[Paper Review] Confidence levels of evolutionary synthesis models

M. Cerviño, V. Luridiana+1 more|arXiv (Cornell University)|Jun 30, 2000

Stellar, planetary, and galactic studies20 citations

TL;DR

This paper quantifies the intrinsic statistical uncertainties in evolutionary synthesis models due to the stochastic nature of the initial mass function (IMF) in young stellar clusters. Using Monte Carlo simulations of IMF sampling, it shows that for a 10⁵ M⊙ cluster, the 90% confidence interval for Hβ equivalent width (EW(Hβ)) spans ±18% around the analytical value at 3.5–5 Myr, highlighting that model uncertainties from IMF sampling are significant and must be accounted for in observational comparisons.

ABSTRACT

The stochastic nature of the IMF in young stellar clusters implies that clusters of the same mass and age do not present the same unique values of their observed parameters. Instead they follow a distribution. We address the study of such distributions, parameterised in terms of their confidence limits, in evolutionary synthesis models. These confidence limits can be understood as the inherent uncertainties of the synthesis models. Here we concentrate on some parameters such as EW(Hb) in emission. For instance, we show that for a cluster where 10^5 Mo have been transformed into stars, the dispersion of EW(Hb) is about 18% within the 90% confidence levels at ages between 3.5 and 5 Myrs.

Motivation & Objective

• To quantify the intrinsic statistical uncertainties in evolutionary synthesis models arising from the stochastic sampling of the initial mass function (IMF) in young stellar clusters.
• To assess how these uncertainties affect key observable parameters such as EW(Hβ), ionizing photon rate Q(H⁰), and the WR bump to Hβ luminosity ratio.
• To determine whether the dispersion in model outputs due to IMF sampling is independent of the synthesis code and represents a fundamental lower bound on model uncertainties.
• To evaluate how the width of parameter distributions scales with cluster mass and whether composite systems of smaller clusters reproduce the statistical behavior of single massive clusters.

Proposed method

• Employed Monte Carlo realizations of the IMF to generate synthetic stellar populations with fixed total masses (10³, 10⁴, and 10⁵ M⊙) in the 2–120 M⊙ range.
• Used the same evolutionary synthesis code (updated from Cerviño & Mas-Hesse 1994) with Meynet et al. (1994) solar-metallicity tracks and compared results with Leitherer et al. (1999) code for consistency.
• Performed 600, 400, and 200 Monte Carlo realizations for 10³, 10⁴, and 10⁵ M⊙ clusters, respectively, to sample the distribution of observable parameters.
• Computed 68% and 90% confidence intervals (equivalent to 1σ and 90% central intervals) for key observables like EW(Hβ), Q(H⁰), and L(WRbump)/L(Hβ) from the resulting distributions.
• Simulated composite 10⁵ M⊙ systems by summing 6 × 100 clusters of 10³ M⊙ and 6 × 10 clusters of 10⁴ M⊙ to test statistical convergence.
• Compared the resulting distributions with analytical model predictions and assessed the consistency of stochastic results with non-detection of WR stars.

Experimental results

Research questions

• RQ1 How do the statistical fluctuations in the IMF affect the distribution of observable parameters in evolutionary synthesis models?
• RQ2 What is the magnitude of the 90% confidence interval for EW(Hβ) in a 10⁵ M⊙ cluster at 3.5–5 Myr, and how does it compare to the analytical model value?
• RQ3 To what extent do the observed parameter distributions in composite clusters (e.g., multiple subgroups) converge toward the analytical model values compared to single massive clusters?
• RQ4 Are the observed dispersions in parameters like Q(H⁰) and L(WRbump)/L(Hβ) consistent with the stochastic sampling of the IMF, and can they be explained by the combination of individual uncertainties?
• RQ5 How do the confidence intervals of key observables scale with the total mass transformed into stars, and what does this imply for model-data comparisons?

Key findings

• For a 10⁵ M⊙ cluster, the 90% confidence interval for EW(Hβ) spans approximately ±18% around the analytical value at ages between 3.5 and 5 Myr.
• The 68% confidence level for the L(WRbump)/L(Hβ) ratio is compatible with no detection of Wolf-Rayet stars, indicating that stochastic sampling can produce non-detections even when the analytical model predicts a strong WR bump.
• The dispersion in EW(Hβ) is proportional to the square root of the number of stars, and thus scales inversely with cluster mass, with the 90% confidence interval width decreasing as cluster mass increases.
• Composite systems made of multiple smaller clusters (e.g., 6 × 10⁴ M⊙ or 60 × 10³ M⊙) show parameter distributions that fall within the 90% confidence interval of the single 10⁵ M⊙ cluster, but not within the 68% interval, indicating statistical convergence at the 90% level.
• The intrinsic stochasticity of the IMF introduces a fundamental lower bound on model uncertainties that is independent of the synthesis code, and must be considered when comparing models to observations.
• The observed dispersion in Q(H⁰) and EW(Hβ) is consistent with the combined dispersion from L(Hβ) and Q(H⁰), confirming internal consistency of the stochastic model approach.