The formation probability of the compound nucleus $P_{\mathrm{CN}}$, predominated by quasifission, is pivotal for synthesizing heavy and superheavy elements via fusion-evaporation reactions. However, $P_{\mathrm{CN}}$ has thus far been formulated only phenomenologically, with substantial discrepancies observed in certain systems. Therefore, accurate prediction of $P_{\mathrm{CN}}$ is essential when designing heavy-ion fusion experiments. In this study, we quantitatively investigate $P_{\mathrm{CN}}$ using a machine learning (ML) algorithm, eXtreme Gradient Boosting. To reproduce entrance-channel properties as accurately as possible, we introduced various input features, including the charge product, the ratio of center-of-mass energies to the Bass barrier, fissility parameters, mass asymmetry, and static deformation parameters of the compound nucleus. The precision of the ML model was characterized using the mean absolute error (MAE), which is the average absolute difference between log ${}_{10}$ of the predicted and experimental $P_{\mathrm{CN}}$ values. The associated uncertainty was evaluated by repeatedly training on randomly split subsets of the dataset. The trained models achieved mean MAEs of 0.056 and 0.070 for the training and validation sets, respectively. For the test set, the mean MAE was 0.077, and the ratios between the predicted and experimental $P_{\mathrm{CN}}$ values were mostly distributed within a factor of ${10}^{\pm 0.2}$. Furthermore, we compared the predictability of our ML approach with $P_{\mathrm{CN}}$ values calculated using the semiempirical formula proposed by Zagrebaev and Greiner, revealing that our ML model yielded better performance on the test set by aligning closer with the experimental results.

Close

PrevNext

View figure in article

• Low & intermediate energy heavy-ion reactions
• Models & methods for nuclear reactions
• Nuclear fusion
• Machine learning
• Nuclear reactions