Bayesian approach for many-body uncertainties in nuclear structure: Many-body perturbation theory for finite nuclei

This paper introduces a Bayesian framework to systematically quantify many-body truncation uncertainties in *ab initio* calculations of finite nuclei using many-body perturbation theory and chiral effective field theory interactions, thereby advancing the comprehensive assessment of theoretical errors in nuclear structure research.

The Gist

Imagine you are trying to predict the weather for a specific city next week. You have a supercomputer model that simulates the atmosphere. But you know the model isn't perfect. It might miss a small cloud here or a breeze there. To be a responsible forecaster, you don't just say, "It will be 72°F." You say, "It will be 72°F, plus or minus 3 degrees."

This paper is about doing exactly that for atomic nuclei (the tiny cores of atoms), but instead of weather, they are predicting how heavy and stable these nuclei are.

Here is the breakdown of their work using simple analogies:

1. The Problem: The "Recipe" is Too Complicated

Physicists have a "recipe" (called Chiral Effective Field Theory) to describe how protons and neutrons stick together. It's like a cookbook with infinite pages.

• The Issue: You can't read the whole infinite cookbook to cook a meal. You have to stop at a certain page.
• The Consequence: If you stop too early, your cake might be raw. If you stop a bit later, it might be perfect. But how do you know how "raw" your cake is?

For decades, physicists have been good at estimating the error from the ingredients (the nuclear forces). But they haven't been very good at estimating the error from stopping the recipe early (the math used to solve the problem). They usually just guessed, saying, "I think we're close enough," based on experience.

2. The Solution: A "Bayesian" Weather Forecast

The authors (I. Svensson and colleagues) decided to stop guessing and start using Bayesian Statistics. Think of this as a "smart learning machine."

Instead of just guessing the error, they built a model that asks: "Based on the pattern of the first few steps of the recipe, how likely is it that the next steps will change the result significantly?"

They used a method called Many-Body Perturbation Theory (MBPT). Imagine this as peeling an onion:

• Layer 1 (Hartree-Fock): The basic shape of the onion.
• Layer 2: A small correction to the shape.
• Layer 3: A tiny tweak to the shape.
• Layer 4: A microscopic adjustment.

The goal is to see how much each new layer changes the final size of the onion.

3. The "Magic" of the Study

The team looked at a bunch of different atomic nuclei (from Oxygen to Lead). They calculated the first few layers of the onion (up to the 3rd layer) and then used their "smart learning machine" to predict the uncertainty.

Here is the cool part they found:

• The "Soft" Interaction: When they used a "soft" nuclear force (like a gentle, squishy interaction), the layers got smaller and smaller very quickly. It was like peeling an onion where each layer is half the size of the previous one. The model was very confident, and the "error bars" (the uncertainty range) were tiny.
• The "Hard" Interaction: When they used a "hard" nuclear force (like a stiff, rigid interaction), the layers didn't shrink as fast. The model realized, "Hey, the next layer might be huge!" So, it gave a much wider "error bar" to warn the scientists: "Be careful, the answer might change a lot if we calculate more layers."

4. Why This Matters

Think of nuclear physics as building a bridge.

• Before: Engineers would build the bridge and say, "It should hold 10 tons." They didn't have a mathematical way to say, "There is a 90% chance it holds between 8 and 12 tons."
• Now: This paper gives them a tool to say, "Based on our math, we are 90% sure the bridge holds between 8 and 12 tons."

This is crucial for:

• Nuclear Energy: Knowing exactly how stable a nucleus is helps design better reactors.
• Neutron Stars: These are giant balls of neutrons. To understand them, we need to know how nuclear forces behave under extreme pressure.
• New Physics: If our predictions are wrong, it might mean there is new, undiscovered physics hiding in the errors.

5. The "Gotcha" (Limitations)

The authors are honest about the limits.

• Correlations: They noted that their data points (different nuclei) are related to each other, like siblings. Their model treated them as independent for now, which is a simplification.
• The "Hard" Case: For very stiff nuclear forces, the math gets messy and might even diverge (explode). Their model can detect this and say, "Warning! The math is breaking down!" but it needs more data to be precise in those extreme cases.

Summary

This paper is a quality control manual for nuclear physics. It moves the field from "We think this is right" to "We know this is right within this specific range of error." By using a statistical "smart guess" system, they can now tell us exactly how much we can trust our calculations of the atomic nucleus, paving the way for more reliable predictions in nuclear energy and astrophysics.

Technical Summary

1. Problem Statement

In ab initio nuclear structure calculations, quantifying theoretical uncertainties is critical for reliable predictions, especially as research pushes toward heavy nuclei and extreme conditions. While uncertainties arising from the truncation of the nuclear interaction (via Chiral Effective Field Theory, EFT) have been systematically addressed using Bayesian methods, uncertainties stemming from the many-body expansion (the method used to solve the Schrödinger equation) have historically relied on expert assessment rather than rigorous statistical modeling.

The specific challenge addressed in this work is the quantification of truncation errors in Many-Body Perturbation Theory (MBPT) applied to finite nuclei. Current MBPT applications are often limited to third or fourth order, and without a robust error model, it is difficult to determine if the perturbative series has converged or if higher-order corrections might significantly alter the results.

2. Methodology

The authors develop a Bayesian statistical framework to model the truncation error of MBPT expansions. The methodology involves the following key components:

• Error Model Formulation:

  • The MBPT ground-state energy is modeled as a power series expansion: $E_0 = E_{\text{ref}} \sum \gamma_i R^i$.
  • $E_{\text{ref}}$ is a scaling factor (set to $-8A$ MeV, representing typical binding energy).
  • $R$ is a convergence parameter representing the "softness" of the interaction (the ratio of successive corrections).
  • $\gamma_i$ are expansion coefficients assumed to be drawn from a normal distribution with zero mean and variance $\bar{\gamma}^2$.
  • The truncation error $\delta E_0$ is modeled as the sum of neglected higher-order terms, governed by the parameters $R$ and $\bar{\gamma}$.

• Bayesian Inference:

  • The authors infer the hyperparameters $R$ and $\bar{\gamma}^2$ using Markov-chain Monte Carlo (MCMC) sampling.
  • Priors: An inverse-gamma prior is used for $\bar{\gamma}^2$ (encoding the belief that coefficients are "naturally sized"), and a uniform prior is used for $R$ (allowing for the possibility of divergent series, i.e., $R > 1$).
  • Data: The inference is trained on MBPT results (orders 0, 1, 2, and 3) for a small subset of closed-shell nuclei ($^{16}\text{O}$, $^{48}\text{Ca}$, $^{132}\text{Sn}$) to avoid artificially narrowing uncertainties due to strong correlations between nuclei.

• Predictive Distribution:

  • Once the posterior distributions for $R$ and $\bar{\gamma}^2$ are established, the framework generates a Posterior Predictive Distribution (PPD) for the ground-state energy. This provides credibility intervals (e.g., 68% and 90%) for the true energy value based on the truncated MBPT calculation.

3. Key Contributions

• First Systematic UQ for MBPT: This work provides the first systematic, Bayesian-based quantification of many-body truncation uncertainties for finite nuclei, moving beyond heuristic expert estimates.
• Robust Error Model: The model successfully captures the size-extensive nature of MBPT corrections, demonstrating that uncertainty scales consistently with mass number ($A$).
• Interaction Sensitivity Analysis: The framework is applied to multiple chiral interactions (1.8/2.0 EM, $\Delta$N2LO, 1.8/2.0 EM7.5, and N2LOsat), revealing how the "softness" of the interaction directly influences the convergence rate and uncertainty magnitude.
• Validation against Non-Perturbative Methods: The uncertainty bands are validated against results from the In-Medium Similarity Renormalization Group (IMSRG), a non-perturbative many-body method, serving as a high-accuracy benchmark.

4. Key Results

• Convergence Behavior: For "soft" interactions like 1.8/2.0 (EM), the inferred convergence parameter is $R \approx 0.15$, indicating rapid convergence. The uncertainty bands for third-order MBPT (MBPT(3)) are small (below 0.2 MeV per nucleon for heavy nuclei) and successfully encompass the IMSRG reference values within 90% credibility intervals.
• Interaction Dependence:
  • 1.8/2.0 (EM): Shows the best perturbative behavior with narrow uncertainty bands.
  • $\Delta$N2LOGO and 1.8/2.0 (EM7.5): These interactions yield wider uncertainty bands (roughly double the width of the EM case), indicating slower convergence.
  • N2LOsat: This "harder" interaction shows a poor reference state (Hartree-Fock) and a complex convergence pattern where accidental cancellations occur. The model correctly identifies this as having larger uncertainties (bands >2x wider than EM).
• Empirical Coverage: The "weather plots" (empirical coverage analysis) show that the uncertainty estimates are well-calibrated for third-order results but slightly conservative for second-order results.
• Nuclear Matter: Including symmetric nuclear matter in the inference had a negligible effect on the uncertainty bands for finite nuclei. However, pure neutron matter (PNM) showed a qualitatively different convergence pattern, suggesting the current error model may need refinement to handle PNM simultaneously with finite nuclei.
• Divergence Handling: The model successfully handles cases of apparent divergence (e.g., a fictitious interaction with $R > 1$), resulting in infinitely wide uncertainty bands, correctly signaling the breakdown of the perturbative approach.

5. Significance

This work represents a major step toward complete uncertainty quantification (UQ) in ab initio nuclear physics. By providing a rigorous statistical framework for many-body truncation errors, the authors enable:

1. Reliable Predictions: Researchers can now assign credible error bars to MBPT calculations, essential for interpreting experimental data and guiding future experiments.
2. Method Selection: The framework helps identify when MBPT is sufficient (soft interactions) and when non-perturbative methods (like Coupled Cluster or IMSRG) are strictly necessary (hard interactions).
3. Future Extensions: The methodology lays the groundwork for extending UQ to more complex many-body expansions (e.g., Coupled Cluster, IMSRG) and open-shell nuclei, as well as for combining many-body uncertainties with interaction uncertainties for a total error budget.

In summary, the paper transforms the assessment of many-body uncertainties from an art based on experience into a science based on statistical inference, significantly enhancing the reliability of nuclear structure theory.