Webb detects quartz crystals in clouds of hot gas giant
phys.org
phys.org
Also as a spectroscopist, I look at the error bars (especially as wavelength increases) and wonder how they make the assignments.
(Yes, I did read the paper)
Worth looking into, ie: request their raw data and crawl over their numerics.
The paper suggests a lot of 5 sigma et al rejections and iterative rolling medians, etc. all of which is all very fine save oft times in the execution via hand rolled exquisitely crafted one off filters that can converge and produce results even if "wrong" (off by one binning errors, etc. etc. etc...).
I'm not saying they're wrong, I am saying (as you likely know) it's easy to make small mistakes that can reinforce noise into signal.
I didn't say I was a good spectroscopist. ;-P
I kid, I kid.
On the one hand, a planet crosses its star. A craptillion miles away, JWST captures the starlight passing through a sliver of atmosphere and spectrally resolves it to whatever degree it can through multiple instruments. That's just incredible.
On the other hand, qualitative chemical analysis of gas mixtures is hard enough with high spectral resolution and these are low resolution spectra. Gas phase lineshapes in lab spectra can be Lorentzian, Gaussian, or usually Voigt (convolution of Lorentzian and Gaussian.) Teasing out the fits usually requires not just high resolution but very high SNR especially in the wings:
https://cefrc.princeton.edu/sites/g/files/toruqf1071/files/F...
Not only are the authors claiming to identify silicates but nanoparticles of quartz vs. other silicates and/or oxides using broad spectral bins and with big error bars. I just don't see how they get there no matter how much they "torture the data."
Having seen a few of these recent spectral announcements, I have to wonder what's really going on. JWST was conceptualized in the late 1980s before exoplanets were even found. AFAIR, it was going to be a "First Light Machine." Then as exoplanets became important and the JWST program ran into trouble, it feels like the exoplanetary characterization became integral to keeping the program sold.
Slide 26 of this presentation by exoplaneteers even states a goal:
"Reclaim JWST from cosmologists
– Win more than our fair share with a goal of capturing 25% of mission time"
https://exoplanets.nasa.gov/exep/files/exep/4%20Beichman.pdf
JWST is a magnificent beast but even at 6.5m, it can only collect so much and stare for so long. It feels like it is a pathfinder that will set requirements for future exoplanet-specific missions.
It all comes down to dwell time - the more statistics the better.
Of course it's not just down to time dwelt on target star X, they can work in all the data from all the stars that come from a similar family and share gross characteristics.
There's a sneaky trick in airborne gamma ray spectroscopy where, while a plane travels at (say) 70 m/s and you can only accumulate one seconds worth of ground source radiometric decay over any 70 m patch (along with cosmic gamma, gamma from aircraft and pilots wrist watch, gamma from gases, etc) .. that "one second snapshot" can be pimped the heck up by 'knowing' what tens of hundreds of thousands of 'typical' one second spectrums look like.
If your ground patch of interest is in the "typical" class, the result is greatly (and accurately sharpened) by training and if the ROI patch (region of interest) is atypical - lights up as "very much not like the others".
This, of course, has parallels in astro spectrums - that paper is littered with ThisName and ThatName inside references.
In a parallel earth based field there are papers like (say)
Minty | Hovgaard Reducing noise in gamma-ray spectrometry using spectral component analysis
I agree in principle about dwell time but for exoplanets that must be some combination of the eclipse time, JWST's intrinsic pointing stability, detector saturation, maximum time allotted on a fiercely oversubscribed telescope, etc.
For those that are still following along the additional exoplanet challenge is to first gather the best possible spectrum for the star .. and then get the best possible spectrum of "the difference" when a (?) planet passes across the star and the total light drops by some factor.
Complications, of which there are many, include having the best possible guess at the question of one planet or many, the same planet or a different one, etc.
Bigger filters have little filters than focus on the differences, Little filters have lesser filters, and so on 'til white noise.
(With deep apologies to Lewis Fry Richardson, of course)
For those that enjoy this sort of thing the very recent 2023 annual Sky at Night Question Time features several exchanges on exo-planet fun - it's pitched to the general public.
https://thetvdb.com/series/the-sky-at-night/seasons/official...
Free to watch in the UK via the BBC, from abroad via a proxy, otherwise there are groups dedicated to sharing such material.
https://docuwiki.net/index.php?title=Question_Time_%28BBC_th...
My understanding of these things, generally, is that they disregard every candidate outside of a narrow set drawn from theory-based atmospheric models. Like: it's 90% Bayesian priors + 10% experimental evidence. I agree with your sense that it's sketchy how they present these results to the public.
I've complained about this at least twice on HN (on unrelated JWST research papers):
https://news.ycombinator.com/item?id=37473122 (dimethyl sulfide bullshit)
https://news.ycombinator.com/item?id=37782393 (polyaromatic hydrocarbon bullshit)
I’ll be looking for ways to use this in casual conversation
Sounds terrible.