How we first found iron in the atmosphere of KELT-9b on paper

First discovery of iron and titanium in the atmosphere of an ultra-hot Jupiter

Go to the profile of Kevin Heng
Aug 15, 2018
Like 2 Comment

The paper in Nature is here:

Being trained classically as a theoretical astrophysicist, I was taught that one of the functions of theory is to predict: to anticipate the outcome of natural phenomena before it happens.  (There is value to explaining phenomena after the fact, often jokingly derided as "post-diction", but I digress.)  In astronomy, this means making a prediction for what astronomers will observe before they actually point their telescopes at an object.  (Or before they realise the phenomenon is hidden in their already-taken data, as in our case.)  Our story centres around an exoplanet named KELT-9b, which belongs to a class of exoplanets known as "ultra-hot Jupiters".  I first encountered this object during my work as one of the core science team members of the Characterising Exoplanet Satellite (CHEOPS) mission, which is a Swiss-led space telescope of the European Space Agency (ESA) dedicated to detecting small exoplanets orbiting the nearest, brightest stars.  

On CHEOPS, I serve informally as the spokesperson for the atmospheric characterisation "theme" (ATMOS.CHAR), which functions as a de facto working group.  (It is a humbling experience for a theoretician to lead a group of observers, but that is another story.)  One of the tasks of ATMOS.CHAR is to identify the best candidates for measuring secondary eclipses and phase curves - CHEOPS GTO (guaranteed telescope observations) programs led by Monika Lendl and Brice-Olivier Demory.  If an exoplanet is sufficiently cool in temperature, then the secondary eclipse translates into the so-called "geometric albedo", which is the reflectivity of the exoplanet at zero phase.  Phase curves are essentially crude, one-dimensional maps of reflected light, when measured in the optical or visible range of wavelengths.  KELT-9b is especially interesting for CHEOPS, because it is so hot that its blackbody peak resides in the middle of the CHEOPS bandpass.  In other words, CHEOPS secondary eclipses and phase curves are recording thermal emission from KELT-9b.

KELT-9b stood out for several more reasons.  For starters, this gas-giant exoplanet has an equilibrium temperature in excess of 4000 degrees Kelvin (!)  The caveat is that the equilibrium temperature is not necessarily the actual atmospheric temperature: Earth's atmospheric temperature at its surface is a balmy 15 degrees Celsius on average, whereas Venus's is an inferno well in excess of 400 degrees Celsius, yet their equilibrium temperatures are not too dissimilar.  KELT-9b probably has the atmospheric temperatures of a star; in fact, the equilibrium temperature of the diminutive TRAPPIST-1 star is only slightly higher than 2500 degrees Kelvin.  At face value, it means that KELT-9b has the mass and radius of a Jupiter-like exoplanet, but potentially the atmospheric chemistry of a star - at least on its hot dayside.  I was further inspired by a paper by Bell & Cowan (2018), which pointed out a fascinating property of KELT-9b that is almost certainly present in its atmosphere: the dayside, which has a z-band brightness temperature of 4600 degrees Kelvin, is so hot that hydrogen exists in its atomic form, unlike for the regular hot Jupiters.  Yet, there is the potential that the nightside of KELT-9b is cool enough to be dominated by molecular hydrogen.  In other words, this Janus-like object may have the chemistry both of a star and an exoplanet - in the same atmosphere!

Intrigued, I begged one of my senior postdocs, Daniel Kitzmann, to make some calculations using his chemistry code (FastChem, which is an open-source community resource).  We reasoned that the high temperatures expected of KELT-9b's atmosphere meant that it was most probably in chemical equilibrium and cloud-free.  Chemical equilibrium is a wonderful thing for a theorist, because it means that chemistry becomes local: it only depends on the temperature and pressure at each point in the atmosphere, as well as its elemental abundances.  It retains no memory of its past.  (None of the planetary atmospheres in the Solar System are in chemical equilibrium.)  Clouds are probably absent because it is difficult to condense out any solid material from the gas at 4000 degrees Kelvin.  All in all, the prediction is that KELT-9b's atmosphere is a tightly constrained chemical system, controlled only by the temperature and elemental abundances (C/H, O/H, N/H, etc).  Processes such as photochemistry or atmospheric mixing, which lead to chemical disequilibrium, are probably negated by the unusually high temperatures.  Daniel's calculations predicted a few fascinating properties of KELT-9b's atmosphere: carbon monoxide is the dominant molecule (and not just the dominant carbon carrier), and whether water is present at the atmospheric limb (which is probed by transmission spectroscopy) depends sensitively on the temperature and the carbon-to-oxygen ratio (C/O).  Furthermore, metals such as iron, magnesium and titanium are not sequestered into molecules such as enstatite or titanium oxide, but rather exist in their atomic form - unbounded to any other atom.  The high temperatures also means that some of these atoms may be ionized by collisions, rather than by ultraviolet radiation from the star (photochemistry) - similar to conditions in a hot plasma.

It is funny sometimes how science works - one can learn a lot by simply assimilating information without having an agenda.  Earlier in my career, I had the privilege to work with one of the living legends of astrophysics, the great Rashid Sunyaev of "alpha-disk" and Sunyaev-Zeldovich Effect fame.  We were studying shocks and supernova remnants, and I distinctly remember Rashid telling me that the spectral lines of neutral, atomic iron are special, because they are numerous and show up across the optical to near-infrared range of wavelengths.  Somehow, this thought stuck in my mind for the next decade - partially because I was starstruck by one of my scientific heroes.  Immediately following Daniel's theoretical predictions, I asked Jens Hoeijmakers, who is a postdoc working with both me and David Ehrenreich, to simulate how these iron lines would look like when measured with a high-resolution spectrograph.  His predictions were that the iron lines would be easily detected using either the HARPS-N or CARMENES spectrograph even with only one transit of KELT-9b.

As is my usual practice, we posted the submitted version of the Kitzmann et al. (2018) manuscript on the archive ( on 19th April 2018.  As a testament to the growing interest in ultra-hot Jupiters, several other papers came tumbling out of the woodwork on 30th April 2018.  Kreidberg et al. (2018) reported phase curve measurements of WASP-103b, which they confronted with a suite of models.  Lothringer et al. (2018) and Parmentier et al. (2018) both explored the theory of ultra-hot Jovian atmospheres - in ways similar to what we did in Kitzmann et al. (2018).  The comparison of data vs. models of HAT-P-7b by Mansfield et al. (2018) followed a day later.  In all of these studies, chemical equilibrium was assumed for the atmosphere, whereas in the case of KELT-9b Kitzmann et al. (2018) showed explicitly that this was not an unreasonable approximation. Despite our manuscript appearing 11 days before  Lothringer et al. (2018) and Parmentier et al. (2018), these in-review manuscripts did not cite our work.  In the second and accepted (by the Astrophysical Journal) version of our manuscript,  we mentioned all of these studies and critiqued the fact that chemical equilibrium was simply assumed and not investigated.  Therefore, a calculation may be "self-consistent" in the sense that chemical and radiative equilibria are iterated with each other, but one or both of these assumptions may still be approximate - it is precision at the expense of accuracy.  Self-consistency will forever remain an ideal and a cudgel that should be wielded with grace.

Shortly after Kitzmann et al. (2018) appeared on the archive, Jens chatted with David about KELT-9b.  A light went off in David's head.  By coincidence, he already had a HARPS-N dataset of KELT-9b hidden in his metaphorical drawer, obtained using Director's Discretionary Time (DDT).  The Geneva astronomers had searched this dataset for atomic hydrogen (expressed via the H-alpha line) and sodium, but lacked the theoretical motivation to conduct a serious search for metals such as iron.  We sought the help of Simon Grimm, who is (among other things) an expert in the computation of the opacities of atoms and molecules.  These opacities are not trivial to compute, because one needs to evaluate the strengths and shapes of millions to billions of spectral lines in a reasonable amount of time - a skill Simon mastered using a combination of high performance computing know-how (including an intimate knowledge of arcane memory management strategies on GPU cards), a careful curation of the spectroscopic databases and meticulous attention to detail.  (Robust calculations of opacities are typically dismissed as an afterthought, but our own experience is that they are a half-time job.)

From these opacities, we then constructed template transmission spectra.  Another peculiar property of KELT-9b is that the continuum of its transmission spectrum is expected to be dominated by the absorption of light by hydrogen anions, unlike for regular hot Jupiters where Rayleigh scattering by molecular hydrogen is expected to be important.  In fact, the hydrogen anion bound-free absorption is so strong that it drowns out the line wings of the metal lines - a convenient property for us, because a complete theory of the shape of these line wings, due to pressure broadening, remains elusive.  (In technical parlance, the shape of spectral lines is usually described by the Voigt profile - this is Astrophysics 101.  The issue with the Voigt profile is that the shape of the far line wings is given by the Lorentz profile, but this only holds true for atoms or molecules in isolation.  In a real situation, collisions alter the shape of these far line wings and the true shape remains challenging to calculate.  This is a question of unknown physics, rather than computation.)  

We then cross-correlated the measured HARPS-N spectrum with the templates of neutral iron, singly-ionized iron, neutral titanium and singly-ionized titanium.  The results were astonishing.  Both the iron atom and its singly-ionized counterpart jumped out at us from the data - at about 9 and 26 standard deviations, respectively.  The titanium atom was absent, but its singly-ionized counterpart was present at 18 standard deviations.  It is interesting that the non-detection of neutral titanium showed up at about 3 standard deviations - a more optimistic set of authors may have claimed it as a detection.  The fact that the ions showed up more strongly than the neutrals implied a high temperature (so as to facilitate collisions that would remove an electron each from neutral iron and titanium), which we estimated to be at least 4000 degrees Kelvin.  This was the first robust detection of iron and titanium in any exoplanetary atmosphere.  Period.  It was a career-defining moment for Jens.

Our next thoughts were strategic.  In parallel, as Jens was confirming his analysis and discovery, I had tabled a pre-submission inquiry to Nature editor Leslie Sage, who judged the discovery to be significant enough to be sent out to reviewers.  We were acutely aware that a competing team had very similar data of KELT-9b (across a slightly different range of wavelengths) from the CARMENES spectrograph, which meant they were capable of making the exact same discovery if they had the same idea of searching for iron and titanium in the data.  As an advisor looking out for his postdoc, I saw no reason to share the glory and dilute the impact of this discovery on Jens's career.  At this point, we had a key decision to make: submit to one of the regular, major journals and post the submitted manuscript immediately on the archive, or take the route of attempting to publish in Nature, which would delay the announcement of the discovery (because submitted manuscripts cannot be posted) but have an important long-term effect on Jens's career prospects.  After deliberating for a day, we stuck to Nature and lobbied for an expedited review process.

The next few days were a blur.  The first draft of the manuscript was written by me in about 36 hours.  (The final version that was acceptable to all of our co-authors took somewhat longer.)  Barely four days after the manuscript was sent out for review, two reviewers responded with glowing verdicts - they were unanimous in their appraisal that these discoveries are important.  The work took the combined effort of all of the co-authors, as we executed a series of tasks using almost military division of labour: everyone had a key task, no matter how small, as the data reduction, analysis and interpretation came together amidst a frenzy of caffeine and email.  Theoreticians and observers learned to show respect for each other.  I ruffled a few feathers by functioning as the "whip", yanking folks away from obsessing about secondary details and sticking to the big picture and our key goal.  It was a truly thrilling experience.

What is the long-term implication of our discovery?  Metals (or more accurately, refractory elements) have long been an ingredient in the theory of exoplanet formation, but they have never been directly detected - and certainly not at high spectral resolutions where their individuals lines are resolved.  A major caveat is certainly that the abundances of metals in the atmosphere of an ultra-hot Jupiter may not reflect the abundances in its interior, and therefore one cannot get a handle on the true bulk abundances of metals using these high-resolution measurements.  Nevertheless, as more ultra-hot Jupiters continue to be discovered and characterised (by, for example, the TESS space mission), we will obtain a statistical sample of gas-giant exoplanets with a corresponding inventory of metals.  CHEOPS will have a key role to play in the characterisation of ultra-hot Jupiters.  With ultra-hot Jupiters, the simplicity of their atmospheric chemistry means there is no place for the theorist to hide her/his free parameters, and theory may be decisively confronted by observations.


1. Artist's impression of KELT-9b's atmosphere by Denis Bajram, showing an observer on a red-hot KELT-9b staring at the blue glow of the KELT-9 star

2. Bell & Cowan, 2018, "Increased Heat Transport in Ultra-hot Jupiter Atmospheres through H2 Dissociation and Recombination," Astrophysical Journal Letters, 857, L20

3. Kitzmann, Heng, Rimmer, Hoeijmakers, Tsai, Malik, Lendl, Detrick & Demory, 2018, "The Peculiar Atmospheric Chemistry of KELT-9b," Astrophysical Journal, in press (arXiv:1804.07137v1 on 19th April 2018)

4. Kreidberg et al., 2018, "Global Climate and Atmospheric Composition of the Ultra-hot Jupiter WASP-103b from HST and Spitzer Phase Curve Observations," Astronomical Journal, 156, 17 (arXiv:1805.00029v1 on 30th April 2018)

5. Lothringer et al., 2018, "Extremely Irradiated Hot Jupiters: Non-Oxide Inversions, H- Opacity, and Thermal Dissociation of Molecules," under review at AAS Journals (arXiv:1805.00038v1 on 30th April 2018)

6. Parmentier et al., 2018, "From thermal dissociation to condensation in the atmospheres of ultra hot Jupiters: WASP-121b in context," under review at Astronomy & Astrophysics (arXiv:1805.00096v1 on 30th April 2018)

7. Mansfield et al., 2018, "An HST/WFC3 Thermal Emission Spectrum of the Hot Jupiter HAT-P-7b," Astronomical Journal, 156, 10 (arXiv:1805.00424v1 on 1st May 2018)

8. Hoeijmakers, Ehrenreich, Heng, Kitzmann, Grimm, Allart, Deitrick, Wyttenbach, Oreshenko, Pino, Rimmer, Molinari & Di Fabrizio, 2018, "Atomic iron and titanium in the atmosphere of the exoplanet KELT-9b," Nature Letters (doi: s41586-018-0401-y)

Go to the profile of Kevin Heng

Kevin Heng

Professor and CSH Director, Center for Space and Habitability (CSH)

Theory, simulation and phenomenology of exoplanetary atmospheres, including radiative transfer, fluid dynamics, chemistry and inversion methods. I am also a writer for the American Scientist magazine and the recipient of the 2018 Chambliss Astronomical Writing Award of the American Astronomical Society. I am the author of "Exoplanetary Atmospheres: Theoretical Concepts & Foundations," a graduate-level textbook published by Princeton University Press.


Go to the profile of Leslie Sage
Leslie Sage about 1 month ago

Actually, Nature places no restrictions on posting to arxiv. There can, however, be complications, so it's worth consulting with your editor and with your local public affairs office.