🔭 astrophysics

A comparison of pendulum models for large-amplitude longitudinal prominence oscillations

This study employs a Bayesian approach to compare original and extended pendulum models for large-amplitude longitudinal prominence oscillations, finding that while the extended model is slightly more plausible across all observed periods, the evidence is insufficient to definitively favor one model, thus recommending the use of model-averaged posteriors for inferring magnetic field strength.

Original authors: Iñigo Arregui

Published 2026-04-02

📖 4 min read☕ Coffee break read

CC BY 4.0

Original authors: Iñigo Arregui

Original paper licensed under CC BY 4.0 (http://creativecommons.org/licenses/by/4.0/). ✨ This is an AI-generated explanation of the paper below. It is not written or endorsed by the authors. For technical accuracy, refer to the original paper. Read full disclaimer

Imagine the Sun is a giant, bubbling pot of soup, but instead of vegetables, it's made of super-hot, electrically charged gas called plasma. Floating on the surface of this soup are massive, dark clouds of cooler gas called solar prominences. Think of these prominences as giant, glowing bridges or arches made of magnetic "spaghetti" holding the gas in place.

Sometimes, these magnetic bridges get shaken by solar storms, causing the gas to swing back and forth like a giant pendulum. This is what scientists call a Large-Amplitude Longitudinal Oscillation (LALO).

The Problem: Two Different Swings

For years, scientists have used a simple rule (a "model") to figure out how strong the magnetic "spaghetti" is, just by watching how long it takes for the gas to swing back and forth. This rule is called the Pendulum Model.

Think of it like this: If you see a child on a swing, you can guess how long the chains are based on how fast they swing.

The Old Model (M1): This model assumes gravity is the same everywhere, like a flat floor. It's a simple, easy-to-use rule.
The New Model (M2): Scientists realized that on the Sun, gravity isn't perfectly flat; it curves slightly because the Sun is a giant ball. This new model adds a tiny correction for that curve, like realizing the swing is actually on a giant, curved trampoline.

The paper asks: Does this tiny correction actually change our answer about how strong the magnetic field is?

The Experiment: A Bayesian Detective Story

The author, Iñigo Arregui, didn't just pick one model and hope for the best. Instead, he acted like a detective using a special toolkit called Bayesian Statistics.

Imagine you are trying to guess the weight of a mystery box.

The Clues: You have the swinging time (the period).
The Guesses: You have two theories (Old Model vs. New Model).
The Uncertainty: You know your measurements aren't perfect (the swing might be wobbly).

The author ran a simulation where he asked: "If the swing takes 30 minutes, 60 minutes, or 120 minutes, what does each model say the magnetic strength is?"

The Findings:

For short swings (under an hour): Both models agree. They are like two friends who both guess the box weighs 10 pounds.
For long swings (over an hour): The models start to disagree. The "New Model" (with the curved gravity correction) says the magnetic field must be stronger to hold the gas up for that long. The "Old Model" thinks it can be weaker.
The Limit: The New Model has a hard stop. It says, "If the swing takes longer than 167 minutes, it's impossible for the magnetic field to hold the gas up at all." The Old Model doesn't have this limit; it thinks the gas could swing forever if the magnet is strong enough.

The Verdict: Who Wins?

The author used math to calculate which model is more likely to be the "truth" given the data we have.

The Result: The New Model (M2) is slightly better. It fits the observed data a bit more often than the Old Model.
The Catch: The difference isn't huge. It's like flipping a coin 10 times and getting 6 heads and 4 tails. You might suspect the coin is slightly weighted, but you can't be 100% sure yet. The evidence is "positive" but not "strong."

The Solution: Don't Pick a Side!

Since neither model is clearly the winner, the author suggests a clever solution: Model Averaging.

Instead of saying, "I choose the New Model," the author says, "Let's take a weighted average of both."

Imagine you have two weather forecasters. One says "Sunny," the other says "Cloudy." Neither is 100% sure.
Instead of picking one, you say, "There's a 60% chance of clouds and a 40% chance of sun."
This gives you a result that accounts for both possibilities and the uncertainty in the data.

The Bottom Line

This paper is a lesson in scientific humility. Even when we have better physics (like accounting for the Sun's curved gravity), the data we have right now isn't perfect enough to say, "The old way is wrong."

So, the best way to understand the Sun's magnetic strength right now is to listen to both models and combine their answers. It's a reminder that in science, sometimes the most accurate answer is the one that admits we aren't entirely sure yet.

1. Problem Statement

Solar prominences exhibit Large-Amplitude Longitudinal Oscillations (LALOs), which serve as a diagnostic tool for "prominence seismology" to infer physical parameters such as magnetic field strength ( $B$ ) and plasma density ( $n$ ).

The Core Issue: Two competing theoretical models exist to describe the restoring force (projected gravity) driving these oscillations:
1. Original Model ( $M_1$ ): Proposed by Luna & Karpen (2012), it assumes uniform gravity and semi-circular flux tube geometry. It predicts that oscillation periods can grow unbounded as magnetic field strength increases.
2. Extended Model ( $M_2$ ): Proposed by Luna et al. (2022), it relaxes the uniform gravity assumption, incorporating non-uniform solar gravity and various flux-tube geometries. This introduces a physical "cut-off" period ( $P_\odot \approx 167$ min) beyond which gravity cannot be supported, regardless of magnetic field strength.
The Question: How do the differences between these models impact the Bayesian inference of the minimum magnetic field strength derived from observed oscillation periods? Furthermore, which model is statistically more plausible given observational data and uncertainties?

2. Methodology

The study employs a rigorous Bayesian framework to address parameter inference and model comparison.

Models Defined:
- $M_1$ (Original): Period $P$ is a function of $B$ and $n$ derived from uniform gravity assumptions (Eq. 4).
- $M_2$ (Extended): Period $P$ is derived using an equivalent radius of curvature ( $R_{eq}$ ) that accounts for solar surface curvature and non-uniform gravity (Eq. 8). This model imposes a hard upper limit on the period.
Inference Process:
- Parameters: The unknown vector is $\theta = \{B, n\}$ .
- Data: Observed periods $D = \{P\}$ with uncertainty $\sigma_P$ .
- Priors: Two types were tested:
  1. Uniform Priors: Assuming $B \in [1, 100]$ G and $n \in [10^9, 10^{11}]$ cm $^{-3}$ .
  2. Gaussian Priors: Using observational estimates for density ( $n$ ) with specific means and standard deviations, while keeping $B$ uniform.
- Likelihood: A Gaussian likelihood function was used to measure the distance between observed data and model predictions.
Model Comparison Metrics:
- Marginal Likelihood (Evidence): $p(D|M)$ , calculated by integrating the likelihood over the entire parameter space.
- Bayes Factor ( $BF_{21}$ ): The ratio of evidences between $M_2$ and $M_1$ . Values $>2$ indicate positive evidence; $>6$ strong evidence.
- Model Averaging: Since neither model was definitively proven superior, the authors computed model-averaged posteriors ( $p_{avg}$ ), weighting the individual model posteriors by their respective probabilities.

3. Key Contributions

Quantification of Model Discrepancies: The study explicitly maps how the assumption of non-uniform gravity alters the relationship between period, magnetic field, and density. It demonstrates that $M_2$ predicts higher minimum magnetic fields than $M_1$ for the same observed period, with the divergence increasing as the period lengthens.
Bayesian Model Comparison: It provides the first quantitative assessment of the plausibility of the extended pendulum model versus the original using Bayes factors, moving beyond qualitative comparisons.
Proposal of Model Averaging: The paper advocates for Bayesian Model Averaging (BMA) as the most robust solution for prominence seismology when competing physical models exist but neither has overwhelming statistical evidence to rule out the other.

4. Key Results

Parameter Inference:
- The Bayesian approach yields well-constrained posteriors for the minimum magnetic field strength ( $B$ ), even when plasma density is poorly constrained.
- Effect of Period: As the oscillation period increases, the inferred $B$ increases for both models.
- Model Difference: For short periods ( $<60$ min), $M_1$ and $M_2$ yield similar results. For long periods ( $>90$ min), $M_2$ infers significantly higher magnetic field strengths. For example, for a 120 min period, $M_1$ infers $B \approx 40$ G, while $M_2$ infers $B \approx 60$ G (under uniform priors).
Model Comparison (Evidence):
- Marginal Likelihood: The extended model ( $M_2$ ) consistently yields a higher marginal likelihood than $M_1$ across the entire range of observed periods (10–120 min).
- Bayes Factors: The Bayes factor ( $BF_{21}$ ) is positive, indicating $M_2$ is more plausible. However, the magnitude is generally below 2 (the threshold for "positive evidence") for observed periods. The threshold of $BF_{21} > 2$ is only reached for theoretical periods $\ge 118$ min, which are rarely observed.
Case Study (Zhang et al. 2017):
- For an observed event with $P = 99$ min, $M_1$ inferred $B \approx 28$ G, while $M_2$ inferred $B \approx 35$ G.
- The model-averaged posterior resulted in $B \approx 32$ G, with model probabilities of $p(M_1) \approx 0.34$ and $p(M_2) \approx 0.66$ .

5. Significance and Conclusion

Physical Implications: The inclusion of non-uniform gravity ( $M_2$ ) is physically more realistic and introduces a necessary cut-off period. While the statistical evidence favoring $M_2$ is not yet "strong" (due to observational uncertainties and the specific range of periods observed), the model is consistently more plausible.
Methodological Shift: The paper argues that relying on a single "best" model is insufficient when the Bayes factors are inconclusive. Instead, model-averaged posteriors should be adopted. This approach naturally incorporates the uncertainty regarding the physical model itself, providing a more conservative and comprehensive estimate of magnetic field strength.
Future Outlook: As observational precision improves and longer-period oscillations are detected, the evidence for the extended model ( $M_2$ ) is expected to strengthen, eventually crossing the threshold for definitive model selection. Until then, BMA remains the most rigorous approach for prominence seismology.