibility of a transport peak at $\omega \to 0$.
• Optimization: The minimization is solved using the Alternating Direction Method of Multipliers (ADMM) algorithm, subject to non-negativity constraints ($\rho \ge 0$).

3. Key Contributions

1. Extension of SpM to Charmonium: This is one of the first applications of SpM to meson spectral functions at finite temperature, moving beyond its previous use in energy-momentum tensor correlations.
2. Covariance Integration: The authors explicitly include the full covariance matrix of the lattice data in the SpM formalism, improving the statistical robustness of the reconstruction.
3. Systematic Benchmarking: The study rigorously tests the method using mock data before applying it to real lattice QCD data. This includes:
   • Testing dependence on data points ($N_\tau$) and noise levels ($\epsilon$).
   • Verifying the method's ability to reconstruct sharp resonance peaks vs. broad transport peaks.
   • Validating that the method does not generate spurious peaks when none exist (using free spectral functions).
4. Kernel Selection Strategy: The paper demonstrates the necessity of choosing specific kernel forms ($K^{(0)}$ vs. $K^{(1)}$) depending on the temperature regime to correctly handle boundary conditions at $\omega=0$.

4. Results

A. Mock Data Analysis

• Resonance Peaks ($T < T_c$): SpM successfully reconstructs the position of resonance peaks (e.g., $J/\psi$). The accuracy improves with higher $N_\tau$ and lower noise. However, sharp peaks are often reconstructed as slightly broader structures because SpM favors solutions with moderate width (requiring fewer high-frequency modes).
• Transport Peaks ($T > T_c$): The method struggles to resolve the sharp transport peak at $\omega \to 0$ without additional assumptions. While the method captures the general shape, the specific value at $\omega=0$ is sensitive to noise and does not always converge to the input value, even with reduced noise.
• False Positives: When tested with "free" spectral functions (no peaks), SpM correctly produced smooth functions without generating spurious peaks, confirming the method's reliability in avoiding overfitting when the data lacks structure.

B. Lattice QCD Data Application

The authors applied SpM to charmonium correlation functions in the pseudoscalar (PS) and vector (VC) channels at $T \simeq 0.73 T_c$ and $T \simeq 1.46 T_c$.

• Below Critical Temperature ($T \simeq 0.73 T_c$):

  • A well-defined resonance peak was observed in both channels.
  • Peak Positions: PS channel $\approx 4.02$ GeV; VC channel $\approx 4.30$ GeV.
  • Comparison: These values are slightly higher than those obtained via MEM in previous studies (3.31 GeV and 3.48 GeV, respectively), but the qualitative hierarchy (PS < VC) and the existence of a bound state are consistent.
  • Using the $K^{(1)}$ kernel ensured $\rho(0)=0$ and provided stable results at low frequencies.

• Above Critical Temperature ($T \simeq 1.46 T_c$):

  • The resonance peaks became broader and shifted to higher energies (PS $\approx 4.87$ GeV, VC $\approx 5.42$ GeV), indicating the melting of charmonium states.
  • Transport Peak: In the vector channel, a transport peak is theoretically expected. However, SpM did not clearly resolve a distinct peak at $\omega \to 0$. The spectral function remained near zero or fluctuated slightly, suggesting that sparsity alone is insufficient to resolve the transport peak without further physical assumptions about its shape.

5. Significance and Conclusion

• Validation of SpM: The study confirms that SpM is a viable, model-independent tool for extracting spectral functions from lattice QCD. It successfully captures the essential physics (resonance existence, channel hierarchy, and broadening with temperature) without relying on the default models required by MEM.
• Limitations: The primary limitation identified is the difficulty in resolving sharp features (like narrow transport peaks or very sharp resonances) solely based on sparsity. The method tends to smooth out abrupt changes in value.
• Future Directions: To accurately extract transport coefficients (like heavy quark diffusion) or resolve transport peaks, the authors suggest that SpM must be combined with additional physical assumptions regarding the shape of the low-frequency behavior, rather than relying on sparsity alone.

In summary, this paper establishes Sparse Modeling as a robust complementary method to MEM for studying heavy-quark observables in the Quark-Gluon Plasma, providing a new perspective on charmonium survival and melting while highlighting the specific challenges in resolving transport phenomena.