Aryana Haghjoo, Shoubaneh Hemmati, Bahram Mobasher, Nima Chartab, Alexander de la Vega, Tim Eifler, Emily Everetts, Hooshang Nayyeri, Zahra Sattari

The information recoverable from galaxy spectra depends fundamentally on spectral resolution, yet assembling large samples at high resolution remains observationally expensive. We present a deep-learning framework for spectral super-resolution that enhances low-resolution galaxy spectra by a factor of $\sim$10 in resolving power ($R\sim100$ to $R\sim1000$). The model is trained on 1,187 paired JWST/NIRSpec observations from the JADES program, where low-resolution prism spectra are matched with medium-resolution grating spectra (G140M, G235M, G395M) combined into a unified reference covering 1-5 $μ$m. Our three-stage architecture performs an initial super-resolution, infers the redshift from the coarse reconstruction, and then applies a physics-informed residual refinement that uses attention across emission-line tokens to learn inter-line relationships and predict parametric line profiles, alongside a convolutional branch for continuum corrections. Evaluated on a 20% held-out sample, the model achieves noise-limited residuals over most of the spectral range and systematically improves the signal-to-noise ratio of key diagnostic lines including [OII], H$β$, [OIII], and H$α$, often by factors of several. The super-resolved spectra successfully deblend features that are entirely unresolved at prism resolution, such as the [OIII] $λ\lambda4959,5007$ doublet and H$β$. As a proof of concept using JWST data, this approach is readily extensible to the low-resolution grism spectroscopy that will be delivered by Euclid and the Roman Space Telescope, potentially enabling population-level diagnostics across millions of galaxy spectra that would otherwise be inaccessible at grism resolution.