Generation of fission yield covariance matrices and its application in uncertainty analysis of decay heat

This study employs a generalized least squares (GLS) updating approach to generate fission yield covariance matrices from major nuclear data libraries, demonstrating that incorporating these correlations significantly reduces decay heat uncertainty and shifts the dominant error source from fission yields to decay energy data.

The Gist

Imagine a nuclear reactor is like a massive, high-speed factory that smashes heavy atoms apart to create energy. When these atoms split (a process called fission), they don't just vanish; they turn into hundreds of different "baby" atoms, known as fission products.

Even after the factory shuts down, these baby atoms keep buzzing around, decaying and releasing heat. This leftover heat is called decay heat. It's a critical safety issue: if we can't predict exactly how much heat is left, the reactor could overheat, or we might not know how to safely store the waste.

To predict this heat, scientists use a "recipe book" (nuclear data libraries) that lists how many of each baby atom are created and how much energy they release. However, there's a problem: the old recipe books had a blind spot. They told scientists the average number of baby atoms, but they didn't tell them how much those numbers might vary or how the different baby atoms were connected to each other.

Think of it like baking a cake. The old books said, "Use 2 eggs." But they didn't say, "If you accidentally use 3 eggs, you probably also used 1.5 cups of flour instead of 1, because the baker tends to make mistakes in pairs." Without knowing these connections (called covariances), scientists had to guess the worst-case scenario, leading to huge safety margins and uncertainty.

What This Paper Did: The "Smart Update"

The authors of this paper decided to fix the recipe books using a clever mathematical trick called Generalized Least Squares (GLS).

Here is how they did it, using a simple analogy:

1. The "Physics Rules" (The Constraints)
Imagine you are trying to guess the ingredients of a secret soup. You know some hard rules:

• Conservation of Mass: The total weight of the soup must equal the weight of the ingredients you started with.
• Conservation of Charge: The total "electric charge" must balance out.
• Chain Yields: You know that if you wait long enough, certain ingredients will turn into others (like a caterpillar turning into a butterfly).

The authors took the existing data from three major international recipe books (ENDF, JENDL, and JEFF) and forced them to obey these physics rules perfectly.

2. The "Update" (GLS)
They used the GLS method to act like a smart editor.

• Step 1: They looked at the original data.
• Step 2: They applied the physics rules (the constraints).
• Step 3: The math "adjusted" the numbers slightly to fit the rules perfectly.
• Step 4 (The Magic): While adjusting the numbers, the math also figured out the connections. It realized, "Ah, if the number of Atom A goes up, Atom B must go down to keep the total weight the same."

This created a Covariance Matrix. In our soup analogy, this is a new page in the recipe book that says: "These ingredients are linked. If one changes, the other changes in a predictable way."

The Results: Why It Matters

The team tested these new, "updated" recipe books by calculating the decay heat for Uranium-235 (the most common fuel). Here is what they found:

• Before the Update (The Old Way): Because the books didn't know about the connections, the uncertainty (the "fuzziness" of the prediction) was high. It was like saying, "The leftover heat could be anywhere between 100 and 200 units." The main source of this fuzziness was the guesswork about how many baby atoms were created.
• After the Update (The New Way): By adding the connections (covariances), the fuzziness disappeared. The uncertainty dropped significantly.
  • At very short times (0.1 seconds after shutdown), the uncertainty dropped from a vague guess to a precise ~5-10%.
  • At longer times (days later), the uncertainty became incredibly small, around 1%.

The Big Surprise:
Once they fixed the "baby atom" counts, the biggest source of uncertainty wasn't the atoms anymore—it was the energy they release. It's like fixing the ingredient list perfectly, only to realize the temperature of the oven is the only thing left that's a bit fuzzy. This tells scientists exactly where to focus their future research: measure the energy release more precisely, not just the atom counts.

The Takeaway

This paper is a major step forward in nuclear safety. By using math to force the data to obey the laws of physics and by mapping out how different atoms are connected, the authors have turned a "fuzzy guess" into a "sharp prediction."

In short: They took a blurry, uncertain picture of nuclear decay heat and sharpened it into a clear, high-definition image. This means engineers can design safer reactors and manage nuclear waste with much greater confidence, knowing exactly how much heat is coming and when it will fade away.

Technical Summary

1. Problem Statement

• Context: Decay heat from fission products is a critical parameter for nuclear reactor safety, design, and waste management. Accurate uncertainty analysis is essential for these applications.
• The Gap: While evaluated nuclear data libraries (ENDF/B-VIII.0, JENDL-5, JEFF-3.3) provide best-estimate fission yields and their individual uncertainties (standard deviations), they lack covariance matrices for fission yields.
• Consequence: Without covariance data, uncertainty analyses must assume fission yields are uncorrelated. This assumption leads to an overestimation of the total uncertainty in decay heat calculations, particularly at longer cooling times, and fails to capture the physical correlations between different fission products.
• Objective: To generate fission yield covariance matrices using a Generalized Least Squares (GLS) updating approach, constrained by physical conservation laws and experimental chain yield data, and to apply these matrices to re-evaluate the uncertainty of decay heat for thermal neutron-induced fission of $^{235}\text{U}$.

2. Methodology

The study employs a multi-step methodology combining nuclear data evaluation, statistical updating, and perturbation theory.

A. Covariance Generation via Generalized Least Squares (GLS)

The authors used the GLS updating approach to generate posterior covariance matrices for Independent Fission Yields (IFY).

• Prior Data: Used original IFY data and diagonal covariance matrices (variances only) from ENDF/B-VIII.0, JENDL-5, and JEFF-3.3.
• Constraints: The updating process enforced five physical constraints to ensure the data remained physically consistent:
  1. Conservation of Mass and Charge: Ensured the sum of mass and charge numbers of fission products matched the compound nucleus (accounting for prompt neutrons and ternary fission light charged particles).
  2. Normalization of IFYs: The sum of yields (excluding light charged particles) must equal 2.
  3. Normalization of Heavy Mass Yields: Yields on either side of a midpoint mass number must sum to 1.
  4. Charge Symmetry: Enforced symmetry in charge distributions for binary fission (with slight adjustments for ternary fission).
  5. Chain Yield Constraints: Used experimental chain yield data (from England & Rider for ENDF/JENDL and Nichols et al. for JEFF) as strong constraints. This required correcting IFYs for delayed neutron emission to match post-delayed-neutron chain yields. A half-life cutoff ($T_{cut} = 1$ min) was applied to define which decay processes contribute to the chain yield calculation.
• Output: The GLS process produced updated IFY values and a full covariance matrix, introducing correlations (primarily negative) between yields of different nuclides.

B. Decay Heat Calculation (Summation Method)

• Target: Thermal neutron-induced fission of $^{235}\text{U}$ ($^{235}\text{U}(n_{th}, f)$).
• Approach: Calculated Fission Pulse Decay Heat (FPDH) using the summation method: $H(t) = \sum \lambda_i \bar{E}_i N_i(t)$.
• Data Sets: Two sets were compared:
  • Set-A: Original library data (uncorrelated yields).
  • Set-B: GLS-updated yields (correlated) combined with original decay data.

C. Uncertainty Analysis

• Theory: Generalized Perturbation Theory (Sandwich formula: $V_H = S^T V_\sigma S$).
• Sources: Uncertainties were propagated from four sources: Fission Yield, Decay Energy, Decay Constant, and Branching Ratio.
• Sensitivity: Sensitivity coefficients were calculated analytically for decay constants/energies and numerically for branching ratios and yields.

3. Key Contributions

1. First Application of GLS to Multiple Libraries: Successfully generated covariance matrices for $^{235}\text{U}$ fission yields across three major international libraries (ENDF/B-VIII.0, JENDL-5, JEFF-3.3) using a unified constraint framework.
2. Physical Constraint Integration: Demonstrated how to rigorously integrate fundamental physical conservation laws (mass, charge, normalization) and experimental chain yield data into the covariance generation process.
3. Quantification of Correlation Impact: Provided the first quantitative assessment of how fission yield correlations reduce decay heat uncertainty compared to the standard uncorrelated assumption.
4. Library Comparison: Identified specific discrepancies between libraries, particularly regarding the contributions of specific nuclides (e.g., $^{98}\text{Zr}$, $^{102}\text{Nb}$) and the underestimation of electromagnetic decay heat in JEFF-3.3.

4. Key Results

A. Impact of GLS Updating on Data

• Data Adjustment: The GLS procedure slightly adjusted the IFY values (mostly <5% for ENDF/JENDL, negligible for JEFF) to satisfy constraints.
• Covariance Structure: Generated strong negative correlations between fission yields of neighboring mass numbers (e.g., A=84-86, 136-137). This reflects the physical reality that if one mass chain yield increases, a neighboring one must decrease to conserve mass.
• Constraint Satisfaction: The $\chi^2$ values for constraints (mass/charge conservation, chain yields) were significantly reduced, indicating the updated data is more consistent with physical laws and experimental benchmarks.

B. Decay Heat Uncertainty Reduction

• Uncorrelated Case (Set-A): Original uncorrelated yield data contributed a constant ~4% uncertainty to the total decay heat at all cooling times. This dominated the total uncertainty for cooling times > 100 s.
• Correlated Case (Set-B): With the generated covariance matrices, the contribution of fission yield uncertainty was strongly reduced.
  • At 0.1 s cooling time: Total uncertainty dropped to ~10% for ENDF/B-VIII.0 and JEFF-3.3, and ~5% for JENDL-5.
  • At 10$^5$ s cooling time: Total uncertainty for all three libraries reduced to ~1%.
• Dominant Source: In the correlated scenario, Decay Energy became the primary contributor to uncertainty, replacing fission yield as the dominant factor.

C. Nuclide Contributions and Sensitivity

• Dominant Nuclides: At 10 s cooling time, 40 specific nuclides accounted for >80% of light particle decay heat.
• Library Discrepancies: Significant differences were found in the contributions of specific nuclides (e.g., $^{102}\text{Nb}$, $^{96m}\text{Y}$, $^{146m}\text{La}$) between libraries.
  • Example: JEFF-3.3 underestimated electromagnetic decay heat (5–50 s) because it assigned zero decay energy to $^{98}\text{Zr}$, whereas ENDF/B-VIII.0 assigned 0.449 MeV.
• Sensitivity Coefficients: The GLS adjustment had a weak influence on sensitivity coefficients for ENDF and JEFF but caused dramatic changes (up to 95% reduction or 5x enhancement) for specific nuclides in JENDL-5 (e.g., $^{102}\text{Nb}$, $^{146}\text{La}$).

5. Significance and Conclusion

• Improved Safety Margins: By correctly accounting for correlations, the study demonstrates that decay heat uncertainties are significantly lower than previously estimated when using uncorrelated data. This allows for more precise safety margins in reactor design and spent fuel management.
• Data Validation: The study validates the GLS approach as a robust method for generating missing covariance data in nuclear libraries, ensuring consistency with physical conservation laws.
• Future Application: The generated covariance matrices and updated yield data are ready for integration into nuclear design codes and safety analysis tools, moving the field beyond the assumption of uncorrelated fission yields.
• Library Recommendations: The authors highlight the need for JEFF-3.3 to re-evaluate decay energy data for specific nuclides (like $^{98}\text{Zr}$) to resolve discrepancies in electromagnetic decay heat predictions.

In summary, this work bridges a critical gap in nuclear data evaluation by providing the first comprehensive set of fission yield covariance matrices for major libraries, fundamentally changing the understanding of uncertainty propagation in decay heat calculations.