Gas-depleted planet formation occurred in the four-planet system around the red dwarf LHS 1903 Thomas G. Wilson, Anna M. Simpson, Andrew Collier Cameron, Ryan Cloutier,, Vardan Adibekyan, Ancy Anna John, Yann Alibert, Manu Stalport, Jo Ann Egger,, Andrea Bonfanti, Nicolas Billot, Pascal Guterman, Pierre F. L. Maxted,, Attila E. Simon, Sérgio G. Sousa, Malcolm Fridlund, Mathias Beck, Anja Bekkelien,, Sébastien Salmon, Valérie Van Grootel, Luca Fossati, Alexander James Mustill,, Hugh P. Osborn, Tiziano Zingales, Matthew J. Hooton, Laura Affer,, Suzanne Aigrain, Roi Alonso, Guillem Anglada, Alexandros Antoniadis-Karnavas,, Tamas Bárczy, David Barrado Navascues, Susana C. C. Barros, Wolfgang Baumjohann,, Thomas Beck, Willy Benz, Federico Biondi, Xavier Bonfils, Luca Borsato,, Alexis Brandeker, Christopher Broeg, Lars A. Buchhave, Maximilian Buder,, Juan Cabrera, Sebastian Carrazco Gaxiola, David Charbonneau, Sébastien Charnoz,, David R. Ciardi, Karen A. Collins, Kevin I. Collins, Rosario Cosentino,, Szilard Csizmadia, Patricio E. Cubillos, Shweta Dalal, Mario Damasso,, James R. A. Davenport, Melvyn B. Davies, Magali Deleuil, Laetitia Delrez,, Olivier D. S. Demangeon, Brice-Olivier Demory, Victoria DiTomasso, Diana Dragomir,, Courtney D. Dressing, Xavier Dumusque, David Ehrenreich, Anders Erikson,, Emma Esparza-Borges, Andrea Fortier, Izuru Fukuda, Akihiko Fukui,, Davide Gandolfi, Adriano Ghedina, Steven Giacalone, Holden Gill, Michaël Gillon,, Yilen Gómez Maqueo Chew, Manuel Güdel, Pere Guerra, Maximilian N. Günther,, Nathan Hara, Avet Harutyunyan, Yuya Hayashi, Raphaëlle D. Haywood,, Rae Holcomb, Keith Horne, Sergio Hoyer, Chelsea X. Huang, Masahiro Ikoma,, Kate G. Isaak, James A. G. Jackman, Jon M. Jenkins, Eric L. N. Jensen,, Daniel Jontof-Hutter, Yugo Kawai, Laszlo L. Kiss, Ben S. Lakeland, Jacques Laskar,, David W. Latham, Alain Lecavelier des Etangs, Adrien Leleu, Monika Lendl,, Jerome de Leon, Florian Lienhard, Mercedes López-Morales, Christophe Lovis,, Michael B. Lund, Rafael Luque, Demetrio Magrin, Luca Malavolta,, Aldo F. Martínez Fiorenzano, Andrew W. Mayo, Michel Mayor, Christoph Mordasini,, Annelies Mortier, Felipe Murgas, Norio Narita, Valerio Nascimbeni,, Belinda A. Nicholson, Göran Olofsson, Roland Ottensamer, Isabella Pagano,, Larissa Palethorpe, Enric Pallé, Hannu Parviainen, Marco Pedani, Francesco A. Pepe,, Gisbert Peter, Matteo Pinamonti, Giampaolo Piotto, Don Pollacco, Ennio Poretti,, Didier Queloz, Samuel N. Quinn, Roberto Ragazzoni, Nicola Rando, David Rapetti,, Francesco Ratti, Heike Rauer, Federica Rescigno, Ignasi Ribas, Ken Rice,, George R. Ricker, Paul Robertson, Thierry de Roche, Laurence Sabin, Nuno C. Santos,, Dimitar D. Sasselov, Arjun B. Savel, Gaetano Scandariato, Nicole Schanche,, Urs Schroffenegger, Richard P. Schwarz, Sara Seager, Ramotholo Sefako,, Damien Ségransan, Avi Shporer, André M. Silva, Alexis M. S. Smith,, Alessandro Sozzetti, Manfred Steller, Gyula M. Szabó, Motohide Tamura,, Nicolas Thomas, Amy Tuson, Stéphane Udry, Andrew Vanderburg, Roland K. Vanderspek,, Julia Venturini, Francesco Verrecchia, Nicholas A. Walton, Christopher A. Watson,, Robert D. Wells, Joshua N. Winn, Roberto Zambelli, & Carl Ziegler Abstract Small exoplanet radii show two populations, referred to as super-Earths and sub-Neptunes, separated by a gap known as the radius valley. This may be produced by the removal of atmospheres due to stellar or internal heating, or lack of an initial envelope. We use transit photometry and radial velocity measurements to detect and characterize four planets orbiting LHS 1903, a red dwarf (M-dwarf) star in the Milky Way’s thick disk. The planets have orbital periods between 2.2 and 29.3 days, and span the radius valley within a single planetary system. The derived densities indicate that LHS 1903 b is rocky, while LHS 1903 c and LHS 1903 d have extended atmospheres. Although the most distant planet from the host star, LHS 1903 e, has no gaseous envelope, indicating it formed from gas-depleted material. Planets form in protoplanetary disk environments and evolve under irradiation from their host stars or via internal mechanisms. Theoretical models of these processes can be built from first principles or constrained by comparing them to the demographics of observed exoplanets. Systems containing multiple planets provide further information because all planets formed together in the same disk and have been exposed to the same star throughout their lifetimes, eliminating factors (i.e. mass, temperature, metallicity) which vary between systems and complicate the interpretation of larger samples. One such demographic trend is the radius valley [ 74 , 195 , 46 ] , a dearth of planets with radii between 1.6 to 1.8 Earth radius (R ⊕ ), is thought to separate smaller, rocky (iron- and silicate-rich) planets from larger bodies that have water or gas layers [ 173 , 211 , 140 ] . Planet formation [ 139 , 200 , 98 , 199 ] and evolution [ 159 , 80 , 38 ] theories have reproduced the radius valley for planets orbiting Sun-like stars [ 195 ] . However, for red dwarf (M-dwarf) host stars, studies disagree on the properties or existence of a radius valley [ 93 , 196 , 163 , 28 ] . Proposed explanations of the radius valley include thermally-driven mass loss (T-DML) or gas-depleted formation (G-DF) [ 159 , 139 , 80 ] . T-DML models predict that the radius valley is evolution-based caused by planet atmospheric escape due to stellar irradiation (via photo-evaporation) [ 159 ] or due to internal planet heating (core-powered mass loss) [ 80 ] . G-DF theory explains the radius valley by a lack of material within the protoplanetary disk during formation [ 139 ] . Those mechanisms would produce different trends in planet composition (rocky, water, or gaseous) as a function of distance from the host star [ 139 ] . However, there are few known multi-planet M-dwarf systems with sufficiently long-period planets to test these theories. Most previously identified multi-planet systems orbiting M-dwarf stars do not have sufficiently precise radius and mass measurements [ 46 , 158 ] to draw strong conclusions. Identification of the LHS 1903 planetary system LHS 1903 (TOI-1730, Gaia DR3 978086481343568128) is a M-dwarf star [ 1 ] that was photometrically observed by the Transiting Exoplanet Survey Satellite (TESS) [ 172 ] between 2019 and 2023 to obtain \sim 69.8 d of science data. Calibrated TESS images were corrected for non-astrophysical signals and processed into light curves that can be used to find planetary systems. Orbiting objects that pass between their host stars and our telescopes produce photometric dips called transits. We searched the TESS data for such signals and found three sets of periodic features that passed diagnostic tests [ 1 ] , as seen in a plot of flux as a function of time (Fig. S1). These were ascribed to three transiting planets with planet radius ( R_{\rm p} ) \sim 1.5, 2.0, and 2.5 R ⊕ , with orbital periods of 2.16, 6.23, and 12.57 day (d), respectively [ 1 ] . With these properties, these objects span the M-dwarf radius valley and thus offer a unique opportunity to test theoretical formation and evolution models. To determine the masses of the planets and characterize the system, we initiated observations of the LHS 1903 planets using multiple facilities. Follow-up observations of LHS 1903 We obtained space- and ground-based photometry from the Characterising Exoplanets Satellite (CHEOPS) [ 25 ] , Las Cumbres Observatory Global Telescope (LCOGT) [ 36 ] , Multicolor Simultaneous Camera for Studying Atmospheres of Transiting Exoplanets 2 (MuSCAT2) [ 154 ] and 3 (MuSCAT3) [ 155 ] , and Search and Characterisation of Transiting Exoplanets (SAINT-EX) [ 57 ] . We also obtained high-resolution spectroscopy of the host star using the High Accuracy Radial Velocity Planet Searcher-North (HARPS-N) [ 50 ] spectrograph. High angular resolution images were obtained using the adaptive optics instruments Palomar High Angular Resolution Observer (PHARO) [ 87 ] and Shane Adaptive Optics Infrared Camera-Spectrograph (ShARCS) [ 175 ] . A summary of the observational datasets is provided in Tables S1 and S2, and Figs. S2-S8. We corrected, calibrated, and processed the PHARO and ShARCS near-infrared time series data into stacked images for each instrument. By performing injection and recovery analyses, we converted these images into sensitivity curves that allowed us to rule out the presence of main sequence stellar companions 8.9 au from LHS 1903 [ 1 ] . The lack of excess noise in Gaia astrometry rejects Jupiter mass planets and brown dwarfs out to 1 and 10 au, respectively [ 1 ] . Therefore, by refuting binarity for LHS 1903, we confirmed that the TESS transits were caused by orbiting planets (hereafter LHS 1903 b, c, and d) and that the host star is a single, old M-dwarf. We conducted spectral analysis on the HARPS-N data to determine the stellar effective temperature ( T_{\rm eff} ) = 3664 \pm 70 kelvin (K), stellar surface gravity ( \log{g} ) = 4.75 \pm 0.12 centimetre per square second (cm s -2 ), and stellar iron abundance ([Fe/H]) = -0.11 \pm 0.09 values [ 1 ] . From broadband photometry and stellar atmospheric and evolutionary models, we executed infrared flux method and isochronal analyses to determine the stellar radius ( R_{\star} ) = 0.539 \pm 0.014 Solar radius (R ⊙ ), stellar mass ( M_{\star} ) = 0.538 {}^{+0.039}_{-0.030} Solar mass (M ⊙ ), stellar density ( \rho_{\star} ) = 3.44 \pm 0.35 Solar density ( \rho_{\odot} ) (4850 \pm 490 kilogram per cubic metre ( {\rm kg\,m^{-3}} )), and stellar age ( t_{\star} ) = 7.08 {}^{+2.87}_{-1.98} gigayear (Gyr) [ 1 ] . The uncertainties listed here and hereafter are 1 \sigma , unless stated otherwise. The measured properties of the host star are listed in Tables 1 and S3. The kinematic properties of the host star were obtained by computing the Galactic velocities from Gaia astrometry and HARPS-N radial velocities (RV) and via comparison to known stellar populations, we found that LHS 1903 is more likely than not (63 \% to 37 \% ) to be part of the thick disk compared to the thin disk [ 1 ] (Fig. S9). The stellar position and velocity highlights that LHS 1903 is in a low-density Galactic region typical of thick disk stars, and the stellar Galactic eccentricity is consistent with the thick disk population, but greater than 99.86 \% of thin disk stars. Our CHEOPS photometric images were corrected for instrumental and environmental systematics with the \sim 300.72 hours (h) of scientific data extracted using the optimal aperture [ 1 ] . Moreover, because there is a background star close to the line of sight towards LHS 1903, the re-derived CHEOPS light curves account for background-induced noise [ 1 , 204 ] . These observations revealed 12, seven, and four transits of LHS 1903 b, c,and d, respectively. Furthermore, the LCOGT, MuSCAT2, MuSCAT3, and SAINT-EX data were processed via standard pipelines [ 1 ] with extracted differential photometry covering planetary transits of the LHS 1903 system. Combining these light curves with the TESS transit data allow for accurate measurement of the physical and orbital properties of the LHS 1903 planets. The CHEOPS observations showed additional transit-like features (Figs. S2 and S3). This prompted us to re-extract the TESS data [ 1 ] with a different data reduction pipeline (Fig. S10, Table S4). These data show two, three, and two further transits in the TESS, CHEOPS, and LCOGT photometry, all from the same object. After excluding potential false positives [ 1 ] , we concluded that these transits are due to a fourth planet (LHS 1903 e) on an orbital period of 29.32 d. The HARPS-N spectra were reduced using the standard data reduction software with the optimal stellar cross correlation function mask [ 1 ] to produce stellar activity indicators. We utilised a template matching pipeline to optimise RV extraction and planetary mass measurement. Following quality control cuts, we obtained 91 high precision RVs spanning 769 d. These data provide supporting evidence for the four LHS 1903 planets (Fig. S5) and show the stellar rotation period to be \sim 40.8 d [ 1 ] . Physical properties of the planets To determine the properties of the LHS 1903 planets, we analysed all the transit photometry and RV data simultaneously [ 1 ] . We fitted the data with a combined astrophysical model consisting of four transiting planets on Keplerian orbits, using the juliet software [ 67 ] . This analysis constrained the planetary orbital period ( P ), transit centre time ( T_{0} ), planet-to-star radius ratio ( R_{\rm p}/R_{\star} ), transit impact parameter ( b ), eccentricity ( e ), argument of peristron ( \omega ) in degrees (deg), and RV semi-amplitude ( K ) in metre per second ( {\rm m\,s}^{-1} ). We explored the parameter space using the nested sampling technique [ 1 ] . From the mass and radius of LHS 1903 (see Table 1), the stellar density is used to constrain the transit model. We accounted for stellar variability or instrumental signals in the TESS and CHEOPS photometry, and HARPS-N RVs using a scalpels [ 47 ] -derived linear noise model [ 1 ] and three Gaussian Process (GP) regressions. The assumed Bayesian prior probabilities and best-fitting model parameters of all data are listed in Tables S5, S6, and S7. All four planets are detected with statistical significance of \geq 40 \sigma in the transit photometry, and \geq 4 \sigma in the RV data [ 1 ] . The model indicates that LHS 1903 b has a radius of 1.382 \pm 0.046 R ⊕ , typical of super-Earth planets [ 173 ] , and an equilibrium temperature ( T_{\rm eq} ) of 796 \pm 20 K. The corresponding values for planet c are 2.046 {}^{+0.078}_{-0.074} R ⊕ , 559 \pm 14 K, and for planet d are 2.500 {}^{+0.078}_{-0.077} R ⊕ , 442 \pm 11 K; both are sub-Neptune size. LHS 1903 e has a radius of 1.732 {}^{+0.059}_{-0.058} R ⊕ , smaller than planets c and d, but larger than planet b. LHS 1903 e has an equilibrium temperature of 333 {}^{+9}_{-8} K, which is colder than most known exoplanets [ 184 ] . The planet masses ( M_{\rm p} ) are constrained by the RVs from the HARPS-N data; for planets b to e they are 3.28 \pm 0.42, 4.55 {}^{+0.73}_{-0.69} , 5.96 {}^{+1.15}_{-1.13} , and 5.79 {}^{+1.60}_{-1.61} Earth mass (M ⊕ ), respectively. The fitted and derived stellar and planet parameters are listed in Table 1, including planet transit depth ( \delta_{\rm tr} ) in parts per million (ppm), stellar radius to semi-major axis ratio ( R_{\star}/a ), planet semi-major axis ( a ) in au, transit duration ( t_{14} ) in h, planet instellation ( S_{\rm p} ) in Earth instellation (S ⊕ ), and planet surface gravity ( g_{\rm p} ) in metre per square second ( {\rm m\,s}^{-2} ), with fitted limb-darkening and noise parameters shown in Tables S6 and S7. The fitted transit model is compared to the observed photometric data in Figs. S1, S2, S3, and S7. The combined fitted Keplerian orbits are compared to the HARPS-N RVs in Fig. S4. The posterior probability distributions for the main model parameters are shown in Figs. S11 to S14. The phase-folded transit photometry and RVs, assuming the best-fitting orbital periods for all four planets, are presented in Figs. 1 and 2. Fig. 3 compares the masses, radii, and densities of the four planets orbiting LHS 1903 to other precisely measured (planet radius uncertainty ( \sigma\,R_{\rm p} ) \,<\, 5% & planet mass uncertainty ( \sigma\,M_{\rm p} ) \,<\, 33%) exoplanets orbiting M-dwarfs. Comparison with radius valley models We compare the LHS 1903 planets to theoretical predictions of the radius valley for exoplanets orbiting M-dwarf stars. The predictions of the T-DML and G-DF models vary as a function of stellar mass. More massive M-dwarfs are expected to have had higher dust masses in their protoplanetary disks [ 112 ] , which would increase the maximum mass and radius of a rocky planet in the G-DF framework. In the T-DML model, lower mass M-dwarfs emit a greater fraction of their bolometric flux at X-ray plus extreme ultraviolet (XUV) wavelengths [ 179 ] , causing more rapid atmospheric loss from any orbiting planets. We corrected for these effects by scaling [ 207 ] the radius valley predictions to the mass of LHS 1903. We find that LHS 1903 b is located below the M-dwarf radius valley [ 46 ] predictions ( R_{\rm p,T-DML} for the T-DML model and R_{\rm p,G-DF} for the G-DF model) for its orbital period ( R_{\rm p,T-DML} = 2.21 R ⊕ , R_{\rm p,G-DF} = 1.69 R ⊕ ). However, planets c and d have larger radii than the predictions ( R_{\rm p,T-DML} = 1.99 R ⊕ , R_{\rm p,G-DF} = 1.74 R ⊕ , and R_{\rm p,T-DML} = 1.84 R ⊕ , R_{\rm p,G-DF} = 1.78 R ⊕ , respectively). With a radius of 1.732 {}^{+0.059}_{-0.058} R ⊕ , LHS 1903 e lies between the predictions of the two models ( R_{\rm p,T-DML} = 1.65 R ⊕ , R_{\rm p,G-DF} = 1.83 R ⊕ ), see Fig. 4. The four exoplanets orbiting LHS 1903 span the M-dwarf radius valley within a single system, so can distinguish between these models. The T-DML [ 159 , 80 ] model, due to either photo-evaporation or core-powered mass loss, would predict gas-rich exoplanets at greater distances from the host star (longer orbital periods), and rocky planets closer to the star (shorter orbital periods). This would lead to a negative gradient in radius as a function of orbital period. Conversely, the G-DF [ 139 ] model produces larger rocky bodies at longer orbital periods, and smaller rocky planets closer to their host stars. In this model, the larger rocky bodies form at later times, after the majority of the gas in the protoplanetary disk has been dissipated by irradiation from the star [ 130 , 131 ] . A similar process is invoked in formation models of the rocky planets in the Solar System, including Earth [ 167 , 151 ] . This model predicts a higher dust-to-gas ratio at larger distances from the star, leading to a positive radius valley gradient. Comparing these model predictions to the radius and orbital period of LHS 1903 e indicates that this planet is predicted to be gas-rich if formed by T-DML [ 159 , 80 ] and rocky if formed by G-DF [ 139 ] mechanisms. We therefore investigate the composition (gas-rich or rocky) of this object. As a first step, we computed the bulk densities ( \rho_{\rm p} ) of all four planets in the LHS 1903 system (Fig. 3 and Table 1). We find that planet b has a density of 1.24 {}^{+0.21}_{-0.19} Earth density ( \rho_{\oplus} ), equivalent to 6.82 {}^{+1.15}_{-1.04} gram per cubic centimetre ( {\rm g\,cm^{-3}} ), which is consistent with being rocky [ 173 , 211 ] . This contradicts a previous proposal [ 116 ] that stars in the Milky Way’s thick disk probably do not host terrestrial planets, as does TOI-561 [ 124 ] , another thick disk system containing a rocky planet. The densities of LHS 1903 c and d are 0.53 {}^{+0.11}_{-0.09} and 0.38 {}^{+0.09}_{-0.08} \rho_{\oplus} (2.91 {}^{+0.60}_{-0.52} and 2.09 {}^{+0.47}_{-0.43} {\rm g\,cm^{-3}} ), which is consistent with a primary rocky body with a lower-density envelope, perhaps of water or hydrogen & helium gas [ 211 ] . LHS 1903 e has a density of 1.11 {}^{+0.33}_{-0.31} \rho_{\oplus} (6.10 {}^{+1.83}_{-1.71} {\rm g\,cm^{-3}} ), which like planet b is consistent with a lack of an extended gaseous atmosphere and being purely rocky. Fig. S15 shows the densities of the LHS 1903 planets, normalised by an Earth-like density, compared other planets that orbit M-dwarfs. Modelling the LHS 1903 planet interior compositions To provide a more robust estimate of the LHS 1903 planet compositions, we conducted an interior structure analysis via comparison of their mass and radius measurements to interior structure models. We used the planetic [ 132 , 204 , 124 , 63 ] which models planets as consisting of four layers: an iron core, silicate mantle, water, and a hydrogen & helium atmosphere. We allowed the core and mantle mass fractions to vary between zero and one, and the water mass fraction between zero and 0.5 [ 1 ] . This model does not include atmospheric water (including steam), due to a lack of observational atmospheric composition constraints of these planets. However, steam could be a notable atmospheric constituent for bodies hotter than 400 K [ 6 , 38 , 199 ] , such as LHS 1903 b, c, and d. The resulting planetic model outputs have an increasing hydrogen & helium gas mass fraction with increasing orbital distance for the three innermost planets (Fig. S15A), from a gas-devoid planet b (hydrogen & helium gas mass fraction \sim 10 -9.5±1.2 ) to \sim 10 -3.1±1.2 (equivalent to \sim 0.1%) for LHS 1903 c and \sim 10 -1.4±0.2 (equivalent to \sim 4%) for LHS 1903 d. However this trend is not followed by the outermost LHS 1903 e, which is modelled as gas-depleted ( \sim 10 -7.7±2.1 ). Using the measured LHS 1903 planet masses, the gas mass fractions translate into absolute gas masses of \sim 10 -9.0±1.3 , \sim 10 -2.4±1.3 , \sim 10 -0.6±0.3 , and \sim 10 -6.9±2.2 M ⊕ , respectively. The gas masses of LHS 1903 d and e therefore differ by > 3 \sigma . For all four planets, the water mass fractions are poorly constrained (Fig. S15B). Testing theory with planetary gas masses Our internal structure models of the three inner planets indicate that the gas-poor planet b and the gas-rich planets c and d are consistent with the radius valley predictions from both the T-DML and G-DF scenarios (Fig. 4). However LHS 1903 e falls between the two model predictions, so can distinguish between them. Our internal structure model indicates that planet e is gas-poor, so should fall below the predicted positions [ 173 ] of the radius valley in Fig. 4. To assess the compatibility of planet e’s measured radius (Table 1) with the two theoretical predictions, we integrated the posterior probability distribution of its radius between zero and the radii predicted by the stellar mass-corrected G-DF and T-DML models at the planet’s orbital period ( R_{\rm p,G-DF} = 1.83 R ⊕ and R_{\rm p,T-DML} = 1.65 R ⊕ , respectively). We find that 94% of the posterior probability distribution is consistent with the G-DF prediction, and 9% is consistent with the T-DML model. As a check, we also computed the probability [ 201 , 166 ] that each of the two radius valley predictions is consistent with the true radius posterior probability distribution of LHS 1903 e. This metric indicates that, the G-DF prediction has a 100% probability of being consistent with the radius of LHS 1903 e whereas the T-DML prediction consistency probability is \sim 10%. We therefore conclude that the bulk densities and compositions of the four planets orbiting LHS 1903 are most likely explained by the G-DF formation scenario. Both the T-DML and G-DF models assume that only a hydrogen & helium dominant atmosphere for a rocky planet; other atmospheric compositions are not considered. A possible interpretation of these conclusions is that the T-DML and G-DF mechanisms are dominant in different regimes (stellar irradiation, age) [ 120 ] of a planetary system. Alternatively, the protoplanetary disk of LHS 1903 might have had an outer edge that receded inwards during planet formation due to the inward radial drift of dust particles ( \sim megayear (Myr) timescales) [ 165 ] . Another possibility is that LHS 1903 e could have formed after the gas had dissipated from the protoplanetary disk, but planets c and d formed before gas dissipation [ 41 ] . Previous models of exoplanet atmospheric evolution via core-powered mass loss and photo-evaporation have shown that small planets with low equilibrium temperatures, such as LHS 1903 d and e, do not undergo substantial atmospheric mass loss [ 120 ] . Other models that seek to reproduce the radius valley have argued that the gap separates rocky and water-rich planets [ 199 ] . In combination, these studies indicate that small, cold, high-density planets cannot be produced by the evolution of gaseous bodies, but must be formed rocky or water-rich without a substantial atmosphere. We therefore infer that the radius, density, and internal structure of planet e indicates that it formed water-rich in a gas-depleted environment [ 140 ] . We tested the potential effect of thermally-driven mass loss on the evolution of the LHS 1903 planetary system, using the Bayesian atmospheric evolution code pasta [ 1 , 30 ] . pasta models planetary atmospheric loss through internal heating (i.e. core-powered mass loss) and high energy stellar irradiation (i.e. photo-evaporation) over the lifetime of the system to compute the resulting gas mass fractions of planets. By comparing these values to the output of our internal structure model of the LHS 1903 planets, we find that the current atmospheric gas mass fractions cannot be reproduced by the evolution processes implemented in pasta . This supports our conclusion that the planets formed in a gas-depleted environment. Trends within the multi-planet system A gas-depleted formation scenario for LHS 1903 e can explain the observed trends in radius and gas mass fraction with the system. The stellar gravitation influence on planets decreases with increasing orbital distance that results in larger exoplanet Hill radii. This means that exoplanets orbiting further from their host stars could attract more material. Planet formation theory predicts that the maximum planetesimal mass attained by accretion from the protoplanetary disk is greater at larger orbital distances, and that the critical mass required for planetesimal gas accretion is lower at greater distances [ 17 ] . This favours larger planets with higher gas mass fractions at longer orbital periods, as found in population synthesis studies [ 65 ] . From our internal structure modelling, we find that the atmospheric mass of the outer planet e is \sim 3 \sigma lower than the inner LHS 1903 d, contrary to the predictions of planet formation theory [ 17 ] . To quantify the trends within the LHS 1903 system, we computed correlation metrics [ 150 ] for the radii, masses, bulk densities, and gas mass fractions of the four planets. These correlation parameters evaluate the logarithmic average variation of each measured quantity over multiple planets. We find that the inner three planets of LHS 1903 have an ordered architecture, as characterised by a coefficient of similarity ( C_{\rm s} ) greater than 0.2, with each parameter are increasing or decreasing uniformly with orbital period ( C_{\rm s} = 0.30, 0.30, 0.59, and 0.95, respectively). However LHS 1903 e does not follow these trends ( C_{\rm s} = 0.08, 0.19, 0.04, and 0.07, respectively), producing a mixed architecture. This observed trend is not consistent with gas accretion during planet formation. It is possible that LHS 1903 e formed later than the other planets, in an evolved radially-truncated disk that had been depleted of its gas. Alternative planet formation models have proposed an inside-out mechanism [ 41 ] whereby the formation of an inner planet produces a local pressure minimum at the planet location. This induces an exterior higher pressure maximum, which subsequently forms another planet. This cycle continues with outer planets forming in sequence. If the first (interior) planet forms \sim 1 Myr after formation of the disk, then the repeating nature of this mechanism causes each exterior planet to form \sim 1 Myr after the previous planet [ 94 ] . Protoplanetary disks around M-dwarf systems, such as LHS 1903, have smaller radii than those around higher mass main sequence stars [ 13 ] so are predicted to have shorter timescales for dust to drift through the protoplanetary disk. For low-mass stars, simulations have predicted that the dust mass within the protoplanetary disk would decrease by 2 orders of magnitude after \sim 1 Myr and become fully-depleted after 3 to 4 Myr, with the disk radial extent reducing by a factor of 3 to 5 over these timescales [ 165 ] . Observations have shown that most protoplanetary disks around M-dwarfs dissipate after \sim 5 Myr [ 164 ] , which is consistent with the inside-out scenario. Comparison to a planet formation simulation We compare the LHS 1903 system to a M-dwarf planetary population synthesis analysis [ 37 ] constructed from a framework of formation and evolution theory [ 65 ] . In that simulation, planetesimals form and grow in a gas and dust coupled disk with the internal structure of the planets monitored during the accretion of dust, ice, and gas. The evolution of each planet is calculated including atmospheric escape and migration through gas-driven, tidal forces, and planet-planet scattering modelled by a N-body code that also includes impacts. This includes host star evolution, over the time scale of of the simulation (5 Gyr). We selected simulated planets with radii, masses, semi-major axes, and gas-mass fractions within 25% or 3 \sigma (whichever was larger) [ 194 , 63 ] of the LHS 1903 planets. We found 45, 72, 61, and 56 simulated planets that are similar to LHS 1903 b, c, d, and e, respectively. For each of these planet samples, we take the average of the gas and dust disk masses, the gas disk outer radii, and the formation location. This determines the average formation environment of each simulated planet sample that represents the formation environment of the LHS 1903 planets. We find consistent parameters for the LHS 1903 b-like and e-like planets: gas disk masses \sim 0.02 M ⊙ , dust disk masses \sim 100 M ⊕ , gas disk outer radii \sim 200 astronomical unit (au), although LHS 1903 e-like planets formed at greater semi-major axes in the simulations. However, the LHS 1903 c- and d-like simulated planets must have formed in a more extensive and massive dust and gas disk to produce their low densities and inferred higher gas-mass fractions, when compared to the rocky, gas-poor LHS 1903 b and e. The simulation study [ 37 ] predicted that planets with radii, masses, and semi-major axes similar to LHS 1903 e, but with larger gas-mass fractions (i.e. akin to LHS 1903 c and d) must form in more massive and radially extended gas and dust disks. We therefore suggest that LHS 1903 b and e might have formed in a lower-mass region of the protoplanetary disk than LHS 1903 c and d. For LHS 1903 b, this could be because planets at smaller orbital distances have less material to accrete, and therefore lower maximum masses, compared to more distant planets that have a larger reservoir of material [ 17 ] . However the opposite effect would occur for LHS 1903 e, so it might instead have formed later [ 41 ] , after the protoplanetary disk has experienced substantial material drift [ 13 ] , resulting in formation in a gas-depleted environment. The rocky Solar System planets formed 10 Myr after the disk formed, in a gas-depleted environment [ 167 , 151 ] . For LHS 1903, this implies that LHS 1903 c and d formed earlier, in gas-rich environments. The LHS 1903 planets might have migrated from their formation sites to the current orbits, but LHS 1903 d and e probably experienced similar of thermal evolution due to their similar masses and low equilibrium temperatures, so the observed gas mass fractions of LHS 1903 d and e are probably unchanged since their formation [ 120 ] . Dynamical history of the LHS 1903 system Dynamical histories of planetary systems may affect planet properties during formation and early evolution. The orbital period ratios of the LHS 1903 planets are close to mean motion resonances (MMRs): at 2:1 for planets c and d, and at 7:3 for planets d and e. The numerical analysis of fundamental frequencies method [ 127 , 1 ] shows that planets c and d are not in MMR, but planets d and e might be in the 7:3 MMR. Given this multi-planet architecture, we assess the orbital evolution of the planets [ 1 ] and find that the LHS 1903 system is dynamically stable against destructive excursions over long timescales. Impacts within planetary systems might remove gas from the planet atmospheres. However, simulations have shown that impacts with the required energy also cause catastrophic disruption of the planet [ 58 , 59 ] . This is because cold, small planets, such as the LHS 1903 bodies, have high escape velocities [ 202 ] and low initial atmospheric thermal excitation that inhibits atmospheric expansion and therefore loss [ 26 ] . Therefore, an impact that would overcome these limitations would likely disrupt the planet. If collisions could remove atmospheres, inner planets would undergo more impacts, due to their shorter orbital periods and therefore more orbit crossing events. We calculated the planet-impactor collision probability using previous methods [ 105 ] with the stellar and planetary properties of the LHS 1903 system, and find impact probabilities over the system age of 0.37%, 0.19%, 0.10%, and 0.02% for LHS 1903 b, c, d, and e respectively. We conclude that LHS 1903 e is unlikely to have experienced a sufficient impact to cause its atmosphere to be stripped. Summary and conclusions The four planets in the LHS 1903 system do not follow simple trends in their radii and gas masses with orbital distance. LHS 1903 e is a gas-depleted long-period planet orbiting a thick disk star, which are rare in previous observations [ 116 ] . The properties of these planets within a single dynamically stable system cannot be explained by atmospheric evolution or planetary impacts alone, but instead indicate that at least one of the planets formed in a gas-depleted region of the protoplanetary disk. Therefore, the gas-depleted formation mechanism is the most likely scenario that can result in the innermost rocky planet LHS 1903 b, whilst preserving the gas-rich nature of planets LHS 1903 c and d, and yielding an outer gas-poor planet, LHS 1903 e. Fig. 1: Detrended phase-folded transit photometry of the LHS 1903  planets . TESS and CHEOPS data phase-folded to the orbital periods from our global analysis [ 1 ] for the four LHS 1903 planets; b is green, c is purple, d is orange, and e is cyan, vertically offset by 0.01 for clarity. ( A & C ) TESS and CHEOPS individual observations are shown as grey points, open circles show the same data binned every 20 minutes (error bars are 1 \sigma uncertainties), and solid lines are the transit models fitted to the data. ( B & D ) TESS and CHEOPS residuals between the data and the model. Fig. 2: Detrended phase-folded radial velocity measurements for the LHS 1903  planets . ( A, C, E, G ) Detrended HARPS-N RV data for LHS 1903 b, c, d, and e, respectively, as black data points, error bars are 1 \sigma . The x-axis has been phase-folded to each planets’ orbital period (Table 1), determined from the joint fitting [ 1 ] . Coloured lines are Keplerian orbit models fitted to the data, with shaded regions indicating the 1 \sigma uncertainty. ( B, D, F, H ) Residuals between the data and the model for LHS 1903 b, c, d, and e, respectively. Fig. 3: Planet masses as a function of bulk densities and radii for selected planets orbiting M-dwarf stars . ( A ) Circular data points are well-characterised planets ( \sigma\,R_{\rm p}\,<\, 5% & \sigma\,M_{\rm p}\,<\, 33%) [ 184 ] orbiting M-dwarf stars, coloured by their equilibrium temperatures. Purple diamonds are the same quantities for M-dwarf stars [ 160 ] . Star symbols with black outlines are the four planets and host star in the LHS 1903 system. Coloured lines are theoretical mass-density curves for small planets (solid lines, see legend) [ 212 , 138 ] , and giant planets (dot-dashed line) [ 19 ] , and M-dwarf stars (black dashed line) [ 20 ] . ( B ) The masses and radii of small ( R_{\rm p}< 3 R ⊕ ) planets orbiting M-dwarf stars, coloured by their equilibrium temperatures. Coloured lines are theoretical mass-radius relations for gas-poor (solid) and gas-rich (dashed) [ 212 , 138 ] planets. The black curve is the collisional stripping limit [ 144 ] . The theoretical models provide general guidance but do not determine the specific composition of each individual system due to difference in stellar parameters, irradiation, etc. Fig. 4: Orbital periods as a function of planet radii for selected planets orbiting M-dwarf stars . Data points are planets with precise radius and mass measurements ( \sigma\,R_{\rm p}\,<\, 5% & \sigma\,M_{\rm p}\,<\, 33%), as a function of radius and coloured by their bulk densities. Error bars are 1 \sigma uncertainties. Star symbols indicate the LHS 1903 planets. The black lines show the LHS 1903 stellar mass-corrected predictions [ 207 ] for gas-depleted formation (G-DF, solid line) [ 139 ] and thermally-driven mass loss (T-DML, dashed line) [ 159 , 80 ] models of the radius valley [ 195 , 146 , 46 , 91 ] . Planets above or below each line are predicted to be gaseous or rocky by that model, as labelled. The LHS 1903 planets are located in three different regions. The mass and radius of LHS 1903 e indicate that it is rocky (see text), which is contrary to the T-DML prediction but consistent with the G-DF prediction of the radius valley. Table 1: Stellar and planetary properties of the LHS 1903 system . Stellar parameters were determined from the spectroscopy [ 1 ] . Fitted and derived stellar and planet parameters were determined by joint fitting of the transit photometry and RV data [ 1 ] . Values and uncertainties are the median and 16th/84th percentiles of the posterior probability distribution. Times are given in Barycentric Julian Date (BJD). Parameter (unit) Value LHS 1903 Spectroscopic stellar parameters T_{\rm eff} (K) 3664 \pm 70 \log{g} (cm s -2 ) 4.75 \pm 0.12 [Fe/H] (dex) -0.11 \pm 0.09 R_{\star} (R ⊙ ) 0.539 \pm 0.014 M_{\star} (M ⊙ ) 0.538 {}^{+0.039}_{-0.030} t_{\star} (Gyr) 7.08 {}^{+2.87}_{-1.98} \rho_{\star} ( \rho_{\odot} ) 3.44 \pm 0.35 \rho_{\star} ( {\rm kg\,m^{-3}} ) 4850 \pm 490 Fitted stellar parameters \rho_{\star} ( \mathrm{\rho_{\odot}} ) 3.79 {}^{+0.17}_{-0.14} \rho_{\star} ( {\rm kg\,m^{-3}} ) 5340 {}^{+240}_{-190} LHS 1903 b LHS 1903 c LHS 1903 d LHS 1903 e Fitted planet parameters P (d) 2.1555098 {}^{+0.0000026}_{-0.0000029} 6.226185 {}^{+0.000028}_{-0.000026} 12.566287 {}^{+0.000032}_{-0.000028} 29.31773 {}^{+0.00028}_{-0.00025} T_{0} (BJD-2457000) 1844.5771 {}^{+0.0012}_{-0.0009} 1844.3692 {}^{+0.0026}_{-0.0021} 1844.3901 {}^{+0.0019}_{-0.0021} 1868.8152 {}^{+0.0071}_{-0.0083} R_{\rm p}/R_{\star} 0.02351 {}^{+0.00047}_{-0.00049} 0.03479 {}^{+0.00097}_{-0.00088} 0.04253 {}^{+0.00072}_{-0.00071} 0.02947 \pm 0.00062 b 0.174 {}^{+0.088}_{-0.091} 0.513 {}^{+0.088}_{-0.093} 0.329 {}^{+0.096}_{-0.112} 0.120 {}^{+0.094}_{-0.079} e 0.015 {}^{+0.014}_{-0.010} 0.089 {}^{+0.036}_{-0.030} 0.112 {}^{+0.055}_{-0.044} 0.014 {}^{+0.015}_{-0.010} \omega (deg) 216 {}^{+80}_{-87} 288 {}^{+38}_{-31} 233 {}^{+29}_{-19} 263 {}^{+63}_{-76} K ( {\rm m\,s}^{-1} ) 2.46 \pm 0.30 2.40 {}^{+0.37}_{-0.35} 2.48 \pm 0.46 1.82 \pm 0.50 Derived planet parameters \delta_{\rm tr} (ppm) 553 {}^{+22}_{-23} 1210 {}^{+69}_{-60} 1808 {}^{+62}_{-60} 868 {}^{+37}_{-36} R_{\star}/a 0.09441 {}^{+0.00320}_{-0.00312} 0.04655 {}^{+0.00158}_{-0.00154} 0.02914 {}^{+0.00099}_{-0.00096} 0.01657 {}^{+0.00056}_{-0.00055} R_{\rm p} (R ⊕ ) 1.382 \pm 0.046 2.046 {}^{+0.078}_{-0.074} 2.500 {}^{+0.078}_{-0.077} 1.732 {}^{+0.059}_{-0.058} a (au) 0.02656 {}^{+0.00055}_{-0.00058} 0.05387 {}^{+0.00112}_{-0.00117} 0.08604 {}^{+0.00178}_{-0.00186} 0.15135 {}^{+0.00314}_{-0.00338} t_{14} (h) 1.562 {}^{+0.058}_{-0.057} 1.986 {}^{+0.125}_{-0.134} 2.762 {}^{+0.127}_{-0.134} 3.780 {}^{+0.134}_{-0.133} S_{\rm p} (S ⊕ ) 66.66 {}^{+7.08}_{-6.46} 16.20 {}^{+1.72}_{-1.57} 6.35 {}^{+0.67}_{-0.62} 2.05 {}^{+0.22}_{-0.20} T_{\rm eq} (K) ∗ 796 \pm 20 559 \pm 14 442 \pm 11 333 {}^{+9}_{-8} M_{\mathrm{p}} (M ⊕ ) 3.28 \pm 0.42 4.55 {}^{+0.73}_{-0.69} 5.96 {}^{+1.15}_{-1.13} 5.79 {}^{+1.60}_{-1.61} \rho_{\rm p} ( \rho_{\oplus} ) 1.24 {}^{+0.21}_{-0.19} 0.53 {}^{+0.11}_{-0.09} 0.38 {}^{+0.09}_{-0.08} 1.11 {}^{+0.33}_{-0.31} \rho_{\rm p} ( {\rm g\,cm^{-3}} ) 6.82 {}^{+1.15}_{-1.04} 2.91 {}^{+0.60}_{-0.52} 2.09 {}^{+0.47}_{-0.43} 6.10 {}^{+1.83}_{-1.71} g_{\mathrm{p}} ( {\rm m\,s}^{-2} ) 16.8 \pm 2.4 10.7 \pm 1.8 9.3 \pm 1.9 18.9 \pm 5.4 ∗ Computed assuming zero Bond albedos and perfect energy redistribution [ 1 ] .