Resumen:

Active regions (ARs) appear in the solar atmosphere as the consequence of the emergence of magnetic flux tubes formed in the solar interior. Several observational evidence, models and simulations have shown that these coherent structures must carry magnetic helicity, forming magnetic flux ropes (FRs). Since these FRs are the most important means by which magnetic energy is transported out to the solar atmosphere, its study is fundamental to fully comprehend energy release processes such as flares and coronal mass ejections. However, acquiring precise estimations of their intrinsic magnetic parameters during the early phase of the ARs evolution is limited to the observed photospheric magnetic flux distribution. In particular, the observed line-of-sight (LOS) component of the photospheric magnetic field can be affected by the amount of twist on these FRs producing a departure of the expected symmetric bipolar configuration. In this work, we aim to model the magnetic parameters of emerging FRs using a Bayesian scheme. We model the 3D structure of the FR with a magnetic field with a geometry defined by a half-torus with uniform torsion. This 8-parameter model can produce a sequence of synthetic LOS magnetograms by projecting the vertical component over transversal planes at different heights relative to the center of the torus. We perform the inference over a sequence of LOS magnetograms of four different emerging ARs. We test and compare submodels in which different temporal correlation of the parameters are considered. We found that the inferred magnetic parameters such as the magnetic helicity and tilt angle obtained for all the studied ARs are consistent with other previous estimations.

Resumen:

The present work is framed within the applications of deep neural networks for modeling phenomena present in the solar photosphere. The proposed research is based on the construction of a deep 3D generative convolutional neural network, known as DCGAN (Deep Convolutional Generative Adversarial Network), utilizing Python's artificial intelligence modules such as Pytorch for the neural network architecture. The goal is to train a neural network capable of generating groups of cubes highly similar to training cubes. These cubes correspond to physical quantities of the solar photosphere, such as density, magnetic field, plasma velocity, temperature, among others, obtained from the MURaM simulation code. The aim is to generate realistic simulations of magnetoconvection processes and magnetic activities that occur in the solar convective zone. This work employs the simulation results as training data for the neural network, generating new data with consistency in the physical parameters. Subsequently, a comparison is made between the original simulated results and the training data, and challenges are proposed and discussed for using these tools in the study of the solar photosphere, flux tubes, and pores.