SDSS-V Local Volume Mapper (LVM): Dithered Data Cube Reconstruction with 3dcubegen

Technical Summary: SDSS-V Local Volume Mapper (LVM): Dithered Data Cube Reconstruction with 3dcubegen

Problem Statement
The SDSS-V Local Volume Mapper (LVM) conducts an integral field spectroscopic (IFS) survey of the Milky Way, Magellanic Clouds (MCs), and nearby galaxies using a strategy of multiple dithered exposures to achieve full spatial coverage and improved sampling. While the LVM Data Reduction Pipeline (LVM-DRP) produces calibrated row-stacked spectra (RSS) for individual pointings, scientific exploitation requires combining these dithered observations into homogeneous three-dimensional data cubes. This combination is necessary to reach deeper sensitivity levels, detect faint emission lines, and recover spatial information lost in single pointings. However, existing methods required adaptation to handle the specific hardware constraints of the LVM, particularly its large fiber diameter (35.3 arcseconds) relative to the seeing conditions, and the need to preserve flux accuracy and propagate uncertainties correctly across thousands of dithers.

Methodology
The authors present 3dcubegen, a flexible and scalable Python package designed to reconstruct data cubes from LVM dithered observations. The methodology extends approaches previously used in surveys like CALIFA and MaNGA but is specifically optimized for the LVM's 9-point dither pattern and fiber geometry. The reconstruction process involves three primary steps:

1. 2D Spatial Interpolation: The core algorithm interpolates the flux of individual fibers, located at sky positions $(x_i, y_i)$, onto a discrete 2D spatial grid. The flux at a spaxel $(x, y)$ is calculated as a weighted mean of all fibers within a threshold distance, using the expression:
   $$F_{x,y}(\lambda) = \frac{\sum w(\lambda, r_i) f_A F_i(\lambda, x_i, y_i)}{\sum w(\lambda, r_i)}$$
   Here, $f_A$ is an aperture correction factor ($A_{spax}/A_{fib}$) ensuring flux conservation, and $w(\lambda, r)$ is a weighted kernel defined as $w = \exp(-0.5(r/\sigma)^\alpha)$. The parameter $\sigma$ controls the kernel size, and $\alpha$ controls the kernel sharpness (where $\alpha=2$ yields a Gaussian shape).
2. Kernel Selection: Unlike previous surveys where the Point Spread Function (PSF) dominated resolution, the LVM's resolution is fundamentally limited by the fiber diameter ($D_{fib} \approx 35.3''$). The authors determine that setting the kernel size $\sigma$ smaller than a specific fraction of the fiber diameter leads to oversampling artifacts.
3. 3D Cube Reconstruction: The 2D interpolation is repeated for every spectral pixel (3600–9800 Å, $\Delta\lambda = 0.5$ Å). The pipeline also includes optional modes for direct 2D map reconstruction (bypassing the full cube) and converting the reconstructed cube back into a spatially co-added RSS format for input into the LVM Data Analysis Pipeline (LVM-DAP).

Key Contributions and Analysis

• Kernel Optimization via $\chi^2$ Analysis: The authors quantified the optimal reconstruction parameters by comparing reconstructed LVM images of NGC 1365 against high-resolution Digitalized Sky Survey (DSS) images. They found that while smaller $\sigma$ and larger $\alpha$ reduce the $\chi^2$ residual, values where $\sigma < 1/4 D_{fib}$ result in oversampling artifacts (visible dither patterns) without recovering additional physical structure. The recommended limit is $\sigma = 8.8''$ ($1/4 D_{fib}$) with $\alpha=2$.
• Resolution Limits and Deconvolution: The study demonstrates that the effective spatial resolution of the reconstructed cubes is a quadratic combination of the fiber diameter and the kernel size ($\sqrt{D_{fib}^2 + 8 \ln 2 \sigma^2}$). Attempts to deconvolve the images using a Gaussian kernel to recover structures below the fiber aperture resulted in spurious structures, confirming that the fiber size sets a hard physical limit on recoverable spatial information for the LVM.
• Noise and Covariance Characterization: The paper details the propagation of errors, noting that the interpolation introduces spatial covariance between neighboring spaxels within a scale of $r \sim 2\sigma$. The authors provide a method to compute the full spatial covariance tensor and a 2-point correlation function, warning that standard aperture photometry assuming independent errors will overestimate the signal-to-noise ratio (S/N).
• Signal-to-Noise (S/N) Gain: The authors analyze the S/N gain from combining $N$ dithers. They show that the gain is not a uniform $\sqrt{N}$ factor but depends on the kernel size relative to the dither offsets. For large kernels ($\sigma \gg D_{fib}$), the gain approaches the theoretical $\sqrt{N}$. For the recommended high-resolution kernel ($\sigma = 8.8''$), the gain is approximately 77% of $\sqrt{N}$ due to the steep weighting of distant fibers, representing a trade-off between spatial resolution and S/N depth.

Results
The authors applied 3dcubegen to approximately 6 million spectra derived from ~6,000 dithered observations of the Large and Small Magellanic Clouds (LMC/SMC) and 26 Local Volume galaxies.

• Data Products: The reconstruction produced ten sets of data cubes with spatial resolutions ranging from ~1.1 kpc to ~4.9 pc (at the distance of the LMC).
• Visualizations: Reconstructed narrowband images of the LMC and SMC (using the Hubble palette: [O III], H$\alpha$, [S II]) demonstrate significant improvements in spatial sampling and the ability to resolve complex ionization morphologies.
• Spectral Analysis: The pipeline successfully generated co-added RSS spectra suitable for the LVM-DAP, enabling parametric fits of stellar populations and emission lines (velocity shifts, dispersions, fluxes) at resolutions of ~17 pc (LMC) and ~21 pc (SMC).

Significance
The paper claims that 3dcubegen provides a robust and scalable solution for LVM data cube reconstruction, essential for exploiting the scientific potential of the SDSS-V Local Volume Mapper. By preserving flux and propagating uncertainties correctly, the tool enables detailed studies of ionized gas, stellar populations, and kinematics across extended regions. The work establishes that while the LVM's large fibers impose a resolution floor, the 9-point dither strategy combined with optimized kernel selection allows for the recovery of spatial information up to the physical limit of the fiber aperture, significantly enhancing the survey's sensitivity to faint emission features compared to single-pointing observations. The methodology offers a key tool for the community to analyze the unprecedented volume of IFS data being generated by the LVM.