Empirical SpecMatch: Stellar Characterization

• Empirical SpecMatch is a family of methods that determines stellar parameters by directly matching observed spectra to calibrated, high-fidelity empirical libraries.
• The approach involves rigorous preprocessing, including blaze correction, continuum normalization, and wavelength registration, followed by χ² minimization for template matching.
• Composite spectrum formation through weighted interpolation enables precise estimates of Tₑff, [Fe/H], log g, R★, and v sin i, though library sparsity can limit performance near parameter boundaries.

Empirical SpecMatch refers to a family of methodologies for stellar parameter determination that leverage direct spectral comparison to high-fidelity empirical libraries, rather than synthetic spectral models. The canonical implementations—SpecMatch-Emp (Yee et al., 2017), its derivatives such as NEIDSpecMatch (Han et al., 20 Mar 2025), and HPFSpecMatch—are optimized for FGKM stars and operate by matching observed spectra against a grid of well-characterized stellar templates. These approaches facilitate robust inference of key parameters ($T_\mathrm{eff}$, $[\mathrm{Fe/H}]$, $\log g$, $R_\star$, and $v\sin i$) with minimal reliance on atmospheric modeling, offering distinct advantages for late-type stars and observational regimes poorly sampled by synthetic codes.

1. Empirical Library Construction and Calibration

The foundation of Empirical SpecMatch pipelines is a carefully curated spectral library. For SpecMatch-Emp (Yee et al., 2017), the library contains 404 HIRES spectra at $R \sim 60{,}000$ spanning $T_\mathrm{eff}$ $\sim 3000$–$7000$ K, $[\mathrm{Fe/H}]$ $[\mathrm{Fe/H}]$0 to $[\mathrm{Fe/H}]$1, and $[\mathrm{Fe/H}]$2 $[\mathrm{Fe/H}]$3–$[\mathrm{Fe/H}]$4, with parameter values from interferometry, asteroseismology, and LTE synthesis. NEIDSpecMatch (Han et al., 20 Mar 2025) utilizes a tailored subset: 78 NEID spectra ($[\mathrm{Fe/H}]$5), $[\mathrm{Fe/H}]$6, and parameters tied to SpecMatch-Emp catalog values.

Library spectra are uniformly processed to ensure consistent calibration. HIRES spectra are blaze-corrected via division by a rapidly-rotating B-star and rebinned onto a common logarithmic wavelength scale. NEID spectra use NEID DRP v1.3.0 for flat-fielding, blaze correction, laser frequency comb-referenced wavelength solution, and adaptive telluric modeling. Each echelle order is continuum-normalized by low-order polynomial or cubic spline fits excluding deep lines and trimmed to avoid edge effects.

2. Preprocessing and Spectral Registration

Unknown target spectra undergo analogous preprocessing to maximize fidelity in inter-comparison. For both HIRES and NEID implementations, key steps include:

• Blaze function removal and continuum normalization using polynomials or splines.
• Wavelength resampling onto the uniform $[\mathrm{Fe/H}]$7 grid of the library.
• Segmental cross-correlation for precise radial velocity registration; pixel lag determination by parabola fitting to maximize template overlap.
• Telluric correction by masking fixed bands (e.g., O$[\mathrm{Fe/H}]$8 “A-band” at 6270–6310 Å) and emission features/star-specific masking (notably in active M dwarfs).

Multiple echelle orders and wavelength segments are processed independently, yielding order-specific parameter estimates.

3. Spectral Matching and χ² Metric

Central to Empirical SpecMatch is direct $[\mathrm{Fe/H}]$9-based evaluation of spectral similarity. For each library spectrum, after registration and continuum correction, the residual

$\log g$0

is computed, where $\log g$1 incorporates pixel-level flux uncertainty for NEIDSpecMatch (Han et al., 20 Mar 2025), while SpecMatch-Emp (Yee et al., 2017) uses unit weights. Local continuum mismatches are mitigated via high-pass filtering, and rotational broadening (Gray kernel up to $\log g$2 km s$\log g$3 for SpecMatch-Emp, stepwise for NEIDSpecMatch) is iteratively applied to each template. The minimal $\log g$4 spectra are identified.

4. Composite Spectrum Formation and Parameter Interpolation

Empirical SpecMatch does not select a single best-match template; rather, the five (or similarly selected) library spectra with the lowest $\log g$5 are linearly combined to synthesize a composite spectrum. The combination coefficients $\log g$6 are determined by inverse-$\log g$7 weighting (NEIDSpecMatch) or nonlinear least squares (SpecMatch-Emp), subject to a normalization constraint ($\log g$8 enforced via a Gaussian prior).

The stellar parameters ($\log g$9, $R_\star$0, $R_\star$1, $R_\star$2) associated with the target spectrum are interpolated as weighted sums over the selected library members:

$R_\star$3

A residual spectrum ($R_\star$4) is evaluated to ensure no systematic mismatches remain. For sparsely sampled library regions, multidimensional polynomial fitting or linear interpolation across neighbor parameters may be employed to minimize local bias (Han et al., 20 Mar 2025).

5. Rotational Broadening and v sin i Estimation

Projected rotational velocity ($R_\star$5) is determined by convolving the composite or individual templates with rotational kernels. The $R_\star$6 value that produces the lowest composite $R_\star$7 is adopted. Instrumental resolution limits precision: NEIDSpecMatch sets upper limits below 2.8 km s$R_\star$8, whereas SpecMatch-Emp is limited by the profile cap for resolutions $R_\star$9. For stars where $v\sin i$0 km s$v\sin i$1, broadening signatures are inadequately represented in the library, and results are excluded (Yee et al., 2017).

6. Validation, Calibration, and Uncertainty Estimation

Pipeline performance is quantified via leave-one-out cross-validation on the empirical library. Median uncertainties (standard deviation of recovered–archival values) across NEIDSpecMatch orders are $v\sin i$2 K, $v\sin i$3 dex, $v\sin i$4 (order 102 achieving up to 64 K, 0.161 dex, 0.042 dex respectively) (Han et al., 20 Mar 2025).

Systematic trends—particularly regression toward mean values near parameter boundaries—are corrected post hoc using polynomial fits, e.g., for $v\sin i$5 and $v\sin i$6 in SpecMatch-Emp (Yee et al., 2017). Empirical zero-point corrections ($v\sin i$70.05 dex, 50 K) may be applied as functions of $v\sin i$8 or color.

Algorithm robustness has been demonstrated for SNRs as low as 10/pixel; accuracy degrades only marginally (e.g., $v\sin i$9 by $R \sim 60{,}000$010 K) (Yee et al., 2017). Lowering resolution below $R \sim 60{,}000$1 reduces performance, with significant compromise in $R \sim 60{,}000$2 estimate and spectral matching.

7. Applicability, Strengths, and Limitations

Empirical SpecMatch techniques are most effective when the parameter domain of the target star is well covered by the empirical library and when both library and target spectra are acquired with the same instrument and reduction pipeline. This empirical approach avoids modeling uncertainties inherent to synthetic codes, particularly in cool star atmospheres—e.g., line list incompleteness in late K/M dwarfs (Yee et al., 2017, Han et al., 20 Mar 2025).

The dominant limitation is library sparsity, especially toward boundaries of $R \sim 60{,}000$3 and $R \sim 60{,}000$4, which can bias interpolated parameters toward average library properties. This is evident in cross-validation residuals and especially acute for very metal-poor, very cool, or chemically peculiar objects.

Extensions to additional spectrographs (e.g., HIRES, HARPS, CARMENES) necessitate instrument-tailored libraries with matching reduction protocols. Once such a library is established, the pipeline—order selection, normalization, $R \sim 60{,}000$5-based ranking, linear-combination composite formation, and cross-validated uncertainty estimation—transfers directly.

Summary Table: Library Statistics and Parameter Precision

Library	Size	Spectral Range ($R \sim 60{,}000$6)	Median Uncertainties
SpecMatch-Emp	404	3000–7000 K (M5–F1)	100 K, 15%, 0.09 dex (Yee et al., 2017)
NEIDSpecMatch	78	3000–6000 K (M–G/K)	115 K, 0.143 dex, 0.073 (Han et al., 20 Mar 2025)

Empirical SpecMatch methods provide a robust empirical pathway for high-precision stellar characterization, particularly for FGKM stars and late-type dwarfs, with documented resilience to moderate SNR and instrumental resolution, and are broadly adopted in both general stellar survey pipelines and specialized applications such as technosignature searches (Zuckerman et al., 2023).