Analysis of hyperfine structure constants of singly ionized cobalt (Co II) were performed on cobalt spectra measured by Fourier transform spectrometers in the region 3000–63,000 cm⁻¹ (33333 – 1587 Å). Fits to over 700 spectral lines led to measurements of 292 magnetic dipole hyperfine interaction A constants, with values between −32.5 mK and 59.5 mK (1 mK = 0.001 cm⁻¹). Uncertainties of 255 A constants were between ±0.4 mK and ±3.0 mK, the remaining 37 ranged up to ±7 mK. The electric quadrupole hyperfine interaction B constant could be estimated for only one energy level. The number of Co II levels with known A values has now increased tenfold, improving and enabling the wider, more reliable, and accurate application of Co II in astronomical chemical abundance analyses.

Aims. Our goal is to accurately determine bound-to-bound transition wavelengths and energy levels of the low-lying open-shell configurations 4f⁴, 4f³ 5d, 4f³6s, and 4f³ 6p of doubly ionised neodymium (Nd III) through high-resolution spectroscopy and semi-empirical calculations. Methods. The emission spectra of neodymium (Nd, Z = 60) were recorded using Penning and hollow cathode discharge lamps in the region 11 500-54000 cm⁻¹ (8695–1852 Å) by Fourier transform spectroscopy at resolving powers up to 10⁶. Wavenumber measurements were accurate to a few 10⁻³ cm⁻¹. Grating spectroscopy of Nd vacuum sliding sparks and stellar spectra were used to aid line and energy level identification. For the analysis, new Nd III atomic structure and transition probability calculations were carried out using the Cowan code parameterised by newly established levels. Results. The classification of 432 transitions of Nd III from the Penning lamp spectra resulted in the determination of 144 energy levels of the 4f⁴, 4f³ 5d, 4f³ 6s, and 4f³ 6p configurations of Nd III, 105 of which were experimentally established for the first time. Of the 40 previously published Nd III levels, one was revised and 39 were confirmed. Conclusions. The results will not only benchmark and improve future semi-empirical atomic structure calculations of Nd III, but also enable more reliable astrophysical applications of Nd III, such as abundance analyses of kilonovae and chemically peculiar stars, and studies of pulsational wave propagation in these stars.

Aims. We aim to accurately determine bound-to-bound transition wavelengths and energy levels of the high-lying open-shell configurations 4f³7s, 4f³6d, and 4f³5f of doubly ionised neodymium (Nd III, Z = 60) through high-resolution spectroscopy and semi-empirical calculations. This study is motivated by lanthanide atomic data requirements in astronomy, such those involved in investigations of kilonova spectra. Methods. Fourier transform spectra of Nd Penning and hollow cathode discharge lamps were recorded within the region 32500–54000 cm⁻¹ (3077–1852 Å) and grating spectra of Nd vacuum sliding sparks were recorded within the regions 820–1159 Å and 1600–3250 Å. New energy levels were found using the observed wavelengths measured accurate to a few parts in 10⁸ in Fourier transform spectra and to a few parts in 10⁷ in grating spectra. Atomic structure and transition probability calculations of Nd III were carried out using the Cowan codes, where energy parameters were adjusted to fit all known Nd III levels. Finally, Nd-rich stellar spectra were also used to aid in the analysis. Results. In total, 355 transitions were classified from observed spectra involving 116 previously experimentally unknown energy levels of the 4f³7s, 4f³6d, and 4f³5f configurations of Nd III, all reported here for the first time. Three newly identified levels of the 4f³5d configuration and one newly identified 4f⁴ level are also reported. Typical level energy uncertainties are 0.01 cm⁻¹ for the 4f³7s and 4f³6d levels and 0.3 cm⁻¹ for the 4f³5f levels. In addition, calculated energy levels up to 130 936 cm⁻¹ are presented, including eigenvector composition and calculated level lifetimes. Calculated transition probabilities and wavelengths between 1900–50 000 Å are also given. Using newly established levels of the 4f³7s configuration and the recently established levels of the 4f³6s configuration, the ionisation energy of Nd III has been estimated at 178 090 ± 330 cm⁻¹. This result offers up to twice the accuracy of the most recently published value.

The atomic spectra and structure of the low-ionisation open d- and f-shell elements are extremely complex, tens of thousands of fine structure lines are observable between the infrared and ultraviolet per species. These transitions underpin most spectroscopic plasma diagnostics, yet their fundamental data remain incomplete and are in high demand in astronomy and fusion research. A key limitation is the substantial human effort required to extract transition wavenumbers and intensities from Fourier transform spectra, as existing peak-detection methods based on thresholding and convolutional kernels frequently fail on blended or weak lines that make up the majority of spectral lines. We approach this challenge by framing spectral line detection as a sequential point-wise binary classification problem to be learned by bidirectional long short-term memory and fully connected neural networks. We trained the model using spectra simulations and evaluated model predictions on experimental spectra of Ni (nuclear charge Z = 28) covering 1800–70 000 cm⁻¹ (5555–143 nm) and Nd (Z = 60) covering 25 369–32 485 cm⁻¹ (394–308 nm) recorded under various experimental conditions. The neural network approach outperforms existing peak-detection methods, particularly for noisy, blended, and/or instrument-distorted lines. In our model evaluation by brief energy level analysis, lines newly detected by the neural networks enabled the confident identification of two Ni II energy levels, 3d 8 ( 3 F 4 ) 6 f [ 2 ] 3 / 2 at 134 261.8946 ± 0.0081 cm⁻¹ and 3d 8 ( 3 F 4 ) 6 f [ 1 ] 3 / 2 at 134 249.5264 ± 0.0054 cm⁻¹, previously deemed to be unidentifiable using the Ni spectra.

Atomic data determined by analysis of observed atomic spectra are essential for plasma diagnostics. For each low-ionisation open d- and f-subshell atomic species, around $10^3$ fine structure level energies can be determined through years of analysis of $10^4$ observable spectral lines. We propose the automation of this task by casting the analysis procedure as a Markov decision process and solving it by graph reinforcement learning using reward functions learned on historical human decisions. In our evaluations on existing spectral line lists and theoretical calculations for Co II and Nd II-III, hundreds of level energies were computed within hours, agreeing with published values in 95% of cases for Co II and 54-87% for Nd II-III. As the current efficiency in atomic fine structure determination struggles to meet growing atomic data demands from astronomy and fusion science, our new artificial intelligence approach sets the stage for closing this gap.