Reading the dust around aging stars

An aging star surrounded by dust presents an interpretation problem. Its observed light comes not just from the stellar surface, but from radiation absorbed, scattered and re-emitted by the surrounding material. How can astronomers work backwards from that mixture to estimate the star’s properties and the rate at which it sheds mass?

In a preprint, Hamidreza Mahani, Atefeh Javadi and Amirhossein Rabiee describe two software pipelines for this task: SEDust and its extension, MCM-SEDust. Their comparison illustrates why fitting a star’s light is not necessarily enough to determine its mass-loss rate uniquely.

Why individual dusty stars matter

The work focuses on asymptotic giant branch, or AGB, stars: evolved stars whose winds can carry material into their surroundings, where circumstellar dust forms. That dust absorbs shorter-wavelength light and re-emits energy in the infrared.

The distribution of a source’s light across wavelengths is its spectral energy distribution, or SED. Interpreting this distribution matters beyond individual stars. Models of entire stellar populations combine contributions from stars of different ages and compositions, together with the effects of dust. Uncertain descriptions of luminous, dusty evolutionary stages therefore feed into estimates of galaxy properties.

The authors’ approach starts at the smaller scale: model individual stars, estimate their contributions, and use those results to constrain the material returned by a population.

Two ways to search for a fit

Both pipelines use DUSTY, a radiative-transfer program that calculates how light travels through a spherical dust shell. It accounts for absorption, scattering and thermal emission. The adopted wind model also calculates the density structure and final wind speed internally.

SEDust searches a precomputed grid of models. Each grid point represents a particular combination of stellar temperature, temperature at the shell’s inner edge and optical depth. Optical depth measures how strongly material obstructs light at a specified wavelength.

The pipeline compares predicted and measured brightnesses, allowing an overall brightness adjustment, and chooses the model with the smallest uncertainty-weighted mismatch. This makes automated fitting of large catalogues practical, but restricts the answer to available grid points.

MCM-SEDust instead runs new DUSTY calculations while exploring those three parameters. It uses Markov chain Monte Carlo sampling: a statistical method that samples combinations according to how well they agree with the observations and the adopted prior assumptions. The resulting posterior distribution describes which combinations remain plausible within that framework, rather than returning only one best-fitting grid point.

The extension compares brightnesses in logarithmic space and calculates the overall brightness adjustment analytically. Measurements without published uncertainties receive an assumed fractional uncertainty of 15%. The authors check whether sampling adequately explores the distributions and flag fits that fail their checks. The tradeoff is computation: thousands of radiative-transfer calculations may be needed for one source.

Similar brightness, different mass loss

The numerical comparison centres on one example AGB star in IC 10, identified as 24472. The authors fit it using separate carbon-rich and oxygen-rich dust assumptions.

For carbon-rich dust, SEDust gives a stellar temperature of 2,400 kelvin, compared with a posterior median of 2,356 kelvin from MCM-SEDust. The inferred luminosities are approximately 70,100 and 65,000 times the Sun’s luminosity—a difference of about 7%.

The mass-loss estimates differ much more: 0.000027 versus 0.000013 solar masses per year, roughly a factor of two. With oxygen-rich dust, the direction reverses: MCM-SEDust gives a higher mass-loss rate than SEDust.

The lesson is not that one method always predicts more mass loss. Rather, comparable luminosities and optical depths can accompany different mass-loss estimates because the inferred rate also depends on dust temperature and the radiative-transfer solution.

What the comparison cannot settle

This is an illustrative comparison, not a broad demonstration of accuracy across many stars. The Bayesian uncertainty intervals also remain conditional on the chosen dust properties, spherical-shell model, prior ranges and measurement uncertainties. Exploring parameter uncertainty does not remove those assumptions.

There is a reporting inconsistency worth noting. The methods section describes a SEDust grid with inner dust temperatures of 500–1,500 kelvin, but the results table lists grid solutions at 100 and 200 kelvin. The text calls 100 kelvin the grid’s lower boundary. The supplied account does not reconcile these descriptions, so the temperature comparison needs clarification.

The broader goal is to connect individual stellar winds with galactic dust budgets. Incorporating those empirical constraints into stellar population models remains future work, rather than a completed result of this preprint.