PaperlayerView exact revision on arXiv

Source-linked AI summary

Determining stellar atmospheric parameters and chemical abundances of FGK stars with iSpec

S. Blanco-Cuaresma, C. Soubiran, U. Heiter, P. Jofré

arXiv:1407.2608v1astro-ph.IMastro-ph.SR

TL;DR

Large spectroscopic surveys require automated derivation of stellar parameters and abundances. This paper presents iSpec, which implements synthetic fitting and equivalent-width pipelines and validates them with Gaia FGK benchmark stars. The synthetic-fitting pipeline gives more accurate and precise results and remains more effective at lower resolution and signal-to-noise.

  • Problem

    Large spectroscopic surveys produce extensive datasets requiring automated methods to derive atmospheric parameters and individual chemical abundances.

  • Method

    iSpec combines synthetic spectral fitting and equivalent-width pipelines and compares their outputs with Gaia FGK benchmark reference values.

  • Results

    The synthetic spectral-fitting pipeline provides more accurate and precise results and is more effective than equivalent widths at resolution 7500 and signal-to-noise ratio 25.

  • Takeaways & Limitations

    The Gaia FGK benchmark library can assess and optimize spectroscopic pipelines, supporting more reliable and comparable analysis results.

  • Takeaways & Limitations

    Chemical-abundance validation was limited to neutral and ionized iron because it was the only element with an available reference value.

Abstract

from arXiv · show

Context. An increasing number of high-resolution stellar spectra is available today thanks to many past and ongoing extensive spectroscopic surveys. Consequently, the scientific community needs automatic procedures to derive atmospheric parameters and individual element abundances. Aims. Based on the widely known SPECTRUM code by R. O. Gray, we developed an integrated spectroscopic software framework suitable for the determination of atmospheric parameters (i.e., effective temperature, surface gravity, metallicity) and individual chemical abundances. The code, named iSpec and freely distributed, is written mainly in Python and can be used on different platforms. Methods. iSpec can derive atmospheric parameters by using the synthetic spectral fitting technique and the equivalent width method. We validated the performance of both approaches by developing two different pipelines and analyzing the Gaia FGK benchmark stars spectral library. The analysis was complemented with several tests designed to assess other aspects, such as the interpolation of model atmospheres and the performance with lower quality spectra. Results. We provide a code ready to perform automatic stellar spectral analysis. We successfully assessed the results obtained for FGK stars with high-resolution and high signal-to-noise spectra.

1. Introduction

iSpec addresses the need for automated stellar-spectrum analysis by unifying synthetic spectral fitting and equivalent-width methods in a flexible framework. Its pipelines are assessed against Gaia FGK benchmark stars.

  • High-resolution surveys generate large spectral datasets that require automated procedures to derive atmospheric parameters and abundances.
  • iSpec was developed to derive atmospheric parameters using synthetic spectral fitting and equivalent-width techniques.
  • The framework supports homogeneous analyses using shared continuum normalization, model atmospheres, atomic information, and the SPECTRUM radiative-transfer code.
  • Synthetic fitting compares observed spectra with spectra generated on-the-fly and uses least-squares minimization, with selected spectral regions available to reduce computation.
  • The equivalent-width pipeline fits Gaussian profiles to Fe I/II lines and determines atmospheric parameters through excitation equilibrium and ionization balance.
  • The Gaia FGK benchmark library is used to compare derived atmospheric parameters and abundances with reference values for pipeline assessment.

2.1. Data treatment

iSpec integrates common spectral-treatment operations before parameter and abundance analysis, while allowing third-party tools to perform these steps. The Gaia FGK benchmark library demonstrates these capabilities.

  • Continuum normalization identifies continuum points with median and maximum filters, fits a polynomial or splines, and divides fluxes by the fitted continuum.
  • The framework can degrade spectral resolution by Gaussian convolution, derive radial velocity by cross-correlation, and identify telluric-line positions.
  • These operations can be applied within iSpec or performed with third-party software before subsequent analysis.
  • The Gaia FGK benchmark stars library was created with iSpec and illustrates its data-treatment capabilities.

2.2. Line selection

Line selection is treated as an important analysis choice because the selected absorption lines affect atmospheric-parameter and abundance results. iSpec supports assessing spectral-region quality before analysis.

  • The selected absorption lines affect the resulting atmospheric parameters and abundances, so atomic-data reliability, including oscillator strengths, must be considered.
  • iSpec supports custom line selection through line synthesis, theoretical equivalent-width calculations, and visual comparison in its user interface.
  • The framework also facilitates quality assessment of spectral regions affected by noise, cosmic rays, and telluric lines.

2.3. Spectral synthesis and abundances from equivalent widths

iSpec uses SPECTRUM-compatible atomic data, abundances, and model atmospheres to synthesize spectra and derive equivalent-width abundances. It expands atmosphere grids through interpolation and extrapolation for parameter-space coverage.

  • iSpec uses SPECTRUM to generate synthetic spectra and determine abundances from equivalent widths.
  • Its line lists encode wavelengths, species, excitation energies, oscillator strengths, broadening adjustments, and transition types.
  • Missing upper excitation energies are calculated from lower energies and line wavelength using physical constants and unit conversions.
  • Ready-to-use line lists cover 300–1100 nm, including SPECTRUM and VALD-derived lists.
  • iSpec scales selected solar abundances to target metallicity and can fix individual abundances to unscaled values for synthetic fitting.
  • MARCS and ATLAS atmospheres are included, with MARCS spanning 2500–8000 K, surface gravity 0.00–5.00 dex, and metallicity −5.00–1.00 dex.
  • Pre-computed grids contain missing parameter combinations and relatively coarse steps, so iSpec completes them using interpolation and extrapolation.
  • Atmosphere quantities are resampled onto a common optical-depth scale, and extrapolated values are bounded by extrema from real atmospheres.

2.4. Atmospheric parameter determination

iSpec determines atmospheric parameters and chemical abundances through synthetic spectral fitting or equivalent widths, using iterative least-squares procedures and diagnostic equilibria.

  • iSpec offers synthetic spectral fitting and equivalent width methods for determining atmospheric parameters and individual chemical abundances.
  • The nonlinear least-squares algorithm iteratively updates free parameters using finite-difference Jacobians until a χ2 convergence threshold or iteration limit is reached.
  • Synthetic fitting compares observed spectra with on-the-fly synthetic spectra and minimizes their differences, optionally restricting the comparison to selected spectral regions.
  • The equivalent width method excludes rotational and macroturbulent broadening because it uses line areas rather than absorption-profile shapes.
  • Robust M-estimator fitting identifies abundance outliers before minimization, discarding measurements with assigned weights below 0.90.
  • Equivalent-width analysis derives abundances from fitted Fe I/II line areas and adjusts parameters to enforce excitation equilibrium, reduced-equivalent-width balance, and ionization balance.

2.5. Error estimation

iSpec estimates atmospheric-parameter errors from the nonlinear-fit covariance matrix, but the resulting uncertainties depend strongly on the quality of the supplied flux-error estimates.

  • Atmospheric-parameter errors are calculated from the covariance matrix constructed by the nonlinear least-squares fitting algorithm.
  • Underestimated spectral flux errors produce unrealistically low atmospheric-parameter uncertainties, whereas ignoring flux errors generally overestimates them.
  • When enough lines are available, chemical-abundance internal errors are estimated from the standard deviation of individually derived line abundances.

3. Pipeline description and validation

The study develops two iSpec pipelines based on synthetic fitting and equivalent widths, with line selection and observation-specific verification used to support parameter determination.

  • Two pipelines were developed using synthetic spectral fitting and the equivalent width method, with tests covering different configurations such as model atmospheres.
  • The analysis used GES atomic data covering 475–685 nm, including medium- and high-quality lines for iron and other elements.
  • Selected absorption lines were fitted with Gaussian profiles and automatically rejected when affected by shifts, poor fits, tellurics, or invalid fluxes.
  • The equivalent-width pipeline filtered weak and strong lines because weak lines are noise-sensitive while strong lines are commonly blended and broadening-sensitive.
  • A synthetic spectral grid provided initial parameters for separating giants from dwarfs and metal-rich from metal-poor stars; rotation was fixed at 2 km s−1.
  • Observation verification adapted the analysis to each spectrum by retaining only the highest-quality regions for final parameter determination.

3.2. Atmospheric parameters determination

iSpec pipelines were tuned on Gaia FGK benchmark stars, prioritizing low dispersion and surface-gravity precision. Synthetic spectral fitting achieved smaller reported errors than the equivalent width method, while equivalent-width performance degraded at lower resolution.

  • Pipeline evaluation: The pipelines were evaluated against Gaia FGK benchmark reference values, prioritizing low dispersion and especially surface-gravity precision.The authors considered low dispersion easier to correct than systematic offsets and identified surface gravity as the most difficult parameter to derive spectroscopically.
  • Synthetic spectral fitting: Synthetic spectral fitting produced average errors of 19 K in effective temperature, 0.05 dex in surface gravity, and 0.02 dex in metallicity.The corresponding average dispersions for stars with multiple observed spectra were 15 K, 0.06 dex, and 0.01 dex, respectively.
  • Equivalent width method: Outlier filtering improved accuracy and precision for all atmospheric parameters, while Gaussian line profiles performed slightly better than Voigt profiles.The equivalent width tests also preferred medium-quality iron-line selection and a relaxed reduced-equivalent-width filter.
  • Method comparison: At lower spectral resolution, blending makes the equivalent width method over-estimate abundances and provide poorer results.The method was also highly sensitive to its starting point, making a good initial guess strongly recommended.
  • Equivalent width method: The equivalent width method produced average errors of 67 K in effective temperature, 0.13 dex in surface gravity, and 0.09 dex in metallicity.The corresponding average dispersions for stars with multiple observed spectra were 38 K, 0.12 dex, and 0.03 dex, respectively.
  • Method comparison: The comparison used the same normalization process, atomic data, model atmospheres, and SPECTRUM radiative-transfer code for both methods.Under this controlled comparison, synthetic spectral fitting was reported as more accurate and precise and became the preferred iSpec strategy.

3.3. Chemical abundances

iSpec determines individual iron abundances using a line-by-line differential analysis, but this validation was limited to neutral and ionized iron. Both methods agreed better with reference values after differential treatment, while metal-poor stars remained overestimated.

  • The analysis was limited to neutral and ionized iron because only iron had an available reference value.
  • iSpec derives differential abundances by calibrating selected lines against solar spectra, subtracting solar abundances, and taking the median relative abundance.
  • Differential treatment improved agreement with LTE and NLTE iron references, partly because some NLTE effects seemed to cancel for solar-type stars.
  • Both methods overestimated iron abundances for metal-poor stars, with a trend depending on the reference iron abundance.
  • A 1% lower continuum placement reduced overall metallicity by ∼0.07 dex in the synthetic spectral fitting pipeline.

4. Additional validations

Additional validations examined model-atmosphere interpolation, iterative convergence, parameter correlations, and performance on lower-quality spectra. Interpolation preserved spectral behavior closely, while synthetic fitting was generally more robust than equivalent widths under difficult conditions.

  • 4.1. Model atmosphere interpolation: Interpolated model atmospheres reproduced the overall shape of pre-computed models with a small shift.
  • 4.1. Model atmosphere interpolation: The average flux difference between interpolated and pre-computed synthetic spectra was −0.22%±2.12%, while equivalent-width differences averaged 0.10%±6.32%.
  • 4.2.1. Iterations: Synthetic spectral fitting produced good average results around iterations five to six, whereas equivalent-width results stabilized around iterations nine to ten.
  • 4.2.1. Iterations: With a maximum of 10 iterations, average computation time was 36 minutes per spectrum for synthetic fitting and about 19 minutes for equivalent widths.
  • 4.2.2. Correlations: A 0.50 dex surface-gravity change implied effective-temperature differences of 129 K and 157 K for synthetic fitting and equivalent widths, respectively.
  • 4.3. Signal-to-noise ratios and spectral resolutions: Synthetic fitting remained similar to reference parameters at signal-to-noise ratio 25, while equivalent widths deviated even at higher signal levels.
  • 4.3. Signal-to-noise ratios and spectral resolutions: Equivalent-width performance was sensitive to noise, continuum placement, starting parameters, and blended lines at lower spectral resolution.

5. Conclusions

iSpec is an integrated framework for deriving stellar atmospheric parameters and individual abundances through synthetic fitting and equivalent-width pipelines. Tests on Gaia FGK benchmark spectra found synthetic fitting more accurate, precise, and effective for lower-quality data, while the benchmark library supported pipeline assessment.

  • iSpec integrates functions for determining effective temperature, surface gravity, metallicity, and individual chemical abundances using SPECTRUM.
  • Two iSpec pipelines based on synthetic spectral fitting and equivalent widths were assessed using high-resolution, high signal-to-noise Gaia FGK benchmark spectra.
  • The synthetic spectral fitting pipeline provided more accurate and precise results, while parameter correlations were similar across both techniques.
  • Synthetic fitting was more effective than equivalent widths at resolution 7500 and signal-to-noise ratio 25.
  • The Gaia FGK Benchmark Star library can be used to assess and optimize spectroscopic pipelines for more reliable and comparable results.
Loading 1407.2608v1…