Nonlocality Effect in the Alpha decay half-lives of superheavy nuclei with XGBRegressor

This paper generalizes the alpha-nucleus nonlocality effect to odd-A and odd-odd nuclei within the two-potential approach framework, utilizing an XGBRegressor to optimize parameters and significantly improve the accuracy of alpha-decay half-life predictions for superheavy nuclei.

The Gist

Imagine the atomic nucleus as a tiny, crowded ballroom where protons and neutrons are dancing. Sometimes, this ballroom gets too crowded or unstable, and it decides to eject a small, tight group of dancers—a pair of protons and a pair of neutrons (known as an alpha particle)—to calm down. This ejection is called alpha decay.

The big question physicists have always asked is: "How long will it take for this nucleus to let go of that dancer?" This time is called the half-life.

This paper is about a team of scientists who built a better "crystal ball" to predict exactly how long this process takes, especially for the heaviest, most unstable nuclei (the "superheavy" ones) that humans have only recently created in labs.

Here is the story of how they did it, broken down into simple concepts:

1. The Old Map vs. The New GPS

For decades, scientists used a standard map (a mathematical formula called the Two-Potential Approach or TPA) to predict these half-lives. Think of this map like a paper road map from the 1980s. It's good, but it has blind spots. It assumes the alpha particle moves in a straight, predictable line through a barrier, ignoring some subtle "wiggles" in the physics.

The scientists realized that the alpha particle doesn't just move like a solid marble; it acts a bit like a ghost that can be in two places at once (a quantum effect called nonlocality). The old map didn't account for this ghostly behavior, leading to errors in the prediction.

2. Bringing in the AI Coach

To fix the map, the authors didn't just tweak the math by hand. They brought in a Machine Learning Coach (specifically an algorithm called XGBRegressor).

Imagine you are trying to teach a robot how to drive a car through a complex maze.

• The Old Way: You give the robot a set of rigid rules.
• The New Way: You show the robot thousands of examples of successful drives (experimental data). The robot learns the patterns, the shortcuts, and the tricky turns that the rigid rules missed.

In this study, the "robot" (the AI) analyzed 599 different nuclei. It learned how to adjust the "ghostly" parameters of the alpha particle's movement to match reality perfectly.

3. The Result: A 75% Improvement

The result was stunning. By letting the AI fine-tune the physics model, they reduced the error in their predictions by 74.8%.

• Before: If the real half-life was 10 seconds, the old model might guess 100 seconds or 1 second.
• After: The new model guesses something very close to 10 seconds.

It's like upgrading from a blurry, old photograph to a high-definition 4K image. You can finally see the details clearly.

4. Predicting the Future (Superheavy Nuclei)

The real test was to use this new, super-accurate model to predict the half-lives of 142 superheavy nuclei (elements with atomic numbers 117 to 120). These elements are so heavy and unstable that they often don't exist long enough to be measured easily, or they haven't been made yet.

The scientists compared their AI-enhanced predictions with two other famous, established models (DZR and MUDL).

• The Verdict: Their new model agreed almost perfectly with the best existing models.
• The Discovery: They confirmed that a specific number of neutrons (184) acts like a "magic number" that makes these superheavy nuclei slightly more stable, like a perfectly balanced stack of blocks.

Why Does This Matter?

You might ask, "Who cares about atoms that disappear in a split second?"

1. Finding New Elements: Scientists are constantly trying to build heavier and heavier elements to see if there is an "Island of Stability"—a place where superheavy atoms might actually last for years or even centuries. This new model acts as a treasure map, telling them exactly where to look and what to build.
2. Understanding the Universe: It helps us understand the fundamental rules of how matter holds together.
3. The Power of AI: This paper is a great example of how Artificial Intelligence isn't just for chatbots or self-driving cars; it's a powerful tool for solving the deepest mysteries of physics, acting as a bridge between complex math and real-world data.

In a Nutshell

The authors took a classic physics theory, realized it was missing a subtle "ghostly" detail, and used a smart computer algorithm to fill in the gaps. The result is a much sharper, more accurate tool for predicting the lifespan of the universe's heaviest atoms, helping scientists navigate the frontier of the periodic table.

Technical Summary

1. Problem Statement

Alpha decay is a critical process for understanding nuclear structure and synthesizing superheavy elements (SHEs). While empirical formulas (e.g., Geiger-Nuttall, UDL, DZR) and theoretical models (e.g., Two-Potential Approach - TPA) exist, they often struggle to accurately predict half-lives, particularly for odd-A and odd-odd nuclei where pairing effects and unpaired nucleons complicate the dynamics.

Specifically, the standard TPA framework often neglects the nonlocality effect of the alpha-nucleus interaction. Previous studies incorporated this effect primarily for even-even nuclei. There is a need to:

1. Generalize the nonlocality effect to odd-A and odd-odd nuclei within the TPA framework.
2. Optimize the coordinate-dependent parameters introduced by nonlocality, which are difficult to determine analytically.
3. Improve predictive accuracy for superheavy nuclei ($Z=117-120$) to guide future synthesis experiments.

2. Methodology

The authors developed an Improved Two-Potential Approach (TPA) enhanced by Machine Learning (ML).

A. Theoretical Framework (TPA with Nonlocality)

• Decay Width: The alpha decay half-life ($T_{1/2}$) is calculated using the decay width $\Gamma$, which depends on the preformation factor ($P_\alpha$), normalization constant ($F$), and penetration probability ($P$).
• Nonlocality Effect: Following Medeiros et al., the reduced mass ($\mu$) of the alpha-daughter system is redefined to include a coordinate-dependent mass $m^*$:
  $$\mu = \frac{m^* M}{m^* + M}, \quad m^* = \frac{m}{1-\rho(r)}$$
  Here, $\rho(r)$ is an isotropic function derived from the gradient of a velocity-dependent potential. It introduces a parameter $\rho_S$ (related to the diffuseness of the alpha-core potential) that varies with the radial coordinate.
• Preformation Factor ($P_\alpha$): Calculated using the Cluster Formation Model (CFM), with distinct energy definitions for even-even, odd-A, and odd-odd nuclei based on binding energies.
• Potentials: The total potential includes Nuclear (cosh parametrized), Coulomb, and Centrifugal terms.

B. Machine Learning Integration (XGBRegressor)

Instead of manually tuning the nonlocality parameter $\rho_S$, the authors employed XGBoost (Extreme Gradient Boosting) regression:

• Input Features: Angular momentum of the parent nucleus, $\alpha$-preformation probability, asymmetry term, and the square root of decay energy ($\sqrt{Q_\alpha}$).
• Target Output: The mass parameter $\rho_S$ (specifically the parameter $a_s$ in the $\rho(r)$ function).
• Training: The model was trained on experimental data to minimize the discrepancy between calculated and experimental half-lives.
• Optimization: Python 3.7 was used with automatic optimization routines. The model performance was validated using 5-fold cross-validation (50 repetitions).

C. Scope of Application

1. Training/Validation: Applied to 599 nuclei with $Z=52-118$ (including even-even, odd-A, and odd-odd).
2. Prediction: Used to predict half-lives for 142 superheavy nuclei with $Z=117-120$.
3. Comparison: Results were benchmarked against the DZR model (Deng-Zhang-Royer) and the MUDL model (Modified Universal Decay Law).

3. Key Contributions

1. Generalization to Odd Nuclei: Successfully extended the nonlocality effect treatment within the TPA framework to odd-A and odd-odd nuclei, a domain where previous nonlocality studies were limited.
2. ML-Driven Parameter Optimization: Introduced a novel workflow where an XGBRegressor model predicts the coordinate-dependent nonlocality parameters ($\rho_S$) based on nuclear properties, replacing heuristic or fixed-parameter approaches.
3. Significant Accuracy Improvement: Demonstrated that incorporating the ML-optimized nonlocality effect drastically reduces the error in half-life predictions compared to the standard TPA.

4. Results

• Accuracy Enhancement:
  • The improved TPA achieved a Root-Mean-Square (RMS) deviation that is 74.8% lower than the original TPA (without nonlocality).
  • Standard deviation ($\sigma$) decreased from 0.82058 (original TPA) to 0.46929 (Improved TPA).
• Model Performance Metrics (XG Model):
  • Mean Squared Error (MSE): $2.68 \pm 1.43$
  • Root Mean Squared Error (RMSE): $1.60 \pm 0.33$
  • Coefficient of Determination ($R^2$): $0.56 \pm 0.18$ (indicating moderate correlation, suggesting room for improvement via additional physical features).
• Superheavy Nuclei Predictions ($Z=117-120$):
  • The predictions for 142 superheavy nuclei showed close agreement with both the DZR and MUDL models.
  • The results from the Improved TPA were nearly identical to the DZR model, validating the approach.
  • The models consistently identified $N=184$ as a probable neutron magic number, evidenced by local minima in half-life values.
• Limitations Observed:
  • For specific nuclei (e.g., $Z=117, N=188$), deviations were observed between the Improved TPA and DZR/MUDL. The authors attribute this to the XG model's reliance on $\alpha$-preformation probability, which may not be predicted with sufficient accuracy for all cases.
  • The $R^2$ value of 0.56 suggests the current ML model captures only a moderate level of correlation, likely due to the omission of other physical factors like nuclear deformation.

5. Significance

• Theoretical Advancement: This work bridges the gap between rigorous quantum mechanical tunneling models (TPA) and modern data-driven techniques (Machine Learning), providing a more robust framework for handling complex nuclear systems like odd-odd nuclei.
• Experimental Guidance: The high-accuracy predictions for superheavy nuclei ($Z=117-120$) provide critical guidance for experimental facilities (e.g., GSI, RIKEN, JINR) in planning synthesis routes and identifying new isotopes.
• Methodological Blueprint: The study demonstrates a viable pathway for using gradient boosting models to optimize non-local parameters in nuclear physics, a technique that can be extended to other decay modes or nuclear properties.

In conclusion, the paper successfully demonstrates that integrating nonlocality effects via XGBoost optimization significantly refines alpha-decay half-life predictions, offering a powerful tool for exploring the limits of the periodic table.