The data & tools

The analysis used the ELODIE archive at Observatoire de Haute-Provence — a 1.93m telescope with a fiber echelle spectrograph at R~42,000. Spectra were retrieved via SPLAT-VO, a virtual observatory tool with a manual GUI for spectrum inspection and normalization.

Stellar atmosphere modeling used SPECTRUM (R. O. Gray) with ATLAS9/Kurucz model grids, and equivalent widths were measured using the BLACKWEL and ABUNDANCE programs — a C++/GUI toolchain developed for classroom use at the time.

The solar reference was Lodders (2003), and no NLTE corrections were applied.

The targets

HD 89307 — a G0V solar analog selected as the cleanest exoplanet host star available in ELODIE. The goal was to establish a baseline comparison star as close to solar as possible, then examine how the chemistry deviated in the other targets.

Gliese 581 — an M3 red dwarf with two candidates in the habitable zone at the time (Gliese 581c and 581d). The low effective temperature (~3500 K) made ATLAS9 modeling unreliable for an M dwarf, and the analysis showed this clearly.

The Sun served as the primary calibration anchor, as it always must.

What the 2010 data said

Despite the limitations of the toolchain, several results held up and a few anomalies appeared that have gone unresolved ever since.

EXOPLANET CODEX — METHODOLOGY

From thesis to frontier:
15 years of pipeline evolution

In 2010 Ryan measured stellar abundances using SPECTRUM and the ELODIE archive for his astrophysics thesis. The Exoplanet Codex rebuilds that work on a modern foundation — better data, better physics, open science.

2010
Senior thesis
2014
iSpec + HARPS era
2021
Asplund 3D NLTE
solar scale
2024
JWST rocky
planets
2026
Codex v1 ← now
Future
3D NLTE + M dwarfs
higher spectral resolution
HARPS vs ELODIE
27
elements measured
vs a limited subset
0.05
dex target precision
vs ~0.1–0.2 dex typical
WHAT CHANGED AND WHY IT MATTERS
2010 THESIS
EXOPLANET CODEX (2026)
Spectrograph
ELODIE R~42,000
Ceres/Vesta solar reference
Spectrograph
HARPS R~115,000
Direct solar feed (Dumusque 2021)
eliminates asteroid systematics
Radiative transfer
SPECTRUM (Gray & Corbally 1994)
No molecular lines
Radiative transfer
Turbospectrum (Plez 2012)
Native molecular + NLTE
correct Mg/Si — 0.1–0.15 dex gain
Model atmospheres
ATLAS9 — FGK only
1D plane-parallel, LTE
Model atmospheres
iSpec + Turbospectrum / ATLAS9 + MARCS spherical
FGK through M5 dwarfs
M dwarfs unlocked (LHS 3844)
NLTE corrections
Not applied
NLTE corrections
C, O, Na, Li, Mg, Ba
(Amarsi+2016/19; Lind+2009/11)
C/O ratio science-grade
EW measurement
Gaussian profiles
Manual continuum
Binary blend flag
EW measurement
Voigt + Gaussian fallback
Adaptive windows + σ-clip
Continuous quality score 0–1
Bayesian sampler for tricky lines
Uncertainty
Basic error propagation
Uncertainty
Type A + Type B solar zero-point
Full quadrature budget per element
publishable error bars
Reproducibility
Manual, thesis only
Reproducibility
Git-versioned, automated pipeline
Open science at exoplanetcodex.org
fully reproducible
Elements measured
13
Elements measured
27 (canonical list: Fe I/II, C, O, N, Mg, Si, S, Ca, Ti, Co, Ni, Na, Al, K, P, Ba, Y, Eu, Mn, Cr, V, Sc, Cu, Zr, Li, Sr)

"The same wall I hit in 2010 — optical HARPS can’t recover carbon well — is still there. The difference now is that we know exactly why, and we have Turbospectrum, UV composites, and NLTE corrections to work around it."

— Ryan Schmitt, Exoplanet Codex
WHY THIS MATTERS FOR ROCKY PLANET SCIENCE

The C/O ratio of a host star sets the mineralogy of its planets. A star with C/O > 0.8 builds carbon-rich worlds (graphite, SiC); below that threshold you get silicate-dominated planets like Earth. SPECTRUM-era pipelines had ~0.1–0.15 dex systematic errors on Mg and Si — large enough to shift a planet from "rocky like Earth" to "different composition" in the models.

Turbospectrum + NLTE corrections + astrophysical log gf calibration targets <0.05 dex precision on the key elements. That’s the threshold where stellar chemistry becomes a reliable input to planetary interior models.

Sixteen years later

What held up
  • The core physics. Stellar atmosphere theory, LTE line formation, and the Kurucz model grid are still the backbone of precision abundance work in 2026.
  • The ionization equilibrium method. Balancing Fe I and Fe II to derive log g is still the gold standard. We were doing it right in 2010, just with too few lines.
  • The scientific question. Alpha enhancement as a window into planet formation is still an active research area. Papers continue to be published on [Mg/Fe], [Si/Fe], and the Mg/Si ratio as proxies for rocky planet interior structure.
  • HD 89307’s Fe abundance. The literature value for HD 89307 ([Fe/H] ≈ +0.02) aligns well with the 2010 result. The solar-analog conclusion was correct.
What we’d do differently
  • Never analyze M dwarfs with ATLAS9. Gliese 581 required a PHOENIX or MARCS model grid designed for cool stars. The anomalous abundances are almost certainly model artifacts, not astrophysics.
  • Use at least 40 Fe lines. Ten lines is not enough to reliably establish excitation equilibrium. The stellar parameter uncertainties were dominated by line-selection noise.
  • Document the uncertainty budget formally. The 2010 thesis reports scatter but not stellar parameter sensitivity (ΔTeff, Δlog g, Δvturb contributions). This is now standard practice.
  • Use a version-controlled, automated pipeline from day one. Manual GUI workflows make results difficult to reproduce and impossible to re-run when a new line list or solar reference supersedes the old one.