Speaker
Description
We modify the method of extrapolating the variational calculation results to the case of the
infinite model space using machine learning of neural networks suggested in Ref. [1]. The main
idea of the modified method is to train an ensemble of artificial neural networks using a
preliminary selection of training data, a subsequent selection of the trained neural networks
according to some criteria, and a statistical processing of the selected network predictions. We
propose a new neural network topology with an appropriate set of learning parameters.
The suggested modified method provides stable results, does not require a division of data
into the training and test sets, ensures the convergence of predictions with increasing the learning
data set by including the results from larger model spaces as well as a high statistical confidence
of the final results.
We extrapolate results obtained within the no-core shell model [2] with NN interaction
Daejeon16 [3] for ground state energies and root-mean-square radii of 6 Li, 6 He and 6 Be nuclei.
We obtain the 6 Li ground state with the same accuracy but higher in energy than the predictions
of Ref. [1]. However, our approach has a higher statistical confidence.
References
- G. A. Negoita et al. // Phys. Rev. C. — 2019. — Vol. 99. — 054308.
- B. R. Barrett, P. Navrátil, J. P. Vary // Prog. Part. Nucl. Phys. — 2013. — Vol. 69. — P. 131.
- A. M. Shirokov et al. // Phys. Lett. B — 2016. — Vol. 761. — P. 87.
| Section | Nuclear structure: theory and experiment |
|---|
