A complete mid-infrared spectrum at the galactic centre moves IRS 3 into the oxygen-rich column

A spectrum from inside the inner parsec

IRS 3 sits about 0.17 parsecs in projection from Sagittarius A*, which is roughly 0.55 light-years, close enough that the black hole's radiation field is part of the star's weather. It is the brightest L band source in the galactic centre and the most prominent asymptotic giant branch star inside the inner parsec, which is to say a dying star far enough along to be shedding its outer layers into a neighbourhood no one expected to be hospitable to what it sheds. In 2025, MIRI's Medium Resolution Spectrometer aboard Webb pointed at it under the guaranteed time programme MICONIC, and for the first time collected a continuous mid-infrared spectrum of this star rather than a set of samples at chosen wavelengths.^[1]

What that continuity buys is visible immediately. The dereddened spectrum carries a broad silicon-oxygen stretching feature at 9.7 micrometres with an optical depth of 2.98, and the matching bending mode at 18.5 micrometres at 0.85. Their ratio, 3.5 with an uncertainty of 0.1, is a chemical fingerprint: a carbon-rich star has no analogue to those two modes and cannot make that ratio. IRS 3 had been proposed as carbon-rich in 2008. One spectrum, taken whole, moves it to the oxygen-rich side, and that reassignment is the firmest thing in the paper.^[1]

Where the envelope comes from

The layered shell of dust that the coverage describes is a model output, arrived at by running the three-dimensional Monte-Carlo radiative transfer code Hyperion across a large grid and keeping what reproduces the spectrum. The surviving configuration has an inner region at 1,200 kelvin, then shells at 280 to 300 kelvin near 949 astronomical units, 180 kelvin near 2,214, and 80 to 100 kelvin near 6,325, with the envelope reaching about 10,000 astronomical units and the star radiating 60,000 times as much as the Sun. The dust is alumina and amorphous silicate. None of this is a picture; it is the arrangement of material that, given the code's assumptions, would send us the light we received.^[1]

The mass-loss rate carries more assumption still. The authors reach roughly six hundred-thousandths of a solar mass a year by measuring how far out the star's wind is stopped by the surrounding medium, then assuming a wind speed typical of these stars, about 15 kilometres per second. They say plainly that telescope diffraction makes their stand-off distance an upper limit, that they do not exclude local departures from that wind speed, and that the gas density in the inner parsec is uneven. A different local density, or a wind moving at a different speed, would give a different rate from exactly the same spectrum, so the figure describes the model's best fit rather than a quantity the observation pins down. The authors are equally candid that the shells themselves could come from mass loss that varies over time, from a tidal interaction, or from a companion star, and that the spectrum does not separate those.^[1]

Why does the water carry less weight?

The water is identified in a narrow stretch of the spectrum, between 6.0 and 6.25 micrometres, by matching the absorption against synthetic spectra from the HITRAN database. The fit returns a temperature of 700 kelvin and a column density of 1.5 times 10 to the power 17 per square centimetre, which is a count of molecules along the line of sight and not a measure of how much water the envelope holds. The reporting was right to say the observations do not reveal how much water is present; the paper's own numbers say the same thing in the other direction, by giving a column and stopping there.^[1]

And the authors go further than their own headline. They write that the exact nature of the absorption around 6.06 micrometres and between 6.12 and 6.25 micrometres remains unclear, and they list what else could sit there: ice such as ammonia or the ammonium ion, large molecules, or simply a residue left behind by the correction for foreground extinction. That last possibility is the prosaic one, and it deserves its strongest version, because the same correction is already known to be imperfect here. The silicate feature comes out at longer wavelengths than any model run predicts, and emission between 5 and 9 micrometres needs a component the spherical model does not contain. A spectrum this good is worth having for what it settles about the dust. What it says about water is a first look at a band that the next independent spectrum will either confirm or dissolve.^[1]