Abstract
Abstract During a solar flare, it is believed that reconnection takes place in the corona followed by fast energy transport to the chromosphere. The resulting intense heating strongly disturbs the chromospheric structure and induces complex radiation hydrodynamic effects. Interpreting the physics of the flaring solar atmosphere is one of the most challenging tasks in solar physics. Here we present a novel deep-learning approach, an invertible neural network, to understanding the chromospheric physics of a flaring solar atmosphere via the inversion of observed solar line profiles in Hα and Ca ii λ8542. Our network is trained using flare simulations from the 1D radiation hydrodynamic code RADYN as the expected atmosphere and line profile. This model is then applied to single pixels from an observation of an M1.1 solar flare taken with the Swedish 1 m Solar Telescope/CRisp Imaging SpectroPolarimeter instrument just after the flare onset. The inverted atmospheres obtained from observations provide physical information on the electron number density, temperature, and bulk velocity flow of the plasma throughout the solar atmosphere ranging from 0 to 10 Mm in height. The density and temperature profiles appear consistent with the expected atmospheric response, and the bulk plasma velocity provides the gradients needed to produce the broad spectral lines while also predicting the expected chromospheric evaporation from flare heating. We conclude that we have taught our novel algorithm the physics of a solar flare according to RADYN and that this can be confidently used for the analysis of flare data taken in these two wavelengths. This algorithm can also be adapted for a menagerie of inverse problems providing extremely fast (∼10 μs) inversion samples.
Highlights
The current and generation of solar observations, with their high spatial, temporal, and spectral resolution, present a significant analysis challenge, as does the increasing complexity and realism of the models with which the data are confronted
We demonstrate the method on data taken by the CRisp Imaging SpectroPolarimeter (CRISP) instrument on the Swedish Solar Telescope (Scharmer et al 2003, 2008)
We look to fully connected artificial neural networks (ANNs)
Summary
The current and generation of solar observations, with their high spatial, temporal, and spectral resolution, present a significant analysis challenge, as does the increasing complexity and realism of the models with which the data are confronted. Sampling different values from the latent space will lead to a sampling of the distribution of the input parameters corresponding to a given output observation This deterministic function x = g (y, z) is invertible, and we can learn the function g−1 as the forward process and g as the inverse process that will directly track where the lost information is obtained from the latent space. The input data is fed through a neural network, where linearities and nonlinearities are applied to it until it reaches the output, where it is compared with the known answers This comparison is surmised by a loss function, which is minimized by changing the values of the weights in each layer of the network to produce a different result (Schmidhuber 2015). The flow of the forward model is shown by the black arrows, and the flow of the inverse is shown by the cyan arrows
Sign-in/Register to view highlights & summary Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
