Quantifying uncertainty in physics-based predictions of rare-isotope production cross sections via Bayesian-inspired model averaging across nuclear mass tables

This paper introduces a Bayesian-inspired model-averaging framework that combines abrasion-ablation calculations from multiple nuclear mass tables to generate statistically weighted predictions and uncertainty estimates for rare-isotope production cross sections, thereby improving accuracy for both interpolation and limited extrapolation in proton-rich fragmentation regimes.

The Gist

Imagine you are trying to bake the perfect cake, but you don't have a single, perfect recipe. Instead, you have a dozen different cookbooks, each with its own slightly different instructions for how much flour, sugar, and heat to use. Some cookbooks are great for chocolate cakes, while others are better for vanilla. If you follow just one, you might end up with a dry sponge or a burnt mess.

This is exactly the problem nuclear physicists face when trying to predict how to create rare isotopes (exotic, unstable versions of atoms) in a particle accelerator. They need to know exactly how many of these rare "atoms" they will get out of a collision to plan their experiments. If they guess wrong, they might waste months of expensive machine time or miss a discovery entirely.

Here is a simple breakdown of what this paper does, using our kitchen analogy:

1. The Problem: Too Many Recipes, No Consensus

The scientists use a computer program called LISE++ to simulate these atomic collisions. This program uses a method called "Abrasion-Ablation" (think of it as a two-step process: first, you chip off a piece of the atom like a cookie cutter, then the remaining piece cools down and loses more bits like steam escaping a pot).

The problem is that the program needs a "mass table" (a list of how heavy every atom is) to work. There are 12 different mass tables available, like 12 different cookbooks.

• Cookbook A might predict you'll get 100 rare atoms.
• Cookbook B might predict only 10.
• Cookbook C might predict 1,000.

In the past, scientists would just pick one cookbook and hope for the best. But sometimes, the real world doesn't match any single cookbook perfectly.

2. The Solution: The "Master Chef" Averaging

The author, O. B. Tarasov, developed a new method called Bayesian-inspired Model Averaging (BMA). Instead of picking one cookbook, he acts like a Master Chef who listens to all 12 of them.

Here is how the Master Chef decides who to listen to:

1. The Taste Test: He takes the 12 cookbooks and tests them against real data from two specific experiments (using Krypton-78 and Xenon-124 beams). He sees which cookbook predicted the results most accurately.
2. Assigning Votes: The cookbooks that got the "taste test" right get more votes (higher weight). The ones that got it wrong get fewer votes.
3. The Final Recipe: He doesn't just pick the winner. He mixes the predictions of all 12 cookbooks together, weighted by how good they were. The result is a single, statistically weighted prediction that is smarter than any single cookbook.

3. The Magic Trick: Scaling to New Ingredients

Now, the scientists want to use this "Master Chef" method for new experiments they haven't done yet (using Molybdenum-92 and Samarium-144 beams). They don't have real data for these yet to do a taste test.

So, they use a scaling rule. Imagine if the Master Chef knows that "Cookbook A" is great for small cakes and "Cookbook B" is great for big cakes. If he wants to bake a medium-sized cake, he can mathematically estimate how much to trust each cookbook based on the size of the cake.

• He takes the lessons learned from the Krypton and Xenon experiments.
• He applies a mathematical "ruler" to figure out how those lessons apply to the new Molybdenum and Samarium experiments.
• This allows him to predict the outcome for the new experiments with a built-in uncertainty estimate (a "confidence interval").

4. Why This Matters: Finding the "Needle in the Haystack"

The ultimate goal is to find new, never-before-seen isotopes. These are like finding a needle in a haystack.

• If the prediction says you will get 1,000 needles, you might not bother looking very hard.
• If the prediction says you will get 1 needle, you need to set up your detector perfectly to catch it.

Using this new averaging method, the paper helps scientists decide:

• Which beam to use: Should they smash Krypton, Xenon, Molybdenum, or Samarium into the target to find a specific rare atom?
• Where to look: It helps them identify "blind spots" where current beams fail, suggesting that a different beam (like Samarium-144) might be better for finding very proton-rich Xenon isotopes.

The Bottom Line

This paper is about moving away from "guessing" with a single model to smartly combining multiple models. By listening to all the experts (the 12 mass tables) and weighing their opinions based on past performance, the scientists can now predict the production of rare atoms with much higher confidence and a clear understanding of how wrong they might be.

It's like going from asking one friend, "How long will this trip take?" to asking a whole group of travelers, averaging their answers based on who has the best map, and getting a much more reliable estimate of your arrival time.

Technical Summary

Here is a detailed technical summary of the paper "Quantifying uncertainty in physics-based predictions of rare-isotope production cross sections via Bayesian-inspired model averaging across nuclear mass tables" by O. B. Tarasov.

1. Problem Statement

Accurate prediction of projectile fragmentation cross sections is critical for planning rare-isotope beam production, optimizing separator settings, and designing experiments to discover new isotopes near the proton and neutron drip lines. However, existing prediction methods face significant limitations:

• Phenomenological parameterizations (e.g., EPAX) lack predictive power in data-scarce regions and do not incorporate nuclear masses.
• Mass-based systematics are often constrained by neutron-rich data and fail in proton-rich regions.
• Physics-based models (specifically Abrasion–Ablation, AA) are sensitive to model assumptions, particularly regarding excitation-energy generation and, most critically, the choice of nuclear mass tables. No single AA realization using a specific mass table is globally reliable across all isotopic regions.

The core problem is the lack of a robust method to quantify systematic uncertainties arising from the choice of nuclear mass models when predicting cross sections for unmeasured, exotic nuclei.

2. Methodology

The author develops a Bayesian-inspired Model Averaging (BMA) framework to combine predictions from multiple nuclear mass tables into a single, statistically weighted estimate with quantified uncertainties.

A. The Abrasion–Ablation (AA) Framework

The study utilizes the LISE++ code, which implements the AA model. This two-step process involves:

1. Abrasion: A fast geometric removal of nucleons to form an excited prefragment.
2. Ablation: Statistical de-excitation via particle evaporation (n, p, $\alpha$) and fission.

Key model settings include:

• Excitation Energy Model: The study adopts the Exponential Deposition Model (EDM) over Gaussian or log-normal forms. The EDM is parameterized by $T = k_1 + k_2\Delta A_{abr} + k_{NZ}(\Delta N_{abr} - \Delta Z_{abr})$, offering better physical transparency and scaling with projectile mass.
• Minimization Strategy: A robust fitting procedure minimizes an objective function $J$ combining a quadratic metric (sensitive to absolute values) and a logarithmic metric (sensitive to low-yield tails).
• Robustness Features: The minimization employs "local-line cycling" to avoid local minima and diagonal scaling to handle parameters with different numerical magnitudes.

B. Calibration and Weighting

1. Calibration Data: The framework is calibrated using high-quality experimental data for $^{78}$Kr + $^{9}$Be and $^{124}$Xe + $^{9}$Be fragmentation.
2. Mass Table Evaluation: AA calculations are performed using 12 different nuclear mass models (e.g., AME2020, FRDM2012, HFB-27, KTUY, WS4RBF).
3. Empirical Weighting: For each mass model $i$, a weight $w_i$ is assigned inversely proportional to the objective function value $J_i$ derived from the calibration data:
   $$w_i = \frac{J_i^{-1}}{\sum_k J_k^{-1}}$$
   This assigns higher weight to models that better reproduce the measured isotopic distributions.
4. Model Averaging: The final cross section is a weighted mean of the predictions in log-space to handle the wide dynamic range, resulting in asymmetric uncertainty bands.

C. Propagation to New Systems

To predict cross sections for $^{92}$Mo and $^{144}$Sm (where no direct calibration data exists), the author employs a controlled scaling procedure:

• Excitation Energy Scaling: A separable size–isospin scaling law is used to interpolate the EDM parameters ($k_1, k_2, k_{NZ}$) from the calibrated projectiles ($^{78}$Kr, $^{124}$Xe) to the new projectiles based on mass ($A$) and asymmetry ($I = (N-Z)/A$).
• Weight Interpolation: The model weights for the new projectiles are inferred by interpolating the goodness-of-fit ($J$ values) between the two calibration points in the $(A, I)$ plane.

3. Key Contributions

• Uncertainty Quantification: The paper moves beyond single-model predictions by providing a rigorous statistical framework that quantifies the systematic uncertainty arising from the choice of nuclear mass tables.
• Hybrid Framework: It successfully integrates physics-based AA calculations with empirical model averaging, preserving the underlying physics while correcting for model bias.
• Scaling Procedure: A novel method is introduced to propagate model weights and excitation-energy trends from calibrated systems to uncalibrated intermediate-Z projectiles ($^{92}$Mo, $^{144}$Sm).
• Improved Predictions: The BMA approach demonstrates superior agreement with experimental data compared to standard phenomenological parameterizations (EPAX3), particularly in the tails of isotopic distributions.

4. Results

• Calibration Performance:
  • For $^{78}$Kr, the KTUY mass model provided the best fit ($J=1.10$), followed closely by AME2020.
  • For $^{124}$Xe, HFB-27 yielded the best fit.
  • The BMA predictions for both systems showed that ~60-64% of data points fell within a factor of 2 of the measurement, with significant improvement in the low-yield tails compared to EPAX3.
• Propagation to $^{92}$Mo and $^{144}$Sm:
  • $^{92}$Mo: The BMA predictions showed a steeper suppression of the most proton-rich tails compared to EPAX3, correcting the overestimation often seen in phenomenological models.
  • $^{144}$Sm: The BMA and EPAX3 systematics were generally closer, with differences restricted to the extreme low-cross-section region.
• Beam Selection Analysis:
  • The study analyzed the production of proton-rich isotopes in the $Z=50-54$ region (Sn, Te, Xe).
  • $^{144}$Sm was identified as the superior projectile for the most proton-rich Xe isotopes due to significant cross-section gain factors, despite slightly lower separator transmission.
  • $^{124}$Xe remains favored for Sn isotopes due to better transmission conditions.
• New Isotope Discovery Potential:
  • By combining BMA production rates with proton-decay half-life estimates (using the Brown-Barker R-matrix formalism), the study identified 10 promising candidates for new isotope discovery:
    • $^{92}$Mo beam: $^{71,72}$Sr, $^{75,76}$Zr, $^{80}$Mo.
    • $^{144}$Sm beam: $^{113}$Ba, $^{117,118}$Ce, $^{122}$Nd, $^{127}$Sm.
  • These candidates are expected to have production rates $\gtrsim 1$ ion/day and measurable proton-decay lifetimes.

5. Significance

This work provides a practical, uncertainty-aware basis for planning future rare-isotope experiments at facilities like FRIB (Facility for Rare Isotope Beams). By moving away from reliance on a single mass table or phenomenological fit, the BMA framework:

1. Reduces Systematic Bias: It mitigates the risk of missing new isotopes due to model-specific failures.
2. Guides Experimental Design: It offers quantitative guidance on projectile selection and separator settings for maximizing yields of exotic nuclei.
3. Identifies "Blind Spots": It highlights specific regions of the nuclear chart (intermediate-Z, proton-rich) where intermediate-Z projectiles ($^{92}$Mo, $^{144}$Sm) are essential for discovery, bridging the gap between light ($^{78}$Kr) and heavy ($^{238}$U) beam capabilities.
4. Data Availability: The authors have made the model-averaged cross-section tables publicly available, facilitating immediate use by the experimental community.