Influence of sub- and super-solar metallicities on the compositions of solid planetary building blocks

The composition of the protoplanetary disc is linked to the composition of the host star, where a higher overall metallicity of the host star provides more building blocks for planets. However, most planet formation simulations only link the stellar iron abundance [Fe/H] to planet formation and [Fe/H] in itself is used as a proxy to scale all elements. But large surveys of stellar abundances show that this is not true. We use here stellar abundances from the GALAH surveys to determine the average detailed abundances of Fe, Si, Mg, O, and C for a broad range of [Fe/H] spanning from -0.4 to +0.4. Using an equilibrium chemical model that features the most important rock forming molecules as well as volatile contributions of H$_2$O, CO$_2$, CH$_4$ and CO, we calculate the chemical composition of solid planetary building blocks. Solid building blocks that are formed entirely interior to the water ice line (T>150K) only show an increase in Mg$_2$SiO$_4$ and a decrease in MgSiO$_3$ for increasing host star metallicity, related to the increase of Mg/Si for higher [Fe/H]. Solid planetary building blocks forming exterior to the water ice line (T<150K) show dramatic changes in their composition. The water ice content decreases from around $\sim$50\% at [Fe/H]=-0.4 to $\sim$6\% at [Fe/H]=0.4 in our chemical model. This is mainly caused by the increasing C/O ratio with increasing [Fe/H], which binds most of the oxygen in gaseous CO and CO$_2$, resulting in a small water ice fraction. Planet formation simulations coupled with the chemical model confirm these results by showing that the water ice content of super-Earths decreases with increasing host star metallicity due to the increased C/O ratio. This decrease of the water ice fraction has important consequences for planet formation, planetary composition and the eventual habitability of planetary systems formed around these high metallicity stars.


Introduction
Observations of exoplanets have reveled that close-in super-Earths are the most abundant type of planet within 1 AU (Mayor et al. 2011;Mulders et al. 2018;Zhu & Wu 2018). These super-Earths have typically masses of a few Earth masses and planets within the same system could be of similar size (Weiss et al. 2018) and thus, presumably mass. Even though recent analysis have questioned this trend (Zhu 2019).
These super-Earths exist around many different type of stars, even around small M-dwarfs (Gillon et al. 2016(Gillon et al. , 2017. In addition, there seems to be no trend with host star metallicity regarding the occurrence rate of these super-Earths (Buchhave et al. 2012). This seems to make super-Earths the most robust outcome of planet formation in nature.
However, Sousa et al. (2019) found that planets below 30 Earth masses seem to have a positive correlation in the host star metallicity-period-mass diagram, meaning that the mass of the planet increases with both, host star metallicity and orbital period. The trend that planetary masses increase with host star metallicity is not surprising, as giant planets are more common around host stars with higher metallicity (Santos et al. 2004;Fischer & Valenti 2005;Johnson et al. 2010). The trend with orbital distance could also be explained by host star metallicity. If more building blocks are available, planetary embryos at larger dis-tances can also grow faster and form planets by the accretion of pebbles (Bitsch et al. 2015b), which could explain these trends and could also give rise to the giant planet eccentricity distribution (Buchhave et al. 2018).
If the planetary radius through transit observations and the planetary masses through RV detections are known, the mean density of the planet can be calculated (e.g. see Dorn et al. 2018 for the Trappist-1 system). This can give important information about the planetary composition through interior structure models (Valencia et al. 2007;Sotin et al. 2007;Seager et al. 2007;Fortney et al. 2007;Selsis et al. 2007;Adams et al. 2008;Zeng & Sasselov 2013;Buchhave et al. 2016), which can give important information about the formation history of the planet and their evolution (Vazan et al. 2018). Planets with mean densities close to the terrestrial planets most likely formed in the inner regions of the protoplanetary disc without significant accretion of water ice, while planets with lower densities could harbour a significant fraction of water ice (Zeng et al. 2019). Of course, the planetary radius measurements could be greatly influenced by planetary atmospheres.
The formation pathways of these super-Earth, on the other hand, is still under debate and a lot of different theories have been proposed (Terquem & Papaloizou 2007;Ida & Lin 2008;Ogihara & Ida 2009;Ida & Lin 2010;McNeil & Nelson 2010;Hansen & Murray 2012;Cossou et al. 2014;Chatterjee & Tan 2014;Ogihara et al. 2015;Lee & Chiang 2016;Izidoro et al. 2017). Studies that form super-Earths in the outer parts of the Article number, page 1 of 17 arXiv:1911.09725v1 [astro-ph.EP] 21 Nov 2019 A&A proofs: manuscript no. FeHplanets disc and then let them migrate into the inner regions have the advantage that they can naturally explain water rich super-Earths (Raymond et al. 2018;Izidoro et al. 2019;Bitsch et al. 2019b;Schoonenberg et al. 2019). However, the formation pathway of forming super-Earths in the outer disc and then migrating them inwards depends crucially on the water fraction within the protoplanetary disc, because it influences the growth and migration pathway of planets (Bitsch & Johansen 2016).
Planet formation simulations that include chemical models have mostly used solar like composition. These models either operate in the classical planetesimal accretion scenario (Bond et al. 2010;Marboeuf et al. 2014;Thiabaud et al. 2015;Ronco et al. 2015) or with planet formation assisted at planet migration traps (Alessi et al. 2017;Cridland et al. 2019) or include pebble accretion (Ali-Dib 2017; Madhusudhan et al. 2017). Some of these models additionally focus on the chemical abundances of the planetary atmospheres (Ali-Dib 2017; Madhusudhan et al. 2017;Cridland et al. 2019), while others mainly focus on the chemical composition of rocky/icy super-Earths (Alessi et al. 2017). Here we want to focus on the chemical composition of solid planetary building blocks, which is set by the underlying chemical composition of the disc and not focus on the underlying physics needed for planet formation theories.
The process of planet formation is thought to happen within the protoplanetary disc surrounding the newly formed star. A key assumption in models of planet formation is that the elemental abundances of the protoplanetary disc are directly correlated to the host star, meaning that the disc and the host star have the same elemental composition. However, in planet formation simulations it is not very often taken into account that stellar elemental abundances show different trends when compared to [Fe/H]. In addition most studies of planet formation only focused on solar like composition for the different elements.
These different elemental trends in stars are caused by the fact that different elements are produced in different kind of stars at different times in the galactic evolution (Burbidge et al. 1957). Elements like O, Mg, Si, Ca, Ti (also known as α-elements) as well as S, are mostly produced in massive stars during their evolution and then released in the interstellar space during supernovae explosions (Type II supernovae, SNII): this means that this enrichment is expected to happen on a short timescale (Matteucci & Greggio 1986). Low mass stars instead can pollute the interstellar medium on a longer time scale via, for example, Type Ia supernovae (SNIa), where significant quantities of Fe are released and only small fraction of α-elements (e.g. Thielemann et al. 2002). SNIa are created from white dwarfs in binary systems, resulting in a time delay relative to SNII (e.g. Matteucci et al. 2009) in a range between 0.3 to 1 Gyr (e.g. Valiante et al. 2009). It is important to know that even if different elements share the same production site (like SNII), the difference in mass of the progenitor star and/or the explosion parameters matters (Kobayashi et al. 2006;Kobayashi & Nakasato 2011) for the elemental yields and thus how they are incorporated to molecular clouds that form new stars.
Several works showed the trend of α-elements in respect to [Fe/H] in different sample sizes and different parts of the Galaxy (e.g. Adibekyan et al. 2012;Smiljanic et al. 2014;Bensby et al. 2014;Bensby et al. 2017;Buder et al. 2018). Elements in different elemental groups (light elements, iron-peak elements and neutron-capture elements) might have different trends simply because they are formed in different sites (see for example Bensby et al. 2014;Battistini & Bensby 2015. However these trends have only been taken into account in a very small sample of studies related to planet formation and composition. Recent studies have focused on the mineralogy of super-Earth planets and how it could vary with stellar abundance (Unterborn & Panero 2017;Hinkel & Unterborn 2018;Putirka & Rarick 2019) or by building the planet very close to its host stars where the temperatures are very high changing the molecules that form the planet . These studies took only the formation interior to the water ice line (T>150K) into account and Putirka & Rarick (2019) focused on the mantle mineralogy. Ligi et al. (2019) linked detailed stellar observations of HD219134 to constrain the properties of its planetary companions and determine the core to mantle ratio of these planets. The work by Santos et al. (2017) followed a different approach by studying the chemical composition of planetary building blocks around thin and thick disc stars in general, showing that the planetary building blocks differ in their iron and water mass fraction. Cabral et al. (2019) used stellar population synthesis simulations to predict the chemical composition of planetary building blocks using the same chemical model as Santos et al. (2017), finding similar results.
Here we combine detailed stellar abundance measurements of C, O, Mg, Si and S 1 , from the GALAH survey to calculate the bulk composition of solid planetary building blocks. We do not distinguish in this study planetary host stars and stars that have no detected planets, because the formation of planets seems to be an universal process, where on average nearly all stars should host planets within 1 AU (Mulders et al. 2018). For simplification, we mainly focus only on the individual solid planetary building blocks rather than incorporating them in a complex planet formation model (see appendix C). We thus focus on solids that either form completely interior or exterior to the water ice line.
Our paper is structured as follows. In section 2 we describe our chemical model and the calculation of the solid planetary building blocks. In section 3 we show the compositions of solids formed around host stars with different metallicities and for solids formed completely interior and exterior to the water ice line. In section 4 we apply the chemical model used to calculate the solid planetary building blocks to a planet formation model and show the final water ice content of the formed planets. We then discuss the implications of our results in section 5 and summarize our findings in section 6.

Stellar abundances
For our study we use stellar abundances derived for the second data release of the GALAH (Galactic Archeology with HER-MES) survey (Buder et al. 2018) that contains 342,682 stars. GALAH is a high-resolution (R ∼ 28,000), large-scale stellar spectroscopic survey of Milky Way stars in the magnitude range 12 < V < 14 probing mostly the Galactic thin and thick disk but also a consistent number of halo stars. The Hermes spectrograph observes in four discrete optical wavelength bands (4713-4903Å, 5648-5873Å, 6478-6737Åand 7585-7887Å) meaning that for our studies sulfur is not included in the observations (the triplet S I lines commonly used for chemical abundance determination are located around 6756 Å).
Stellar parameters and elemental abundances are derived using the data-driven approach of T he Cannon (Ness et al. 2015) that first uses a training sample to derive several stellar labels that are going to be propagated to the entire sample (see Buder et al. 2018 for the details). The training sample is analysed using SME (Spectroscopy Made Easy, Valenti & Piskunov 1996;Piskunov & Valenti 2017) that performs spectrum synthesis with 1D LTE (Local Thermal Equilibrium) hydrostatic stellar atmosphere MARCS models. It is important to mention that GALAH DR2 incorporates non-LTE line formation for several elements using departure coefficients from LTE. In respect to our study, O (Amarsi et al. 2016a), Mg (Osorio et al. 2015), Si (Amarsi & Asplund 2017) and Fe (Amarsi et al. 2016b) are corrected for 1D non-LTE.
We selected only dwarf stars (T eff > 5000 and log g > 4.0) that have a good parameter flag value from The Cannon (flag_cannon == 0). At this point we select for each element only the stars for which The Cannon gives a good abundance flag (flag_x_fe == 0): this lead to a sample of 142,000 stars with good O abundance, 27,638 stars for C, 148,765 for Mg and 121,105 for Si.
The fact that so few stars from the original sample have good C abundances, is caused by the combination of detection limits and the flagging algorithm of The Cannon. In fact only one C line is used for the abundance determination (at 6588 Å) and only a small training sample for C is available (see Tab.2 in Buder et al. 2018).
We then grouped the stars selected for each element in metallicity bins of 0.1 dex as shown in Fig. 1 Even if our values are corrected for non-LTE, it is important to remember that they are based on 1D models. Recently new determination of non-LTE values are derived using 3D model atmospheres. In this respect Amarsi et al. (2019) shows new abundances for C, O and Fe corrected for 3D non-LTE for 187 F and G stars in the Galactic disk and halo. It seems that even if the correction for O are more severe in the metal-rich regime, the correction from our 1D non-LTE values should not be big, because our sample is made of dwarf stars where this effect is not that large.
As previously said, there are no S abundance determination from GALAH and we assume that it scales as Si as shown in Chen et al. (2002). The same trend is present in several studies (Caffau et al. 2005;Jönsson et al. 2011;Takeda et al. 2016;Duffau et al. 2017) regarding dwarfs and giants, reassuring us to scale S with the Si abundance trends.

Chemical model
In order to account for the chemical composition of the planet, we include only the major rock and ice forming species. The mixing ratios (by number) of the different species as a function of the elemental number ratios is denoted X/H and corresponds to the abundance of element X compared to hydrogen for solar abundances, which we take from Asplund et al. (2009) and are given as follows: He/H = 0.085; C/H = 2.7 × 10 −4 ; O/H = 4.9 × 10 −4 ; Mg/H = 4.0 × 10 −5 ; Si/H = 3.2 × 10 −5 ; S/H = 1.3 × 10 −5 ; Fe/H = 3.2 × 10 −5 .  (Buder et al. 2018). The error bars are the mean deviations of the observations. The data is given in tab. A.1. Sulfur (not shown) scales in the same way as silicon (Chen et al. 2002).
These different elements can combine to different molecular species. We list these species, as well as their condensation temperature and their volume mixing ratios v Y in Table 1. More details on the chemical model can be found in Madhusudhan et al. (2017) and Bitsch et al. (2018a).
To account for different host star metallicities, we scale the solar values from Asplund et al. (2009) with the corresponding scaling from Fig. 1. In addition, sulfur scales the same as silicon (Chen et al. 2002).
As we focus on solid formation close to the water ice line, the solids in our simulation still form at temperatures larger than 70K, meaning that CO, CH 4 and CO 2 are in gaseous form and they thus do not contribute to the solid planetary building blocks. However, including the gaseous CO and CO 2 binds a lot of oxygen, which is then not available to form water ice, in contrast to the study by Santos et al. (2017) and Cabral et al. (2019). This binding of oxygen can greatly reduce the amount of available water around stars with high C/O ratios. We discuss how the amount of carbon bound in molecules that do not bind any oxygen (e.g. methane) influences our results in appendix B.

Planet formation
As outlined in the introduction, the formation of super-Earths can happen through many different formation channels. However, all core accretion theories agree that solid material, either in the form of pebbles or planetesimals (for a review see Johansen & Lambrechts 2017), must be accreted to reach masses of a few Earth masses. In this work, we focus on the elemental and molecular composition of planetary building blocks formed completely interior (T>150K) and completely exterior (T<150K)to the water ice line, before applying it to a planet formation model. In principle growing planets migrate through the disc (for a review see Baruteau et al. 2014), but accretion, especially in the pebble accretion scenario, can easily be fast enough that the planet is fully grown before it starts to migrate significantly (Bitsch et al. 2015b). However, in reality, planets formed exterior to the water ice line can also cross the water ice line Article number, page 3 of 17  Lodders (2003). For Fe 2 O 3 the condensation temperature for pure iron is adopted (Lodders 2003). Volume mixing ratios v Y (i.e. by number) are adopted for the species as a function of disc elemental abundances (see e.g. Madhusudhan et al. 2014). We note that the Mg abundance is always larger than the Si abundance.
during their assembly, which allows them to have a lower water content compared to planets formed completely exterior to the water ice line (Bitsch et al. 2019b), which can help to explain the formation of the Trappist-1 system (Schoonenberg et al. 2019). The planetary composition in this case is just a mixture between the composition acquired exterior and interior to the water ice line. We show in section 4 and appendix C the results of simple planet formation model including pebble accretion, disc evolution and planet migration. The final water ice content of planets in this model is directly correlated to the temperature at which the solids are accreted and is as such always a mixture between the minimum and maximal allowed water ice content of the model, which we discuss in section 3.

Composition of solid planetary building blocks formed completely interior or exterior to the water ice line
In this section we show the elemental and molecular composition of solid planetary building blocks formed entirely interior or exterior to the water ice line (but interior to the CO 2 ice line) to give an overview of the extreme cases. We then apply the same chemical model in section 4 and study the compositions of planets that grow by pebble accretion and migrate through the disc, where they can cross the ice lines. We first discuss the elemental and molecular composition of solids formed at [Fe/H]=0 and then move to lower and higher [Fe/H] values. We then summarize the findings in Fig. 10 and Fig. 11.

Solar metallicity
In Fig. 2  larger than the solar values. This results in an overabundance of magnesium for our planetary composition. If we instead use directly the solar abundances (Asplund et al. 2009) our model is much closer to the Earth's composition, but does not match to exactly. We attribute these differences to our simple chemical model which only traces the main carriers of material. In addition, our formation model does not account for the impact history on Earth, where impactors originated from different regions in the solar system.
In Fig. 3 we show the molecular and elemental mass fractions of solid planetary building blocks that formed entirely exterior to the water ice line (T<150K). The solid mass is now mainly dominated by water ice (H 2 O) and thus oxygen. The total elemental composition is then obviously not Earth like at all, confirming that the Earths did not form exterior to the water ice line, even though the water ice line can evolve interior to 1 AU during the evolution of the protoplanetary disc (Oka et al. 2011;Bitsch et al. 2015a). In the pebble accretion scenario this problem could be solved by the growth of Jupiter that blocks the inflow of water rich particles into the inner disc , even though the water ice line sweeps interior to 1 AU after 1 Myr according to viscous disc evolution models (Oka et al. 2011;Bitsch et al. 2015a;Baillié et al. 2015

Sub-solar metallicities
In Fig. 4  In Fig. 5 we show the molecular and elemental abundances of solid planetary building blocks formed exterior to the water ice line (T<150K) for sub-solar metallicities. For lower and lower [Fe/H], water ice becomes more and more dominant for the total mass budget of the solids and with it, oxygen. This is clearly related to the fact that the oxygen abundances is enhanced compared to the other elements at low [Fe/H], see Fig. 1. In our chemical model this implies that the relative abundances of rock forming molecules are similar to stars with [Fe/H]=0, but the iron content decreases for lower overall metallicities. This is related to the overabundance of oxygen allowing the formation of more H 2 O, which thus becomes more and more dominant for lower [Fe/H]. As a consequence the water ice fraction in the solid planetary building blocks can be above 50% for [Fe/H]=-0.4.

Super-solar metallicities
In Fig. 6 we display the molecular and elemental mass fractions of solid planetary building blocks formed interior to the water ice line (T>150K) around stars with super-solar metallicity. As can be seen from Fig. 1, the magnesium abundance increases strongest with [Fe/H], resulting in an overall increase of Mg 2 SiO 4 in the mass fraction of the formed solids. However, the overall mass fractions for the different elements only changes slightly, because the relative mass of Mg and O inside of the rock forming molecules MgSiO 3 and Mg 2 SiO 4 changes only slightly as well, leaving the overall elemental trend with [Fe/H] quite constant.
Additionally, the oxygen abundance is reduced compared to the other elements, but there is still enough oxygen to fully oxidize iron, magnesium and silicon. This implies that the reduction of the oxygen abundance relative to the other elements has no influence on the elemental composition of solids formed entirely Article number, page 5 of 17 interior to the water ice line. If the galactic trends for these elements continue for even larger [Fe/H] abundances, we expect a deviation from the here derived mass fractions of the elements inside the solids formed interior to the water ice line, because not enough oxygen to fully oxidize Fe, Mg and Si might be available.
In Fig. 7 we show the elemental and molecular mass fractions of solid planetary building blocks formed entirely exterior to the water ice line (T<150K) around stars with super-solar metallicity. In contrast to solids formed around stars with low [Fe/H] exterior to the water ice line (Fig. 5, the water ice fraction of the solids has decreased dramatically. In fact for [Fe/H]=0.4, the water ice mass fraction is only of a few percent. The low water ice mass fraction is caused by the reduced oxygen abundances compared to the C, Si, Mg, Fe and S abundances around stars with large [Fe/H], see Fig. 1. This lower oxygen abundances still allows Si, Mg and Fe to be fully oxidized, binding a lot of oxygen. In addition, the large carbon abundance (Fig. 12) binds a lot of oxygen into CO and CO 2 , leaving only very little oxygen that can then form water ice (see table 1). This results in a very low water ice abundance in solids formed exterior to the water ice line around metal rich stars. As a consequences, the elemental abundances show an increase of Fe, S, Mg, Si and a reduction of H and O with increasing [Fe/H] due to the reduced availability of water ice. If the same trend for all elements holds true for even larger [Fe/H], our chemical model predicts that water ice would not be available at overall higher metallicities than [Fe/H]>0.4.

Planet formation
In this section we show the results of a simple planet formation model based on pebble accretion to study the final position, masses and water ice content of formed planets. The details of the planet formation model are given in appendix C and follow standard models used in the past (e.g. Bitsch et al. 2015b). Our model contains a decreasing pebble flux with time, an evolving protoplanetary disc model which cools in time, meaning that the ice lines move inward in time and type-I migration for the growing planetary cores. We start our planetary embryos with 0.01 Earth masses. The growth of the planets stops at the pebble isolation mass, when planets start to open a small gap in the protoplanetary discs, which stops the inward flux of pebbles and thus accretion (Lambrechts et al. 2014;Bitsch et al. 2018b). We do not include gas accretion and type-II migration, because our work focuses on the solid composition of material and thus our planet formation model focuses on the formation of super-Earths.
Past simulations have shown that the water ice content in the protoplanetary disc can greatly influence the disc structure and the planet formation pathway regarding growth and migration (Bitsch & Johansen 2016). However constructing a detailed disc model based on the different abundances is beyond the scope of this work. We therefore rely on the disc model derived for solar metallicities presented in Bitsch et al. (2015a) and use this for all our simulations. This approach might not be as accurate, but it allows an easy comparison between the different scenarios, because the planet migration rates are the same due to the same disc model.

Solar metallicity
In Fig. 8  Planets growing in the early stages of the disc (small t 0 ) receive the largest pebble flux and thus grow fastest. In addition, in the early stages of the disc evolution, the pebble isolation mass, at which pebble accretion stops, is highest (e.g. Bitsch et al. 2015b;Bitsch et al. 2019b). This results in the most massive planets that can reach masses of up to 5-10 Earth masses within this model (t 0 < 500 kyr). Due to the model-choice of a constant Stokesnumber for the pebbles, pebbles are much smaller in the outer disc where the gas density is lower, reducing the pebble accretion rate for planets further away from the star.
As the pebble flux decreases in time, the planetary growth rate reduces. Additionally, the disc cools and it's aspect ratio reduces, resulting in a smaller pebble isolation mass at later times. Thus planets formed at large initial times t 0 grow slower and to lower masses than planetary embryos injected at earlier starting times.
During their growth, the planets migrate through the disc. The used disc model (Bitsch et al. 2015a) features a region of outward migration exterior to the water ice line, where planets can be trapped as they grow. These planets could then only be released from the disc when the disc dissipates and the region of outward migration can only hold small planets (Bitsch et al. 2015a). As a result nearly all of the planets in our simulations are trapped exterior to 0.1 AU, which is the inner edge of our disc model. However, planetary systems most likely contain more than one planet, which then results in a different dynamical history of the system. The N-body simulations of Izidoro et al. (2019) and Bitsch et al. (2019a) use the same disc model and show that planetary embryos can migrate inwards in a convoy. This is caused by the mutual interactions between the bodies, which increase their eccentricity. An increase in the planetary eccentricity quenches the entropy driven corotation torque (Bitsch & Kley 2010) resulting in inward migration of these planets. As a consequence, multiple planetary embryos that grow exterior to the water ice line can migrate all the way to the inner edge in a convoy (Izidoro et al. 2019) in contrast to the single planetary embryos used in the here presented work. We thus think that the final semi-major axes of the planets in our simulations might be different if multibody simulations are used, however, the growth of these embryos by pebble accretion is very fast and thus local, so that the water ice content is not affected much by the multiplicity (Izidoro et al. 2019).
The planets that grow entirely exterior to the water ice line and are trapped until they reach the pebble isolation mass have the largest water ice content, as shown by Bitsch et al. (2019b). The water ice content of the planets that formed completely interior to the water ice line is by construction zero, while the water ice content is largest for planets formed just exterior to the water ice line. Planets forming even further away from the water ice line have a lower water ice content, because these planets might accrete CO 2 , CH 4 and CO reducing the relative mass fraction of water ice. However, these planets would have a higher ice fraction. In fact the water ice content of planets formed just exterior to the water ice line and planets formed completely exterior to the CO ice line differ by about a factor of two. The planets formed far away from the star are dominated in CO 2 , CH 4 and CO ice instead of water ice. As expected the water ice content of the planets is between the maximal allowed value (Fig. 3) and zero.
Another factor that can influence the water ice content of the formed planets is the direction of migration. Bitsch et al. (2019b) showed that planets have a lower water ice content in a scenario where planets can only migrate inwards. We see the same effect in our simulations here (see appendix C).

Sub-and super-solar metallicity
In Fig. 9 we show the final masses, positions and water ice fractions of planets formed at different initial distances r 0 and at different initial times t 0 around stars with sub-and super-solar metallicity. The exact elemental abundances for each [Fe/H] value are shown in Fig. 1 and  There are, however, some other small differences for the planet formation pathway. In the low metallicity cases, the planets grow very slow, meaning that they just reach 2-3 Earth masses. Planets of that mass range are trapped in the region of outward migration until disc dissipation at 5 Myr and thus do not migrate to the inner edge. As a consequence at low [Fe/H], basically no planet reaches the inner edge in our disc model. However, chains of multiple planets can migrate inwards due to their mutual increase in eccentricity which reduces outward migration due to the entropy driven corotation torque (Bitsch & Kley 2010;Izidoro et al. 2017;Izidoro et al. 2019). If the metallicity is slightly larger ([Fe/H]>-0.2), some planets become larger than 5 Earth masses, which allows them to migrate to the inner edge of the disc towards the end of the discs lifetime (Planets below the yellow lines at low t 0 in Fig. 9 migrate all inwards to 0.1 AU).
As already shown in Fig. 8, the water ice content of the planets reduces if they are formed further away and the water ice content of the planets formed close to the water ice line is largest. This is, as stated before, caused by the fact that the planets formed exterior to the CO 2 , CH 4 or CO ice line accrete these ices as well. As a result the water ice content of the planet decreases, but the total ice fraction can increase (not shown).
The total water ice content of the planets formed in the simulations is between the maximal amount of water allowed by the chemical model for a given stellar composition (see Fig. 10) and zero. Planetary migration across the water ice line can reduce the water ice content of growing planets, see appendix C. In particular if planetary migration is only directed inwards, the water ice content can be reduced compared to a scenario where planet migration is outwards at the water ice line (Bitsch et al. 2019b). Nevertheless, the highest water ice content of planets can be achieved at low host star metallicity, due to the lowest C/O ratio, which allows the largest water ice content, independently if migration is inwards or outwards.  The decrease in the water ice fraction at large initial distances is caused by the accretion of CO 2 , CH 4 and CO into the planet, which reduces the relative water ice fraction of the planet, but increase the total ice fraction of the planet.

Discussion
In Fig. 10 and Fig. 11 we show the molecular and elemental mass fractions of solid planetary building blocks formed interior (T>150K) and exterior (T<150K) to the water ice line as function of [Fe/H]. These figures summarize the trends described in section 3 to give a better overview over the trends. In addition we show in Fig. 12 the C/O ratio as function of [Fe/H]. In the following we discuss the assumptions and implications of the results of our model on planet formation, where we also show the results of a planet formation model in appendix C.

Planet formation
One of the main questions in planet formation theories is where the first planetesimals and planetary embryos formed that then grew to fully grown planets. In recent years, the water ice line has become the prime suspect for the first planetesimals to form (Ida & Guillot 2016;Drażkowska & Alibert 2017), also because of the large water abundance in the solar system (Lodders 2003). Condensation of water onto grains allows them grow larger than with pure coagulation (Ros & Johansen 2013;Schoonenberg & Ormel 2017;Ros et al. 2019), which allows planetesimals to form easier. This is caused by the larger particle sizes which allow gravitational collapses of pebble clouds to planetesimals at lower overall metallicities (Yang et al. 2017). In contrast, the CO ice line is not thought to harbour the same effect as the water ice line, because the grains drift inwards faster than they can grow through condensation at these distances (Stammler et al. 2017). However all these simulations have been conducted with solar abundances of water and CO. As the host star metallicity increases, our model predicts a significant reduction in water ( Fig. 10 and an increase in CO (Fig. 12)). A change in the water ice fraction has large implications for the formation of planets around the water ice line (Bitsch & Johansen 2016).
In the case of low host star metallicity, the water ice fraction is much larger than for solar metallicity, which could increase the effects of condensation and could thus allow grains to grow larger. This could significantly help planet formation around the ice line in low metallicity environments, where build-ing blocks are rare, but the increased pebble sizes due to condensation allow an easier collapse of pebble clouds to planetesimals (Yang et al. 2017). However, this effect might be counteracted by the overall low metallicity which hinders the gravitational collapse of pebble clouds to planetesimals as well (Yang et al. 2017). This implies that the effect of condensation can not be so strong that giant planet cores grow quickly and easily, due to the lack of giant planets around low metallicity host stars (Santos et al. 2004;Fischer & Valenti 2005;Johnson et al. 2010). Additionally, it could imply that super-Earths formed around low metallicity stars could be water rich, in contrast to super-Earths formed around high metallicity stars, where water ice is rare.
On the other hand, the water ice fraction is very low around stars with super-solar metallicity, which implies that maybe condensation is not the main driver of grain growth and thus planetesimal formation around these stars. Condensation at the water ice line is reduced, but it could be compensated by the overall increased metallicity around these stars, which can trigger planetesimal formation (see Johansen et al. 2014 for a review). The simulations of Yang et al. (2017) show that even small particles with Stokesnumbers around 10 −3 can trigger planetesimal formation, if the overall metallicity is super-solar, indicating that the help of condensation to particle growth is not necessarily required to form planetesimals.
The giant planet occurrence rate increases with host star metallicity (Santos et al. 2004;Fischer & Valenti 2005;Johnson et al. 2010), which is confirmed by many simulations in the core accretion scenario (Ida & Lin 2004;Mordasini et al. 2009;Ndugu et al. 2018). However, these simulations did not take the effects of different elemental scalings with host star metallicity into account and should be revisited due to the effects found here. In addition, the amount of CO and CO 2 around these stars is increased compared to stars with solar abundances due to their elevated C/O ratio (Fig. 12). This could allow condensation at the CO snow line to outperform radial drift, implying that giant planet formation around stars with solar and sub-solar metallicity starts at the water ice line where condensation can increase the pebble sizes, while giant planet formation around high metallicity stars could start at the CO ice line, where our model predicts a larger amount of CO available  Fig. 9. Same meaning as Fig. 8, except that the planets are formed around stars with different [Fe/H] values and thus chemical abundances for the all elements (Fig. 1). Stellar abundances increase from top to bottom. As the stellar abundances increase, the water ice content of the planet decreases due to the lack of water ice, caused by the increasing C/O ratio (Fig. 12).
Article number, page 9 of 17  for for condensation compared to solar and sub-solar metallicity. This, however, needs to be tested with future models like of Drażkowska & Alibert (2017) that combine grain growth and drift with condensation and planetesimal formation. In addition the C/O ratio of giant planet host stars could help to distinguish if water ice condensation is the main driver of giant planet formation.

Composition of super-Earths and gas giants for different host star abundances
In our model we study the composition of solid material around host stars with different metallicities. We use this to calculate the composition of solid planetary building blocks that form interior or exterior to the water ice line. In section 4 we show the water ice content of planets using a planet formation model featuring planet migration, disc evolution and pebble accretion, which resulted in water ice contents of the formed planets between zero and the maximal content allowed for the specific host star abundance. In appendix C we explain our planet formation model in detail. Our planet formation simulations coupled to our chemical model shows that the water ice content of the super-Earths is between zero and the maximal allowed water content of the model ( Fig. 9 and Fig. 10). This result is consistent with other simulations of planet formation including compositions and planet migration (Alessi et al. 2017;Bitsch et al. 2019b).
In Fig. 10 we show the mass fraction of the molecules that contribute to solid planetary building blocks for changing host star metallicities. For planets forming completely interior to the water ice line (T>150K), the molecular composition of the solids only changes slightly, with an increase of Mg 2 SiO 4 by about 15% from [Fe/H]=-0.4 to 0.4, while MgSiO 3 decreases by the same amount over the same range of [Fe/H]. This is caused by the relative stronger increase of magnesium compared to silicon with [Fe/H], see Fig. 1. In the elemental mass fraction of the solids (Fig. 11), on the other hand, the relative mass fractions of all elements stay roughly constant with [Fe/H]. This is caused by a very similar oxygen and magnesium fraction inside of MgSiO 3 and Mg 2 SiO 4 . This implies that the overall composition of rocky planets that formed completely interior to the water ice line (T>150K) Fig. 1  at solar composition our simulations actually predict a composition that is very close to the Earth (McDonough & Sun 1995). On the other hand, some super-Earths have densities much higher than the Earth (Guenther et al. 2017), similar to Mercury. These high densities are caused by a large iron fraction, which we do not reach in our model. For Mercury, the high density could be explained by stripping its mantle through a collision, leaving an iron rich remnant (Benz et al. 1988). In simulations that study the formation of close-in super-Earths, collisions are basically inevitable (Ogihara et al. 2015;Izidoro et al. 2017;Ogihara et al. 2018;Izidoro et al. 2019;Lambrechts et al. 2019), but collisions are normally treated as perfect mergers in these simulations and they thus do not account for mantle stripping. However, this process could help to explain the large densities of some super-Earth planets.
Planets formed completely exterior to the water ice line (70K<T<150K), on the other hand, would show a strong decrease in water ice with increasing host star metallicity ( Fig. 10 and Fig. C.1). This is caused by the binding of oxygen with carbon to form CO and CO 2 and the other rock forming elements, which results in only a small fraction of oxygen that can form water ice. As water ice decreases, Mg 2 SiO 4 increases in a similar fashion as for planets formed interior to the water ice line. The elemental trend here gives a clear picture as well (Fig. 11). As the water ice decreases, the oxygen mass fraction decreases and the mass fraction of the other elements increases accordingly.
The consequences for the composition are quite dramatic and super-Earths formed exterior to the water ice line have a very significant water ice fraction if they form around low metallicity stars and a very tiny water ice fraction if they form around high metallicity stars (Fig. 9). This has important consequences for the formation mechanism of close in super-Earths.
In the Kepler sample, recent detailed observations have revealed a gap in the radii of the observed super-Earths (Fulton et al. 2017;Van Eylen et al. 2018). This gap is generally interpreted as atmospheric mass loss of small planets either by photoevaporation (Owen & Wu 2013;Lopez & Fortney 2014) or by mass loss directly from the cooling of the core (Gupta & Schlichting 2019). As a consequence, it is thought that most close-in super-Earths are of mostly rocky nature (Owen & Wu 2017;Jin & Mordasini 2018), putting constraints on planet formation models. However, some models predict that these super-Earths could have water contents up to 10-20% (Gupta & Schlichting 2019), in agreement with recent observations (Zeng et al. 2019).
For example, the N-body simulations of Izidoro et al. (2019) show that planets assembled exterior to the water ice line should have a large water ice content (50% in their model), mostly inconsistent with the observations. On the other hand, if the assembly of the planets starts exterior to the water ice line, but finishes interior to the water ice line, the final composition could be in line with the observations (Raymond et al. 2018;Bitsch et al. 2019b). For super-Earths formed around high metallicity stars with [Fe/H]>0.2, our model predicts a maximum water ice content of 20%, which is naturally consistent with the observed gap in the planetary radius distribution. This could imply that the radius gap could only tell something about the formation channel of super-Earths around stars with [Fe/H]<0.2. Thus a more detailed analysis of the host stars within the Kepler sample is of crucial importance to constrain planet formation theories. We apply our model to the solid building blocks of planets and have not focused on the gaseous component of the protoplanetary discs and what this could imply for the composition of giant planets. Giant planets that grow and migrate in protoplanetary discs do not only accrete hydrogen and helium in the gas phase, but also heavy elements. The accretion of these elements can be enhanced due to the evaporation of volatiles (e.g. H 2 O, CO 2 , CO) at ice lines (Öberg et al. 2011;Madhusudhan et al. 2017;Booth et al. 2017). As already mentioned before, the C/O ratio in the gas phase changes with host star metallicity ( Fig. 1 and Fig. 12). In addition, hot Jupiters, which are the gas giants whose atmosphere we can characterize, are more common around high metallicity stars (Buchhave et al. 2018). This could naturally imply a change in the C/O ratio of the observed giant planet atmospheres compared to solar composition and should be taken into account in the modeling of the planet formation pathway of these planets, where previous simulations were not tailored to match the exact host star abundances ; Ali-Dib 2017).

Habitability
Planets that are suspect to harbour life, as we know it, reside in the habitable zone around their host star (e.g. Seager 2013), which is defined as a region around the star where water could be in liquid form 2 . This implies that a planet inside the habitable zone should actually harbour water. In the solar system, the Earth formed dry in the inner regions of the protoplanetary disc and water was delivered via impacts of asteroids and comets triggered by the giant planets (Raymond et al. 2009;. Simulations normally assume a water ice composition gradient around the water ice line, where the most water rich objects can have a substantial amount of water ice. However, around stars with high metallicity, the water ice fraction is dramatically reduced (Fig. 10), which could have important implications for the potential habitability of planets inside the habitable zone around high metallicity stars. The study by Abbot et al. (2012), on the other hand, indicates that too much water could actually be potentially dangerous for the habitability of a given planet, implying that planets around stars with a lower metallicity might be more suitable for life. However, as already hinted above, water delivery onto planets is a complicated process that needs much further investigation. Nevertheless, our model suggests that the characterization of potential habitable worlds with future instruments should take the host star composition into account.

Chemical Model
Our used chemical model (table 1) is a simple equilibrium model that does not take the chemical evolution during the protoplanetary disc stage into account, which can alter the chemical abundances of different molecules (Henning & Semenov 2013;Eistrup et al. 2016). Our used model also does not account for the formation of molecules during the infall stage, however simulations using solar-like composition (Furuya et al. 2015) seem to be in agreement with our model at solar composition. However, the impact of a more advanced model seems to be much more profound on the gaseous component (Eistrup et al. 2018) and thus the planetary atmosphere composition (Cridland et al. 2019) than on the solid composition. However, to calculate the solid composition of the material, also the gaseous component has to be taken into account. This is especially important for the oxygen fraction, as a large fraction of oxygen is bound in CO and CO 2 , especially for high C/O ratios.
In the study by Santos et al. (2017) the effect of gas phase CO and CO 2 was not taken into account and they find that the protoplanetary discs should harbour a large water fraction independently of the host star metallicity. We attribute this difference in their model, originally presented in Santos et al. (2015), to ours to the more detailed chemical model we use here. In particular we include the effects of oxygen binding in CO and CO 2 , which binds most of the oxygen, especially for high metallicity stars with large C/O ratio (Fig. 12). Additionally, we include binding of oxygen with iron to form Fe 3 O 4 and Fe 2 O 3 . This results in the low water ice fraction around high metallicity stars in our model compared to Santos et al. (2017).
As the water ice fraction is closely related to the binding of oxygen with carbon to form CO and CO 2 , the water ice fraction would change if a lot of carbon grains would be bound in molecules that do not feature any oxygen, for example in organics or CH 4 (which accounts for 45% of the carbon abundance in our nominal model). We present a chemical model in appendix B, where we decrease the CO abundance to 0.25 of the C/H and instead include an organic component consisting of pure carbon grains with evaporation temperature of 500K (Semenov et al. 2003), but leaving the other parts as in our nominal model (table 1). We find that the water ice content at [Fe/H]=0.4 can increase to 15% for solid planetary building blocks formed exterior to the water ice line. However this is still a factor of 3-4 smaller than the water ice content at [Fe/H]=-0.4, reproducing the trend shown in Fig. 10 from our nominal model. As the water ice content of planets formed exterior to the water ice line is crucial for planet formation, planetary composition and even habitability, future studies of planet formation around stars with different metallicity should take even more detailed chemical models into account. However, some models (e.g. Eistrup et al. 2018) seem to indicate that carbon is most abundant in the form of CO or CO 2 rather than organic material without any oxygen, indicating a trend more consistent with our nominal chemical model (Fig. 10).

Future observations
In the near future GALAH will publish its third data release. This release will contain more data collected after the DR2 but also re-analysis of the previous stars: this will include abundances of new elements (some neutron-capture elements not treated in DR2) and might incorporate new non-LTE corrections coming from 3D models, which could influence the here drawn conclusions.
In addition to this, in the next years other big spectroscopic surveys will start their operation. In particular 4MOST (4m Multi-Object Spectroscopic Telescope, more information on all the surveys can be found in de Jong et al. (2019)) will start to observe at the end of 2021 from the VISTA telescope in Chile and it is planned to observe almost 10 million stars in the Milky Way. Related to this star-planet connection, more than 2.5 millions of objects will be observed in the Galactic disk, both in high (R ∼ 20,000) and low resolution (R ∼ 5000): in particular for the high resolution disk and bulge survey, at least 15% of the targets will be observed also by TESS. For the high-resolution observations more than 20 elemental abundances can be potentially derived and in this list all the elements that are included in this work are going to be observed, including sulfur. The potential coming from 4MOST is incredible: the massive sample homogeneously analysed together with the big overlap with the planet transit mission TESS and other surveys will give enormous amount of constrains and data for planet formation theory, in particular in respect to the star-planet connection.

Summary
In this study we have combined observations of elemental abundances of stars ( Fig. 1) to the composition of planet forming material and planet formation. Our model is based on the assumption that the host star abundance is directly linked to the abundances of elements within the planet forming protoplanetary disc. To calculate the exact composition of solid planetary building blocks interior and exterior to the water ice line we have used a simple equilibrium chemical model (table 1) and show variations of that model in appendix B.
Our model predicts that the elemental composition of planetary building blocks formed completely interior to the water ice line (T>150K) is very similar across all host star metallicities (Fig. 11). However, the elemental mass fractions of planetary building blocks forming completely exterior to the water ice line (T<150K) are very different depending on the host star metallicity. In particular the water ice mass fraction reduces from close to 50% at [Fe/H]=-0.4 to only ∼6% at [Fe/H]=0.4 (Fig. 10) within our chemical model. This has important consequences for the formation, composition and habitability of planets.
Planet formation models (section 4 and appendix C) show that the water ice content of the formed planets can be mixed between the extreme states of zero water ice and the maximal content allowed for the specific disc composition due to planetary migration and the evolution of the water ice line (see also Bitsch et al. 2019b). This trend has also been found in other planet formation simulations just using solar composition (e.g. Alessi et al. 2017).
Additionally, a change in the water ice fraction alters the formation pathway of planets (Bitsch & Johansen 2016). In particular, Bitsch & Johansen (2016) showed that discs with similar metallicity, but higher water ice fraction form more giant planets. Our here presented simulations show that the water ice fraction around high metallicity stars is very low and thus condensation of water ice molecules to assist grains growth (Ros et al. 2019) might not be the main reason to trigger giant planet formation. Instead planetesimal formation might be triggered by the overall large metallicity (Yang et al. 2017). The effects of water ice condensation for grain growth might thus be more important for stars with [Fe/H]<0.2, while its effect might be very small around stars with even higher metallicity. We thus speculate that giant planet formation around stars with [Fe/H]<0.2 might be triggered at the water ice line, but it could be triggered instead at the CO ice line for stars with even higher metallicity due to the large CO abundance caused by the large C/O ratio (Fig. 12), which could cause grain growth at the CO line due to CO condensation.
The change of the water ice fraction with host star metallicity has also direct impacts on the habitability of planets formed in these systems. Systems with intrinsically lower or no water abundance might not be able to form planets with enough water for them to be habitable. Future characterization of atmospheres of exoplanets in the habitable zone should thus take the host star abundances into account.
Our study clearly indicates a large impact on the planetary composition and planet formation pathway depending on the chemical composition of the host star. We thus encourage future studies of planet formation that try to span a wide range of host star metallicities to include the effects of different abundances of different elements.

Appendix A: Stellar abundances
We show in Fig. A.1 the stellar abundances of carbon, oxygen, magnesium and silicon for the GALAH sample of stars used in our work. The overall trends agree with the previous study of Bensby et al. (2014) and the values used for our calculations are shown in table A.1 and plotted in Fig. 1.

Appendix B: Carbon contribution
The carbon fraction of minor bodies in the solar system is much higher than the carbon content within the Earth (e.g. Bergin et al. 2015), implying that the history of the chemical composition could change with orbital distance and planet formation models have to account for these effects. In our model, the water ice abundance is governed by the amount of free oxygen that does not form rock forming molecules or CO and CO 2 and not by orbital distance. Thus an elevated C/O ratio (Fig. 12) reduces the amount of free oxygen that can form water ice. As a consequence the amount of carbon that forms carbon chain molecules without oxygen can influence the water ice content in our model. In our nominal model (table 1) we included a fraction of 45% of the C/H in organics (methane). We relax this approach here for testing purposes and reduce the CO content to 25% and include a 20% C/H fraction in pure carbon grains with an evaporation temperature of 500K (e.g. Semenov et al. 2003), bringing the total carbon content that forms non-oxygen bearing molecules to 65% of the whole C/H, thus increasing the water content. We show these trends in Fig. B.1.
Even with these assumptions, the general trend shown in Fig. 10 (Fig. 12), so that the effect of forming less CO compared to the total amount of oxygen that can form water is not significant. For high [Fe/H], the C/O ratio is much larger, resulting that the effect of forming less CO can actually increase the amount of oxygen that can from water. On the other hand, at high [Fe/H] also the rock forming species that bind oxygen like Fe, Si and Mg are more abundant than oxygen implying that the water ice ratio will decrease with increasing [Fe/H] independently of how much oxygen is bound in carbon to form CO or CO 2 . The exact water content though can be estimated more accurately with more detailed chemical models, however, detailed models seem to indicate that the ration between CO to CO 2 to non-oxygen bearing carbon molecules is maximal equal, where in most cases the non-oxygen bearing carbon molecules are less abundant than CO and CO 2 (Eistrup et al. 2018) which points to a water ice ratio as predicted by our nominal model (Fig. 10).  (Buder et al. 2018). This data is used to plot Fig. 1 where η is the radial pressure gradient at the pebbles position and v K is the Keplerian speed of the gas. We use a pebble flux oḟ The 2D Hill accretion rate onto the planetary core is then given aṡ with r H being the planetary Hill radius and v H = Ω K r H being the Hill speed. The planetary growth is stopped at the pebble isolation mass (Lambrechts et al. 2014;Bitsch et al. 2018b;Ataiee et al. 2018), when the planet starts to carve a small gap in the protoplanetary disc which exerts a pressure bump exterior to it, stopping the inward flux of pebbles (Paardekooper & Mellema 2006) and thus pebble accretion. The slightly simplified pebble isolation mass (Bitsch et al. 2018b) is given as As soon as the planet reaches its pebble isolation mass we stop pebble accretion, but at the same time we do not model the gas accretion component, because we are only interested in the composition of the planetary core. The migration of the planet is determined by the formula of Paardekooper et al. (2011). In this formula, the migration direction and speed depends crucially on the radial gradients of gas surface density and temperature/entropy. In addition outward migration by the entropy driven corotation torque can only happen, if the viscosity is large enough. In the disc model of Bitsch et al. (2015a) this leads to a region of outward migration attached to the water ice line, if α mig > 0.001, where the nominal value in this model is α mig = 0.0054. We also test a variation with α mig = 0.0001 which results in only inward migration at the water ice line, changing the water content to smaller values, because the assembly of the planets can be finished interior to the water ice line (Bitsch et al. 2019b). In section 4 we use α mig = 0.0054, meaning that planets can migrate inwards and outwards. Here we show the results of different migration prescriptions and their influence on the water ice content of planets in a similar fashion as in Bitsch et al. (2019b).
We place planetary embryos of initially 0.01 Earth masses at different initial positions r 0 in discs that are initially already 0.5 Myr old (in order to account for the formation of the planetary embryo). We then vary the [Fe/H] of the host star and calculate the water ice fraction of the formed planets via our nominal chemical model. In Fig. C.1 we show the water ice content of formed planets in our model for three different assumptions, (i) a non-evolving disc with only inward planetary migration modelled with α mig = 0.0001 (top), (ii) an evolving disc with only inward planetary migration (middle) and (iii) an evolving discs where planets can migrate inwards and/or outwards with α mig = 0.0054 (bottom). The final planetary masses are between 3 and 10 Earth masses in all models and if planetary migration is only inwards, planets reach the inner edge at 0.1 AU, while in the scenario with outward migration, planets are trapped at around 1.8AU, which corresponds to the outer edge of the region of outward migration in that disc model (Bitsch et al. 2015a).
For the non-evolving disc with only inward migration, the results are as expected. Planets starting to form exterior to the water ice line initially accrete water ice and then migrate into the inner disc where they finish their assembly and thus feature a low water ice content. All planets in this case reach 0.1 AU.
Article number, page 15 of 17 A&A proofs: manuscript no. FeHplanets Due to the disc's evolution in time (middle and bottom), the water ice line moves inwards, so that the water ice line can sweep over slow growing planets initially interior to the water ice line allowing them to have a water ice contribution.
Planets forming exterior to the CO 2 ice line feature a higher water ice content, because they can finish most of their growth exterior to the water ice line and thus accrete water ice. This is also reflected by the water ice gradient of the planets with increasing initial planetary position, meaning that the farther the planetary embryo is placed exterior to the water ice line, the higher its water ice content. This result is consistent with Bitsch et al. (2019b) derived for solar metallicity. In addition with increasing [Fe/H] the water ice content decreases, as shown in the main paper.
In case the disc evolves, the water ice line moves inward in time due to the decrease in viscous heating and even planetary embryos that are initially placed interior to the water ice line can accrete water ice, because the evolution of the water ice line is initially faster than planet migration (Bitsch et al. 2019b). If planetary accretion is faster, planets grow quicker and they can migrate inwards faster, avoiding the sweep of the inward moving water ice line. This is for example the case at high [Fe/H], where the pebble flux is larger and thus accretion is faster and only planets initially slightly interior to the water ice line can accrete some water ice. The planets in this scenario all migrate to the inner edge of the disc at 0.1 AU.
In the case of an evolving disc where planets can migrate inward and outwards, the water ice content of the fully grown planets is highest of these three models. This is caused by the fact that planets in this model migrate outwards exterior to the water ice line where they then finish their assembly and thus accrete all their mass with a water ice component (Bitsch et al. 2019b). In the end, planets either migrate to the inner edge of the disc at 0.1 AU or stay trapped in the region of outward migration exterior to 1 AU until disc dissipation. The water ice component of the planets decreases slightly for planetary seeds forming exterior to a few AU, because the disc becomes so cold that the planets also accrete a fraction of CO 2 ice. This decreases the relative water ice component of the accreted material and results in a lower water ice fraction of the formed planets.
In any case, all these simulations clearly show that the water ice content of the formed planets is a mixture and always in between the states of zero water ice (planets formed completely interior to the water ice line) or up to the maximal water ice content allowed in the chemical model (planets that form completely exterior to the water ice line). The planet formation simulations also clearly demonstrate the water ice gradient from low [Fe/H] to high [Fe/H], as outlined in the main part of the paper and section 4.