3D Full Spectrum Fitting: Algorithm Comparison

TL;DR

3D full-spectrum fitting for IFS data addresses the extraction of stellar kinematics and populations by modeling spatial and spectral dimensions jointly. The study compares PNKR, a joint 3D reconstruction method, with Bayes-LOSVD, a Bayesian LOSVD approach with spatial regularisation, using mock counter-rotating galaxies across SNRs 20–200. Bayes-LOSVD with spatial correlations yields markedly better kinematic recoveries and uncertainties, while PNKR-3D provides little gain over 1D; PNKR-1D can recover qualitative metallicity–velocity trends but introduces metallicity biases. These results highlight the value of 3D modelling and joint population–kinematic analyses for detecting and characterising stellar substructures in IFS data, and suggest paths towards integrating the strengths of both approaches.

Abstract

Full spectrum fitting is the prevailing method for extracting stellar kinematic and population measurements from 1D galaxy spectra. 3D methods refer to analysis of Integral Field Spectroscopy (IFS) data where spatial and spectral dimensions are modelled simultaneously. While several 3D methods exist for modelling gas structures there has been less investigation into the more computationally demanding problem of 3D full spectrum fitting for stellar recoveries. This work introduces and compares two algorithms for this task: the Projected Nesterov Kaczmarz Reconstruction method (PNKR) and a version of the Bayes-LOSVD software which has been modified to account for spatial correlations. We aim to understand strengths and weaknesses of both algorithms and assess the impact of 3D methods for stellar inferences. We apply both recovery algorithms to a mock IFS data over a signal-to-noise ratio (SNR) range from 20-200 and evaluate the quality of the recoveries compared to the known ground truth. Accounting for spatial correlations in Bayes-LOSVD significantly improved the accuracy and precision of kinematic recoveries. 3D modelling with PNKR did not provide any significant improvement over 1D fits however, for SNR>40, PNKR did recover the most accurate kinematics overall. Additionally, by modelling the joint distribution over kinematics and populations, PNKR could successfully infer trends between these quantities e.g. inferring local metallicity-velocity trends, albeit with a significant bias on the absolute metallicity. Having demonstrated advantages of (i) 3D modelling with Bayes-LOSVD, and (ii) joint kinematic-population analyses with PNKR, we conclude that both methodological advances will prove useful for detecting and characterising stellar structures from IFS data.

Figures (21)

• Figure 1: Toy demonstration of LOSVD recovery with PNKR. The left panel shows the number of active wavelengths per iteration of the algorithm, with iterations 300 and 1000 highlighted. Corresponding LOSVD recoveries are shown in the right panels: at iteration 300, the recovery is smooth and matches the truth well, while extending to iteration 1000 introduces noise.
• Figure 2: Demonstration of LOSVD recovery with BLOSVD-1D. Each panel shows BLOSVD-1D recoveries via their posterior median (thick blue line), 99% credible interval (shaded region) and 10 randomly selected posterior samples (thin blue lines). For the left panel, the template spectrum contains a single absorption line exactly one pixel wide; here, the recovery tightly encloses the truth. For the right panel, the absorption line is two pixels wide, which introduces a degeneracy between neighbouring velocity bins, evidenced by the zig-zagging posterior samples. This degeneracy results in noisy median recoveries and inflated credible intervals.
• Figure 3: Our ground truth galaxy model in age-metallicity space (left panel), as a light-weighted image (center) and mean velocity map (right). The model consist of two components. The dominant component (blue contours) is a young, metal-rich thin disk with negative LOS velocity. The counter-rotating weaker component (orange contours) is older, more metal-poor and more vertically extended. Successive contours in the central panel show changes in flux by 15%. The grid of crosses in the right panel indicate spaxels used for illustration in Figures \ref{['fig:pnkr_vs_blosvd']}, \ref{['fig:b3d']}, and \ref{['fig:metal_map']}.
• Figure 4: Comparison of LOSVDs recovered from 1D spectral fits. True LOSVDs (black) are shown alongside recoveries from PNKR-1D (red) and BLOSVD-1D (blue). At SNR 200 (left grid) PNKR recovers the LOSVD well everywhere, while BLOSVD recoveries show spike artifacts in some cases e.g. panels 1, 2 and 9. At SNR 20 (right grid) the recoveries are worse. Both algorithms produce LOSVDs with extended, flat wings which reach the edge of the velocity range. For spaxels with bimodal LOSVDs (panels 5-9) we see that BLOSVD-1D captures the dip between the two components whereas PNKR-1D oversmooths this feature.
• Figure 5: Error in recovered LOSVDs from 1D spectral fits as a function of SNR. At high SNR, PNKR (red) achieves the smallest error, while BLOSVD (blue) is best at low SNR. By modifying PNKR to retain all wavelengths throughout the recovery (pink line), we see an improvement at low SNR, with a recovery error similar to BLOSVD, at the expense of larger errors at high SNR.